Desenvolvimento Económico Português No Espaço Europeu - 2008

Conferência do Banco de Portugal
16 de Maio de 2008
Desenvolvimento Económico Português

no Espaço Europeu
Portuguese Economic Development

in the European Context
proceedings
Banco de Portugal
Av. Almirante Reis, 71
1150-012 Lisboa
Economic Research Department
Printed by
Selenova - Artes Gráficas, Lda.
Number of copies printed
500 issues
Legal Deposit no. 236983/06
ISBN 978-989-8061-28-7
Nota de Apresentação
Foreword
NOTA DE APRESENTAÇÃO
A Conferência “Desenvolvimento Económico Português no Espaço Europeu: Determinantes e Políticas”

constitui uma iniciativa do Banco de Portugal que agora conhece a sua 4ª edição. Com ela pretende o Banco
incentivar economistas oriundos da academia, muitos deles com posições em universidades estrangeiras, a
reflectirem sobre os desafios do desenvolvimento económico em Portugal. Para além de reflexões com sólida
sustentação teórica e grande rigor técnico, pretendeu-se que os estudos apresentados fossem ainda relevantes do
ponto de vista de política económica para Portugal.
Esta publicação apresenta os trabalhos seleccionados para apresentação na 4ª Conferência do

“Desenvolvimento Económico Português no Espaço Europeu: Determinantes e Políticas” realizada a 16 de Maio
na Fundação Calouste Gulbenkian, em Lisboa.
As áreas cobertas pelos estudos seleccionados para a presente edição da Conferência poderão ser
consideradas menos tradicionais no contexto da análise dos problemas de desenvolvimentos económico,
sobretudo quando comparados com as de anteriores edições.
Podemos agrupar os trabalhos em três grupos temáticos: Regulação e Concorrência, Sustentabilidade da

Segurança Social e Ambiente. Um dos trabalhos em “Regulação e Concorrência” salienta a importância que a
melhoria de protecção legal aos investidores teve na performance da economia portuguesa desde a adesão à EU;
outros, fornecem quadros analíticos que permitem analisar quantitativamente o impacto no bem-estar de fusões
ocorridas (ou tentadas) em certas indústrias como a banca, as comunicações móveis, os cimentos e o retalho de
combustíveis. A qualidade das instituições legais e políticas que enquadram as actividades económicas são
comummente identificadas como das mais importantes determinantes dessa variável complexa que é a
produtividade. Estes trabalhos ajudam a conhecer com mais precisão os mecanismos dessa influência.
O aumento da esperança de vida registado nas últimas décadas tem criado importantes problemas na
sustentabilidade dos Sistemas de Segurança Social não capitalizados. Alguns dos trabalhos do tema relativo à
Segurança Social mostram como o aumento da poupança privada induzido pela maior longevidade permite aos
VI Vítor Constâncio
agentes auto-segurarem, ainda que parcialmente, esse risco. Outro trabalho simula, precisamente, a aplicação em
Portugal de um sistema de subsídios de desemprego geridos em contas individuais em moldes que garantiriam a
sua sustentabilidade bem como a adequação dos incentivos ao trabalho.
O clima é um bem público, cujo uso está cada vez mais presente no cálculo económico de governos e
empresas, e cuja natureza global requer abordagens globais. No tema do “Ambiente” discutem-se os principais
desafios económicos criados pelas alterações climáticas. Em particular, abordam-se as questões dos instrumentos
económicos a usar para controlar emissões e a arquitectura dos acordos internacionais pós-Quioto.
Os estudos que estão na base da Conferência provieram quer de convites directos a alguns economistas quer
de um concurso público dirigido à academia. Seguiu-se um processo de selecção dos estudos, tarefa que foi da
responsabilidade do Comité Científico, com a participação de Mário Centeno (Banco de Portugal e Universidade
Técnica de Lisboa), Isabel Horta Correia (Banco de Portugal e Universidade Católica de Portugal), José Ferreira
Machado (Universidade Nova de Lisboa), Carlos Robalo Marques (Banco de Portugal) e Pedro Portugal (Banco
de Portugal e Universidade Nova de Lisboa).
Agradecimentos são devidos a todos pelo trabalho realizado, tal como ao Departamento de Estudos
Económicos do Banco de Portugal é devido o nosso reconhecimento pela tarefa de organização desta iniciativa.
Vítor Constâncio
Governador do Banco de Portugal
FOREWORD
The Conference “Portuguese economic developments in the European area: determinants and policies” was
created on the initiative of Banco de Portugal and is the fourth of a conference cycle. The purpose of this
Conference is to encourage academic economists, many holding relevant positions at foreign universities, to
reflect on the challenges facing economic developments in Portugal. In addition to reflections based on solid
theoretical work and high technical precision, the research work presented is also relevant for Portugal from an
economic policy perspective.
This publication presents the papers selected for the 4th Conference on “Portuguese economic developments
in the European area: determinants and policies” held on 16 May 2008 at the Calouste Gulbenkian Foundation,
in Lisbon.
The areas covered by the papers selected for this edition of the Conference may be considered less
traditional in the context of the economic development analysis, mainly when compared with the previous
editions.
Papers may be grouped in three areas: Regulation and competition, Social security sustainability and
Environment. One of the papers on the theme “Regulation and competition” stresses the importance that the
improvement of the legal protection of investors has had for the performance of the Portuguese economy since
the accession to the EU; other papers provide analytical tables enabling a quantitative assessment of the impact
of actual (or attempted) mergers on welfare, in some industries such as banking, mobile communications, cement
and fuel retail trade. The quality of the legal and political institutions underlying economic activities is usually
identified as a major determinant of productivity, a complex variable. These papers contribute to a more accurate
knowledge of the mechanisms through which this influence works.
Increasing life expectancy over the last decades has raised major problems regarding the sustainability of
unfunded social security schemes. Some of the papers on Social security show that the increase in private
savings induced by rising longevity enables agents to self-insure such risk, albeit only partially. Another paper
simulates the implementation in Portugal of an unemployment benefit scheme managed through individual
accounts, in such a manner that its sustainability and the adequacy of work incentives are guaranteed.
VIII Vítor Constâncio
Climate is a public good. Its use is increasingly taken on board in the economic calculations of governments
and companies, and its global nature requires global approaches. The theme “Environment” includes discussions
on key economic challenges raised by climate change. In particular, it covers issues related to the economic
instruments to be used with a view to controlling emissions and the architecture of the post-Kyoto international
agreements.
The papers presented at the Conference are the result of invitations directly addressed to some economists
and a call for papers addressed to the academy. The papers were selected by the Scientific Committee, comprised
of: Mário Centeno (Banco de Portugal and Universidade Técnica de Lisboa), Isabel Horta Correia (Banco de
Portugal and Universidade Católica de Portugal), José Ferreira Machado (Universidade Nova de Lisboa), Carlos
Robalo Marques (Banco de Portugal) and Pedro Portugal (Banco de Portugal and Universidade Nova de Lisboa).
Our sincere thanks are due to all those involved for their work. We are also grateful to the Economic
Research Department of Banco de Portugal for the organisation of this initiative.
Vítor Constâncio
Governor of Banco de Portugal
Contents
Contents XI
Nota de Apresentação / Foreword
Nota de Apresentação ..................................................................................................................................... V

Foreword......................................................................................................................................................... VII
Vítor Constâncio
Session 1
Hedging Longevity Risk ................................................................................................................................ 3
João Cocco (London Business School)
Francisco Gomes (London Business School)
The Impact of Firm Size and Market Size Asymmetries on National Mergers in a Three-Country
Model ............................................................................................................................................................. 27
Luís Santos-Pinto (Faculdade de Economia, Universidade Nova de Lisboa)
Merger Analysis in the Banking Industry: the Mortgage Loans and the Short Term Corporate Credit
Markets ........................................................................................................................................................... 67
Duarte Brito (Faculdade de Ciências e Tecnologia, Universidade Nova de Lisboa)
Pedro Pereira (Autoridade da Concorrência)
Tiago Ribeiro (Indera - Estudos Económicos, Lda)
Session 2
The Economic Effects of Improving Investor Rights in Portugal................................................................... 125
Rui Castro (Université de Montréal)
Longevity Risk, Retirement Savings and Individual Welfare......................................................................... 173

João Cocco (London Business School)
Francisco Gomes (London Business School)
Simulation of Unemployment Insurance Savings Accounts in Portugal ........................................................ 203

Ricardo Rodrigues (European Centre for Social Welfare Policy and Research)
Session 3
The Economics of Climate Chance: an Overview .......................................................................................... 235
Maria A. Cunha-e-Sá (Faculdade de Economia, Universidade Nova de Lisboa)
Session 1
HEDGING LONGEVITY RISK *
João F. Cocco†
London Business School
Francisco J. Gomes‡
April 2008
Abstract
We study hedging of longevity risk in an empirically parameterized life-cycle model of consumption

and savings decisions. In particular we study the optimal life-cycle portfolio allocation to longevity bonds
and the welfare gains from investing in this asset. We first show that, if households know the exact
parameters of the stochastic process for mortality rates, then the benefits from investing in this asset are very
small. In fact, a negative risk premium of 10 basis points (in line with recent estimates) is enough to deter
investors from holding these bonds almost completely. However, we also show that, small misperceptions
about future improvements in mortality rates are enough to induce very large welfare gains from investing in
longevity bonds: between 0.5% and 1% of annual consumption. This indicates that investors might not
actually buy these bonds even if they actually stand to gain significantly from investing in them.
*
We would like to thank Mario Centeno, David Laibson, Alex Michaelides, Olivia Mitchell and seminar participants at the London School
of Economics, Stockholm School of Economics, Tilburg University, Warwick University, and The Wharton School for comments.
†
London Business School, Regent’s Park, London NW1 4SA, UK and CEPR. Tel (020) 70008216. Email jcocco@london.edu.
‡
London Business School, Regent’s Park, London NW1 4SA, UK and CEPR. Tel (020) 70008215. Email fgomes@london.edu.
4 Joao Cocco - Francisco Gomes
1. Introduction
Over the last few decades there has been an unprecedented increase in life expectancy. These large increases
in life expectancy were, to a large extent, unexpected, and as a result they have often been underestimated by
actuaries and insurers. These unprecedented longevity increases are also, to a large extent, responsible for the
underfunding of pay as you go state pensions,1 and of defined-benefit company sponsored pension plans.
Longevity risk, or the risk that the individual might live longer than he expected when making her savings
decisions, may be reduced by the purchase of annuities at retirement age. However, although the purchase of
annuities at retirement age provides insurance against longevity risk as of this age, a young individual saving for
retirement still faces substantial uncertainty as to what aggregate life expectancy and annuity prices will actually
be when he retires. This problem is made significantly worse, by the fact that the payouts of defined benefit
pension plans are likely to be negatively correlated with aggregate survival rates. As a result exactly when
longevity increases, and households need more wealth to finance their consumption at retirement, they are
actually likely to receive a lower pension. Furthermore, markets may be incomplete in the sense that individuals
may lack the financial assets that would allow them to insure against this risk.
There have been recent attempts to address this market incompleteness, and offer economic agents financial
instruments that would allow them to hedge against longevity risk. In December 2003, Swiss Re. issued a $400m
three-year life catastrophe bond.2 This was a direct attempt by Swiss Re. to insure itself against a catastrophic
mortality deterioration (e.g. a pandemic). This bond offered an opposite hedge to pension funds and other
annuity providers, since when aggregate mortality rates increase their liabilities decrease. This issue was well
received and fully subscribed, and followed by a second bond placement in April 2005. The second bond had a
five-year maturity, and a total principal amount of $362m. This second issue was also fully subscribed. In the
meantime, in November 2004, The European Investment Bank (EIB) announced the issuance of a 25-year
£540m longevity bond.3 Despite receiving substantial public attention, the issue was not well perceived by
market participants, and was eventually withdrawn.
With these financial innovations in mind, we extent the set of available assets in the model, and allow the
agent to invest in financial assets whose returns are correlated with longevity shocks, which can therefore be
used to hedge longevity risk. Such investments could be made directly or directly or indirectly, by owning shares
1. The decrease in birth rates that has occurred over this period has also contributed to the underfunding.
2. Designing financial instruments for hedging longevity risk was first proposed by Blake and Burrows (2001), who recommend that
governments should issue survival bonds, to allow the private sector to hedge this source of risk.
3. The bond was supposed to be issued by the EIB, with Partner Re. acting as the longevity reinsurer and BNP Paribas structuring, managing,
and marketing the placement.
Hedging Longevity Risk 5
in mutual funds that buy them. We study the potential benefits from these investments, and draw implications for
security design.4, 5
We first show that, if households know the exact parameters of the stochastic process for mortality rates,
then the benefits of being able to trade longevity bonds are quite small. If the risk premium on these bonds is
zero, although they will invest a significant fraction of their wealth in this asset, the welfare gain from doing so
is relatively small. In fact, a negative risk premium of 10 basis points is enough to deter investors from holding
these bonds almost completely, depending on the exact calibration. Unfortunately, since longevity bonds are a
fairly recent financial development, it is hard to compare this number with empirical estimates. Using the
CCAPM and mortality rates forecasts (also based on the Lee-Carter model) Friedberg and Webb (2005) obtain
an estimate of -0.02% with a risk aversion coefficient of 10.
However, small misperceptions about future improvements in mortality rates are enough to induce very
large welfare gains from investing in longevity bonds. More precisely, if households guide their decisions using
official period life tables, which do not take into account future improvements in longevity, the true benefit of
investing in this asset can be as large as 1% of annual consumption. Naturally, from the household's perspective,
her ex-ante estimate of this gain is very small, since she believes period life tables allow for future mortality
improvements. This difference is extremely important because it suggests that investors might not buy these
bonds even if they actually stand to gain significantly from investing in them.
The paper is organized as follows. In section 2 we use long term data for a cross section of countries to
document the existing empirical evidence on longevity. In sections 3 and 4 we setup and parameterize a life
cycle model of the optimal consumption and saving choices of an individual who faces longevity risk. The
results of the model are discussed in section 5. In section 6 we expand the set of assets that the agent has
available to include longevity bonds. The final section concludes and discusses extensions for future research.
2. A Model of Longevity Risk
The model in this section is described in more detail in a companion paper (Cocco and Gomes (2008)). In
this paper we use the same basic set-up but introduce longevity bonds.
4. There is a growing literature that studies the optimal pricing of longevity bonds and related instruments (see, for example, Dahl (2004) and
Carins, Blake and Dowd (2006)). Here we instead take bond prices as given and discuss their potential role in household portfolios.
5. Menoncin (2007) also introduces longevity bonds in an optimal savings and portfolio choice problem. However, in his model there is no
labor income or retirement.
2.1. Survival Probabilities
We solve a life-cycle model of consumption and savings, similar to Carroll (1997) and Gourinchas and
Parker (2001), but in which survival probabilities are stochastic. We let t denote age, and assume that the
individual lives for a maximum of T periods. Obviously, T can be made sufficiently large, to allow for
increases in life expectancy in very old age. We use the Lee-Carter (1992) model to describe survival
probabilities. This is the leading statistical mortality model in the demographic literature, and it has been shown
to fit the data relatively well. In addition, it has the advantage of being a relatively simple model. Mortality rates
are given by:
lnm t,x a t b t k x
(1)
where m t,x is the death rate for age t in period x . The a t coefficients describe the average shape of the
lnm t,x surface over time. The b t coefficients tell us which rates decline rapidly and which rates decline
slowly in response to changes in the index k x . The b t are normalized to sum to one, so that they are a relative
measure. The index k x describes the general changes in mortality over time. If k x falls then mortality rates
decline, and if k x rises then mortality worsens. When k x is linear in time, mortality at each age changes at its
own constant exponential rate.
Lee and Carter (1992) show that a random walk with drift describes the evolution of k x over time well.
That is:
k x 6 k k x"1 /kx
(2)
where 6 k is the drift parameter and / kx is normally distributed with mean zero and standard deviation @ k . This
model can be used to make stochastic mortality projections. The drift parameter 6 k captures the average annual
change in k , and drives the forecasts of long-run change in mortality. A negative drift parameter indicates an
improvement in mortality over time.
2.2. Preferences
Let p t denote the probability that the individual is alive at age t 1 , conditional on being alive at age t , so
that p t 1 " m t . For a given individual age and time are perfectly co-linear, so that in order to simplify the
exposition from now on we include only age indices. We assume that the individual's preferences are described
by the time-separable power utility function:
T t"2
C 1"2
E1 ! - t"1
pj p t"1 t
,
1"2
t1 j0
(3)
where - is the discount factor, C t is the level of age/date t consumption, and 2 is the coefficient of relative
risk aversion.
2.3. Labor Income
During working life age- t labor income, Yt , is exogenously given by:
logYt ft, Z t v t /t for t t K ,

(4)
where ft, Z t is a deterministic function of age and of a vector of other individual characteristics, Z t , / t is an
idiosyncratic temporary shock distributed as N0, @ 2/ , and v t is a permanent income shock, with
v t v t"1 u t , where u t is distributed as N0, @ 2u and is uncorrelated with /t . Thus before retirement, log
income is the sum of a deterministic component that can be calibrated to capture the hump shape of earnings
over the life cycle, and two random components, one transitory and one persistent.
The individual retires at age t R , and after this age income is modeled as a constant fraction 5 of permanent
labor income in the last working-year:
logYt log5 fK, Z K v K for t K ,

(5)
The parameter λ measures the extent to which the individual has defined benefit pensions, which implicitly
provide insurance against longevity risk. In the current version of our model we assume away labor supply
flexibility, but we plan to consider this possibility later. Retiring later in life may be an additional natural
mechanism to insure against increases in life expectancy.
2.4. Asset Markets
In the baseline version of the model households can invest in two alternative assets: a riskless asset with
(gross) interest rate R , and a longevity bond. In practice the payoffs on longevity bonds are correlated with
aggregate mortality rates but there is no perfect (or even linear) relationship between the two.6 Here we consider
a best case scenario where the bond returns are perfectly negatively correlated with the mortality rate
innovations, thus providing the investor with a perfect hedge. More precisely, the return on the longevity bond
L
( Rt ) also follows a two-state process, and it is given by
L
RLt 6 L @ k /kt
@
where 6 L and @ L correspond to the mean and standard deviation, respectively.
2.5. The Optimization Problem
In each period the timing of the events is as follows. The individual starts the period with wealth W t . Then
labor income and the shock to survival probabilities are realized. Following Deaton (1991) we denote cash-on-
hand in period t by X t W t Yt . We will also refer to X t as wealth: it is understood that this includes labor
income earned in period t . Then the individual must decide how much to consume, C t . The wealth in the next
period is then given by the budget constraint:
Wt1 ¡) t RLt 1 " ) t R¢W t Yt " C t

(6)
where ) denotes the share of wealth invested in longevity bonds. The problem the investor faces is to maximize
utility subject to the constraints. The control variable is consumption/savings. The state variables are age, cash-
on-hand, and the current survival probabilities. In our setup the value function is homogeneous with respect to
permanent labor income, which therefore is not a state variable.
6. These features are discussed in more detail in the calibration section.

The model was solved using backward induction. In the last period the policy functions are trivial (the agent
consumes all available wealth) and the value function corresponds to the indirect utility function. We can use
this value function to compute the policy rules for the previous period and given these, obtain the corresponding
value function. This procedure is then iterated backwards.
To avoid numerical convergence problems and in particular the danger of choosing local optima we
optimized over the space of the decision variables using standard grid search. The sets of admissible values for
the decision variables were discretized using equally spaced grids. The state-space was also discretized and,
following Tauchen and Hussey (1991), approximated the density function for labor income shocks using
Gaussian quadrature methods, to perform the necessary numerical integration.
In order to evaluate the value function corresponding to values of cash-on-hand that do not lie in the chosen
grid we used a cubic spline interpolation in the log of the state variable. This interpolation has the advantage of
being continuously differentiable and having a non-zero third derivative, thus preserving the prudence feature of
the utility function. The support for labor income realizations is bounded away from zero due to the quadrature
approximation. Given this and the non-negativity constraint on savings, the lower bound on the grid for cash-on-
hand is also strictly positive and hence the value function at each grid point is also bounded below. This fact
makes the spline interpolation work well given a sufficiently fine discretization of the state-space
3. Calibration
3.1. Time and preference parameters
The initial age in our model is 30, and the individual lives up to a maximum of 110 years of age. That is T
is equal to 110. Retirement age, K , is set equal to 65, which is the typical retirement age. We assume a discount
factor, - , equal to 0.98, and a coefficient of relative risk aversion, 2 , equal to three.
3.2. Survival probabilities
Undoubtedly, the calibration of the parameters for the mortality process are likely to be the most
controversial. This is in itself a sign that there is a great deal of uncertainty with respect to what one can
reasonably expect about future increases in life expectancy. In order to parameterize the stochastic process for
survival probabilities we do two things. First, we estimate the parameters of the Lee-Carter model using
historical data. Second, we try to determine which are the parameters of such model that match the projected
increases in life expectancy made by the UK government actuaries department. The latter projections are
forward looking measures that reflect historical data, other information, and expectations of future improvements
in mortality.
For the estimation of the Lee-Carter model using historical data, we use US data from 1959 to 2002 which is
the data period available in the Human Mortality Database, and estimate:
lnm t,x a t b t k x /t,x

(7)
where / t,x is an error term with mean zero and variance @ 2/ , which reflects particular age-specific historical
influences not captured by the model. This model is undetermined: k x is determined only to a linear
transformation, b t is determined only up to a multiplicative constant, and a t is determined only up to an
additive constant. Following Lee and Carter (1992) we normalize the b t to sum to unity and the k x to sum to
zero, which implies that the a x are the simple averages over time of the lnm t,x . We estimate the model using
the singular value decomposition method, and then use the time series data of k x to estimate the parameters of
the random walk. The estimated drift parameter 6 k is -0.7095 and the standard deviation of the shocks @ k is
1.299 (both of these parameters are reported in Table 1. As in Cocco and Gomes (2007), we also carry out a
second calibration, where we adjust the estimated parameter values to match the forward looking forecasts from
the UK's Government Actuaries Department.
3.3. Labor Income
To calibrate the labor income process, we use the parameters estimated by Cocco, Gomes and Maenhout
(2005) for individuals with a high school degree, which are also reported in Table 1. Deaton and Paxson (2001)
have investigated the correlation between aggregate labor income and mortality improvements using UK and US
data and concluded that the two aggregate series do not seem to be correlated. Therefore we assume zero
correlation between labor income shocks and shocks to longevity.
One important possibility that we also consider, is that the retirement replacement ratio is correlated with
improvements in life expectancy. This is motivated by recent events: the large improvements in life expectancy
that have occurred over the last decades have led governments to reduce the benefits of pay as you go state
pensions. Therefore we will consider a parameterization in which the replacement ratio is reduced when there is
an improvement in life expectancy, such that the present value of the retirement benefits that the individual
receives is unchanged.
Table 1: Parameters of the model
3.4. Asset Parameters
Returns on longevity bonds are linked to an aggregate mortality index. The bonds issued by Swiss Re. pay a
quarterly fixed coupon equal to 3-month US dollar LIBOR plus 135 basis points. The principal is then repaid in
full if the mortality index does not exceed a given threshold, but the payment decreases linearly with the index if
that threshold is reached. The EIB bond would have floating coupons that were directly linked to a mortality
index. Unfortunately, since these bonds are a fairly recent financial development, it is hard to estimate their risk
premia and volatility empirically. Using mortality rate forecasts, also based on the Lee-Carter model (Lee and
Carter (1992)), Friedberg and Webb (2005) compute the hypothetical returns for the EIB longevity bond, if the
issue had indeed occurred. They obtain a return volatility of 3%, which we use as the baseline value in our
L
calibration ( @ 3% ). They also find that the returns are negatively correlated with aggregate consumption
and, in a CCAPM framework with relative risk aversion of 10, the implied risk premium would be -0.02%.
L
Therefore, we set our baseline risk premium to zero ( 6 1. 5% ), but will also consider alternative values.
We assume that the real interest rate is equal to 1.5 percent.
3.4.1. The Portuguese Case
In our analysis we also want to measure welfare gains for the Portuguese case. For these experiments we
take the estimated values from our companion paper (Cocco and Gomes (2008)). Cocco and Gomes (2008) use
data from the Human Mortality Database to estimate the parameters of the stochastic process for mortality
shocks in the Lee-Carter model for Portugal. They restrict the sample to the post 1970 period, due to the lower
data quality for the 1940 to 1970 period. In addition they use panel data from the European Community
Household Panel to estimate labor income profiles for Portugal. They find that the labor income for a Portuguese
individual peaks later in life than for a US individual, and it is also characterized by a relatively higher
replacement ratio. This has important implications, since defined benefit state pensions implicitly provide
insurance against longevity risk.
3.5. A 3-period model with longevity bonds
We start by presenting a 3-period model which is better suited to understand the intuition behind our results,
and the determinants of the demand for longevity bonds.
3.5.1. Model set-up
The agent's objective is
C 1"2 C 1"2 C 1"2

MaxE¡U¢ E 1
p1 - 2 p1 p2 -2 3
1"2 1"2 1"2
where p 1 and p 2 denote the conditional survival probabilities, as before.
In period 1 she consumes ( C 1 ) and allocates her savings between riskless bonds (as before) and longevity
bonds:
W 2 W1 " C 1 ¡)1 RL2 1 " ) 1 R ¢ Y2

where ) denotes the fraction of wealth invested in longevity bonds. Her decisions are a function of the current
level of wealth ( W 1 ), expected future labor income ( Y2 and Y3 ), and expected future survival probabilities
( p 1 and p 2 ). Although she already knows p 1 , she does not yet know p 2 for sure. Longevity bonds allow her
to insure against this uncertainty. In period 2 she again chooses her optimal consumption ( C 2 ), but now based
on the exact realization of the survival probability p 2 , while in period 3 she simply consumes all available
wealth ( C 3 W 3 ). We are interested in understand the behavior of ) for different assumptions about labor
income, which will mimic the different stages of the life cycle.
3.5.2. Demand for longevity bonds
In Figure 1 we plot the optimal share invested in longevity bonds when labor income is riskless, Y2 Y3
( 10 ), and the bonds earn a zero risk premium. This figure captures some features of the retirement period in
our life-cycle model. Intuitively, the policy function is decreasing in financial wealth. With constant relative risk
aversion households always want to insure a constant fraction of their total wealth. As financial wealth increases,
relative to their future labor income, this is achieved by decreasing the fraction invested in longevity bonds. It is
important to note that the optimal fraction is not 100%. Longevity bonds are not riskless annuities. When agents
invest in longevity bonds they know these will yield a low return if p 2 happens to be lower than expected. From
the perspective of period 1, this implies that the agent will anticipate having less total resources in period 2 in
such scenario. Since she cannot off-set this by borrowing against her future labor income then she must trade-off
the increase in consumption risk in period 2, versus the decrease in conditional consumption risk in period 3.
Figure 1: Portfolio Rule in the 3-Period Model ("Baseline Case")

Note: This figure plots the optimal share invested in longevity bonds in the 3-period model, under
the baseline case (second period income riskless and equal to third period income).
This trade-off is clearly illustrated in Figure 2, where we also plot the policy function for case of
Y3 2 ' Y2 , for comparison. In this second case the investor is much less willing to transfer additional
resources to period 3, and thus the demand for longevity bonds decreases significantly, except for very high
values of wealth.7 Riskless bonds allow the agent to transfer the desired level of wealth from period 1 to period 2
without risk. Investing in longevity bonds would introduce risk. The compensation for this risk, which is
potential for higher savings for period 3 if p 2 happens to be higher than expected, is almost worthless since the
agent doesn't want to save for period 3 anyway ( Y3 is much higher than Y2 ). These policy functions will allow
us to explain the behavior of the portfolio allocation during retirement.
7. The share invested in longevity bonds is higher for very high values of wealth, because there is a lower level of savings (and thus a lower
ratio of financial wealth to future labor income).
Figure 2: Portfolio Rule in the 3-Period Model: Different Levels of Period 3-Income
Note: This figure plots the optimal share invested in longevity bonds in the 3-period model, for
different levels of income in the third period.
In our third experiment we try to capture the behavior during working life and thus consider a case with
labor income risk ( Y2 is either equal to 1 or 19, with equal probabilities). The results are shown in Figure 3. For
very low levels of cash-on-hand there is no demand for longevity bonds: labor income risk is much more
important. From the perspective of hedging this risk, riskless bonds clearly dominate. Only as wealth increases
does the household start to invest in longevity bonds.
Figure 3 - Portfolio Rule in the 3-Period Model: The Impact of Income Risk
Note: This figure plots the optimal share invested in longevity bonds in the 3-period model, for the
baseline case (second period income riskless and equal to third period income), and for an
alternative scenario where the second period labor income has the same mean but is now risky (it
can be either 90% above or 90% below its mean).
3.6. Life-Cycle Model
We now introduce the longevity bond in the life-cycle model discussed in section 2.
3.6.1. Baseline model
Figure 4 plots the share of wealth invested in the longevity bond over the life cycle. As expected from our
discussion of the 3-period model, early in life the demand for this asset is crowded-out by labor income risk.
Young households haven't accumulated much wealth and therefore they are significantly exposed to labor
income shocks. Naturally the riskless asset is better suited to hedge this risk, since it has no volatility. However,
as households accumulate more wealth and start saving for retirement, the allocation to longevity bonds quickly
converges to 100%. This exposure decreases again as they approach retirement since wealth accumulation is
increasing rapidly and, as previously discussed, the portfolio rule is decreasing in wealth (see figure 3).
Figure 4: Portfolio Allocation in the Baseline Case

Note: This figure plots the average portfolio share invested in longevity bonds over the life-cycle
in our baseline model. This series represents the average across 10000 simulations of the model.
During retirement the shape of the allocation to longevity bonds first increases, then decreases and finally
increases again. The intuition for the first two effects is explained by the second policy function in figure 2 (with
Y3 2Y2 . Initially, households have a significant amount of accumulated wealth and therefore they are
(mostly) on the decreasing part of the policy function. As they get older, financial wealth decreases at a faster
rate than future labor income, and therefore the portfolio allocation to longevity bonds is increasing in age.
Eventually, however, as wealth is significantly reduced and since the discount rate for future consumption is
very high, households effectively become liquidity constrained. They would like to be able borrow against their
future retirement transfers but are unable to do so. Therefore, any significant transfer of resources for future
years is suboptimal: they are now in the increasing part of the policy function. As a result the optimal portfolio
allocation is now decreasing in age: as wealth keeps decreasing they become more and more constrained over
time. To explain the final increasing pattern, during the last years of retirement, before financial wealth is
completely depleted, we now need to consider both policy functions in figure 2. At this stage of the life cycle,
the ratio of the agent's illiquid future labor income versus her next-period's income is decreasing fast which, in
the language of the 3-period model, is equivalent to Y3 becoming closer to Y2 . Therefore, the policy function
quickly converges to the one with Y3 Y2 , and the increasing pattern for low levels of wealth almost
disappears.
Figure 5: Portfolio Allocation for Different Values of the Expected Return

Note: This figure plots the average portfolio share invested in longevity bonds for different values
of the expected return on longevity bonds (mu). These series represent averages across 10000
simulations of the model.
3.6.2. Alternative return calibrations
As previously discussed, the empirical evidence on the returns of longevity bonds is very limited. Therefore,
in this section, we consider alternative values for both its expected return and volatility. Figure 5 plots the
L
optimal portfolio share invested in longevity bonds for three different calibrations of its expected return ( 6 ):
our baseline value, 1.5% and two cases with a small negative risk premium, 1.475% and 1.45%. The choice of
these exact alternative numbers is motivated by the results. Naturally, as we decrease their expected return, the
demand for longevity bonds falls. Interestingly, the results in Figure 5 show that a 2.5 basis points reduction is
enough to decrease the demand by more than half, and a 5 basis points negative risk premium is enough to deter
almost all investors from buying those bonds.
Figure 6: Portfolio Allocation for Different Values of Return Volatility

Note: This figure plots the average portfolio share invested in longevity bonds over the life-cycle
for two different values of the standard deviation of the returns on the longevity bonds (sigma),
and a third series which plots the portfolio share for the sigma=3% case multiplied by 0.3 (which
constitutes the frictionless equivalent to the sigma=10% case). These series represent averages
across 10000 simulations of the model.
L
Next we increase the volatility of longevity bond's return ( @ ) to 10%. The results are shown in Figure 6.
Since the returns are perfectly correlated with the mortality shocks, a higher volatility effectively increases the
investor's hedging position for a given portfolio allocation. Therefore, when we increase @ L , the share invested
in longevity bonds is naturally lower at every age. In a frictionless world this would be a simple re-scaling effect,
and nothing else should change. Therefore, for comparison purposes, we also plot in Figure 6 the scaled down
baseline portfolio allocation. Comparing this with the actual allocation for the @ L 10% case, we find that
there is something else going on in our model. Since the investor is facing short-selling constraints this result no
longer holds. The higher volatility case allows her to achieve levels of hedging that were previously unfeasible.
Therefore, we do not have a simple proportional shift. When the short-selling constraints were binding in the
baseline case, we observe a much smaller difference between the two optimal allocations. This shows that, under
the previous calibration, the investor would have liked to be able to invest more than 100% in the longevity
bonds at those ages. We could have increased @ L even further, but this would not have made any meaningful
difference, since the short-selling constraints are no longer binding for almost any agent in the simulations.
Interestingly, early in life we have the opposite result, as the portfolio allocation for the @ L 10% case is
actually slightly lower than the simple scaled-down version of @ L 3% case. This, is due to the borrowing
constraints and labor income risk. As previously discussed, in this stage of the life cycle, households are less
willing to buy longevity bonds because they constitute a poor hedge against labor income risk. As we increase
their return volatility this result becomes stronger. We believe that these are important conclusions to keep in
mind for the optimal design of longevity bonds. In general, if the correlation with aggregate mortality risk is kept
constant, then it is better to develop bonds with significant return volatility. However, excessive levels of
volatility might deter young households from buying this asset.
3.6.3. Welfare gains
Model with a fixed replacement ratio. In this section we compute the welfare gains from having access to
the longevity bond. Table 2 reports the age-30 utility gains, measured in certainty equivalent consumption units.
Table 2: Welfare gains from investing in the longevity bond (percent).

Note: The gains are measured as a percentage increase in annualised certainty equivalent (CE)
consumption levels.
L
For our baseline case ( 6 1. 5% and @ L 3% ) we find very low welfare gains: 0.032% of annual
consumption. On one hand, this could be seen as a striking result, based on the evidence in figure 4, which
shows that the investor will optimally allocate a large fraction of her wealth to this asset. However, on the other
hand, this is consistent with the evidence in figure 5, which shows that a small change in the bond's expected
return is enough to drive the demand close to zero. As also expected from the results in figure 6, when we
compute the welfare gains with a negative risk premium of -0.025%, they are indeed almost identical to zero
(0.005%). Households only invest a very small fraction of their wealth in longevity bonds and therefore, from
their perspective, this asset is not worth very much. On the other hand, if we increase the volatility of longevity
bond's returns, the household is now better off: the welfare gain increases to 0.038%. As previously explained,
due to the existence of short-selling constraints, a higher return volatility allows the investor to increase its
hedging position for a given portfolio allocation. In the previous section we have also found a negative effect for
young households, those concerned about labor income risk, but these results show that, for these levels of
volatility, the benefits are more important.
Model with a stochastic retirement replacement ratio. In this section we consider the version of the
model with a stochastic retirement replacement ratio. More precisely, we now consider the likely impact of
demographic changes (via changes in mortality rates) on households' retirement income, as described in sections
4.3 and 5.3. Table 3reports the corresponding welfare gains. In this case, the benefits from having access to
longevity bonds are almost an order of magnitude higher than before. For the zero risk premium case we obtain
increases of 0.15% to 0.24% of annual consumption, depending on the level of volatility (respectively 3% or
10%). In fact, in the previous case a negative risk premium of 5 basis points was enough to drive the demand for
longevity bonds down to essentially zero, while now this is no longer the case, and there is still a welfare gain of
0.054%. Naturally, if we decrease the risk premium further, eventually there is no demand for longevity bonds
and no welfare gain. In this case a negative risk premium of -15 basis points will deliver that result.
Table 3: Welfare gains from investing in the longevity bond with stochastic replacement ratio (percent)
consumption levels.
Model with mis-perceptions. We now consider the set-up of section 5.2. That is: we investigate the welfare
gains for an agent who uses official period life tables to make his consumption and saving decisions, without
allowing for future mortality improvements. More precisely, we consider an agent who in each period looks up
mortality rates in official period life tables, understands that they are stochastic, but does not acknowledging that
they are likely to improve in the future (i.e. assumes that the drift for k t is zero). As previously discussed, this is
a common mistake made by users of life tables.
Table 4 reports the welfare gains from having access to longevity bonds in this case. It is important to
mention that these welfare gains are computed using the objective probability measure, and thus reflect the real
welfare gains that the typical investor will realize. Naturally, from the investor's own ex-ante perspective, the
welfare gains are still the ones reported in the previous subsection (Tables 2 and 3), since she thinks that her
expectations about future survival probabilities are correct. The results in Table 4 show that the real welfare
gains are now very large and significantly higher than the previous ones. Even for the case with a fixed
replacement ratio the welfare gain is now 1.20% of annual consumption, and if we take into account for the
likely effects of the demographic changes on retirement income this number increases to 2.04%. It is important
to note that these are welfare gains as of age 30.
Table 4: Welfare gains from investing in the longevity bond with mis-perceptions (percent).
consumption levels. The agent assumes that the expected drift in mortality rates is equal to zero.
Results are shown for the case both with and without a stochastic replacement ratio (respectively
fixed and stochastic).
This combination of results, very large real welfare gains and very small perceived ex-ante welfare gains, is
extremely important. The results in the previous section show that household demand for longevity bonds is
likely to be very small unless these assets earn a zero (or positive) risk premium. Since the very limited existing
empirical evidence (Friedberg and Webb (2005)) suggests that they earn a negative risk premium, even though a
very small one, such demand is indeed likely to be close to zero. However, we have now shown that, unless
investors are actually able to make a good assessment of the expected future path of survival probabilities, the
real gains from investing in those bonds are actually extremely large.
The Portuguese Case In our finally set of experiments we calibrate the model using Portuguese data for
both mortality rates and the labor income process. Given our previous results we only consider the case in which
agents incorrectly assume that mortality rates are expected to remain constant in the future (i.e. use the current
official period life tables to make their future projections). Table 5 reports the welfare gains for both the case
with a constant replacement ratio, and the case with a stochastic replacement ratio. We find that the welfare
gains are larger than the ones obtained under our US calibration. This reflects the higher expected improvement
in mortality rates implicit in the estimation with Portuguese data: 6 K is -0.949 as opposed to -0.7095 in the US
estimation.
Table 5: Welfare gains from investing in the longevity bond with mis-perceptions (percent): The
Portuguese Case
consumption levels. The agent assumes that the expected drift in mortality rates is equal to zero.
Results are shown for the case both with and without a stochastic replacement ratio (respectively
fixed and stochastic).
4. Conclusion and Future Research
We study the potential role of longevity bonds in allowing households to hedge longevity risk in an
empirically parameterized life-cycle model of consumption and savings decisions. We find that the benefits from
investing in this asset are very small, if households know the exact parameters of the stochastic process for
mortality rates. In fact, a small negative risk premium (10 basis points, in line with recent estimates) is enough to
deter investors from holding these bonds almost completely. However, we also found that, in the presence of
small biases in expectations about future improvements in mortality, the welfare gains from investing in
longevity bonds are actually very large: potentially up to 1% of annual consumption. These results show that,
while very large might choose not buy these bonds if they are given the option to do so, they could actually stand
to gain significantly from investing part of their portfolios in this asset.
References
Blake, David and William Burrows, 2001, Survivor Bonds: Helping to Hedge Mortality Risk, The Journal of
Risk and Insurance, 68 (2), 339-348.
Cairns, Andrew, David Blake and Kevin Dowd, 2006, A Two-Factor Model for Stochastic Mortality with
Parameter Uncertainty: Theory and Calibration, The Journal of Risk and Insurance, 73 (4), 687-718.
Carroll, Christopher D., 1997, Buffer-Stock Saving and the Life-Cycle/Permanent Income Hypothesis,
Quarterly Journal of Economics 114, 433- 495.
Cocco, Joao and Francisco Gomes, 2008, Longevity Risk, Retirement Savings, and Individual Welfare,
Working Paper, London Business School.
Cocco, Joao, Francisco Gomes, and Pascal Maenhout, 2005, Portfolio Choice Over The Life-Cycle, Review
of Financial Studies 18: 491-533.
Dahl, Mikkel, 2004, Stochastic Mortality in Life Insurances: Market Reserves and Mortality-Linked
Insurance Contracts, Insurance: Mathematics and Economics, 35, 113-136.
Deaton, Angus S., 1991, Savings and Liquidity Constraints, Econometrica 59, 1221- 1248.
Deaton, Angus S. and Christina Paxson, 2001, Mortality, Income, and Income Inequality in Britain and The
United States NBER working paper 8534.
Friedberg, Leora and Anthony Webb, 2005, Life is Cheap: Using Mortality Bonds to Hedge Aggregate
Mortality Risk, Working Paper, Center for Retirement Research, Boston College.
Gourinchas, Pierre-Olivier, and Jonathan Parker, 2002, Consumption over the life cycle, Econometrica, 70,
47-89.
Government Actuary's Department, National Population Projections: Review of Methodology for Projecting
Mortality, National Statistics Quality Review Series Report No. 8.
Human Mortality Database. University of California, Berkeley (USA), and Max Planck Institute for
Demographic Research (Germany). Available at www.mortality.org or www.humanmortality.de (data
downloaded on 10/17/2006).
Lee, R.D. and L.R. Carter, 1992, Modelling and Forecasting U.S. Mortality, Journal of the American
Statistical Association, 87, 419, p. 659-671.
Menoncin, Francesco, 2006, The Role of Longevity Bonds in Optimal Portfolios, working paper, University
of Brescia.
Tauchen, G. and R. Hussey, 1991, Quadrature-Based Methods for Obtaining Approximate Solutions to
Nonlinear Asset Pricing Models, Econometrica, 59, 371-396.
Zeldes, Stephen, 1989, Optimal Consumption with Stochastic Income: Deviations from Certainty
Equivalence, Quarterly Journal of Economics 104, 275-298.
THE IMPACT OF FIRM SIZE AND MARKET SIZE ASYMMETRIES ON NATIONAL MERGERS IN
A THREE-COUNTRY MODEL *
Luís Santos-Pinto†
Faculdade de Economia - Universidade Nova de Lisboa
March 2008
Abstract
This paper studies incentives for national mergers in a three-country partial equilibrium model
where firms and markets may have different sizes. The paper finds that if firm size asymmetries are
sufficiently high or firm size asymmetries are small and the third-country market is much larger than home,
then there is no conflict of interest between national firms and governments since all of them are in favor of
the creation of a national champion. However, if firm size asymmetries are small or moderate and the third-
country market is not much larger than home, then firms wish to merge but governments oppose mergers.
Keywords: Mergers, International trade, Merger policy, Size asymmetry.
JEL Codes: F13, H77, L11, L41.
*
I am thankful to Eileen Fumagalli, Peter Neary, Helder Vasconcelos, and Pedro Pita Barros for helpful comments and suggestions. I am
also thankful to Tiago Pires and João Jalles for excellent research assistance.
†
Universidade Nova de Lisboa, Faculdade de Economia, Campus de Campolide, PT-1099-032, Lisboa, Portugal. Ph: +351-213801640. Fax:
+351-213870933. E-mail address: lspinto@fe.unl.pt.
28 Luís Santos-Pinto
1. Introduction
In many European countries there is a heated debate over whether governments and competition authorities
should favor or oppose the creation of national champions.1 An argument often put forth in favor of national
champions is that bigger firms will be in a better position to compete against foreign firms in world markets.2 A
merger of domestic firms can improve national welfare if it increases the market share of domestic firms in
world markets and/or raises prices for consumers in world markets.
This paper contributes to this debate by setting up a simple open economy model with firm size
asymmetries, market size asymmetries, and cost reductions due to mergers. The model is used to make
comparisons between endogenous mergers and mergers that improve national welfare.
The paper considers a partial equilibrium model where competitors in each of two countries, home and
foreign, serve their respective domestic markets, and all firms compete jointly in a third (world) market.3 The
main novelty, by comparison with the existing literature on three-country models of mergers, is that this model
allows for both firm size as well as market size asymmetries. Firm size asymmetries are modeled as different
marginal costs for firms in each country. Market size asymmetries are modeled as different demand curves in
home, foreign and the third-country. Home is assumed to be the small country in that, for any given price,
demand in foreign and in the third-country is larger than demand in home. As is standard in the literature firms
compete à la Cournot and markets are segmented.
The paper starts by providing conditions under which a merger of domestic firms is profitable conditional on
a given market structure in foreign. The paper shows that this happens if either (i) firm size asymmetries are
sufficiently large or (i) firm size asymmetries are small and the domestic market is not much smaller than the
third-country market. The intuition for this result is straightforward. A merger leads to less competition in the
domestic market and in the third-country market, therefore the market power of the firms involved in the merger
increases in these two markets. This allows firms to raise mark-ups since the less efficient firm that takes part in
the merger transfers production to the most efficient firm. This effect creates an incentive for mergers to take
place. However, in the third-country the merger implies that the market share of the merged firm is lower than
the sum of the pre-merger market shares of the firms involved in the merger. This effect reduces the incentive for
1. For example, the French government advocated a merger between the electricity and gas company SUEZ with the firm GAZ DE
FRANCE.
2. A recent example in Germany has been the aproval of the merger between the E.ON and RUHRGAS corporations where the German
Minister of Economics argued that size was very important at the onset of the energy market liberalization in Europe.
3. A partial equilibrium analysis assumes that the sector in question is very very small and therefore has little if any impact on other sectors
of the economy.
The Impact Of Firm Size And Market Size Asymmetries On National Mergers In A Three-Country Model 29
mergers to occur. Thus, in this model a merger always increases profits in domestic markets whereas it may or
may not increase profits in the third-country market.
The above conditions are used to characterize the set of Nash equilibria of the merger game played by
domestic and foreign firms. The paper shows that for most firm and market size parameter configurations the
merger game played by firms has a unique pure strategy Nash equilibrium where mergers take place in home and
in foreign. Only if firm size asymmetries are small and the size of the third-country market is relatively large will
the merger game played by firms have a unique pure strategy Nash equilibrium where no mergers take place in
both countries. For a small set of firm and market size configurations the merger game played by firms has
multiple equilibria.
Next the paper provides conditions under which a merger in one of the countries improves that country's
welfare for a given market structure in the other country. These conditions are used to characterize the set of
Nash equilibria of the merger game played by national governments. The paper finds that governments prefer not
to merge national firms when firm size asymmetries are (i) small or (ii) moderate and the size of the third-
country market is not much larger than the home market. However, governments prefer to merge national firms
when firm size asymmetries are (i) high or (ii) moderate-high and the size of the third-country market is much
larger than the home market. There is also a set of firm and market size configurations where the merger game
played by governments has multiple equilibria.
In a final step the paper explains when will firms and governments interests regarding the desirability of
national mergers be aligned or in conflict. The paper finds that if firm size asymmetries are sufficiently high,
then there is no conflict of interest between national firms and governments: all favor the creation of a national
champion. The interests of national firms and governments are also aligned if firm size asymmetries are small
and the third-country market is much larger than home: all oppose the creation of a national champion. However,
if firm size asymmetries are small or moderate and the third-country market is not much larger than home, then a
conflict of interest arises between national firms and governments: firms wish to merge but governments oppose
the merger.
The questions that this paper addresses have many links with the existing literature on merger and
competition policy, specially with papers which extend the analysis to the context of open economies. 4 This
literature has taken two different routes. A first set of papers focuses on nationally optimal merger policies and
merger profitability when trade policy instruments are available to national governments--e.g., Richardson
4. The traditional analysis of mergers and acquisitions in industrial organization--Salant et al. (1983) and Deneckere and Davidson (1985)--
usually neglects the effects of country borders.
(1999), Horn and Levinsohn (2001), and Huck and Conrad (2004). A second line of research is based on the
concept of `èxternal effects'' of a merger to outsiders. An important early contribution to this topic is Farrel and
Shapiro (1990). This concept was extended to open economies by Barros and Cabral (1994). This literature has
derived rather general conditions under which a merger benefits, or harms, the parties not participating in the
merger. It does not, however, explicitly consider that a merger may lead to cost reductions and so it can not
provide a complete characterization of post-merger equilibrium.
This paper also has links to the recent literature on merger waves. Qiu and Zhou (2007) attempt to model
endogenous mergers more completely, assuming an arbitrary number of heterogeneous firms which differs in
their marginal costs and Cournot competition. They endogenize the order of mergers and consider that firms
choose independently whether, when and whom they merge. Qiu and Zhou (2007) show that mergers are
efficient since firms surviving the mergers are the most efficient firms in the industry. They demonstrate that
some mergers are strategic and occur in waves.
Suedekum (2006) analyzes the profitability of mergers in a model with symmetric firms and trade between
all countries but with iceberg transport costs. Suedekum uses this framework to study the profitability and social
desirability of national versus international mergers. He finds that national mergers can have a negative impact in
world surplus, and so national competition policy can be seen as too permissive. However, the promotion of
national mergers can be in the interest of individual countries if rent extraction possibilities are strong enough.
He also shows that cross-border mergers are more attractive than domestic mergers.
Haufler and Nielsen (2005) is the paper that is most closely related to this one. Like here, they also consider
a three-country model where firms of two countries serve their respective domestic market and a third market.
By contrast to this paper, Haufler and Nielsen assume that firms are symmetric and that a merger creates
synergies which imply a reduction of marginal cost.
The remainder of the paper is organized as follows. Section 2 sets-up the model. Section 3 analyses
conditions under which a merger in one country, conditional on a given market structure in the other country, is
in the interest of the merging firms. Section 4 describes the set of equilibria of the simultaneous move merger
game played by firms. Section 5 analysis conditions under which a merger in one country, conditional on a given
market structure in the other country, increases national welfare. Section 6 describes the set of equilibria of the
simultaneous move game played by governments. Section 7 states conditions under which firms and
governments interests regarding the desirability of national mergers are aligned or in conflict. Section 8
discusses possible extensions of the model. Section 9 illustrates the model by applying it to the cement, mobile
telecommunications, and fuel industries of Portugal and Spain. Section 10 concludes the paper. The Appendix
contains the proofs of propositions.
2. Set-up
Consider three countries: home, foreign and a third-country (or the rest of the world). Before any merger
takes place there are 2 firms in home and 2 firms in foreign. Firms in each country sell in domestic markets and
there is no bilateral trade between home and foreign. However, home and foreign firms compete in the third-
country market.5
Home and foreign firms are fully owned by residents and produce a homogeneous good. The inverse
demand function in the home country is given by P h a " Qh , with a 1 while the inverse demand
function in foreign is P f a " +Qf , with + 0, 1¢. The inverse demand function in the third-country is
P 3 a " *Q3 , with * 0, 1¢. The previous assumptions imply that, for any given price, demand in the
home market is smaller than or equal to demand in foreign and in the third-country markets.6 Marginal costs are
assumed to be nonnegative and constant. We also assume that there are no fixed costs (this rules out gains from
economies of scale in mergers). Marginal costs of firms are given by c l1 c, c l2 c , where
l h, f and ¡0, a " c /3¢ . The parameter is an index of cost asymmetry. If 0 all firms
have the same cost. We assume that must be smaller than or equal to a " c /3 so that, in the absence of
mergers, even the less efficient firm makes nonnegative profits in all markets. It is useful to define
- /a " c and use it as a summary measure of asymmetry.

Following Barros (1998) we assume that when a merger between two firms occurs the less efficient firm
ceases production.7 Because marginal costs are constant, when two firms merge the merged entity will shut
down the high-cost unit and use only the low-cost unit for production. Let i j stand for the merger between
5. Transportation costs between home and the third-country and between foreign and the third country are assumed to be equal to zero.
Transportation cost for third-contry market can be greater than zero (but lower than a ) without changing qualitatively the results in the
paper.
6. Normalization of demand for home simplifies the problem and does not change qualitatively the results.
7. Barros (1998) proposes a framework with three asymmetric firms and where mergers are endogenously determined. He tries in this way to
explain the intuition behind the relationship between initial market concentration and size asymmetry of firms, showing that a negative
relation should be expected.
firms i and j . Then, the merged entity's marginal cost is equal to minc i , c j . Therefore, a merger can be
viewed as an acquisition of a high-cost firm by a low-cost firm.8
Firms compete in each market à la Cournot, that is, each firm chooses noncooperatively and simultaneously
the quantity that maximizes its individual profit.9 Each firm sees the markets it serves as segmented, that is, it is
responsible for the choice of how much to produce in each market and it considers not just the output of other
firms but their own choices about where to produce that output as unaffected by its actions. Thus, firms play
separate Cournot games in each market as they take as given the output of each rival in each market and not the
total output of each rival in all markets. This means that each market can be analyzed independently of the other
markets.
The starting point of the analysis is a situation where no merger has yet taken place. So, at the start, the
problem of home firm i is given by
max a " Qh " c hi q ih a " *Q' " c hi q 'hi ,

q hi ;q 'hi
where the quantities with an asterisk are sold in the third-country. The first-order conditions to this problem
are
a " 2q hi " ! q hj " c hi 0.

jpi
a " 2*q 'hi " * ! q 'hj ! q ',fk " c hi 0.

jpi k
Solving the first equation with respect to q hi we obtain the best reply of domestic firm i in the domestic
market
a " c hi 1
q hi " ! q hj .
2 2 jpi
Solving the second equation with respect to q 'hi we obtain the best reply of home firm i in the third-
country market
8. Barros (1998) approach also used by Qiu and Zhou (2007). However, Perry and Porter (1985) and Farrell and Shapiro (1990) use different
approaches to model the impact of a merger on an industry's cost structure.
9. The assumption of Cournot competition is in line with much of the literature on mergers. Theoretical and empirical arguments in defence
of the Cournot model are presented by Haufler and Nielsen (2005). The model proposed by Kreps and Sheinkman (1983) in which firms
choose in capacities in the first period and compete in prices in the second period generates Cournot outcomes.
a " c hi 1
q 'hi " ! q' ! q' .
2* 2 jpi hj k fk
Similarly, at the start, the problem of foreign firm i is given by
maxa " +Qf " c fi q fi a " *Q' " c fi q 'fi ,

q fi ,q 'fi
The first-order conditions to this problem are given by
a " 2+q fi " + ! q fj " c fi 0

jpi
a " 2*q 'fi " * ! q 'fj ! q 'hk " c fi 0

jpi k
Thus, the best reply of foreign firm i in the foreign market is

a " c fi 1
q fi " ! q fj ,
2+ 2 jpi
and the best reply of foreign firm i in the third-country market is
a " c fi 1
q 'fi " ! q' ! q' .
2* 2 jpi fj k hk
This set-up captures the idea that domestic markets are less competitive than the third-country market (or
rest of the world).
3. Profitability of Conditional Mergers
In this section we characterize the conditions under which a merger in one of the countries is profitable for a
given market structure in the other country. We start by proving conditions under which a domestic merger is
profitable conditional on a given market structure in foreign.
If home firms are not merged they sell q h1 a " c /3 and q h2 a " c " 2 /3 in the home
market. In this case, profits of home firms in the home market are given by = hh1 a " c 2 /9 and
= hh2 a " c " 2 2 /9. If foreign firms are not merged they sell q f1 a " c /3+ and
q f2 a " c " 2 /3+ in the foreign market. Profits of foreign firms in the foreign market are
f f
= f1 a " c 2 /9+ and = f2 a " c " 2 2 /9+. The market equilibrium in the third-country
market is given by:
q 'h1 a " c " 1 q' q 'f1 q 'f2

2* 2 h2
q 'h2 a " c " " 1 q 'h1 q 'f1 q 'f2
2* 2
q 'f1 a " c " 1 q' q 'h2 q 'f2
2* 2 h1
q 'f2 a " c " " 1 q 'h1 q 'h2 q 'f1
2* 2
Solving this system we obtain q 'h1 q 'f1 a " c 2 /5* and q 'h2 q 'f2 a " c " 3 /5* .
The profits of home and foreign firms in the third-county market are given by
= 'h1 = 'f1 a " c 2 2 /25* and = 'h2 = 'f2 a " c " 3 2 /25* .
If home firms merge the home market becomes a monopoly and the equilibrium quantity is
q h 1 h 2 a " c /2. The monopoly profits are = hh1h2 a " c 2 /4. If home firms merge and foreign
firms are not merged, then the equilibrium in the third-country market is given by
q 'h1h2 a " c " 1 q' q'

2* 2 f1 f2
q 'f1 a " c " 1 q' q 'f2

2* 2 h1h2
q 'f2 a " c " " 1 q 'h1h2 q 'f1
2* 2
' ' '

Solving this system we obtain q h1 h2 q f1 a " c /4* and q f2 a " c " 3 /4*. The
profits of the merged home firm in the third-country market are = 'h1h2 a " c 2 /16*. A merger of
home firms is profitable when foreign firms are not merged if the total profits of the merged home firm are
greater than the sum of the profits of the home firms before the merger, that is,
a " c 2 a " c 2 a " c 2

u
4 16* 9
a " c " 2 2 a " c 2 2 a " c " 3 2
.
9 25* 25*
(1)
If home firms are not merged but foreign firms are the equilibrium in the third-country market is given by
q 'h1 a " c " 1 q' q'

2* 2 h2 f1f2
q 'h2 a " c " " 1 q 'h1 q 'f1f2

2* 2
q 'f1f2 a " c " 1 q ' q '
2* 2 h1 h2
The solution to this system is q 'f1f2 q 'h1 a " c /4* and q 'h2 a " c " 3 /4*. The
profits of h1 in the third-country market are = 'h1 a " c 2 /16* and the profits of h2 are
= 'h2 a " c " 3 2 /16*. If home firms merge and so do foreign firms we have a duopoly in the third-
country market. In this case the equilibrium quantities in the third-country market are
'
q 'h1h2 q f1f2 a " c /3* and profits of the merged home firm by = 'h1h2 a " c 2 /9*. Thus, a
merger of home firms is profitable when foreign firms are merged if
a " c 2 a " c 2 a " c 2

u
4 9* 9
a " c 2 a " c " 2 2 a " c " 3 2
.
16* 9 16*
(2)
We use conditions (1) and (2) to state our first result.

Proposition 1:
7
(i) A merger of domestic firms is profitable when foreign firms are not merged if either (a) - ¡0, 61
¢
f1,f2 63 7"82-183- 2
, 1¢ , with * h
f1,f2 7 1
and * ¡* h 100 756-"140- 2 or (b) - ¡ 61 , 3 ¢.
1
(ii) A merger of domestic firms is profitable when foreign firms are merged if either (a) - ¡0, 15
¢ and
f1f2 f1f2 50 1"18-45- 2
* *h , 1 , with * h 100 18-"20- 2 or (b)
1 1
- ¡ 15 , 3¢ .
Proposition 1 part (i) tells us that when firm size asymmetries are sufficiently high a merger of home firms is
profitable when foreign firms are not merged. However, when firm size asymmetries are small, a merger of
home firms when foreign firms are not merged is only profitable if the third-country market is not too large by
comparison to home. The intuition for this result is as follows. The merger always raises profits in the home
market since it implies moving from a duopoly to a monopoly. However, the impact of the merger on profits in
the third-country market may be favorable or unfavorable to home firms depending on the size of firm size
asymmetries.10 If firm size asymmetries are large enough the merger raises profits in the third-country market.
This happens because large firm size asymmetries imply a large efficiency gain which raises mark-ups (the
difference between price and marginal cost) enough to make up for the loss of market share. By contrast, if firm
size asymmetries are small, a merger of home firms reduces profits in the third-country market. In this case, a
merger of home firms is only profitable if the gains in home are bigger than the losses in the third-country
market. For this to happen the third-country market must be not be excessively large by comparison to home.
Proposition 1 part (ii) says that a merger of home firms is profitable when foreign firms are merged if either
firm size asymmetries are high or firm size asymmetries are low but the third-country market is not much larger
than the home market. Comparing part (i) to (ii) of Proposition 1 we see that a domestic merger is profitable
under less restrictive conditions when foreign firms have merged than when they have not merged.11 This
happens because the gain from a domestic merger in the third-country market is larger when foreign firms are
merged than when foreign firms are not merged. When foreign firms are not merged and domestic firms merge
there is a move from four to three firms in the third-country market. By contrast, when foreign firms are merged
and domestic firms merge there is a move from three to two firms in the third-country market. A move from
10. The merger leads to gains in the home market for any δ ∈ [ 0,1/ 3] . The merger only leads to gains in the third country market if
δ > 7 / 61.
11. The set [ 151 , 31 ] contains the set [ 617 , 13 ] and β hf 1+ f 2 ≤ β hf 1, f 2 in the relevant range of parameters.
three to two firms has associated a larger increase in mark-ups and a smaller loss of market share for domestic
firms than a move from four to three firms.
We now provide conditions under which a foreign merger is profitable conditional on a given market
structure in home. A merger of foreign is profitable when home firms are not merged if
a " c 2 a " c 2 a " c 2

u
4+ 16* 9+
a " c 2 2 a " c " 2 2 a " c " 3 2
.
25* 9+ 25*
(3)
A merger of foreign firms is profitable when home firms are merged if
a " c 2 a " c 2 a " c 2

u
4+ 9* 9+
a " c 2 a " c " 2 2 a " c " 3 2
.
16* 9+ 16*
(4)
We use conditions (3) and (4) to state the following result.
Proposition 2:
7
(i) A merger of foreign firms is profitable when domestic firms are not merged if either (a) - ¡0, 61
¢
h1,h2 63+ 7"82-183-2
and * ¡* h1,h2
f , 1¢ , with * f 100 756-"140-2 or (b)
7 1
- ¡ 61 , 3 ¢.
1
(ii) A merger of foreign firms is profitable when domestic firms are merged if either (a) - ¡0, 15
¢ and
50+ 1"18-45- 2
* * h1h2
f , 1 , with * h1h2
f 100 18-"20- 2 or (b)
1 1
- ¡ 15 , 3 ¢.
Proposition 2 provides conditions under which a foreign merger is profitable when domestic firms are not
merged-part (i)-and when domestic firms are merged-part (ii). The intuition is the same as the one behind
Proposition 1. The only difference here is that if firm size asymmetries are small and the foreign market is larger
than home, then the conditions for a foreign merger to be profitable are less restrictive. Recall that for small firm
size asymmetries a merger of domestic firms implies a gain in the domestic market and a loss in the third-
country market. If the market in foreign is larger than the home market, then, for any given size of the third-
country market, the gain in the foreign market of a merger of foreign firms is greater than the gain in the home
market of a merger of home firms. In short, this result shows that when two countries compete in a third-country
market, mergers are more attractive for firms in the country with the largest domestic market.
4. Merger Game Played by Firms
We will now characterize the equilibrium decisions of home and foreign firms assuming that governments
do not interfere in markets. Later on we will analyze the opposite scenario, where governments decide whether
firms merge or not and firms play no role in merger decisions.
At the start we assume that no merger has taken place in either country. Home firms decide jointly whether
they wish to merge or not. Foreign firms also decide jointly if they wish to merge or not. The joint decisions of
home and foreign firms are taken simultaneously. This means we have a simultaneous move game where we can
use the Nash equilibrium concept to make predictions about behavior.
Let the generic merger game played by firms be denoted by F n,+ where n is the number of firms in each
country. The relevant payoffs of F 2,+ are summarized in Table I for the two strategies of merge, M, and not
merge N . The upper left part of each cell in Table I displays the profits of the merged home firm (when home
firms choose to merge) or the sum of profits of the two home firms (when home firms choose not to merge) for
each of the two possible market configurations in foreign. The lower right part of each cell displays the profits of
the merged foreign firm (when foreign firms choose to merge) or the sum of profits of the two foreign firms
(when foreign firms choose not to merge) for each of the two possible market configurations in home.
h\f M N
2 2
a"c a"c
4 4
2
a"c 2
9*
, a"c
16*

,
M
a"c 2 a"c 2 a"c"2 2
4+ 9+
a"c 2 a"c 2 a"c"3 2
9*
16*
a"c 2 a"c"2 2 a"c 2 a"c"2 2

9 9
2 a"c"3 2 a"c2 2 a"c"3 2
a"c 16*
, 25*
,
N
a"c 2 a"c 2 a"c"2 2
4+ 9+
a"c 2 a"c2 2 a"c"3 2
16*
25*
Table I
The next result characterizes the set of Nash equilibria of F 2,+ when home and foreign markets have the
f1f2
same size, that is, + 1 . In this case the game is symmetric with *h * h1h2
f *M
f1,f2
*h * h1,h2
f *N .
Proposition 3:
1
(i) If - ¡0, 15
¢ and * 0, * M ¢, then NE F 2,1 N, N .
1
(ii) If - ¡0, 15
¢ and * ¡* M , * N ¢, then NE F 2,1 £M, M , N, N , p, p ¤, with
2
63"100* "738800* -16472000* -
p .
13162-"603-2
1
(iii) If - ¡0, 15
¢ and * ¡* N , 1¢, then NE F 2,1 M, M .
1 7
(iv) If - ¡ 15 , 61
¢ and * 0, * N ¢, then NE F 2,1 £M, M , N, N , p, p ¤.
1 7
(v) If - ¡ 15 , 61
¢ and * ¡* N , 1¢, then NE F 2,1 M, M .
7 1
(vi) If - ¡ 61 , 3 ¢, then NE F 2,1 M, M .
1
This result tells us that if firm size asymmetries between firms are small, - ¡0, 15
¢ , then the set of
Nash equilibria of F 2,1 depends on the size of the third-country market. If the size of the third-country market
is much larger than the home market, * 0, * M ¢ , then F 2,1 has a unique pure strategy Nash equilibrium
(PSNE from now on) where home firms and foreign firms do not merge. For small firm size asymmetries and a
large third-country market, the profit gains in the home market are less than the losses in the third-country
market and so the merger is not profitable. However, if firm size asymmetries are small and the size of the third-
country market is moderately larger than the home market, * ¡* M , * N ¢, then F 2,1 has two PSNE and
one mixed strategy Nash equilibrium (MSNE from now on) where firms in each country merge with probability
p . If firm size asymmetries are small but the size of the third-country market is not much larger than the home
market, * ¡* N , 1¢ , then F 2,1 has a unique PSNE where mergers take place in home and in foreign.
1 7
Proposition 3 also tells us that if firm size asymmetries between firms are moderate, - ¡ 15 , 61
¢, then
we may have two situations. If the size of the third-country market is sufficiently larger than the home market,
* 0, * N ¢ , then F 2,1 has two PSNE and one MSNE where firms in each country merge with probability
p . However, if the size of the third-country market is not sufficiently larger than the home market, then F 2,1
has a unique PSNE where mergers take place in home as well as in foreign. Finally, Proposition 3 tells us that if
7 1
firm size asymmetries are large, - ¡ 61 , 3 ¢ , then F 2,1 has a unique PSNE where mergers take place in
home as well as in foreign. This happens no matter the size of the third-country market. These findings are
summarized in Figure 1.
Figure 1
Figure 1 displays on the horizontal axis the range of values of the summary index of firm size asymmetries,
- /a " c , and on the vertical axis the range of values of the size of the third-country market, * . We
see that for most parameter configurations the merger game played by firms has a unique PSNE where mergers
take place in both countries--area ( M, M ). Only if firm size asymmetries are small and the size of the third-
country market is relatively large will the merger game played by firms have a unique PSNE where mergers do
not take place in both countries--area ( N, N ). For the remaining configurations of the parameter space--the
area that separates area ( M, M ) from area ( N, N )-the merger game played by firms has two Nash equilibria
in pure strategies and one in mixed strategies.
Next we characterize the set of Nash equilibria of the merger game played by firms when the home market is
smaller than the foreign market. To do that we start by introducing a lemma.
50 7"126-315-2
+' 63 7"82-183- 2
.
Lemma 1: Let
f1f2 f1,f2
(i) If + 0, +' ¢ then * f
h1h2
* h1,h2
f t *h *h .
f1f2 f1,f2
'
* h1h2
f *h * h1,h2
f *h .
(ii) (ii) If + , 1 then
Proposition 4 characterizes the set of Nash equilibria of F 2,+ when the home market is smaller than the
foreign market, that is, + 0, 1 .
Proposition 4:
(i) If - ¡0, 1
15
¢, + 0, +' ¢ and * 0, * h1,h2
f ¢, then NE F 2,+ N, N .
f1f2
(ii) If - ¡0, 1
15
¢, + 0, +' ¢ and * ¡* h1,h2
f , *h ¢ , then NE F 2,+ N, M .
1 f1 f2
(iii) If - ¡0, 15
¢, + 0, +' ¢ and * * h , 1 , then NE F 2,+ M, M .
1
(iv) If - ¡0, 15
¢, + +' , 1 and * 0, * f1f2
h ¢ , then NE F 2,+ N, N .
f1 f2
(v) If - ¡0, 1
15
¢, + +' , 1 and * ¡* h , * hf 1,h 2 ¢ , then NE
63+"100* "738+800* -1647+2000* -2
F 2,+ £M, M , N, N , p h , p ¤, with p h 13162-"603-2 +
.
(vi) If - ¡0, 1
15
¢, + +' , 1 and * ¡* h1,h2
f , 1¢ , then NE F 2,+ M, M .
(vii) If
1
- ¡ 15 , 7
61
¢ and * 0, * h1,h2
f ¢, NE F 2,+ £M, M , N, N , p h , p ¤.
(viii) If
1
- ¡ 15 , 7
61
¢ and * ¡* h1,h2
f , 1¢, then NE F 2,+ M, M .
7 1
(ix) If - ¡ 61 , 3 ¢ , then NE F 2,+ M, M .
Figure 2 summarizes the findings of Proposition 4.

Figure 2
Area ( M, M ) in figure 3 represents the parameter configurations for which the asymmetric merger game
played by firms has a unique PSNE where national firms in each country decide to merge. Area N, N
represents the configurations of parameters for which the game has a unique PSNE where national firms in each
country decide not to merge. Area N, M represents the configurations of parameters for which the game has
a unique PSNE where foreign firms decide to merge but home firms do not. For the remaining values of the
parameter space--the small area to the southeast of area ( N, M ) the game has two Nash equilibria in pure
strategies and one in mixed strategies.
Proposition 4 tells us that if the foreign market is not much larger than the home market, + +' , 1 ,
then the set of Nash equilibria of the asymmetric merger game played by firms is similar to that of the symmetric
merger game played by firms.
There are only two novelties novelty here. First, in the mixed strategy Nash equilibria home firms merge
with a smaller probability than foreign firms.12 Second, if the foreign market is substantially larger than the home
market, + 0, +' , then there will be a range of parameters for which we can find asymmetric pure strategy
Nash equilibria in which foreign firms merge but domestic firms do not merge. This happens when firm size
12. It follows from the definition of ph and p that ph < p in the relevant range of parameters for which the MSNE are defined.
1
asymmetries are small, - ¡0, 15
¢, and the size of the third-country market is moderately larger than the
f1f2
size of the home market, * ¡* h1,h2
f , *h ¢.
The intuition for this result comes solely from the fact that if the foreign market is larger than the home
market, then the foreign market profit gains that a merger of foreign firms induces are greater than the home
market profit gains induced by a merger of home firms.
5. Welfare Impact of Conditional Mergers
In this section we provide conditions under which a merger in one of the countries improves that country's
welfare for a given market structure in the other country. As it is usual in the partial equilibrium literature we
measure national welfare as the sum of consumer and producer surplus.
We start by proving conditions under which a domestic merger improves national welfare conditional on a
given market structure in foreign. Consumer surplus at home is given by CS h a " p h Qh /2 Q2h /2,
where Qh is total output produced by home firms. If home firms do not merge, then
Qh 2a " 2c " /3 and CS hh1,h2 2a " 2c " 2 /18. If home firms merge, then
Qh a " c /2 and CS hh1h2 a " c 2 /8 .

Thus, a domestic merger will improve national welfare when foreign firms are not merged if
a " c 2 a " c 2 a " c 2 2a " 2c " 2

u
8 4 16* 18
a " c 2 a " c 2 2 a " c " 2 2 a " c " 3 2
.
9 25* 9 25*
(5)
A domestic merger will improve national welfare when foreign firms are merged if
a " c 2 a " c 2 a " c 2 2a " 2c " 2

u
8 4 9* 18
a " c 2 a " c 2 a " c " 2 2 a " c " 3 2
.
9 16* 9 16*
(6)
From inequalities (5) and (6) we obtain the following result.
Proposition 5:
(i) A merger of domestic firms improves national welfare when foreign firms are not merged if either (a)
f1,f2 9 "782-"183- 2
7
- ¡ 61 , 5
¢ and * 0, b fh1,f2 , with b h 50 5"32-44-2
5 1
- ¡ 22 , 3 ¢.
22 or (b)
(ii) A merger of domestic firms improves national welfare when foreign firms are merged if either (a)
f1f2 "118-"45-2
1
- ¡ 15 , 5
¢ and * 0, b f1f2 , with b h 5 1
- ¡ 22 , 3¢ .
22 h 5"32-44- 2 or (b)
Part (i) of Proposition 5 says that a merger of domestic firms improves national welfare when foreign firms
are not merged if firm size asymmetries are sufficiently high. It also says that when firm size asymmetries are
moderate a merger of domestic firms only improves national welfare when foreign firms are not merged if the
third-country market is substantially larger than the home market. Finally, it says that for small firm size
asymmetries a domestic merger never raises national welfare when foreign firms are not merged.
Proposition 5 part (ii) states conditions under which a merger of domestic firms improves national welfare
when foreign firms are merged. The conditions are similar to those found in part (i) only less restrictive. This
happens because the gains in the third-country market that are obtained with a merger of domestic firms are
larger when foreign firms are merged than when foreign firms are not merged.
We will now provide conditions under which a foreign merger improves foreign welfare conditional on a
given market structure in home. Consumer surplus in foreign is given by CS f a " p f Qf /2 +Q2f /2,
where Qf is total output produced by foreign firms. If foreign firms do not merge, then
f
Qf 2a " 2c " /3+ and CS f1,f2 2a " 2c " 2 /18+. If foreign firms merge, then
f
Qf a " c /2+ and CS f1f2 a " c 2 /8+ .
So, a foreign merger will be welfare improving for foreign when home firms are not merged if
a " c 2 a " c 2 a " c 2 2a " 2c " 2

u
8+ 4+ 16* 18+
a " c 2 a " c 2 2 a " c " 2 2 a " c " 3 2
.
9+ 25* 9+ 25*
(7)
A foreign merger will improve foreign welfare when home firms are merged if
a " c 2 a " c 2 a " c 2 2a " 2c " 2

u
8+ 4+ 9* 18+
a " c 2 a " c 2 a " c " 2 2 a " c " 3 2
.
9+ 16* 9+ 16*
(8)
From (7) and (8) we derive the following result.
Proposition 6:
(i) A merger of foreign firms improves foreign welfare when domestic firms are not merged if either (a)
h1,h2 9+ "782-"183- 2
7
- ¡ 61 , 5
22
¢ and * 0, b h1,h2
f , with b f 50 5"32-44- 2 or (b)
5 1
- ¡ 22 , 3 ¢.
(ii) A merger of foreign firms improves foreign welfare when domestic firms are merged if either (a)
2
1
- ¡ 15 , 5
¢ and * 0, b h1h2
f
h1h2
, with b f + "118-"45-2 5 1
- ¡ 22 , 3¢ .
22 5"32-44- or (b)
Proposition 6 provides conditions under which a foreign merger improves foreign welfare when domestic
firms are not merged--part (i)--and when domestic firms are merged--part (ii). The intuition is similar to that of
Proposition 5. The only difference here is that a merger of foreign firms improves foreign welfare for moderate
firm size asymmetries only if the size of the third-country market is larger than the minimum size of the third-
country market necessary for a merger of home firms to raise welfare at home.
6. Merger Game Played by Governments
We now assume that firms play no active role in merger decisions. Instead national governments are
completely free to determine the market structure in each country.13
Like before we assume that the starting point of this game is a situation where no merger has taken place.
The home government decides whether home firms should merge or not. Similarly, the foreign government
decides whether foreign firms should merge or not. The decisions of home and foreign governments are taken
simultaneously. Given the choice in the other country, each government takes the decision that maximizes its
welfare: the sum of consumer and producer surplus.
13. This is an extreme scenario. The most realistic model would be one where firms in each country propose mergers to a national anti-trust
authority and the anti-trust authority allows or blocks mergers. In this case merger decisions are determined by firms' proposals and by the
decisions of anti-trust authorities.
Let the generic merger game played by governments be denoted by G n,+ where n is the number of firms
in each country. The relevant payoffs for G 2,+ are summarized in Table II for the two strategies of merge, M ,
and not merge, N.

h\f M N
2
a"c 2 a"c
8 8
2 a"c 2
a"c
4
4
2 a"c 2
a"c
9*
, 16*
,
M
a"c 2 2a"2c" 2
8+ 9+
a"c 2 a"c 2 a"c"2 2
4+
9+
a"c 2 a"c 2 a"c"3 2
9*
16*
2a"2c" 2 2a"2c" 2
9 9
a"c 2 a"c"2 2 2 a"c"2 2
a"c 9
9
2 a"c"3 2 a"c2 2 a"c"3 2
a"c 16*
, 25*
,
N
a"c 2 2a"2c" 2
8+ 9+
a"c 2 a"c 2 a"c"2 2
4+
9+
a"c 2 a"c2 2 a"c"3 2
16*
25*
Table II
The upper left part of each cell in Table II displays the sum of consumer surplus at home with profits of the
merged home firm (when home firms are merged) or with profits of the two home firms (when home firms are
not merged) for each market configuration in foreign. The lower right part of each cell displays the sum of
consumer surplus in foreign with profits of the merged foreign firm (when foreign firms are merged) or with
profits of the two foreign firms (when foreign firms are not merged) for each market configuration in home.
We now characterize the set of Nash equilibria of the merger game played by governments when +1.
f1,gf2 f1f2
Note that if + 1 we have b h b h1,h2
f bN bh b h1h2
f bM .
Proposition 7:
1
(i) If - ¡0, 15
¢ , then NE G 2,1 N, N .
1 7
(ii) If - ¡ 15 , 61
¢ and * 0, b M ¢ , then NE G 2,1 £M, M , N, N , q, q ¤, with
21350* "246800* -549800* - 2
q3 .
13162-"603- 2
1 7
(iii) If - ¡ 15 , 61
¢ and * ¡b M , 1¢ , then NE G 2,1 N, N .
7 5
(iv) If - ¡ 61 , 22
¢ and * 0, b N ¢ , then NE G 2,1 M, M .
7 5
(v) If - ¡ 61 , 22
¢ and * ¡b N , b M ¢ , then NE G 2,1 £M, M , N, N , q, q ¤.
7 5
(vi) If - ¡ 61 , 22
¢ and * ¡b M , 1¢ , then NE G 2,1 N, N .
5 1
(vii) If - ¡ 22 , 3 ¢ , then NE G 2,1 M, M .
Proposition 7 shows that in equilibrium of the merger game played by governments, governments decide not
to merge national firms when firm size asymmetries are small, when firm size asymmetries are moderate-low
and the size of the third-country market is not much larger than the home market, or when firm size asymmetries
are moderate-high and the size of the third-country market is not much larger than the home market. By contrast,
governments decide to merge national firms when firm size asymmetries are high or when firm size asymmetries
are moderate-high and the size of the third-country market is much larger than the home market. In the
remaining cases we have multiple equilibria: both governments decide to merge national firms, both
governments decide not to merge national firms, and each government merges national firms with probability
q . The findings are summarized in figure 3.

Figure 3
Area ( M, M ) in figure 3 represents the parameter configurations for which the symmetric merger game
played by governments has a unique PSNE where both governments decide to merge national firms. Area
N, N represents the configurations of parameters for which the game has a unique PSNE where both
governments decide not to merge national firms. For the remaining values of the parameter space--the area that
separates area ( M, M ) from area ( N, N ) the game has two Nash equilibria in pure strategies and one in mixed
strategies.
7. Firms versus Governments
We are now in a position were we can use the model to clarify when there will be a convergence or a
divergence of interests between national firms and governments regarding the desirability of national mergers.
The findings are summarized in Proposition 8.
Proposition 8:
(i) National firms and governments agree that a merger of national firms should not take place when
1
- ¡0, 15
¢ and * 0, * M ¢.
(ii) National firms prefer to merge but governments prefer that national firms do not merge when (a)
1 1 7
- ¡0, 15
¢ and * ¡* N , 1¢, (b) - ¡ 15 , 61
¢ and * ¡max- * N - , b M - , 1¢, or (c)
7 25 " 91
- ¡ 61 , 89
¢ and * ¡b M , 1¢.
(iii) National firms and governments agree that a merger of national firms should take place when
5 1
- ¡ 22 , 3 ¢.
This result provides conditions on cost and market size asymmetries under which the interests of national
firms and of national governments will be either alined or in conflict. It says that if firm size asymmetries are
small and the third-country market is much larger than the national market, then the interests of national firms
and governments are alined: all are opposed to the creation of a national champion. It says that the interests of
national firms and governments are also aligned when firm size asymmetries are high: all favor the creation of a
national champion. However, if firm size asymmetries are small or moderate and the third-country market is not
much larger than the national market, then there is a conflict of interest between national firms and governments:
firms prefer to merge but governments are opposed to the merger.14
8. Extensions
There are many possible directions in which one could extend this model. For example, one could relax the
assumption of no trade between home and foreign. In this case competition in domestic markets would resemble
competition in the third-country market and the only difference would be that incentives for mergers would be
weaker.
Another possible extension of the model would be to break the assumption that home and foreign firms have
the same size distribution by assuming, for example, that foreign firms are uniformly more or less efficient than
home firms by a parameter ) . This extension complicates the analysis since it is no longer possible to find
closed form solutions for the market size thresholds. However, it is possible to parameterize the model
numerically to analyze this situation.
14. Proposition 8 and can be easily understood from visual inspection of Figures 1 and 3.
The most obvious extension would be to look at a situation where we have three firms in home and three
firms in foreign. This extension also makes the analysis harder. If there are three firms in each country we would
need to consider all possible merger combinations. We would need to state not only individually rational
constraints for mergers to be viable but also stability conditions under which the firms outside the mergers would
not make a better offer to one of the participants in the merger.
Another possible modification of the model would be to model explicitly a game between national firms and
competition authorities where firms propose mergers and competition authorities accept or reject mergers
proposed by firms. This extension introduces a sequential aspect to merger analysis in open economies that has
not yet been sufficiently explored.
9. Application: Portugal and Spain
In this section we apply the model to three industries in Portugal and Spain: cement, mobile
telecommunications an fuel retail. There are several characteristic of these three sectors that fit quite well into
the framework of the model. Cement, telephone calls, and fuel are very close to being homogenous products.
The Portuguese and Spanish markets in these three sectors have high market concentration indexes. A large
percentage of profits of Portuguese and Spanish firms in these sectors comes from third-country markets (South
America and Africa).
The question we ask is whether the current industry configuration in each country and in each of these two
industries is stable or unstable. By a stable industry configuration we mean a situation where firms will not
propose mergers or where a merger will be proposed by firms but will be rejected by national governments. In
all other cases we say that the industry configuration is unstable.
It is important to bear in mind that the analysis that follows depends on the assumptions of the model. If
some of the assumptions of the model do not hold (e.g., linear demand, constant marginal costs, homogeneous
products, ect.), then the analysis may no longer be valid.
9.1. Cement
The data collected was obtained directly from different sources (annual report of Secil, annual report of
Cimpor, annual report of Uniland, report of cement industry in Arab countries, European Commission for Latin
America and Caribbean, Sector Report of Cement Industry by Caixa BI) and it refers to the year 2005/2006.
The market shares of Portuguese firms in the Portuguese cement market are: Cimpor 52.2% and Secil
47.8%. The market shares of Spanish firms in the Spanish cement market are: Cia Valenciana de Cementos
(CEMEX) 24.8%, Cementos Portland (Uniland - Group Portland Valderrivas) 21.7%, Arland (Lafarge) 9.3%,
Cementos Cosmos (Cimpor) 7.1%, Financera y Minera 6.4% and others 30.7%.
The similar market shares of the Portuguese cement firms in the domestic market suggest that firm size
asymmetry is quite small in Portugal. We can confirm this using the model. The summary index of asymmetry
for Portugal is the solution to
q Cimpor /QPT
0. 522 1 - ,
q Secil
/Q PT 0. 478 1 " 2-
or - PT 0. 028909 1/15. In Spain the two biggest cement firms also have very similar market
shares. The index of asymmetry for these two Spanish firms is
q Cemex /QSP
0. 248 1 - ,
q Portland
/Q SP 0. 217 1 " 2-
or - SP 0. 043478 1/15.
The two largest firms in Portugal and the two largest firms in Spain compete in third countries. Only 50% of
Cimpor's profits before taxes and depreciation come from sales in the Portuguese market.15 A large percentage of
profits before taxes and depreciation of Cemex and Portland comes from sales in South America and Africa (less
than 50% of total profits). The size of the Spanish cement market is approximately 4 times the size of the
Portuguese cement market and the size of third-country markets is much larger than the size of both domestic
markets put together. This means that + should be close to 0.25 and * must be smaller than 0.25. Take
- PT X - SP 0. 028909 0. 043478 /2 0. 0362. We have that
50 7"126-315-2
+' 63 7"82-183- 2
0. 529. + 0. 25 0. 529 +' . So, assuming
-0.0362 So, we have that
that * is less than 0.25, it follows from Proposition 4 that Cimpor and Secil will not wish to merge. However,
it is not clear whether Cemex and Portland wish to merge or to remain separated.
15. Approximately 25% of Cimpor's profits before taxes and depreciation come from sales in Brazil, 11% from sales in Spain and the
remaining 14% from sales in Africa.
9.2. Mobile Telecommunications
In Portugal there exist three operators: Vodafone, Optimus and TMN. In Spain the operators are Vodafone,
Movistar and Orange. Both TMN and Movistar have quite significant parts of their profits coming from third
country markets.16
The data collected were obtained directly from Annual Reports of the respective companies and it refers to
the year 2005. The market shares for the Portuguese market are: TMN 44%, Vodafone 35% and Optimus 21%.
For the Spanish case we have: Movistar 46.6%, Vodafone 29.3% and Orange 24.1%.
The summary index of firm size asymmetry for the two largest Portuguese firms is the solution to
q T MN /QPT
0. 44 1 - ,
q Vodafone
/Q PT 0. 35 1 " 2-
or - PT 0. 0732. We have that 1/15 - PT 0. 0732 7/61. The summary index of firm
size asymmetry for the two largest Spanish firms is the solution to
q Movistar /QSP
0. 466 1 - ,
q Vodafone
/Q SP 0. 293 1 " 2-
or - PT 0. 14122. We have that 7/61 - SP 0. 14122 5/22. It follows from Proposition 4

that Portuguese firms would like to merge as long as the third-country markets is not too large. It also follows
from Proposition 4 that Spanish firms would like to merge, no matter the size of the third-country market.
Should national governments promote these mergers? According to Proposition 8 Portuguese and Spanish
governments should not allow these mergers since the losses for domestic consumer surplus overtake the
efficiency gains or any profit gains in third-country markets.17
16. In Brasil the three major brands are Vivo, Claro and TIM. It is well known that TMN is owned by Portugal Telecom and Movistar by
Telefónica, each company has a participation of 50% in the shareholding that controls 63% of Vivo.
17. If the third-country market is very large we have a region of multiple equilibria. However, for this region to be attained the third-country
markets must be really large by comparison with domestic markets. This is clearly not the case in the mobile telecommunications industry in
Portugal and Spain.
9.3. Fuel Retailing
The market shares of the two largest Portuguese fuel retailers in the Portuguese market are: Galp 37% and
BP 17%. The market shares of the two largest Spanish fuel retailers in the Spanish market are: Repsol 41%
Cepsa 20%. The summary index of firm size asymmetry for the two largest Portuguese firms is the solution to
q Galp /QPT
0. 37 1 - ,
BP
q /Q PT 0. 17 1 " 2-
or - PT 0. 21978. We have that 7/61 - PT 0. 21978 5/22. The summary index of firm
size asymmetry for the two largest Spanish firms is the solution to
q Repsol /QSP
0. 41 1 - ,
q Cepsa
/Q SP 0. 20 1 " 2-
or - SP 0. 20588. We have that 7/61 - SP 0. 20588 5/22. We see that firm size
asymmetries of the two largest firms in Portugal and Spain are much larger in fuel retailing industry than in the
cement industry.
It follows from Proposition 4 that no matter the sizes of the Portuguese, Spanish and third-country markets,
Galp and BP would like to merge and so would Repsol and Cepsa. Should national governments promote these
mergers? According to Proposition 7 they should since the large firm size asymmetries are a signal of large
efficiency gains in such mergers. This is enough to compensate the increase in market power associated with
these mergers.
10. Conclusion
This paper studies incentives for national mergers in a partial equilibrium model where firms of two
countries compete in a third-country market. The main novelty of the paper is that it allows for both firm size
asymmetries as well as market size asymmetries. This allows us to characterize the joint impact of cost and
market size asymmetries on incentives for national firms to merge and on incentives for governments to promote
national mergers.
The paper shows that if firm size asymmetries between national firms are sufficiently high, then the interests
of national firms and governments are aligned in favor of creating a national champion. However, firms and
governments agree that no national champion should be created when firm size asymmetries are small and the
third-country market is much larger than the national market. If firm size asymmetries are small or moderate and
the third-country market is not much larger than the national market, then there is a conflict of interest between
national firms and governments: firms would like to merge but governments oppose mergers.
References
Barros, P. and L. Cabral (1994). “Merger Policy in Open Economies”, European Economic Review 38:
1041-1055.
Barros, P. (1998). “Endogenous mergers and size asymmetry of merger participants”, Economic Letters 60,
113-119.
Brander, J. and B. Spencer (1985). “Export subsidies and international market share rivalry”, Journal of
Internacional Economics 18, 83-100.
De Stefano, M. and M. Rysman (2004). “Competition policy as strategic trade with differentiated products”,
mimeo, Boston University.
Dixit, A.(1984), International Trade Policy for Oligopolistic Industries, Economic Journal 94, 1-16.
Farrell, J. and C. Shapiro (1990). “Horizontal Mergers: An Equilibrium Analysis”, The American Economic
Review 80, 107-126.
Fauli-Oller, R. (2000). “Takeover Waves”, Journal of Economics and Management Strategy 9, 189-210.
Forslid, R., Häckner, J. and A. Muren (2005). “When do Countries Introduce Competition Policy?”, CEPR
Discussion Paper No. 5162.
Gowrisankaran, G. (1999). “A dynamic model of endogenous horizontal mergers”, Rand Journal of
Economics 30, 56-83.
Haufler, A. and S.Nielsen (2005). “Merger policy to promote global players?” A Simple Model,'' CESifo
Working Paper No.1523.
Horn, H. and J. Levinsohn (2001). “Merger Policies and Trade Liberalization”, Economic Journal 111, 244-
276.
Horn, H. and L. Persson (2001). “Endogenous Mergers in Concentrated Markets”, International Journal of
Industrial Organization 19, 1213-1244.
Huck, S. and K.Konrad (2004), Merger profitability and trade policy, Scandinavian Journal of Economics
106, 107-122.
Huck, S., K. Konrad and W. Mueller (2005). “Merger without Cost Advantages”, CESifo Working Paper
No 1461.
Kreps, D. and J. Sheinkman (1983). “Capacity Precommitment and Price Competition Yield Cournot
Outcomes”, Bell Journal of Economics 14, 326-337.
Perry, M. and R. Porter (1985). “Oligopoly and the Incentive for Horizontal Merger”, The American
Economic Review, Vol. 75, No. 1, pp. 219-227.
Qiu, L. and W.Zhou (2006). “Merger waves: A model of Endogenous Mergers”, mimeo.
Richardson, M. (1999). “Trade and Competition Policies: Concordia Discors?”, Oxford Economic Papers
51, 649-664.
Rysman, M.(2001). “Competition Policy as Strategic Trade”, mimeo, Boston University.
Salant, S., S. Switzer, R. Reynolds (1983). “Losses from Horizontal Merger: the Effects of an Exogenous
Change in Industry Structure on Cournot-Nash Equilibrium”, Quarterly Journal of Economics 98, 185-199.
Suedekum, J. (2006). “Cross-Border Mergers and National Champions in an Integrating Economy”, IZA
Discussion Paper N. 2220.
Appendix
Proof of Proposition 1: (i) We start the proof by analyzing separately the profitability of the merger in each
market. The change in profits of home firms in the home market is given by the following function
a " c 2 a " c 2 a " c " 2 2

CP h " "
4 9 9
2
a " c 8a " c " 20 2
.
36 (9)
CP h is concave in and CP h 0 when " a"c

10 and a"c
2 . Therefore CP h is
greater than 0 , that is, the merger has a positive effect in profits in the home market for - ¡"1/10, 1/2¢ .
Since - ¡0, 1/3¢ by the nonnegative profits condition and by the restriction that u 0 then it follows
from this that the merger leads to gains in the home market for any - ¡0, 1/3¢ .
The change in profits of home firms in the third-country market due to the merger is given by
a " c 2 a " c 2 2 a " c " 3 2

CP 3 " "
16* 25* 25*
7a " c 2 " 82a " c 183 2
" .
400*
CP 3 is concave in and CP 3 0 when 7a " c /61 and a " c /3. Therefore
CP h is greater than 0 for - ¡7/61, 1/3¢ and so it follows from the above expression that the merger leads
7 1 7
to gains in the third-country market if 61 - 3 and losses if - 61
. This implies that the merger is
always profitable when - ¡7/61, 1/3¢ independently of any other parameter. When - ¡0, 7/61¢, the
merger leads to gains in home but losses in the third-country market. In this case, to evaluate when the merger is
profitable we need to compare the gains in the home market with the losses in the third-country market. The
merger will be profitable for this range of values of - when CP h u CP 3 or
a " c 2 8a " c " 20 2

7a " c 2 " 82a " c 183 2
u
36 400*
Solving the previous equation with respect to * we obtain that the gains in home are greater than the
63 7 "82-183- 2 f1,f2
*u 100 756-"140- 2
* h 1h 2 . - ¡0, 7/61¢ and
losses in the third-country market if Therefore, if
f1,f2
* ¡* h1h2 , 1¢ the merger is profitable because the losses in the third-country are smaller than the gains in
the home market.
(ii) The change in profits of home firms in the home market due to the merger is given by (chm2) and so the
merger leads to gains in the home market for any - ¡0, 1/3¢ . The change in profits of home firms in the
third-country market due to the merger is given by
a " c 2 a " c 2 a " c " 3 2

" " .
9* 16* 16*
1
It follows from the above expression that the merger leads to gains in the third-country market if - 15
1
and losses if - 15
. This implies that the merger is always profitable when - ¡1/15, 1/3¢. If
- ¡0, 1/15¢, then the merger leads to gains in home but losses in the third-country market. It follows from
(merger3) that the gains in home are greater than the losses in the third-country market if
50 1"18-45- 2 f1f2 f1f2
*u 100 18-"20-2
* h1h2 . - ¡0, 1/15¢ * * h1h2 , 1
Thus, If and the merger is
profitable. Q.E.D.
Proof of Proposition 2: The proof is similar to that of Proposition 1. Q.E.D.
Proof of Proposition 3: (i) If - ¡0, 1/15¢ and * ¡0, * M ¢ , then it follows from Proposition 1 that
N is a dominant strategy for home firms. Since F 2,1 is symmetric N is also a dominant strategy for foreign
firms. Therefore, NE F 2,1 N, N when - ¡0, 1/15¢ and * ¡0, * M ¢. (ii) If - ¡0, 1/15¢
and * ¡* M , * N ¢ , then it follows from Proposition 1 that the best reply of domestic firms to a merger by
foreign firms is for domestic firms to merge. This and the fact that F 2,1 is symmetric imply that M, M
NE F 2,1 when - ¡0, 1/15¢ and * ¡* M , * N ¢. If - ¡0, 1/15¢ and * ¡* M , * N ¢ , then it
follows from Proposition 1 that the best reply of domestic firms to N is to play N . This and the fact that
F 2,1 is symmetric imply that N, N NE F 2,1 when - ¡0, 1/15¢ and * ¡* M , * N ¢. So, for
this range of parameters F 2,1 has two symmetric PSNE. It is a well know result that the number of Nash
equilibria of this type of game must be odd. Since there is no other PSNE we must have a MSNE. By definition,
in a MSNE foreign firms randomize between M and N to make domestic firms indifferent between M and
N . Thus, in the MSNE we must have that
a " c 2 a " c 2 a " c 2 a " c 2

p 1 " p
4 9* 4 16*
a " c 2 a " c 2 a " c " 2 2 a " c " 3 2
p
9 16* 9 16*
a " c 2 a " c 2 2 a " c " 2 2 a " c " 3 2
1 " p
9 25* 9 25*
Solving this equation for p we obtain
63 " 100* " 738 800* - 1647 2000* - 2

p .
13 162- " 603- 2 (10)
Thus, NE F 2,1 £M, M , N, N , p, p ¤ if - ¡0, 1/15¢ and * ¡* M , * N ¢. (iii) If
- ¡0, 1/15¢ and * ¡* N , 1¢, then it follows from Proposition 1 that M is a dominant strategy for
home firms. Since F 2,1 is symmetric it is also a dominant strategy for foreign firms to play M . Therefore,
NE F 2,1 M, M when - ¡0, 1/15¢ and * ¡* N , 1¢. (iv) If - ¡1/15, 7/61¢ and
* 0, * N ¢, then it follows from Proposition 1 that the best reply of domestic firms to a merger by foreign
firms is for domestic firms to merge. This and the fact that F 2,1 is symmetric imply that M, M NE
F 2,1 when - ¡1/15, 7/61¢ and * 0, * N ¢. If - ¡1/15, 7/61¢ and * 0, * N ¢ , then it
follows from Proposition 1 that the best reply of domestic firms to N is to play N . This and the fact that
F 2,1 is symmetric imply that N, N NE F 2,1 when - ¡1/15, 7/61¢ and * 0, * N ¢. In this
case there also exists a MSNE where the probability of playing M is the same as the one obtained in (ii). Thus,
NE F 2,1 £M, M , N, N , p, p ¤ when - ¡1/15, 7/61¢ and * 0, * N ¢. (v) If
- ¡1/15, 7/61¢ and * ¡* N , 1¢, then it follows from Propositions 1 that it is a dominant strategy for
domestic firms to play M . Since F 2,1 is symmetric it is also a dominant strategy for foreign firms to play
M . Therefore, NE F 2,1 M, M when - ¡1/15, 7/61¢ and * ¡* N , 1¢. (vi) If
- ¡7/61, 1/3¢ , then it follows from Propositions 1 that M is a dominant strategy for domestic firms. Since
F 2,1 is symmetric it is also a dominant strategy for foreign firms to play M . Therefore, NE
F 2,1 M, M when - ¡7/61, 1/3¢ . Q.E.D.
Proof of Lemma 1: It follows directly from the definitions of +' , *h

f1,f2
, * h1,h2
f , *h
f1f2
and
* h1h2
f . Q.E.D.
h 1, h 2
Proof of Proposition 4: (i) If - ¡0, 1/15¢, +t + ' and * ¡0, * f ¢ , it follows from
Propositions 1 and 2 and from Lemma 1(i) that N is a dominant strategy for domestic firms and also for
1 h 1 h 2
foreign firms. Therefore NE F 2 , + N, N when - 0, 15
, + t +' and * 0, * f
h 1, h 2 f1 f2
. (ii) If - ¡0, 1/15¢, +t +' and * ¡* f , *h ¢, it follows from Propositions 1 and 2 and
Lemma 1(i) that N is a dominant strategy for domestic firms whereas M is a dominant strategy for foreign
h 1, h 2 f1 f2
firms. Therefore, NE F 2 , + N, M when - ¡0, 1/15¢, +t +' and * ¡* f , *h ¢ .
f1 f2
(iii) If - ¡0, 1/15¢, +t +' and * * h , 1 , it follows from Propositions 1 and 2 and from
Lemma 1(i) that M is a dominant strategy for foreign firms and also for foreign firms. Therefore, NE
f1 f2
F 2 , + M, M when - ¡0, 1/15¢, +t +' and * * h , 1 . (iv) If - ¡0, 1/15¢,
f1 f2
+ +' and * 0, * h ¢ , it follows from Propositions 1 and 2 and from Lemma 1(ii) that N is a
dominant strategy for home firms and also for foreign firms. Therefore, NE F 2 , + N, N when
f1 f2
- ¡0, 1/15¢, + +' and * 0, * h ¢. (v) If - ¡0, 1/15¢, + +' and
f1 f2
* ¡* h , * hf 1,h 2 ¢, it follows from Proposition 1 and from Lemma 1(ii) that the best reply of domestic
firms to a merger by foreign firms is for domestic firms to merge. Additionaly, the best reply of domestic firms
to an absence of merger of foreign firms is for domestic firms not to merge. The same holds for foreign firms.
From here it follows that in this range of parameters the merger game has two PSNE. In one PSNE domestic
firms merge and foreign firms merge in the other PSNE domestic firms do not merge and foreign firms do not
merge. As we have said before given that we have a even number of PSNE we must have one MSNE. In the
MSNE home firms merge with probability p h defined by
a " c 2 a " c 2 a " c 2 a " c 2

ph 1 " p h
4+ 9* 4+ 16*
a " c 2 a " c 2 a " c " 2 2 a " c " 3 2
ph
9+ 16* 9+ 16*
a " c 2 a " c 2 2 a " c " 2 2 a " c " 3 2
1 " p h
9+ 25* 9+ 25*
Solving this equation for p h we obtain
63+ " 100* " 738+ 800* - 1647+ 2000* - 2

ph .
13 162- " 603- 2 +
On other hand, in the MSNE foreign firms merge with probability p defined by (p). Thus, NE
f1 f2
F 2,1 £M, M , N, N , p, p ¤ when - ¡0, 1/15¢, + +' and * ¡* h , * h1,h2
f ¢. (vi)
h1,h2
If - ¡0, 1/15¢, + +' and * ¡* f , 1¢ , it follows from Propositions 1 and 2 and from Lemma
1(ii) that M is a dominant strategy for both domestic and foreing firms. Therefore, NE F 2,+ M, M
when - ¡0, 1/15¢, + +' and * ¡* hf 1,h 2 , 1¢. (vii) If - ¡1/15, 7/61¢, and
* 0, * hf 1,h 2 ¢, it follows from Proposition 1 that the best reply of home firms to M is M and the best
reply of home firms to N is N . The same holds for foreign firms. So, we have one PSNE where domestic
firms merge and foreign firms merge and another PSNE where domestic firms do not merge and foreign firms do
not merge. The MSNE is obtained as in part (v). Thus, NE F 2,+ £M, M , N, N , p h , p ¤ when
h 1,h 2
- ¡1/15, 7/61¢, and * 0, * f ¢. (viii) If - ¡1/15, 7/61¢ and * ¡* hf 1,h 2 , 1¢, it follows
from Propositions 1 and 2 that it is a dominant strategy for domestic firms to merge. The same is true regarding
h 1, h 2
foreign firms. Thus, NE F 2,+ M, M when - ¡1/15, 7/61¢ and * ¡* f , 1¢ . (ix) If
- ¡7/61, 1/3¢ , it follows from Propositions 1 and 2 that it is a dominant strategy for domestic firms to
merge. It is also a dominant strategy for foreign firms to merge. Thus, NE F 2,+ M, M when
- ¡7/61, 1/3¢ . Q.E.D.
Proof of Proposition 5: (i) The change in welfare in the home market given the merger is given by the
following function
3a " c 2 2a " 2c " 2 a " c 2 a " c " 2 2

CW h " "
8 18 9
"5a " c 2 32a " c " 44 2
.
72 (11)
5a"c a"c
CWh is concave in and CW h 0 when 22 and 2 . Therefore CW h is
greater than 0 for - ¡5/22, 1/2¢ , that is, the merger has a positive welfare impact in the home market for
- ¡5/22, 1/2¢ . Since - ¡0, 1/3¢, the merger leads to welfare losses in the home market for
- ¡0, 5/22¢ and to gains when - ¡5/22, 1/3¢. We know from Propositon 1 that, when foreign firms
have not merged, the merger of domestic firms leads to gains in the third-country market if - ¡7/61, 1/3¢
and losses if - ¡0, 7/61¢. Therefore, conditional on foreign firms not having merged, a merger of domestic
firms improves national welfare when - ¡5/22, 1/3¢ and reduces national welfare when - ¡0, 7/61¢ .
When - ¡7/61, 5/22¢ we have losses at home and gains in the third-country market. In this case the merger
of domestic firms improves national welfare if losses at home are smaller than gains in the third-country market.
1 "1 18 -"45 -2 f1,f2
* 3 7"16 -16 -2
bh .
This happens if So, conditional on foreign firms not having merged, a merger
of domestic firms improves national welfare if either (a) - ¡5/22, 1/3¢ or (b) - ¡7/61, 5/22¢ and
f1,f2
* 0, b h ¢.
(ii) We know from (i) that the merger of domestic firms leads to welfare losses in the home market for
- ¡0, 5/22¢ and to gains when - ¡5/22, 1/3¢. We also know from Proposition 1 that, when foreign
firms have merged, the merger of domestic firms leads to gains in the third-country market if - ¡1/15, 1/3¢
and losses if - ¡0, 1/15¢. Therefore, conditional on foreign firms having merged, a merger of domestic
firms improves national welfare when - ¡5/22, 1/3¢ and reduces it when - ¡0, 1/15¢. When
- ¡1/15, 5/22¢ we have losses at home and gains in the third-country market. In this case the merger of
domestic firms improves national welfare if losses at home are smaller than gains in the third-country market.
"118 -"45 - 2 f1f2
* bh .
That happens if 5 "32 -44 -2 So, conditional on foreign firms having merged, a merger of
domestic firms improves national welfare if either (a) - ¡5/22, 1/3¢ or (b) - ¡1/15, 5/22¢ and
f1f2
* 0, b h . Q.E.D.
Proof of Proposition 6: The proof is similar to that of Proposition 5. Q.E.D.
Proof of Proposition 7: (i) If - ¡0, 1/15¢, then it follows from Proposition 5 that N is a dominant
strategy for home. Since G 2,1 is symmetric, then N is also a dominant strategy for foreign. Therefore, NE
G 2,1 N, N when - ¡0, 1/15¢. (ii) If - ¡1/15, 7/61¢ and * 0, b M ¢, then it follows
from Proposition 5 that the best reply of home to N is to play N. Since the game is symmetric the best reply
of foreign to N is to play N . This shows that N, N NE G 2,1 when - ¡1/15, 7/61¢ and
* 0, b M ¢. If - ¡1/15, 7/61¢ and * 0, b M ¢, then it follows from Proposition 5 that the best
reply of home to M is to play M. Since the game is symmetric the best reply of foreign to M is to play M
. This shows that M, M NE G 2,1 when - ¡1/15, 7/61¢ and * 0, b M ¢. It is a well know
result that the number of Nash equilibria of this type of game must be odd. Since there is no other equilibrium in
pure strategies we must have a MSNE. By definition, in a MSNE foreign randomizes between M and N to
make home indifferent between M and N . Thus, in the MSNE we must have that
a " c 2 a " c 2 a " c 2

q
8 4 9*
a " c 2 a " c 2 a " c 2
1 " q
8 4 16*
2a " 2c " 2 a " c 2 a " c 2 a " c " 2 2
q
9 9 16* 9
a " c " 3 2 2a " 2c " 2 a " c 2
q 1 " q
16* 9 9
a " c 2 2 a " c " 2 2 a " c " 3 2
1 " q
25* 9 25*
Solving this equation for q we obtain
21 350* " 246 800* - 549 800* - 2

q3 .
13 162- " 603- 2
Thus, NE G 2,1 £M, M , N, N , q, q ¤ when - ¡1/15, 7/61¢ and * 0, b M ¢. (iii) If
- ¡1/15, 7/61¢ and * ¡b M , 1¢, then it follows from Proposition 5 that N is a dominant strategy for
home. Since G 2,1 is symmetric, then N is also a dominant strategy for foreign. Therefore, NE
G 2,1 N, N when - ¡1/15, 7/61¢ and * ¡b M , 1¢. (iv) If - ¡7/61, 5/22¢ and
* 0, b N ¢ , then it follows from Proposition 5 that M is a dominant strategy for home. Since G 2,1 is
symmetric, then M is also a dominant strategy for foreign. Therefore, NE G 2,1 M, M when
- ¡7/61, 5/22¢ and * 0, b N ¢. (v) If - ¡7/61, 5/22¢ and * ¡b N , b M ¢, then it follows from
Proposition 5 that the best reply of home to N is to play N. Since the game is symmetric the best reply of
foreign to N is to play N . This shows that N, N NE G 2,1 when - ¡7/61, 5/22¢ and
* ¡b N , b M ¢. If - ¡7/61, 5/22¢ and * ¡b N , b M ¢, then it follows from Proposition 5 that the best
reply of home to M is to play M. Since the game is symmetric the best reply of foreign to M is to play M
. This shows that M, M NE G 2,1 when - ¡7/61, 5/22¢ and * ¡b N , b M ¢. In this case there
also exists a MSNE where the probability of playing M is the same as the one obtained in (ii). Thus, NE
G 2,1 £M, M , N, N , q, q ¤ when - ¡7/61, 5/22¢ and * ¡b N , b M ¢. (vi) If
- ¡7/61, 5/22¢ and * ¡b M , 1¢, then it follows from Proposition 5 that N is a dominant strategy for
home. Since G 2,1 is symmetric, then N is also a dominant strategy for foreign. Therefore, NE
G 2,1 N, N when - ¡7/61, 5/22¢ and * ¡b M , 1¢. (vii) If - ¡5/22, 1/3¢ , then it follows
from Proposition 5 that M is a dominant strategy for home. Since G 2,1 is symmetric, then M is also a
dominant strategy for foreign. Therefore, NE G 2,1 M, M when - ¡5/22, 1/3¢. Q.E.D.
Proof of Proposition 8: (i) If - ¡0, 1/15¢ and * 0, * M ¢, then it follows from Proposition 3(i)
and 7(i) that NE F 2,1 NE G 2,1 N, N . (ii) If - ¡0, 1/15¢ and * ¡* N , 1¢, then it
follows from Proposition 3(iii) and 7(i) that NE F 2,1 M, M p N, N NE G 2,1 . If
- ¡1/15, 7/61¢ and * ¡max- * N - , b M - , 1¢, then it follows from Proposition 3(v) and 7(iii)
that NE F 2,1 M, M p N, N NE G 2,1 . If - ¡7/61, 25 " 91 /89¢ and
* ¡b M , 1¢, then it follows from Proposition 3(v) and 7(vi) that NE F 2,1 M, M p N, N NE
G 2,1 . (iii) If - ¡5/22, 1/3¢, then it follows from Propositions 3(vi) and 7(vii) that NE F 2,1 NE
G 2,1 M, M . Q.E.D.
MERGER ANALYSIS IN THE BANKING INDUSTRY:
THE MORTGAGE LOANS AND THE SHORT TERM CORPORATE CREDIT MARKETS*
Duarte Brito†
Universidade Nova de Lisboa
Pedro Pereira‡
Autoridade da Concorrência
Tiago Ribeiro§
Indera
January 2008
Abstract
This article, assesses the unilateral and coordinated effects on the Portuguese mortgage loans and
short term corporate credit markets of the merger between the banks BCP and BPI. We use a rich cross-
section of micro level data and a discrete choice model to estimate the price elasticities of demand and the
marginal costs of mortgage loans and short term corporate credit. Based on these estimates, we simulate the
impact of the merger on prices and welfare. Regarding unilateral effects, our results indicate that the merger
would lead to an average increase in the prices of mortgage loans of 3.1%, and to an average increase in the
spread of 9.9%. Additionally, the merger would lead to an average increase in the prices of short term
corporate credit of 7.4%. Regarding coordinated effects, our results indicate that the merger would
significantly increase the profitability of collusion between the three largest banks. We also simulate the
effects of one of the remedies proposed by BCP: selling-off 10% of the branches of BPI.
Keywords: Mortgage loans, Short run credit, Merger, Prices.
JEL Classification: L25, L41, G21, G34.
* The opinions expressed in this article reflect only the authors' views, and in no way bind the institutions to which they are affiliated.
†
DCSA, Faculdade de Ciências e Tecnologia da Universidade Nova de Lisboa, Quinta da Torre, 2829-516 Caparica, Portugal. e-mail:
dmb@fct.unl.pt
‡
Autoridade da Concorrência, Rua Laura Alves, nº 4, 6º, 1050-188 Lisboa, Portugal, e-mail: jpe.pereira@netcabo.pt
§
Indera - Estudos Económicos, Lda, Edifício Península, Praça Bom Sucesso, 127/131, Sala 202, 4150-146 Porto, Portugal, e-mail:
tiago.ribeiro@indera.pt
68 Duarte Brito – Pedro Pereira – Tiago Ribeiro
1. Introduction
In March 2006, the bank BCP proposed the acquisition of the bank BPI. The operation was approved in
phase II by the Portuguese Competition Authority, with remedies.1 This article, assesses for the Portuguese
mortgage loans and short term corporate credit markets, the unilateral and the coordinated effects of the merger
on the prices, i.e., on the interest rates.2
According to the Portuguese Banking Association, in the first semester of 2006 there were 48 active banks in
Portugal. These banks owned assets worth 327.035 million euros, of which 204.334 million euros were loans.
Despite the large number of banks, most of the assets were concentrated in a small number of institutions. Seven
banks accounted for 95% of the total loans, and 91% of the industry assets.3
In June 2006, mortgage loans represented 79.9% of total loans to households, which in turn, amounted to
55.1% of the loans to the private sector, i.e., loans to households and firms.4 The seven banks in our sample were
responsible for more than 85% of the value of mortgage loans contracted in 2004. Considering only these banks,
the joint market shares of the participants of the value of mortgage loans belonged to the interval [30% - 40%].
Table 1: Market shares for the banks in the sample

Loans to non-financial firms represented 44.8% of total loans to the private sector, i.e., loans to households
and firms. In the subset of six banks that make up our sample for the short term corporate credit market, the joint
market shares of BCP and BPI belonged to the interval [40% - 50%].
1. The remedies were: (i) the disposal of BCP's and BPI's participations in the largest Portuguese credit card acquirer and the launch of an
acquiring card operation, (ii) the sell-off of 60 branches of BPI, and (iii) the introduction of measures to reduce client switching costs.
2. See Ivaldi, Jullien, Rey, Seabright, and Tirole (2003b) for a review of the literature on unilateral effects, and Kovacic, Marshall, Marx,
and Schulenberg (2006), Ivaldi, Jullien, Rey, Seabright, and Tirole (2003a), and Davies (2006) for a review of the literature on
coordinated effects. See also the EC Merger Regulation Council Regulation (EC) No 139/2004 of 20 January 2004 (OJ L 24, 29.01.2004,
p.1), and the US Merger Guidelines (DoJ and FTC, 1997).
3. These numbers could be larger because these banks control some smaller banks.
4. Source: Boletim Estatístico do Banco de Portugal - Outubro/06.
Merger Analysis in the Industry: The Mortgage Loans and the Short Term Corporate Credit Market 69
Confidentiality requirements prevent us from identifying these banks. Hence, we will refer to the two
participants in the merger as banks I1 and I2 and to the other banks as banks O1 to O5.
Table 1 presents the distribution of market shares for the two markets within the group of banks in our
samples.
We use a rich cross-section of consumer level data and a multinomial logit model to estimate the price
elasticities of demands for mortgage loans and short term corporate credit.5 In both cases, the demand functions
of the banks are elastic with respect to price, but the market demand is inelastic. Assuming that firms play a
Bertrand game, we estimate marginal costs. Given the estimates of the firms' pricing strategies and the cost
estimates, we simulate the unilateral and coordinated effects of the merger on prices, market shares, profits, and
consumer surplus.
Regarding the unilateral effects in the mortgage loans market, our results indicate that the merger would
increase the prices of mortgage loans on average by 3.1%. The prices of bank I1 and bank I2 would increase by,
respectively, 7.1% and 16.3%. The average increase in prices is associated with an average increase in the spread
between the Euribor and the interest banks charge of 9.9%. For bank I1 and bank I2, the spread would increase
by, respectively, 17.3% and 61.5%.6 On average, the consumer surplus per household would decrease by 87
euros per year, the profits of bank I1 and bank I2 per household would increase by 7 euros per year, the profits of
the remaining banks per household would increase by 73 euros per year, and the social welfare per household
would decrease by 7 euros per year.7
For the short term corporate credit, our results indicate that the merger would lead to an increase in price of
7.4%. The prices of bank I1 and bank I2 would increase by, respectively, 6.3% and 35.6%. On average, the
average client firm's profit would decrease by 1,001 euros per year, the profits of bank I1 and bank I2, per firm,
5. According to Crooke, Froeb, Tschantz, and Werden (1999), predicted post-merger price changes vary greatly with the demand
specification. The price increases predicted by the logit model are lower than those predicted by the log-linear and AIDS models, but
higher than those predicted by the linear demand.
6. All contracts in our sample have adjustable rates. These contracts feature adjustments of the interest rate at regular time intervals, based
on the evolution of a predetermined index, e.g., the Euribor. The interest rate is adjusted to a rate that equals the current index value plus
a predetermined margin, or spread. According to Low, Sebag-Montefiore, and Dübel (2003), variable rate contracts represented 95% of
new lending in Portugal, in 1999, a figure that clearly contrasts with other European countries.
7. We did not consider merger induced cost efficiency gains because: (i) BCP did not claim them, and (ii) the literature does not clearly
support their existence in the banking industry. Berger and Humphrey (1992) and Srinivasan and Wall (1992) analyzed the efficiency
effects of bank mergers, and found that these do not, on average, result in efficiency gains. Rhoades (1998) summarizes nine case studies
on the efficiency effects of bank mergers, selected among those that seemed more likely to result in efficiency gains, and reports that only
four of these were successful. Peristiani (1997) finds no evidence that in-market merger leads to improvements in bank efficiency. Wang
(2003) introduces a measure of bank output, accounting for risk, that has the potential to identify merger indued cost savings.
would increase by 171 euros per year, the profits of the remaining banks per firm would increase by 735 euros
per year, and the social welfare, per firm, would decrease by 94 euros per year.
Regarding the coordinated effects, we follow the approach proposed by Kovacic, Marshall, Marx, and
Schulenberg (2006). This approach considers that a change in market structure increases the incentives for a set
of firms to collude, if the change in market structure increases the profits of the colluding firms. Our results
indicate that the merger would increase the aggregate profits in the mortage loan market under collusion of the
three largest banks, bank I1, bank O4, and bank O5, by 54.2%. On average, the merger would increase the
consumer surplus loss from collusion by 37.5%, or by an additional 456 euros, per household, per year.
For the short term corporate credit, the merger would increase the profitability of the collusion between the
three largest banks by 275.3%. The collusion after the merger, compared to the collusion without the merger,
would decrease, on average, profit per firm by an additional 8,791 euros per year. The merger would increase the
average client firm loss from collusion by 226%.
We also simulate the effects of one of the remedies proposed by BCP: selling-off 10% of the branches of
BPI. Our results indicate that the impact of the remedy is negligible, both on the unilateral and coordinated
effects of the merger.
Our methodological approach draws on the discrete choice literature, represented among others by
Domencich and McFadden (1975), McFadden (1974), McFadden (1978), and McFadden (1981). In the industrial
organization literature, Berry (1994), Berry, Levinsohn, and Pakes (1995), Goldberg (1995), and Nevo (2001)
applied discrete choice models to the analysis of market structure. Dube (2005), Ivaldi (2005), Ivaldi and
Verboven (2005), Nevo (2000), Pereira and Ribeiro (2007b) and Pinkse and Slade (2004) analyzed the impact of
mergers in a framework similar to ours.8 These studies used aggregate data, with the exception of Dube (2005).
Pereira and Ribeiro (2007a) analyzed the effects on broadband access to the Internet of the divestiture, the
opposite of a merger, of the Portuguese telecommunications incumbent from the cable television industry.
To our knowledge, there are no applications of merger analysis or discrete choice models to the mortgage
loans market. However, our research relates to four strands of the empirical literature on the banking industry.
First, our research relates to the literature that uses discrete choice models to estimate the demand for several
types of banking deposits and loans, e.g., Dick (2002), Molnár, Nagy, and Horváth (2006), and Nakane, Alencar,
and Kanczuk (2005).9 Second, our research relates to the literature that evaluates ex-post the competitive impact
of mergers, e.g., Focarelli and Panetta (2003), Prager and Hannan (1998), Sapienza (2002). The findings of this
literature are largely inconclusive. Third, our research relates to the literature that evaluates the level of
competition for various banking products, e.g., Adams, Roeller, and Sickles (2002), Berg and Kim (1998),
Shaffer (1993), Shaffer (1989), and Neven and Röller (1999). This literature shows that competitive conditions
vary greatly across products, countries and even periods. Fourth, our research relates to studies of the mortgage
loans market that analyze various aspects such as the determinants of demand or price discrimination, e.g., Jones
(1995), Breslaw, Irvine, and Rahman (1996), Follain and Dunsky (1997), Ling and McGill (1998), Gary-Bobo
(2003), Gary-Bobo and Larribeau (2004), Leece (2006), Moriizumi (2000), and Paiella and Pozzolo (2006).
The rest of the article is organized as follows. Section 2 describes the data. Section 3 presents the model.
Section 4 describes the econometric implementation and presents the basic estimation results. Section 5 analyzes
the unilateral effects of the merger, section 6 analyzes the coordinated effects of the merger, and section 7
analyses the effects of a remedy proposed by BCP. Finally, section 8 concludes.
2. Data
The data used in this study consists of a rich micro level cross-section. For the mortgage loans market, we
obtained data from seven banks, which are among the eight largest banks in terms of the value of mortgage
loans, for mainland Portugal.10
For each bank, we obtained two samples of client data. The first sample included 1,000 clients from each
bank, to whom the bank granted, in 2004 or 2005, some form of mortgage credit. The second sample, also of
1,000 clients from each bank in 2004 or 2005, was extracted from the universe of clients of the bank, for the
same period, regardless of the type of credit granted. We obtained for each individual: (i) the age, (ii) the
income, (iii) the total amount of debt in the banking system, (iv) the amount of credit granted, (v) the value and
8. See also Baker and Bresnahan (1985), Hausman, Leonard, and Zona (1994), and Ashenfelter, Ashmore, Baker, Gleason, and Hosken
(2004).
9. Dick (2002) estimates a multinomial and nested logit models for commercial bank deposit services for the US. The results indicate that
consumers respond to deposit rates, and to a lesser extent, to account fees, when choosing their depository institution. Nakane, Alencar,
and Kanczuk (2005) use a multinomial logit model to study the demand for time deposits in Brazil, for an aggregate of demand and
passbook savings deposits and for loans. Molnár, Nagy, and Horváth (2006) analyze competition in the Hungarian household credit and
deposit markets, estimating multinomial logit deposit and loan demand functions for each bank.
type of the mortgaged asset, (vi) the term of the credit contract, (vii) the index rate, (viii) the spread, and (ix) the
residence location. We also obtained the number of elements in the population from which both samples were
drawn, as well as the population average for these variables, to assess the representativeness of the sample.11
Table 2 presents the summary statistics of the variables used in the mortgage loans model.12
Table 2: Descriptive statistics – Mortgage loans sample
For the short term corporate credit market we obtained two similar samples of information on the small and
medium firms (SMF) clients of six banks, which are among the eleven largest banks in terms of total credit, for
mainland Portugal.13 The first sample included 1,000 small and medium firms clients from each bank, to whom
the bank granted, in 2004 and 2005, short run corporate credit in the form of overdrafts and secured current
accounts. The second sample, also of 1,000 small and medium firms clients of each bank, was extracted from the
universe of bank clients, regardless of the type of credit granted. We obtained for each firm: (i) number of years
10. Jointly, these seven banks accounted for 85.6% of the value of mortgage loans contracted in 2004. Only one bank outside our sample had
a larger market share, 7.4%, than that of the smallest bank we considered. The remaining 7% were scattered among smaller banks.
11. We excluded observations considered errors or outliers, according to the criteria (values in million euros): (i) annual income >1, (ii) total
debt to the banking sector <.015 or > .5, (iii) loan amount <.01 or >.5, (iv) collateral <.02 or >.75, (v) duration <5 or > 45 years, and
(vi) non-positive spread.
12. We also obtained for each individual: (a) the number of years as client of the bank, (b) the professional occupation, (c) the credit rating,
(d) the commissions paid, and (e) other assets, liabilities or products held at the bank. The occupation and the residence were used to
impute a small number of income observations, but were otherwise not statistical significant in the model. The other variables collected
had inconsistent codings across banks or a large amount of missing values, and were therefore not used.
as client of the bank, (ii) location and number of employees, (iii) total assets and total liabilities, (iv) cash and
cash equivalents, (v) accounts receivable, (vi) fixed capital, (vii) net revenue, (viii) net results, (ix) total bank
loans payable, (x) credit rating, (xi) average credit outstanding, (xii) average interest rate, (xiii) commissions
paid, and (xiv) other assets and liabilities or products held at the bank.
Table 3 presents the summary statistics of the variables used in the short term corporate credit model.14
Table 3: Descriptive statistics – Short term credit
We completed our data set with the number of branches of each bank by municipality on December 2004.15
13. Of the other five banks, only one is significantly larger than the smallest bank in our sample.
14. Some of the variables collected presented inconsistent coding across banks and/or had a significant amount of missing values and were
therefore not used.
15. The source is the Boletim Informativo, Associação Portuguesa de Bancos, Ano 18, Nº 35, Julho de 2005.
3. Economic Model
In this section, we present the econometric models we use to analyse the two markets. The two models are
similar and we start with a brief introduction to the discrete choice model we estimate for the mortgage loans
market. A second subsection highlights the differences between the two models.
3.1. The Mortgage Loans market
3.1.1. Demand
Utility of Mortgage Loans - Index consumers with subscript n = 1,..., N , and mortgage loan products
with subscript i = 1,..., I . A consumer chooses among a set of alternative mortgage loan products. The products
differ in: (i) the price, i.e., the interest rate,16 (ii) the bank that provides the credit, (iii) the distribution of bank
branches throughout the country, and (iv) the term of the contracts. The demand of each consumer for a given
alternative depends on his type. The type of a consumer is defined by a K dimensional vector of characteristics,
zn , which includes: (i) the age, (ii) the annual income, (iii) the amount of credit required, (iv) the value of the
asset, (v) the total debt in the banking system, and (vi) the place of residence.17
Consumer n derives from alternative i utility:
U ni r in , z n , x i , /ni , 2 V ni r in , z n , x i , 2 /ni ,
(1)
where rin is the price of alternative i for consumer n , xi is a J dimensional vector of the other characteristics
of alternative i , θ is a vector of parameters, and finally ε ni is a random disturbance independent across products,
consumers, and identically distributed. We assume additionally that:
V ni r in , z n , x i , 2 : r in )z n , 2 ) 5x i , 2 5 ,
(2)
16. The interest rate interacts with the amount of credit required, and is therefore a proxy for the monthly installments of a given credit
contract, which is what consumers care for.
17. We treat the value of the asset being acquired and the amount of credit required as exogenous variables. Consumers make their decisions
in two stages. First, they decide which asset to purchase and how much credit they require. Second, given the amount of credit required,
they choose a mortgage loan product.
where
K
§ ·
α ( zn ,θα ) := − exp ¨ ¦znkθα k ¸ , (3)
© k =1 ¹
J
λ ( xi ,θ λ ) := ¦xijθ λ j , (4)
j =1
θ := (θα ,θ λ ), (5)
and where α (⋅) is the price coefficient, i.e., the negative of the marginal utility of income, which depends on
individual characteristics. The exponential transformation imposes the restriction that this coefficient is
negative.18 Expression λ (⋅) is a linear combination that summarizes the utility component associated with the
product characteristics other than price. The parameters θα translate the effect of the characteristics of the
consumers on the price coefficient. The parameters θλ translate the valuation of the consumers for the product
characteristics other than price.
Assuming that the random disturbance ε ni has an extreme value Type I distribution, one obtains the
standard multinomial logit model.
Choice Probabilities – Consumer n chooses product i if U ni > U nj , for all j ≠ i . This occurs with
probability:
P ni : Pr V ni " V nj /ni /nj , for allj p i, j 1, T , I
; F i V ni " V n1 u, T , u, T , V ni " V nI u du,

where F (⋅) is the joint distribution function of (ε n1 ,! , ε nI ) , and Fi (⋅) is its partial derivative with respect to
th
the i argument.
18. The average α is almost unchanged by this restriction, since the restriction only binds in the tails for very low values of α .
Neverevertheless, imposing that the values of α are always consistent with economic theory gives more stability to the numerical
simulations performed later in the paper.
If F (⋅) is an extreme value type I distribution, one obtains the standard multinomial logit expression for the
choice probabilities:
P ni e Vni .
! j e Vnj
(6)
Choice Set Generation - We do not observe the choice set of each individual. In principle, a consumer could
have chosen any of the products observed to have been selected, possibly conditional on his individual
characteristics. It is unfeasible to enumerate all possible choices and estimate models with such large choice sets.
Instead, we construct a choice set for each individual by drawing product characteristics from pre-specified
probability distributions to define the different alternatives, and add to this set the choice effectively made.19
Denote by An the set of possible alternatives assigned to consumer n , and denote by G ( An | i ) the probability
of assigning the set An to consumer n , given that the chosen alternative was i . We omit subscript n whenever
possible. The probability of alternative i being chosen, given that the choice is restricted to set A is:
GA|i P i
Pi|A .
! jA GA|j P j
(7)
In the multinomial logit model this probability becomes:
expV i lnGA|i
Pi|A .
! jA expV j lnGA|j
(8)
We draw product characteristics from a joint distribution of (ri , xi ) , independent of the chosen alternative,
whose density we denote by h(⋅) . Hence, we have:
lnGA|i ! lnhr j , x j ! lnhrj , x j " lnhri , x i k " lnhr i , x i ,

jA,jpi jA
(9)
19. This methodology was used by, e.g., Train, McFadden, and Ben-Akiva (1987).
where k is a constant. The observed probabilities become:
expV i " lnhr i , x i

Pi|A .
! jA expV j " lnhr j , x j
(10)
Likelihood Function - The log-likelihood function corresponding to the multinomial logit is:
L ! ! y ni lnP n i|A n ,
n i
(11)
where yni = 1 if individual n chooses product i , and yni = 0 otherwise.
Calibration - Our sample is choice based, and is, therefore, not random. Thus, the predicted probabilities do not
reflect market shares, and product dummy variables reflect the sample composition, and not the market. One can
obtain consistent estimates for the coefficients of the product dummy variables by a calibration process that
adjusts the product dummy variables, so that the model's predicted market shares match the actual market
shares.20 Partition the vector of coefficients, θ λ , into (θ λ , θ λ ) , where θλ1

1 2
represents the coefficients associated
with product dummy variables, and θλ2 represents all the remaining coefficients in θλ . Let si represent the
2
correct market share of product i , θλ the estimated value of θλ2 , and θα the estimated value of θα . The
calibrated value of θλ1 , denoted by θλ1 , is defined by:
§ §2 2
I N
¨1 ! Mn P ni 2 ) , 2 15 , 2 5
2 5 : arg min
1
! si "
I
n1
§ §2 ,
25
i1 ! j1 ! Nn1 Mn P nj 2 ) , 2 15 , 2 5
(12)
where M n is the amount of credit required by individual n .
20. There are several alternative techniques to correct the bias of some of the coefficients of the model. See, e.g., Manski and McFadden
(1981), in particular chapters 1 and 2. The first method that appeared in the econometrics literature addressing this issue was the WESML
estimator of Manski and Lerman (1977).
It is possible to add new products not present in the initial sample, provided that one can compute Pni . All
that is required is that one: (i) knows the value of the exogenous variables that characterize these products, and,
(ii) includes new product dummy variables. We use this mechanism to introduce an outside option of no demand
for credit, and report results based on different assumptions about the market share of the outside option.21 All
variables that define this option are set to zero, except the dummy variable.
Price Elasticities of Demand - One of our goals is to determine the price changes caused by the merger. As
an intermediate step, we compute the price elasticities of demand for the products of each bank. These
expressions should be taken merely as indicative of the price changes that may occur, since the computation of
the price, detailed below, cannot be expressed as a function of the elasticities reported here.
Denote by ε nij , the elasticity of demand of product i with respect to the price of product j for consumer n :
P ni r n t 1 .
/nij :
t j P
t1 ni (13)
In the multinomial logit model this expression simplifies to:
) n r i 1 " P ni if i j
/nij
") n r j P nj if i p j.
(14)
Let Qi := ¦ nN=1M ni Pni (rn ) represent the expected total volume of credit of product i . Denote by ε ij , the
elasticity of the total volume of credit of product i with respect to the price of product j:
! Nn1 Mni P ni r n t 1 .
/ij :
t j Qi
t1 (15)
21. The outside option refers to the alternatives to mortgage loans, which are: renting or self-financing.
Finally let Q := ¦ iI=1Qi represent the total volume of credit granted. Denote by ε , the elasticity of the total
volume of credit with respect to the average market price:22
I N
/ :
Qrt 1 ! i1 ! n1 Mni P ni r n t 1.
t t1 Q t Q
t1 (16)
Consumer Welfare Variation - Denote by Vnj0 and Vnj1 , the utility levels before and after the merger,
respectively. The merger may imply three types of changes. First, prices may change, which requires computing
the market equilibrium after the merger. Second, the characteristics of the products other than price, xi , may
change. Third, the number of products offered, I , may change. We assume that the number of products offered,
as well as the product characteristics other than price, do not change. The generalized extreme value model, of
which the multinomial logit model is a particular case, provides a convenient computational formula for the
exact consumers' surplus, up to a constant, associated with a policy that changes the attributes of the products in
the market. Such expression, known as the “log sum” formula, is:
CS n 1 ln' e V1n1 , T , e V1nJ 0

" ln' e Vn1 , T , e VnJ
0

)Mn
J J
1 ln ! e V1nj
" ln ! e Vnj
0
,
)Mn
j1 j1
(17)
where Ψ (⋅) is the probability generating function of the generalized extreme value distribution.23
This formula is valid only when the indirect utility function is linear in income, i.e., when prices changes
have no income effects, which is the case assumed here.
22. This elasticity only applies when there is an outside option.

23. Expression (17) was developed by Domencich and McFadden (1975), and McFadden (1974) for the multinomial logit model, and by
McFadden (1978) and McFadden (1981) for the nested logit model. Small and Rosen (1981) discuss the connection between (17) and
standard measures of consumer surplus.
3.1.2. Supply
Price Equilibrium - Index firms with subscript b = 1,..., B . We assume that banks choose prices, and play
a static non-cooperative game, i.e., play a Bertrand game. Denote by cb ( z ) , the constant marginal cost of bank
b providing credit to an individual with characteristics z , denote by φ (⋅) the density of z , and denote by Fb ,
the fixed costs of b .24 Assume that banks can observe the vector of consumer characteristics. In these
circumstances, banks can price discriminate between types of consumers. The strategy of firm b is then a rule,
rb ( z ) , that says which price the firm should charge for each consumer with vector of characteristics, z . Denote
th
by r ( z ) the B dimensional vector with b element rb ( z ) , that includes the strategies of all banks for that type
of consumer. Variable M is one of the elements of z that represents the amount required by the individual and
thus enters directly in each bank's profit function. The payoff of bank b , up to normalization of the market size,
is given by:
$ b ;¡r b z " c b z ¢MP b rz , z Cz dz " F b .

(18)
Given that the demand and costs of each type of consumer do not depend on the prices offered to the other types
of consumers, the firms' profit maximization problem is separable across types.
We assume that the current prices are equilibrium prices, and compute the marginal costs such that the
current first-order conditions are satisfied. The computation of the price changes after a merger is done by fixing
the computed marginal costs, and solving the new first-order conditions that emerge when the market structure
changes. Under different scenarios of ownership and collusion, the objective of bank b is to maximize the
= ¦ B γ Π , with respect to prices,
function Π
b k =1 bk k
rb , where γ bk = 1 if bank b takes bank k 's profit into
account when setting prices, and γ bk = 0 otherwise. The property matrix Γ consists of the elements
Γbk := γ bk .
24. Most empirical articles on the cost structure of banks, e.g., Bernstein (1996), Hunter, Timme, and Yang (1990), Beard, Caudill, and
Gropper (1991), Gropper, Caudill, and Beard (1999), Karafolas and Mantakas (1996), Cebenoyan (1990), and Mitchell and Onvural
(1996), conclude that the technology of the banking industry presents no economies of scope, and that economies of scale can only be
found in the smaller banks. Girardone, Molyneux, and Gardener (2004), however, finds economies of scale.
We restrict the pricing rules rb (⋅) to belong to a certain set based on: (i) economic reasons, and (ii) statistical
reasons. From an economic perspective, we conjecture that banks discriminate between individuals according to
simple pricing rules. This is justified by the computational burden of setting complex price schemes, that may be
prohibitive for banks, or may be unwarranted if the information on the individuals at time of contracting is
incomplete or contains errors.25 From a statistical perspective, it is unfeasible to identify complex pricing rules
from a limited number of observations. Thus, even if firms follow complex pricing rules, the data limitations
force one to approximate them through simple pricing rules.
Next we describe how we implement the general procedure defined above. The empirical counterpart of the
~
objective function Π defined for each bank is:
b
B N
b
$ ! +bk N1 !r nk " cnk Mn P nk r n ,
k1 n1
(19)
where rnb , cnb and rn and denote the discrete counterparts of rb ( z ), cb ( z ) and r ( z ) . Let:26
nb
$ P B P
f nb r n ;c n , : M n P nb r nb " c nb M n nb ! k1 +bk r nk " c nk M n nk .
r nb r nb kpb
r nb
(20)
25. Some regularity conditions on the set of functions to which rb (⋅) belongs, such as that the functions do not oscillate too much, are also
necessary for the stated problem to be meaningful.
26. Function f nb (⋅ ) measures the impact on the objective function of bank b of an increase in its interest rate, rnb , which can be divided in
three effects. First, by increasing its interest rate, bank b collects a higher interest from individual n ' s loan, M n , with the probability that
this individual selects bank b , Pnb . This profit margin effect is represented by the first term: M n Pnb . Second, by increasing its interest
rate, bank b decreases the utility of consumer n when selecting bank b . As a consequence, the probability that individual n selects
∂Pnb
bank b decreases. This volume of sales effect is represented by the second term: ( rnb − cnb ) M n . Third, by increasing its interest rate,
∂rnb
bank b raises the probability that consumer n chooses another bank l . If bank b takes the profit of bank l into account when setting its
prices, i.e., if γ bl =1 , this effect is positive for bank b , otherwise this effect is irrelevant. This consumer switching effect is represented by
∂P
the third term: ¦γ bk ( rnk − cnk ) M n nk .
∂r
nb
Denote the period before and after the merger by, respectively, t = 0,1 . Regarding the pricing rule, we
assume that rnb = rnb + δ b . This means that after the merger, i.e., at t = 1 , each bank increases the interest rate
t 0
by a given value, common to all of its customers, i.e., the prices paid by all the customers of a bank vary by the
same fixed amount, which we denote by δ b . This reduces the dimension of the firm's problem.
Regarding the cost structure, we assume that cnb = cb , for all n . This means that the marginal costs do not
differ across customers of the same bank.
Hence, the current equilibrium can be characterized by the first-order conditions with respect to each δ b at
the point δ b = 0 . Denote by c , a B dimensional vectors with b element cb , and denote by r a NB
th t
dimensional vector that results from the piling-up of vectors rn . At any time t , and for b = 1,..., B , the first-
t
order conditions that characterize the Nash equilibrium are given by:27
N N
nb r nb
$
t
E b r ;c, : ! 1 ! f nb r tn ;c, 0.
r nb - b N
n1 n1
(21)
Denote by ψ ( r ; c, Γ ) , the B dimensional vector with generic element ψ b (r t ; c, Γ) . The Nash equilibrium at
t
time t is thus characterized by ψ ( r ; c, Γ t ) = 0 . At t = 0 , if the observed prices are equilibrium prices, these
t
conditions must be verified when δ b = 0 , i.e., ψ (r 0 ; c, Γ 0 ) = 0 . We use these first-order conditions to obtain
estimates of the marginal costs that are consistent with the assumption that the observed prices are equilibrium
prices. These estimates are given by the solution to:28
min
c
Er 0 ;c, 0 U Er 0 ;c, 0 .
(22)
Denote by ĉ the estimates of the marginal costs, and denote by δ , the B dimensional vector with b element δ b .
th
The price increases at t = 1 can then be obtained by solving:
27. We assume that a Nash equilibrium exists. Caplin and Nalebuff (1991) proved existence in a general discrete choice model, with single
product firms. Anderson and de Palma (1992) proved existence for the nested logit model with symmetric multiproduct firms.
28. Solving the minimization problem is equivalent to solving a system of equations. The formulation merely defines marginal costs, and
subsequently price changes, as GMM estimators.
§ §
min Er 1 ; c, 1 U Er 1 ; c, 1 .
-
(23)
Initially there are seven firms. Thus, Γ 0 is the identity matrix: Γ 0 = I 7 . In the course of the analysis, we
will assume two alternative forms for the matrix Γ , associated with the cases of: (i) the merger of bank I1 and
bank I2, and (ii) perfect collusion between alternative sets of banks.
Profit Variation - The profit variation for product j is then:

N
$j ! r 1nj " ƙ j P nj r 1n " r 0nj " ƙ j P nj r 0n Mn .
n1
(24)
3.2. The Short Term Corporate Credit Market
3.2.1. Demand
The main difference between the two markets we analysed results from the fact the SMF's maximize profit
instead of utility. For this market a SMF chooses among a set of alternative banking products in order to
maximize its profits. For a given banking product with price (interest rate) r and loan amount M the firm's
profit are Π( r , Μ, z , x ) = Π 0( M , z , x ) − rM , where Π0 are the profits that result from optimization taken
M as given, z are variables that characterize this optimization operation and can include prices of inputs and
outputs whose determination is considered to occur at the same time as the choice of the credit product,
quantities of predetermined inputs and outputs and other factors that characterize the environment in which the
firm operates and x is a vector with product characteristics other than price and amount. The products differ in
several characteristics.
Firm n has a total profit if it opts for alternative i = 1,..., I expressed as:
$ ni r i , z n , x i , /ni , 2 V ni r i , z n , x i , 2 /ni ,
(25)
where ri is the price of alternative i , xi is a J dimensional vector of the other characteristics of alternative i ,
zn is a K dimensional vector of firm characteristics, θ is a vector of parameters and, finally, ε ni is a random

disturbance independent across products, firms, and identically distributed.
The price of an alternative is considered to be the interest rate interacted with the amount of credit associated
with a given option. Other product characteristics are the bank that provides the credit. The firm is characterized
by the composition of its assets and liabilities, by the total volume of sales and the amount of credit that it seeks.
All other assumptions regarding choice set generation and calibration issues are as in the mortgage loans
case described above, with the obvious difference that individual n should be replaced by SMF n.
The expected maximum profit of the firm n is given by:
I
E$ n E maxV 0nj , T , V 0nI ln ! expV 0ni k.
i1
(26)
0
where k is a constant independent of V . Since Vnj is not measured in monetary units as the inequalities that
define the choice problem are unchanged by linear transformations of the profit levels we convert the measure
obtained above into monetary units dividing by αM , i.e., the coefficient on interest costs.
3.2.2. Supply
For the short term corporate credit market we considered a more flexible pricing rule, by setting
rnbt = rnb0 + ¦ kK=1β rbk

t
S k (rnb0 ) , where Sk (r ) is a basis-Spline function k defined on the variable r , and β rbk
t
are
the coefficients, to be estimated, that bank b assigns to the basis function k . See the Boor (1978) and Wahba
(1990). Hence, the current equilibrium can be characterized by the first-order conditions with respect to each
β rbk
0
at the point β rbk
0
= 0. The departures from this equilibrium can be made as flexible as desired by
increasing the number of basis functions. The extent to which one makes the approximation more flexible, i.e.,
increase K , is limited by our data. Likewise, we set cnb = ¦ k =1β cbk S k ( rnb ) . The average price increase for the
K 0
mortgage loans estimated under this more flexible approach with the case of cubic basis-splines with knots at
quartiles of the distribution of r as basis functions, i.e., K = 7 , does not differ much from those reported in this
article using the simple pricing rule.
4. Econometric Implementation
In this section we present the basic estimation results first for the mortgage loans market and then for the
short term corporate credit market.
4.1. The Mortgage Loans market
4.1.1. Basic Estimation Results
We estimated the model by maximum likelihood.29 Table 4 presents the results.

The estimates of most coefficients are statistically significant at a 1% confidence level.
The coefficients presented in Table 4 above the line refer to the linear index that defines the price
coefficient, α (⋅) . The estimates reveal that the price coefficient is decreasing in the total debt in the banking
system, and increasing in the amount of credit required, the value of the asset being acquired, and the individual
income. Individuals value longer loan terms. This is true, not only for younger individuals, individuals with
lower incomes, and individuals with larger debts, but also for individuals requiring a larger amount of credit to
purchase a lower valued asset.
29. All procedures were coded in MATLAB.

Table 4: Model 1 – Mortgage loans

The median of the distribution of the estimate of the price coefficient is 51.076. The estimates of the
calibrated coefficients of product characteristics, θλ , reflect the median consumer's incremental valuation of
1
these attributes relative to those of the products of bank I1. These estimates are also presented in Table 4. For
instance, the negative coefficient -0.438 for bank O5 translates into a negative median interest rate premium of
0.438
51.076 = 0.008 for the products of this bank relative to those of bank I1. For the median individual, the disutility
associated with the products of bank O5 compared to those of bank I1 can be compensated by an interest rate
reduction of 0.008.30
The estimates of the coefficients of the products of bank or product dummy variables with the individuals'
income level are all negative, or not statistically significant. This means that individuals with larger income
levels have a relative preference towards bank I1. The estimates of all the coefficients of the products of bank
dummy variables with the amount of credit required are positive. This means that the higher the amount of credit
needed, the lower the preference towards any bank relative to bank I1. With the exception of bank I2, the
estimates of the coefficients of the products of bank dummy variables with the debt level in the banking system
are positive, and the estimates of the coefficients of the products of bank dummy variables with the value of the
asset are negative. Thus, individuals with higher debts in the banking system and purchasing assets of lower
value tend to show a larger preference towards other banks than bank I1 and bank I2. The estimates of the
coefficients of the interaction of bank dummy variables and individual age is inconclusive.
Regarding the local branch network, the estimates reveal that individuals have a negative valuation for the
banks that have no branches in the their municipality of residence. For consumers residing in the Lisbon District,
the estimate of the coefficient of the number of local branches is not statistically significant. An increase in the
number of local branches has no value for these consumers. This may be the result of over-branching in the
Lisbon district. For consumers residing in the rest of the country, the estimate of the coefficient of the number of
local branches is positive and statistically significant. The number of local branches has a positive impact on
consumer valuation.
30. The model has no bank characteristics other than the dummy variables and the number of local branches. Thus, the bank dummy
variables capture most bank characteristics other than the contract characteristics. As a consequence, expressing the value of these
dummy variables in an interest rate metric may generate unrealistic premiums, which reflect merely the fact that differences in market
shares are due to more than differences in the prices of the mortgage loans, and that changing this variable does not, by itself, equate
market shares.
As it is well know, the multinomial logit model has the property of independence of irrelevant alternatives,
IIA. We performed three types of tests to determine if the data complied with the IIA property, and the direction
of any eventual departure from this property. First, we performed Hausman-McFadden type of tests (Hausman
and McFadden (1984)), using different subgroups of the choices. The tests were implemented in their Lagrange
Multiplier version as described in McFadden (1987). The null hypothesis that the data complies with the IIA
property was not rejected. Second, we performed Lagrange Multiplier tests with the nested logit model as the
alternative, also described in McFadden (1987). The null hypothesis was also not rejected in most cases. Third,
we performed Lagrange Multiplier specification tests for the multinomial logit against mixture models described
in McFadden and Train (2000). The null hypothesis was, again, not rejected.
4.1.2. Price Elasticities of Demand
Table 5 presents the elasticity of the total volume of credit granted by bank i with respect to changes in
price j , ε ij , assuming that there is no outside option. The weights used are the amount of credit required
multiplied by the probability of each individual making the loan at a given bank.
Table 5: Elasticities for calibrated model without outside option – Mortgage loans
The demand functions of the banks for the mortgage loan products are elastic with respect to price. The
own-price elasticities of demand ranges from -1.34 for bank O4 to -4.02 for bank O1. The own-price elasticity of
demand of bank I1 and bank I2 are, respectively, -1.40 and -2.28.
Bank O4 is the bank for which a price increase results in the largest increases in the demands of the other
banks. As bank O4 is the largest bank, this is the expected result.31 Similarly, an increase in the price of bank O2,
the smallest bank in our sample, results in the smallest increases in the demands of the other banks. Interestingly,
when bank I1 increases its price by 1%, bank I2 is the bank whose demand increase the most, 0.47%. If,
however, bank I2 increases its price, the demand of the other large banks increases almost in the same
proportion, with a slight preference towards bank O3.
Tables 6-9 present the price elasticities of demand of the total volume of credit granted, for market shares of
the outside option ranging from 5 to 30%, i.e., for sizes of the market for mortgage loans relative to the amount
of the volume of credit granted ranging from 1.05 to 1.43, respectively.
Table 6: Elasticities for calibrated model with outside option at 5% – Mortgage loans
Table 7: Elasticities for calibrated model without outside option at 10% – Mortgage loans
31. By construction, this is always the case for the individual price elasticities of demand.
As the share of the outside option increases, the firms' price elasticities of demand decrease, but only
slightly. If the market share of the outside option is 30%, the own-price elasticities of demand for bank I1, bank
O4, and bank I2 become, respectively: -1.27, -1.23, and -1.70. Bank O1 is the bank with the highest elasticity,
-2.45.
For each individual, as the market share of the outside option increases, the own-price elasticity of demand
increases in absolute terms, while the cross-price elasticities of demand decrease. However, the weighted
average over all individuals of the own-price elasticities, described in section 3.1.1, may decrease with the
market share of the outside option.32 The unweighted average of the individual price elasticities of demand, ε nij ,
for different values of the market share of the outside option, is available upon request.
32. The weights, the amount of credit required multiplied by the probability of each individual making the loan at a given bank, change with
the market share of the outside option. As the market share of the outside option rises, the more price sensitive individuals tend to be
increasingly assigned to this alternative. Hence, if the probability of a high elasticity individual signing the contract with a given bank
decreases with the introduction of the outside option more than the elasticity of a low elasticity individual, the bank's own-price elasticity
of demand may decrease, in absolute terms, with the market share of the outside option.
Table 10 presents the elasticity of the total market demand for mortgage loans, ε , for market shares of the
outside option ranging from 5 to 30%.
Table 10: Market elasticity – Mortgage loans
The larger the size of the outside option, the more elastic the market demand is. However, the market
demand is rigid and does not change much with the market share of the outside option. For market shares of the
outside option in the 5 to 30% range, the market own-price elasticities of demand vary between -0.417 to -0.629,
respectively. A larger share of the outside option means that it is a more attractive alternative. Hence, if the
market prices increase, a higher percentage of consumers shift towards it.
The more reasonable values for the market share of the outside option are those in the range of 0 to 10%,
based on the following three arguments. First, between June 2003 and June 2005, the index rates were at
historically low levels. Hence, most consumers interested in obtaining mortgage loans had the ideal conditions to
do so. Second, the Portuguese renting market has some peculiarities regarding rent control. As a result, the
supply of housing for rent is scarce.33 Third, the literature on the demand for mortgage loans, e.g., Moriizumi
(2000) and Leece (2006), presents very inelastic demands, which are compatible with ours for small market
shares of the outside option.
4.2.1. Basic Estimation Results
Table 11 presents the results of the maximum likelyhood estimations for the short term corporate credit
market.
33. According to the INE, Recenseamento Geral da População e Habitação - 2001 (Resultados Definitivos), of the 3.551.229 existing
lodgings, only 605.288, i.e., 17%, were rented.
The estimates of most coefficients are statistically significant at a 1% confidence level. The estimates
indicate that the small and medium firms' price sensitivity is decreasing in the firm's total liabilities and
increasing in the amount of short run corporate credit, total assets and overall debt in the banking system. The
median of the distribution of each firm's sensitivity to price, α , is -17.011. Relative to bank I1, the median firm
places a positive extra valuation on the products of bank O5 and a negative one on the products of bank O2.
These valuations seem not to depend on the total amount of debt in the banking system. Considering only the
significant coefficients, all the coefficients associated with the interaction between the bank dummy variables
and cash and cash equivalents are positive, meaning that firms with larger cash balances tend not to favour bank
I1. On the opposite, the same coefficients for total assets, turnover and short run debt are negative, implying that
firms with a larger assets, turnover and balances of short run credit tend to show a preference towards bank I1.
As in the case of the mortgage loans market, the null hypothesis that the data complies with the IIA property
was not rejected. However, the Lagrange Multiplier tests with the nested logit model as a more specific
alternative rejected the null hypothesis in most cases, not giving much guidance as to a possible pattern of
nesting to be implemented. We performed the Lagrange Multiplier specification tests for the multinomial logit
against mixture models described in McFadden and Train (2000). The null hypothesis is rejected and the
variables associated with missing random components are the product dummies suggesting existence of
heterogeneity of firms beyond those captured by the individual characteristics already included. A model which
this mixing structure will be implemented in the following version of the current paper.
4.2.2. Price Elasticities of Demand
Table 12 presents the elasticity of the total volume of credit granted by bank i with respect to changes in
price j , ε ij , for the parameter estimates of Model 2, and assuming that there is no outside option.
Table 11: Model 2 – Short term credit

Table 12: Elasticities for calibrated model – Short term credit
The bank's demands for the products considered are elastic with respect to price, with the own-price
elasticities of demand ranging from -1.37 for bank I2 and -2.29 for bank O3. The own-price elasticities of
demand for bank I1 is -1.80.
Contrary to the mortgage loans case, when bank I1 increases its price by 1% the bank whose demand
increase the least, 0.53%, is bank I2. If, however, bank I2 increases its price, the demand of the other large banks
increases almost in the same proportion, with a slight preference towards bank O5.
Tables 13-16 present the price elasticities of demand of the total volume of credit granted, for market shares
of the outside option ranging from 5 to 30%, i.e., for sizes of the market for short term credit relative to the
amount of the volume of credit granted ranging from 1.05 to 1.43, respectively.
Table 13: Elasticities for calibrated model with outside option – Short term credit
As the share of the outside option increases, the firms' price elasticities of demand decrease, but only
slightly. If the market share of the outside option is 30%, the own-price elasticities of demand for bank I1, bank
O3, and bank I2 become, respectively: -1.66, -1.77, and -1.09.
Table 17 presents the elasticity of the total market demand for short term corporate loans, ε , for market
shares of the outside option ranging from 5 to 30%.
Table 17: Market elasticity – Short term credit
The results are qualitatively similar to the mortgage loans case.
5. Analysis: Unilateral Effects
In this section, we analyze the effects of the concentration operation, assuming Nash behavior ex-ante.
5.1. The Mortgage Loans Market
The merger of bank I1 and bank I2 leads to a market with six firms. Thus, the merger consists of a change in
the property matrix from Γ 0 to Γ1 , given by:
1 1 0 0 0 0 0
1 1 0 0 0 0 0
0 0 1 0 0 0 0
1 0 0 0 1 0 0 0 .
0 0 0 0 1 0 0
0 0 0 0 0 1 0
0 0 0 0 0 0 1
Table 18 reports the estimates of the marginal costs and the impact of the unilateral effects of the merger on
prices, for the case where there is no outside option.
Table 18: Marginal cost and price variation estimates – Mortgage loans
After the merger, the prices of mortgage loans increase on average 3.1%. The prices of bank I1 and bank I2
increase by, respectively, 7.1% and 16.3%. The prices of bank O2, bank O3 and bank O5 decrease. However,
these latter price variations are not statistically significant at a 5% level. These price variations are associated
with an average increase in the spread between the Euribor and the interest banks charge of 9.9%34. For bank I1
and bank I2 the spread increases by, respectively, 17.3% and 61.5%.
34. The price each individual pays is the sum of the index rate plus the spread. The increase in price is a weighted average of the increase in
the index rate and the increase in spread. The weights are the relative importance of both terms in the pre-merger price. The index rate is
assumed not to change with the merger. Hence, the percentage variation in price is equal to the percentage variation in the spread
multiplied by the weight of the spread in the original price.
Table 19 reports the impact of the unilateral effects of the merger on prices, for market shares of the outside
option ranging from 5 to 30%.
Table 19: Percent change in prices with alternative outside option assumptions – Mortgage loans
The average price increase is very robust to changes in the market share of the outside option. With the
exception of bank I2, the expected increase in each bank's price after the merger does not change significantly
for different values of the market share of the outside option35. Depending on the market share of the outside
option, the estimated increase in the price of bank I1 varies between 9.5% and 6.7%, and the estimated increase
in the price of bank I2 price varies between 19.9% and 14.2%.
Table 20 reports the impact of the unilateral effects of the merger on welfare, for market shares of the
outside option ranging from 0 to 30%.
Table 20: Welfare effects for different market shares of the outside option – Mortgage loans
35. Some prices increase more when the outside option has a larger market share because of the reduction in the banks' own-price elasticities,
referred in section 4.2.
Without an outside option, after the merger, on average, the consumer surplus per household decreases by 87
euros per year, the profits of bank I1 and bank I2 per household increase by 7 euros per year, the profits of the
remaining banks per household increase by 73 euros per year, and the social welfare per household decreases by
7 euros per year. As expected, in the absence of merger induced cost reductions, the merger has a larger impact
on the non-merging banks than on the merging banks. The non-merging banks benefit from the larger increase in
the prices of the merging banks. This increases their market shares, even when they raise their own prices.
If the market share of the outside option is 10%, on average, the consumer surplus per household decreases
by 94 euros per year, the profits of bank I1 and bank I2 per household increase by 7 euros per year, the profits of
the remaining banks per household increase by 66 euros per year, and the social welfare per household decreases
by 22 euros per year.
Table 21 reports the estimates of the marginal costs and the impact of the unilateral effects of the merger on
prices, for the case where there is no outside option in the short term corporate credit market. Figures 1 and 2
plot the marginal costs and price changes against current prices that result from using a more flexible pricing rule
as described in section 3.1.2. The results for the average of the population are the same if ones uses the more
simple rule of estimating one marginal cost and one price change per bank, however the graphs suggest that one
should estimate a constant markup for each bank and not a constant marginal cost or a constant price change.
Table 21: Marginal cost and price variation estimates – Short term credit
Figure 1: Marginal Costs
Figure 2: Price Changes
After the merger, the prices of short term corporate loans increase on average 7.4%. The prices of bank I1
and bank I2 increase by, respectively, 6.3% and 35.6%. The prices of bank O3, bank O4, and bank O5 increase
slightly. However, these latter price variations are not statistically significant at a 5% level.
Table 22 reports the impact of the unilateral effects of the merger on prices, for market shares of the outside
option ranging from 5 to 30%.
Table 22: Percent change in prices with alternative outside option assumptions – Short term credit
With the exception of bank I2, the expected increase in each bank's price after the merger does not change
significantly for different values of the market share of the outside option.36 The estimated increase in bank I2's
price varies between 35.6% and 48.3%, depending on the market share of the outside option. It should be noted
that the average price increase is very robust to changes in the market share of the outside option.
Table 23 reports the unilateral impact of the merger on welfare, for market shares of the outside option
ranging from 0 to 30%.
Table 23: Welfare effects for different market shares of the outside option – Short term credit
Without an outside option, after the merger, the average SMF profit decreases by 1,001 euros per year, the
profits of bank I1 and bank I2 per SMF increase by 171 euros per year, the profits of the remaining banks per
SMF increase by 735 euros per year, and the social welfare per SMF decreases by 94 euros per year.
36. Some prices increase more when the outside option has a larger market share because of the reduction in the banks' own-price elasticities,
explained above in section 4.2.
If the market share of the outside option is 10%, the average SMF profit decreases by 1,051 euros per year,
the profits of bank I1 and bank I2 per SMF increase by 187 euros per year, the profits of the remaining banks per
SMF increase by 724 euros per year, and the social welfare per SMF decreases by 141 euros per year.
5.3. Plausibility of the Nash ex-ante Assumption
We assumed that before the merger, firms played a Bertrand game. But firms could have played a game that
led to either more or less competitive outcomes, than those implied by a Bertrand game. Table 24 presents the
estimates of the marginal costs consistent with Nash behavior, the quarterly average of capital costs reported by
the banks, and the Euribor interest rates.37
Table 24: Estimated and Reported Capital Costs
The estimated marginal costs do not differ much from either the reported average of capital costs, or the
Euribor. This gives some support to the assumption that firms play a Nash equilibrium, but of course this
comparison is only a very crude validation of the assumption.38
Assuming that firms colluded in prices before the merger results in non-positive marginal cost estimates,
except when the outside option has a large market share, but even in this case they are small. We take this as
indirect evidence that firms did not collude perfectly on prices.
37. There is no simple relation between the marginal costs of mortgage loans and the reported costs of funding. On the one hand, the
marginal costs of mortgage loans include the costs of other inputs, such as labour and physical capital, which means that the estimated
marginal costs should be larger than the reported capital cost. On the other hand, mortgage loans involve a low risk, which implies that
the cost of financing this type of credit may be lower than the cost of financing the bank's average credit.
38. If we had estimates of the marginal costs we could test our estimates of the marginal costs assuming Nash behaviour against the observed
marginal costs (Pereira and Ribeiro (2007b)).
6. Analysis: Coordinated Effects
To analyze the coordinated effects of the merger, we follow the approach proposed by Kovacic, Marshall,
Marx, and Schulenberg (2006).39 Rather than focusing on whether collusion is more easily sustained after a
merger, we analyze how the merger affects the profitability of collusion.40 More specifically, we simulate the
effects of an hypothetical collusion between a set of banks in two alternative scenarios: (i) with the merger, and
(ii) without the merger. First, we estimate the impact of collusion on prices and market shares for both scenarios.
Second, given the previous values, we evaluate the impact of collusion on profits for both scenarios. Third, we
compare the increase in profits from collusion for both scenarios to obtain a measure of the incentives firms have
to collude.
We consider the possibility that the three largest banks, bank O4, bank I1 and bank O5, collude. The results
obtained would not differ significantly: (i) if bank O5 was replaced by bank O3, given the similarities between
these banks in terms of market shares and elasticities, and (ii) if the smaller banks were included in the collusive
agreement. Adding bank O3 to the colluding trio would be very close to perfect collusion.
Table 25 presents the change in prices, profits, and consumer surplus caused by the collusion of the three
largest banks, when there is no outside option.
39. When cost data is available one can also use the approach of Nevo (2001) and Slade (2004).
40. The methodology consists of extending the procedure used to estimate the unilateral effects to the analysis of the coordinated effects. In
the absence of cost synergies or changes in the products characteristics, both a merger or a collusive arrangement result in a group of
firms setting their prices to maximize joint profits. In these circumstances, the prices set by firms j and k should be the same, regardless
of these two firms merging or setting collusive prices. To estimate the effects of price collusion, the generic element of matrix Γ 's , γ bk ,
should be set to 1 if firms b and k either merge or collude. Following the same procedure as in the case of the merger, it is possible to
estimate the increase in prices and profits resulting from a given hypothetical collusive agreement, in the presence and absence of the
merger.
Table 25: Coordination effects with no outside option – Mortgage loans
Without the merger, if the three banks collude, their prices increase by, respectively: 64.7%, 59.9% and
76.3%, as can be seen in column (0) → (2) . With the merger, if bank I1, bank O4 and bank O5 collude, their
prices and those of bank I2, increase by, respectively: 74.0%, 75.0%, 99.0%, and 78.9%, as seen in
column (1) → (3) .41 The price increases caused by collusion are different with and without the merger for two
41. These price increases should be accumulated to those resulting solely from the merger. If the merger occurs, the prices of bank I1 and
bank I2 increase by, respectively, 7.1% and 16.3%, as reported in Table 18 or column (0) →(1) of Table 25.
reasons. First, with the merger, the colluding banks take bank I2's profit into account when setting their prices.
Second, with the merger, the colluding banks control an additional instrument: bank I2's price.
Without the merger, if bank I1, bank O4 and bank O5 collude, their aggregate profit increases by 4,099.5
euros per million euros of total market size, as presented in column (0) → (2) . With the merger, if these banks
collude, their aggregate profit increases by 6,320.4 euros per million euros of total market size, as can be seen in
column (1) → (3) , row AGG2 . In other words, the increase in profits of these three banks from collusion is
54.2% larger with the merger than without the merger. If the profits of bank I2 are considered, the merger
increases the change in profits from collusion of the four banks by 24.3%, because the profits of bank I2 increase
relatively less than the other colluding banks. The increase in the profits of bank O4 and bank O5 from the
collusion between the three largest banks are, respectively, 56.9% and 36% larger with the merger than without
the merger. This can be seen by comparing the values presented in columns (0) → (2) and (1) → (3) for each
bank. As for the merging banks, the increase in their aggregate profit from collusion is almost the same with the
merger than without the merger.42
Without the merger, if bank I1, bank O4 and bank O5 collude, the consumer surplus per household decreases
by 1,217 euros per year. With the merger, if these banks collude the consumer surplus per household further
decreases by 1,673 euros per year, to which should be added the loss of 87 euros per year that results from the
merger alone. Hence, with the merger the effect of collusion on consumer surplus, per household, is 456 euros
per year, or 37.5%, larger.
Tables 26 and 27 present the estimated variation in profits and consumer surplus from the collusion of bank
O4, bank I1 and bank O5, for market shares of the outside option ranging of 10% and 30%43.
42. We compare the 2,806.1 increase in the profits from collusion of both bank I1 and bank I2 with the merger, with the 2,845.9 increase in
profits from collusion without the merger. This last figure may overestimate the incentives to collude because it includes the gain to bank
I2, which, without the merger, was not part of the set of colluding firms. The alternative of comparing the gains of both bank I1 and bank
I2 from collusion with the merger, with the 1,365.2 increase in the bank I1 profits from collusion without the merger may be misleading,
as it compares the increase in the profits of two banks with the increase in the profits of a single bank. This problem occurs because the
merger has two effects on collusion: (i) it changes the increase in profits from collusion for each bank, and (ii) it changes the number of
colluding firms.
43. Similar tables for market shares of the outside option of 5% and 20% are available upon request.
Table 26: Coordination effects with outside option at 10% – Mortgage loans
The effect of the merger on the increase in profits from collusion does not vary much with changes in the
market share of the outside option. For instance, suppose that the outside option has a market share of 10%.
Without the merger, if the three banks collude, their aggregate profits increase by 2,780.1 euros per million euros
of total market size, while with the merger, the profits from collusion of these banks increase by 4,594.4 euros
per million euros of total market size. Hence, with the merger the profits from collusion of these banks increase
65.3%.
Table 27: Coordination effects with outside option at 30% – Mortgage loans
Table 28 summarizes the impact of the merger on the profitability of collusion between the three largest
banks for different values of the market share of the outside option.
Table 28: Merger impact on the profitability of collusion – Mortgage loans
As table 28 illustrates, the effect that the merger has on the increase in the aggregate profits from the
collusion of bank I1, bank O4 and bank O5, presented on row AGG2 , does not depend much on the market
share of the outside option.
Table 29 presents the change in prices, market shares and profits in the short term corporate credit market
caused by the same collusive arrangement, when there is no outside option.
Without the merger, if bank I1, bank O4 and bank O5 collude, their prices increase by, respectively: 32.2%,
46.2% and 53%, as can be seen in column (0) Æ (2). After the merger, if bank I1, bank O4 and bank O5 collude,
their prices and those of bank I2, increase by, respectively: 51.7%, 166.5%, 139.4%, and 211.8%, as seen in
column (1) → (3) 44.
44. Again, these price increases should be accumulated to those resulting solely from the merger. If the merger occurs, the prices of bank I1
and bank I2 increase by, respectively, 6.3% and 35.6%, as reported in Table 21 or column (0) →(1) of Table 29.
Table 29: Coordination effects with no outside option – Short term credit
Without the merger, if bank I1, bank O4 and bank O5 collude, their aggregate profit increases by 3,676.6
euros per million euros of total market size, as presented in column (0) → (2) . After the merger, however, the
increase in the aggregate profits of the three banks is 13,798.5 euros per million euros of total market size, as can
be seen in column (1) → (3) , row AGG 2 . In other words, the increase in profits from collusion is 275.3%
larger with the merger than without the merger. If the profits of bank I2 are considered, the merger increases the
change in the four firm's profits resulting from collusion by 123.8%, because bank I2's profits increase relatively
less than the other colluding banks. If banks are considered individually, the increase in the profits of bank O4
and bank O5 from the collusion between the three largest banks are, respectively, 36.1% and 259.5% larger if the
merger occurs than if it does not. This can be seen by comparing the values presented in
columns (0) → (2) and (1) → (3) for each bank. As for the merging banks, the increase in their joint profit
from collusion is 129.7% larger if the merger occurs.
Without the merger, the collusion between bank I1, bank O4 and bank O5 reduces the average SMF profit
by 3,890 euros per year. After the merger, the collusion further reduces the average SMF profit by 12,681 euros
per year, to which should be added the loss of 1,001 euros per year that results from the merger alone. Hence,
after the merger the effect of collusion on SMF profits, per firm, is 8,791 euros per year larger. In other words,
the merger amplifies by 226% the reduction in SMF profits due to collusion.
Table 30 summarizes the impact the merger has on the profitability of collusion between the three largest
banks for different values of the market share of the outside option.45 As can be seen, the impact of the merger
on the profitability of collusion changes substantially with the market share of the outside option. Despite of
these changes, the merger seems to increase significantly the profitability of collusion, regardless of the size of
the outside option.
Table 30: Merger impact on the profitability of collusion – Short term credit
45. Similar tables for market shares of the outside option of 5 %, 10%, 20% and 30% similar to those presented for the mortgage loans case
are available upon request.
7. A Structural Remedy
In this section, we analyze the unilateral and coordinated effects of a structural remedy proposed by BCP:
selling-off 60 branches of BPI, which represent about 10% of its branch network.46 The sell-off of the branches
includes the physical capital and the employees, as well as the client contracts whose demand deposits were
subscribed on the branch.
There is no information on the percentage of mortgage loans contracted in 2004 and 2005 that will be
transferred to other banks as a result of this remedy, nor is there any information on which banks will purchase
the branches. According to BCP, these 60 branches represent 11.9% of the total volume of credit, or 17.1% of
the total volume of credit if the capitals of the local municipalities are excluded. First, we assume that the 60
branches of BPI are sold to the largest bank in our sample. Second, we assume that the branches are sold to the
smallest bank. Throughout the section, we assume that there is no outside option.
Table 31 presents the unilateral effects of the merger if 60 branches of BPI are sold to bank O4, the largest
bank.
46. These 60 branches were selected from those in geographical areas in which both BCP and BPI were present to ensure that, after the
merger: (i) there is no decrease in the number of competitors in geographical areas which currently have less than four active banks, and
(ii) BCP will not own, after the merger, more than 40% of the branches in any geographical area. BCP defines geographical areas as
townships, Freguesias, unless these belong to the capital of a local municipality, sede de Concelho. In that case, the “geographical area” is
the whole capital of the local municipality.
The increase in the prices of bank I1 caused by the merger drops from 7.1%, without the remedy, to 7.0%,
with the remedy, while the increase in the prices of bank I2 drops from 16.3%, without the remedy, to 15.9%,
with the remedy. The impact of the remedy on the average change in prices is small: the estimated average
increase in prices caused by the merger changes from 3.1%, without remedy, to 3.2%, with the remedy. This
occurs because bank O4 hikes its price by 1.0%, with the remedy, instead of by 0.6%, without the remedy.
Table 32 presents the unilateral effects of the merger if 60 branches of BPI are sold to bank O1, the smallest
bank.
The results are similar to those of the previous case. With the remedy, the estimated increase in the prices of
bank I1 and bank I2 caused by the merger are, respectively, 7.1% and 18.3%, and the increase in the average
price is 3.4%. As expected, the prices of bank O1 increase more.
To sum up, the increase in the average price is larger with the remedies than without the remedies although
the difference is not statistically significant.
Table 33 presents the coordinated effects of the merger if 60 branches of BPI are sold to bank O4.
The increase in the profits of bank I1, bank O4 and bank O5 from collusion caused by the merger drops from
54.2%, without the remedy, to 53.1%, with the remedy.47 If the profits of bank I2 are included, the increase in
the profits of the four banks from collusion caused by the merger drops from 24.3%, without the remedy, to
24%, with the remedy. Similar values hold if the banks are considered individually.
47. This value is obtained by comparing the increase in profits from collusion after the merger is approved with remedies, presented in
column (1)→(3) , row AGG2 on Table 33, with the increase in profits from collusion when no merger takes place, presented in
column (0)→(2) , row AGG2 on Table 25.
Without the remedy, the merger amplifies the reduction in consumer surplus due to collusion by 37.5%,
whereas with the remedy, the merger amplifies the reduction in consumer surplus due to collusion by 37.4%,
almost the same value as without the remedy.
Table 34 presents the coordinated effects of the merger if 60 branches of BPI are sold to bank O1.
The increase in the profits of bank I1, bank O4 and bank O5 from collusion caused by the merger drops from
54.2%, without the remedy, to 50.7%, with the remedy. If the profits of bank I2 are included, the increase in the
profits of the four banks from collusion caused by the merger drops from 24.3%, without the remedy, to 22.2%,
with the remedy.
With the remedy, the merger amplifies the reduction in consumer surplus due to collusion by 35.8%.
To sum up, if 60 branches of BPI are sold to bank O4 or bank O1, the incentives for collusion decrease very
slightly. The impact of the remedy is modest, particularly on the consumer surplus.
Table 35 presents the unilateral effects of the merger after 17.1% of BPI clients are transferred to bank O4,
the largest bank.
The prices of bank I1 and bank I2 still increase with the merger, but by a smaller amount. bank I1's price
increase changes from 6.3%, without the remedies, to 6.0% while bank I2's price increase drops from 35.6% to
27.4%. The impact on average price, however, is small. When the structural remedy is considered, the estimated
increase in the average price drops from 7.4% to 6.9%. This results from the fact that bank O4 reacts by hiking
its price 4.2% instead of 1.1% in the case of no remedies.
Table 36 presents the unilateral effects of the merger after 17.1% of bank I2's clients are transferred to bank
O2.
The results are similar to those of the previous case. The estimated increase in bank I1's and bank I2's price
is, respectively, 5.1% and 27.5% and the increase in the average price is 5.8%.
Table 37 presents the coordinated effects of the merger after 17.1% of the value of BPI short term loans are
transferred to bank O4.
With collusion, the increase in the aggregate profits of the three colluding banks is 285% larger with the
merger than without the merger.48 In the absence of remedies, the same percentage was estimated at 275.3%. If
48. This value is obtained by comparing the increase in profits from collusion after the merger is approved with remedies, presented in
column (1)→(3) , row AGG2 on Table 37, with the increase in profits from collusion when no merger takes place, presented in
column (0)→(2) , row AGG2 on Table 29.
the profits of bank I2 are included, the merger, with remedies, increases the change in the profits resulting from
collusion of the four firms by 123.6% which should be compared to 123.8% when no remedies are implemented.
If banks are considered individually, the effect of the merger in the profits from collusion of bank O4 and bank
O5 changes from 36.1% and 259.5%, respectively, to 72.6% and 266.9%. As for the merging banks, joint profit
from collusion is 122.2% larger if the merger occurs and remedies are enforced than when it does not. In the
absence of remedies, bank I1 -bank I2 joint profit from collusion was 129.7% larger if the merger occurred.
Without the merger, the collusion between bank I1, bank O4 and bank O5 reduces the average SMF profit
3,890 euros per year. After the merger with remedies, the collusion reduces the SMF profits by 12,663 euros per
year. Hence, the merger amplifies by 225.5% the reduction in SMF profits due to collusion, whereas without
remedies this amplification effect amounted to 226%.
Table 38 presents the coordinated effects of the merger after 17.1% of the value of BPI short term loans are
transferred to bank O2.
If the three largest banks collude, the increase in profits from collusion is 310.7% larger after the merger
takes place, with remedies, than without the merger. This is larger than the 275.3% obtained without remedies.
Additionally, if the profits of bank I2 are considered, the merger increases the change in the four firm's profits
resulting from collusion by 136.5%, instead of 123.8% without remedies, when compared to the case of no
merger. The increase in the profits of bank O4 and bank O5 from the collusion between the three largest banks
are, respectively, 205.6 and 455.8% larger if the merger occurs. Without the remedies, the corresponding
percentages were smaller, 36.1% and 259.5%, respectively. The increase in the merging banks joint profit from
collusion is almost 99.5% larger if the merger occurs with remedies than when there is no merger. In the absence
of remedies, bank I1-bank I2 joint profit from collusion was 129.7% larger if the merger occurred.
As for SMF, the merger with remedies amplifies by 188.6% the reduction in profits due to collusion, instead
of 226%, when approved without remedies. Without the enforcement of these remedies, the average SMF loses,
due to collusion alone, 12,681 Euros per year, whereas with remedies this loss is 11,225.
8. Concluding Remarks
In this article, we evaluated the unilateral and coordination effects on the mortgage loans and the short term
corporate credit markets of the proposed merger between BCP and BPI. We used a rich cross-section of
consumer level data from a number of banks, that account for 85% of the mortgage loans' market and for 95% of
the loans to the private sector, and a discrete choice model to estimate the price elasticities of demand and the
marginal costs of mortgage loans and short term corporate credit. Given these estimates, we simulated the effects
on prices, market shares, and welfare of the merger. The general picture that emerges is that the unilateral effects
of merger are relatively small, both in terms of price increases and changes in consumer surplus. However, the
merger greatly enhances the benefits from subsequent collusion between the remaining banks.
References
Adams, R., L. Roeller, and R.Sickles (2002): “Market Power in Outputs and Inputs: An Empirical
Application to Banking,”.
Anderson, S. P., and A. de Palma (1992): “Multiproduct Firms: A Nested Logit Approach,” Journal of
Industrial Economics, 40(3), 261 - 276.
Ashenfelter, O., D. Ashmore, J. B. Baker, S Gleason, and D. S. Hosken (2004): “Econometric Methods in
Staples,”.
Baker, J., and T. Bresnahan (1985): “The Gains from Merger or Collusion in Product-Differentiated
Industries,” Journal of Industrial Organization, 33(4), 427-44.
Beard, T. R., S. B. Caudill, and D. M. Gropper (1991): “Finite Mixture Estimation of Multiproduct Cost
Functions.,” Review of Economics and Statistics, 73(4), 654-64.
Berg, S., and M. Kim (1998): “Banks as Multioutput Oligopolies: An Empirical Evaluation of the Retail and
Corporate Banking Markets,” Journal of Money, Credit and Banking, 30(2), 135-53.
Berger, A., and D. Humphrey (1992): “Megamergers in Banking and the Use of Cost Efficiency as an
Antitrust Defense,” Antitrust Bulletin, 37, 541 - 600.
Bernstein, D. (1996): “Asset Quality and Scale Economies in Banking,” Journal of Economics and Business,
48(2), 157-66.
Berry, S. (1994): “Estimating Discrete-Choice Models of Product Differentiation,” RAND Journal of
Economics, 25(2), 242 - 262.
Berry, S., J. Levinsohn, and A. Pakes (1995): “Automobile Prices in Market Equilibrium,” Econometrica,
63(4), 841-90.
Breslaw, J., I. Irvine, and A. Rahman (1996): “Instrument Choice: The Demand for Mortgages in Canada,”
Journal of Urban Economics, 39(3), 282 - 302.
Caplin, A., and B. Nalebuff (1991): “Aggregation and Imperfect Competition: On the Existence of
Equilibrium,” Econometrica, 59(1), 25 - 29.
Cebenoyan, A. S. (1990): “Scope Economies in Banking: The Hybrid Box-Cox Function,” Financial
Review, 25(1), 115-25.
Crooke, P., L. Froeb, S. Tschantz, and G. Werden (1999): “Effects of Assumed Demand Form on Simulated
Postmerger Equilibria,” Review of Industrial Organization, 15, 205 - 217.
Davies, P. (2006): “Coordinated Effects Merger Simulation with Linear Demands,” Discussion paper,
Competition Commission.
Davis, P. (2006): “Coordinated effects merger simulation with linear demands,” Discussion paper,
Competition Commission.
De Boor, C. (1978): A Practical Guide to Splines. Springer-Verlag, Berlin.
Dick, A. A. (2002): “Demand estimation and consumer welfare in the banking industry,” Discussion Paper
2002-58.
Domencich, T., and D. McFadden (1975): Urban Travel Demand: A Behavioral Analysis. North-Holland
Publishing.
Dube, J. (2005): “Product Differentiation and Mergers in the Carbonated Soft Drink Industry,” Journal of
Economics and Management Science, 14(4), 879 - 904.
Focarelli, D., and F. Panetta (2003): “Are Mergers Beneficial to Consumers? Evidence from the Market for
Bank Deposits,” American Economic Review, 93(4), 1152-72.
Follain, J. R., and R. M. Dunsky (1997): “The Demand for Mortgage Debt and the Income Tax,” Journal of
Housing Research, 8(2), 155-99.
Gary-Bobo, R. J., and S. Larribeau (2004): “A Structural Econometric Model of Price Discrimination in the
French Mortgage Lending Industry,” International Journal of Industrial Organization, 22 (1), 101 - 134.
Gary-Bobo, R. J. and Larribeau, S. (2003): “The Bank’s Market Power and the Interest-Rate Elasticity of
Demand for Housing: An Econometric Study of Discrimination on French Mortgage Data,” Annales d'Economie
et de Statistique, N° 71-72, 377-398.
Girardone, C., P. Molyneux, and E. P. M. Gardener (2004): “Analysing the Determinants of Bank
Efficiency: The Case of Italian Banks,” Applied Economics, 36(3), 215-27.
Goldberg, P. (1995): “Product Differentiation and Oligopoly in International Markets: The Case of the U.S.
Automobile Industry,” Econometrica, 63(4), 891 - 951.
Gropper, D. M., S. B. Caudill, and T. R. Beard (1999): “Estimating Multiproduct Cost Functions Over Time
Using a Mixture of Normals,” Journal of Productivity Analysis, 11(3), 201-18.
Hausman, J., G. Leonard, and J. Zona (1994): “Competitive Analysis with Differenciated Products,”
Annales d'Economie et de Statistique, 34, 159-80.
Hausman, J., and D. McFadden (1984): “A Specification test for the multinomial logit model,”
Econometrica, 52(5), 1219 - 1240.
Hunter, W. C., S. G. Timme, and W. K. Yang (1990): “An Examination of Cost Subadditivity and
Multiproduct Production in Large U.S. Banks,” Journal of Money, Credit and Banking, 22(4), 504-25.
Ivaldi, M. (2005): “Study on Competition Policy in the Portuguese Insurance Sector: Econometric
Measurement of Unilateral Effects in the CAIXA / BCP Merger Case,” Discussion Paper 7, Autoridade da
Concorrência Working Papers.
Ivaldi, M., B. Jullien, P. Rey, P. Seabright, and J. Tirole (2003a): “The Economics of Tacit Collusion,”
Discussion paper, DG Competition, European Commission.
_________ (2003b): “The Economics of Unilateral Effects,” Discussion paper, DG Competition, European
Commission.
Ivaldi, M., and F. Verboven (2005): “Quantifying the Effects from Horizontal Mergers in European
Competition Policy,” International Journal of Industrial Organization, 23(9), 699 - 702.
Jones, L. D. (1995): “Net Wealth, Marginal Tax Rates and the Demand for Home Mortgage Debt,” Regional
Science and Urban Economics, 25(3), 297 - 322.
Karafolas, S., and G. Mantakas (1996): “A Note on Cost Structure and Economies of Scale in Greek
Banking,” Journal of Banking and Finance, 20(2), 377-87.
Kovacic, W., R. Marshall, L. Marx, and S. Schulenberg (2006): “Quantitative Analysis of Coordinated
Effects,”.
Leece, D. (2006): “Testing a Theoretical Model of Mortgage Demand on UK Data,” Applied Economics, 38,
2037 - 2051.
Ling, D. C., and G. A. McGill (1998): “Evidence on the Demand for Mortgage Debt by Owner-Occupants.,”
Journal of Urban Economics, 44(3), 391 - 414.
Low, S., M. Sebag-Montefiore, and A. Dübel (2003): “Study on the Financial Integration of European
Mortgage Markets,” Discussion paper, Mercer Oliver Wyman.
Manski, C. F., and S. Lerman (1977): “The Estimation of Choice Probabilities from Choice Based Samples,”
Econometrica, 45(8), 1977 - 1988.
Manski, C. F., and D. L. McFadden (1981): Structural Analysis of Discrete Data and Econometric
Applications. Cambridge: The MIT Press.
McFadden, D. (1974): “Conditional Logit Analysis of Qualitative Choice Behavior,” in Frontiers in
econometrics, ed. by P. Zarembka, chap. 4, pp. 105-142. Academic Press: New Yo.
_________ (1978): “Modeling the choice of residential location,” in Spatial interaction theory and planning
models, ed. by A. Karlkvist, L. Lundkvist, F. Snickars, and J. Weibull, pp. 75-96. North-Holland, Amsterdam.
_________ (1981): “Structural Discrete Probability Models Derived from Theories of Choice,” in Structural
Analysis of Discrete Data and Econometric Applications, ed. by C. F. Manski, and D. L. McFadden,
chap. 5. Cambridge: The MIT Press.
_________ (1987): “Regression Based Specification Tests for the Multinomial Logit Model,” Journal of
Econometrics, 34, 63 - 82.
McFadden, D., and K. Train (2000): “Mixed MNL Models for Discrete Response,” Journal of Applied
Econometrics, 15(5), 447 - 470.
Mitchell, K., and N. M. Onvural (1996): “Economies of Scale and Scope at Large Commercial Banks: Evidence
from the Fourier Flexible Functional Form.,” Journal of Money, Credit and Banking, 28(2), 178-99.
Molnár, J., M. Nagy, and C. Horváth (2006): “A Structural Empirical Analysis of Retail Banking Competition:
The Case of Hungary,” .
Moriizumi, Y. (2000): “Current Wealth, Housing Purchase, and Private Housing Loan Demand in Japan.,”
Journal of Real Estate Finance and Economics, 21(1), 65 - 86.
Nakane, M. I., L. S. Alencar, and F. Kanczuk (2005): “Demand for bank services and market power in Brazilian
banking,” Discussion paper.
Neven, D., and L.-H. Röller (1999): “An Aggregate Structural Model of Competition in the European Banking
Industry,” International Journal of Industrial Organization, 17, 1059-74.
Nevo, A. (2000): “Mergers with Differentiated Products: The Case of the Ready-to-Eat Cereal Industry,” RAND
Journal of Economics, 31(3), 395 - 421.
_________ (2001): “Measuring Market Power in the Ready-to-Eat Cereal Industry,” Econometrica, 69(2), 307-
42.
Paiella, M., and F. P. Pozzolo (2006): “Mortgage Lending and the Choice Between Fixed and Adjustable Rate,”.
Pereira, P., and T. Ribeiro (2007a): “The Impact on Broadband Access to the Internet of the Dual Ownership of
Telephone and Cable Networks,” Discussion Paper Discussion Paper Working Paper No. 17, Autoridade
da Concorrência.
_________ (2007b): “Mergers in Mobile Telephony,” .
Peristiani, S. (1997): “Do Mergers Improve the X-Efficincy and Scale Efficiency of U.S. Banks? Evidence from
the 1980s,” Journal of Money, Credit and Banking, 29(3), 326 - 337.
Pinkse, J., and M. Slade (2004): “Mergers, Brand Competition, and the Price of a Pint,” European Economic
Review, 48(3), 617-43.
Prager, R. A., and T. H. Hannan (1998): “Do Substantial Horizontal Mergers Generate Significant Price Effects?
Evidence from the Banking Industry.,” Journal of Industrial Economics, 46(4), 433-52.
Rhoades, S. A. (1998): “The Efficiency Effects of Bank Mergers: An Overview of Case Studies of Nine
Mergers.,” Journal of Banking and Finance, 22(3), 273-91.
Sapienza, P. (2002): “The Effects of Banking Mergers on Loan Contracts,” Journal of Finance, 57(1), 329-67.
Shaffer, S. (1989): “Competition in the U.S. Banking Industry,” Economics Letters, 29 (4), 321-23.
_________ (1993): “A Test of Competition in Canadian Banking,” Journal of Money, Credit and Banking,
25(1), 49 - 61.
Slade, M. (2004): “Market Power and Joint Dominace in U.K. Brewing,” The Journal of Industrial Economics,
LII, 133 - 163.
Small, K., and H. Rosen (1981): “Applied Welfare Economics with Discrete Choice Models,” Econometrica,
49(1), 105-30.
Srinivasan, A., and L. Wall (1992): “Cost Savings Associated with Bank Mergers,” Discussion Paper Discussion
Paper Working Paper 92-2, Federal Reserve Bank of Atlanta.
Train, K. E., D. L. McFadden, and M. Ben-Akiva (1987): “The demand for local telephone service: a fully
discrete model of residential calling patterns and service choices,” RAND Journal of Economics, 18(1),
109 - 123.
Wahba, G. (1990): “Spline Models for Observational Data,” Series in Applied Mathematics, 59.
Wang, J. C. (2003): “Merger-related cost savings in the production of bank services.,” Discussion Paper 03-8.
Session 2
THE ECONOMIC EFFECTS OF IMPROVING INVESTOR RIGHTS IN PORTUGAL*
Rui Castro†
Université de Montréal
March 2008
Abstract
The Portuguese economy has performed remarkably well since joining the EU in 1986. Output per
worker grew at an annual rate of 2.25%. The relative price of investment has declined. Real investment has
increased compared to output, in part fuelled by an increase in capital inflows. At the same time, resource
allocation seems to have improved as well: firm-level data shows a significant decline in the dispersion of
labor productivity and size across firms. This paper argues that improvements in outside investor rights that
have taken place since Portugal joined the EU is a prime candidate to explain this set of facts.
Key Words: Macroeconomics, Investment rate, Relative prices, Resource misallocation, Investor protection, Optimal contracts, Portugal.
JEL Codes: E22, F43, G32, G38, O16, O17, O41.
*
This paper is being prepared for the Bank of Portugal Conference on “Desenvolvimento Económico Português no Espaço Europeu." I
would like to thank the members of the Scientific Committee, as well as Sílvia Gonçalves, António Antunes, Gian Luca Clementi, and Pedro
Portugal for comments and suggestions. A very special thanks to Gian Luca Clementi, for many many insightful conversations throughout
the years. This paper relies very heavily on joint research. All remaining errors are my own responsibility. Parts of this paper were completed
while I was visiting the Bank of Portugal and the Department of Economics at Stern School of Business, NYU. I wish to thank their very
generous hospitality.
†
Department of Economics and CIREQ, Université de Montréal. Email: rui.castro@umontreal.ca Web:
http://www.fas.umontreal.ca/sceco/castroru.
126 Rui Castro
1. Introduction
The changes to the Portuguese economy since joining the European Union in 1986 (then the European
Economic Community) have been dramatic.
First, the Portuguese economy has performed remarkably well since 1986 compared to the rest of the EU, at
least until recently. The Portuguese economy grew faster than the EU average, nearly doubling the level of
income in 20 years. Labor productivity increased significantly during the period. The price of investment relative
to the price of consumption declined. Investment rates have increased in real terms, although more modestly.
This was in part fuelled by a larger inflow of foreign capital.
This paper also reports evidence on the degree of resource misallocation across Portuguese firms. Since
joining the EU, there has been a significant decline in the dispersion of firm-level productivity, and in the
dispersion of firm size. This suggests an improvement in micro-level resource allocation in Portugal.
Second, joining the EU led to many important reforms. This paper puts particular emphasis on institutional
reforms, namely those affecting corporate governance and investor protection. Investor protection is the term
given by La Porta, Lopez-de Silanes, Shleifer, and Vishny (1998) to the provisions of the commercial law, as
well as its enforcement, that protect outside investors against the expropriation of their returns by managers and
other firm insiders. Before joining the EU, the Portuguese commercial code dated from 1888. However, the new
company law enacted in 1986 has become much more protective of outside investors' rights. In 1991, a modern
securities law was introduced. The Portuguese bankruptcy laws were also significantly revised in 1993. At
around the same time, transparency was greatly increased with the creation of a centralized system of
information about credit risks. Arguably there have been improvements in law enforcement as well.
Can the improved performance of the Portuguese economy since joining the EU be at least in part attributed
to this improvement in investor protection? The main goal of this paper is to answer this question.
The strategy to answer this question is to rely on a model for the Portuguese economy. The model contains
an explicit description of agents' motives and constraints, as well as a description of how they interact. The
model may then be used as a laboratory, where agents are subject to some of the policy changes that the
Portuguese economy has faced since joining the EU. Chiefly among them is the improvement in investor
protection. Another potentially important policy change is the movement started in 1987 towards the creation of
the Single European Market, by abolishing the restrictions on the intra-EU movements of people, goods,
services, and capital. The model traces individual and economywide responses to the different policy changes,
which can be compared with the evidence. The main goal of this paper is to see which policies can account, in a
The Economic Effects of Improving Investor Rights in Portugal 127
qualitative sense, simultaneously for the key macro and micro-level facts that characterize the Portuguese
economy since the mid-1980's.
The model economy is the one developed in Castro, Clementi, and MacDonald (2004) and Castro, Clementi,
and MacDonald (2008). It is a simple extension of the standard Overlapping Generations growth model featuring
imperfect investor protection. The model predicts that an economy which provides better legal protection to
outside investors is able to channel more saving to the entrepreneurial sector, including foreign saving, and to do
it in a more efficient way. When funds are channelled to firms more efficiently, this leads to less distortions at
the micro level, resulting in a better aggregate performance.
The main conclusion of this exercise is that the improvement in investor protection is a prime candidate to
explaining the post-1986 Portuguese evidence. Among the alternative policy changes considered, no other is
consistent with all the facts jointly. The liberalization of capital movements alone would have implied a
significant outflow of capital away from the Portuguese economy. This is because the pre-EU standard of
investor protection in Portugal was much poorer compared with the rest of Europe, and so Portuguese saving
would have fled. Also, policies leading to improvements in either aggregate or investment-specific productivities
cannot account for the improvement in micro-level resource allocation. Productivity improvements simply shift
the production possibilities frontier outward, without affecting the wedge between private and social returns.
This paper is related to a recent literature that studies the economic implications of imperfect investor
protection. Apart from Castro, Clementi, and MacDonald (2004) and Castro, Clementi, and MacDonald (2008),
other papers in this literature are Amaral and Quintin (2007), Antunes, Cavalcanti, and Villamil (2008), Erosa
and Hidalgo-Cabrillana (2007), and Albuquerque and Wang (2008). All these papers are mostly concerned in
explaining the cross-country variation in economic outcomes. A few, like Albuquerque and Wang (2008) and
Antunes, Cavalcanti, and Villamil (2008), consider steady-state effects of hypothetical institutional reforms,
namely those achieving the perfect protection benchmark. I am not aware of any study attempting to achieve
something similar to this paper. That is, tie down an economic model to a specific example of institutional
reform (in this case Portugal), trace down the economic dynamics following the reform, and compare the
implications of the model with the data for a broad range of variables.1
1. Lopez-de Silanes (2002) describes some investor protection reforms in Latin America since the mid-1990s. See also Tavares (2004) for a
model-free strategy to gauging the potential effect of further investor protection reform on Portuguese economic growth. A model-free
approach has the disadvantage of not taking into consideration neither how agents respond to reforms, nor equilibrium effects. Moreover, a
model is typically needed in order to make normative assessments. Finally, any such approach forces the researcher to rely on available
indicators of investor protection, such as the ones provided by La Porta, Lopez-de Silanes, Shleifer, and Vishny (1998), to measure the
benefits and the costs of reform. This is problematic since these indicators are ordinal in nature.
128 Rui Castro
The remainder of this paper is organized as follows. Section 2 describes some facts about the performance of
the Portuguese economy since joining the EU, both macro and micro-level facts, and discusses the main policy
changes that have taken place. Section 3 describes the model, and Section 4 calibrates it to the Portuguese
economy. Section 5 describes the policy experiments and presents the results. Section 6 concludes. Appendices
A and B present some details about the data and the model's computation, respectively.
2. Portugal, 1978-2006
This section presents some empirical observations about the performance of the Portuguese economy since
1978. The focus is both on macro-level and on micro-level data. Although macro-level data extends from 1978
until 2006, micro-level data only spans the 1981-2005 period. All the data are yearly. See Appendix A for further
details.
2.1. Macro Data
This section presents the evolution of some key macroeconomic variables for the Portuguese economy
during the 1978-2006 period.
Figure 1 plots the time-series of Portugal's real GDP per worker, that is real GDP divided by the total labor
force.2 Portugal's real GDP per worker grew at an average annual growth rate of 2.24 percent between 1978 and
2006. This growth was not evenly distributed through time. Growth was moderate at 0.93 percent per year in the
years prior to joining the EU. Between 1986 and 2000 growth accelerated significantly to 2.59 percent per year.
However, since 2000 growth has been slightly negative at -0.45 percent per year.
2. Real variables are in millions of 2000 chained euros.

3.2
Real GDP per Worker (log)
2.8 2.6 3
1980 1985 1990 1995 2000 2005

year
0.93% (1978−1986) 2.59% (1986−2000)

−0.45% (2000−2006)
Source: Banco de Portugal
Figure 1: Real GDP per worker
Figure 2 plots Portugal's investment rate. The left panel reports the ratio of nominal gross fixed capital
formation over nominal GDP - nominal investment rate for short. The nominal investment rate declined at an
average rate of about 1 percent per year. Its evolution through time was very uneven, although much of the
variation can be attributed to the high cyclical variability of investment. The right panel of Figure 2 plots the
ratio of real gross fixed capital formation to real GDP - real investment rate for short.3 Over the whole 1978-
2006 period, the real investment rate in Portugal rose at an yearly average of 0.74 percent. Growth was faster
between 1986 and 2000, at about 1.4 percent per year.
Figure 2 can be interpreted as follows. The left panel tells us the expenditure effort that went into domestic
investment: for each euro of income, how many cents were spent into investment. In Portugal, less and less
resources have been devoted to investment. Is this bad news? Does it mean actual investment in Portugal has
suffered? As the right panel of Figure 2 shows, the answer is no. Real investment did actually grow faster than
real output. The reason is that investment goods in Portugal have become cheaper over time, relative to
consumption goods. Therefore, even if Portugal is putting less effort into investing, it is actually getting more out
of it.
3. Only the nominal investment rate has a share interpretation. Although the real investment rate does not have a share interpretation, it still
provides useful information about the relative growth rates of real investment and real output. See Whelan (2002).
130 Rui Castro
−1.3
−.8
−1.4
−1
Nominal Investment Rate (log)
Real Investment Rate (log)

−1.2
−1.5
−1.4
−1.6
−1.6
−1.7
1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005
year year
−5.98% (1978−86) −0.51% (1986−00) −3.74% (1978−86) 1.40% (1986−00)

−4.43% (2000−06) −3.68% (2000−06)
Source: Banco de Portugal Source: Banco de Portugal
Figure 2: Investment rate

.3
Relative Price of Investment (log)
0 .1 −.1.2
1980 1985 1990 1995 2000 2005

year
−0.06% (1978−1986) −1.37% (1986−2000)

−2.06% (2000−2004)
Source: Penn World Tables 6.2
Figure 3: Relative price of investment

Figure 3 plots precisely the price of investment relative to the price of consumption, from 1978 until 2004.
The data is from Heston, Summers, and Aten’s (2006) version 6.2 of the Penn World Tables. The relative price
of investment has indeed declined throughout the sample period, at an average annual rate of 1.12 percent. Prior
to 1986, the relative price of investment was essentially flat. The decline has been significant since 1986.
.05
0
Nominal Trade Balance relative to GDP
−.05
Real Trade Balance relative to GDP

0
−.1
−.05
−.15
−.1
−.2 −.25
−.15
1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005
year year
1978−86 1986−00 1978−86 1986−00

2000−04 2000−04
Source: Banco de Portugal Source: Banco de Portugal
Figure 4: Trade balance relative to GDP
Figure 4 plots the trade balance relative to GDP. The left panel reports the ratio of the nominal trade balance
to nominal output - nominal trade balance over output for short. The trade balance is total exports minus total
imports of goods and services. The nominal trade balance over output has increased around 1986, and has
displayed a slight negative trend since then. It is hard to attribute any long-term significance to the plunge in net
exports in the early to mid 1980's, given that Portugal was in recession back then. Since 1986, Portugal has been
spending a slightly higher fraction of its income on foreign goods. The right panel of Figure 4 displays the real
trade balance over real GDP - real trade balance over output, for short.4 In real terms, net exports have declined
significantly relative to output since 1986. One way to interpret the two panels is as follows. The left panel tells
us that the Portuguese economy has been spending slightly more on imports, in net terms. The right panel tells us
4. The comment of footnote 3 applies here as well.

132 Rui Castro
that, in spite of this slightly higher expenditure, the real value of those imports has been increasing a lot faster
than real GDP. This is, once again, due to the behavior of relative investment prices.
2.2. Micro Data
I now report some facts about the micro-level allocation of resources in Portugal. The data is from the
Portuguese Ministry of Employment's Quadros do Pessoal, and spans the 1981-2004 period.
Figure 5 plots, for the whole economy, the evolution of the firm-level dispersion in labor productivity and in
firm size. The measure of dispersion is the standard deviation of the logarithm (both series normalized to take on
the value of 1 in the first period).
1 .9
stddev log
.8 .7
1980 1985 1990 1995 2000 2005

year
Dispersion in firm−level productivity Dispersion in firm size

Source: Quadros do Pessoal
Figure 5: Dispersion in firm-level production and size (all sectors)
Size is the total number of employees. Measuring labor productivity is problematic, due to absence of data
on firm-level value-added. My measure of labor productivity is real sales per employee. As argued in Appendix
A.1, under some assumptions value added per employee is proportional to sales per employee. The constant of
proportionality is a technological parameter (the share of intermediate goods in production) which may be
assumed to be constant across firms, but only within narrowly-defined industries. This is obviously not the case
when comparing firms across all the sectors in the economy, as in the solid line of Figure 5. The reader should be
cautious about interpreting it - it is nevertheless comforting that the behavior of dispersion in firm size, which is
not subject to this concern, displays a consistent picture. Dispersion in both firm size and productivity has been
declining steadily since 1981. The decline picked up in the early to mid 1990s, and slowed down afterwards.
Is the evidence of Figure 5 accounted for by a decline in dispersion within sectors, or across sectors?
Although I won't provide a conclusive answer to this question here, Figure 6-Figure 9 present some
disaggregated evidence, at the 2-digit level.5 The different figure captions indicate how the 2-digit sectors are
grouped in terms of 1-digit categories. Table 3 in Appendix a.2 provides the full set of sector names associated
with each sector code.
01 02
1 1.2 1.4
1
.8
.6
.8
.6
.4
05 13
1.4
3
1.2
2
1
1
.8
1981 1986 2000 2005 1981 1986 2000 2005

year
Graphs by 2−digit sector (CAE)
Figure 6: Dispersion in firm-level productivity and size (agriculture, animal production, hunting forestry
(01,02), fishing (05), and extractive industries (13,14))
Visual inspection of these figures reveals that the same pattern that was observed in the aggregate tends to be
observed at the 2-digit sectoral level as well. There are a few exceptions though: Tobacco Manufacturing (16),
Radio, TV, and Communication Material Manufacturing (32), Automobile Manufacturing (34), Recycling (37),
essentially all sectors in Transportation, Storage, and Communications (60-64) sector, and essentially all sectors
in Real Estate, Rentals, Service Provision to Firms (70-74).
5. Ideally one would like to work with a finer classification. Unfortunately, the number of firms in several sectors rapidly becomes too small.
134 Rui Castro
15 16 17 18 19
.6 .8 1 1.2
.6 .8 1 1.2
1 1.2
.4 .6 .8 1
0 1 2 3 4
.8
20 .6 .8 1 1.2 21 22 24 25
.7 .8 .9 11.1
.8 .9 1 1.1
1
1
.8
.6 .8
.6
26 27 28 29 31
.7 .8 .9 11.1
.6 .8 1 1.2
.8 .9 1 1.1
.7.8.9 11.1
.6 .7 .8 .9 1
32 33 34 35 36
.8 11.21.41.6
.6 .8 1 1.2
.6 .8 1 1.2
.7 .8 .9 11.1
1 1.21.41.6
19811986 20002005 19811986 20002005 19811986 20002005
37 45
0 .5 1 1.5
.6.7.8.9 1
19811986 20002005 19811986 20002005
year
Figure 7: Dispersion in firm-level productivity and size (manufacturing (15-37) and construction (45))
In a few other sectors, for example Retail and Wholesale Trade (50-52) and Accommodation, Restaurants
and Kindred Activities (55), the decline in productivity dispersion is marked, however the decline in size
dispersion is either very modest or non-existent.
Sectoral heterogeneity is clearly not the whole story behind the pattern in Figure 5. That is, part of the
improvement in resource allocation appears to have taken place within 2-digit sectors. However, for the most
part, this paper will abstract from within-sector misallocation. The analysis will proceed as if all firms within a
sector were identical. My main focus will be on misallocation that occurs between high risk sectors (e.g.
investment good sectors like machinery manufacturing), and low risk sectors (e.g. consumption good sectors like
apparel manufacturing). Section 5.5 will, however, discuss a specific setting where the within-sector
improvement in resource allocation displayed in Figure 6-9 is particularly relevant.
50 51 52
.8 .9 1 1.1
.7 .8 .9 1
.7 .8 .9 1
55 60 61
.6 .8 1 1.2
1 1.5 2 2.5
.7 .8 .9 1
62 63 64
.7 .8 .9 1 1.1
.5 1 1.5 2
0 1 2 3
1981 1986 2000 2005 1981 1986 2000 2005 1981 1986 2000 2005
year
Figure 8: Dispersion in firm-level productivity and size (retail and wholesale trade (50-52),
accommodation, restaurants and kindred activities (55), and transportation, storage, and
communications (60-64))
2.3. Policy Reforms
This section provides a brief chronology of the main policy changes that took place in Portugal during the
1978-2006 period. I will focus first on the main improvements in corporate governance and investor protection,
and then on other types of policy change likely to be important.
There were major improvements in all the main areas related to corporate governance and investor
protection: Company Law, Securities Law, Bankruptcy Law, as well as in law enforcement. Most of these major
reforms were concentrated on a short period after Portugal joined the EU.
136 Rui Castro
70 71 72
1 1.2
.6 .8 1 1.2
.6 .8 1 1.2
.8
73 74 92
.8 1 1.21.4
.7 .8 .9 11.1
.6 .8 11.21.4
1981 1986 2000 2005 1981 1986 2000 2005
93
.6 .8 1 1.2
1981 1986 2000 2005

year
Figure 9: Dispersion in firm-level productivity and size (real estate, rentals, service provision to firms
(70-74), and various services (92,93))
• 1986: Major revision of the existing Commercial Code of 1888, with the introduction of a new
Company Law (Código das Sociedades Comerciais). The code became closer to the rest of Europe,
and significant improvements were made in terms of mandatory accounting practices and investor
rights. The code has been further revised on a regular basis since 1986.
• 1991: Creation of an independent Securities Commission (CMVM - Comissão do Mercado de

Valores Mobiliários) in charge of regulating and supervising financial markets. The basic
regulatory framework (Securities Law - Código do Mercado de Valores Mobiliários) was
introduced in 1991, and significantly revised in 2000. The Bank of Portugal continued being the
main regulator and supervisor of financial intermediaries, but a new Banking Law (Regime Geral
das Instituições de Crédito e Sociedades Financeiras) was enacted in 1992.
• 1993: New corporate Bankruptcy and Reorganization Law (Código da Insolvência e Recuperação
de Empresas) enacted, and subsequently revised in 1998. A further comprehensive reform took
place in 2004, aimed at speeding up the bankruptcy process and dealing with poor enforcement.
• 1993: Creation of a centralized database of the individual credit positions and credit ratings of
every household and firm in Portugal (Central de Responsabilidades de Crédito). Information about
firms started to be gathered in 1978, but the scope became much broader starting in 1993. Since
then it covers essentially the universe of borrowers from Portuguese financial institutions. The
database is managed by the Bank of Portugal, and the information is provided to lending financial
institutions at their request.
• There were gradual improvements in law enforcement as well, for which there is no specific
turning-point date. In spite of this progress, issues of poor law enforcement are still the Achilles
heel of overall institutional development in Portugal, particularly the slow speed of the judicial
system.
Other major policy changes have taken place in Portugal during this period. Like the improvements in
investor protection, most of these reforms took place soon after Portugal joined the EU. They are also potential
candidates to explain the performance of the Portuguese economy since 1986.
• 1987: The Single European Act is put to practice, preparing the ground for the Single European
Market. Like its EU partners, Portugal began abolishing the barriers to the free intra-EU movement
of people, capital, goods, and services. The Single European Market finally took shape in 1993.
• 1988: Beginning of large deregulation and privatization wave, with the approval of the Privatization
Law. Large impact on the financial intermediation industry, heavily regulated and largely state-
owned prior to this date.
• 1999: The Portuguese currency fixes its exchange rate against the Euro. In 2002, the Portuguese
currency is replaced by the Euro.6
The model presented in the next section will be able to sort out the effects of some of these other policies
from the effects of improving investor protection. Section 5 will compare the effects the different policies might
have had in the Portuguese economy since 1986.
6. Portugal joined the European Exchange Rate Mechanism in 1992. Participants to this arrangement were supposed to maintain a relatively
stable value of their currency against each other. As is well-known, the success was mitigated. A period of instability during 1992 and 1993
lead to a relaxation of the commitment towards exchange rate stability.
138 Rui Castro
3. Model
The benchmark model is the one developed in Castro, Clementi, and MacDonald (2008). It is a simple
extension of the standard two-period Overlapping Generations model of capital accumulation. It features two-
sectors, consumption and investment, and institutions that imperfectly protect outside investor rights.
3.1. Preferences
Population is normalized to one. Each period a new generation of individuals is born. An individual's utility
from time t consumption is
c 1"@
t "1
uc t ,
1"@
where @ 0 is the coefficient of relative risk-aversion. Future utility is discounted at rate * 0 .

Individuals are ex-ante identical. They are born with no resources. They are born, however, with one unit of
time and with managerial talent. Young individuals employ their unit of time as managers of a firm. For this
reason we call young individuals entrepreneurs.
3.2. Technology
After being born, an entrepreneur can decide whether to manage a firm that produces consumption goods
( C ), or a firm that produces investment goods ( I ). The production of good j = C , I is done according to
y jt z jt k )jt ,
where k jt is capital, 0 ) 1 is the share of capital in production, and z jt ¡0, . is the entrepreneur-
specific productivity for producing good j . We assume lnz jt is normally distributed with mean 6 j and standard
deviation 1 j . We also call z jt the quality of the entrepreneur's investment project. The two technologies differ
only in the two parameters governing the productivity distribution.
3.3. Lending Contracts
After deciding in which sector to produce, an entrepreneur needs to borrow funds from financial
intermediaries, to be able to get capital for production. Once capital is installed, the entrepreneur produces,
receives profits, and consumes and saves out of these profits. The saving is used towards the purchase of capital,
which is held until next period and then rented out to the next generation of entrepreneurs. For this reason, we
call old individuals capitalists.
A central element of the model is the interaction between entrepreneurs and financial intermediaries. There
are two important ingredients.
3.3.1. Information
First, when a financial intermediary lends funds to an entrepreneur, neither of the two knows the quality of
the investment project. After funds are lent, the project quality becomes known. However, it is private
information of the entrepreneur. This creates an incentive for the entrepreneur to under-represent the true quality
of his investment project, since lower-quality projects are required to pay back lower returns to the intermediary.
The entrepreneur stands to profit from the difference between the return that he should have paid, and the return
he actually pays - an activity that we call stealing.
3.3.2. Institutions
Second, how easy it is to steal depends upon the institutions governing outside investor rights. Countries
with good institutions make stealing very inefficient - most of the resources stolen get lost in the process of
stealing. In countries with poor institutions, instead, the return from stealing is high.
3.3.3. Intermediation Industry
We assume a perfectly competitive intermediation industry with free entry. As such, the lending contracts
offered by intermediaries will be constrained-efficient. That is, they will provide the maximum expected utility
for entrepreneurs, subject to being not only resource but also incentive feasible.
140 Rui Castro
3.3.4. Optimal Consumption-Saving Behavior
To formally characterize the lending contract, we first need to consider an entrepreneur's consumption-
saving decision. Let vm t , r t1 denote the indirect utility of an agent born at time t , conditional on having
received an income m t and on facing an interest rate r t1 . Then,
vm t , r t1 q u¡m t " sm t , r t1 ¢ *u¡1 r t1 sm t , r t1 ¢,
where the optimal saving function sm t , r t1 is
sm t , r t1 q arg max

s
£um t " s *u¡1 r t1 s¢¤.
3.3.5. Optimal Contracting
We model financial intermediaries as if they directly rented out the services of capital to entrepreneurs, on
behalf of capitalists. We also model them as if they collected production from entrepreneurs. Intermediaries are
then responsible for providing entrepreneurs with an income which is already net of the total loan repayment.7
Due to perfect competition and free entry, there won't be any scope for cross-subsidization between
contracts offered to different sectors. With cross-subsidization, any financial intermediary could steal the market
by offering better terms to the entrepreneurs operating in the sector where current contracts are making losses.
Without loss of generality, we can think of a single intermediary dealing with every entrepreneur operating in the
same sector.
The optimal lending contract offered to an entrepreneur operating in sector j at time t then solves8
max
k jt u0,A jtz u0
; vAjt z , rt1 f j z dz
(P1)
subject to
7. An alternative would be to suppose entrepreneurs receive capital, and then pay back a return to intermediaries. Although more realistic, it
turns out this formulation is formally equivalent to ours. Moreover, our formulations leads to a simpler characterization of the problem.
8. Unless otherwise specified, integration is over the whole domain [0,+∞).
v¡A jt z , r t1 ¢ u v¡A jt z U 8p jt z " z U k )jt , r t1 ¢ for all z u z U (1)
; Ajt z f j z dz p jt z j k )jt " rt - p It k jt , (2)
where p jt is the relative price of good j in terms of consumption. Hence p Ct 1 , and we denote the
relative price of investment simply by p It p t .
Upon observing the true z , an entrepreneur must report what his productivity was to the intermediary. This
announcement, let's call it z , may or may not correspond to z . Since the income transfer provided by the
intermediary, A jt z , must be based upon the entrepreneur's announcement, it may be in his the best interest to
misreport z.
The optimal lending contract is a capital advance k jt , and a schedule of income transfers A jt z
conditional on reported project quality, which maximize the entrepreneur's expected lifetime utility subject to
two constraints.
Equation (2) is a resource constraint. It requires that total transfers paid out by the intermediary must equal
total output obtained from entrepreneurs net of rental payments to owners of capital. Because the risk
entrepreneurs face is purely idiosyncratic, a law of large numbers applies and there is no uncertainty about (2).
Equation (1) is an incentive-compatibility constraint. It follows from the Revelation Principle, which allows
us to restrict attention to contracts that induce truthful revelation. (1) ensures that if an entrepreneur receives
z u z U , he is better-off by announcing z than misreporting a lower productivity level z U .9 In the latter case,
the entrepreneur would only receive an income A jt z U . However, he would also be required to surrender just
p jt z U k )jt back to the intermediary. The difference p jt z " z U k )jt is the amount of resources stolen away from
the intermediary. The entrepreneur only gets to enjoy a fraction 8 ¡0, 1¢ of this amount. The remainder, we
assume, is a deadweight loss. Countries with good institutions have lower 8 's, with 8 0 corresponding to
perfect investor protection. This is a situation in which stealing is never profitable for an entrepreneur.
U
9. We do not need to impose an analogous incentive compatibility for z z . That is, no entrepreneur would ever report a higher
productivity than the true one. This would require surrendering a higher amount of resources than the one the entrepreneur has, something
which would not be feasible.
142 Rui Castro
The parameter 8 therefore captures the level of investor protection, or more generally the quality of
institutions. It is a deep parameter of the model, like preference and technology parameters. It also has attached
to it a very specific interpretation: it measures the loss in resources incurred by borrowers in credit relationships,
should they decide to steal from lenders. Factors that affect 8 are reforms of the provisions of the commercial
code that are protective of outside investor rights, either shareholders or creditors. Also improvements in law
enforcement, including better monitoring technologies, will affect 8 . The bottom line is that the model provides
a foundation for interpreting 8 as the level of investor protection.
3.4. International Trade
We assume only investment goods are traded internationally. Since consumption goods are not traded, there
is no scope for intratemporal trade between consumption and investment goods. Each country faces the autarkic
relative price of investment.
All trade is intertemporal, each country simply engages in borrowing or lending with the rest of the world.
The open economy is small. Let K St be the aggregate capital supplied by the residents, and K Dt be the
domestic aggregate demand for capital. Then
Bt q p t K St " K Dt
is the net current outflow of capital towards the rest of the world, measured in consumption. In other words, Bt
equals the small open economy's net investment abroad, and "Bt is net foreign debt (when Bt 0 there is
foreign investment in the small open economy). Let r ' be the constant world interest rate. The trade balance is
TBt Bt1 " 1 r ' Bt .
We concentrate on a scenario of limited international capital mobility. An economy with net foreign
investment Bt incurs trading costs in units of capital. Measured in consumption, these costs are
2
I Bt Yt ,
Yt
for I u 0 , where Yt is economy's GDP, in consumption units. Letting N t ¡0, 1¢ denote the
equilibrium fraction of entrepreneurs that produce investment goods at time t then
Yt p t N t z I k )It 1 " N t z C k )Ct .
Trading costs are a convex function of the size of net investment relative to output. They are a stand-in for
frictions such as the risk of sovereign default or restrictions to currency conversion. Our specification
accommodates any degree of international capital mobility, ranging from fully open ( I 0 ) to fully closed
economy ( I . ). Trading costs induce a wedge between the world interest rate and the domestic interest rate
( r t ), which depends not only on I but also on equilibrium variables. Countries with different institutions will
thus have different equilibrium interest rates.
3.5. Competitive Equilibrium
We now turn to the formal definition of the competitive equilibrium for this economy.
Definition 1. Given an initial aggregate supply of capital K S0 0 , a competitive equilibrium is a non-negative
consumption level of the initial old c o0 and sequences of young and old agents' non-negative consumption
y .
allocations £c jt z ¤ .t0 and £c ojt z ¤ .t1 , contracts £k jt , A jt z ¤ t0 , measures of entrepreneurs in the
investment good sector £N t ¤ .t0 , aggregate capital demand £K Dt ¤ .t0 , aggregate capital supply £K St ¤ .t0 ,
relative investment prices £p t ¤ .t0 , and domestic interest rates £r t ¤ .t0 , such that
1. c o0 p 0 K S0 1 r 0 ;
y
2. c jt z A jt z " sA jt z , r t1 and c oj,t1 z sA jt z , r t1 1 r t1 , for j C, I ;
3. £k jt , A jt z ¤ solve problem (P1) for j = C, I ;

4. Young individuals are indifferent between the two sectors:
; vAIt z , rt1 f I z dz ; vACt z , r t1 f C z dz;

(3)
144 Rui Castro
5. The aggregate supply of capital equals aggregate savings:
p t K St1 N t ; sA It z , r t1 f I z dz 1 " N t ; sA Ct z , r t1 f C z dz;
(4)
6. Aggregate consumption equals the production of consumption goods:
N t"1 ; c oIt z f I z dz N t ; c It z f I z dz

y
1 " N t"1 ; c oCt z f C z dz 1 " N t ; c Ct z f C z dz 1 " N t z C k )Ct ;
y
(5)
7. The market for capital clears:
K Dt N t k It 1 " N t k Ct ;
(6)
8. The aggregate profit from accessing world credit markets equals the aggregate cost:
2
r ' " r t Bt I Bt Yt .
Yt (7)
It is worth elaborating a little on (7). Suppose r t r ' . In this case the residents of the small open
economy wish to invest their capital abroad, so that Bt 0 . The term r ' " r t Bt 0 is the aggregate
profit from investing Bt abroad. In equilibrium, the residents as a whole must have no further interest in
investing abroad. This happens when the aggregate profit equals the aggregate trading costs. An analogous
argument applies to r t r ' , in which case residents must have no further interest in borrowing from abroad.
3.6. Model Solution
This section describes very briefly the solution to the model. See Castro, Clementi, and MacDonald (2008)
for further details.
Because of isoelastic preferences, we can write vA jt z , r t1 uA jt z Cr t1 , up to a constant. It
follows that the solution to (P1) does not depend upon r t1 . Further, it has a particularly simple structure.
Transfers are given by A jt z p jt g j z k )jt for j = C, I , where g j z u 0 is a sector-specific function that
only depends on model parameters. The capital advance in sector j = C , I solves:
p jt r t - )k )"1 if 8 t 8 'j (8)

jt z
j " 8F j
p jt r t - )k )"1
jt 1 " 8 z
j if 8 u 8 'j (9)
where
; u U ¡g j z ¢z j " z f j z dz

Fj q u 0,
; u U ¡g j z ¢f j z dz
8 'j q 1 0, 1 .
) 2
1 1")
e "@1 j
'
Let's begin with condition (8). The first-best capital advance obtains for 8 0 . When 0 8 t 8 j , a
wedge is introduced relative to first-best, governed by F j 0 . The wedge arises because, when investor
protection is imperfect, financial intermediaries must induce entrepreneurs to truthfully reveal their
productivities. This is achieved by forcing entrepreneurs to share part of the firm-level risk. To see this in more
detail, notice that the incentive-constraint (1), which binds for 8 0 , implies
A jt z " Ajt z U 8p jt z " z U k )jt for all z, z U
and so
stdA jt z 8p jt k )jt stdz ,

(10)
where std is the standard deviation. The key point is that the risk faced by entrepreneurs, measured by the
standard of transfers, is increasing in firm size k jt . Hence, advancing an extra unit of capital to entrepreneurs
must be accompanied by a transfer schedule that imposes on them a larger fraction of the idiosyncratic firm risk.
In other words, entrepreneurs receive less insurance against that risk. The term 8F j 0 captures precisely the
utility loss for entrepreneurs of facing this higher risk. Their private marginal gain of employing an extra unit of
capital (right-hand-side of (8)) is lower compared to first-best.
'
Condition (9) arises because, for 8 high enough ( 8 8 j ), spreading transfers further apart eventually
violates the limited liability constraint A jt z u 0 for some z . Intermediaries must further reduce the capital
146 Rui Castro
advance, to be able satisfy this constraint. In this case, the marginal benefit of increasing capital by an additional
unit (again the right-hand-side of (8)) becomes larger than the marginal cost. This implies a further wedge in the
allocation of capital relative to first-best.
Replacing the transfer schedule in (3) and (8)-(9), it is easy to verify that the relative price of investment p
and the relative size of consumption good firms Q q k Ct /k It are both time-invariant. This feature dramatically
simplifies the computation of the economy's transition to the steady-state, starting from K S0 0 .10 Appendix B
provides further details on the computation of the economy's transition path.
4. Calibration
4.1. Exogenous Growth
For the purpose of calibrating the model, it is useful to consider exogenous growth. Let TFP in sector j be
0 jt +t z jt for j I, C , where + 1 is the common gross growth rate and z jt is defined as before.
Since there are enough homogeneity assumptions, growth is easily accommodated in the model. A balanced
growth path exists where the growth rate of aggregate output is +1/1") . As it is standard in the business cycle
literature, one can derive the analogue of the equilibrium conditions of Section 3.5 in terms of detrended
variables.
4.2. Parameter Values
I will adapt the basic calibration methodology of Castro, Clementi, and MacDonald (2008) to the case of
Portugal. The model's parameters are summarized in Tables 1 and 2. The latter contains the policy parameters
which characterize Portugal prior to joining the EU.
β σ α δ γ r* ηC ηI
0.1428 1.5 1/3 0.7099 1.3542 0.4986 0.9136 1.4736
Table 1: Parameter values
10. Castro, Clementi, and MacDonald (2008) concentrate their attention on steady-states. Here I will be interested in the transitional
dynamics as well.
ξ ϕ zI zC
0.1428 1.5 1/3 0.7099
Table 2: Pre-EU policy parameters
Individuals are assumed to have a productive life of 40 years. This implies a 20-year model period. The
relative risk aversion coefficient @ is set to 1.5, a standard value in quantitative analysis. The parameter ) is
1/3, the value considered by other studies of entrepreneurial behavior such as as Burstein and Monge-Naranjo
(2007) and Buera (2003). I set - so that the annual depreciation rate is 6%. The world average annual growth
rate of real GDP per worker in the Penn World Tables is about 2.3%, about the same as in Portugal during the
1978-2006 period. This amounts to a 20-year growth rate of output of 57.58%, implying + 1. 5758 1") .
Next, I need to assign values to the variance parameters 1 I and 1 C . I will calibrate these two parameters
so that the model's implied annual standard deviation of firm growth rates are equal to the corresponding
estimates in the data.11 For a US panel of Compustat firms, Castro, Clementi, and MacDonald (2008) report
those estimates to be 0.0646 for consumption good firms, and 0.1042 for investment good firms. This implies
1 C 0. 0646 20/ 2 and 1 I 0. 1042 20/ 2 .

Castro, Clementi, and MacDonald (2008) show that this model exhibits the same sort of invariance to scale
as the standard neoclassical model, in the sense that relative variables are independent from aggregate TFP. For
that reason, I normalize 6 j "1 2j /2 , so that z C z I 1 .

To assign a level of investor protection for Portugal, I rely on the procedure developed in Castro, Clementi,
and MacDonald (2008). The model implies a one-to-one mapping between investor protection and relative
11. As pointed out by Castro, Clementi, and MacDonald (2008), strictly speaking this model does not produce implications for the growth
rate of firm sales. This is because firms operate for one model's period only. We define the time- t annual detrended average growth rate in
U) ) 2
sector j as ; ; 1/20 log z k j,t1 " log zk jt f j z f j z U dzdz U . In balanced-growth steady state, such statistic is always zero
by construction. Therefore the model's annual standard deviation of the sales growth rate is the square root of
; ; 1/20 ¡logz U " logz¢2 f j z f j z U dzdz U , or 1 j 2 /20 . Our definition of growth rate can be rationalized by assuming that the
technologies are infinitely lived and are passed down from generation to generation.
148 Rui Castro
prices, p8 . It is then in principle possible to infer 8 i for any country i from data on its relative price.12 Castro,
Clementi, and MacDonald (2008) consider the mapping:
p8 i P I /P C i
,
p0 P I /P C 1
where P I /P C i is the relative price data from the Penn World Table. In practice, P I /P C i from the Penn
World Table is country i 's relative investment price levels over the world's. Dividing P I /P C i by the relative
price of a benchmark country ( i 1 ) eliminates the effect of world prices. Choosing the country with the
lowest relative price as the benchmark, and assuming it corresponds to perfect investor protection ( 8 1 0 ), the
above mapping allows one to recover a value of 8 i from a country's relative price of investment.13
Castro, Clementi, and MacDonald (2008) focus on the 1996 Penn World Table, in which case Singapore is
the country with the lowest relative investment price. No matter the year, however, Singapore turns out to always
have a relative price of investment which is among the very lowest reported by the Penn World Tables. This
motivates me to always pick Singapore as the benchmark country. The average relative price of investment in
Portugal relative to Singapore is 1.84 between 1978 and 1985, that is, before joining the EU. The above mapping
implies 8 0. 9402 for Portugal, before joining the EU.
There are three parameters left to be calibrated, * , r ' , and I . I follow Castro, Clementi, and MacDonald
(2008) and select them jointly so that the model, with a 8 for each country computed as described previously,
closely matches the following three moments: (i) a cross-country average PPP-investment rate of 0.146 (the
1996 figure in the Penn World Table), (ii) an average interest rate of 4 percent among the top 5 percent richest
countries (a figure commonly used in the business cycle literature), and (iii) a 4.2 percent interquartile range for
the world interest rate (the figure obtained from Lustig and Verdelhan’s (2007) data during the 1990's).
It follows that the model is not tightly calibrated to the Portuguese data. In particular, other than the initial
relative price of investment, I am not making an effort to match the pre-reform level of any other variable. This
12. This identification is valid under the assumption that 8 is the only source of cross-country differences. If there are cross-country
differences in sectoral productivities as well (actual, not measured productivities as in Hsieh and Klenow (2007)), then relative price
differences also contain information about sectoral productivity differences. I will return to this issue in Section 5.1. Another potential
difficulty would arise if the capital shares in the consumption and the investment sectors were significantly different. However, neither Chari,
Kehoe, and McGrattan (1996) nor Hsieh and Klenow (2007) find this to be the case.
13. As long as the relative price is not too high, otherwise there may not exist a 8t1 consistent with it.
should not be a cause for much concern, given that the main goal of present analysis is mainly qualitative. It is
worth pointing out, however, that the model's quantitative implications are simply suggestive.
5. Policy Experiments
I use the model to conduct four separate policy experiments: (i) an improvement in investor protection, (ii)
an increase in international capital mobility, (iii) an increase in aggregate TFP, and (iv) an increase in
investment-specific TFP. These experiments are meant to mimic some of the main policy changes associated
with Portugal having joined the EU. I consider each policy experiment in isolation. Because different policy
experiments turn out to have different qualitative effects, this exercise will be informative about which among
them is most likely to have generated the data, both the macro and the micro-level observations.
I assume the Portuguese economy was in steady-state until 1985. I will then assume that agents faced a
surprise once-and-for-all policy change in 1986. Needless to say, it is neither true that all policy changes
occurred in 1986 (some took place afterwards, some even before), nor that they came as a surprise when they
were implemented. It is also not true that the Portuguese economy was in steady-state before 1986. These
extreme assumption are, however, a very useful benchmark to get the discussion going.
5.1. Improvement in Investor Protection
Consider the effects of a permanent reduction in 8 . I compute the post-1986 level of 8 like in Section 4.2,
but now to match the 2000-2003 average relative price of investment in Portugal over that of Singapore.14 This
yields 8 U 0. 8018 .15
14. Unfortunately the Penn World Table does not report the price levels for Singapore in 2004.
15. As pointed out in Section 4, identifying a change in 8 with a change in Portugal's relative price of investment relative to Singapore's is
potentially problematic. If, at the same time, Portugal has become relatively more productive at producing investment goods compared to
Singapore, then the change in relative prices would also contain information about the increased productivity. The experiment considered in
Section 5.5 illustrates this problem: changes in investment-specific productivity map into relative prices in a way qualitatively similar to
investor protection. Since independent measures of either sectoral productivities or investor protection are very hard to come by, this appears
to be a very difficult problem to solve. A way out is to check overidentifying restrictions. That is, perhaps relative productivities and investor
protection have a different impact on other endogenous variables. This could be used to infer which change is most likely to have generated
the data. This paper explores this alternative.
150 Rui Castro
Figure 10: Improvement in investor protection

−1.5−1.4−1.3−1.2−1.1
3.2 3.4
3
2.6 2.8
1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005
year year
Real GDP per Worker (log) Nominal Investment Rate (log)

Data Model Data Model
.3
0
−.15 −.1 −.05
.2
.1
0
−.1
1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005
year year
Nominal Trade Balance over GDP Relative price of investment (log)

Data Model Data Model
Figure11: Improvement in investor protection
In the model, the relative price of investment is increasing in 8 . Bad investor protection is particularly
harmful for investing in very risky sectors, like those producing investment goods. In equilibrium, the relative
price of investment must increase in order to encourage both entrepreneurs and capitalists to actually invest in
those sectors. It follows that the sharp drop in the relative price of investment in Portugal since 1986 (Figure 3)
provides indirect support for an improvement in investor protection. The remaining two policy parameters are
held constant at their pre-1986 levels.
Figure 10 shows the paths for different variables starting from the initial steady-state (year 1966). To
emphasize the qualitative implications of the model, all variables are normalized by their initial values, except
for 8 t , Qt and r t . The policy change takes place in 1986, with some adjustment taking place right then. The
economy subsequently converges to the new steady-state.
Figure 10 reveals that convergence is very fast: after 2006, that is after only a couple of model periods, there
is hardly any adjustment left. The economy is essentially on the new balanced-growth path.
152 Rui Castro
The improvement in investor protection removes distortions in the allocation of capital, no matter the sector.
Private returns to capital increase, and so does the real interest rate. However, the allocation of capital improves
relatively more in the investment good sector. For the present parameterization, Qt becomes close to 1, which
means the allocation of capital becomes close to first-best.
The large reduction in distortions in the investment good sector encourages entrepreneurs into producing
those goods, at the expense of less risky consumption goods. N t therefore increases. The share of investment in
total expenditure also increases, in spite of the drop in the relative price. This increase in investment
expenditures is financed in part by an inflow of foreign capital, attracted by the higher domestic return. This is
why the trade balance drops relative to output.16
It is also useful to directly compare the model's implications with the data. This task is somewhat
complicated by the fact that a model's period is 20 years. To go around this issue, assume that all variables
remain constant within a 20-year period. This allows me to infer yearly observations for every variable in the
model. I focus on the 1978-2005 period, matching the data, and consider a partition into three subperiods: (I)
1978-1985, (II) 1986-1999, and (III) 2000-2005. (I) is the pre-reform subperiod, (II) is the subperiod where most
of the reforms occurred, and (III) is the post-reform subperiod. Both in the data and in the model, I compute the
average of every variable in each subperiod.
Figure 11 plots four key variables. For each, the solid line is the original series already discussed in Section
2.1. The squares and the circles are the subperiod averages in the data and in the model, respectively.
Although the improvement in investor protection was not calibrated to match the growth in output per
worker in Portugal since 1986, it is remarkable how close the model comes to matching the data. Qualitatively,
there are two inconsistencies between model and data. First, the nominal investment rate has declined in
Portugal, whereas the model predicts a slight increase. Second, the nominal trade balance to GDP ratio increased
around 1986 and the declined slightly. The model predicts well the slight decline, however it does not feature the
large increase around 1986. Recall, however, that the sharp drop in the trade balance in the early 1980's might
not be very significant from a long-term perspective.
16. Although this is not apparent from, Figure 10 the improvement in investor protection has a permanent effect on the trade balance over
GDP, which goes from a small positive number to a small negative number. In the new steady-state, Portugal becomes a net importer of
capital.
1.5
stddev log firm size
.5 0 1
1980 1985 1990 1995 2000 2005

year
dispersion in firm size

Data Model
Figure 12: Policy comparison - firm size dispersion
Figure 12 turns to the micro-level evidence. I focus on the dispersion in firm-size. The data series in Figure
12 is the unconditional firm size dispersion, the dotted line in Figure 5. The model is qualitatively consistent
with the observed decline in size dispersion, although the magnitude is high compared with the data.
5.2. Increased Openness to International Capital Flows
Consider a permanent drop in ϕ of 5 percent ( ϕ ′ = 20.7768 ). The new value for this parameter was selected
to loosely match the evolution of output per worker in Portugal since 1986. This policy change is meant to
capture the liberalization of intra-EU capital flows, one of the pillars of the Single European Market
implemented in 1993. Figure 13 displays the dynamic response of the economy.
154 Rui Castro
Figure 13: Incresead openness to international capital flows
The first point is that a reduction in barriers to international capital mobility has no effect neither on the
relative investment price, nor on relative firm size. Second, Portugal protected outside investors poorly before
1986, particularly compared to the rest of Europe. Opening to international capital flows in 1986, while keeping
investor protection at its original level, would have given incentives for Portuguese residents to invest their funds
abroad, rather than in Portugal. As result, the nominal investment rate goes down, and the trade balance goes up
relative to output. Capital flows away from Portugal.
Because the it is now less costly for the Portuguese to invest abroad, as less investment goods are lost in the
process, less resources end up being allocated to home-production of investment goods ( N t declines slightly).
Since this is the sector that is most distorted, moving resources away from it raises the aggregate return on
capital ( r t increases slightly).

I'll postpone the direct comparison with the data of this and the remaining policy changes until the very end
of this section.
5.3. Increase in Aggregate TFP
Consider an increase in both 6 I and 6 C , leading to a permanent increase in aggregate TFP of 1 percent
U
( z I z UC 1. 01 ). The new value for this parameter was selected to loosely matches the evolution of output
per worker in Portugal since 1986. This policy change is meant to capture improvements in the access of
Portuguese firms to the state-of-the-art production technology, better management, better access to distribution
points (for example because of better public infrastructure like roads). These improvements are assumed not to
discriminate between investment and consumption good firms. The connection between actual policy changes
and changes in z is obviously less tight than in the previous two cases. Figure 14 displays the dynamic
response of the economy.
The real interest rate increases, reflecting the higher aggregate productivity for given aggregate capital. The
higher return encourages faster capital accumulation. However, as capital gets accumulated and reaches the new
steady-state, the interest rate returns to its initial level. The same is true with the investment rate - just like in the
standard neoclassical growth model. Everything else constant, the higher domestic return would work to
encourages some inflow of capital into the domestic economy. The trade balance over GDP, however, actually
increases. This is because the Portuguese economy still protects its investors badly compared to other European
countries, and its return on capital is still low compared to the rest of Europe. As a result, Portuguese residents
would like invest the extra domestic saving abroad. Like the remaining effects, however, this one is short-lived.
156 Rui Castro
Figure 14: Increase in aggregate TFP

Figure 15: Increase in investment-specific TFP

158 Rui Castro
5.4. Increase in Investment-Specific TFP
Now consider a permanent increase in 6 I , leading to a permanent increase in z I of 5 percent

U
( z I 1. 05 ). Like in the previous two cases, this value was chosen to roughly match the post-1986 evolution
of output. This policy variable captures the same kind of improvements as the change in aggregate TFP. The
main difference is that I now assume the improvements were biased towards investment good firms. The effects
are displayed in Figure 15.
−1.5−1.4−1.3−1.2−1.1
2.6 2.8 3 3.2 3.4
1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005
year year
Real GDP per Worker (log) Data Nominal Investment Rate (log) Data
Investor Protection TFP Investor Protection TFP
Capital Mobility TFP−I Capital Mobility TFP−I
−.1 0 .1 .2 .3
−.15 −.1 −.05 0
1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005
year year
Nominal Trade Balance over GDP Data Relative price of investment (log) Data
Figure 16: Policy comparison
The only qualitative difference between in increase in aggregate versus an increase in investment-specific
TFP is that the latter generates a drop in the relative price of investment. The relative price drops to encourage
agents into investing in the consumption good sector, in spite of a relatively lower productivity. The economic
mechanisms behind the remaining effects are the same as for the increase in aggregate TFP.17
Figure 16 now compares the effects of all four policy changes with the data. Once again, by construction,
both TFP increases and also the increase in capital mobility roughly match the post-1986 evolution of output per
worker. The improvement in investor protection is not constructed in such a way. In spite of this, it is also able to
match the post-1986 growth of output per worker remarkably well.18
The only policy change qualitatively consistent with the drop in the nominal investment rate is the increase
in capital mobility, although the effect is much too small quantitatively. All the other policy changes predict an
increase in this variable. It is worth pointing out, however, that while an increase in the nominal investment rate
is a robust implication of both TFP changes, it is not a robust implication of an improvement in investor
protection. Castro, Clementi, and MacDonald (2008) show that the nominal investment rate can remain flat or
even decline with an improvement with investor protection. A key parameter governing this effect is I . With a
slightly higher I (i.e., if the Portuguese economy were initially less open to international capital flows that
what assumed here), one can potentially match the post-1986 behavior of the nominal investment rate.
Instead, only the improvement in investor protection is consistent with a sustained drop in the nominal trade
balance. All the other policies changes predict this variable to increase, and then to either return back to its initial
level, or to remain above it.
Finally, both the improvement in investor protection and the increase in investment-specific TFP are
obviously the only ones capable of generating a decline in the relative investment price.
It is also interesting look at the implications for the real investment rate and real trade balance over GDP.
Figure 17 tries to achieve this. In the model, I compute “real” variables by holding the relative price of
investment constant at its initial level.19 Qualitatively, only the improvement in investor protection and the
increase in investment-specific TFP are consistent with the observations for the real investment rate. However,
only the improvement in investor protection explains a drop in the real trade balance over GDP.
17. To further understand the mechanics of the increase in aggregate and investment-specific TFP's, it might be useful to point to a formal
result in Castro, Clementi, and MacDonald (2008). They show that, in steady-state, Qt and Nt are invariant to either μI or μC. Furthermore, pt
is proportional to z C /z I . It follows that the steady-state investment rate and the trade balance to output ratio are also invariant to either μI or
μC.
18. Still, the model is always far from accounting for the post-2000 slowdown in productivity. In the context of the present model, only a
negative shock of some sort would be able to deliver near zero productivity growth after 2000. So, none of the policy experiments considered
is successful in this regard.
19. I thus compute constant-price series instead of replicating in the model the chain-weighted measurement procedure.
160 Rui Castro
−1.2
.05
−1.3
0
−1.4
−.05
−1.5
−.1
−1.6
−.15
1980 1985 1990 1995 2000 2005 1980 1985 1990 1995 2000 2005
year year
Real Investment Rate (log) Data Real Trade Balance over GDP Data
Figure 17: Policy comparison - constant price ratios
Figure 18 compares the implications of the different policy experiments for the dispersion in firm size. In
addition to the improvement in investor protection, also the increase in capital mobility manages to improve
micro-level resource allocation. In the case of greater capital mobility, this is fully due selection effects. That is,
we know that this policy does not affect relative size, which means Qt stays well below 1. However, it moves
resources away from the investment sector and into the consumption sector. Hence, away from the most
distorted sector. This change in sectoral composition is what explains the slight drop in size dispersion in
response to an increase in capital mobility. The improvement in TFP ends up generating slightly higher
dispersion in firm size, also due to changes in sectoral composition. The same is true for the increase in
investment-specific TFP.
1.5
1
.5
0
1980 1985 1990 1995 2000 2005

year
dispersion in firm size Data

Investor Protection TFP
Capital Mobility TFP−I
Figure 18: Policy comparison - firm size dispersion
Taken together, these observations suggest the improvement in investor protection is a prime candidate for
rationalizing the macro-level and micro-level observations for the Portuguese economy since joining the EU.
Qualitatively at least, the improvement in investor protection is able to account for the higher output growth,
together with an increase in real investment and a sustained larger inflow of foreign capital. The largest
discrepancy is in the behavior of the nominal interest rate. But, once again, it would be interesting to know
whether a tighter calibration of the model to the Portuguese data might alter this implication.
5.5. Additional Policy Changes: A Discussion
What about additional policy changes that were abstracted from? Several potentially important policies,
listed in Section 2.3, come to mind, (i) increasing intra-EU labor mobility, (ii) liberalizing intra-EU trade in
162 Rui Castro
goods and services, (iii) deregulation and privatization, and (iv) achieving exchange rate stability and joining the
Euro.
Labor Mobility - Lower barriers to intra-EU labor mobility, it seems fair to say, hasn't had a large impact on the
Portuguese economy. First, population outflows from Portugal were much larger prior to joining the EU,
including outflows to other European countries. Indeed, lower outflows were a major contributor to a steady
increase in net migration into Portugal throughout the 1980s, which actually became positive around 1993 (Table
II.10 of Instituto Nacional de Estatística (2007)). Second, although population inflows increased sharply in the
1990's, Fonseca, Caldeira, and Esteves (2002) and Instituto Nacional de Estatística (2007) also document a
decline in the share of EU residents. The larger inflows are mostly accounted for by Africans, Eastern-
Europeans, and Brazilians. It then seems fair to say that migration flows in and out of Portugal reacted little to
policies aimed at increasing intra-EU labor mobility. This is in fact a broader European feature, documented in
Krueger (2000).
Liberalizing trade in goods and services - The liberalization of trade in goods and services appears to have had
a larger impact. A few facts illustrate the point. Portugal's trade share of output increased significantly since
1986 (Amador, Cabral, and Maria (2007)), and along with it the intra-EU trade share (Lima- (2000)). Most of
this increase in trade was of the intra-industry type (Amador, Cabral, and Maria (2007)).
There is a recent resurgence in trade theory, originating in Melitz (2003) and Bernard, Eaton, Jensen, and
Kortum (2003), emphasizing the intra-industry resource allocation effects of trade. These papers establish how
trade leads the most productive firms in a narrowly-defined industry to grow in size and eventually export, while
the least productive firms either remain confined to the domestic market, or exit. While the cross-firm size
distribution could either increase or decrease (large firms grow larger, small firms shrink further; the net effect
will depend on the relative mass of small versus large firms), the dispersion in firm-level productivities would
certainly decrease (only less productive firms exit). By improving the micro-level allocation of resources, trade
may thus contribute to a increase in aggregate productivity.
Figures 7-9 are indeed consistent with a decline in the dispersion of firm-level productivities, conditional on
industry. It then seems plausible that trade liberalization might also account for the micro-level, and at least some
of the macro-level observations of the Portuguese economy since 1986. Introducing the features necessary to
account for such type of effects in the current framework (e.g. introducing within-industry productivity
dispersion) is both challenging and outside of the scope of this paper. It would, however, be a fascinating topic
for further research.
Still, it might be possible to sort out the effects of improving investor protection from the effects of
liberalizing trade in goods, at a purely empirical level. The idea is that the sectors affected by one and the other
policy are likely to be different. For the improvement in investor protection, Castro, Clementi, and MacDonald
(2008) emphasize that most of the micro-level effects should occur between high and low firm-level risk sectors.
Along similar lines, Erosa and Hidalgo-Cabrillana (2007) emphasize that most of the micro-level effects should
occur between high and low fixed cost sectors. In reality, either criterion to sorting sectors, risk or fixed-costs, is
likely to give very similar results. So, these two stories are likely to amount to the same empirical implication.
Consider now trade liberalization. An implication from Melitz (2003) and Bernard, Eaton, Jensen, and
Kortum (2003) would be that most of the improvement in resource allocation should occur in sectors with high
“tradability.” Betts and Kehoe (2001) propose a definition of a good's tradability in terms of (i) how substitutable
similar goods produced in different countries are to this good, and (ii) its trading costs. They argue, however,
that tradability is well-proxied by observed trade volumes.
It follows that one could in principle verify empirically whether most of the improvement in Portugal's
micro-level resource allocation occurred within high-tradability sectors, or across high and low risk/fixed-cost
sectors. A full analysis of this issue is beyond the scope of this paper. However, consider the main export-
oriented manufacturing sectors identified by Amador, Cabral, and Maria (2007). These are Textiles, leather, and
footwear (2-digit CAE 17 and 19), Food products, beverages and tobacco (2-digit CAE 15 and 16), Wood, paper,
and printed products (2-digit CAE 20 and 21), and finally Motor vehicles and trailers (2-digit CAE 34). On the
surface, the evidence is somewhat mixed. For example, productivity dispersion seems to have declined in
Textiles, leather, and footwear, still the largest exporting sector in 2003. However, it has stayed essentially flat in
Motor vehicles and trailers, a sector whose weight in total exports grew significantly since 1986, becoming a
close second in 2003. The bottom line is that, without further research, it is not possible at this stage to identify
which policy change explains the Portuguese data the best.
Adopting a Common Currency - According to the common currency literature, two main benefits were likely
for the Portuguese economy of having stabilized the exchange rate vis-a-vis other European countries, and of
having adopted the Euro. First, this has reduced international transaction costs. Coupled with the absence of
exchange rate uncertainty, this has potentially promoted trade between Portugal and the other Euro countries.
Frankel and Rose (2002) do find evidence for this sort of effect from common currencies. The consequences of
this, however, are the same discussed previously for intra-EU trade liberalization (i.e. the lowering of other types
of barriers).
164 Rui Castro
The second benefit is lower inflation. Can lower inflation account for the evidence of Section A? It is hard to
think so. For one thing, researchers have had a hard time establishing a causal effect from inflation to long-run
growth, even though countries with lower inflation do tend to grow faster (see Kocherlakota (1996)). It also
appears the Portuguese economy started performing the worst just around 2000, just when monetary integration
fully materialized, and with inflation already at a very low level.
Deregulation and privatisation - The Portuguese economy went through a big wave of market deregulation and
privatization after joining the EU. Whole sectors, like the financial sector, have been privatized. Barriers to entry
20
are lower, as well as all sorts of regulations to economic activity. Can increased competition due to lower
regulations explain the facts of Section 2?
I start by recalling that my sample of firms in Section 2.2 excludes firms that were mostly state-owned at
some point in the sample, and also excludes firms in the most heavily regulated sectors. Still, it could be that
higher competition in these sectors lead to higher competition in other sectors. Moreover, clearly the aggregate
data of Section 2.1 includes all firms.
In principle, deregulation-induced competition would have the same effect on resource allocation as trade-
induced competition. That is, both could potentially account for the facts of Section 2.2. So, my answer to this
question is going to be related to the previous discussion for trade liberalization: deregulation could plausibly
account for the micro-observations, and at least some of the macro-observations of Section 2. And, like before, it
should be in principle possible to verify empirically whether most of the improvements in resource allocation
occurred within highly regulated sectors, or across high and low risk/fixed-cost sectors.
6. Conclusion
Since joining the EU, Portugal's GDP per worker grew at a much faster rate. Although investment spending
declined as a fraction of output, real investment grew much faster than real output. This was possible because
investment goods became much cheaper relative to consumption. Net capital inflows have also increased in real
terms compared to output. Finally, there is evidence of a significant improvement in micro-level resource
allocation.
This paper argues that a prime candidate to rationalize this evidence is the improvement in the protection of
outside investors that took place in Portugal since joining the EU. Better investor protection improves the way
firms have access to external finance. With better access to external funds, firms may operate at a more
20. In spite of these improvements, Portugal still tends to display higher levels of regulation compared to similarly developed countries,
particularly regarding the time it takes for an entrepreneur to obtain legal status. On this, see Cabral (2007).
efficiently scale. This reduction in firm-level distortions is particularly significant in sectors whose production is
more risky. Such is the case of sectors producing investment goods. Because of this, investment goods become
less expensive compared to consumption goods. Investment rises as a fraction of output, particulary at constant-
prices. This comes in part from a greater ability to attract foreign capital.
There are at least two areas in which the analysis could be improved. First, within the confines of the
framework presented here, it would be important to tighten the match between the model and the Portuguese
data. This means both calibrating the model to specifically match long-run facts on the Portuguese economy, and
also analyzing a setting in which several policy changes operate at the same time. This would allow one to make
more precise quantitative statements about the relative contribution of the different policy changes.
Second, it would be important to allow the model to have something to say about the effects of trade
liberalization and market deregulation. On the surface, these policies appear to be able to account for at least
some of the Portuguese evidence as well as the improvement in investor protection.
More generally, this paper has hopefully contributed to sharpening some of the policy discussion in
Portugal. It is not uncommon to hear analysts identify very broad policy changes that plausibly affect outcomes,
such as “improving productivity” or “joining the EU”. However, it is never laid out how exactly one goes about
improving productivity, or which among the many changes brought about by joining the EU have had an
important effect. Moreover, it is often not laid out how exactly these changes impact on the macroeconomy, or
how agents respond to it. At the very best, this makes it impossible to test the different hypothesis. More likely,
statements about policy effects which ignore changes in individual behavior and general equilibrium effects
simply cannot be trusted.
The aim of this paper has been to make this policy discussion a bit more tangible, and also to bring
additional evidence to the table bearing on the effects of the different policies. The view put forth here is that the
spotlight should be on policies, such as those affecting laws and law enforcement that protect outside investor
rights, or such as trade liberalization and market deregulation. These policies work mostly at the micro-level, by
improving resource allocation. They impact on macroeconomic outcomes by almost silently building up on a
series of micro-level improvements. The ultimate macroeconomic impact, however, turns out to be quite
significant.
166 Rui Castro
References
Albuquerque, R., and N. Wang (2008): “Agency Conflicts, Investment, and Asset Pricing,” Journal of
Finance, 63(1), 1–40.
Amador, J., S. Cabral, and J. R. Maria (2007): “International Trade Patterns over the Last Four Decades:
How Does Portugal Compare with the Other Cohesion Countries?,” Banco de Portugal, Working Paper 14 |
2007.
Amaral, P., and E. Quintin (2007): “Financial Intermediation and Economic Development: A Quantitative
Assessment,” Federal Reserve Bank of Dallas.
Antunes, A., T. Cavalcanti, and A. Villamil (2008): “The Effect of Financial Repression and Enforcement
on Entrepreneurship and Economic Development,” Journal of Monetary Economics, forthcoming.
Bernard, A. B., J. Eaton, J. B. Jensen, and S. Kortum (2003): “Plants and Productivity in International
Trade,” American Economic Review, 93(4), 1268–1290.
Betts, C. M., and T. J. Kehoe (2001): “Tradability of Goods and Real Exchange Rate Fluctuations,” Federal
Reserve Bank of Minneapolis.
Buera, F. (2003): “A Dynamic Model of Entrepreneurship with Borrowing Constraints,” Northwestern
University.
Burstein, A., and A. Monge-Naranjo (2007): “Foreign Know–How, Firm Control, and the Income of
Developing Countries,” NBER Working Paper #13073.
Cabral, L. M. B. (2007): “Small Firms in Portugal: a Selective Survey of Stylized Facts, Economic Analysis,
and Policy Implications,” Portuguese Economic Journal, 6(1), 65–88.
Castro, R., G. L. Clementi, and G. MacDonald (2004): “Investor Protection, Optimal Incentives, and
Economic Growth,” Quarterly Journal of Economics, 119(3), 1131–1175.
_____ (2008): “Legal Institutions, Sectoral Heterogeneity, and Economic Development,” Université de
Montréal.
Chari, V., P. Kehoe, and E. McGrattan (1996): “The Poverty of Nations: A Quantitative Exploration,”
NBER Working Paper # 5414.
Erosa, A., and A. Hidalgo-Cabrillana (2007): “On Finance As A Theory Of TFP, Cross-Industry
Productivity Differences, And Economic Rents,” International Economic Review, forthcoming.
Fonseca, M. L., M. J. Caldeira, and A. Esteves (2002): “New Forms of Migration into the European South:
Challenges for Citizenship and Governance - the Portuguese Case,” International Journal of Population
Geography, 8(2), 135–152.
Frankel, J., and A. Rose (2002): “An Estimate of the Effect of Common Currencies on Trade and Output,”
Quarterly Journal of Economics, 117(2), 437–466.
Heston, A., R. Summers, and B. Aten (2006): Penn World Table Version 6.2. Center for International
Comparisons of Production, Income and Prices at the University of Pennsylvania.
Hsieh, C.-T., and P. Klenow (2007): “Relative Prices and Relative Prosperity,” American Economic Review,
97(3), 562–585.
Instituto Nacional de Estatística (2007): Portugal - 20 Years European Integration. Insti-tuto Nacional de
Estatística, Lisbon, Portugal.
Kocherlakota, N. (1996): “Inflation and growth - commentary,” Federal Reserve Bank of St Louis Review,
(May), 170–172.
Krueger, A. B. (2000): “From Bismarck to Maastricht: The March to European Union and the Labor
Compact,” Labor Economics, 7(2), 117–134.
La Porta, R., F. Lopez-de Silanes, A. Shleifer, and R. Vishny (1998): “Law and Finance,” Journal of
Political Economy, 106, 1113–55.
Lima, M. A. (2000): “Portugal in the European Union: What Can We Tell the Central and Eastern European
Countries?,” World Economy, 23(10), 1395–1408.
Lopez-de Silanes, F. (2002): “NAFTA and Mexico’s Reforms on Investor Protection,” World Bank.
Lustig, H., and A. Verdelhan (2007): “The Cross-Section of Currency Risk Premia and US Consumption
Growth Risk,” American Economic Review, 97(1), 89–117.
Melitz, M. J. (2003): “The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry
Productivity,” Econometrica, 71(6), 1695–1725.
Tavares, J. (2004): “Institutions and Economic Development in Portugal: A Quantitative Exploration,”
Portuguese Economic Journal, 3(1), 49–79.
Whelan, K. (2002): “A Guide to U.S. Chain Aggregated NIPA Data,” Review of Income and Wealth, 48(2),
217–33.
A Data
All macro-level data used in this paper was obtained from the Bank of Portugal's Boletim Económico, and
spans the 1978-2006 period. The data on the relative price of investment is from Heston, Summers, and Aten’s
(2006) version 6.2 of the Penn World Tables, and it only spans the 1978-2004 period.
168 Rui Castro
The micro-level data set is the Quadros do Pessoal, an annual survey conducted by the Portuguese Ministry
of Employment which is mandatory for all Portuguese firms. I focus on the maximum available period length of
data collection, 1982-2005, except for 2001 because no data is available for this year.21 The Quadros do Pessoal
contains a wealth of information on Portuguese firms (and workers). In this paper I focus on nominal sales,
number of employees, and 2 and 3-digit sector of activity.
A.1 Sales
Ideally one would like to have information on value-added per firm. Unfortunately, Quadros do Pessoal does
not have data on the cost of intermediate inputs. To circumvent this problem, I'll make some assumptions and I'll
use some economic theory. Suppose firm i in sector j at time t produces according to
) + 1") j "+ j
y ijt z ijt k ijtj x ijtj e ijt ,
where y ijt is gross output (total sales), z ijt is total factor productivity, k ijt is capital services, e ijt are labor
services, and x ijt are the intermediate inputs. The parameters ) j , +j 0, 1 , the shares of capital and
intermediate goods in production, are potentially sector-specific. Value-added (net output) is given by
y
va ijt p ijt y ijt " p xt x ijt ,
y
where p ijt is the price of firm i 's output and p xt is the price of the intermediate input. If the market for
intermediate goods is perfectly competitive, then
y y ijt
p ijt +j x p xt .
ijt
Replacing in the definition of value-added

y
va ijt 1 " +j p ijt y ijt .
In other words, value-added is proportional to nominal sales, and the constant of proportionality is sector-
specific. Up to this constant, one can measure (real) value added by (real) sales. In most scenarios considered in
this paper, this turns out to be very convenient - for sufficiently low levels of sectoral disaggregation, the results
21. I note that until the mid to late 1980s, the survey was not mandatory for firms with less than 10 workers. This does not appear to
introduce significant censoring, since one cannot detect a significant discreet increase in the number of firms with less than 10 workers at any
year starting in 1992.
do not hinge upon knowing +j . Due to the absence of comprehensive data on firm or even sectoral level price
deflators, nominal sales were deflated by the GDP deflator.
One issue in Quadros do Pessoal, relevant for computing labor productivity, is that sales and the number of
employees measure in a given year do not refer to the same time period. For the following discussion, it's
important to distinguish between the data collection year (the year attached to the variables in Quadros do
Pessoal), and the observation year (the year the variable corresponds to). In any given data collection year,
Quadros do Pessoal gathers information in the month of October. The information collected on nominal sales is
always for the whole previous year. The observation year for sales is thus the year before the data collection
year. Until 1993, the number of employees is for the month of March of that data collection year. However, since
1994, the number of employees is for the month of October of that data collection year - implying a one-year
observation lag between sales and the number of employees since 1994. I'll assume that, until 1993, the
information on the number of employees is coincidental with the information on sales. In this case, the
observation year for both variables is the one before the data collection. After 1994, I need to lag the number of
employees one year, so that sales collected in a given year coincides with the number of employees collected the
year before. In this case, after 1994, sales and lagged employees refer both to the observation year prior to the
data collection year. This procedure implies a missing observation for employees in the observation year 1993. I
compute the number of employees in this year as the average between the number of employees in March 1993
(collected in 1993) and the number of employees in October 1994 (collected in 1994). It follows that there is no
missing observation, due to this procedure, for the number of employees from observation year 1981 until 2005.
A.2 Sectoral Codes
The sectoral codes are CAE codes (Portuguese Classification of Economic Activities), Revision 2 (REV2).
The challenge is to obtain codes that are consistent through time. The codes were revised twice since 1982, in
1995 and again in 2003. The 2003 change (from REV2 to REV2.1) is minimal, and affects only a couple of 3-
digit codes (516 and 519, which can be easily recoded back to their REV2 values). The 1995 change (from
REV1 to REV2) is more comprehensive, and sometimes there is not a one-to-one mapping between REV1 and
REV2 codes at the 3-digit level, and more often at the 2-digit level. Another issue is that there is a non-negligible
fraction of firms that change sectoral code for reasons unrelated to the official revision. I assigned REV2 codes
to REV1 codes using the following rule. From 1994 to 1995, I assumed that every code change was due to the
170 Rui Castro
official revision. For each REV1 code in 1994, I computed the modal REV2 code, to which the largest number
of firms switched in 1995. I attributed this code to firms that exited the sample before 1994.
A.3 Sample Selection
I eliminated observations with missing number of employees. I also eliminated firms with a share of public
capital larger than 50 percent at any point in the sample. Finally, I eliminated firms in sectors that tend to be
highly-regulated: utilities (2-digit codes 40 and 41), public mail (3-digit sector 641), financial (2-digit codes 65,
66 and 67), public administration (2-digit code 75), education (2-digit code 80), health (2-digit code 85), public
cleaning (2-digit code 90), individual's associations (2-digit code 90), and international organizations (2-digit
code 99). I also eliminated firms with sector code 000000.
Table 3 contains the subset of sector codes that are used in this study, after applying the sample selection
criteria.
B Transitional Dynamics
I consider the economy with exogenous TFP growth, as described in Section 4.1. For any variable x t , let its
§ 1
detrended value be x t +" 1") t x t .
The price level p and the relative size Q are constant in a balanced-growth path. Also, detrended
g j z functions. As outlined in Since these objects are time-invariant,

transfers depend on the time-invariant
S
they may be computed independently from initial conditions. Given an initial level of capital supply K , the 0
economy's transition path is then fully characterized by sequences £K St1 ¤ .t0 , £K Dt ¤ .t0 , £k It ¤ .t0 ,
£N t ¤ .t0 , £r t ¤ .t0 which solve:
pK St1 + 1") 4r t1 k )It pN t Eg I z 1 " N t Q) Eg C z
1
1 " N t z C Q) N t pEg I z 1 " N t Eg C z Q)

" k ")
1
S " 1 r t K St
It p +
1") K
t1
K Dt k It ¡N t 1 " N t Q¢
r t - )k )"1 It z I " 8F I
2
B t
r ' " r t B t I Ǔt
Ǔt
where
4r t1 q 1
@"1
" @1
1* 1 r t1 @
Ǔt k )It ¡N t pz I 1 " N t z C Q) ¢
B t pK St " K Dt .
The above system of equations defines the economy's transition mapping, from N t , K St into
N t1 , K St1 . Given K S0 0 , one needs to compute the unique value of N 0 that puts the economy on the
saddle path. In practice, N 0 is computed as the value such that the economy converges to the steady-state
starting from N 0 , K S0 . The full solution sequences are obtained in the process of solving for N 0 , by
iterating forward on the economy's transition function.
172 Rui Castro
A Agriculture, animal production, hunting and forestry

01 Agriculture, animal production, hunting, and related activities
02 Forestry and related activities
B Fishing
05 Fishing, aquaculture and related activities
C Mining
13 Metal Ore Mining
14 Other Mining
D Manufacturing
15 Food and beverage manufacturing
16 Tobacco manufacturing
17 Textile manufacturing
18 Apparel manufacturing
19 Leather and allied product manufacturing; Luggage manufacturing; Personal Leather Good Manufacturing; Footwear
manufacturing
20 Wood and cork product manufacturing; Basketry
21 Pulp, paper and paperboard manufacturing
22 Editing, printing and reproduction of pre-recorded information-supporting material
24 Chemical manufacturing
25 Rubber and plastic product manufacturing
26 Other non-mineral product manufacturing
27 Primary metal manufacturing
28 Metal product manufacturing, except machinery and equipment
29 Machinery and equipment manufacturing
31 Electrical equipment and machinery manufacturing
32 Radio, television, and communications equipment and appliance manufacturing
33 Surgical and orthopedic instrument manufacturing; Watch, Optics, and precision instrument manufacturing
34 Motor vehicle manufacturing; trailer and semi-trailer manufacturing
35 Other transportation equipment manufacturing
36 Furniture manufacturing; Other manufacturing
37 Recycling
F Construction
45 Construction
G Wholesale and retail trade; Automotive and personal and household goods repair and maintenance
50 Motor vehicle dealers; Motor vehicle repair and maintenance; Motor vehicle fuel retail
51 Wholesale trade (except motor vehicles)
52 Retail trade (except motor vehicles, fuel, and maintenance and repair of personal and household goods)
H Accommodation and food services (restaurants and similar
55 Accommodation and food services (restaurants and similar establishments)
I Transportation, warehousing and communications
60 Terrestrial transportation; Pipeline transportation of oil and gas
61 Water transportation
62 Air transportation
63 Support activities for transportation; Travel and tourism agencies, other support activities for tourism
64 Mail services and telecommunications
K Real estate, rentals and leasing, and professional and technical services
70 Real estate
71 Machinery and equipment rentals and leasing; Rentals and leasing of personal and household goods
72 Computer related services
73 Research and development
74 Other activities and services supplied to firms
O Other activities related to collective, social, and personal services
92 Recreational, cultural and sports activities
93 Other activities and services
Table 3: Two-digit sector codes (CAE, rev. 2)

LONGEVITY RISK, RETIREMENT SAVINGS, AND INDIVIDUAL WELFARE*
João F. Cocco†
Francisco J. Gomes‡
April 2008
Abstract
Over the last couple of decades there have been unprecedent, and to some extent unexpected,
increases in life expectancy which have raised important questions for retirement savings. We study optimal
consumption and saving choices in a life-cycle model, in which we allow for changes in the distribution of
survival probabilities according to the Lee-Carter (1992) model. We use historical empirical evidence and
actuary's projections on longevity to parameterize the model. We show that when agents use official period
life tables, which do not allow for future improvements in life expectancy, to make their savings decisions, the
effects of longevity improvements on individual welfare can be significant. This is particularly so in the
context of declining payouts of defined benefit pensions, which are correlated with improvements in life
expectancy.
*
We would like to thank Mario Centeno, David Laibson, Alex Michaelides, Olivia Mitchell and seminar participants at the London School
of Economics, Stockholm School of Economics, Tilburg University, Warwick University, and The Wharton School for comments.
†
London Business School, Regent’s Park, London NW1 4SA, UK and CEPR. Tel (020) 70008216. Email jcocco@london.edu.
‡
London Business School, Regent’s Park, London NW1 4SA, UK and CEPR. Tel (020) 70008215. Email fgomes@london.edu.
174 João Cocco - Francisco Gomes
1. Introduction
Over the last few decades there has been an unprecedented increase in life expectancy. For example, in 1970
a 65-year-old United States male individual had a life expectancy of 13.04 years.1 Three decades later, a 65-year-
old male had a life expectancy of 16.26 years. This represents an increase of 1.12 years per decade. To
understand what such increase implies in terms of the savings needed to finance retirement consumption,
consider a fairly-priced annuity that pays $1 real per year, and assume that the real interest rate is 2 percent. The
price of such annuity for a 65 year old male would have been $10.52 in 1970, but it would have increased to
$12.89 by 2000. This is an increase of roughly 23 percent. Or in other words, a 65 year old male in 2000 would
have needed 23 percent more wealth to finance a given stream of real retirement consumption than a 65 year old
male in 1970.
The numbers for Portugal are an order of magnitude similar to those for the US. Using the same data source,
we have that in 1970 a 65-year-old Portuguese male individual had a life expectancy of 12.38 years. Three
decades later, a 65-year-old male had a life expectancy of 15.43 years, which represents an increase in life
expectancy of 1.02 years per decade. This would result in an increase of 24% in the price of a fairly-priced
constant real annuity.
These large increases in life expectancy were, to a large extent, unexpected and as a result they have often
been underestimated by actuaries and insurers. This is hardly surprising given the historical evidence on life
expectancy. In the US, from 1970 to 2000 the average increase in the life expectancy of a 65 year old male was
1.12 years/decade, but over the previous decade the corresponding increase had only been 0.15 years. In
Portugal, from 1970 to 2000 the average increase was 1.02 years/decade, but over the previous three decades the
average increase was only 0.2 years/decade. In the United Kingdom, a country for which a longer-time series of
data on mortality is available, the average increase in the life expectancy of a 65 year old male was 1.23
years/decade from 1970 to 2000, but only 0.17 years/decade from 1870 to 1970. These unprecedented longevity
increases are to a large extent responsible for the underfunding of pay as you go state pensions,2 and of defined-
benefit company sponsored pension plans. For individuals who are not covered by such defined-benefit schemes,
and who have failed to anticipate the observed increases in life expectancy, a longer live span implies a lower
average level of retirement consumption.
1. The data in this paper on life-expectancy was obtained from the Human Mortality Database.
2. The decrease in birth rates that has occurred over this period has also contributed to the underfunding.
Longevity Risk, Retirement Savings, And Individual Welfare 175
The response of governments has been to decrease the benefits of state pensions, and to give tax and other
incentives for individuals to save privately, through defined contribution pension schemes. Likewise, many
companies have closed company sponsored defined benefit plans to new members, and instead offer to
contribute towards personal pensions that tend to be defined contribution in nature. There has been a
considerable amount of research comparing defined benefit and defined contribution pension plans (Campbell
and Feldstein, 2001). Some have argued that individuals may be able to replicate the risk/return characteristics of
defined benefit plans, within defined contribution plans, through an appropriate choice of financial assets.
Investment risk, or the risk associated with the returns on the portfolio of invested assets, may be reduced or
even eliminated by investing retirement savings in long-term inflation indexed bonds.
Longevity risk, or the risk that the individual might live longer than average, may be reduced by the
purchase of annuities at retirement age. However, whereas the purchase of annuities at retirement age provides
insurance against longevity risk as of this age, a young individual saving for retirement faces substantial
uncertainty as to the level of aggregate life expectancy, and consequently annuity prices, that she will face when
she retires. Furthermore, markets may be incomplete in the sense that they may lack the financial assets that
would allow individuals to insure against this risk. This paper studies how much are individuals affected by
longevity risk.
We use historical data to parameterize a standard life-cycle model of consumption and saving choices. The
novelty of the model is that the survival probabilities are stochastic and evolve according to the Lee-Carter
model (1992), which is the leading statistical model of mortality in the demographic literature. We study how the
individual's consumption, savings, and welfare are affected by longevity risk. We find that agents respond to
longevity improvements by increasing their savings, and in this way are able to at least partially self insure
against longevity shocks. We say partially because when agents guide their decisions using official period life
tables, which do not take into account future improvements in longevity, the effects of longevity risk on welfare
can be substantial, particularly as of retirement age. More generally, we show that longevity risk can have
significant welfare implications when agents underestimate the probability of future mortality improvements.
This result is suggestive of the importance of household financial literacy, a point emphasized by Lusardi and
Mitchell (2006). It also suggest that governments should try to make projections of future improvements in life
expectancy available to economic agents, to guide them in their savings decisions.
This is particularly important since the welfare losses of longevity risk are substantially higher when the
payouts of defined benefit pension plans are negatively correlated with aggregate survival rates. In this case,
which is motivated by recent events, when longevity increases and households need more wealth to finance their
retirement consumption, they are more likely to receive a lower pension.
The paper is organized as follows. In section 2 we use long term data for a cross section of countries to
document the existing empirical evidence on longevity. In sections 3 and 4 we setup and parameterize a life
cycle model of the optimal consumption and saving choices of an individual who faces longevity risk. The
results of the model are discussed in section 5. The final section concludes and discusses extensions for future
research.
2. Empirical Evidence on Longevity
In this section we consider the existing empirical evidence on longevity. The data is from the Human
Mortality Database, from the University of California at Berkeley. At present the database contains survival data
for a collection of 28 countries, obtained using a uniform method for calculating such data. The database is
limited to countries where death and census data are virtually complete, which means that the countries included
are relatively developed.
We focus our analysis on period life expectancies. These life expectancies are calculated using the age-
specific mortality rates for a given year, with no allowance for future changes in mortality rates. For example,
period life expectancy at age 65 in 2006 would be calculated using the mortality rate for age 65 in 2006, for age
66 in 2006, for age 67 in 2006, and so on. Period life expectancies are a useful measure of mortality rates
actually experienced over a given period and, for past years, provide an objective means of comparison of the
trends in mortality over time. Official life tables are generally period life tables for these reasons. It is important
to note that period life tables are sometimes mistakenly interpreted by users as allowing for subsequent mortality
changes. This is an important issue when we analyze, in the context of our model, the welfare costs associated
with an underestimation of future mortality improvements.
We focus our analysis on life expectancy at ages 30 and 65. Over the years there have been very significant
increases in life expectancy at younger ages. For example, in 1960 the probability that a male US newborn
would die before his first birthday was as high as 3 percent, whereas in 2000 that probability was only 0.8
percent. In Portugal, and in 1940, the live expectancy at birth for a male newborn was 49 years. By 2000 it had
increased to 73 years. In England, and in 1850, the life expectancy for a male newborn was 42 years, but by 1960
the life expectancy for the same individual had increased to 69 years. Our focus on life expectancy at ages 30
and 65 is due to the fact that we are interested on the relation between longevity risk and saving for retirement.
Furthermore, the increases in life expectancy that have occurred during the last few decades have been due to
increases in life expectancy in old age. This is illustrated in Figure 1, which plots life expectancy for a male
individual for the United States, England and Portugal over time, at birth, age 30, and at age 65.
90
80
Life expectancy in years
70
60
50
40
30
20
10
0
1841 1851 1861 1871 1881 1891 1901 1911 1921 1931 1941 1951 1961 1971 1981 1991 2001
Year
England at birth England at age 30 England at age 65

US at birth US at age 30 US at age 65
Portugal at birth Portugal at age 30 Portugal at age 65
Figure 1: Life expectancy in the United States, Portugal and England for a male individual at selected ages
Note: This figure shows period life expectancy over time and at selected ages (birth, age 30, and
age 65) for the United States, for Portugal, and for England. The data is from the Human Mortality
Database. The data for the United States is from 1959 to 2002, for Portugal from 1940 to 2003,
and for England from 1841 to 2003.
Table 1 reports average annual increases in life expectancy for a 65 year old male for selected countries
included in the database and for different time periods. It is important to note that the sample period available
differs across countries. This table shows that there have been large increases in life expectancy since 1970, and
these have not been confined to a few countries. Furthermore, and overall, there does not seem to be evidence
that the increases in life are becoming smaller over time: in the US, and in the 1970s, the average annual increase
was 0.13 years, whereas in the 1990s it was 0.11 years. In England, the corresponding values are 0.07 and 0.15.
These increases in life expectancy have been attributed to changes in lifestyle, smoking habits, diet, and
improvements in health care, including the discovery of new drugs.
Table 1: Average annual increases in life expectancy in number of years for a 65 year old male
Note: This table shows average annual increases in life expectancy for a 65 year old male over
time and for different countries. The data is from the Human Mortality Database.
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
30 34 38 42 46 50 54 58 62 66 70 74 78 82 86 90 94 98 102 106 110
Age
Y1960 Y1970 Y1980 Y1990 Y2000
Figure 2: Conditional probability of death for a male US individual

Note: This figure shows the conditional probability of death over the life-cycle for selected years
(1960, 1970, 1980, 1990, and 2000) and for a male United States individual. The data is from the
Human Mortality Database.
Figure 2 shows the conditional probability of death for the a male US individual, for different years, and for
ages 30 to 110. This figure shows, for each age, the probability that the individual will die before his next
birthday. As it can be seen from this figure the probability of death has decreased substantially from 1970 to
2000, mainly after ages 65. This confirms the results in figure 1, that the increases in life expectancy that have
occurred over the past few decades have been due to decreases in mortality at old age.
The uncertainty as to future increases in life expectancy in old age is reflected in the projected life
expectancies for a 65 year old UK male individual released by the Government Actuary's Department (GAD) in
October 2005, and shown in Figure 3. Importantly, this figure plots cohort life expectancies, and not period life
expectancies. Cohort life expectancies are calculated using age-specific mortality rates which allow for known or
projected changes in mortality in future years. If mortality rates at a given age and above are projected to
decrease in future years, the cohort life expectancy at that age will be greater than the period life expectancy at
the same age.
35.0
Life expectancy in number of years
30.0
25.0
20.0
15.0
10.0
1981 1986 1991 1996 2001 2006 2011 2016 2021 2026 2031 2036 2041 2046 2051
Year
Principal (cohort) High/low variant (cohort)
Figure 3: Projected cohort life expectancy for a 65 year old United Kingdom male individual
Note: This figure plots projected cohort life expectancy over time for a 65 year old male United
Kingdom individual. This figure plots a principal, a high and a low variant. The projections were
done by the UK government actuaries department and are available at www.gad.gov.uk.
This figure plots both a principal projection, and a high and low variants, which allow for high and low
increases in life expectancy, respectively. In the low variant, the future increases in life expectancy are assumed
to go to zero. For the cohort life expectancies, which capture the uncertainty in future increases in life
expectancy, and by year 2037, the high variant is 26.7 years whereas the low variant is considerable lower and
equal to 19 years. We will use these projections in order to parameterize the model.
3. A Model of Longevity Risk
We solve a life-cycle model of consumption and savings, similar to Carroll (1997) and Gourinchas and
Parker (2001), but in which survival probabilities are stochastic. We let t denote age, and assume that the
individual lives for a maximum of T periods. Obviously, T can be made sufficiently large, to allow for
increases in life expectancy in very old age. We use the Lee-Carter (1992) model to describe survival
probabilities. This is the leading statistical mortality model in the demographic literature, and it has been shown
to fit the data relatively well. In addition, it has the advantage of being a relatively simple model. Mortality rates
are given by:
lnm t,x a t b t k x
(1)
where m t,x is the death rate for age t in period x . The a t coefficients describe the average shape of the
lnm t,x surface over time. The b t coefficients tell us which rates decline rapidly and which rates decline
slowly in response to changes in the index k x . The b t are normalized to sum to one, so that they are a relative
measure. The index k x describes the general changes in mortality over time. If k x falls then mortality rates
decline, and if k x rises then mortality worsens. When k x is linear in time, mortality at each age changes at its
own constant exponential rate.
Lee and Carter (1992) show that a random walk with drift describes the evolution of k x over time well.
That is:
k x 6 k k x"1 /kx
(2)
where 6 k is the drift parameter and / kx is normally distributed with mean zero and standard deviation @ k . This
model can be used to make stochastic mortality projections. The drift parameter 6 k captures the average annual
change in k , and drives the forecasts of long-run change in mortality. A negative drift parameter indicates an
improvement in mortality over time.
3.2. Preferences
Let p t denote the probability that the individual is alive at age t 1 , conditional on being alive at age t , so
that p t 1 " m t . For a given individual age and time are perfectly co-linear, so that in order to simplify the
exposition from now on we include only age indices. We assume that the individual's preferences are described
by the time-separable power utility function:
T t"2
C 1"2 D1"2
E1 ! - t"1
pj p t"1 t
b1 " p t"1 t ,
1"2 1"2
t1 j0
(3)
where - is the discount factor, C t is the level of age/date t consumption, 2 is the coefficient of relative risk
aversion, and Dt is the amount of wealth the individual bequeaths to his descendants at death. The parameter b
controls the intensity of the bequest motive.
3.3. Labor Income
During working life age- t labor income, Yt , is exogenously given by:
logYt ft, Z t v t /t for t t K ,

(4)
where ft, Z t is a deterministic function of age and of a vector of other individual characteristics, Z t , / t is an
idiosyncratic temporary shock distributed as N0, @ 2/ , and v t is a permanent income shock, with
v t v t"1 u t , where u t is distributed as N0, @ 2u and is uncorrelated with /t . Thus before retirement, log
income is the sum of a deterministic component that can be calibrated to capture the hump shape of earnings
over the life cycle, and two random components, one transitory and one persistent.
The individual retires at age t R , and after this age income is modeled as a constant fraction 5 of permanent
labor income in the last working-year:
logYt log5 fK, Z K v K for t K ,

(5)
The parameter 5 measures the extent to which the individual has defined benefit pensions, which implicitly
provide insurance against longevity risk. In the current version of our model we assume away labor supply
flexibility, but we plan to consider this possibility later. Retiring later in life may be an additional natural
mechanism to insure against increases in life expectancy.
3.4. The Optimization Problem
We assume that there is a single riskless asset in which the individual can invest with interest rate R . In each
period the timing of the events is as follows. The individual starts the period with wealth W t . Then labor income
and the shock to survival probabilities are realized. Following Deaton (1991) we denote cash-on-hand in period
t by X t W t Yt . We will also refer to X t as wealth: it is understood that this includes labor income earned
in period t . Then the individual must decide how much to consume, C t . The wealth in the next period is then
given by the budget constraint:
Wt1 1 R Wt Yt " C t .

(6)
The problem the investor faces is to maximize utility subject to the constraints. The control variable is
consumption/savings. The state variables are age, cash-on-hand, and the current survival probabilities. In our
setup the value function is homogeneous with respect to permanent labor income, which therefore is not a state
variable.
4. Calibration
4.1. Time and preference parameters
The initial age in our model is 30, and the individual lives up to a maximum of 110 years of age. That is T
is equal to 110. Retirement age, K , is set equal to 65, which is the typical retirement age. We assume a discount
factor, - , equal to 0.98, and a coefficient of relative risk aversion, 2 , equal to three. In the baseline model we
assume that there is no bequest motive.
Undoubtedly, the calibration of the parameters for the mortality process are likely to be the most
controversial. This is in itself a sign that there is a great deal of uncertainty with respect to what one can
reasonably expect about future increases in life expectancy. In order to parameterize the stochastic process for
survival probabilities we do two things. First, we estimate the parameters of the Lee-Carter model using
historical data. Second, we try to determine which are the parameters of such model that match the projected
increases in life expectancy shown in Figure 3, made by the UK government actuaries department. The latter
projections are forward looking measures that reflect historical data, other information, and expectations of
future improvements in mortality.
4.2.1. Estimates from Historical Data: United States
For the estimation of the Lee-Carter model using historical data, we use US data from 1959 to 2002, which
is the data period available in the Human Mortality Database, and estimate:
lnm t,x a t b t k x /t,x

(7)
where / t,x is an error term with mean zero and variance @ 2/ , which reflects particular age-specific historical
influences not captured by the model. This model is undetermined: k x is determined only to a linear
transformation, b t is determined only up to a multiplicative constant, and a t is determined only up to an
additive constant. Following Lee and Carter (1992) we normalize the b t to sum to unity and the k x to sum to
zero, which implies that the a x are the simple averages over time of the lnm t,x . This model cannot be fit by
ordinary regression methods, because there are no given regressors. On the right side of the equation there are
only parameters to be estimated and the unknown index k x . We apply the singular value decomposition method
to the logarithms of the mortality rates after the averages over time of the log age-specific rates have been
subtracted to find a least squares solution.
Figure 4 shows the actual and estimated mortality rates for two different years, namely 1959 and 2002, the
first and the last year in our sample. From this figure we see that the model fits the data relatively well. In Figure
5 we plot the evolution over time of the k x parameter. This figure confirms, from a different perspective, the
data shown in Table 1. The parameter k x was relatively constant during the 1960's, a period during which there
were not mortality improvements. However, after 1970 there have been a decrease in k x reflecting the
significant decreases in mortality rates that have taken place since then.
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
30 34 38 42 46 50 54 58 62 66 70 74 78 82 86 90 94 98 102 106 110
Age
Data Y1959 Estimated Y1959 Data Y2002 Estimated Y2002
Figure 4: Actual and estimated conditional survival probabilities for the United States
Note: This figure shows the data and the estimated conditional probabilities of death for a United
States male individual at selected years. The data used in the estimation is from the Human
Mortality Database from 1959 to 2002. The estimation is done using the Lee-Carter model.
20
15
10
0
1959 1962 1965 1968 1971 1974 1977 1980 1983 1986 1989 1992 1995 1998 2001
-5
-10
-15
-20
-25
Year
Figure 5: Estimated k(x) parameter in the Lee-Carter model for the United States
Note: This figure shows the estimated k(x) parameter in the Lee-Carter model. The data used in
the estimation is for a male US individual, from the Human Mortality Database from 1959 to
2002.
We use the time series data of k x to estimate the parameters of the random walk. The estimated drift
parameter 6 k is -0.7095 and the standard deviation of the shocks @ k is 1.299 (both of these parameters are
reported in Table 2). In order to understand what such estimated parameters imply in terms of future
improvements in life expectancy, we use them to make projections. More precisely, we ask the following
question: consider an individual who is 30 years old in 2002 (the starting age in our model and the latest year for
which we have data from HMD database respectively). How does life expectancy at age 65 in 2002 compare to
life expectancy at age 65 when the individual reaches such age (i.e. in year 2037)? In other words, we ask which
is the increase in the life expectancy of a 65 year old individual that the model forecasts to take place over the
next 35 years.
Table 2: Parameters of the model
Obviously, such increase will be stochastic as it will depend on the realization of the shocks to survival
probabilities that will take place over the next 35 years. Therefore, in Table 3 we report increases/decreases in
life expectancy for several percentiles of the distribution of the shocks to life expectancy (10, 25, 50, 75, and 90).
The first row of Table 3 shows that for the values of 6 k and @ k estimated using historical data, the median
increase in the life-expectancy of a 65 year old over a such a period is 2.57 years. The 10th and 90th percentiles
are 1.51 and 3.59 years, respectively.
Table 3: Increase in life expectancy at age 65 between 2002 and 2037 at different percentiles of the
distribution and GAD projections
Note: This table shows the increases in life expectancy at age 65 predicted by the model over a 35
year period for diff erent percentiles of the distribution and for different model parameters. The
second part of the table shows the projected increase in cohort life expectancy at age 65 between
2002 and 2037 by the UK Government Actuary’s Department for the low, principal and high
variants.
4.2.2. Estimates from Historical Data: Portugal
We have obtained data on mortality rates from the Human Mortality Database for Portugal. The data
available is from 1940 to 2005, but the quality of the data for 1940-1970 are lower than in later years. The main
data source for the Portuguese data in the HMD database is the Instituto Nacional de Estatistica.
Figure 6 shows the actual and estimated mortality rates for two different years in our sample, namely 1940
and 2005, the first and the last year in our sample. From this figure we see that the model fits the Portuguese data
relatively well for the year 2005, but not for 1940. This is likely to be related to the previously mentioned lower
quality of the data for the early part of the sample. Therefore, for calibrating the model we use data from 1971 to
2005. Figure 7 plots the evolution over time of the k t parameter. We see that the parameter k t decreased
throughout this period, reflecting the significant decreases in mortality rates that have taken place.
We use the time series data of k t to estimate the parameters of the random walk. The estimated drift
parameter 6 k is -0.949 and the standard deviation of the shocks @ K is 2.803. These parameters are reported in
Table 3, below those previously estimated for the US. Comparing the two we see that the drift for Portugal is
lower, which reflects larger mortality improvements during the sample period. This is in part due to the fact that
the sample period for the US estimation included data from the 1960s, when there were not significant
improvements in life expectancy.
0.6
0.5
0.4
0.3
0.2
0.1
0
30 34 38 42 46 50 54 58 62 66 70 74 78 82 86 90 94 98 102 106
Age
Data Y1940 Data Y2005 Estimated Y1940 Estimated Y2005
Figure 6: Actual and estimated conditional survival probabilities for Portugal

Note: This figure shows the data and the estimated conditional probabilities of death for a
Portuguese male individual at selected years. The data used in the estimation is from the Human
Mortality Database from 1940 to 2005. The estimation is done using the Lee-Carter model.
Another feature of the estimated parameters of the stochastic process for k t is that the standard deviation of
the shocks for Portugal is higher than the estimated standard deviation for the US. This does not necessarily
mean that there is greater long run uncertainty with respect to longevity in Portugal than in the US. In fact the
evidence suggests that the recent increases in longevity have been due to medical advances that eventually find
their way around the world. The higher standard deviation may simply be due to higher measurement error in the
Portuguese than in the US data, or alternatively the Lee-Carter model may fit the US data better than the
Portuguese experience.
20
15
10
0
1971 1974 1977 1980 1983 1986 1989 1992 1995 1998 2001 2004
-5
-10
-15
-20
Year
Figure 7: Estimated k(x) parameter in the Lee-Carter model for Portugal

Note: This figure shows the estimated k(x) parameter in the Lee-Carter model. The data used in
the estimation is for a male Portuguese individual, from the Human Mortality Database from 1970
to 2005.
4.2.3. Estimates from Forward Looking Projections: United Kingdom
The calculations in the first two rows of Table 3 are based on a statistical model that extrapolates for the
future based on the history of mortality improvements that have taken place in the past, without conditioning on
any other information.
We carry out a different calibration exercise in which we choose parameters for the Lee-Carter model such
that when we simulate the model we are able to generate increases in life expectancy that roughly match the
forward looking projections of the UK government actuaries department shown in Figure 3. At the bottom of
table 3, and based on the same data that we have used in figure 3, we report the UK GAD projected increases in
life expectancy of a 65 year old over a 35 year period (from 2002 to 2037). We report the projected increase for
the low, principal, and high variants.
These principal, high and low variant projections are not carried out in the context of a model, but they
agreed upon by the government actuaries. They are calculated by assuming annual rates of mortality
improvement of 1, 0.5 and zero percent at all ages for the high, principal and low life-expectancy variants,
respectively. Therefore, the GAD does not assign probabilities to these different variants. They report that
``these (high and low variants) are intended as plausible alternative assumptions and do not represent lower and
upper limits.'' (GAD report no. 8, page 28).
It is not clear what plausible alternative assumptions means exactly, but we think that it is reasonable, and
conservative, to compare the high and low variant projections to the 10th and to the 90th percentiles of the
distribution of improvements in life expectancy that is generated by the model. We argue that this is a
conservative assumption because this means that the probability of life improvements predicted by the model
being higher (lower) than those predicted by the GAD under the high (low) variant (which are described as
plausible alternatives) is only 10%.
Comparing the mortality improvements projected using the estimated historical parameters of the Lee-Carter
model (shown in the first row of Table 3), to the projections of the GAD we see that the former lead to lower
median mortality improvements, and to lower dispersion in the forecasts. This suggests that the historically
estimated parameters of the Lee-Carter model do not reflect the forward-looking views of the GAD.3 To
motivate this dispersion the GAD reports that `ìt could be argued that uncertainty over long-term mortality
levels is higher than ever given current research in areas such as mapping the human genome and gene therapy.''
((GAD report no. 8, page 25)
We have experimented with different drift and volatility parameters for the Lee-Carter model, and projected
simulated mortality improvements under such alternative parameterizations. In the second row of table 3 we
report values for the drift and volatility parameters that generate mortality improvements that roughly match
those projected by the GAD. These parameters involve a smaller (more negative) drift parameter and a much
higher volatility than those estimated from historical data. Thus the GAD projections entail higher risk and
higher average rates of mortality improvement than those that have taken place historically. We solve our model
both for the parameters estimated using historical data and for this alternative parameterization based on GAD
projections.
3. The projections made by the GAD are for the UK, whereas we have estimated the Lee-Carter model for the US, but Deaton and Paxson
have shown that the UK and the US have similar histories of mortality improvements.
Figure 8 plots the probability distribution of life expectancy at age 65 predicted by the model, for both the
US historical estimated parameters of the Lee-Carter model and for the parameters chosen based on the GAD
projections.
30
28
26
24
22
Years
20
18
16
14
12
10
0.00% 5.00% 10.00% 15.00% 20.00% 25.00%
Probability
Historical GAD Projections
Figure 8: Model implied age 65 life-expectancy for different parameters of the k(x) stochastic process
Note: This figure plots the model implied life expectancy at age 65 for different parameters of the
stochastic process for k(t) and the probability that such life expectancy will occur.
4.3. Labor Income and Asset Parameters
4.3.1. Labor Income Process: United States
To calibrate the labor income process for the United States, we use the parameters estimated by Cocco,
Gomes and Maenhout (2005) for individuals with a high school degree, which are also reported in Table 2.
These estimates were obtained using data from the Panel Study of Income Dynamics. Deaton and Paxson (2001)
have investigated the correlation between aggregate labor income and mortality improvements using UK and US
data and concluded that the two aggregate series do not seem to be correlated. Therefore we assume zero
correlation between labor income shocks and shocks to longevity. We assume that the real interest rate is equal
to 1.5 percent.
4.3.2. Labor Income Process: Portugal
We follow a similar procedure to the one followed by Cocco, Gomes, and Maenhout (2005) to estimate
labor income profiles for Portugal. The data source is the European Community Household Panel (ECHP), for
the years 1994 to 1999. This data is obtained from an annual survey of a representative sample of individuals.
For a more detailed description of the data we would like to refer the reader to Addison, Centeno, and Portugal
(2008).
We first obtain a broad measure of labor income, so as to implicitly allow for (potentially endogenous) ways
of self-insuring against pure labor income risk. If one were to include only labor income, the risk an agent faces
would be overstated for several reasons: multiple welfare programs effectively set a lower bound on the support
of non-asset income available for consumption and savings purposes, both the agent and his spouse can vary
their labor supply endogenously, help from relatives and friends might be used to compensate for bad labor
income shocks and so on. For this reason we defined labor income as total reported wage and salary earnings,
self-employment earnings, and total social transfer receipts, all this for both head of household and if present his
spouse. Observations which still reported zero for this broad income category were dropped.
The estimation controls for family-specific fixed effects. We chose this technique over the synthetic-cohort
approach, because the latter one might overstate the variance of the income shocks as many sources of
heterogeneity are not properly accounted for. We do not remove households with less than six observations and
estimate an unbalanced panel. Labor income profiles differ across education groups. The vast majority of
individuals in our sample (over eighty five percent), have attained a first stage of secondary education. Therefore
we restrict the sample to this education group.
The logarithm of labor income was then regressed on dummies for age, family size, marital status, sex of the
household head, and on household fixed effects. We included households whose head was between 30 and 65
years old. We fit a third-order polynomial to the age dummies to obtain the labor income profiles for the
numerical solution. Finally, the replacement ratio used to determine the amount of retirement income, was
calibrated as the ratio of the average of our labor income variable defined above for individuals above 65 years
of age to the average labor income for individuals in the age group 60 to 64. The estimated replacement ratio for
Portugal is 73 percent which is higher than the one reported by Cocco, Gomes, and Maenhout (2005) for the US.
The estimated labor income profile for Portugal is plotted in Figure 9, together with that for the United
States. The currency units of the two are different, so that we focus on the shape rather than the level of these
income profiles. There are two main differences in the shape of these profiles. In Portugal the peak of household
income occurs later in life, and there is a smaller drop in income at retirement.
3.5 3.5
3 3
Log thousands of US dollars

Log thousands of PTE
2.5 2.5
2 2
1.5 1.5
1 1
31 35 39 43 47 51 55 59 63 67 71 75 79 83 87 91 95 99 103 107
Age
Profile Portugal Profile US
Figure 9: Estimated Income Profiles

Note: This figure plots the estimated income profiles for the United States and for Portugal.
Our procedure for estimating the variance of temporary and permanent labor income shocks follows closely
the variance decomposition described by Carroll and Samwick (1997). Using this procedure, we have estimated
unrealistically large variances of temporary and permanent labor income shocks, equal to 0.13 and 0.07,
respectively. These large estimates are due to measurement error in the data, which is a common problem
encountered when using micro data. Therefore, we have decided to use lower values for the variance of
temporary and permanent labor income shocks when calibrating the model.
4.3.3. Stochastic Replacement Ratio
One important possibility that we also consider, is that the retirement replacement ratio is correlated with
improvements in life expectancy. This is motivated by recent events: the large improvements in life expectancy
that have occurred over the last decades have led governments to reduce the benefits of pay as you go state
pensions. Therefore we will consider a parameterization in which the replacement ratio is reduced when there is
an improvement in life expectancy, such that the present value of the retirement benefits that the individual
receives is unchanged. Figure 10 shows the probability distribution that we obtain for the retirement replacement
ratio, at age 65, for both the historical US parameters and for those based on GAD projections.
0.7
0.65
0.6
Replacement Ratio
0.55
0.5
0.45
0.4
0.35
0.3
0.00% 5.00% 10.00% 15.00% 20.00% 25.00%
Probability
Historical GAD
Figure 10: Replacement ratio at retirement

Note: This figure plots the replacement ratio at age 65 for different parameters of the stochastic
process for k(x) and its probability.
4.4. Solution Technique
The model was solved using backward induction. In the last period the policy functions are trivial (the agent
consumes all available wealth) and the value function corresponds to the indirect utility function. We can use
this value function to compute the policy rules for the previous period and given these, obtain the corresponding
value function. This procedure is then iterated backwards.
To avoid numerical convergence problems and in particular the danger of choosing local optima we
optimized over the space of the decision variables using standard grid search. The sets of admissible values for
the decision variables were discretized using equally spaced grids. The state-space was also discretized and,
following Tauchen and Hussey (1991), approximated the density function for labor income shocks using
Gaussian quadrature methods, to perform the necessary numerical integration.
In order to evaluate the value function corresponding to values of cash-on-hand that do not lie in the chosen
grid we used a cubic spline interpolation in the log of the state variable. This interpolation has the advantage of
being continuously differentiable and having a non-zero third derivative, thus preserving the prudence feature of
the utility function. The support for labor income realizations is bounded away from zero due to the quadrature
approximation. Given this and the non-negativity constraint on savings, the lower bound on the grid for cash-on-
hand is also strictly positive and hence the value function at each grid point is also bounded below. This fact
makes the spline interpolation work well given a sufficiently fine discretization of the state-space.
5. Results
5.1. Life-Cycle Profiles
We use the optimal policy functions to simulated the consumption and savings profiles of thirty thousand
agents over the life-cycle. In Figure 11 we plot the average simulated income, wealth and consumption profiles.
We see that households are liquidity constrained during, roughly, the first five years. Consumption tracks income
very closely and a small level of savings is accumulated to use as insurance-cushion against negative labor
income shock. As labor income increases, and its profile becomes less steep, agents start accumulating wealth
for retirement. The consumption profile ceases to be increasing as agents get older, reflecting the fact that the
liquidity constraint becomes less binding. Finally, during retirement effective impatience increases due to
mortality risk and the consumption path slopes down, while wealth is depleted at a fast rate. The standard hump-
shaped consumption profile emerges.
In order to better understand the effects of longevity improvements on individual choices we simulate the
profiles for two different individuals who face exactly the same labor income shocks over the life-cycle, but
differ in terms of the shocks to life expectancy. For one agent the realization of the survival probabilities shocks
is negative in each period, so that in each period there in an improvement in life expectancy (this is the agent
labeled as ìncrease' in Figure 12). For the other agent we set the realizations of shocks to survival probabilities
as positive in each period.
100
90
80
Thousands of US dollars
70
60
50
40
30
20
10
0
30 34 38 42 46 50 54 58 62 66 70 74 78 82 86 90 94 98 102 106 110
Age
Consumption Wealth Income
Figure 11: Simulated Consumption, Income and Wealth in the Baseline Model
Note: This figure plots the simulated consumption, wealth, and income in the baseline model for
the US income profile. The figure plots an average across 30,000 simulated profiles.
In figure 12 we see that the agent who faces an improvement in life expectancy decides, from age 40
onwards, to save more than the individual who faces no improvement. Just before retirement there is a large
difference in accumulated savings between the two. Thus, the optimal response of agents to an increase in life-
expectancy is to save more in order to self-insure against the fact that they expect to live longer, and need to
finance more retirement consumption out of accumulated savings. Because longevity risk is realized slowly over
the life-cycle, the agent can, through his consumption and saving decisions self insure against these shocks.
5.2. The costs of underestimating future mortality improvements
When agents are at each point in time aware of the increases in longevity, and rationally take into account
any expected future increases, they can through their consumption and saving decisions partially self-insure
against longevity risk. This is in spite of the fact that the agent in our model does not have at his disposal
financial assets or securities that would allow him to hedge such risk.
160
140
120
Thousands of US dollars
100
80
60
40
20
0
30 34 38 42 46 50 54 58 62 66 70 74 78 82 86 90 94 98 102 106
Age
Cons Wealth Income Cons Increase Wealth Increase
Figure 12: Simulated Consumption, Income and Wealth for two different individuals
Note: This figure plots the simulated consumption, wealth, and income in the baseline model for
two different individuals who differ in the shocks to longevity.
We now investigate the welfare losses suffered by an agent who uses official period life tables to make his
consumption and saving decisions, without allowing for future mortality improvements. That is, we investigate
the consumption and saving choices of an agent who in each period looks up mortality rates in official period life
tables, and in this way is informed about survival rates, but who fails to recognize that such survival rates are
likely to improve in the future. This is a common mistake made by users of life tables, who think these official
life tables allow for future mortality improvements. In terms of our model this means that the agent thinks that
6 k is zero when making his consumption/saving decisions.

We calculate welfare losses under the form of consumption equivalent variations. That is for each scenario
(with and without mistakes) we compute the constant consumption stream that makes the individual as well-off
in expected utility terms as the consumption stream that can be financed by his decision. Relative utility losses
are then obtained by measuring the percentage difference in these equivalent consumption streams.
The results for different parameterizations are shown in Table 4. The first row of table 4 shows the welfare
losses associated with such mistakes for the parameters for the k x process calibrated to the projections of the
GAD. It shows that, as of age 30, the individual looses the equivalent of 0.028 percent of his annual consumption
for not recognizing future mortality improvements in his decisions. The reason for such relatively low welfare
loss is in part due to the fact that we are calculating welfare losses as of age 30, and the losses associated with an
underestimation of future mortality improvements are incurred late in life, and are small when discounted to age
30. Therefore, in the following column of table 4 we calculate welfare losses as of age 65. We see that as of this
age the agent looses 1.30 percent of his annual consumption for having failed to recognize future mortality
improvements.
Table 4: Welfare Gains in The Form of Consumption Equivalent Variations (Percent)

Note: This table reports the welfare gains at ages 30 and 65 for different parameters of the model,
for the US income profile, and for the Portuguese income profile (last panel). The last column
reports the percentage difference in wealth accumulation at retirement age.
This reason for this relatively large welfare loss as of age 65 is due to the fact that the agent by failing to
anticipate future mortality improvements saves significantly less. The last column of Table 4 reports the
percentage difference in financial wealth accumulated at age 65 between agents who correctly anticipate future
improvements in life expectancy, and those who fail to do so. This percentage difference is as high as 8.78
percent.
In the second row of table 4 we consider the case of an agent who is informed about the current survival
probabilities (from period life tables), starts his life thinking that 6 k is zero, but who updates this value based on
what has happened during his life, More precisely, we assume that agent thinks that the value of 6 k is given by:
t
6 kj
1 " F it !
k,Agent k,Agent
6t F t 6 t"1
t
j1
(8)
Thus in each period t , the expected value of 6 k,Agent is a weighted average of the initial prior probability
and the observed average increase in life expectancy during his life. We set F it t/100 , so that each annual
observation has weight of one percent.
The motivation for this scenario is to model the situation of an individual who looks at the past one hundred
years of data, and concludes that the probability of an improvement in life expectancy in old age is relatively
small. As time passes, and there are large improvements to life expectancy, the agent revises his initial
probability assessment. The second row of Table 4 shows the welfare results. The losses are still very close to
the ones previously obtained, even though the agent learns over time the value of the probability of an increase
in life expectancy.
The second and third panels of Table 4 also show the welfare losses associated with mistakes for the values
of 6 k and @ k estimated from historical data for both the US and Portugal. As expected, the welfare losses are
smaller than for the parameters that better approximate the GAD projections, but they are still significant,
particularly as of retirement age. Obviously, there is a mapping between welfare losses and financial savings as
of retirement age: the larger are welfare losses the larger the percentage difference in such savings.
5.3. Comparative statics
In the recent years there has been a trend away from defined benefit pensions, and towards pensions that are
defined contribution in nature. Faced with the severe projected underfunding of pay as you go state pensions
systems, many governments have reduced the level of benefits of such schemes, or are planning to do so. In
addition, many companies have closed their defined benefit schemes to new employees. This means that in the
future the level of benefits that individuals will derive from defined benefit schemes are likely to be smaller than
the one that we have estimated using historical data. This is important since defined benefit pension plans,
because of their nature, provide insurance against longevity risk. Thus one might argue that for a high level of
defined benefit pensions longevity risk does not matter much, but given the expected reductions in the level of
such benefits, longevity risk will become more important.
We use our model to investigate the extent to which that is likely to be the case. More precisely, we carry
out two different exercises. In the first we decrease the replacement ratio from the baseline value of 0.68 to 0.50,
and investigate the effects of such lower replacement ratio on welfare. The results are shown in table 4. The
welfare costs associated with the mistakes are considerably higher.
In the second, and probably more realistic scenario, we allow for negative correlation between the
replacement ratio and longevity shocks. In this case, when there are improvements in mortality rates the level of
retirement benefits is decreased. This is motivated by recent events: the providers of defined benefit pension
plans have in recent years reduced the benefits that are to be paid out, also through and increase in the retirement
age, as a response to the large increases in life expectancy that have occurred over the last decades. When the
replacement ratio is correlated with longevity shocks the welfare losses of underestimating future mortality
improvements are very substantial, and as high as 3.51 percent of annual consumption as of age 65, reflecting the
fact that the agent has saved 15 percent less than if the mortality improvements had been correctly forecast.
A final comparative statics exercise is to allow for a bequest motive, with the parameter b set equal to one.
From table 4 we see that as of age 30 the welfare losses associated with mistakes regarding the rate of mortality
improvement are lower. This suggests that individuals with a bequest motive may be in a better position to
provide insurance against longevity risk to those who do not have a bequest motive.
6. Conclusion and Future Research
The objective of this paper was two fold. First, to document that existing evidence on life expectancy.
Second to solve a life cycle model with longevity risk, and investigate how much such risk affects the
consumption and saving decisions, and the welfare of an individual saving for retirement. We have found that
when the agent is informed of the current survival probabilities and correctly anticipates the probability of a
future increase in life expectancy, the agent can through his consumption and saving decisions self insure against
longevity shocks. Since longevity risk is realized slowly over the life-cycle, the agent optimally saves more in
response to an improvement in longevity.
However, we also show that when agents are informed about life expectancy, but make an incorrect
assessment of the probability of future improvements in life expectancy, the effects of longevity risk on
individual welfare can be substantial. This is a common mistake since official life tables usually are period life
tables which do not allow for future mortality improvements. We have shown that the welfare losses associated
with such mistakes are particularly large in the context of declining payouts of defined benefit pensions,
especially when such declining payouts are correlated with longevity improvements, as suggested by recent
events. There are several extensions that are worth exploring in future research. Perhaps the most important one
is to consider labor supply flexibility, and in particular endogeneize retirement age.
References
Addison, John T., Mario Centeno, and Pedro Portugal, 2008, `Ùnemployment Benefits and Reservation
Wages: Key Elasticities from a Stripped-Down Job Search Approach,'' Economica, forthcoming.
Campbell J. and M. Feldstein, 2001, ``Risk Aspects of Investment-Based Social Security Reform,'' The
University of Chicago Press.
Carroll, Christopher D., 1997, Buffer-Stock Saving and the Life-Cycle/Permanent Income Hypothesis,
Quarterly Journal of Economics 114, 433-495.
Cocco, Joao, Francisco Gomes, and Pascal Maenhout, 2005, Portfolio Choice Over The Life-Cycle, Review
of Financial Studies 18: 491-533.
Deaton, Angus S., 1991, Savings and Liquidity Constraints, Econometrica 59, 1221-1248.
Deaton, Angus S. and Christina Paxson, 2001, Mortality, Income, and Income Inequality in Britain and The
United States NBER working paper 8534.
Gourinchas, Pierre-Olivier, and Jonathan Parker, 2002, Consumption over the life cycle, Econometrica, 70,
47-89.
Government Actuary's Department, National Population Projections: Review of Methodology for Projecting
Mortality, National Statistics Quality Review Series Report No. 8.
Human Mortality Database. University of California, Berkeley (USA), and Max Planck Institute for
Demographic Research (Germany). Available at www.mortality.org or www.humanmortality.de (data
downloaded on 10/17/2006).
Lee, R.D. and L.R. Carter (1992), Modelling and Forecasting U.S. Mortality, Journal of the American
Statistical Association, 87, 419, p. 659-671.
Lusardi, Annamaria and Olivia S. Mitchell (2006), Baby Boomers Retirement Security: The Roles of
Planning, Financial Literacy, and Housing Wealth, Journal of Monetary Economics, forthcoming.
Tauchen, G. and R. Hussey, 1991, Quadrature-Based Methods for Obtaining Approximate Solutions to
Nonlinear Asset Pricing Models, Econometrica, 59, 371-396.
Zeldes, Stephen, 1989, Optimal Consumption with Stochastic Income: Deviations from Certainty
Equivalence, Quarterly Journal of Economics 104, 275-298.
SIMULATION OF UNEMPLOYMENT INSURANCE SAVINGS ACCOUNTS IN PORTUGAL*
Ricardo Rodrigues
European Centre for Social Welfare Policy and Research
January 2008
Abstract
This paper simulates the introduction of Unemployment Insurance Savings Accounts (UISA) in
Portugal, similar to those proposed by Feldstein and Altman (1998) to the United States. We analyse the
viability conditions and potential redistributive consequences of replacing the current unemployment benefits
system with UISA. The simulation was performed on hypothetic lifetime labour market histories of individuals
estimated from a representative sample from the Portuguese Employment Survey. We conclude that the UISA
would be viable, as 64 percent of individuals would retire with positive balances, and a change to the UISA
would result in a benefit for individuals of all income quintiles on average, although better off individuals
would benefit the most from it.
Keywords: Unemployment insurance, Individual accounts, Unemployment duration, Wage determination models, Simulation models.
JEL Classification: J64, J65.
___________________________________________
* I am very grateful to Mário Centeno, Álvaro Novo and Vítor Escária for their inestimable comments and suggestions as members of the
jury of the dissertation thesis for obtaining the Master degree of Economics, Instituto Superior de Economia e Gestão (ISEG) – Universidade
Técnica de Lisboa, on which this paper is based. I also wish to thank the anonymous referee of this conference for helpful comments.
204 Ricardo Rodrigues
1. Introduction
Unemployment benefits (UB) play an important role in the labour markets of most developed countries
providing insurance against the hardships of unemployment, enhancing welfare and consumption smoothing and
allowing unemployed more time to search for an adequate job match. However, it is commonly argued that by
reducing the cost of remaining unemployed, UB such as we know it cause a problem of moral hazard:
beneficiaries will inefficiently increase the duration of their unemployment spells in order to search longer for a
job or simply to take up more leisure. In favour of this argument a number of empirical studies have found a link
between exhaustion of UB and increasing escape rates from unemployment, as well as differences in the escape
rates of beneficiaries and non-beneficiaries of UB1. Brown et al (2006) argue that this arises from the fact that
UB place a burden on those employed through the taxes needed to finance UB, while rewarding those who
remain unemployed.
In order to tackle this adverse incentive arising from UB a number of reforms has been set forward, mostly
involving some kind of benefit cuts, either through reduced duration or reduced benefit amount. More recently,
Feldstein and Altman (1998) and Kling (2006) have set forward proposals to deal with the adverse incentive of
UB without cutting social protection levels: the replacement of traditional UB by a system of Unemployment
Insurance Savings Accounts (UISA). They simulate an UISA system for the United States by applying it on the
contributory careers of a sample of individuals. In the wake of such proposal, this paper will simulate an UISA
system for Portugal and analyse its financial viability. The simulations will be based on the hypothetical
contributory records of a sample of individuals whose employment and unemployment spells, as well as wages
throughout their working lives, will be estimated by econometric models. The rest of this paper is organized as
follows: section 2 briefly explains the Portuguese UB system in 2001 and the functioning of generic UISA, in
section 3 we present the methodology and data used to estimate the hypothetical contributory records, section 4
describes the baseline UISA system to be simulated and its alternative designs, section 5 presents the results of
the simulation of UISA for Portugal and section 6 concludes and summarizes.
2. Portuguese current UB system and functioning of an UISA system
The Portuguese UB system in place in 2001 insures wage earners against unemployment by providing
Unemployment Insurance (UI) that amounts to 65 percent of previous average wage income. The amount is
caped at three times the National Minimum Wage (NMW) and may not be lower than the NMW, or the previous
1. See Krueger and Meyer (2002) for a recent review on this subject.
Simulation of Unemployment Insurance Savings Accounts in Portugal 205
wage if this is lower. In order to qualify for UI, unemployed must have a record of 18 months of social
contributions paid in the 24 months prior to unemployment. Duration of benefits depends only on the
beneficiaries’ age and may extend from 12 months for those aged under 30 years old, to 30 months for those
with 45 years of age or older. The duration can be further extended to a maximum of 38 months for the latter, if
they have paid contributions during the 20 years prior to unemployment. Those who are not eligible for UI can
apply for unemployment Social Assistance (SA). This is a flat rate benefit whose amount equals 80 percent of
the NMW for single individuals or 100 percent of the NMW if they have a family, or equal to the previous wage
if it is lower. Eligibility conditions for SA include a record of at least 6 months of social contributions paid in the
12 months prior to unemployment and a means test: per capita income may not exceed 80 percent of the NMW.
The SA has the same duration as UI, except if it is claimed after the exhaustion of the latter, in which case its
maximum duration is half of that of the UI to which it succeeded. UB benefits are financed through social
contributions paid by employees and employers.
The UISA proposed by Feldstein and Altman (1998) and the one implemented in Chile in 2002 function in
the following way. Individuals and/or employers deposit a contribution equal to a percentage of their wage into
an individual account, which earns a market interest rate. When unemployed and if the individual meets the
eligibility conditions, the amount deposited in the individual account finances the benefits that the individual is
entitled to. If the balance is not enough to finance the benefits, the State lends the necessary amount and the
individual’s account balance thus turns negative. The amount lent by the State pays a market interest rate as well
and when employed again the contributions made to the account will begin by paying this debt. At the moment
of retirement or death, the individual receives the amount deposited in the account if its final balance is positive,
while if the final balance is negative the State forgives the debt. In the Chilean system, part of the contributions
paid by employers is deposited in a Solidarity Fund which finances UB of unemployed workers whose
individual accounts have insufficient funds2.
UISA guarantee the exact same social protection as traditional UB, as the amounts, duration and eligibility
conditions under both systems are the same. However, under UISA the individual internalizes the cost of her
own unemployment, thus reducing the problem of moral hazard inherent to the traditional UB and thereby
improving the individual’s incentive to job search, while maintaining the social protection guaranteed by the
traditional UB. Potentially this would lead to a lower unemployment rate and therefore would lower the amount
of taxes needed to finance UB. Brown et al (2006) built a theoretical model in which they demonstrate that UISA
2. See Acevedo et al. (2006) and Walker and Roa (2003) on the functioning of unemployment protection in Chile.
improve incentives both to keep the current job and to seek a new job when unemployed, compared to the
traditional UB.
Under the UISA, individuals retiring with a positive balance in their accounts will indeed have internalized
the cost of their unemployment spells, but for those retiring with a negative balance the adverse incentives would
be the same of the traditional system, since the State forgives the amounts in debt. The proportion of individuals
retiring with a negative balance in their accounts is an important indicator as to just how viable is the UISA as an
alternative to the traditional UB system. If a significant proportion of individuals concentrate unemployment
spells and they are unable to finance their unemployment benefits through the monthly contributions made to
their individual accounts, then the UISA would entail little improvement, since those most affected by
unemployment and adverse incentives would face no changes in their incentives. This is an important viability
condition and it is one that is addressed by empirical studies simulating the implementation of UISA systems,
namely the above mentioned Feldstein and Altman (1998) and Kling (2006), as well as Vodopivec and Rijec
(2001) for Estonia.
Besides the viability conditions, there is also the issue of the redistribution effects that would arise from the
replacement of the traditional UB with the UISA, i.e., which groups would be better off with the UISA than with
traditional UB and who wouldn’t benefit from this change. Net gainers and losers can be found by quantifying
the differences between what each person would pay and receive under each system. Under UISA, taxes
collected by the State only need to cover the negative final balances, whereas under traditional pay-as-you-go
UB these taxes are supposed to pay for all UB. While individuals have to pay a contribution to their accounts in
the UISA, the accumulated final balance when retiring would be theirs, if it is positive. Social benefits are
exactly the same under both systems so they don’t need to be quantified. Therefore, individuals who experience
very little or no unemployment during their working lives would probably stand to benefit the most from the
change. Most or everything of what they had contributed to their accounts would be returned when they retire,
while at the same time saving in the amount of taxes paid to the State. On the other hand, individuals ending up
with negative final balances would probably be worst off since, although taxes paid to the State would be lower,
the contributions made to their individual accounts would be no different from another tax imposed on their
income.
In order to simulate the implementation of UISA it is necessary to have information on the contributory
records of individuals and indeed both Feldstein and Altman (1998) and Kling (2006) base their findings on a
representative sample of workers for whom that information is available. Unfortunately, such data is not yet
accessible in Portugal so a different approach was followed: hypothetical contributory records were constructed
with information estimated from econometric models, based on data from the Inquérito ao Emprego
(Employment Survey) of the Portuguese Instituto Nacional de Estatística (National Statistical Institute).
3. Data, methodology and simulation algorithm for hypothetical contributory records
3.1. Data and econometric methodology
The econometric models used to estimate employment and unemployment decisions, as well as wages, were
based on the data available in the Portuguese quarterly Inquérito ao Emprego (IE) for all quarters of 2001 to
2003. This survey characterizes Portuguese labour market and each sampled individual is followed during six
consecutive quarters, thus providing longitudinal data. This survey harbours information about present labour
market situation (employment, unemployment, inactivity) and transitions that may occur, an ample array of
individual and family characteristics and also information about the past, namely time spent in the current
employer (tenure) or in the last employer if individuals are currently without a job, time spent since the first job
(experience) and elapsed durations of non-employment periods and job search. Unemployed individuals are also
questioned about UB received but without discriminating the type of UB, namely if it is UI or SA. Tenure on the
previous job was used to identify which subsidy was being paid to those who claimed to be beneficiaries of UB:
those with at least 18 months of tenure in the previous job were labelled as receiving UI, while those with 6 to 18
months of tenure were classified as SA beneficiaries. Furthermore, those who were unemployed for longer
periods of time than the maximum duration of UB were classified as non-recipients of UB. Those who had at
least 18 months of tenure in the previous job and whose elapsed unemployment spell was still within the
maximum combined duration of UI and subsequent SA were classified as receiving Subsequent SA (SSA).
From the above data two subsets of samples were drawn: sample A with 149102 observations from
individuals of all quarters who were aged 15 to 64 years old and were either employed as wage earners (136751
observations) or unemployed (12531 observations); and sample B with 3904 observations (1938 individuals)
from individuals that were unemployed when first sampled into the IE between 2001 and 2003, disregarding
those aged under 15 or over 64 years and the inactive. Given the longitudinal character of the IE, sample B may
comprise more than one interview per individual, i.e., each individual may have up to six observations.
Self-employed workers were not considered for estimation purposes, and therefore were also excluded from the
simulation of UISA for Portugal. Not only are self-employed workers not entitled to UB in Portugal, but their
inclusion in the sample could cause imprecision in the results, given that self-employment is more easily related
to atypical forms of work, heterogeneity and situations that could bring further difficulties in quantifying or
classifying some important variables, not least of all wages.
Sample A was used to estimate a wages and employment decisions using a two-step Heckman sample
selection model based on Heckman (1979). The employment decisions were modelled by a probit maximum
likelihood model (1st step) which estimated the employment probability given a set of covariates x2j:
P[hi=1|x2ij]= x2ijȕ2j + ui. (1)
φ (x 2ij β 2 j )
Φ (x 2ij β 2 j )
Besides employment probability, this probit model allowed the inverse Mills ratio, Ȝi = , to be
estimated by λ̂i . The estimated inverse Mills ratio was then integrated into the 2nd step, the estimation of the
linear wage equation by Ordinary Least Squares (OLS), using the above mentioned 136751 observations of
employed individuals from sample A, in which the logarithm of the net wage plus social contributions was
regressed on a set of covariates, x1j, plus the inverse Mills ratio:
wi*=x1ijȕ1j + ıİu Ȝi + Și, (2)
where the error term Și= İi – E{İi|xij, hi=1}.
Although x1ij and x2ij could be identical, covariates in x1ij where as much as possible excluded from x2ij in
order to prevent problems of identification in the 2nd step or collinearity which could lead to an imprecise
^
estimation of β 1 j . Table 1 shows the results for the Heckman model.
Having modelled employment decisions and wages, the next step was to estimate a model that could enable
the prediction of unemployment durations. These were estimated using a methodology based on the work of
Prentice and Gloecker (1978).
A piecewise constant hazard rate was assumed for the baseline hazard function. By using this flexible form
of the hazard function we can accommodate the possibly contradictory effects over the duration profile of
unemployment that could arise from some variables. With this framework the baseline hazard rate is constant
within each of the (c0, c1], ..., (cK-1, cK] intervals of time which were set at 2, 3, 6, 12, 18 and 24 months:
h1i exp( β j z1ij ) if t ∈ ( 0, c1 ]

°
°h exp( β j z2ij ) if t ∈ ( c1, c2 ¼º
hi (t , zij ) = ® 2i (3)
°...
°h exp( β z ) if t ∈ c , c
¯ ki j kij ( k −1 k ]
For estimation purposes we relied on a proportional hazards specification,
hi (t , z ij ) = h0 (t ) exp(β j z ij ) , (4)
where zij is a set of explanatory variables, h0(t) is the baseline hazard rate and coefficients βk can be interpreted
in this kind of model as the proportional change in risk arising from a change in the absolute value of the
explanatory variable zk.
Two different destinations out of unemployment where considered: employment with a permanent contract
and employment with other contract types, as the transition profiles to these two different destinations may
differ. Taking the example mentioned by Portugal and Addison (2003), p. 4, «if individuals place a higher value
on permanent job offers than shorter-term employment opportunities they may be expected (initially at least) to
search more intensively over the former type of vacancies», leading to different baseline hazard rates to each
destination that would not be picked by a single aggregate hazard function. The model has a competing risks
structure, in which only the minimum duration till exit from unemployment into one of m destinations, or until
the observation is censored, is observed. Assuming that these different risks are independent, all the other m-1
destinations can be set as censored and a separated duration model can be fitted to each destination. The
aggregated hazard function will then be:
2
h (t , z ij ) = ¦ hi (t , z ij ) (5)
m =1
Wage equation Participation equation

Log (net wage + social Employed
contributions)
Estimated Standard Estimated Standard
coefficients errors coefficients errors
Married 0.13158 -0.004 0.41609 -0.016
Female -0.18679 -0.004 -0.06376 -0.015
Female x Married -0.14485 -0.005 -0.19499 -0.020
Madeira 0.00807 -0.006 - 0
Açores 0.05166 -0.005 - 0
Algarve 0.05359 -0.004 - 0
Alentejo 0.12716 -0.005 - 0
Lisboa 0.15635 -0.004 - 0
Centro 0.04786 -0.005 - 0
Dummy permanent contract 0.10142 -0.004 - 0
Dummy basic education 0.23466 -0.006 0.08521 -0.023
Dummy secondary education 0.5779 -0.007 0.23038 -0.027
Dummy higher education 1.13064 -0.007 0.31772 -0.028
Dummy firm size small 0.11649 -0.005 - 0
Dummy firm size medium 0.14036 -0.006 - 0
Dummy firm size large 0.10484 -0.003 - 0
Tenure (months) 0.00110 -3.7E-05 - 0
Tenure2 -1.10E-06 -8.6E-08 - 0
Experience (months) 0.00131 -3.2E-05 - 0
Experience2 -2.20E-06 -5.2E-08 - 0
Number of working adults in household - 0 0.07679 -0.002
Number of underaged individuals in household - 0 -0.0059 -0.002
Dummy for currently attending professional training - 0 -0.27207 -0.022
Age (years) - 0 0.00976 -0.00049
Constant 5.86113 -0.008 0.39443 -0.032
Number of observations: 149,102
Table 1: Results for the Heckman model

Source: Adapted from Rodrigues (2007)
Notes: LR test of independent equations (rho = 0): chi2(1) = 4.20, Prob > chi2 = 0.0405. Dummy
variables missing: (wage equation) Norte, other types of contract, no education, firm size micro;
(participation) no education. 12,531 censored observations and 136,571 non-censored
observations.
Two important notes regarding some particularities of the data used to model unemployment duration are
worth mentioning. First, the IE questions unemployed both about elapsed duration since the last job held as well
as duration of job search and since the former may contain periods of unemployment as well as inactivity, which
could blur estimations, the latter was chosen to model non-employment durations, although the term
unemployment duration is used throughout this paper for simplification. Second, this data provides a stock
sample of unemployed as date of beginning of job search is known, but date of exit out of unemployment may
not be known as individuals may leave the sample before it. Observed unemployment spells are therefore
incomplete and while longer spells are over-represented, the average duration of those spells is underestimated.
However, given the longitudinal characteristics of the data (each individual may be observed more than once, up
to six times), one can condition the observed spells to the probability of having stayed unemployed to that date
thus allowing the correct estimation of conditioned density function, i.e. the hazard function.
Although each individual is interviewed every three months, the exact dates of both beginning of job search
and transitions out of unemployment are known. Therefore transitions may occur in any moment during the time
span in which the individual is part of the survey. The generic likelihood contribution from a single individual,
given the sampling characteristics of the data, namely its stock sampling feature, and resorting to Lancaster
(1990) and Jenkins (2005), would be similar to equation (6):
1− d i d im
ª S (t ) º 2
ª f (t i ) º
Ai = « i » ∏ « » , (6)
¬ S (τ i ) ¼ m =1 ¬ S (τ i ) ¼
where d im equals 1 if individual i leaves at time t for destination m or else equals 0, and d i equals 1 if
individual i exits t at that point and 0 if the observation is censored, d i being equal to ¦ m
d im . Equation (6)
considers both the two destinations out of unemployment and the need to condition on being unemployment for a
period long enough to be sampled at moment τ . The stock sampling nature of the data could induce a problem
of length bias, as the sample is more likely to pick up the individuals that stay unemployed for longer periods,
which is why the conditioning on having remained unemployed (“survived”) up to the sample date is needed.
A common issue in estimating unemployment duration models is the possibility of allowing for unobserved
individual heterogeneity to be incorporated in the model. The usual approach is to take the hazard function and
add a multiplicative error to it, assuming this error term has a known distribution, usually Gamma. However,
“the effects of unobserved heterogeneity are mitigated, and thence estimates more robust, if the analyst uses a
flexible baseline hazard specification” (Jenkins (2005), p. 89), such as the above mentioned piecewise constant
hazard rate, and incorporating unobserved heterogeneity would make the simulation algorithm described in point
3.2 to cumbersome to run. The tests performed with the addition of a multiplicative error term were not entirely
conclusive as to its significance. Consequently, unobserved individual heterogeneity was not included in the
models estimated here, although the possibility of a downward bias in the estimated hazard rates cannot be
entirely ruled out.
The results of the duration model for both exit to employment with a permanent contract and employment
with other contracts are shown in table 2, whereas figure 1 shows the graphic representation of the baseline
hazard function for each destination.
Exit:
Permanent contract Other types of contract
Estimated Standard Estimated Standard
coefficients errors coefficients errors
Dummy for receiving
SA -0.83426 -0.46 -0.78848 -0.206
Dummy for receiving -1.57902 -1.019 -0.43803 -0.271
subsequent SA
Experience (months) -0.61074 -1.027 -0.73967 -0.458
Experience2 -0.00561 -0.003 -0.00111 -0.001
Number of underage
individuals in household 8.53E-06 4.8E-06 -8.49E-08 -2.3E-06
Dummy no education 0.11019 -0.122 0.0084 -0.057
Dummy basic education
1.03238 -0.635 0.18289 -0.28
Algarve 0.54779 -0.374 -0.23603 -0.138
Alentejo -0.94014 -0.544 -0.04395 -0.193
Lisboa -0.34196 -0.389 -0.12113 -0.179
Centro -0.62716 -0.367 0.10153 -0.145
Madeira 0.4128 -0.441 0.33565 -0.221
Açores 0.5305 -0.496 -0.32522 -0.35
Dummy looking for 1st
job -0.57356 -0.743 0.43036 -0.267
Number of individuals: 1,938
Number of failures: 63 336
Table 2: Estimated Piecewise Constant Hazard rate – Competing Risks Model

Notes: Dummy variables missing: secondary education, higher education, Norte.
The coefficients have the signs predicted by economic theory albeit most are not statistically significant. A
different specification was tried where distinction between the different types of UB was replaced by a single
dummy variable denoting subsidized unemployment and although this produced an improvement in the overall
significance of variables, the specification of table 2 was maintained since distinction between different UB was
important for the simulation of hypothetical contributory records.
Employment permanent contract Employment other contracts
Figure 1: Baseline hazard function for each destination out of unemployment

Source: Adapted from Rodrigues (2007).
3.2. Simulation algorithm
Having modelled employment, non-employment and wages, the next step was to build a simulation
algorithm for hypothetical contributory records. A sample of 15473 individuals with ages ranging from 15 to 64
years old, who were either employed as wage earners or unemployed, was taken from the first quarter of the IE
for 2001. For each one of these 15473 individuals an individual hypothetical contributory record was simulated.
A few important constrains were imposed, namely: there are no transitions to inactivity once individuals get
their first job, except when individuals retire at the legal retirement age (65 years old); there are no deaths and
early or deferred retirement is not possible; employed individuals are full-time wage earners; and finally, the
majority of independent variables maintain their observed value in the first quarter of 2001 throughout the
simulation, which is not altogether unrealistic regarding region of residency or education, but may not be entirely
realistic when it concerns household composition or marital status. Probably the most important constrain is the
exclusion of transitions to/from inactivity between the moment individuals begin their working careers and the
moment they retire. Limitations of time and the focus on employment/unemployment were central in the
exclusion of inactivity, although this exclusion may be potentially hurtful for the precision of simulations
involving women, as inactivity rates are traditionally higher among them.
Bearing the above mentioned restrictions in mind, the algorithm functions in the following manner. Each
sampled individual i enter the algorithm as “employed” either in her 15th anniversary, or in the jth quarter and kth
year in which she declared to the IE to have started her first job, if the latter date is older. At that point her tenure
and experience are set to zero and her wage is predicted by the wage equation of the Heckman model. In the next
quarters, until her 65th anniversary is reached, the algorithm will sequentially determine in each quarter:
First step: Probability of being employed, given by the selection equation of the Heckman model, from
which two outcomes are possible: either remaining employed, in which tenure and experience are
added another quarter; or losing the job and the tenure of the lost job is used to determine which
UB is the individual entitled to, if any at all.
Second step: In case individual i has lost her job in the first step, or is already non-employed from the
previous quarter, the probability of exiting non-employment to one of the possible two destinations
is calculated by each of the destination specific hazard functions and then compared to a randomly
generated figure (0,1]. If the hazard is higher than the randomly generated figure, than the
individual becomes employed, with the firm size determined also randomly. It is possible for the
same individual to lose her job in the first step and immediately get a new job in the same quarter
through this second step, in which case she would be merely shifting between jobs with no UB
being paid. If the hazard is lower and the individual remains unemployed during that quarter, her
non-employment spell is increased by three more months and UB are paid for during those three
months, if eligibility conditions are met.
Third step: If individual i remain employed after the first step or employed again after the second step, the
wage equation of the Heckman model will determine her wage for that quarter.
Variables such as tenure, experience and age all vary endogenously throughout the algorithm and time. As
much information as possible regarding the past was taken from the individuals’ responses to the IE and
incorporated into the hypothetical contributory records. For example, the IE provides information as to the date
in which the job held by sampled individuals who were employed in 2001 started, which means that from that
date onwards, and until the first quarter of 2001 is reached, only the above third step is ran for those individuals.
To increase robustness the algorithm was run five times for the baseline scenario and for each of the
alternatives, which will be described in section 4, and the results averaged.
The hypothetical contributory records have yielded the lifelong monthly earnings profiles by quintiles
depicted in figure 2, which are in accordance with what economic theory suggests. The monthly average wage
estimated is €853.66 in real terms for 2001, which compares with €819.71 surveyed in the Quadros de Pessoal
(2003) of the Ministry of Labour and Social Solidarity3.
1600
1400
1200
1000
Euros
800
600
400
200
0
15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63
Age
Total 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile
Figure 2: Estimated lifelong monthly earnings profiles by income quintiles

Comparing the average non-employment incidence of 9.17 percent estimated by the algorithm with the
unemployment rate of 4 percent to 7 percent observed for Portugal between 2001 and 2003 it seems that the
former may be slightly overestimated. However, the duration of simulated non-employment spells may be
underestimated as suggested by the difference between estimated long-term non-employment depicted in figure
3 and observed long-term unemployment for 2001 to 2003, in which the figures ranged from 37.2 percent to 39.8
3. Quadros de Pessoal is a matched longitudinal employer-employee dataset that surveys workers employed in almost every firm in
Portugal.
percent. A possible explanation for this outcome is the fact that the algorithm predicts two independent exit
probabilities out of unemployment thus providing individuals with two chances to exit unemployment4.
18
16
14
12
Percentage
10
0
15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63
Age
Total
Figure: 3 Estimated incidence of long-term non-employment (Percentage)

Figure 4 depicts non-employment incidence by age and income quintiles and not surprisingly, individuals
from the lower income quintiles have a higher incidence of non-employment. The profile for estimated non-
employment incidence is increasing with age, only apparently contradicting reality as the algorithm considers
young individuals looking for their first job as inactive and not unemployed. This has no influence on the UISA
results because individuals pay no social contributions and are not entitled to UB before being employed for the
first time.
The overestimated non-employment incidence will undoubtedly decrease the chances of the UISA system
being viable in Portugal, while the underestimation of the duration of non-employment spells will have mixed
effects as, on one hand, non-employment will be less concentrated, but on the other hand, most non-employment
spells will be subsidized. In fact, the coverage rate of UB estimated by the algorithm is extremely high at 94.3
percent, owing namely to the hypothesis that take-up rate is 100 percent and to the absence of means testing in
the SA modelled due to difficulties in measuring household income in the IE.
4. For each of the possible two destinations separately, the probability of exiting non-employment is calculated and then compared to a
randomly generated figure (0,1], providing two chances of exiting non-employment. In case of a tie in results, the destination is the one
with the highest probability.
16
14
12
10
Percentage
0
15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63
Age
Figure 4: Estimated lifelong incidence of non-employment by income quintiles

4. UISA systems simulated to Portugal
In the UISA system designated as baseline scenario, UB have exactly the same duration, eligibility
conditions and amounts as those of the Portuguese unemployment social protection system in 2001 described in
section 15. Therefore, under the baseline scenario there are no changes in social protection levels. Under this
scenario, individuals pay 5.22 percent of their wages into their own unemployment accounts without any
contributory caps. This contributory rate is slightly higher than the 4 percent used by Feldstein and Altman
(1998) and the 3 percent of Vodopivec and Rejec (2002), but in our simulation there is no grace period. The
individual account balance earns a market interest rate in each quarter which is equivalent to a real annual rate of
2 percent. The accumulated amount finances UB and in the event that this is insufficient, the State will lend the
necessary amount, which will pay a quarterly market interest rate equal to the one above mentioned. At
retirement age, individuals receive the accumulated balance in their accounts, if it is positive, while the State
forgives the negative balances.
5. Under the Portuguese UB system it’s possible to extend the duration of SA until conditions to apply for early retirement are met and there
is also a partial UI for unemployed individuals who accept a part-time job, therefore allowing them to combine part of the UI amount and
the wage received in the part-time job. However, since both early retirement and part-time work were ruled out as possibilities in
simulating hypothetical contributory records, these particular features of the UB system were disregarded.
Besides the baseline scenario that mimics the Portuguese UB system in 2001, four alternative UISA systems
were also simulated to Portugal. In the first alternative, a contributory ceiling is introduced at approximately
4.615 times the NMW, the wage that gives rise to an UB equal to the maximum amount of UB (three times the
NMW), so that the fraction of the wage exceeding this limit is exempted from deducting 5.22 percent to the
individual account. In the second alternative we change the maximum duration of UB so that it reflects not only
age, as in the 2001 UB system, but also the contributory record (table 3), while at the same time abolishing the
possibility to receive SA after exhaustion of UI. Under alternative three we devise a sort of experience rating
mechanism in which the employer is responsible for paying part of the bill resulting from unemployment for
which he is responsible for, in this case by financing the two first monthly UB the former employee is entitled to
receive. Finally, under alternative four, individuals pay a portion of 5.22 percent of their wages into their
individual accounts while the remaining is deposited in a solidarity fund, much like in the Chilean system, which
will earn the same market interest rate and whose balance will be used to finance the individual accounts with a
negative final balance.
Age (years) Contributory record Maximum duration Maximum duration (months) of UI

(months) (months) of UI and SA and SA - baseline scenario and
- alternative 2 traditional UB system
Less than 30 Less than 24 9
12
Equal or more than 24 12
Equal or more than 30 Less than 48 12

and less than 40 18

and less than 45 24

30
Table 3: Maximum duration of UB: alternative two and the baseline scenario
5. Results of the UISA systems simulated
Based on the hypothetical contributory records estimated by the algorithm, we simulated the above
described baseline scenario and alternatives to analyse if an UISA system could be a viable alternative to the UB
system for Portugal. Three viability conditions were built. The first, due to the reasons mentioned in section 2, is
the proportion of individual accounts with negative final balances. A second and more stringent measure of
viability is the proportion of individual accounts that ever had a negative balance at any point of time. Those
who always have positive balances will have no incentive to extend their unemployment spells as they will be
perfectly aware at any time that they will be the ones financing it. Finally, and as Feldstein and Altman (1998)
did, we calculate the fraction of all benefits paid by the system that were funded with borrowings from the State
and that the individuals could not pay back. For this we used the total amount of benefits paid to everyone as
denominator, while for the numerator we only considered individuals with negative final balances in their
accounts and for these we have: summed their balances in the moment where they became irreversibly negative
with the UB paid from that moment onwards and subtracted contributions paid by them to their accounts from
that moment. The numerator is consequently the total amount lent by the State and not returned by individuals,
providing a measure of individuals’ capability to finance their own UB. This last viability condition represents as
well, at least for the baseline scenario and alternatives 1 and 3, the cost bore by the State with the UISA as a
percentage of the total cost to the State of the traditional UB system.
We first present and analyse with greater detail the results for the baseline scenario whose viability
conditions for the total, as well as different income quintiles, are shown in table 4.
In the baseline scenario, only 36 percent of individual accounts have a negative final balance which means
that two thids of individuals internalize their unemployment costs and are thus less keen on prolonging their
unemployment spells. The percentage of individual accounts that ever have a negative balance is somewhat
higher at 61.1 percent, so that 40 percent of individuals would have perfect knowledge of bearing the costs
arising from their unemployment. Nevertheless for 41.1 percent of these accounts, their negative balances would
only be transitory as their balances would end up being positive at retirement. As it would be expected, the
proportion of these accounts recovering to positive is increasing with income quintiles. From the total amount of
UB paid, only 19.9 percent could not be paid by contributions and were therefore financed by the State. The
UISA is therefore clearly viable for Portugal as these results can be thought as the lower limit of viability for
UISA since it doesn’t incorporate the behavioural changes that would arise from shifting to the UISA, which
would translate into a lower unemployment rate and thus have an even more favourable impact on viability of
the UISA.
Income quintiles
Total
1st 2nd 3rd 4th 5th
Individual accounts with negative
final balances (%) 36.0 60.3 46.2 42.5 18.1 13.0
Individual accounts that were ever

negative (%) 61.1 74.9 69.6 68.0 52.2 40.9
Individual accounts that were ever

negative and have a negative final
balance (%) 58.9 80.4 66.4 62.4 34.7 31.8
Average final balance € 9,157.05 -€ 8,061.63 -€ 559.48 € 2,688.77 € 18,598.39 € 33,119.13

Average final balance as a
multiple of the estimated average 10.7 13.7 0.8 3.4 21.5 24.8
monthly wage (a)
Benefits paid by the State and not

returned by individuals (b) 19.9 - - - - -
Benefits paid to individuals with

negative final balances (b) 57.9 - - - - -
Table 4: Viability conditions for the baseline scenario

Notes: (a) Estimated average monthly wage for each quintile. (b) Percentage of total benefits paid.
Figure 5 depicts the histogram of final balances and it’s clear the concentration of balances just above the
null figure, signalling that the above mentioned results may be sensitive to small chances in some of the
parameters of the UISA system. The distribution of final balances shows a close resemblance to a normal
distribution, only with a very slight positive skew (Skewness =0.122, Kurtosis =3.732).
8
6
Percent
4
2
0
-100000 -50000 0 50000 100000 150000

Final balances (euros)
Percent
kdensity base
Figure 5: Histogram of final balances – baseline scenario

The profile of negative balances is clearly decreasing with income quintiles (figure 6), which is an
anticipated outcome given each quintile’s income and unemployment incidence, as individuals with lower
income are also those who have a greater frequency of unemployment throughout their hypothetical contributory
careers. Besides this, it’s worth remembering that UB design also contributes to that result, as minimum amounts
of UI may give rise to replacement rates higher than the statutory 65 percent for those with lower income,
whereas higher income individuals may face lower replacement rates due to the cap on the UI amount, despite
contributing with 5.22 percent of all their wages to the accounts. Hence, individuals from the fifth quintile may
be over accumulating into their individual accounts while those from the first and second quintiles have an
income level which hampers their ability to pay back the amounts borrowed from the State: only 20 percent and
33.6 percent of those that ever have a negative balance manage to recover to a final positive one, in that order.
70
60
50
Percentage
40
30
20
10
0
15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65
Age
Figure 6: Profile of negative individual accounts by age and income quintiles

Advancing to the alternatives to the baseline scenario, their viability conditions are depicted in table 5. It
seems evident that the viability of UISA tested is not jeopardized in any of the alternatives, however there are a
few nuances regarding each one.
Alternative one imposed a cap on the wage base to which the 5.22 percent contributory rate was applied and
as anticipated, the effects of this cap were limited to individuals of the highest income quintile. The introduction
of this cap lowers the average final balance of the accounts overall and especially the average final balance of the
individuals of the upper quintile. However, the viability conditions of this scenario, as well as the proportion of
accounts with negative final balances among the individuals from the upper quintile, show no significant
changes compared with the baseline scenario. The cap on contributions doesn’t restrain the individuals’ ability to
accumulate sufficient funds in their accounts. We can therefore conclude that a contributory cap set at this level
would prevent individuals from the upper quintile from saving an inefficiently high amount into their savings
accounts, while at the same time leaving the viability conditions of the system virtually unaffected when
compared to the baseline scenario.
Unlike the other alternatives, the changes introduced with alternative two could impact on the hypothetical
contributory records themselves since it could lead to a lower maximum UB duration than under the Portuguese
UB system in 2001, thus affecting the duration and incidence of unemployment spells. The algorithm was
therefore ran with changes of alternative two in place. There were only insignificant differences in the
hypothetical contributory records, since overall average non-employment incidence was 9.07 percent, against
9.17 percent of the baseline scenario and only a slight decrease of the long-term non-employment incidence, but
only until the age of 40 years old and with differences bellow three p.p.. The UB coverage also remained well
above the 90 percent, even for individuals of the lower quintiles, whose higher incidence of unemployment could
make them more prone to a lower coverage. This strengthens the possibility that the algorithm may
underestimate the duration of individual non-employment spells as changes in UB durations, liable to hurt those
with very long non-employment spells, seem to produce little impact in the hypothetical contributory careers.
The same holds for the viability conditions of the UISA, because the results of the baseline scenario remain
unchanged under alternative two.
Alternative 1 Alternative 2 Alternative 3 Alternative 4

Total 4th income 5th income Total Total Total
quintile quintile
Individual accounts with negative 36.5 18.2 15.2 35.0 10.6 50.5
final balances (%)
Individual accounts that were
ever negative (%) 61.3 51.9 42.2 60.6 32.9 70.4
Individual accounts that were
ever negative and have a negative 59.5 35.1 36.1 57.8 32.3 71.7
final balance (%)
Average final balance € 8,018.56 € 18,517.32 € 27,610.16 € 9,787.35 € 24,750.09 -€ 1,736.24
Average final balance as a
multiple of the estimated average 9.4 21.4 20.7 11.5 29.0 -2.0
monthly wage (a)
Benefits paid by the State and not
returned by individuals (b) 20.3 - - 18.9 3.5 31.6
Benefits paid to individuals with
negative final balances (b) 58.9 - - 56.4 14.0 73.5
Table 5: Viability conditions for the alternative scenarios

Notes: (a) Estimated average monthly wage for each quintile. (b) Percentage of total benefits paid.
The experience rating mechanism introduced under alternative three strongly improves the viability
conditions of UISA, as 90 percent of individuals now internalize the cost of their unemployment. Even among
individuals of the lowest income quintile, only one fourth of these have a final negative balance in their
accounts. Under alternative three there is a third party financing UB, which is the employer, and although only
the first two monthly benefits of each spell are financed by this third party, in the overall the employer pays 44.5
percent of the total amount of UB paid. The State’s financial effort to support the accounts with a negative final
balance is also greatly diminished. This alternative causes the employer to internalize part of the costs arising
from unemployment but has a potential side-effect. In fact, alternative three creates a further dismissal cost,
which according to economic theory may decrease the employment probabilities of jobseekers envisaged as
more prone to unemployment by employers. However, given its strong positive impact on the viability
conditions of UISA and the advantage of causing an internalization of unemployment costs by the employer, this
experience rating mechanism could be introduced while reducing other dismissal cost as not to hinder
employability of some groups.
Finally, under alternative four, individuals deposit approximately 1.31 percent6 of their wages into a
solidarity fund, whose deposited amounts earn the same market interest rate as the individual accounts, while the
remaining 3.91 percent of their wages is deposited into their individual accounts. These percentages were
calculated as to assure that in the end of their working lives the solidarity fund will have just enough funds to
cover the individual accounts with negative final balances. Therefore, it won’t be necessary to levy any further
tax on the individuals in order to finance these negative accounts, unlike the previous scenarios, including the
baseline. Given the lower contributory rate to the individual accounts, the viability conditions of alternative four
clearly deteriorate but still, half the accounts would have a positive final balance.
Taking the results from Feldstein and Altman (1998), Vodopivec and Rejec (2002) and Kling (2006) and
comparing those with the results obtained here it’s obvious that the former and Vodopivec and Rejec (2002) low-
unemployment scenario7 show more positive results, with less than 10 percent of negative final balances.
However, that is accomplished in a context of much shorter unemployment durations and specially much less
generous benefits than in the baseline simulation carried out for Portugal. The results for Portugal are much more
in line with the high-unemployment scenario of Vodopivec and Rejec (2002) or even Kling (2006), both with
negative final balances accounting for one third of the total, although the latter includes a compensation for wage
losses after unemployment in UISA simulated. The main conclusion from this comparison is that UISA remain
viable even when simulated over a rather generous UB system as the Portuguese in 2001, just as long as this is
combined with a little higher contribution rate to the individual accounts8.
Apart from the viability conditions, it’s also worth investigating the redistribution effects that could arise
from replacing the existing UB system, which spreads unemployment risk throughout the entire pool of
taxpayers, with an UISA system that spreads unemployment risk throughout an individuals’ own life cycle.
Notice that this is a different issue from discussing the redistribution of the UISA by itself. In the UISA
6. Actually, the exact figure is 1.309050828 percent, which allows for the balance of the solidarity fund to be just enough to cover the
individual accounts with negative final balances, on average.
7. In Vodopivec and Rejec (2002) unemployment rate is around 4 percent in the low-unemployment scenario and 9.3 percent in the high-
unemployment scenario.
8. It should be borne in mind that Feldstein and Altman (1998) impose a five year grace period during which individuals contribute to their
UISA while UB are still paid by the State.
simulated in this paper there is still room for redistribution through the negative final balances whose debt is
forgiven by the State.
As stated in section 2, the shift from the current UB system to UISA could have an impact in the amounts
paid and received by each income quintile, presumably leaving those with higher positive final balances better
off under the UISA, while those with negative final balances would be worst off. In the simulation carried out in
the present paper, the baseline scenario UISA pays the same amount of benefits as the current UB system.
Individuals finance their individual accounts through a contribution of 5.22 percent, which is absent in the
current system, while receiving back whatever final balance is deposited in their accounts if it is positive. In the
current UB system the State levies a tax of 5.22 percent9 on wages to finance UB, while under the simulated
UISA this is replaced by an anticipated much lower tax, which only has to cover the accounts with a negative
final balance:
¨ ¦ min (0, final balance )¸

§ 15473 ·
Tax under UISA =

© i =1 ¹. (7)
15473 k
§ ·
¦ ¨ ¦ wages ¸
i =1 © m = 0 ¹
The tax rate necessary to finance the negative final balances would be very low in any case: 1.33 percent in
the baseline scenario, 1.36 percent for alternative one, 1.23 percent in alternative two and 0.22 percent under
alternative three. As we said, alternative four allows for a different way to finance the accounts with a negative
final balance. Following the Chilean example, the solidarity fund is responsible for financing these negative
accounts which are funded through the allocating to this fund of a portion of the contributions paid to individual
accounts. Under this latter alternative, no individual would have to pay any additional amount under the UISA
than with the current UB system. Moreover, transfer of resources from more affluent individuals with lengthier
periods of employment to poorer and more unemployment prone individuals would occur through this solidarity
fund.
Owing to Feldstein and Altman (1998), the redistributive effects of replacing the current UB system with
UISA are estimated by calculating the following formula for each income quintile:
( )
− ¦ ¨ ¦ contributi ons accounts UISA ¸ + ¦ max (0, final balance ) + ¦ ¨ ¦ tax traditiona l UB system − taxUISA ¸
§ k § k
3095
· 3095 3095
·, (8)
j =1 © m = 0 ¹ j =1 j =1 © m = 0 ¹
9. Portuguese employers and employees pay social contributions equivalent to 34.75 percent of gross wages to finance the public pay-as-
you-go, mostly earnings related, “General Scheme” of social protection for private sector employees. Indicatively, 5.22 p.p. is the share
of this contributory rate allocated to finance unemployment benefits.
where “tax UISA“ is calculated according to equation (7). Under alternative four, the terms “contributions
accounts UISA”, which include the contributions made to the individual account plus the solidarity fund, and
“tax traditional UB system” cancel out since both amount to 5.22 percent of the wage earned. Since no “tax UISA” is
needed in this alternative four, the redistributive effects are determined solely by the final balance of the
accounts.
Redistribution effects arising from the replacement are shown in table 6. In the baseline scenario individuals
from all income quintiles would stand to gain from the change, even if these gains would be only marginal for
individuals of the lowest quintile as these would gain in average over their working life only the equivalent to 20
percent of the monthly average wage. However, it should be stressed again that this doesn’t incorporate the
behavioural changes that would follow the introduction of UISA and which would reinforce the gains, namely by
reducing the tax needed to finance negative final accounts as these would diminish with the expected lower
unemployment rate. These results for the baseline scenario are practically a Pareto style improvement, and did
not occurred under Feldstein and Altman (1998) simulation, despite being regressive, that is, more affluent
individuals would gain the most from replacing the 2001 Portuguese UB system by UISA.
Individuals from all income quintiles would stand to gain more from replacing the conventional UB system
by UISA as simulated in alternative three, as part of the benefits would be financed by the former employer. The
same holds true for alternative four, although this latter alternative manages to limit the gains of individuals of
the upper quintiles, while benefiting more those with lower or average incomes in comparison with the baseline
scenario, due to the transfer of resources operating through the solidarity fund.
Changing to the UISA system of alternatives one and two10 would have very similar redistributive
consequences as changing to the UISA of the baseline scenario. The main exceptions would be that more
affluent individuals wouldn’t gain so much from the change to alternative one, as their capacity to accumulate
assets in the accounts would be limited and individuals from the lowest quintile would be marginally worst off
under alternative two. This however, could be reversed for better if the behavioural effects of replacing current
UB system with UISA could be considered.
10. Under alternative two the differences between the total amount of UB received in comparison with the traditional UB system are also
quantified in the redistribution effects, as modifying the maximum duration of UB produces differences in the total amounts received.
1st quintile 2nd 3rd quintile 4th quintile 5th quintile

quintile
Shifting to UISA baseline scenario

(1) Contribution accounts UISA 19.8 24.7 29.2 33.7 49.4
(2) Positive final balances 5.2 9.8 11.8 25.1 41.7
(3) Diference between taxes under the traditional UB
14.7 18.4 21.7 25.1 36.8
system and under the UISA
Total redistribution effect: -(1)+(2)+(3) 0.2 3.5 4.3 16.5 29.1

Shifting to UISA alternative 1
Total redistribution effect: 0.0 3.4 4.2 16.2 27.5
Total redistribution effect: -0.1 3.7 4.7 17.0 30.0
Table 6: Redistribution effects of replacing traditional UB system with UISA

Note: Figures are multiples of the estimated average monthly wage for the economy: € 853.66.
6. Summary and conclusions
In this paper we have simulated the application to Portugal of Unemployment Insurance Savings Accounts
such as devised in the paper of Feldstein and Altman (1998), in which individuals deposit a percentage of their
wages in individual accounts to which they can resort to in case of unemployment to finance their unemployment
benefits. The State may lend the necessary funds if the account balance is insufficient and although negative
final balances are forgiven, the individual receives the accumulated balance when retiring if this is positive. The
advantage of this system is to provide the same level of social protection while at the same time reducing moral
hazard problems associated with traditional UB systems.
The UISA was simulated to Portugal by applying it to hypothetical contributory records estimated through
econometric models that simulate employment and unemployment decisions and wages, based on a
representative sample of workers from the Portuguese Inquérito ao Emprego (Employment Survey) between
2001 and 2003. The results obtained under the baseline scenario show that the UISA is a viable alternative to the
conventional UB system for Portugal, as only 36 percent of accounts would have a negative final balance and
only 19 percent of all benefits paid by the system would be financed by the State. Alternative designs of UISA
were also simulated, including ones with caps on the wage base, UB duration as a function of both age and
contributory records, an UISA incorporating an experience rating mechanism and one with a solidarity fund
financing accounts with negative final balances. These alternatives also proved to be feasible possibilities for
Portugal and provided some insights into possible variations from the baseline scenario.
When quantifying the differences between the amounts paid and received under both systems we find that
replacing current UB system with UISA would result in a Pareto benefit, as individuals from all income quintiles
would be net gainers under the baseline scenario and in all alternatives but one (alternative two), although more
affluent individuals would benefit the most from the change of system.
The simulation exercise presented here doesn’t attempt to provide any estimate on the effects that the change
in job search efforts arising from replacing the current UB system with the UISA could have over employment
levels, or other significant economic variables, in a general equilibrium framework. This is an important aspect
and one that would be worth exploring in latter research. Nevertheless, one could expect that the results
presented here could be improved if the behavioural effects of replacing current UB system with UISA were to
be considered.
This paper shows that UISA are a viable option even when considering relatively generous UB system, such
as the Portuguese, reinforcing their relevance as an alternative reform of UB systems which allows for the
maintenance of social protection levels. This feature may render it a particularly interesting alternative for
Portugal, where poverty levels are still high and consequently scaling back the generosity of UB may not be an
option.
References
Acevedo, G., Eskenazi, P. e Pagés, C. (2006), “Unemployment Insurance in Chile: A New Model of Income
Support for Unemployed Workers”, World Bank, Social Protection Discussion Paper No. 0612.
Brown, A., Orszag, J. e Snower, D. (2006), “Unemployment Accounts and Employment Incentives”,
Institute for the Study of Labor (IZA), Discussion Paper no. 532.
Direcção-Geral de Estudos, Estatística e Planeamento (2003), Quadros de Pessoal 2002, Colecção
Estatísticas, Lisboa: DGEEP-CID.
Feldstein, M. e Altman, D. (1998), “Unemployment Insurance Savings Accounts”, National Bureau of
Economic Research, Working Paper 6860.
Heckman, J. J. (1979), “Sample Selection Bias as a Specification Error”, Econometrica, 47(1), pp. 153-161.
Jenkins, S. (2005), “Survival Analysis”, unpublished manuscript, Institute for Social and Economic
Research, University of Essex, Colchester, UK.
Kling, J. (2006), “Fundamental Restructuring of Unemployment Insurance: Wage-Loss Insurance and
Temporary Earnings Replacement Accounts”, The Hamilton Project Discussion Paper, mimeo.
Krueger, A. e Meyer, B. (2002), “Labor Supply Effects of Social Insurance”, in: Auerbach, A. e Feldstein,
M. (ed.) Handbook of Public Economics, Volume 4, Amsterdam: North-Holland.
Lancaster, T. (1990), The Econometric Analysis of Transition Data, Cambridge: Cambridge University
Press.
Portugal, P. e Addison, J. (2003), “Six Ways to Leave Unemployment”, Institute for the Study of Labor
(IZA), Discussion Paper no. 954.
Prentice, R. e Gloeckler, L. (1978), “Regression Analysis of Grouped Survival Data with Application to
Breast Cancer Data”, Biometrics, 34(1), pp. 57-67.
Rodrigues, R. (2007), “Subsídios de desemprego geridos em contas individuais: uma aplicação a Portugal”,
dissertation thesis, ISEG, Universidade Técnica de Lisboa.
Vodopivec, M. e Rejec, T. (2001), “Unemployment Insurance Savings Accounts: Simulation Results for
Estonia”, presented at the International Workshop on Severance Pay Reform: Towards Unemployment Savings
and Retirements Accounts, World Bank/Institute for Applied Systems Analysis/Ludwig Boltzmann Institute for
Economic Analysis, Laxenberg, Austria, 7th and 8th November 2003.
Walker, F. e Roa, R. (2003), “Seguro de Cesantía: Nueva Legislación Chilena”, Economía e Administración,
146, pp. 24-37.
Employed (used to estimate the

logarithm of the net wage plus Unemployed
social contributions)
Variables Mean Standard error Mean Standard error
Log (wage + social contributions) 6.204 0.585
Married 0.674 0.491
Female 0.474 0.546
Female x Married 0.318 0.294
Madeira 0.436 0.040
Açores 0.100 0.041
Algarve 0.108 0.105
Alentejo 0.104 0.132
Lisboa 0.214 0.277
Centro 0.096 0.070
Norte 0.334 0.335
Permanent contract 0.796
Dummy no education 0.047 0.049
Dummy basic education 0.720 0.735
Dummy secundary education 0.133 0.133
Dummy higher education 0.100 0.084
Dummy firm size micro 0.319
Dummy firm size small 0.087
Dummy firm size medium 0.044
Dummy firm size large 0.550
Tenure (months) 124.672 120.225
Tenure2 29997.18 48,398.11
Experience (months) 256.937 161.183
Experience2 91996.75 93,423.37
Number of working adults in household 5.575 3.572 4.075 3.352
Number of underaged individuals in household 2.418 3.123 2.212 3.123
Dummy for currently attending professional
0.035 0.072
training
Age (years) 38.341 11.818 34.525 13.010
Number of observations: 136,571 12,531
Table 7: Descriptive statistics for sample A

Variables Mean Standard error

Dummy for receiving UI 0.210
Dummy for receiving SA 0.063
Dummy for receiving subsequent SA 0.046
Experience (months) 251.833 175.855
2
Experience 94,336.94 102,076.2
Number of underaged individuals in household 0.704
Dummy no education 0.049
Dummy basic education 0.762
Dummy secondary education 0.119
Dummy higher education 0.071
Madeira 0.039
Açores 0.036
Algarve 0.108
Alentejo 0.130
Lisboa 0.308
Centro 0.054
Norte 0.326
Dummy looking for 1st job 0.048
Number of individuals (number of observations): 1,938 (3,904)
Table 8: Descriptive statistics for sample B

Employed Unemployed
Variables Mean Standard error Mean Standard error
Log (wage + social contributions) 7.407 2.708
Married 0.650 0.508
Female 0.450 0.574
Female x Married 0.295 0.308
Madeira 0.076 0.041
Açores 0.090 0.049
Algarve 0.096 0.111
Alentejo 0.083 0.113
Lisboa 0.234 0.307
Centro 0.102 0.079
Norte 0.319 0.300
Permanent contract 0.805 -
Dummy no education 0.056 0.050
Dummy basic education 0.712 0.753
Dummy secundary education 0.130 0.135
Dummy higher education 0.103 0.062
Dummy firm size micro 0.190 -
Dummy firm size small 0.064 -
Dummy firm size medium 0.030 -
Dummy firm size large 0.715 -
Tenure (months) 124.109 121.374 -
Tenure2 30,133.88 49,977.67 -
Experience (months) 245.363 161.109 216.635 174.738
Experience2 86,157.01 91,678.60 77,426.61 97,727.42
Number of working adults in household 1.234 0.945 1.164 0.908
Number of underaged individuals in household 0.830 0.986 0.840 1.070
Dummy for currently attending professional
0.041 0.058
training
Age (years) 37.620 11.938 34.608 13.117
Dummy for receiving UI - 0.136
Dummy for receiving SA - 0.067
Dummy for receiving subsequent SA - 0.023
Dummy looking for 1st job 0.335 0.034
Number of individuals: 14,652 821
Table 9: Descriptive statistics for sample of 15473 individuals from IE used to estimate hypothetical contributory records
Session 3
THE ECONOMICS OF CLIMATE CHANGE: AN OVERVIEW *
Maria A. Cunha-e-Sá†
Faculdade de Economia - Universidade Nova de Lisboa
April 2008
Abstract
This paper discusses the main economic challenges raised by climate change. Climate is a global
public good, requiring a global perspective to deal with its consequences. In this context, and based on
recently disclosed new scientific evidence, three main questions are addressed: (i) how much to emit (ii) what
economic instruments to use, and, finally (iii) what are the consequences to the architecture of international
agreements for the post-Kyoto era. The experience of the EU ETS in the context of Kyoto is also examined, as
it represents a unique experiment with reputational impact on future design of climate change policy.
*
Paper prepared for the Bank of Portugal Conference “Desenvolvimento Económico Português no Espaço Europeu”, Lisbon, May 2008.
I gratefully acknowledge helpful comments by Clara Costa-Duarte and Vasco Santos on an preliminary version of the paper.
†
Associate Professor, Faculdade de Economia, Universidade Nova de Lisboa, Campus de Campolide, P-1099-032 Lisboa, Portugal; Email:
mcunhasa@fe.unl.pt.
236 Maria A. Cunha-e-Sá
1. Introduction
The issue of global climate change and the choice of the appropriate actions to be undertaken is a challenge
to economics and economists. Despite that concentration of GHGs in the future depends on past decisions, given
the long time residence of greenhouse gases in the atmosphere, its time path will also be influenced by the
evolution of emissions, which in turn depends on economic growth, technology and policy.
Markets do not allocate nonrival goods efficiently. Climate is a global public good and, as any other global
public good, there are no strong economic and political mechanisms that can provide an efficient solution to
dealing with the consequences of its changes. Moreover, too little knowledge is likely to be produced as
innovators cannot fully capture the benefits of their innovations. These are major challenges of the New Scarcity
era in contrast to the Old Scarcity one where most of the attention was focused on resources traded in markets.
In the case of global warming, unique issues are raised. First, the time horizon is probably longer than any
other major public policy problem. While some costs are already being felt, the greatest costs will be felt in the
future, in a half a century or more. Therefore, when applying intertemporal discounting to benefits and costs
several interesting issues are raised. There are different aspects involved underlying the choice of the discount
rate: the tendency to value benefits today more than tomorrow, the ethical observation that people in the future
are likely to be richer than people today, and the fact that through investments it is possible for society to obtain
more in the future out of a given amount saved in the present. Moreover, there is a huge uncertainty coupled with
potentially large irreversibilities surrounding global warming and its economic consequences. Not only benefits
and costs are uncertain, but also the consequences of climate change following an increase in the average earth
temperature are poorly known. In particular, in the presence of disruptive events it may be become meaningless
to apply traditional cost-benefit analysis in this context.
The large time horizon involved together with uncertainty makes it harder to structure decision making. In
fact, announcements of future plans are not credible in such a context. This lack of credibility with respect to
current commitments to the future does not create the appropriate environment to deal with the economic
consequences of climate change, namely, in what concerns the required incentives to foster innovation and R&D
investment in low-carbon technologies. By making emissions reduction affordable, technological development
has an important role to play.
Finally, global warming requires international cooperation at a worldwide scale, in a context where its
impacts are not uniformly spread among the different countries or regions. Thus, incentives to cooperate are, in
general, very different, implying that the issues of participation and compliance are very hard to tackle with, as
The Economics Of Climate Change: An Overview 237
the Kyoto Protocol illustrates. Despite that China, India, and some developing countries will represent up to two-
thirds of global CO2 emissions in this century, exceeding the expected contribution of the OECD of about one-
quarter of global emissions, the developing countries claim that they have only contributed about 20 per cent of
the CO2 that has accumulated in the atmosphere from industrial activity in the past 150 years. So, the distribution
of the burden of controlling emissions between developed and developing countries now and in the future is far
from being consensual. However, without the participation of developing countries, efforts to curb down
emissions will have only a small impact on climate at a high cost. As their high priority is to increase living
standards of their population, and as, often, they do not have enough resources to pay for abatement, new
schemes will have to be designed for those countries to participate and commit to future agreements.
Climate change can have significant impact on ecosystems, wildlife and the well-being of humans now, and,
especially, in the future. Recently, new scientific evidence was disclosed confirming somehow the worst
expectations. In order to act, democratic governments need the support of public opinion. Interestingly, the
public attitudes toward climate change and its consequences have changed, namely, in the US. Given the
1
relevance of the North-American economy, one of the largest current emitters, this is an important fact.
A 2006 survey led by MIT aimed at tracking public understanding of the problem, as well as support for and
opposition to eventual policy intervention, shows that not only a relevant majority recognizes global warming as
a problem, but also that the willingness to pay for intervention has risen by 50 percent, when compared to the
results obtained in a similar survey in 2003. More recently, a survey by ABC News, The Washington Post and
Stanford University (April 20, 2007) found that 84 per cent of Americans now believe that global warming has
been occurring, while 94 per cent say they are willing to change some of their behavior to help the environment.
In the same survey, one in every three Americans cited climate change as the biggest environmental problem
facing the world, much higher than one year earlier. Moreover, this survey also indicates that large numbers of
Americans would like governments to be devoting substantial attention to the problem, favoring a range of
policies in which governments could either encourage or require businesses to reduce greenhouse gas
2
emissions .
This new attitude is also present at the highest political level, in the state-of-the-union speech on January
23rd, in the context of energy security, and, more recently, last May. In this last case, President Bush announced
that the US will work with other nations to establish a new, post-2012 framework on greenhouse gas emissions,
with an emphasis on adaptation and energy-efficient technologies. In fact, various mitigation initiatives have
1. See Auffhammer and Carson (2008).

2. Since this study did not associate different policies with specific greenhouse gas reductions, Bannon, DeBell, Krosnick, Kopp and Aldhous
(2007) have recently conducted a new one where that is examined. See also Viscusi and Zeckhauser (2005).
been launched at the state level, putting pressure on the federal government to act. For the first time climate
change is a high priority issue in the US Congress. Several bills are now discussed in the 110th US Congress
targeting greenhouse gas emissions control.
The remainder of the paper is organized as follows. Section 2 presents the relevant scientific data recently
released. Section 3 discusses the choice of the target to control emissions. Section 4 identifies the different
arguments concerning the choice of the economic instruments to control emissions. Section 5 examines the main
challenges facing the design of an international climate agreement in the post-Kyoto era. Section 6 concludes the
paper.
2. The Relevant Data
The prospect of global climate change has emerged as a major scientific issue in recent years. The
relationship between global warming and increased concentration of greenhouse gases (GHGs) such as carbon
dioxide CO2, produced by the burning of fossil fuels, is suggested by much accumulated scientific evidence.
Recently, the Working Group I contribution to the IPCC (Intergovernmental Panel on Climate Change)
Fourth Assessment Report (AR4), presents scientific evidence according to which warming seems to be
accelerating. Building upon past IPCC assessments, it incorporates new findings from the past six years of
research. Scientific progress since the Third Assessment Report (TAR) is based upon “... large amounts of new
and more comprehensive data, more sophisticated analysis of data, improvements in understanding of processes
and their simulation in models and more extensive exploration of uncertainty ranges.” According to this report,
global atmospheric concentrations of carbon dioxide, methane and nitrous oxide have increased significantly as a
result of human activities since 1750 and now far exceed pre-industrial values determined from ice cores
spanning many thousand of years.
The most important anthropogenic greenhouse gas is carbon dioxide. The global atmospheric concentration
3
of carbon dioxide has increased from a pre-industrial value of 280 ppm to 379 in 2005. The atmospheric
concentration of carbon dioxide in 2005 exceeds by far the natural range over the last 650,000 years (180 to 300
ppm) as determined by ice cores. The annual carbon dioxide concentration growth rate was larger during the last
ten years (1995-2005 average: 1.4 ppm per year) although there is year-to-year variability in growth rates.
The primary source of the increased atmospheric concentration of carbon dioxide since the pre-industrial
period results from fossil-fuel use, with land-use change providing another significant but smaller contribution.
3. ppm (parts per million) or ppb (parts per billion) is the ratio of the number of greenhouse gas molecules to the total number of molecules
of dry air.
Annual fossil fuel carbon dioxide emissions increased from an average of 6.4 [6.0 to 6.8] GtC (23.5 [22.0 to
25.0] GtCO2) per year in the 1990s to 7.2 [6.9 to 7.5] GtC (26.4 [25.3 to 27.5] GtCO2) per year in 2000-2005
4
(2004 and 2005 data are interim estimates).
The understanding of the contribution of human activities to climate change has improved since the TAR.
Therefore, it is estimated that the global average net effect of human activities since 1750 has been one of
warming with very high confidence, that is, with at least a 9 out of 10 chance of being correct, with a radiative
5
forcing of +1.6 [+0.6 to +2.4] Wm-2. Palaeoclimatic information supports the interpretation that the warmth of
6
the last half century is unusual in at least the previous 1 300 years. The last time the polar regions were
significantly warmer than present for an extended period (about 125,000 years ago), reductions in polar ice
volume led to four to six meters of sea level rise.
Eleven of the dozen years in the period 1995-2006 were among the warmest since 1850, when temperatures
began to be recorded. Therefore, the estimate for the average increases in global temperature for the 20th century
was updated from 0.6ºC [0.4ºC to 0.8ºC] in the third assessment report to 0.74ºC [0.56ºC to 0.92ºC] in the fourth
one. The linear warming trend over the last 50 years (0.13ºC [0.10ºC to 0.16ºC] per decade) is nearly twice that
for the last 100 years. The total temperature increase from 1850-1899 to 2001-2005 is 0.76ºC [0.57ºC to 0.95ºC].
This last report claims that most of the observed increase in global average temperatures since the mid-20th
century is very likely due to the observed increase in anthropogenic greenhouse concentrations. This is an
advance since the TAR's conclusion that “most of the observed warming over the last fifty years is likely to have
been due to the increase in the greenhouse concentrations.” Analysis of climate models together with constraints
from observations allows for the identification of an assessed likely range to be given for climate sensitivity for
7
the first time, providing increasing confidence of the climate system response to radiative forcing. “The global
average surface warming following a doubling of carbon dioxide concentrations it is likely to be in the range 2ºC
to 4.5ºC with a best estimate of about 3ºC, and is very unlikely to be less than 1.5ºC. Moreover, values
substantially higher than 4.5ºC cannot be excluded, but agreement of models with observations is not as good for
those values”. Based on the AR4 (Fig.1 in Box 10.2), Weitzman (2007b) points out to the fact that the upper tails
4. Fossil carbon dioxide emissions include those from the production, distribution and consumption of fossil fuels and as a by-product from
cement production. An emission of 1 GtC corresponds to 3.67 GtCO2.
5. Radiative forcing is a measure of the influence that a factor has in altering the balance of incoming and outgoing energy in the Earth-
atmosphere system and is an index of the importance of the factor as a potential climate change mechanism. Positive forcing tends to warm
the surface while negative forcing tends to cool it. In the referred report, radiative forcing values are for 2005 relative to pre-industrial
conditions defined at 1750 and are expressed in watts per square meter (Wm-2).
6. Palaeoclimatic studies use changes in climatically sensitive indicators to infer past changes in global climate on time scales ranging from
decades to millions of years. See Summary for Policymakers, Working Group I, 2007.
7. Climate sensitivity is a measure of the climate system response to radiative forcing. It is a multiplier for converting changes in
concentration levels into equilibrium temperature response.
of the 18 pdfs (probability density functions) for climate sensitivity tend to be long and fat. Based on these
results, the author estimates that the probability that the average temperature increases by 6ºC is around 5 per
cent, and by 8ºC is about 2 per cent. Also, in his own words, “...given current trends in GHG emissions, a
doubling of pre-industrial concentration levels will be attained in 30-40 years, and will keep increasing if nothing
is done. Projecting current trends in business-as-usual GHG emissions, a tripling will be attained in a century and
a steady-state equilibrium global-mean-temperature response might in this case involve for the tail something
like a 1 per cent chance above 10ºC.”
Since IPCC's first report in 1990, assessed projections have suggested global average temperature increases
between 0.15ºC and 0.3ºC per decade for 1990 to 2005. This can now be compared with observed values of
about 0.2ºC per decade, increasing the confidence in near-term projections. Even if all radiative forcing agents
were held constant at year 2000 levels, a further warming would occur in the next two decades at a rate of about
0.1ºC per decade, due mainly to the slow response of the oceans.
The sea level rose on average 3.1 [2.4 to 3.8] mm a year between 1993 and 2003, in contrast to 1.8 [1.3 to
2.3] mm a year from 1961 to 2003. Besides, recent studies show acceleration of glacier-melting in Greenland.
Some trends on climate seem now clear. Extreme weather events, like droughts, heat waves, coastal storms,
heavy rainfall, flooding, have grown more frequent and intense. Agricultural, fishery and forest productivity will
be affected, despite that the consequences will be very different in distinct regions of the world. Biodiversity loss
may also result, despite that the potential for adaptation should not be ignored.
Despite that more scientific evidence has been collected uncertainty in the science of climate change remains
a critical element. The latest IPCC report makes it plain that, while understanding of climate mechanisms has
improved, predictions may become less certain, as new feedback effects may impact them positively or
negatively. Thus, the IPCC's likely range of predictions of temperature rise by 2100 has changed from 1.4-5.8ºC
in the 2001 report to 1.1-6.4ºC in the more recent one, where the lowest level is associated to the low scenario
8
and the highest to the high one. But still more important is the fact that eventual disruptive events caused by
climate change cannot be ignored despite that the probability attributed to them may be tiny, that is, low-
probability high-impact catastrophic events. The existence of structural uncertainty and how it should be dealt
with in the economics of climate change is a subject of current debate and research, as shown in Section 3.
8. In fact, the best estimate for the low scenario (B1) is 1.8ºC (likely range is 1.1ºC to 2.9ºC), while the best estimate for the high scenario
(A1F1) is 4.0ºC (likely range is 2.4ºC to 6.4ºC).
3. How Much To Emit?
The implementation of any climate change control policy requires the quantification of the damages of
climate change from a lack of control (benefits) and the costs of greenhouse gas emissions reduction, or
mitigation costs.
To motivate the discussion, we take as a departure point Figures 1, 2, and Figure 3. From Figure 3,
concentration targets are related to the likelihood of different changes in temperature, while from Figure 1
(SPM.7 from Synthesis Report AR4), the impacts associated with different levels of warming are presented.
Also, it is worth comparing Figure 1 with Figure 2 (Fig. 19-8-1 from IPCC (2001)), where the broad categories
of climate risk are presented at a five-year difference. It is clear that new science identifies now a wide range of
high risk impacts for almost every temperature threshold.
The United Nations Framework Convention on Climate Change (UNFCCC) has defined a qualitative “long-
term goal of atmospheric stabilization to prevent dangerous anthropogenic interference with the global climate”,
but the challenge consists of translating this qualitative goal into quantitative targets for emissions reductions. To
achieve these quantitative targets costs are incurred but benefits also accrue. So, each society evaluates costs and
benefits, and by balancing them under different scenarios it may take more informed choices. In this context, the
results produced by the state-of-the-art on climate change science define the framing context in which estimates
of damages or benefits and mitigation costs have to be considered and evaluated.
3.1. Benefits
In what concerns the damages of climate change, we may distinguish between market and non-market
impacts. While the former are the welfare impacts that result from changes in prices or quantities of marketed
goods, and, therefore, are observable, for the latter there are no markets functioning. Therefore, the estimation of
welfare costs related to eventual lost ecosystem services or lost biodiversity require researchers to employ stated-
preference or survey techniques, namely, the contingent valuation method, to elicit the willingness to pay to
avoid non-market damages. As Tol's (2005) meta-analysis for twenty-eight studies' estimated benefits from
reduced climate change has shown most estimates of the benefits of mitigating GHG emissions do not exceed
9
$14 per metric ton of CO2-eq in the near-term.
9. Sterner and Persson (2007) argue that nonmarket damages from climate change are probably underestimated and that future scarcities that
will be induced by the changing composition of the economy and climate change should lead to rising relative prices for certain goods and
services. Thus, the estimated damage of climate change is larger, acting as if the discount rate was low.
Examples of impacts associated with global average temperature change

(Impacts will vary by extent of adaptation, rate of temperature change, and socio-economic pathway)
Figure 1: Examples of impacts associated with projected global average surface warming. Upper panel:
Illustrative examples of global impacts projected for climate changes (and sea level and atmospheric CO2 where
relevant) associated with different amounts of increase in global average surface temperature in the 21st century.
The black lines link impacts; brokenline arrows indicate impacts continuing with increasing temperature. Entries
are placed so that the left hand side of text indicates the approximate level of warming that is associated with the
onset of a given impact. Quantitative entries for water scarcity and flooding represent the additional impacts of
climate change relative to the conditions projected across the range of SRES scenarios A1FI, A2, B1 and B2.
Adaptation to climate change is not included in these estimations. Confidence levels for all statements are high.
Lower panel: Dots and bars indicate the best estimate and likely ranges of warming assessed for the six SRES
marker scenarios for 2090-2099 relative to 1980-1999.
Figure 2: Sources of Concern and Color-Coded Indications of Vulnerability: Relative levels of vulnerability
along five “Lines of Evidence” or “Sources of Concerned” and their sensitivity to increases in global mean
temperature were assessed based on the literature available through the middle of 2000. Low vulnerability was
indicated by a white or very pale yellow coloration. High vulnerability was highlighted by red coloration; and
intermediate vulnerabilities by various shades of yellow and orange.
Source: Figure 19-18-1 in IPCC (2001).
Figure 3: Likely global warming from stabilization at different greenhouse gas concentrations
Source: U.S. Climate Mitigations in the context of Global Stabilization, RFF Backgrounder
3.2. Costs
When estimating mitigation costs, we may distinguish between “bottom-up” and “top-down” models to
evaluate the cost-effectiveness of alternative policies to achieve a given emission target.
The recently published contribution of Working Group III to the Fourth Assessment Report (AR4) of the
10
IPCC, namely, the Summary for Policy Makers (SPM) released May 4, 2007, summarizes the main results of
studies that have estimated mitigation costs in the past six years, based on bottom-up and top-down models. The
majority of studies of mitigation costs use either the payment for tons reduced or the impact on income to
measure those costs. Therefore, the report also concentrates on these two measures. The main results are
summarized in the report as follows:
11 12
• both “bottom-up” and “top-down” studies indicate that there is substantial economic potential
for the mitigation of global GHG emissions over the coming decades, that could offset the projected
growth of global emissions below current levels;
• in 2030 macro-economic costs for multi-gas mitigation, consistent with emissions trajectories
towards stabilization between 450 and 710 ppm CO2-eq, are estimated at between a 3 per cent
decrease of GDP and a small increase, compared to the baseline. However, regional costs may
differ significantly from global averages.
Tables 1 and 2, reproduced from the SPM, summarize much of the work that has been developed in the last
six years providing the amounts that would have to be paid per ton, for different volumes of reductions. While
the rows identify opportunities based on maximum estimated cost, the columns state the volume of reductions
available at or below the per-ton cost indicated in each row, first in actual tons and then in relation to two
different estimates of emissions in 2030. From these tables, we observe that increasing the price by 150 per cent,
that is, from $20 to $50, contributes about 50 per cent more to emissions reductions. However, doubling to $100
13
determines another 20 per cent of reductions, suggesting that it gets harder to reduce emissions.
10. More recently published the Summary for Policymakers of the Synthesis Report of the IPCC Fourth Assessment Report (ARA),
Valencia, 2007/11/16.
11. Bottom-up studies are based on assessment of mitigation options, emphasizing specific technologies and regulations. They are typically
sectoral studies taking the macro-economy as unchanged. Sector estimates have been aggregated, as in the TAR, to provide an estimate of
global mitigation potential for this assessment.
12. Top-down studies assess the economy-wide potential of mitigation options. They use globally consistent frameworks and aggregated
information about mitigation options and capture macro-economic and market feedbacks.
13. Also, as pointed out by Pizer (2007), from these figures it is possible to obtain rough estimates of costs, by halving the product of price
times the volume of reductions, as these are maximum prices. Moreover, the data on prices can be easily used to calculate the impact of those
reductions on the price of electricity or gasoline, based on how much a one dollar per ton of carbon dioxide increases the price of a gallon of
gasoline.
Carbon price Economic Reduction Reduction

mitigation relative to relative to
potential SRES A1 B SRES B2
(68 GtCO2-eq/yr) (49 GtCO2-eq/yr)
(US$/tCO2-eq) (GtCO2-eq/yr) % %
0 5-7 7-10 10-14
20 9-17 14-25 19-35
50 13-26 20-38 27-52
100 16-31 23-46 32-63
Table 1: Global economic mitigation potential in 2030 estimated from bottom-up studies
Source: AR4, Working Group III, SPM
Carbon price Economic Reduction Reduction

potential relative to relative to
SRES A1 B SRES B2
(68 GtCO2-eq/yr) (49 GtCO2-eq/yr)
(US$/tCO2-eq) (GtCO2-eq/yr) % %
20 9-18 13-27 18-37
50 14-23 21-34 29-47
100 17-26 25-38 35-53
Table 2: Global economic potential in 2030 estimated from top-down studies

Both kinds of approaches, bottom-up and top-down models, have weaknesses and strengths. Bottom-up are
sectoral studies that estimate the cost per ton of a given action or technology, while top-down models are models
of the global economy that include emissions of greenhouse gases. Therefore, while the former may include a
wider variety of technological options, since they are restricted at the sectoral level, they cannot capture the
implications and feedback of those options at the economy level. Two aspects, however, deserve some
14
comment. From the two tables, first, we observe that while in the bottom-up studies the first line corresponds
to the case of zero-cost mitigation opportunities, in the top-down models there are no such cases. Hence, the
bottom-up studies do not account for the costs associated with the removal of barriers that often prevent or
14. See Pizer (2007).

increase the costs of the actions. This may underestimate the lower end of the bottom-up range of estimates, and
more important, it may suggest that market-based instruments that put a price on carbon dioxide emissions are
not enough, requiring the need for direct regulation or other interventions. This will be discussed in the next
section.
Second, the range of estimates between the two approaches is very similar. The TAR estimates from the top-
down models showed higher costs and lower mitigation potential. The differences found are explained by the
inclusion of other greenhouse gases and endogenous technological change in the AR4. In fact, when comparing
to the TAR, the new AR4 presents a range of estimates with a similar upper bound, but with a new lower bound
even including possible economic gains. Figure 4 (Fig. 8.19 reproduced from TAR), shows an average of results
from six models of stabilization at 550 ppm carbon dioxide and using different baselines. The horizontal axis
indicates the different volume of reductions required to meet the 550 concentration target, based on the different
baseline scenarios. The vertical axis shows costs as a percent of global GDP. The figure suggests costs of
between one-tenth of 1 per cent and almost 2 per cent of global GDP, depending on the baseline. For
comparison, we can examine Figure 5 (Fig. 3.25 in the AR4). Unlike the TAR, this figure shows some models
independently, while the full range of results, excluding the top and bottom 10 per cent, is shaded grey. In this
figure, the GDP effects for different levels of eventual stabilization are shown. For comparison with the TAR,
we focus on the “category C” levels, or about 650 carbon dioxide equivalent (i.e, including the other gases that
were excluded in the TAR). By inspection, we observe that the upper end of the range is still around 2 per cent,
but now the lower end includes a net gain of 1 per cent.
A new development since the TAR is the use of endogenous technological change (ETC) in some models. In
contrast to previous models, where the cost of technological change was given, recently, some models determine
endogenously the optimal timing of adopting new technologies, in response to spending on research and
increased demand. However, as Pizer and Popp (2007) mention, there is little evidence about how technological
change will respond to increased spending or increased demand for emissions reductions.
Figures 6 and 7 (Fig. 11.9 in the Working Group III report) summarize the results obtained from different
studies concerning the impact on carbon prices and the effect on global GDP from cost-effectively stabilizing
atmospheric CO2 at two different concentration levels, 450 and 550 ppm, with and without endogenous
technological change. It makes a huge difference considering the possibility of endogenous technological change
or not. Without endogenous technological change the costs are much higher, whether expressed in the carbon
price increase or on global GDP. Also, the more stringent the target, the higher is the cost. By inspection, we
may conclude that to stabilize concentrations at 550 ppm global carbon prices should rise to $9-$30 per metric
ton of CO2 in the next 20 years, to $20-$90 per metric ton by 2050, and continuing to rise thereafter.
Figure 4: Average GDP and CO2 reductions in 550 ppmv stabilization scenarios; year 2050 (labels identify
different scenario groups)
Source: TAR, Working Group III, Chapter 8
Figure 5: Relationship between the cost of mitigation and long-term stabilization targets (radiative forcing
compared to pre-industrialization level, W/m2 and CO2-eq concentration)
Note: Individual lines show selected studies; grey shaded region represents the 80th percentile of scenarios.
Source: AR4, Working Group III, Chapter 3.
Figure 6: Averaged effects of including ETC (Endogenous Technological Change) on carbon prices
Source: AR4, Working Group III, Chapter 11
Figure 7: Averaged effects of including ETC (Endogenous Technological Change) on GDP

Source: AR4, Working Group III, Chapter 11
Given the characteristics of the top-down models, namely, its ability to capture the interaction of mitigation
policies with other fiscal policies, trade and other possible market failures, as well as to consider how costs are
spread over time, all the studies on mitigation portfolios and macro-economic costs presented in the AR4 are
based on top-down modeling, as stated in Box SPM.3, pg 8 of SPM. These models also allow deriving
trajectories for emissions that stabilize greenhouse gas concentrations in the atmosphere.
According to Table 3 (Table SPM.4), the rows show the estimated costs of stabilization at three different
ranges of concentrations of carbon dioxide and other gases, expressed in carbon dioxide equivalent. These ranges
are given in the first column. Costs in the second and in the third column are expressed as a share of global GDP,
which is a measure of the opportunity cost to control emissions. The last column represents the corresponding
effect on growth. Therefore, for the lowest concentration range the cost in terms of global GDP would be not
higher than 3 per cent, and for concentrations between 445 and 710 ppm CO2-eq, are estimated at between a 3
per cent decrease of GDP and a small increase, compared to the baseline. However, regional costs may differ
significantly from global averages.
Stabilization Median GDP Range of GDP Reduction of

19 20
levels (ppm CO2-eq) reduction (%) reduction (%) average annual GDP
growth rates
19
(percentage points)
21
590-710 0.2 -0.6-1.2 <0.06

535-590 0.6 0.2-2.5 <0.1
22
445-535 Not available <3 <0.12
Table 3: Estimated global macro-economic costs in 203016 for least-cost trajectories towards different long-term
stabilization levels 17,18
Notice that, since the cost estimates obtained from top-down studies use a global least cost approach, as
stated in Box SPM.3 of the SPM, they are underestimated, by lack of participation of countries and the absence
of efficient policies.
3.3. Integrated Assessment Models
In order to relate mitigation costs and benefits (avoided damages), Integrated Assessment Models (IAM) of
Climate Change have been developed in the literature. The IAM are computer-simulated models consisting of
optimal economic growth models with endogenous greenhouse warming, and are variants of the relatively
aggregated optimal growth model with GHGs accumulation and damage developed by Nordhaus in the 1990s
(DICE Model). The main question addressed by these models is the derivation of the optimal level of emissions
reduction, that is, the level of emissions reduction that maximizes the present value of net benefits. There are
different examples of such models, as Cline's (1992), PAGE2002 by Hope (2006) and more recently, the CCSP
15
and the EMF-21 studies. Integrated assessment models differ considerably in their representation of impacts,
16
as examined in Tol and Fankhauser (1998).
Nordhaus (2006), based on an updated version of the DICE model to 2005 data and 2006 prices (DICE-2006
17
model), estimates that the optimal price on GHG emissions in 2005 is $17.12 per metric ton, rising over time to
$84 in 2050 and to $270 in 2100. The recent Stern Review obtains a larger estimate for the social cost of carbon,
about $85 per metric ton of CO2, which is clearly an outlier in the marginal damage cost literature.
The vast array of results obtained from different studies is the consequence of different assumptions,
reflecting the underlying uncertainty that characterizes climate change and its economic impacts.
Given the large time-spans involved (many decades or even centuries) and the fact that the benefits from
mitigation will only be felt after a delay of decades, what to do now is highly sensitive to the discount rate used.
This has opened a debate about what should be “ethical” values to use for the rate of time preference and the
elasticity of marginal utility for intergenerational discounting in the deterministic version of the Ramsey
18
equation, on which most of the studies on the economics of climate change have been based. To motivate the
discussion, in particular, to show how the choice of basic parameter values by the researcher is relevant to the
19
results, we use the figures of the Stern Review, which are far from reaching any consensus in the profession,
and can be used as an illustrative example.
According to the Stern Review, “the overall costs and risks of climate change will be equivalent to losing at
least 5 per cent of global GDP each year, now and forever [...] damage could rise to 20 per cent of GDP or
more.” These are “risks of major disruption to economic and social activity, on a scale similar to those associated
with the great wars and the economic depression of the first half of the 20th century.” It concludes that “the costs
of [...] reducing greenhouse gas emissions [to stabilize concentrations at 500-550 ppm CO2-eq] [...] can be
limited to around 1 per cent of global GDP each year.” As a consequence, the authors conclude in the “Summary
15. The CCSP study provides an analysis of stabilization scenarios developed for the federal government's inter-agency Climate Change
Science Program (CCSP), while EFM-21 is the Stanford Energy Modeling Forum's EMF-21 study. These studies modeled scenarios that
assume U.S. participation in a cost-effective global effort to stabilize atmospheric GHG concentrations at different levels. In contrast, two
other recent studies by the Energy Information Administration (EIA) and the MIT, only consider a range of emission targets consistent with
recent U.S. policy proposals.
16. Limitations of PAGE2002 are identified in Tol and Yohe (2006). Since our purpose is not to discuss the results of the Stern Review, we
will not address this issue in detail.
17. The optimal price of carbon, or carbon tax, sometimes called the “social price of carbon”, is the calculated price of carbon emissions that
will balance the incremental costs of reducing carbon emissions with the incremental benefits of reducing climate damages.
18. The Ramsey equation is given by r = ȡ+ηg, where r represents the annual interest rate, ȡ is the rate of pure time preference, η is the
elasticity of marginal utility or the coefficient of relative risk aversion, and, finally, g is the per-capita growth rate of consumption.
19. The economic impacts estimates of the Stern Review were based on PAGE2002.
of Conclusions” that there is a strong case for emission reduction, as the reported benefit-cost ratio is between
20
five and twenty. These results are at odds with the economic literature on climate change. A common finding
of the integrated models that use the same basic data and analytical structure is the “climate-policy ramp” in
Pizer's words, “...in which efficient or “optimal” economic policies to slow global warming increasingly tighten
or ramp overtime.'' This result is a direct consequence of discounting: the higher the interest rate the more
willing economic agents are to increase current utility at the expense of future disutility. This is in contrast to the
Stern Review which suggests raising overall GHGs reductions by imposing a very low discount rate, thus
21
magnifying enormously the impacts in the distant future.
Figure 8: Estimates of the damage costs of climate change (left panel) and the costs of emission reduction (right
panel) according to the Stern Review and according to previous studies
Source: Figure 1, Tol and Yohe [32]
According to Yohe and Tol (2007), the estimates of the damages associated with the impact of the climate
change have never been so high and estimates of the cost of reducing emissions have never been so low (see
Figure 8). In fact, while the Stern Review discounts utility at 0.1 per cent many other economic assessments have
used rates as high as 3 per cent. This explains the critical difference in the findings of the Stern Review and the
22
other studies. As shown in Nordhaus (2006), when using a more conventional value for the discount rate, the
surprising results of the Stern Review replicate the more consensual “climate-policy ramp”. In fact, following
Weitzman (2007a), it is clear how any result can be supported, from large immediate emissions reductions to no
20. In general, in all other studies, the benefit-cost ratio has been taken somewhere in the range of about 1 per cent to 4 per cent.
21. In a review of integrated-assessment models, Kelly and Kolstad (1999) state that: “Perhaps the most surprising result is the consensus
that given calibrated interest rates and low future economic growth, modest controls are generally optimal”.
22. Just by moving from a discount rate of 0.1 per cent to 1 per cent would lower damage estimates by nearly 60per cent; moving to 2 per
cent would reduce them by another 20 per cent and moving into 3 per cent by another 15 per cent. Therefore, using the same data with a 3
per cent discount rate would produce damage estimates between 10 and 20 per cent of those obtained in the Review.
23
current costs. But the problem of what discount rate to choose in problems with these characteristics is a
difficult one. Underlying the choice of the discount rate is the huge uncertainty surrounding the problem. So the
discount rate is uncertain which implies that the deterministic Ramsey equation is not appropriate, as no
distinction is made among rates of return on various assets. As shown by Weitzman (2007b), by making the
growth rate of consumption a random variable, the stochastic version of the problem highlights how sensitive the
discount rate is to the way economic agents evaluate the impact of climate change on the different sectors of the
economy, and how this affects the outcome.
The huge uncertainty that surrounds climate change is reflected in the low and highly unknown probability
high-impact disasters associated with the evolution of the average global temperature. This determines a thick
left tail for the posterior-predictive probability distribution of the growth rate of consumption, as it is very
difficult to learn extreme bad-tails probabilities from finite samples, given that there are no enough data on
catastrophic events. In the presence of structural uncertainty, “...where there do not exist prior limits on damages,
expected present discounted utility analysis of costs and benefits is likely to be dominated by catastrophic-
insurance considerations rather than the consumption-smoothing consequences of long-term discounting at one
or another particular interest rate.” Since there is no way to assess relative probabilities of extreme events from a
finite sample, the expected utility of a Bayesian agent with strict relative risk aversion may be determined by the
incapacity of learning, unless potential exposure is limited by prior information, generating a kind of embodied
“generalized precautionary principle”. Ultimately, the recommendations of the Stern Review may turn out to be
correct.
A direct implication of the presence of structural uncertainty is that Monte Carlo simulations that have been
used in IAM to address uncertainty do not represent adequately the low-utility impacts of fat-tailed pdfs. Thus,
24
these models may “...constitute a serious misapplication of Expected Utility Theory”.
From what was said, not only more research is needed to better understand the deep uncertainty surrounding
those catastrophic events, but also it would be worthwhile to study the different feasible alternatives to deal with
those events. In this context, geoengineering, that is, deliberate climate modification by solar radiation
management, has recently been considered as a rather cheap and effective substitute of mitigation that could be
used in case of emergency in the short-run. The possibility of considering geoengineering to combat the effect of
23. The Stern Review values are ȡ =0.1 per cent, η=1, g=1.3 per cent, implying that r=1.4 per cent. Weitzman's (2007a) values are a “triplet
of twos”, such that ȡ=2 per cent, η=2, g=2 per cent, implying that r=6 per cent. While in the first case the B/C value is 4.5, in the second case
is 1/20. Therefore, the answer to the question whether it is worthwhile or not to sacrifice 1 per cent of GDP today to save 5 per cent of GDP a
century from now can be totally reversed.
24. See Weitzman (2007b), pg. 29.
GHGs concentration on earth temperature is associated with a different perspective concerning the objective of
the United Nations Framework Convention on Climate Change (UNFCCC). Instead of focusing on stabilization
of atmospheric concentrations of GHGs, it could target the need to reduce climate change risk, encompassing not
only mitigation, but also adaptation, geoengineering and R&D into new energy technologies. This is also
contemplated in AR4. (See Barrett (2007a, 2007b)).
4. How to Reduce Emissions? The Choice of the Economic Instrument
Policymakers may consider different instruments to curb down emissions of greenhouse gases. Alternatives
include emission taxes, emission quotas, abatement subsidies, tradable emissions allowances, and performance
standards. Also, the choice can be undertaken between a given instrument that has direct impact on emissions, as
in the case of an emissions trading program or, indirectly, through pollution-related goods or services, like in the
fuel tax case. All of them address the most relevant market failure in the economics of climate change, that is,
the environmental externality of global warming, forcing the economic agents to internalize the cost imposed on
society by the combustion of fossil fuels and other greenhouse gases.
When targeting CO2 emissions, there is a general consensus among economists that market-based policy
instruments should be at the centre of any climate change policy, as they are considered superior to command-
and-control approaches. Among these, most attention has focused on a carbon tax (price approach) and a cap-
and-trade system (quantitative approach), and, more recently, on hybrid trading systems. These hybrid systems
are basically tradable permit systems where both price volatility and the issue of raising government revenue are
addressed by replicating solutions from the carbon tax regime.
Despite that there are similarities between CO2 taxes and permits or tradable allowances, as both can lead to
efficient emissions reductions by establishing a uniform price on emitting, that is, the marginal cost of
abatement, and can either be administered on up-stream fossil-fuel producers or on downstream emitters, there
also differences.
With carbon taxes the price for emissions is known, and the level of emissions is uncertain, varying with
economic conditions. In contrast, in the case of tradable permits, the level of emissions is fixed by the cap while
permit prices are volatile. Under a pure cap-and-trade system permit prices can be volatile because the supply of
allowances is fixed, while demand may vary considerably from year to year with changes in the demand for
energy, in fuel prices, among others. Price volatility is costly, with very negative consequences to long-term
capital and R&D investments in low-carbon technologies that have high up-front costs (as carbon capture and
sequestration technologies), as the returns of those investments become very uncertain.
Also, taxes raise government revenue, while permits typically have not risen, as they have been allocated for
free to the regulated sector. While the revenue raised in the case of taxes can then be used to reduce the cost of
the policy, through revenue-recycling, contributing to increasing efficiency, the distributional consequences of
using those revenues in the case of the tax are very different from those associated with free allocation of
permits. In this last case, benefits are appropriated by the recipients of the allowances, typically the sectors with
higher emissions. Therefore, by freely distributing permits, the government compensates the sectors over which
the burden of the trading program has fallen. This is more difficult to achieve with the revenue generated from a
CO2 tax. Because with taxes redistributions are more difficult to be undertaken, they may increase the risk of
exemptions from the emissions control regime to industries lobbying for it. So, rent-seeking is very likely in this
case, ultimately, increasing the cost of the policy. The revenue-raising policies (taxes and auctioned permits) are
the most cost-effective, while the non-revenue-raising policies (freely allocated permits) have distributional
consequences that may reduce political resistance.
Emissions trading systems require new institutions to function properly, namely, markets where firms can
buy and sell permits and obtain information about current and future permit prices as well as institutions that
monitor, report and verify emissions which represents an additional burden. However, the administration of a
carbon tax, namely, monitoring and enforcement, is also very difficult, as countries could use compensatory
schemes to offset the impact of the tax.
In a context of uncertainty, whenever costs are highly nonlinear compared to benefits, Weitzman's (1974)
seminal article has shown that expected welfare losses are smaller from a price-based instrument (carbon tax)
rather than from a quantity-based instrument, as in the case of a system of tradable emissions allowances.
Theoretical and empirical work by Kolstad (1996), and Newell and Pizer (2003) have shown that the marginal
benefit schedule for emissions reductions is relatively flat, as benefits are related to the stock of GHGs while the
costs are related to the flow of emissions. Thus, price-based instruments perform better than quantity-based ones.
However, quantity-based regimes, like cap-and-trade, have in general been preferred, not only in EU but also in
US, as the most recent debate makes plain. Besides the easy way of providing compensation to regulated sectors
that a cap-and-trade system allows when compared to taxes, and the lack of credibility of governments in using
efficiently the revenues generated by taxes, policymakers typically prefer to have a given amount of emissions
reductions guaranteed over time, that is, quantity-based instruments. In the 110th US Congress several bills are
discussed, calling for mandatory caps on GHG emissions and either mandate or recommend a cap-and-trade
system. Moreover, they also include provisions to accelerate research, development and deployment of climate-
friendly technologies.
This does not preclude the existence of voices in favor of a carbon tax as a better strategy to reduce GHGs
emissions. (See Nordhaus (2007)).
Recently, hybrid trading schemes gained momentum in the international arena. These hybrid schemes have
the ability to overcome the disadvantages of a cap-and-system, namely, price volatility and the incapacity of
raising government revenue. While price volatility can be addressed by introducing cost-containment
mechanisms, such as a “safety-valve” together with allowance banking and borrowing, government revenues can
be raised by auctioning at least part of the tradable allowances, instead of allocating them for free.
The “safety-valve” works as a cap on permit prices to be used whenever demand for permits and abatement
costs are high, as firms can buy additional permits from the government when the permit price achieves a certain
trigger level. Banking allows firms to keep allowances whenever the demand for permits is slack because
abatement costs are low. Thus, firms abate more, keeping the permits to be used in the future when prices are
expected to be higher. Therefore, a floor under permit prices is created. Also, borrowing allows firms to
smoothen the impact of transitory high prices by borrowing permits from the government whenever prices are
high and paying them back through more stringent emissions control in the future.
Finally, both regimes can accommodate incentives for downstream activities that partly offset emissions
releases through tax credits or emissions offsets provisions, such as carbon capture and storage at power plants
and industrial facilities, increasing forested area, among others. Credits can be used to encourage emissions
reductions from activities outside the scope of the cap-and-trade system, lowering costs. Carbon capture and
storage technologies (CCS) and biological sequestration through land use changes, namely, increasing
afforestation, improving forest management and decreasing deforestation are good examples. In these last cases,
by creating a market for carbon, the Kyoto Protocol has indirectly contributed to the development of a market for
one of the most valuable services of forests, that is, their contribution to the carbon cycle through carbon
25
sequestration.
4.1. The European Union Trading System (EU ETS)
The Kyoto Protocol includes a provision for an international hybrid cap-and-trade system among EU
countries, as well as two systems of project-level offsets, the Clean Development Mechanism (CDM) and the
25. In general, these credits play a very important role through the Joint Implementation (JI) and Clean Development Mechanism (CDM)
provisions of the Kyoto Protocol.
Joint Implementation (JI). Therefore, the cap-and-trade system is a regional system, the so-called European
Union Emissions Trading Scheme (EU ETS) for CO2 allowances, and it is the largest in the world.
Despite some similarities with US emissions trading schemes, as the US sulfur dioxide (SO2) trading system,
under provisions of the US Clean Air Act, the EU ETS also presents significative differences. While the
European Commission decides who participate in the market, each member country decides the national cap
level, allocates the permits to the different sources, creates the institutions that monitor, report and verify its
emissions, and makes choices about some basic aspects of the design, such as, auctions and banking. Therefore,
it is a rather decentralized system, when compared to the SO2 US trading system.
Within the EU, each country has its own target for emissions reduction according to its own contribution to
the Kyoto Protocol under the EU-burden sharing agreement, which is decided centrally. Each member country
submits for approval the National Allocation Plans (NAP), which are now finalized and approved by the
European Commission for Phase 2.
NAPs describe three decisions each country must make: (i) how much of a country's Kyoto target will be
assigned to the trading sector, (ii) how much of the cap will be allocated within the trading sector (for example,
electric power versus iron and steel)? and (iii) how will the sector allocation be divided among the different
firms? Notice that the part of the Kyoto target not allocated by the EU ETS must be met by the non-trading
sectors, as transportation and agriculture. Therefore, while the demand of allowances is centralized, the supply
depends upon the decisions of all member countries.
The EU ETS began in January 2005, having operated for two years with reasonable success, despite some
problems at specific points in time. The cap-and-trade system is a downstream system, not an economy-wide
(upstream) one, regulating about 11 500 emissions sources, including oil refineries, iron and steel, energy
production (including electric power generation and refining), cement, glass, ceramics, and pulp and paper.
These sources account for about 46 per cent of all EU emissions, not covering sources in the transportation,
26
commercial or residential sectors.
27
The downstream point of regulation is somehow responsible for that. While in the First Phase (2005-
2007), a trial period, trade was limited to CO2, in the Second Phase from 2008-2012, the Kyoto commitment
period, other greenhouse gases will also be included.
As pointed out by Kruger, Oates and Pizer (2007), some important consequences result from this innovative
structure. Not only the different member states cannot predict the price of the allowances based on the NAPs, but
26. Aviation will be included in 2011.

27. One of the reasons for not including transport is the large revenue that would be lost by governments from taxing fuels.
also the allocation decision between the trading and the non-trading sectors is not cost-minimizing, as the
member countries do not have enough information to decide efficiently at the time the NAPs are set, as the total
amount of allowances is centrally determined while the allocation decision between the trading sectors and
within each trading sector of each member country is decentralized.
While in Phase 1 it was not possible to auction more than 5 per cent of the total allowances allocated to a
given member country, in Phase 2 the upper limit increased to 10 per cent. However, only four member states
opted for an auction: Denmark 5 per cent; Hungary 2.5 per cent; Ireland 0.75 per cent; Lithuania 1.5 per cent. In
Phase 2, according to the NAPs, there are a few more countries choosing an auction, from 0.5 per cent in Ireland
and the Flanders region of Belgium to 7 per cent in the UK. However, neither Germany nor France have chosen
to auction allowances.
28
Despite the arguments favoring auctioning rather than free allocation of allowances, the majority of
countries have chosen not to engage in auctions. Recently, the debate in the literature about this issue has been
undertaken in the context of the analysis of the data available from the First Phase.
In fact, free allocation has been used by member states as an instrument to protect firms or sectors from the
internalization of the cost of carbon, leading in some cases to significant windfall profits, namely, in the power
sector, as prices of electricity increased due to higher energy prices and the price of carbon. In countries with
liberalized power markets, firms have passed through most of the opportunity costs, making huge profits at the
expense of final consumers. On the other hand, in regulated markets, as in France, Spain or Portugal, the retail
prices are set by government contracts or regulators, implying that prices are not adjusted. Consequently, price
signaling is lost, undermining the incentives to reduce CO2 emissions both by firms and final consumers.
Other arguments often used against auctions and in favor of free allocation are the protection of the
competitiveness of EU industry relative to countries that do not impose CO2 controls, determining leakage
effects that may lead to re-location. Despite that many of the involved sectors, such as the electricity sector, are
not directly exposed to international competition, except in some cases, related to closure and investment
decisions, competitiveness concerns may be present in some sectors. In particular, in those that face large
increases in costs or that are exposed to competition from outside the EU. These include industries such as
cement, steel, non-ferrous metal and some chemical products. In such cases, the access to free allocation can be
relevant, at least in the short-run, sometimes at the cost of domestic leakage. Herrnstadt, Ho, Morgenstern and
Pizer (2007) confirm these results by comparing two recent studies that have estimated the competitiveness
28. See Kopp (2007), among others.

29
impacts of the EU ETS after making them compatible. After taking into account the free allocation of permits,
the net cost burdens fall significantly (85-90 per cent) in the case of basic oxygen furnace (BOF) steel and
cement, reflecting the substantial primary fuel consumption and only modest use of electricity in these industries.
In contrast, electricity-intensive industries with important indirect emissions from electricity use, like electric arc
furnace (EAF) steel, only show a modest decline in the net cost burden (10 per cent). Since the EU ETS
allocation is based on direct CO2 emissions, not considering cost increases from increasing electricity prices, the
result follows. Also, aluminum producers do not gain from free allocation as their increase in costs is mostly due
to increasing electricity prices. Other industries with a mix of direct and indirect emissions, such as segments of
the pulp and paper industry, fall in between. As expected, when cost pass-through rates are considered the
impacts are still lower, in general. However, in the case of the highly competitive aluminum industry prices
cannot be adjusted (zero cost pass-through rate). Thus, the largest competitive effects are felt in this case.
Therefore, other instruments can be used with efficiency gains, as export taxes levied by competing countries,
sectoral agreements, among others. Moreover, it is clear that the potential impacts on competition depend upon
the rule used to allocate (free) allowances.
An additional problem with free allocation relates to the presence of perverse dynamic incentives.
Independently of the rule that is used to freely allocate the allowances among the individual installations within
the different sectors, the fact that governments cannot commit credibly to the allocation for more than one
commitment period, determine that allowances are allocated sequentially. Therefore, firms are given a clear
incentive to act strategically, in order to maximize the number of future allowances, as what they did in the past
determines what they get in the future. This updating problem arises from expectations about future free
allowance allocation.
The allocation of allowances to the different individual installations in the First Phase was based on their
shares of emissions within the sector, defined by some baseline period that sometimes extended back as far as six
30
years. Obviously, this allocation rule creates an incentive to emit more now in order to obtain more allowances
in the future. Also, if free allocations to the existing installations are relatively generous, while allocations to
new installations are more restrictive, as it is the case in many member states, in principle, incentives do not
favor plant modernization and/or replacement of old ones.
In many member states allowances are given for free to new entrants, by setting aside or reserving a certain
number of allowances for those cases. In general, distribution was made based on a “first-come-first-served”
29. One study was conducted by McKinsey Company and Ecofys for the European Commission and another one was conducted by Reynaud
for the International Energy Agency (IEA).
30. Benchmarking was difficult to establish due to the large heterogeneity observed.
basis, but what countries do if the reserve is exhausted differs from country to country, as well as if the new
entrant reserve is not fully used. In general, those who arrive late have to go to the market. However, in the case
of Italy and Germany, the government will buy permits in the market to all new entrants. Therefore, in order to
support new investments, an investment subsidy is provided to these new entrants, as governments typically
claim. In the case the reserve is not fully used, some countries will sell those permits on the market, while others
like Germany, France, Spain, and Portugal representing 23 per cent of the total new entrant reserves, have
decided to eliminate them. Therefore, competition between the incumbent firms and the potential new entrants is
eliminated.
However, the rule that is used to allocate those allowances to the new entrants determines the nature of the
incentive given to the new firms. In general, the rule was not the same as the one used for the incumbents, as the
new entrants were not producing emissions before, and it was different from country to country. All the states
showed an intention to do the allocation according to some benchmark based on some definition of best practice
or technology multiplied by expected production or new capacity. In the electricity sector, they typically differ
by fuel or technology used. For instance, while in the case of Spain, Denmark and UK the benchmark was the
same for all fuels, Germany and Italy differentiate by fuel.
Finally, closure provisions were set such that firms that close should give up their allowances. This prevents
firms from shutting down and/or relocating, providing the wrong incentive by biasing technology choices against
CO2 reduction and against innovation. Some member states, including Portugal, have adopted a so-called
transfer rule according to which the owner of a closed facility could transfer unused allowances to a new one
within the same member states. In this case, it is not eligible for allowances from the new entrant reserve. This
basically attempts to prevent major impacts on employment and on the level of activity.
All of these problems could be reduced or even removed in some cases if, instead of using free allocation,
based on grandfathering (that is, based on past behavior), benchmarking or any other rule, the countries have
chosen to make a greater use of auctioning. Based on the recently released information on the approved NAPs
for the first Kyoto commitment period, it does not seem that this is occurring.
Moreover, while there is a common requirement that banking will be allowed after 2008, member countries
are free to choose whether and how to allow banking in each of the two Phases. However, since any banked
allowances by any member country will reduce its cap in the Second Phase, it is not expected that they will play
an important role. Despite that member states cannot introduce a “safety-valve” mechanism, noncompliance is
penalized by the European Commission. The fine for noncompliance is of 40 euros per carbon ton in the First
Phase, and of 100 euros after 2008. However, installations will always have to make up their missed emission
reductions during the next year, implying that fines are not a “safety-valve”.
Finally, member states have the possibility of establishing domestic compliance procedures, including
monitoring, reporting and verification, which is of increasing relevance as new member states with cultural and
institutional differences are joining the EU ETS. However, this issue turned out to be of great relevance in the
First Phase. During this period, the prices of permits presented high volatility, which have culminated in April
2006, in which the price has declined about 54 per cent in four days. While in the middle of January 2005 the
price was around 6.8 euros, the prices increased substantially during that year, achieving 30 euros six months
later. Between July 2005 and April 2006 the prices were in the 21-30 euros range. The huge decline in April
2006 was due to a lower than expected reported emissions and the resulting surpluses of allowances in many
member states. As member countries began reporting compliance results and verified emissions were lower than
31
expected prices dramatically decreased. This has occurred before the European Commission released the data,
as countries were allowed to disclose their own verified emissions. Therefore, the adjustments on prices had
already taken place by the time the European Commission disclosed the data. The way reporting was conducted
32
has also contributed to the dramatic price fall. This could suggest that some overallocation was present in this
First Phase. However, Ellerman and Buchner (2008) consider larger than expected abatement as a possible
alternative explanation. The prices have recovered in mid-August to 17 euros, and then felt again. Sustained low
prices have characterized the current year (See Figure 13).
EU ETS may increase the supply of allowances by obtaining offsets for EU emissions from outside Europe,
under the Kyoto Protocol provisions, that is, CDM and JI. Therefore, the market for offsets is intimately related
to that of EU allowances. The purchase of carbon offsets helped alleviating the burden of stricter targets in the
non-trading sectors, as generous allocations to the trading sector have required. In fact, it is almost consensual
that the only way the EU will be able to meet Kyoto is by buying emissions allowances to Russia and Ukraine,
as well as by taking advantage of the CDM and JI provisions.
The effort to improve upon the design of the EU ETS is important not only to increase efficiency in the
region, but also because it is very likely that other national or regional regimes will be linked to the EU ETS in
the future. Recently, proposals for similar schemes have followed in other countries, as in Norway, Canada,
Japan, as well in the north-eastern US states, known as the Regional Greenhouse Gas Initiative (RGGI), and
31. Six countries, Austria, Ireland, Italy, Greece, Spain and the United Kingdom had annual emissions larger than annual allowances.
32. The European Commission has taken new decisions concerning monitoring and data disclosure by member countries in July 2007
(C(2007) 3416).
most likely, at the federal level in the near future. The issues raised by linking are discussed in Kruger, Oates and
Pizer (2007).
4.2. The Role of Technology
The long-term nature of the climate change problem makes technological change a central issue in policy
design. Integrated assessment studies have analyzed how to design caps on GHG emissions that would lead to
stabilization of atmospheric concentrations of greenhouse gases. These studies have shown that the time scale
involved in order to deal with climate change is measured in centuries. Also, they have found that cost-
effectiveness is achieved through relatively small emission reductions in the medium run, and much larger
emission reduction in the more distant future. The cost of achieving significant greenhouse gas emissions
reductions in the future will depend on the availability of low or zero cost emitting technologies. These emission
reductions in the long-run require the development of new technologies not yet available, implying a complete
33
changeover of the capital stock. In order to provide incentives for firms to invest in the development of those
technologies, credible commitments to meeting long-run emissions targets without imposing unnecessarily high
near-term costs have to be provided. Since the caps to be set in the future are those that are relevant for
innovation, and since it is not possible to credibly impose in the present strict future targets, additional policies to
increase government funding or incentives for private funding of such a research are necessary (grants and tax
incentives, prizes, R&D consortia). Some recent proposals reflect this concern. The case of the recently
presented “European Strategic Energy Technology Plan” (SET-PLAN) by the European Commission is a good
example. Underlying this proposal is a R&D protocol in the line of the so-called Technology oriented
Agreements (TOAs), as referred below. In these proposals there is a clear intention of linking the willingness to
pay to finance R&D to the diffusion of any technology that results.
Therefore, in addition to the appropriability R&D problem according to which social benefits from invention
of a new technology can only be in part appropriated by the inventor, the legitimate expected reward of
undertaking innovation also depends on a credible announcement of future caps capable of generating adequate
price signals to which governments cannot commit, raising a dynamic consistency problem, as discussed in
Montgomery and Smith (2007). Notice also that from the policymakers' perspective, it is important to keep
33. One frequently cited study by Pacala and Socolow (2004) found that stabilizing carbon dioxide CO2 concentrations at about 500 ppm
would require a 50 per cent reduction in global CO2 emissions below baseline within the next fifty years and then a more significant decline
over the following fifty years. This would require substantial technological development.
flexibility in adjusting the long-term caps as learning about benefits and costs of mitigating climate change is
taking place.
Despite that market-based instruments, as carbon taxes and emission trading systems, are typically superior
from a cost-effective point of view, traditional regulations, such as technology standards that dictate the use of a
specific technology, and performance standards that limit emissions generated per unit of output, the so-called
command-and-control, are often proposed as alternatives or complements to emission taxes or tradable permits.
In fact, performance or technology standards in general do not impose an economy-wide carbon price,
preventing an efficient distribution of the regulatory burden across economic agents. Also, the incentives to
reduce emissions are limited to specific sectors. In contrast to market-based instruments, conservation is less
promoted under those command-and-control instruments. Therefore, while some sources may be out of
regulatory control, there are often overlapping incentives for reduction in others. However, performance
standards are frequently used as complements to market-based instruments, either to address additional market
failures and/or because the price incentive to reduce CO2 emissions under the market-based system does not
34
reflect the full value of those reductions to society. Also, politicians may object to higher energy prices and
reduced competitiveness of energy-intensive industries that result from appropriately pricing emissions
reductions.
4.3. The (Short) Experience of EU ETS
Recent data released by the European Environment Agency (EEA) show that emission of GHGs decreased
by one and a half per cent and by 7.9 per cent in the context of the EU-15 and EU-27, respectively, between
1990 and 2005. The overall EC GHG emission trend is dominated by the two largest emitters Germany and UK,
accounting for about one third of total EU-27 GHG emissions. While in Germany it was the increasing
efficiency in power and heating plants that explains the significant decline in emissions, in the UK it was mainly
due to energy liberalization of energy markets and the subsequent fuel switches from oil and coal to gas in
electricity production and N2O emissions reduction measures in adipic acid production. Italy and France were
the third and fourth largest emitters both with shares of 11 per cent. Italy's GHG emissions were about 12 per
cent above the 1990 level in 2005, due to road transportation, electricity and heat production and petrol refining.
While in France large reductions were achieved in N2O emissions, CO2 emissions from road transportation
increased considerably between 1990 and 2005. Spain and Poland were the fifth and sixth largest emitters in the
34. See Stavins (2007).

EU-27, accounting for 9 per cent and 8 per cent of total EU-27 GHG emissions, respectively. Again, this was
due to emissions increases from road transportation, electricity, and manufacturing industries.
Between 2004 and 2005, in the EU-15, emissions of GHGs decreased by 0.8 per cent, mainly due to
decreasing CO2 emissions of 0.7 per cent. Germany, Finland and the Netherlands contributed most to the EU-15
reduction in absolute terms. The reduction of CO2 emissions drove the overall decrease of greenhouse gas
emissions in those countries. Germany reduced emissions by 2.3 per cent as a consequence of shifting from coal
to gas in the production of public electricity and heat. Moreover, emissions from road transportation and from
households and services also declined substantially. Finland reduced emissions by 14.6 per cent, mainly due to a
significant decrease in the use of fossil fuels in the production of public electricity and heat, having relied on
imports. Coal, namely, has decreased. Finally, the Netherlands reduced emissions by 2.9 per cent, as less fossil
fuel was used for the production of public electricity and heat. Other EU-15 countries that have reduced
emissions are Belgium, Denmark, France, Luxembourg, Sweden and UK. Therefore, in absolute terms, the main
sectors contributing to emissions reductions were public electricity and heat production, households and services,
and road transportation. Spain was responsible for the biggest increase in GHGs emissions in absolute terms,
corresponding to 3.6 per cent. In fact, GHG emissions were 52 per cent above the 1990 level in 2005. This
increase originated mainly from public electricity and heat production, and was due to a rise in electricity
generation from fossil thermal power stations (17 per cent) and a decrease in electricity generation from
hydropower plants (-33 per cent). Other countries also have increased emissions in that period: Austria, Greece,
Ireland, Italy and Portugal.
In summary, these figures show that, in general, the southern European countries are performing less well
than the northern European ones.
Figure 9: Short and long positions by member state in percent (average 2006/06)
Source: Ellerman and Bruchner (2008)
Figure 10: Short and long positions by by EU-wide sectors

First Phase
Based on the data released at the installation level for verified emissions and allowance allocations for the
first two years of the First Phase, CO2 emissions were on average 3 per cent lower than the number of
allowances distributed to installations for those years. Based on Ellerman and Buchner (2008), we may conclude
about the different situations faced by the member states (See Figure 9). The group of countries where almost all
the installations are long and by an amount that is larger than 15 per cent of the average annual allocation include
35
Lithuania, Estonia, Latvia, Luxembourg, Slovakia, France and Czech Republic. The second group composed
by Hungary, Sweden, Poland and Finland show net long positions between 10 per cent and 15 per cent of the
allowance allocation, but has more short positions, especially in Sweden. The third group consists of the
Netherlands, Belgium, Denmark, Cyprus, Portugal, Germany, Slovenia and Greece that are long on balance but
by modest amounts. The final group includes Austria, Spain, Italy, Ireland, and the UK that are on balance short.
The gross shorts are situated mostly in four countries, the UK, Spain, Italy and Germany. The net balances imply
that 56 per cent of the demand for international trading came from the UK while Italy represented 22 per cent,
Spain 17 per cent, and the remaining 5 per cent from Ireland and Austria. The potential suppliers were more
evenly distributed with Poland, France, Germany and the Czech Republic accounting for 64 per cent of the total.
By economic sector, and based in Figure 10, the power sector is in the aggregate moderately short (3 per cent),
while all the other sectors are long by more significant percentages. Among these industrial sectors, ceramics,
brick and tiles; iron, steel and coke; and pulp and paper, are long by more than 15 per cent of the allocation to
these sectors. In 2005, almost all of the demand for allowances came from the power sector, as well as about half
of the potential supply. When combining sectors (power and heat and all others) and regions (EU-15, and EU-8)
36
in Figures 11 and 12, it is possible to conclude that the power and heat sectors in the EU-15 are the only
presenting an overall net short position. Given the delay in establishing the registries in Eastern Europe,
according to the authors, it is very likely that most of the short position was probably covered by purchases from
the EU-15 industrial sectors.
In what concerns the CDM and JI provisions, many criticize the nature of the projects supported, as only one
third are energy-related projects. Also, their real impact is not clear, as it very difficult to establish baselines for
individual projects. The number of projects has increased substantially during the First Phase, as well as the
amounts traded. Also, the capitalization of carbon funds worldwide is a consequence of that fast-growing
35. Gross long (short) represent the differences for all the installations having long (short) positions. Each aggregate has either a net long or a
net short position indicated by the darker shaded area and is equal to the difference between the gross long and gross short data points for that
aggregate.
36. EU-15: France, Germany, Italy, Belgium, Luxembourg, the Netherlands, UK, Ireland, Denmark, Greece, Spain, Portugal, Austria,
Finland, and Sweden. EU-8: Czech Republic, Latvia, Poland, Hungary, Lithuania, Slovakia, Slovenia and Estonia.
market. In general, the market is dominated by the Europeans, followed by Japanese buyers, and by the private
sector. UK and US funds have also been increasing.
Figure 11: Short and long positions by by EU-wide sectors in absolute terms
Figure 12: A breakdown of short and long positions by major sectors and regions
In what concerns the recipients, China represents the most important one, having captured almost two-thirds
of the market for project-based transactions in the first two years of the First Phase. Latin America and Asia
follow, and, finally, Africa. However, the majority of the projects in China respect to the destruction of HFC-23,
which in any case would have occurred; without HFC-23, China still dominates. In fact, China, Brazil and India
account for about 80 per cent of supply. Besides HFC-23 destruction projects, methane capture from landfill gas
and coal mines are the second largest, while renewable energy projects constitute the third more important.
Energy efficiency, including fuel switching, and N2O destruction follow. Therefore, projects abating gases other
37
than CO2 account for at least 70 per cent of the volume. Therefore, Land-Use, Land-Use Change and Forestry
(LULUF) projects are not very important, as noncarbon projects benefit not only from a higher price because of
conversion factors, but also because the maturity of these projects is achieved in a shorter period. Besides,
LULUF activities are limited by the EU ETS.
Second Phase
In the case of Portugal, and according to Kyoto targets, GHGs emissions should not exceed 27 per cent of
those registered in 1990 for the first-phase period agreement, that is, 2008-2012. Therefore, the country should
not emit more than 76.39 million tons of carbon dioxide per year between 2008 and 2012. To achieve that target
there are three main policy instruments: the PNAC (Plano Nacional para as Alterações Climáticas) which
defines national policies and measures to reduce GHGs emissions in the different sectors, the PNALE (Plano
Nacional de Atribuição de Licenças de Emissão) which can be applied to a set of heavy emitters, and, therefore,
included in the CELE (Comércio Europeu de Licenças de Emissão), and, finally, the Portuguese Carbon Fund
created by law in 2006 to develop activities to obtain carbon credits by investing in the flexibility mechanisms of
the Kyoto Protocol.
Initially, the forecasts for the total of national emissions for the average year of the period 2008-2012, 2010,
showed that Portugal would emit 87.96 million tons per year. After taking into account all the policy measures
implemented up to January 1, 2005, including the abatement of 3.36 million tons per year from changes in
emissions/carbon sequestered from Forestation, Reforestation and Deforestation (article 3.3 of Kyoto Protocol),
the amount reduced to 84.61. Therefore, there was still a negative balance of 8.22 million of tons per year.
Based on these figures, it was clear that the country would not be able to comply, without additional
intervention. In this context, the PNAC 2006 has considered additional measures that are expected to abate 3.69
million tons a year to that amount, which include the activities from Forest Management, Agricultural
37. See Lecocq and Ambrosi (2007).

Management and Grazing Management under the conditions of article 3.4 of Kyoto Protocol that Portugal has
decided to consider.
Transportation, energy production, namely, electricity and heat production, households and services, as well
as the industry are the sectors that contribute the most to the national balance of GHGs emissions in the period
1990-2010. However, agriculture and waste disposal have their weights reduced in that period.
Despite that the relative position of the different sectors in the balance does not change significantly, after
considering the potential of the additional measures to help meeting Kyoto targets, energy and LULUF activities,
namely, forestry, have reduced their contribution in 3 per cent and 5 per cent, respectively. In this context, the
expected contribution of the sector of renewable energy is important, as well as the role played by natural carbon
sinks as forests.
Figure 13: Price developments on the EU carbon market (OTC Market: Nov 2005-Sept 2007)
In January 2007, some of the targets defined by the PNAC 2006 were revised. The new 2007 targets reflect
Portugal's intention to further increase the part of energy originated in renewable sources from 39 per cent to 45
per cent till 2010, the increase of the part of biofuels from 5.75 per cent to 10 per cent in the transportation
sector, and a 5 to10 per cent substitution of coal by biomass in thermo-electric energy power plants. These
measures are expected to contribute to an additional 2 per cent improvement in the national balance translating
into an additional reduction of 1.56 million tons per year, determining a deficit of 2.97. This deficit will be
compensated by making use of the flexibility mechanisms, either by imposing additional reductions on the
activities under the conditions of the EU Emissions Trading System (EU ETS), or through investments supported
by the Portuguese Carbon Fund. In the context of the EU ETS, Portugal has chosen to allocate all allowances for
free.
The purpose of Portugal's Carbon Fund is mainly to take advantage of the flexibility mechanisms of the
Kyoto Protocol. By investing in mechanisms for so-called clean development, or projects that help reduce
pollution in other countries, as in developing natural carbon sinks as forests, credits can be generated, helping to
meet Kyoto. In 2007 the Fund has invested 25 million euros in other carbon funds (3 million in the Luso Carbon
Fund, 10 million in the Carbon Fund for Europe of BIRD, and 12 million in the Asia Pacific Carbon Fund) that
invest in projects that create carbon credits. In the second semester of the year, projects will be directly
supported by the Fund both in Portugal and in developing countries, as Brazil, Mexico, Argentina, Colombia, El
Salvador, Mozambique, Guinea and Tunisia, among others.
In this context, Portugal is not different from the other EU countries, in particular, from Spain. This country
was allowed a 15 per cent increase in emissions relative to 1990, as the economy was growing faster than other
EU members. In order to comply with Kyoto, Spain is required to offset an expected 37 per cent increase in its
own emissions between 1990 and 2008-12, implying a negative balance of 22 per cent. To this end, the Spanish
government, besides promoting efficiency in energy use and clean energy sources, is also relying heavily on
investing in carbon credits and clean energy projects in developing countries (Brazil, Colombia, Chile, Peru),
which will contribute 20 per cent to that objective. In fact, by taking advantage of the CDM, the energy company
Endesa will invest around 3.2 billion dollars to expand hydroelectric plants in South America as a means to
compensate for the emissions of its fossil fuel-fired plants in Spain. The remainder 2 per cent is expected to be
cancelled out through reforestation projects.
Recently the European Commission has completed the assessment of the NAPs for the 25 member states that
have been part of the EU ETS since 2005. Portugal was one of the last ones, while there still remain those of
Bulgaria and Romania, which will be finalized shortly. According to the data released, the European
Commission has authorized only 89 per cent of CO2 emissions for this Second Phase relative to what was
proposed by the member states. The figures show that emissions will be reduced by 2 per cent relative to 2005
38
and are 9.5 per cent lower than what was established to the First Phase. In the case of Portugal, the European
Commission is requiring Portugal to reduce its proposed cap by 1.08 million tons of CO2-eq per year, 3.1 per
cent lower than the last Portuguese proposal, implying a total of 34.81 million tons of CO2-eq per year, instead
38. Seven countries, Poland, Czech Republic, Estonia, Latvia, Lithuania, Slovakia and Hungary have announced that they will appeal in
court. These countries have proposed in their NAPs caps above their reported levels in 2005. Consequently, the European Commission has
decided to keep the 2005 levels instead.
of the proposed 35.82. The proposed amount is divided between 30.5 to be allocated to the incumbent
installations (33.1 in 2006), while the remainder 4.3 will be reserved to new installations. In case the latter are
not used they will be cancelled. Some sectors, namely, ferrous metals, cement, glass and ceramics, have
presented some doubts concerning their reduction potential and/or competition within the sector, and the number
39
of allowances that are allocated to them.
Also, the European Commission has raised some doubts about the accountability of net carbon sequestration
by forests, through increased forested land, reduced afforestation (Article 3.3 of Kyoto) and forest management
(Article 3.4 of Kyoto), due to the fires that have occurred in the recent years. Also, it has limited the overall
amount of Kyoto project credits from emission-saving projects in third countries to 10\per cent of the annual
allocations to each installation. A key measure to achieve Kyoto's targets is the government purchase of JI and
CDM credits. The maximum amount of allowances approved was based in the commitment of the Portuguese
government to invest a minimum value of 348 million euros from 2007 onwards from the Carbon Fund to
purchase those credits. Also, an effort has to be made to reduce waste, according to the Directive on Packaging
and Packaging Waste 94/62/EC. Finally, the treatment given to new entrants has also to be clarified. Therefore,
the Portuguese government is now in the process of making the adjustments required by the European
Commission.
Third Phase: recent developments

After the endorsement by the European Parliament and by EU Leaders at the March 2007 European Council
of the strategy on energy and climate change, the European Commission came forward, in January 2008, with a
package of concrete policy proposals, including how efforts could be shared among member states to achieve the
proposed targets. Underlying these proposals is the EU commitment to tackling climate change by taking the
lead in pointing the way to an international climate agreement after 2012. Therefore, the challenge in adapting to
the demands of a low-carbon economy with secure energy supplies should also be taken by EU as an opportunity
to boost innovation and growth through exploiting first-mover advantage.
The targets proposed in the beginning of 2007 include:
• an independent EU commitment to achieve at least a 20 per cent reduction of GHGs by 2020
compared to 1990 levels and an objective for a 30 per cent reduction by 2020 subject to the
conclusion of a comprehensive international climate change agreement;
39. See Borrego, Martins and Lopes (2005).

• a mandatory EU target of 20 per cent renewable energy in EU energy consumption by 2020

including a 10 per cent biofuels target.
Five key principles inspire the package: respecting targets to ensure credibility, implementing monitoring
and compliance mechanisms, fairness to all member states, using competitiveness to minimize the costs to the
European economy, promoting a comprehensive international agreement to cut greenhouse emissions, and
starting to work now to cut global emissions by half by 2050, namely, by stimulating technological development,
encouraging innovation.
The tools to deliver the targets include an improved Emissions Trading System (ETS), an emission reduction
target for industries not covered by the ETS, legally enforceable targets for increasing the share of renewable
resources in the energy mix, and a goal to increase energy efficiency. Besides, a proposal for a legal framework
on Carbon Capture and Storage (CCS), a Communication on the demonstration of CCS and new guidelines for
environmental state aid were also part of the package.
Based on the experience of the period 2005-2007, the EU ETS was subject to review. The proposed review
has also implications to the EU-wide GHGs reduction commitments. Currently, the member states propose
through their NAPs the total GHG emission cap for their ETS sectors and the individual allocations to the
different installations, which are subject to approval by the Commission. Therefore, the GHG emission cap for
the sectors not covered by the EU ETS and the amount of JI/CDM credits that the member states will acquire to
comply with their emission reduction commitments and that their industries are allowed to use for compliance
under the EU ETS are implicitly determined.
However, several problems were identified with the existing NAP procedure. The incentive effect of the
current ETS has been reduced by the generous number of allowances issued between 2005 and 2007. The
structure of the ETS, with national allocation plans, has also raised the risk of distortion in terms of competition
and the internal market, introducing uncertainty, lack of predictability and reduced transparency in the allocation
process. From the different options whose impact assessments were examined, the preferred one is a single EU-
wide cap for the emissions covered by the EU ETS, with each installation's individual allocation decided by
Brussels, overcoming most of the problems detected in the previous phase. Since there will be allowances
distributed for free, the EU ETS cap has to be distributed between the different sectors EU-wide that will receive
free allocations, while for those that are auctioned the sharing of auctioning rights has to be defined. Moreover,
the emission reduction commitment of at least 20 per cent has to be first distributed between the EU ETS and
those sectors not covered by the EU ETS, and the effort sharing between member states will be limited to these
non EU ETS sectors. Hence, for Phase three, from 2013 to 2020, emission allocations for the current ETS sectors
will be reduced 21 per cent below 2005 levels by the end of the period, while the sectors not covered by the EU
ETS will need to reduce emissions by around 10 per cent compared to 2005.
In addition, auctions will be open. Any EU operator will be able to buy allowances in any member state.
However, free allocation should be progressively replaced by auctioning of allowances by 2020. Revenues
resulting from the ETS will accrue to member states and should be used to support innovation in areas such as
renewable sources, carbon capture and storage (CCS) and R&D. Part of the revenues should also go towards
helping developing countries to adapt to climate change. The Commission estimates that the revenues from the
auctioning could amount to 75 billion Euros by 2020, representing 0.5 per cent of GDP. Member states should
40
commit to use at least 20 per cent of their auctioning income for this purpose.
The power sector, representing a large part of emissions, would be subject to auctioning from the beginning
of the Third Phase, that is, 2013, while oil refiners and airlines will see free permits phased out in favor of
auctioning up to 2020, starting with 20 per cent from 2013. Overall, about 60 per cent of allowances would be
auctioned off in 2013, against 10 per cent currently. The steel, aluminum and cement industries have won a
temporary reprieve and a decision on how their permits will be allocated is postponed to 2010. This is the result
of lobbying from these industries for ongoing free permits, given that they have to compete in international
markets with China, India, and US, not subject to the same low-carbon measures. The possibility of levying
carbon tariffs on imports to protect them will await developments on a future global climate agreement beyond
2012. Also, the impact of increased electricity prices on energy intensive sectors has also to be addressed. So,
competitiveness and leakage effects are issues of concern.
Moreover, the use of carbon offset credits from Kyoto's CDM and JI markets will only be allowed to help
meet targets but will not be extended from current levels unless a global agreement is achieved. Therefore, the
current limit Kyoto credits for 2008-2012 will become the limit for 2008 to 2020.
In what concerns emissions from sectors not included in the EU-ETS, such as transportation, buildings,
services, smaller industrial installations, agriculture and waste, they will be reduced by 10 per cent of 2005 levels
by 2020, with specific targets for each member state. Some of these will be driven by EU measures, as tougher
standards on CO2 emissions from cars and fuel, EU-wide rules to promote energy efficiency, Common
Agricultural Policy or waste legislation. Otherwise, member states will be free to implement their own policies.
Moreover, member states will also have access to CDM credits covering almost one third of their reduction
effort. Also, the Commission confirms that banking from the Second to the Third Phase will be allowed.
Therefore, unused allowances issued in 2008-2012 can be used up to 2020.
40. See on this subject Cunha-e-Sá and Reis (2007), among others.
Finally, in the previous phase, the scope in terms of sectors included and gases has also limited the ability to
emission cuts. To this end, the scope of ETS will be extended to with the inclusion of aviation and GHGs other
than CO2, such as nitrous oxide and perfluorocarbons. To reduce the administrative burden, industrial plants
emitting less than 10 000 ton of CO2 will not participate in the ETS.
In March 2007, the European Council has put special emphasis on renewable energy, not only to reduce
emissions but also to improve energy security. Currently, the share of renewable energy in the EU's final energy
consumption is 8.5 per cent. An increase of 11.5 per cent is needed on average to meet the target of 20 per cent
in 2020. Given the heterogeneity among different member states, the Commission's proposal is based on a
methodology that accounts for those differences. While half of the additional effort is equally shared by all
Member states, the other half is adjusted according to GDP per capita. Also, the targets are modified according
to the proportion of effort already made by the Member states that have increased their share in the recent years.
Each Member state will develop a national action plan, where the efforts to be undertaken are stated. Since the
cost of exploiting renewable energy potential varies among Member states, as long as the overall EU's target is
met, Member states are allowed to contribute outside their own borders, by making use of transferable
guarantees of origin.
Finally, a minimum target for sustainable biofuels of 10 per cent of overall petrol and diesel consumption is
required for the transport sector. To this end, binding criteria for biodiversity and certain land-use changes are
imposed, in order to satisfy sustainability criteria. Besides, the Commission is also committed to promote
policies that favor the development of second generation biofuels.
From this brief analysis, it is clear that differentiation of target levels between member states was introduced
as an attempt to fairly distribute the required effort to comply with the proposed targets, namely, in the sectors
not covered by the EU ETS, for the deployment of renewable energy, and by allowing for partial redistribution
of auctioning rights.
The EU intends to save 20 per cent of energy consumption by 2020 by investing in energy efficiency. In this
context, Member states were required to submit by the end of June 2007 the National Energy Efficiency Action
Plans (NEEAP), where the national strategies on energy savings are presented. Portugal only very recently has
presented the NEEAP. In fact, the targets concerning the use of renewable energy sources and energy efficiency
set for the National Strategy for Energy defined in 2005 were revised in 2007, becoming more ambitious. In
particular, by 2010, energy consumption produced from renewable sources will increase from 39 per cent to 45
per cent. To this end, the wind, solar, hydro and biomass energy sources contribute significantly to that end.
Moreover, the recently presented NEEAP expects a 10 per cent reduction in energy consumption by 2015.
In the context of the EU Strategic Energy Technology Plan, one of the objectives is to ensure that future
power generation becomes near zero in carbon emissions. To this end, a communication was adopted on
“Sustainable Power Generation from Fossil Fuels” to accelerate the development and deployment of clean fossil
fuel power generation with carbon capture and storage (CCS) technologies, that is, allowing the carbon dioxide
emitted by industrial processes to be captured and stored underground. In fact, fossil fuels will remain the
primary source of energy worldwide in future decades. In particular, the target of halving 1990 global GHG
emissions by 2050 will require the use of coal, implying that emissions have to be controlled. Not only coal is a
cheaper source but is also abundant, and coal resources are distributed in regions of the world other than the
41
Persian Gulf. In particular, the United States, China and India have immense coal reserves. This explains why
the European Council has supported the choice of CCS technology for new power plants, including the setting
up of up to twelve demonstration plants by 2015. An important part of the package is that investors in CCS will
save the costs of ETS allowances faced by their competitors. Since CO2 sequestered underground using CCS
technologies will not be counted as emissions, they do not have to be covered by allowances. However,
significant investment is required. Besides public-private partnerships fed by national budgets and private sector
investment, the revenue generated by the auctioning of ETS allowances can also be used for that purpose.
In fact, the Commission has also adopted new State aid guidelines on environmental protection, which not
only broaden the previous ones but also increase the aid intensities. The new guidelines recognize that state aid
may be justified where higher production costs result in obstacles to market entry for renewable energies.
All these efforts reflect the commitment of the EU to maintain international leadership on climate change
and energy in order to secure a global post-2012 agreement on climate change at Copenhagen in 2009 consistent
with the EU 2ºC objective. At the level of the EU institutions several steps have yet to be undertaken in the near
future, “...as the deliberations by the Spring European Council 2007 should result in an agreement on these
proposals as a coherent package before the end of 2008 and consequently allow for their adoption within the
42
current legislative term, at the latest early in 2009.”
5. Designing Climate Change Policy: International Climate Agreements-Kyoto and Post-Kyoto
Kyoto Protocol was the first major international agreement on climate change. The fundamental claim for
engaging in a global agreement on emissions in Kyoto, December 1997, is the global nature of the problem, as
41. See “The Future of Coal”, An Interdisciplinary MIT Study, 2007.

42. Cover Note, Brussels European Council 13/14 March 2008, Presidency Conclusions, pg. 12.
emissions in one country affect the climate in every other. In other words, reductions in emissions in one country
are a public good. Four years after Kyoto, US announced that it would not sign it, basically because China and
India were not part of it and because it seriously harmed the United States economy.
Despite its innumerous deficiencies, the Kyoto Protocol is perceived as a first step in the policy design to
address global warming and its consequences to climate change. Therefore, a second step is expected. Since
some nations are now finalizing the preparation of the Kyoto’s Protocol first commitment period (2008-2012), a
great deal of discussion in the international arena is currently taking place. Obviously, the recent release of
scientific evidence has also fueled these discussions, as the public opinion is now more alert welcoming
government intervention, even in the US, as several recent studies clearly show.
More than a decade ago, the first international treaty on climate change, the Framework Convention on
Climate Change or FCCC, was signed. Later on, in February 2005, the Kyoto Protocol became the first
significant international effort to reduce GHG emissions. While basic economic theory suggests that the solution
to a global environmental externality like climate change requires global cooperation, the recent experience with
the Kyoto Protocol presents challenges to that theory.
Since the ambitious targets imposed by Kyoto apply only on the short-run, that is, from 2008 to 2012 and
only to industrialized countries (excluding the United States, among others) with a few exceptions (e.g, transition
economies like Russia and Poland), most Kyoto participant countries are above their targets. By taking
advantage of the Clean Development Mechanism (CDM) and investing in projects in developing countries,
overall compliance may improve. However, major concerns have been raised with respect to the architecture of
the agreement.
In this context, several issues have to be revisited. Since by 2020, developing countries with growing
43
population and wealth are likely to generate the largest share of emissions, and since they are considered to be
the largest source of low-cost emissions' reductions in global trading, climate policies cannot ignore them. In
fact, three of the five largest greenhouse gas emitters are not constrained on their emissions. China and India do
not have quantitative targets and Russia's commitment is so generous that it is unlikely to be binding before
2012. Moreover, the United States, the major contributor, has not ratified the agreement. Therefore, the Kyoto
Protocol imposes costs on sources located in the countries with emission commitments, but no costs on sources
outside those countries, thus reducing the cost-effectiveness of the policy, and increasing the costs of future
participation. Though, emissions leakage and competitiveness impacts can be caused by that imbalance, further
43. According to recent estimates by Auffhammer and Carson (2008) China's CO2 emissions has already surpassed the US in 2006.
44
reducing efficiency and the environmental benefits of the agreement. Therefore, the need to ensure the
participation of the industrialized and developing countries is crucial, which requires the appropriate incentives.
45
However, this is a major task.
An important aspect that was already mentioned before concerns the difficulty that developing countries
have in committing to limit the growth of their emissions under any effective, long-term climate policy. In
contrast to the current Kyoto Protocol where countries either have quantitative emission commitments (Annex
I/Annex B) or no commitments (Non-Annex I/Non-Annex B), some authors have proposed rules for
“graduation” into a system of quantitative emission commitment, where the “graduation” criteria typically use
per capita income or per capita emissions as the basis for determining when individual developing countries
should be required to take on commitment. Besides, the stringency of commitment could vary by per capita
emissions and income. An explicit rule that accounts for population, income, historical emissions and other
factors could be the basis for graduation. Despite that the notion of progressivity is already embodied in the
Kyoto Protocol, increasing with per capita income, by adjusting the weights of the different factors, “growth”
targets could be obtained, entailing less stringent goals and more generous permit allocations to developing
countries, keeping, at the same time, the incentives aligned in the long-run to motivate technological change and
bring costs down over time. Therefore, a change from the short-run to long-run time path of action is desirable to
allow for cost-effectiveness.
Establishing market-based institutions could promote substantial cost-savings, but the failure to include the
United States, China and India eliminate much of the potential gains from trade. When combined with a system
of international emissions trading, these less stringent targets allow developing countries to become net exporters
of emission permits, as they are the main source of cheap emission reductions. This would generate an important
wealth transfer to these countries. Despite that these transfers may trigger participation, it is not clear whether
developed countries would be willing to accept it, as it could essentially feed a huge rent-seeking rush, given the
cultural gap between developed and developing countries, with impact on the credibility of institutions in
developing countries. Several examples illustrate the perverse incentives generated by such transfers.
Alternatively, and as an attempt to overcome that problem, some authors have recently suggested that
countries could fund directly the development of technologies through a R&D protocol, as the so-called
Technology-Oriented Agreements (TOAs), which include international agreements that aim at advancing
44. There is no consensus about how important are leakage effects. Recent empirical studies have found pollution-haven effects where trade
is between high and low-standard countries and industries are mobile Ederington, Levinson and Minier (2005), pg. 175. Moreover, it is too
early to conclude about eventual leakage effects.
45. See Olmstead and Stavins (2006).
research, development, demonstration, and/or deployment of technologies. Efforts under TOAs may involve
efforts to “push” technologies by subsidizing or otherwise fostering R&D or efforts that “pull” technologies into
the market by providing incentives for or mandating their use. The US has developed several partnerships and
fora for promoting specific technologies: in 2003, the Carbon Sequestration Leadership Forum (CSLF) and the
International Partnership for Hydrogen Economy (IPHE) were created, and in 2005 the Asia Pacific Partnership
on Clean Development and Climate (APP) was launched with Australia, China, India, Japan and South Korea,
46
among others.
47
This is based on previous experience in other areas, and could at the same time establish the conditions
under which international technology standards could be negotiated to guarantee world-wide adoption and
diffusion. Thus, the appropriate targeting of the second market failure surrounding climate change related to the
48
development and adoption of new technologies can be undertaken by promoting long-run cooperation in R&D.
In fact, there is growing recognition not only that mitigation policy by itself cannot provide the desired
technology development, but also that there is a trade-off between short-run mitigation and long-run technology
development. This basically applies the economics principle according to which the number of instruments
should be at least the number of targets.
Interestingly, the European Union has presented in November 2007 the so-called “European Strategic
Energy Technology Plan” (SET-PLAN). Its main goal is to establish the general policies and measures under
which the development and implementation of low carbon technologies can be accelerated, to meet the 2020
targets (to reduce greenhouse gas emissions by 20 per cent and to ensure 20 per cent of renewable energy
sources in EU, more than doubling the current 8.5 per cent), and the 2050 ones (reduce greenhouse gas emissions
by 60-80 per cent). In particular, this plan stresses the importance of cooperation, not only between the different
member countries at both the public and private levels, but it also opens the possibility of cooperation in general
with other developed countries as well as developing countries. This is of utmost importance, before fast-
growing economies, as China and India, under demand pressure for energy, invest in long-lived carbon intensive
capital stocks. Eventually, renewed CDM and JI provisions could serve that purpose. Also, LULUF activities,
and, in particular, deforestation, should be addressed more seriously, as it accounts for twice as much emissions
as transport. This should also be considered in association with the development of the market for biofuels, as
46. See Coninck, Fisher, Newell and Uno (2007).

47. Examples of Research, Development and Demonstration Agreements are the European Organization for Nuclear Research (CERN), or
the ITER Fusion Reactor, among others. In the case of Technology Transfer Agreements, the Montreal Fund for Implementation of the
Montreal Protocol or the Global Environment Facility, are good examples.
48. See Jaffe (2005), and Golombek (2005).
the boom in this market, highly subsidized, may jeopardize any efforts to control deforestation or increase
forested land, as in the case of Brazil.
One interesting feature that may be observed in the context of climate change policy is the different
domestic responses that have been implemented or proposed in the context of the Annex B countries, namely,
EU, Australia, Canada, Japan, New Zealand and the United States, giving support to a larger flexibility in
domestic actions. Countries may prefer to choose taxes, tradable permits, standards, regulation or some
combination of these. The focus on domestic or national policies when dealing with climate change, preserving
the specificities of the different nations involved is the basis of a different perspective concerning the
architecture of a climate change agreement. According to Victor's (2008), the agreements should be set in a small
49 50
club basis, in which countries share common interests, then evolving to larger participation. This alternative
approach is based on a much more intensive negotiating and review process, and draws experience from other
areas of international economic policy, as the WTO or IMF, or even in the context of environmental policy as in
the case of the North Sea and the Baltic Sea. This focus on institutions rather than on emissions is in the line of
Schmalensee's (1998) seminal assessment of climate policy architecture “When time is measured in centuries,
the creation of durable institutions and frameworks seems both logically prior to and more important than choice
of a particular policy program that will almost surely be viewed as too strong or too weak within a decade”.
Therefore, a future climate regime building upon the experience accumulated in the last decade will have to
find satisfactory solutions to the issues raised. Since post-Kyoto is now being discussed and negotiated, this is
the appropriate time to act.
6. Conclusion
The policy challenges of reconciling rapid economic growth and reduced risks of climate change will have
major consequences to the future of both the developed and developing economies.
The role to be played by technology in this context is almost consensual. In order to be able to keep with the
stringent targets that have been proposed technological breakthroughs are required. In particular, a shift to
renewable energy or other cleaner sources imposes a serious pushing by governments. The necessary political
will and financial support is only possible when societies decide that the costs of pollution are larger than the
benefits from having access to low prices for energy. Since climate is a global public good, it requires action at
49. The club review scheme, such as the “L20” concept proposed by former Canadian Prime Minister Paul Martin reflects the same concern.
50. See also Schelling (2007).
the global level as well. This represents an enormous challenge for policymakers and society in general, as the
discussion on the Post-Kyoto agenda makes plain.
References
Auffhammer, M., and R. Carson, “Forecasting the Path of China's CO2 Emissions Using Province Level
Information”, Journal of Environmental Economics and Management, forthcoming.
Bannon, B., M. DeBell, J. Krosnick, R. Kopp, and P. Aldhous, “Americans' Evaluations of Policies to
Reduce Greenhouse Gas Emissions”, mimeo, June 2007.
Barrett, S., “Proposal for a New Climate Change Treaty System”, Economists' Voice, www.bepress.com/ev,
October 2007a.
Barrett, S., “A Multitrack Climate Treaty System”, J. Aldy and R. Stavins (eds.) in Architectures for
Agreement, Cambridge University Press, 2007b.
Borrego, C., H. Martins and M. Lopes, “Portuguese Industry and the EU Trade Emissions Directive:
Development and Analysis of CO2 Emissions”, Environmental Science & Policy 8:75-84, 2005.
Cline, W., The Economics of Global Warming, Washington DC: Institute for International Economics, 1992.
Coninck, H., C. Fischer, R. Newell, and T. Uno, “International Technology-Oriented Agreements to Address
Climate Change”, RFF Discussion Paper 06-50, January 2007.
Cunha-e-Sá, M. and Ana B. Reis, “The Optimal Timing of Adoption of Green Technologies”,
Environmental & Resource Economics 36(1):35-55, 2007.
Ederington, J., A. Levinson, and J. Minier, “Footloose and Pollution Free”, Review of Economics and
Statistics 87(1):92-99, 2005.
Ellerman, A., and B. Buchner, “The European Union Emissions Trading Scheme: Origins, Allocation and
Early Results”, Review of Environmental Economics and Policy 1(1), Winter 2007.
Ellerman, A., and B. Buchner, “Over-Allocation or Abatement? A Preliminary Analysis of the EU ETS
Based on the 2005-06 Emissions Data”, Environmental & Resource Economics, forthcoming.
Golombek, R., and M. Hoel, “Climate Policy under Technology Spillovers”, Environmental & Resource
Economics 31:201-227, 2005.
Herrnstadt, E., Mun Ho, R. Morgenstern, and W. Pizer, “Competitiveness Impacts of Carbon Dioxide
Pricing Policies on Manufacturing”, RFF Backgrounder, September 2007.
Hope, C., “The Marginal Impact of CO2 from PAGE2002: An Integrated Assessment Model Incorporating
IPCC's the Five Reasons for Concern”, Integrated Assessment Journal 6(1):19-56, 2006.
Jaffe, A., R. Newell, and R. Stavins, “A Tale of Two Markets-Technology and Environmental Policy”,
Ecological Economics 54:164-174, 2005.
Kelly, D., and C. Kolstad, “Bayesian Learning, Growth and Pollution”, Journal of Economic, Dynamics and
Control 23: 491-518, 1999.
Kolstad, C., “Learning and Stock Effects in Environmental Regulation: The Case of Greenhouse Gas
Emissions”, Journal of Environmental Economics and Management 23:491-518, 1996.
Kopp, R., “Allowance Allocation”, RFF Backgrounder, September 2007.
Kruger, J., W. Oates and W. Pizer, “Decentralization in the EU Emissions Trading Scheme and Lessons for
Global Policy”, RFF Discussion Paper 07-52, February 2007.
Lecocq, F., and P. Ambrosi, “The Clean Development Mechanism: History, Status, and Prospects”, J. Aldy
and R. Stavins (eds.) in Architectures for Agreement, Cambridge University Press, 2007.
Montgomery, W., and A. Smith, “Price, Quantity, and Technology Strategies for Climate Change Policy”, in
M. E. Schlessinger and others (eds.), Human Induced Climate Change: an Interdisciplinary Assessment,
Cambridge University Press, 2007.
Newell, R., and W. Pizer, “Discounting the Distant Future: How Much Do Uncertain Rates Increase
Valuations”, Journal of Environmental Economics and Management 46:52-71, 2003.
Nordhaus, W., “The Stern Review on the Economics of Climate Change”, NBER Working-Paper 12741,
December 2006.
Nordhaus, W., “To Tax or Not to Tax: Alternative Approaches to Slowing Climate Change”, Review of
Environmental Economics and Policy 1(1), Winter 2007.
Olmstead, S., and R. Stavins, “An International Architecture for the Post-Kyoto Era”, American Economic
Review Papers and Proceedings 96(2):35-38, 2006.
Pacala, S., and R. Socolow, “Stabilization Wedges: Solving the Climate Problem for the Next 50 Years with
Current Technologies”, Science 305:968-972, 2004.
Pizer, W., “The State of Climate Change in 2007”, prepared for the House of Representatives Committee on
Science and Technology, RFF, May 16, 2007.
Pizer, W., and D. Popp, “Endogenizing Technological Change”, RFF Discussion Paper 07-11, September
2007.
Schelling, T., “Epilogue: Architectures for Agreement”, J. Aldy and R. Stavins (eds.) in Architectures for
Agreement, Cambridge University Press, 2007.
Schmalensee, R., “Greenhouse Gas Policy Architectures and Institutions” in W. Nordhaus (ed.), Economics
and Policy Issues in Climate Change, Washington DC, RFF Press, pg. 137-158, 1998.
Stavins, R., “A US Cap-and-Trade System to Address Global Climate Change”, The Brookings Institution,
Discussion Paper 2007-13, 2007.
Sterner, T., and U. Persson, “An Even Sterner Review”, RFF Discussion Paper 07-37, July 2007.
Tol, R.,”The Marginal Damage Costs of Carbon Dioxide Emissions: An Assessment of the Uncertainties”,
Energy Policy 33(6):2064-2074, 2005.
Tol, R. and S. Fankhauser, “On the Representation of Impact of Integrated Assessment Models of Climate
Change”, Environmental Modelling and Assessment 3:63-74, 1998.
Tol, R. and G. Yohe, “A Review of the Stern Review”, World Economics 4, October-December 2006.
Victor, D., “Fragmented Carbon Markets and Reluctant Nations: Implications for the Design of Effective
Architectures”, J. Aldy and R. Stavins (eds.) in Architectures for Agreement, Cambridge University Press, 2007.
Viscusi, W. Kip, and R. Zeckauser, “The Perception and Valuation of The Risks of Climate Change: A
Rational and Behavioral Blen”, Discussion Paper 537, 12/2005.
Yohe, G., “Some Thoughts on the Damage Estimates Presented in the Stern Review-An Editorial”, The
Integrated Assessment Journal 6, Issue 3:65-72, 2006.
Yohe, G., and R. Tol, “The Stern Review: Implications for Climate Change”, Environment 49(2), March
2007.
Weitzman, M., “Prices versus Quantities”, Review of Economic Studies 41:477-491, 1974.
Weitzman, M., “The Stern Review of the Economics of Climate Change”, Journal of Economic Literature
45(3), September 2007a.
Weitzman, M., “Structural Uncertainty and the Value of Statistical Life in the Economics of Climate
Change”, mimeo, October 2007b.

Desenvolvimento Económico Português No Espaço Europeu - 2008

Enviado por

Dados do documento

Título original

Direitos autorais

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Desenvolvimento Económico Português No Espaço Europeu - 2008

Enviado por

Direitos autorais:

Conferência do Banco de Portugal

Desenvolvimento Económico Português

Portuguese Economic Development

Economic Research Department

Selenova - Artes Gráficas, Lda.

Number of copies printed

Legal Deposit no. 236983/06

A Conferência “Desenvolvimento Económico Português no Espaço Europeu: Determinantes e Políticas”

Esta publicação apresenta os trabalhos seleccionados para apresentação na 4ª Conferência do

Podemos agrupar os trabalhos em três grupos temáticos: Regulação e Concorrência, Sustentabilidade da

Nota de Apresentação / Foreword

Nota de Apresentação ..................................................................................................................................... V

Longevity Risk, Retirement Savings and Individual Welfare......................................................................... 173

Simulation of Unemployment Insurance Savings Accounts in Portugal ........................................................ 203

We study hedging of longevity risk in an empirically parameterized life-cycle model of consumption

2. A Model of Longevity Risk

2.1. Survival Probabilities

2.3. Labor Income

During working life age- t labor income, Yt , is exogenously given by:

logYt  ft, Z t  v t  /t for t t K ,

logYt  log5  fK, Z K  v K for t  K ,

2.4. Asset Markets

where 6 L and @ L correspond to the mean and standard deviation, respectively.

2.5. The Optimization Problem

Wt1  ¡) t RLt  1 " ) t R¢W t  Yt " C t

6. These features are discussed in more detail in the calibration section.

3.1. Time and preference parameters

3.2. Survival probabilities

lnm t,x  a t  b t  k x  /t,x

transformation, b t is determined only up to a multiplicative constant, and a t is determined only up to an

3.3. Labor Income

Table 1: Parameters of the model

3.4. Asset Parameters

3.4.1. The Portuguese Case

3.5. A 3-period model with longevity bonds

3.5.1. Model set-up

The agent's objective is

C 1"2 C 1"2 C 1"2

where p 1 and p 2 denote the conditional survival probabilities, as before.

W 2  W1 " C 1 ¡)1  RL2  1 " ) 1  R ¢  Y2

3.5.2. Demand for longevity bonds

Figure 1: Portfolio Rule in the 3-Period Model ("Baseline Case")

3.6. Life-Cycle Model

3.6.1. Baseline model

Figure 4: Portfolio Allocation in the Baseline Case

Figure 5: Portfolio Allocation for Different Values of the Expected Return

3.6.2. Alternative return calibrations

Figure 6: Portfolio Allocation for Different Values of Return Volatility

3.6.3. Welfare gains

Table 2: Welfare gains from investing in the longevity bond (percent).

4. Conclusion and Future Research

Keywords: Mergers, International trade, Merger policy, Size asymmetry.

JEL Codes: F13, H77, L11, L41.

economies of scale in mergers). Marginal costs of firms are given by c l1  c, c l2  c  , where

-  /a " c and use it as a summary measure of asymmetry.

problem of home firm i is given by

max a " Qh " c hi q ih  a " *Q' " c hi q 'hi ,

a " 2q hi " ! q hj " c hi  0.

a " 2*q 'hi " * ! q 'hj  ! q ',fk " c hi  0.

Similarly, at the start, the problem of foreign firm i is given by

maxa " +Qf " c fi q fi  a " *Q' " c fi q 'fi ,

The first-order conditions to this problem are given by

a " 2+q fi " + ! q fj " c fi  0

a " 2*q 'fi " * ! q 'fj  ! q 'hk " c fi  0

Thus, the best reply of foreign firm i in the foreign market is

and the best reply of foreign firm i in the third-country market is

logYt ft, Z t v t /t for t t K ,

logYt log5 fK, Z K v K for t K ,

Wt1 ¡) t RLt 1 " ) t R¢W t Yt " C t

lnm t,x a t b t k x /t,x

W 2 W1 " C 1 ¡)1 RL2 1 " ) 1 R ¢ Y2

economies of scale in mergers). Marginal costs of firms are given by c l1 c, c l2 c , where

- /a " c and use it as a summary measure of asymmetry.

max a " Qh " c hi q ih a " *Q' " c hi q 'hi ,

a " 2q hi " ! q hj " c hi 0.

a " 2q 'hi " ! q 'hj ! q ',fk " c hi 0.

maxa " +Qf " c fi q fi a " *Q' " c fi q 'fi ,

a " 2+q fi " + ! q fj " c fi 0

a " 2q 'fi " ! q 'fj ! q 'hk " c fi 0

q 'h1 a " c " 1 q' q 'f1 q 'f2

q 'h1h2 a " c " 1 q' q'

q 'f1 a " c " 1 q' q 'f2

a " c 2 a " c 2 a " c 2

q 'h1 a " c " 1 q' q'

q 'h2 a " c " " 1 q 'h1 q 'f1f2

a " c 2 a " c 2 a " c 2

a " c 2 a " c 2 a " c 2

a " c 2 a " c 2 a " c 2

a"c 2 a"c"2 2 a"c 2 a"c"2 2

foreign market, that is, + 0, 1 .

Qh a " c /2 and CS hh1h2 a " c 2 /8 .

a " c 2 a " c 2 a " c 2 2a " 2c " 2

a " c 2 a " c 2 a " c 2 2a " 2c " 2

a " c 2 a " c 2 a " c 2 2a " 2c " 2

a " c 2 a " c 2 a " c 2 2a " 2c " 2

or - PT 0. 14122. We have that 7/61 - SP 0. 14122 5/22. It follows from Proposition 4

a " c 2 a " c 2 a " c " 2 2

CP h is concave in and CP h 0 when " a"c

a " c 2 a " c 2 2 a " c " 3 2

a " c 2 8a " c " 20 2

a " c 2 a " c 2 a " c " 3 2

NE F 2,1 when - ¡0, 1/15¢ and * ¡* M , * N ¢. If - ¡0, 1/15¢ and * ¡* M , * N ¢ , then it

a " c 2 a " c 2 a " c 2 a " c 2

63 " 100* " 738 800* - 1647 2000* - 2

a " c 2 a " c 2 a " c 2 a " c 2

63+ " 100* " 738+ 800* - 1647+ 2000* - 2

- ¡7/61, 1/3¢ . Q.E.D.

3a " c 2 2a " 2c " 2 a " c 2 a " c " 2 2

a " c 2 a " c 2 a " c 2

21 350* " 246 800* - 549 800* - 2

P ni : Pr V ni " V nj /ni /nj , for allj p i, j 1, T , I

; F i V ni " V n1 u, T , u, T , V ni " V nI u du,

lnGA|i ! lnhr j , x j ! lnhrj , x j " lnhri , x i k " lnhr i , x i ,

expV i " lnhr i , x i

CS n 1 ln' e V1n1 , T , e V1nJ 0