Escolar Documentos
Profissional Documentos
Cultura Documentos
< 1 given by
n
0
(j
) = a
0
(j
) (1 j
) /
0
(j
) = 0 (10)
and generating enough surplus that even low types can be protably employed, namely
n(j
) 0
L
1
l. (11)
In cases where (underprovision of) the ethical activity a also has spillovers on the rest of society
be they technological (pollution), pecuniary (imperfect competition in the product market) or
scal (cost of government bailouts, taxes or subsidies) total social welfare becomes n(j) c a(j),
where c is the per-unit externality. Clearly, this will only strengthen our main results about the
competitive overincentivization of the other activity, /.
3 Competing for Talent:
Throughout most of the paper (except for Section 7), is known while 0 0
H
, 0
L
is private
information, with mean 0 =
L
0
L
H
0
H
.
13
We rst consider the polar cases of monopsony and
perfect competition, which make most salient the basic forces at play, then study the full spectrum
of imperfect competition. Before proceeding, it is worth noting that if agents types i = H, 1 were
observable, the only impact of market structure would be on the xed wages .
i
, whereas incentives
would always remain at the ecient level, j
i
= j
.
3.1 Monopsony Employer
A monopsonist (or set of colluding rms) selects a menu of contracts (j
i
, .
i
) aimed at type i
1, H. We assume that it wants to attract both types, which, as we will show, is equivalent to
L
exceeding some threshold. The rm thus maximizes expected prot
max
f(y
i
; z
i
)g
i=H;L
_
i=H;L
i
[(j
i
) (1 j
i
)0
i
.
i
[
_
subject to the incentive constraints
n(j
i
) 0
i
j
i
.
i
_ n(j
j
) 0
i
j
j
.
j
for all i, , H, 1 (12)
13
Asymmetric information about ability remains a concern even in dynamic settings where performance generates
ex-post signals about an agents type. First, such signals may be dicult to accurately observe for employers other
than the current one, especially given the multi-task nature of production. Second, many factors can cause 0 to
vary unpredictably over the life-cycle: age (which aects peoples abilities and preferences heterogeneously), health
shocks, private life issues, news interests and priorities, etc. Finally, dierent (imperfectly correlated) sets of abilities
typically become relevant at dierent stages of a career e.g., being a good trader or analyst, devising new securities,
bringing in clients, closing deals, managing a division, running and growing an international company, etc.
10
and the low types participation constraint, n(j
L
) 0
L
j
L
.
L
_ l. This program is familiar from
the contracting literature. First, the combined incentive constraints yield (0
i
0
j
)(j
i
j
j
) _ 0 :
a more productive agent must receive a higher fraction of his measured output. Second, the low
types participation constraint is binding, and the high types rent above l is given by the extra
utility obtained by mimicking the low type: (^0)j
L
. Rewriting prots, the monopsonist solves:
max
f(y
i
; z
i
)g
i=H;L
_
i=H;L
i
[n(j
i
) 10
i
[ l
H
(^0)j
L
,
_
yielding j
m
H
= j
L
^0, implying j
L
< j
. (13)
The principal reduces the power of the low-types incentive scheme, so as to limit the high-types
rent. It is optimal for the rm to hire both types if and only if
L
_
n(j
m
L
) 10
L
l
_
H
j
m
L
^0, (14)
meaning that the prots earned on low types exceed the rents abandoned to high types. By (13),
the dierence of the left- and right-hand sides is increasing in
L
, so the condition is equivalent to
L
_
L
, where
L
is dened by equality in (14).
Proposition 1 (monopsony) Let (14) hold, so that the monopsonist wants to employ both types.
Then j
m
H
= j
and j
m
L
< j
is given by n
0
(j
m
L
) = (
H
,
L
)^0, with corresponding xed payments
.
m
H
=
l j
m
L
^0 n(j
) 0
H
j
and .
m
L
=
l n(j
m
L
) 0
L
j
m
L
. The resulting welfare loss is equal to
1
m
=
L
[n(j
) n(j
m
L
)[ . (15)
It increases with ^0, but need not be monotonic in or 1.
Note that since total social welfare is
H
[n(j
H
) 10
H
[
L
[n(j
L
) 10
L
[ , a mean-preserving
increase in the distribution of 0 always reduces it, by worsening the informational asymmetry. In
contrast, an increase in (or a decrease in 1) has two opposing eects on 1
m
: (i) it makes any
given amount of underincentivization on the 1 task less costly, as the alternative task is now
more valuable; (ii) the ecient bonus rate j
and .
c
L
= (j
) (1 j
)0
L
.
He should then not benet from mimicking the high type, nor vice-versa,
n(j
) 10
L
_ n(j
H
) 0
L
j
H
.
H
= n(j
H
) 10
H
j
H
^0, (18)
n(j
H
) 10
H
_ n(j
) 10
L
j
^0, (19)
implying in particular that j
H
_ j
) (1 j
c
H
) ^0. (20)
By strict concavity of n, this equation has a unique solution j
c
H
to the right of j
, satisfying
j
< j
c
H
< 1. The inequality in (19) is then strict, meaning that only the low types incentive
constraint is binding. Note that, as illustrated in Figure I, this is exactly the reverse of what
occurred under monopsony.
15
Absent cross-subsidies, the low type cannot receive more than the total surplus n(
, :L = :
c
L
-) for - small, leading to a prot - on this type (and an even larger
one on any high type who also chose this contract). Low types must thus be oered utility equal to n(
) + 0L1,
which only their symmetric-information ecient allocation achieves.
12
Figure I: Distortions under monopsony and perfect competition
The intuition for this reversal is simple. A rm with monopsony power seeks to capture the
rents of its workers, who cannot seek a better deal from a competitor. For the less productive
types it achieves this (l
L
=
l) through a low enough xed wage .
L
, but for the more productive
ones its ability to keep l
H
low via .
H
is limited by the fact that they could always pretend
to be 1 types, thereby achieving
l j
L
^0. To extract rents from the most productive agents,
the rm must therefore oer a low rate of variable pay j
L
, so as to make this mimicking strategy
unappealing. Competing employers, by contrast, seek to attract the workers by oering them high
rents, l
H
; they cannot increase xed compensation .
H
too much, however, otherwise 1 types would
masquerade as H, achieving utility l
H
j
H
^0. To deter such behavior, rms must compensate
high-productivity agents mostly with a high bonus rate j
H
, while limiting their xed pay.
Existence and uniqueness. When is this least-cost separating (LCS) allocation indeed an equilib-
rium, or the unique equilibrium of the competitive-oer game? The answer, which is reminiscent
of Rothschild and Stiglitz (1976), hinges on whether or not a rm could protably deviate to a
contract that achieves greater total surplus by using a cross-subsidy from high to low types to
ensure incentive compatibility.
Denition 1 An incentive-compatible allocation (l
i
, j
i
)
i=H;L
is interim ecient if there exists
no other incentive-compatible (l
i
, j
i
)
i=H;L
that:
(i) Pareto dominates it: l
H
_ l
H
, l
L
_ l
L
, with at least one strict inequality.
(ii) Makes the employer(s) break even on average:
i
i
[n(j
i
) 0
i
1 l
i
[ _ 0.
For the LCS allocation to be an equilibrium, it must be interim ecient. Otherwise, there is
another menu of contracts that Pareto dominates it, which one can always slightly modify (while
preserving incentive-compatibility) so that both types of workers and the employer share in the
overall gain; oering such a menu then yields strictly positive prots. The converse result is also
true: under interim eciency there can clearly be no positive-prot deviation that attracts both
types of agents, and by a similar type of surplus-sharing argument one can also exclude those that
attract a single type. These claims are formally proved in the appendix, where we also show that
13
when the LCS allocation is interim ecient, it is in fact the unique equilibrium. Furthermore, we
identify a simple condition for this to be case:
Lemma 1 The least-cost separating allocation is interim ecient if and only if
H
n
0
(j
c
H
)
L
^0 _ 0. (21)
This condition holds whenever
L
exceeds some threshold
L
< 1.
The intuition is as follows. At the LCS allocation, we saw that the binding incentive constraint
is the low types: l
c
L
= l
c
H
j
c
H
^0. Consider now an employer who slightly reduces the power of
the high types incentive scheme, cj
H
= -, while using lump-sum transfers c.
H
= (/(j
H
) 0
H
)-
-
2
to slightly more than compensate them for the reduction in incentive pay, and c.
L
= -^0 -
2
to preserve incentive compatibility. Such a deviation attracts both types (cl
H
= cl
L
= -
2
, since
n
0
(j) = /(j)) and its rst-order impact on prots is
H
__
0
(j
H
) 0
H
_
cj
H
c.
H
L
c.
L
=
_
H
n
0
(j
c
H
)
L
^0
(-)
Under (21) this net eect is strictly negative, hence the deviation unprotable. When (21) fails,
conversely, the increase in surplus generated by the more ecient eort allocation of the high types
is sucient to make the rm and all its employees strictly better o. A higher
L
= 1
H
means
fewer high types to generate such a surplus and more low types to whom rents (cross-subsidies)
must be given to maintain incentive compatibility, thus making (21) more likely to hold.
We can now state this sections main result.
Proposition 2 (perfect competition) Let
L
_
L
. The unique competitive equilibrium involves
two separating contracts, both resulting in zero prot:
1. Low-productivity workers get (j
, .
c
L
), where .
c
L
is given by (17).
2. High-productivity ones get (j
c
H
, .
c
H
), where .
c
H
is given by (16) and j
c
H
j
by
n(j
) n(j
c
H
) = (1 j
c
H
)^0.
3. The eciency loss relative to the social optimum is
1
c
=
H
[n(j
) n(j
c
H
)[ = (1 j
c
H
)
H
^0. (22)
It increases with ^0 and , but need not be monotonic in 1.
These results conrm and formalize the initial intuition that competition for talent will result
in an overincentivization of high-ability types. As shown on Figure I, this is the opposite distortion
from that of the monopsony case, which featured underincentivization of low-ability types. When
14
the degree of competition is allowed to vary continuously (Section 4), we therefore expect that there
will be a critical point at which the nature of the distortion (reecting which incentive constraint
is binding) tips from one case to the other.
Skill-biased technical change. New technologies and organizational forms that raise the relative
productivity of more skilled workers are generally seen as playing a major role in the rise of wage
inequality. Implicit in most models and discussions is the premise that this is the distributional
downside of important gains in productive eciency. Sometimes this is even explicit, as in the
accounts of superstar or CEO compensation by proponents of the ecient-pay hypothesis (e.g.,
Rosen 1981, Gabaix and Landier 2008, Kaplan and Rauh 2010). Our results also call this view
into question. A higher 0
H
exacerbates the competition for talented agents, resulting in a higher
bonus rate j
c
H
that makes their performance-based pay rise more than proportionately to their
marginal product. This market response to technical change is inecient, however, as it worsens
the underprovision of long-term investments and prosocial eorts inside rms, thereby reducing the
social value of the productivity increase. For a mean-preserving spread in the distribution of 0s,
such as information technologies that substitute for low-skill labor and complements high skills,
only the deadweight loss remains, so overall social welfare actually declines.
What happens when the LCS allocation is not interim ecient, that is, when
L
<
L
? We saw
that it is then not an equilibrium, since there exist protable deviations to incentive-compatible
contracts (involving cross-subsidies) that Pareto-dominate it. We also show in the appendix that
no other pure-strategy allocation is immune to deviations, a situation that closely parallels the
standard Rothschild and Stiglitz (1976) problem: the only equilibria are in mixed strategy.
16
Since
such an outcome is not really plausible as a stable labor-market outcome, we assume from here on
L
_ max
L
,
L
=
L
. (23)
3.3 Welfare: Monopsony versus Perfect Competition
Single task ( = 0). As a benchmark, it is useful to recall that competition is always socially
optimal with a single task. The competitive outcome is then the single contract j
c
= j
= 1, .
c
=
0 : agents of either type are residual claimants for their production and therefore choose the ecient
eort allocation.
17
Monopsony, by contrast, leads to a downward distortion in the power of the
incentive scheme. Hence competition is always strictly welfare superior.
16
An alternative approach is to assume that it is workers who make take-it-or-leave oers, instead of a competitive
industry making oers to them. From Maskin and Tirole (1992) we know that for qL ~ qL, the unique equilibrium of
the resulting informed-principal game is the LCS allocation, so the result is the same as here with competitive oers.
By contrast, for qL < ~ qL, the set of equilibrium interim utilities is the set of feasible utilities (incentive compatible and
satisfying budget balance in expectation) that Pareto dominate (l
c
L
, l
c
H
). A second alternative is to use a dierent
equilibrium concept from the competitive screening literature, as in Scheuer and Netzer (2010); again, this has no
bearing on the region where the separating equilibrium exists. A third alternative would be to introduce search, free
entry by principals and contract posting as in Guerrieri et al. (2010). Self-selection then makes type proportions
among searchers in the market endogenous, in such a way that a separating equilibrium always exists.
17
Similar results holds if 0 but = 0 : since o 0 for all , the socially optimal bonus rate is
= 1, even
though it results in an inecient eort allocation. This is clearly also the competitive outcome.
15
Multitasking. From (15) and (20), 1
m
< 1
c
if and only if
L
[n(j
) n(j
m
L
)[ <
H
[n(j
) n(j
c
H
)[ . (24)
Consider rst the role of labor force composition. As seen from (13) and (20), the monopsony
incentive distortion j
j
m
L
is increasing with
H
,
L
(limiting the high types rents becomes more
important), whereas the competitive one, j
c
H
j
L
and
H
2
L
_
L
<
_
1
2
__
^0
_
. (25)
The underlying intuitions are quite general.
18
First, competition entails a larger eciency loss
when the unrewarded task long-run investments, cooperation, avoidance of excessive risks, etc.
is important enough and the two types of eort suciently substitutable. If they are comple-
ments ( < 0), in contrast, competition is always eciency-promoting. Second, the productivity
dierential ^0 scales the severity of the asymmetric-information problem that underlies both the
monopsony and the competitive distortions. A monopsonistic rm optimally trades o total surplus
versus rent-extraction, so (by the envelope theorem) a small ^0 has only a second-order eect on
overall eciency. Under competition the eect is rst-order, because a rm raising its j
H
does not
internalize the deterioration in the workforce quality it inicts on its competitors or, equivalently,
the fact that in order to retain their talent they will also have to distort incentives and the
allocation of eort. This intuition explains why (25) is more likely to hold when ^0 decreases.
19
4 Imperfect Competition
4.1 A Full-Spectrum Hotelling Model
To understand more generally how the intensity of labor market competition aects the equilibrium
structure of wages, workers task allocation, rms prots and social welfare, we now develop a
variant of the Hotelling model in which competitiveness can be varied continuously over the whole
18
In particular, the models solution with quadratic costs (given in Appendix A) also correspond to Taylor approx-
imations of the more general case when 0 is small, provided 1(1
2
) is replaced everywhere by n
00
(
).
19
As shown in Appendix B, for small 0 the lower bound ~ qL above which (21) holds (and the competitive equilibrium
thus exists) is such that 1 ~ qL is of order
p
0. Thus, to rigorously apply the above reasoning involving rst- versus
second-order losses for small 0, one needs to also let qH become small. This further reduces
m
L
while leaving
c
H
i=H;L
, in which workers who opt for this employer self-select.
Let l
k
i
denote the utility provided by rm / to type i :
l
k
i
= n(j
k
i
) 0
i
j
k
i
.
k
i
. (26)
A worker of type i, located at r, will choose rm / = 0 if and only if
l
k
i
tr _ max
_
l tr,
l t(1 r), l
`
i
t(1 r)
_
. (27)
The rst inequality reduces to l
k
i
_
l : a rm must at least match its local outside option. If both
20
Alternatively, each agent could produce
l at home but then have to travel (or adapt his human capital) to
one or the other marketplace to sell his output.
17
attract 1-type workers, l
`
i
_
l as well, so the third inequality makes the second one redundant.
We shall focus the analysis on the (unique) symmetric equilibrium, in which each rm attracts
half of the total labor force. To simplify the exposition, we take it here as given that: (i) each
rm prefers to employ positive measures of both types of workers than to exclude either one;
(ii) conversely, neither rm wants to corner the market on any type of worker, i.e. move the
corresponding cuto value of r all the way to 0 or 1. In Appendix D we show that neither exclusion
nor cornering can be part of a best response by a rm to its competitor playing the strategy
characterized in Proposition 4 below, as long as
L
_
L
, (28)
where
L
[
L
, 1) is another cuto, independent of t. Assuming (28) from here on, we can focus
on utilities
_
l
k
i
, l
`
i
_
resulting in interior cutos, so that rm /s share of workers of type i is
r
k
i
_
l
k
i
, l
`
i
_
=
1
2
l
k
i
l
`
i
2t
. (29)
The rm then chooses (l
L
, l
H
, j
L
, j
H
) to solve the program:
max
_
H
(l
H
l
`
H
t)[n(j
H
) 0
H
1 l
H
[
L
(l
L
l
`
L
t)[n(j
L
) 0
L
1 l
L
[
_
(30)
subject to the constraints (with Lagrange multipliers in parentheses):
l
H
_ l
L
j
L
^0 (j
H
) (31)
l
L
_ l
H
j
H
^0 (j
L
) (32)
l
L
_
l (i) (33)
To shorten the notation, let
i
= n(j
i
) 0
i
1l
i
denote the rms prot margin on type i = H, 1.
The rst-order conditions, together with the symmetric-equilibrium condition l
i
= l
`
i
, are:
H
(
H
t) j
H
j
L
= 0 (34)
L
(
L
t) j
L
j
H
i = 0 (35)
t
H
n
0
(j
H
) j
L
^0 = 0 (36)
t
L
n
0
(j
L
) j
H
^0 = 0. (37)
Note that j
H
and j
L
cannot both be strictly positive: otherwise (31) and (32) would bind, hence
j
H
= j
L
, rendering (36)-(37) mutually incompatible. This suggests that only one or the other
incentive constraint will typically bind at a given point.
Constructing the equilibrium: key intuitions. Solving the above problem over all values of t
is quite complicated, so we shall focus here on the underlying intuitions. The solution to (31)-
(37) is formally derived in Appendix B; because the objective function (30) is not concave on the
relevant space for (l
L
, l
H
, j
L
, j
H
), Appendix D (online) then provides a constructive proof that
18
this allocation is indeed the global optimum. These and other technical complexities (exclusion,
cornering) are the reasons why we conne our analysis to the symmetric separating equilibrium.
(a) For large t, the equilibrium should resemble the monopsonistic one: the main concern is
limiting high types rent, so rms distort j
L
< j
= j
H
to make imitating low types unattractive.
Conversely, for small t, the equilibrium should resemble perfect competition: the main concern is
attracting the H types, leading employers to oer them high-powered incentives, j
H
j
= j
L
.
(b) As t declines over the whole real line, the high types responsiveness to higher oered utility
l
H
rises, so rms are forced to leave them more rent. Since that rent is either j
L
^0 or j
H
^0
(depending on which of the above two concerns dominates, i.e. on which types incentive constraint
is binding), j
L
and j
H
must both be nonincreasing in t.
(c) Firms 0 and 1 are always actively competing for the high types. If t is low enough, they also
compete for 1 types, oering them a surplus above their outside option: l
L
l. At the threshold
t
1
below which l
L
starts exceeding
l, j
H
has a convex kink: since the purpose of keeping j
H
above
j
is to maintain a gap l
H
l
L
= j
H
^0 just sucient to dissuade low types from imitating high
ones, as l
L
begins to rise above
l, the rate of increase in j
H
can be smaller.
These intuitions translate into a characterization of the equilibrium in terms of three regions,
illustrated in Figure III and formally stated in Proposition 4.
21
Figure III: equilibrium incentives under imperfect competition
Proposition 4 (imperfect competition) Let
L
_
L
. There exist unique thresholds t
1
0 and
t
2
t
1
such that, in the unique symmetric market equilibrium:
1. Region I (strong competition): for all t < t
1;
bonuses are j
L
= j
< j
I
H
(t), strictly decreasing
in t, starting from j
I
H
(0) = j
c
H
. The low types participation constraint is not binding, l
L
l, while his incentive constraint is: l
H
l
L
= j
I
H
(t)^0.
21
With quadratic costs one can show (see Appendix A) that each of the curves is convex, as drawn in the gure.
19
2. Region II (medium competition): for all t [t
1
, t
2
), bonuses are j
L
= j
< j
II
H
(t), with
j
II
H
(t) < j
I
H
(t) except at t
1
and strictly decreasing in t. The low types participation constraint
is binding, l
L
=
l, and so is his incentive constraint: l
H
l
L
= j
II
H
(t)^0.
3. Region III (weak competition): for all t _ t
2
, bonuses are j
L
= j
L
(t) < j
= j
H
, with j
L
(t)
strictly decreasing in t and lim
t!+1
j
L
(t) = j
m
L
. The low types participation and the high types
incentive constraints are binding : l
L
=
l, l
H
l
L
= j
L
(t)^0.
Welfare. For each value of t, either j
L
or j
H
is equal to the (common) rst-best value j
, while
the other bonus rate, which creates the distortion, is strictly decreasing in t. Recalling from (7)-(8)
that \ =
H
n(j
H
)
L
n(j
L
) 1
)0
H
(1j
)0
L
j,
where j
L
= j
= j
H:
Figure IV: competition and social welfare
Note that we do not subtract from \ the total mobility cost t,4 incurred by agents (equivalently,
we add it to their baseline utility), as it is paid even when neither rm attracts workers, who instead
all fetch the nearest outside option. This is also consistent with using t as a measure of pure market
competitiveness, without introducing an additional wealth eect. In particular, it is required to
yield back the monopsony levels of utility as t .
22
International dierences. Another interesting interpretation of t is as an index of labor and/or
product market regulation. Besides direct restrictions on occupational mobility and entry (certi-
cation, licensing or nationality requirements, etc.) this also includes limits on pay-for-performance
22
One can think of t as a tax on mobility rebated to agents, or as the prots of a monopolistic transportation or
human-capital-adaptation sector with zero marginal cost, engaged in limit pricing against a competitive fringe with
marginal cost t. Alternatively, in contexts where variations in t also involve a net resource cost, one could subtract it
from social welfare (as in Villas-Boas and Schmidt-Mohr 1999). In Appendix A we show that increases in t can raise
aggregate welfare even under this more demanding denition.
20
(or ring for misperformance) imposed by unions, civil-service rules or social norms. A number of
European countries can thus be thought of as having many sectors situated in Region III, where
less productive workers are underincentivized and more competition and deregulation would be
benecial. Conversely, important sectors in the US and UK, particularly nance, t the description
of Regions I-II; high skill workers are overincentivized and greater competition for talent, spurred
for instance by deregulation, further aggravates the bonus culture.
23
4.2 Inequality
We next examine how the gains and losses in total welfare (under either denition) are distributed
among the dierent actors in the market.
Proposition 6 (individual welfare and rm prots) As the labor market becomes more com-
petitive (t declines), both l
H
and l
L
increase (weakly for the latter), but inequality in workers
utilities, l
H
l
L
always strictly increases; rms total prots strictly decline.
In Regions III and II, l
L
=
l. In Region I, l
L
is decreasing in t, as we show in the appendix.
Since (l
H
l
L
) ,^0 is equal to j
H
(t) over Regions I and II and to j
L
(t) over Region III, it follows
directly from Proposition 4 that 0l
H
,0t _ 0 (l
H
l
L
) ,0t < 0. As to prots, they must clearly
fall as t declines over Regions II and I, since overall surplus is shrinking but all workers are gaining.
In Region III, Fernandes et as j
L
(t) rises rms reap some of the eciency gains from low-type
agents more ecient eort allocation, but the rents they must leave to high types increase even
faster (as shown in the appendix), so total prots decline here as well.
Income inequality. Consider now the eects of a more competitive labor market on earnings,
which is what is measured in practice. For most sectors in a market economy the empirically
relevant range is that of medium to high mobility, namely Regions I-II in Figure III. Indeed, this is
where rms are more concerned with retaining and bidding away from each other the workers of
high ability who can easily switch (r close to 1,2) than with exploiting their captive local labor
force (r close to 0 or 1). Region III, in contrast is particularly relevant to public-sector employment
and, more generally, to countries and industries with heavily regulated labor markets. We compare
how the two types of workers i = H, 1 fare in terms of total earnings 1
i
= [/(j
i
) 0
i
[ j
i
.
i
, as
well as the separate contributions of performance-based and xed pay.
Proposition 7 (income inequality) Let
L
_
L
. As the labor market becomes more competitive
(t declines), both 1
H
and 1
L
increase (weakly for the latter). Furthermore,
1. Over Regions I and II (medium and high competition), inequality in total pay 1
H
1
L
rises,
as does its performance-based component. Inequality in xed wages declines, so changes in
performance pay account for more than 100% of the rise in total inequality.
23
Fernandes et al. (2012) nd large variations across countries in the use of incentive pay for CEOs, which they
attribute to dierences in the weight of institutional shareholders and independent boards. Our model shows that
these could also reect dierences in labor market institutions aecting the competition for executive talent. Corollary
1 below provides a simple test of which side of the optimum a given sector is on.
21
2. Over Region III (low competition), inequality in performance pay declines, while inequality in
xed wages rises. As a result, inequality in total pay need not be monotonic. With quadratic
costs, a sucient condition for it to rise as t declines is 1 _ (1
2
)^0.
These results are broadly consistent with the ndings of Lemieux et al. (2009) about the driving
role of performance pay in rising earnings inequality, as well as their hypothesis that the increased
recourse to performance pay also serves a screening purpose. They are also in line with Frydmans
(2007) evidence linking increased mobility (skills portability) of corporate executives to the rise in
both the level and the variance in their compensation.
24
The above properties also imply that the
fraction of the income dierential 1
H
1
L
that is due to incentive pay, 1 (.
H
.
L
),(1
H
1
L
),
is l-shaped in t and minimized at t
2
, where it equals 1. Whereas this value reects the specic
assumption (additive separability of talent and eort) making the rst-best incentive rate j
type-
independent, the l shape is a more robust result of competitions opposing eects on the two types
incentive constraints. For this reason we state the next result in terms of changes rather than levels.
Corollary 1 (testing for eciency) An increase in market competition reduces (raises) aggre-
gate eciency when it is accompanied by an increase (decrease) in the share of earnings inequality
accounted for by performance-based pay.
This result provides a simple test, based on observables, to assess whether competition is in
the range where it is benecial, or detrimental. Subject to the caveats inherent in interpreting
empirical data through the lens of a simple, two-type model, the evidence discussed earlier points
to the latter case, especially as ones gets into the upper deciles and then centiles of the earnings
distribution (top 80%, executive pay, nancial sector, etc.).
4.3 Human Capital Investment and Worker Sorting
Endogenous skill distribution. Let the cost to a marginal worker of acquiring high skills, given
that a fraction
H
already have, be G
0
(
H
), where the aggregate aggregate cost G(
H
) is such that
G(0) = G
0
(0) = 0, G
0
0 and G
0
(
H
) j
c
H
^0, where
H
= 1
L
and
L
is given by (28).
Given a utility dierential l
H
l
L
0, the supply of H types is
H
= (G
0
)
1
(l
H
l
L
)
(0,
H
). On the demand side, employer competition leads to a skill premium of l
H
l
L
= j
H
^0
in Regions I-II and l
H
l
L
= j
L
^0 in Region III, which we show in appendix to be everywhere
decreasing in
H
, as one would expect. Recalling from Proposition 7 that it is also strictly decreasing
in t leads to the following results.
Proposition 8 When workers can invest in human capital:
1. The fraction of high-skill workers
H
(t) is the unique solution to G(
H
) = (j
H
j
L
)(
H
, t)^0,
lies in (0,
H
) and is strictly increasing with competition (declines with t over R
+
).
24
See also Bloom and Van Reenen (2010), Bell and Van Reenen (2013) and Frydman and Saks (2005).
22
2. Social welfare,
\(t) = \(t;
H
(t)) G(
H
(t)), is again maximized at an interior level of
competition,
t
2
(0, t
2
). In particular, perfect competition is locally and globally suboptimal.
The intuition for the rst result is familiar: competition for talent protects workers from the
expropriation of their human-capital investment, and thus spurs the acquisition of skills. Consider
next the welfare eects of a small increase in
H
. At xed bonuses (j
H
, j
L
), a marginal worker
acquiring high skills increases productive surplus by n(j
H
) n(j
L
) 1^0 but only appropriates
l
H
l
L
, compensating for his investment cost G
0
(
H
). His employer thus reaps extra prot
H
L
_ 0, with strict inequality except at t = 0, where
H
=
L
= 0. A marginal high-skill worker
also contributes to reducing the skill premium (lowering j
H
or raising j
L
), and this pecuniary
externality alleviates the multi-task distortion at each rm, except at t = 0, where we show that
0j
H
,0
H
= 0. Each worker who invests thus generates two positive externalities, and more of them
do as t falls, so the socially optimal level of competition is higher than with xed types:
t
2
< t
2
.
As one approaches perfect competition, however, both externalities vanish, leaving only the bonus
culture distortion. Therefore,
\
0
(0) 0 and
t
2
0, demonstrating the robustness of our main
conclusions to an endogenous distribution of skills.
25
Firm heterogeneity and worker sorting. While we have focussed here on a symmetric equilibrium,
our main results are also robust to workers sorting dierentially across rms. This is most easily
seen in the case of perfect competition (t = 0).
26
As noted earlier, the equilibrium allocation can
then indierently be achieved as the symmetric outcome of Bertrand competition among two (or
more) rms, or as an equilibrium in which a fraction
H
of rms employ only H types and the
remaining fraction
L
only 1 types. Proposition 3, which provides a simple condition determining
when monopsony or competition is more ecient, applies to both cases.
27
5 Regulating Compensation
Bonus caps. We focus here on the case of perfect competition, for which the issue is most
relevant. When the regulator is able to dierentiate between the performance-based and xed parts
of compensation, and absent other margins that could be distorted, policy can be very eective. As
shown below, if bonuses are capped at j
, (j
) (1 j
and a wage :L that absorbs all remaining prot; all 1 agents, meanwhile, work in rms with
1 1
1
1
(qL), receiving a distorted bonus H(1)
) absorbs all prots of the marginal rm. The same reasoning as that preceding Proposition
3, again shows that for qH small enough, the eciency loss under monopsony is smaller than under competition.
23
In practice, things may not be so simple. For instance, rms might switch to alternative forms
of rewards that, at the margin, appeal dierentially to dierent types but are even less ecient
screening devices than performance bonuses. Plausible examples include latitude to serve on other
companies boards, to engage in own practice (doctors) or consulting (academics), and lower lock-in
to company (low clawbacks, easier terms for quitting). Let $1 paid in the alternative currency
yield utility $`
i
to a type-i employee, with `
L
< `
H
< 1. We assume that, absent regulation,
employers prefer to use incentive pay rather than inecient transfers to screen workers:
n
0
(j
c
H
)
^0
<
1 `
H
^`
, (38)
where ^` = `
H
`
L
.
28
Suppose now that regulation constrains bonuses, j _ j, and consider
a rm that leaves incentives unchanged but substitutes in high types contract $1 of alternative
transfer for a $`
H
reduction in .
H
; it can then also reduce .
L
by $^` while preserving incentive
compatibility. The strategy is protable if
H
(1 `
H
)
L
^` 0, or
1 `
H
^`
<
L
H
. (39)
Proposition 9 (bonus cap) Assume
L
_
L
and (38). Under a bonus cap at any j [0, j
c
H
[,
the unique competitive equilibrium has j
L
= minj
, j = j
and j
H
= j. Furthermore:
1. If (39) does not hold, alternative transfers are not used:
H
=
L
= 0, l
H
l
L
= j^0 and
.
L
.
H
= n( j)n( j
)0
L
( j j
). As j is reduced from j
c
H
to j
. As j is reduced still
further the equilibrium becomes a pooling one, and welfare strictly decreases.
2. If (39) holds, the equilibrium is always a separating one. Low types receive their constrained-
symmetric-information contract (j
L
= j
, .
L
= n( j
)(1 j
)0
L
) while high types get bonus
j, a non-monetary transfer
H
= [(1 j)^0n( j)n( j
)[,(1`
L
) and a monetary transfer
.
H
= n( j)(1 j) 0
H
j/( j)
H
. Social welfare is strictly decreasing (in the Pareto sense)
as j is reduced, and thus maximized when no binding regulation is imposed ( j = j
c
H
).
In the rst case, using the alternative currency to screen is too onerous: it entails a substantial
deadweight loss 1 `
H
, or would have to be given in large amounts to achieve separation (^`
small), or to too many high types (
H
,
L
large). A bonus cap j < j
c
H
will then successfully limit
rms ability to poach each others high-skill workers through escalating incentive pay, without
triggering other distortions. Such a policy achieves Pareto improvements all the way down to
28
The left-hand side is the surplus gained on each high type when decreasing H by $10. This raises the low
types utility from mimicking by $1, so in order to preserve incentive-compatibility the high types contract must
include $1(`) in the inecient currency, while :H is adjusted to keep lH unchanged. Monetary wages (xed plus
variable) thus decrease by $`H`, resulting in a net cost equal to the right-hand side of (38), exceeding the benet.
24
j = j
, with the benets accruing to both types of workers as higher xed pay, which is the margin
where competition now takes place.
In the second case, rms increasingly substitute toward inecient transfers as the bonus cap is
reduced. By (38), even at j
c
H
where the marginal bonus distortion is maximal, it is still smaller
than that from using the alternative currency. A fortiori, the further down j forces j
H
, the less
is gained in productive eciency, while the marginal distortion associated with the alternative
screening device remains constant: a Pareto-worsening welfare loss.
29
Earnings caps. If rms are able to relabel xed and variable compensation, the only cap the
regulator can impose is on total earnings 1. As we show in Proposition 19 (see online Appendix D),
this leads to a set of results parallel to those obtained above for bonus regulation. When rms have
relatively easy access to alternative rewards allowing them to screen workers (meaning that (39)
holds; otherwise, alternative transfers are again not used, and regulation is ecient), an earnings cap
1 leads to a constrained-LCS allocation in which: (i) low types receive their symmetric-information
contract (j
, .
L
); (ii) high types get a package with bonus j
< j
r
H
< j
c
H
, a xed wage .
H
set so
that total pay adds up to
1 , and a nonmonetary transfer
r
H
0. Any tightening of the earnings
cap (reduction in 1 ) then lowers j
r
H
but increases
r
H
, resulting in a Pareto deterioration.
Taxation. Although a conscatory tax of 100/ above a ceiling
1 is unambiguously welfare
reducing (under (38)-(39)), some positive amount of taxation is always optimal to improve on the
laissez-faire bonus culture. While characterizing the optimal tax in this setting is complicated
and left for future work, we can show:
Proposition 10 A small tax t on total earnings always improves welfare: d\,dt[
=0
0.
The intuition is as follows. To start with, condition (38) ensures that, for t suciently small,
the rm does not nd it protable to resort to inecient transfers, hence still uses performance pay
to screen workers. Taxing total earnings then has two eects. First, under symmetric information,
it distorts (net) incentives downward from the private and social optimum, j
. Second, it shrinks
the compensation dierential received by the two types under any given contract. This reduces low
types incentive to mimic high ones, thus dampening rms need to screen through high-powered
(net) incentives and thereby alleviating the misallocation of eort. For small t the rst eect is of
second-order (a standard Harberger triangle), whereas the second one is of rst order, due again
to the externality between rms discussed earlier.
6 Multidimensional Incentives and Noisy Task Measurement
Performance in activity was so far taken to be non-measurable or non-contractible. Consequently,
eort a was driven solely by intrinsic motivation, or by xed outside incentives such as potential
legal liability or reputational concerns. In the other version of the multitask problem studied by
29
Allowing the inecient transfers to have increasing marginal cost (in the form of (1 `H)` rising with
H
)
would combine the two cases.
25
Holmstrm and Milgrom (1991), every dimension of performance can be measured but with noise,
and this uncertainty limits the extent to which risk-averse agents can be incentivized. We now
extend our theory to this case, where there need not be any intrinsic motivation. This variant of
the model is particularly applicable to the issue of short- versus long-term performance and the
possible recourse to deferred compensation, clawbacks and other forms of long-term pay.
Technology and preferences. Outputs in tasks and 1 are now 0
A
a -
A
and 0
B
/ -
B
,
where 0
A
, 0
B
are the employees talents in each one, a and / his eorts as before, and -
A
, -
B
independent random shocks with -
A
~ A(0, o
2
A
) and -
B
~ A(0, o
2
B
). A compensation package is
a triple (j
A
, j
B
, .) where j
A
and j
B
are the bonuses on each task and . the xed wage. As in
Holmstrm and Milgrom (1991), agents have mean-variance preferences:
l(a, /; 0
A
, 0
B
, j, .) = (0
A
a)j
A
(0
B
/)j
B
. C(a, /)
r
2
_
(j
A
)
2
o
2
A
(j
B
)
2
o
2
B
, (40)
where r denotes the index of risk aversion and the cost function C has the same properties as
before. Given an incentive vector j = (j
A
, j
B
), the agent chooses eorts a(j) and /(j) that solve
C
a
(a(j), /(j)) = j
A
and C
b
(a(j), /(j)) = j
B
. It is easily veried that a(j) is increasing in j
A
and decreasing in j
B
, while /(j) has the opposite properties. The rms prot function remains
unchanged, so total surplus is n(j) 0
A
10
B
, where the allocative component is now
n(j) = a(j) 1/(j) C(a(j), /(j))
r
2
_
(j
A
)
2
o
2
A
(j
B
)
2
o
2
B
. (41)
Assuming strict concavity and an interior solution, the vector of rst-best bonuses j
= (j
A
, j
B
)
solves the rst-order conditions:
0n
0j
A
_
j
A
, j
B
_
=
0n
0j
B
_
j
A
, j
B
_
= 0, (42)
which is shown in the appendix to imply that j
A
< and j
B
< 1.
There are again two types of workers, H and 1, in proportions
H
and
L
, who each select their
preferred contract from the menus
_
(j
A
i
, j
B
i
, .
i
)
_
i=H;L
oered by rms. Denoting ^j
= j
H
j
L
and ^0
= 0
H
0
L
for each task t = , 1, incentive compatibility requires that
=A;B
(^j
)(^0
) _ 0. (43)
To simplify the analysis, we assume H types to be more productive in both tasks: ^0
A
_ 0 and
^0
B
0 (otherwise, which type is better depends on the slopes of the incentive scheme).
Monopsony. Denoting 1
i
= 0
A
i
10
B
i
for i = H, 1, a monopsonistic employer solves
max
f(U
i
;y
i
)g
i=H;L
H
[n(j
H
) 1
H
l
H
[
L
[n(j
L
) 1
L
l
L
[, subject to
l
L
_ l,
l
H
_ l
L
j
A
L
^0
A
j
B
L
^0
B
.
26
This yields j
m
H
= j
, while j
m
L
is given by
1
^0
A
0n
0j
A
(j
m
L
) =
1
^0
B
0n
0j
B
(j
m
L
) =
L
. (44)
As before, the incentives of low types (only) are distorted downward, now in both activities. Note
also how the eciency losses, normalized by their osetting rent reductions, are equalized across
the two tasks. As before, one can show that it is indeed optimal to employ both types as long as
L
is above some cuto
L
< 1, which we shall assume.
Perfect competition. We look again for a least-cost separating equilibrium. Denoting l
SI
L
=
n(j
) 1
L
type 1s symmetric-information utility, such an allocation must solve:
max
f(U
i
;y
i
)g
i=H;L
l
H
, subject to
l
H
= n(j
H
) 1
H
,
l
SI
L
_ l
H
j
A
H
^0
A
j
B
H
^0
B
.
Let i
c
denote the shadow cost of the second constraint. The rst-order conditions are then
1
^0
A
0n(j
H
)
0j
A
H
=
1
^0
B
0n(j
H
)
0j
B
H
= i
c
, (45)
while the binding incentive constraint takes the form
n(j
) n(j
H
) =
_
j
A
H
_
^0
A
_
1 j
B
H
_
^0
B
. (46)
Hence, a system of three equations determining (j
A;c
H
, j
B;c
H
, i
c
), independently of the prior probabil-
ities, as usual for the LCS allocation. Clearly, high-ability agents are again overincentivized, now
in both tasks. Note also that even though competitive and monopsonistic rms use screening for
very dierent purposes, resulting in opposite types of distortions, both equalize those distortions
(properly normalized by unit rents) across the two tasks.
The LCS allocation is, once again, the (unique) equilibrium if and only if it is interim ecient.
In the appendix we generalize Lemma 1 to show:
Lemma 2 The LCS allocation j
c
H
is interim ecient if and only if
1
^0
A
0n(j
c
H
)
0j
A
H
=
1
^0
B
0n(j
c
H
)
0j
B
H
_
H
(47)
or, equivalently, i
c
_
L
,
H
.
This condition generalizes (21) and has the same interpretation, which can now be given in
terms of either task. Intuitively, the larger the distortion in the partial derivatives, the higher
the welfare loss relative to rst best; condition (47) requires that it not be so large as to render
protable a deviation to a more ecient contract sustained by cross-subsidies.
27
Welfare. To demonstrate the robustness of our main result competition for talent can be
excessive, resulting in signicant eciency losses to both tasks being incentivized and performed
only for money, it suces to compare again monopsony and perfect competition. As before, a
simple sucient condition for 1
m
=
L
[n(j
) n(j
m
L
)[ <
H
[n(j
) n(j
c
H
)[ = 1
c
is, that
H
,
be small enough. Indeed n(j
) n(j
c
H
) is independent of this ratio, whereas under monopsony
the distortion becomes small as the high types from whom it seeks to extract rents become more
scarce: as
H
,
L
tends to zero, (44) shows that j
m
L
tends to j
and j
j
m
L
is of order (
H
,
L
) .
Therefore n(j
) n(j
m
L
) is of order (
H
,
L
)
2
, implying that 1
m
<< 1
c
.
Proposition 11 There exist
H
such that for all
H
_
H
, welfare is higher under monopsony
than under competition.
Quadratic cost. From here on, we focus on the specication
C(a, /) = a
2
,2 /
2
,2 a/, with 0, (48)
as it allows for many further results, particularly on comparative statics.
30
Eort levels are thus
a(j) =
_
j
A
j
B
_
,
_
1
2
_
and /(j) =
_
j
B
j
A
_
,
_
1
2
_
, non-negative as long as j
A
,j
B
[, 1,[ . The key properties of rst-best incentives parallel those in Holmstrm and Milgrom (1991):
Proposition 12 The rst-best incentive j
A
is decreasing in 1 and o
2
A
, and increasing in and
o
2
B
, whereas j
B
has the opposite properties. Both are decreasing in risk aversion, r.
Turning now to monopsony and competition, the system (A.17)-(A.18) can also be rewritten in
terms of the price distortions j
^0
B
^0
A
_
1 r (1
2
) o
2
B
^0
A
^0
B
= j
_
o
2
A
, o
2
B
; ^0
A
,^0
B
; r
_
. (49)
It is greater:
(i) The greater the noise o
2
A
in task and the lower the noise o
2
B
in task 1;
(ii) The greater the comparative advantage ^0
B
,^0
A
of H types in task 1, relative to task ;
(iii) The greater workers risk aversion if o
2
A
,o
2
B
^0
A
,^0
B
(and the smaller if not).
These results are intuitive: more noisy measurement makes a task a less ecient screening de-
vice whether for rent-extraction or employee-selection purposes while a higher ability dierential
of low and high types makes it a more ecient one. As to the mirror image property of relative
price wedges under monopsony and competition, it reects the fact that both types of rms equalize
the (normalized) marginal distortions across tasks. We next consider workers eort allocations.
30
It also makes Proposition (11) an if and only if statement; details are available upon request.
28
Proposition 14 (eort distortions) (1) Competition distorts high-skill agents eort ratio away
from task , and monopsony away from task 1, a(j
c
H
),/(j
c
H
) < a(j
),/(j
) < a(j
m
L
),/(j
m
L
), if and
only if
1
1
^0
A
^0
B
. (50)
(2) Competition reduces the absolute level of eort on task , a(j
c
H
) < a(j
), while increasing
that on task 1 (and monopsony has the opposite eects), if and only if
ro
2
A
1 ro
2
B
^0
A
^0
B
. (51)
The message of Proposition 14 accords with that of Sections 3.1-3.2, but it also yields several
new insights about how the misallocation of eorts is shaped by the measurement error in each
the two tasks, their substitutability in eort, high-skill agents comparative advantage in one or the
other, and the degree of risk aversion. The second result is particularly noteworthy: even though
both tasks are more strongly incentivized under competition, eort in task still declines, because
task 1 becomes disproportionately rewarded.
31
Technology and monitoring. In the last few decades, a number of technical and deregulatory
changes may have decreased ^0
A
,^0
B
and increased o
A
,o
B
, making (51) more likely to hold and
magnifying the relative wedge in (49). Financial innovation and leverage, expensive high-tech med-
ical procedures, online tools allowing instant counts of researchers publications and citations, all
arguably raise high-ability agents productivity advantage and its measurability more in individual,
revenue-generating tasks than in diuse ones such as cooperation, public goods provision or avoid-
ing collective risks. As to the increased monitoring of managers by more independent corporate
boards (Hermalin 2005) and of rms performance by the nancial media, its impact hinges on
whether they focus more on earnings, costs and market share or on long-term safety, environmental
and legal liabilities.
7 Competition for the Motivated
We now return to the benchmark specication of Section 3 (task is non-contractible, task 1
is perfectly measurable, agents are risk-neutral) and study the polar case where all workers have
the same productivity 0 (normalized to 0 without loss of generality) in task 1 but dier in their
ethical motivations for task : a fraction
L
has =
L
and the remaining
H
have =
H
.
When an agent of type
i
is employed under a compensation scheme (j, .), his net utility is
n
i
(j) . and his employers prot
i
(j) ., where
n
i
(j) = max
(a;b)
i
a j/ C(a, /), (52)
i
(j) = a
i
(j) (1 j)/
i
(j). (53)
31
Note also that when (A.20) holds, so that o(
) 0, /(
i
= aig maxn
i
(j) = n
i
(j)
i
(j). (54)
Note next that, when confronted with an incentive-compatible menu of options, the more pro-
social type (
H
) chooses a less powerful incentive scheme: j
H
_ j
L
and .
H
_ .
L
.
33
This, in turn,
implies that if a
L
and a
H
are the two types equilibrium eorts on task , then a
L
_ a
H
. The more
pro-socially inclined employee exerts more eort on both because he is more motivated for it and
because he selects a lower-powered incentive scheme.
Monopsony. The monopsonist oers an incentive-compatible menu (j
L
, .
L
) and (j
H
, .
H
), or
equivalently (j
L
, l
L
) and (j
H
, l
H
) so as to solve:
max
f(y
i
; z
i
)g
i=H;L
i=H;L
i
[n
i
(j
i
) l
i
[ , subject to
l
L
_ l
l
H
_ l
L
n
H
(j
L
) n
L
(j
L
)
l
L
_ l
H
n
L
(j
H
) n
H
(j
H
).
The rst two constraints must clearly be binding, while the third imposes j
H
_ j
L
, as seen above.
Substituting in, the solution satises j
m
H
= j
H
and, when interior,
n
0
L
(j
L
) =
H
L
_
n
0
H
(j
L
) n
0
L
(j
L
)
=
H
L
[/
H
(j) /
L
(j)[ . (55)
More generally, the left-hand side of (55) must be no greater than the right-hand side. Since the
latter is strictly negative, one must have j
m
L
j
L
in any case: the monopsonist oers a higher-
powered incentive scheme than under symmetric information so as to limit the rent of the more
prosocial types, who clearly benets less from an increase in j.
Perfect competition. Because employees intrinsic-motivation benets a are private, rms have
no reason to compete to select more prosocial types. As a result, the kind of escalating incentive
distortion seen earlier does not arise, and the competitive equilibrium is the symmetric-information
outcome. Employers oer the menu (j
i
, .
i
)
i=H;L
, where for each type j
i
is the ecient incentive
rate dened by (54) and .
i
=
i
(j
i
), leaving the rm with zero prot. Type i = H, 1 then selects
max
j2fH;Lg
_
n
i
(j
j
)
i
(j
j
) = n
i
(j
j
)
_
. (56)
32
In the quadratic-cost benchmark, however,
L
=
H
= 1 . In general, the variation of with involves the
third derivatives of C and is thus ambiguous.
33
Adding up the two incentive constraints, lH lL +&H(L) &L(L) and lL lH &H(H) +&L(H), yields
0
R
y
L
y
H
[&
0
H
() &
0
L
()] d =
R
y
L
y
H
[/H() /L()] d. Since /H() < /L() for all , the result follows.
30
By (54), choosing , = i is optimal, so the symmetric-information outcome is incentive compatible.
Proposition 15 When agents are similar in measurable talent 0 but dier in their ethical values
, monopsony leads to an overincentivization of low-motivation types, j
m
L
j
L
(with j
m
H
= j
H
),
whereas competition leads to the rst-best outcome, j
c
L
= j
L
, j
c
H
= j
H
.
Would conclusions dier under an alternative specication of the impact of prosocial hetero-
geneity? Suppose that instead of enjoying task more, a more prosocial agent supplies more
unmeasured positive externalities on the rm (or on her coworkers, so that their productivity is
higher, or their wages can be reduced due to a better work environment). In other words, agents
i = H, 1 share the same preferences but have dierent productivities in the activity:
n(j) = max
(a;b)
a j/ C(a, /), (57)
i
(j) = (a(j) i
i
) (1 j)/(j). (58)
Under this formulation there is no way to screen an agents type, so the outcome under both
monopsony and competition is full pooling at the ecient incentive power:
j
i
= j
= 1 , (A.3)
n(j
) n(j) =
_
y
y
n
0
(.) d. =
_
y
y
_
j
.
1
2
_
d. =
(j j
)
2
2 (1
2
)
. (A.4)
1. Monopsony. Substituting the last two expressions into Proposition 1 yields
j
m
L
= j
(1
2
)
L
^0, (A.5)
1
m
=
1
2
2
H
L
(1
2
)(^0)
2
. (A.6)
2. Perfect competition. From (A.4) and (20), we get:
1
2 (1
2
)
(j
c
H
j
)
2
= (1 j
c
H
)^0. (A.7)
Let i = j
c
H
j
= j
c
H
1 0 and . =
_
1
2
_
^0. Then Q(i) = i
2
2. (i ) = 0 and
solving this polynomial yields i = .
_
.
2
2. 0, or
j
c
H
= 1 .
_
.
2
2.. (A.8)
Note that j
c
H
< 1, since .
_
.
2
2.. Using (20), the resulting eciency loss relative
to the social optimum is
1
c
=
H
[n(j
) n(j
c
H
)[ = (1 j
c
H
)
H
^0 =
_
.
_
.
2
2.
_
H
^0. (A.9)
Finally, the least-cost separating allocation is interim ecient if (21) holds, which here becomes
1
1
L
_
_
1
2
.
, or equivalently (A.10)
.
_
1
2
_
H
_
2
H
. (A.11)
36
3. Welfare under monopsony versus competition. Using (A.6) and (A.9), condition (24) becomes:
_
2
H
2
L
_
_
1
2
_
(^0)
2
< (1 j
c
H
)
H
^0 ==
H
2
L
. < i == i <
H
2
L
..
Substituting into the polynomial equation Q(i) = 0, this is equivalent to:
_
H
2
L
.
_
2
2.
_
H
2
L
.
_
=
H
L
.
2
,
which yields (25). This inequality and the interim eciency condition (A.11) are simultaneously
satised if and only if
'(
L
) =
1
L
2
L
_
1
L
L
<
1
2
^0
_
1
2
_
L
1
L
_
2
L
1
L
=
'(
L
). (A.12)
Note that:
(i) '(
L
) <
'(
L
) if and only if
H
,
L
< 1, so for any
L
1,2, (A.12) denes a nonempty
range for (,^0)
_
,(1
2
)
.
(ii) As
L
1, '(
L
) 0 and
'(
L
) , so arbitrary values of (,^0)
_
,(1
2
)
become feasible, including arbitrarily large values of or arbitrarily low values of ^0. In particular,
imposing < 1(1
2
)
H
^0,
L
to ensure 0 < j
< j
m
L
is never a problem for
L
large enough.
4. Imperfect competition. In Region I, j
H
(t) is dened as the solution to (B.23) in Appendix
B, which here becomes:
(j
H
j
)
2
2(1
2
)
(1 j
H
)^0
t
L
^0
(j
H
j
)
(1
2
)
= 0 ==
i
2
2 (t,
L
^0) i 2.(i) = i
2
2(. t,
L
^0)i 2. = 0,
with the above denitions of . and i = j
H
j
. Solving, we have:
j
H
(t) = 1 t,
L
^0 .
_
(. t,
L
^0)
2
2.. (A.13)
It is easily veried that j
H
(0) = j
c
H
and
L
^0 j
0
H
(t) = 1
. t,
L
^0
_
(. t,
L
^0)
2
2.
< 0. (A.14)
Moreover, this expression is increasing in t, so j
H
(t) is decreasing and convex over Region I.
In Region II, j
H
(t) is dened as the solution to (B.26) in Appendix B, which here becomes:
n(j
) 0
L
1
(j
H
j
)
2
2(1
2
)
(1 j
H
)^0
t
^0
(j
H
j
)
(1
2
)
l t = 0 ==
37
i
2
2(1
2
)
(i )^0
t
^0
i
(1
2
)
0
L
1
l t n(j
) = 0 ==
i
2
2(i ).
2ti
^0
2(1
2
)
_
l t 0
L
1 n(j
)
_
= 0 ==
i
2
2i(. t,^0) 2. 2(1
2
)
_
n(j
) 0
L
1
l t
_
= 0.
Solving, we have:
j
H
(t) = 1 t,^0 .
_
(. t,^0)
2
2[. (1
2
)
_
n(j
) 0
L
1
l t
_
[, (A.15)
noting that the expression under the square root can also be written as .
2
(t,^0)
2
2[.
(1
2
)
_
n(j
) 0
L
1
l
_
[ 0. Moreover,
j
0
H
(t)(^0) = 1
1
_
1 2[. (1
2
)
_
n(j
) 0
L
1
l t
_
,(. t,^0)
2
[
< 0
and it is increasing in t, so j
H
(t) is decreasing and convex over Region II.
In Region III, j
L
(t) is dened as the solution to (B.30). Denoting i = j
L
j
) 0
L
1
l t (i)^0
t
L
H
^0
i
(1
2
)
= 0 ==
(1
2
)
_
n(j
) 0
L
1
l t ^0
= i
_
.
t
L
H
^0
_
.
Solving, we have:
j
L
(t) = 1
(1
2
)
_
l t n(j
) 0
L
1
.
. t
L
,
H
^0
. (A.16)
It is easily veried that lim
t!+1
j
L
(t) = 1 (1
2
)
H
^0,
L
= j
m
L
. Moreover, by (11),
j
0
L
(t)(^0) =
(
L
,
H
)
_
l n(j
) 0
L
1 ^0
.^0
(. t
L
,
H
^0)
2
,(1
2
)
< 0
and this function is increasing in t, implying that j
L
(t) is convex.
Welfare eects of transport costs (claim following Proposition 5). Let \(t) =
n( j
H
(t))
L
n( j
L
(t)) 1
0 and
\(t) = \(t) t,4. By Proposition 5, \
0
(t) 0 for all
t < t
2
. We now nd conditions ensuring that
\
0
(t) 0 for t small enough. For t _ t
1
, \
0
(t) =
H
n
0
( j
H
(t)) j
0
H
(t). With quadratic costs, and using (A.4), (A.8) and (A.14),
H
n
0
(j
c
H
) j
0
H
(0) =
(
H
,
L
)
_
_
1 2,. 1
__
1 1,
_
1 2,.
_
, which for small ^0 is equivalent to (
H
,
L
)
_
2,.. As seen from (A.10), interim eciency requires
H
_ (1 2,.)
1=2
-
_
.,2.
Letting
H
/
_
.,2 yields
H
\
0
(0) j
0
H
(0) - 1,
L
1,4, hence the result.
38
Proof of Propositions 12 and 13. The rst-order conditions of the rst-best, monopsony and
competitive problems lead to very similar systems of linear equations,
1 j
A
i
j
B
i
r
_
1
2
_
o
2
A
j
A
i
= i
_
1
2
_
^0
A
, (A.17)
1 j
B
i
j
A
i
r
_
1
2
_
o
2
B
j
B
i
= i
_
1
2
_
^0
B
, (A.18)
with the only dierence being that (j
i
= j
) _ 0 and /(j
)
_ 0, is then equivalent to
43
ro
2
A
1 ro
2
B
_
1
1
_
1 ro
2
A
ro
2
B
. (A.20)
Next, subtracting the rst-best solution from (A.17)-(A.18) and denoting r
= j
, t =
, 1, yields
[1 r
_
1
2
_
o
2
A
[r
A
r
B
= i
_
1
2
_
^0
A
, (A.21)
r
A
[1 r
_
1
2
_
o
2
B
[r
B
= i
_
1
2
_
^0
B
, (A.22)
from which j = r
B;c
H
,r
A;c
H
= r
B;m
L
,r
A;m
L
is easily obtained. Its comparative statics follow from
direct computation.
Proof of Proposition 14. (1) It easily seen that a(j
c
H
),/(j
c
H
) < a(j
),/(j
) < a(j
m
L
),/(j
m
L
) if
and only if j
B
,j
A
< r
B;c
H
,r
A;c
H
= r
B;m
L
,r
A;m
L
= j. Using (A.19) and (49), this means:
_
1 r
_
1
2
_
o
2
B
^0
A
^0
B
_
1 r (1
2
) o
2
A
^0
B
^0
A
<
ro
2
B
(1)
ro
2
A
(1 ) 1
.
This can be rewritten as
^0
A
^0
B
<
_
1 r
_
1
2
_
o
2
A
_
ro
2
B
(1)
_
ro
2
A
(1 ) 1
_
1 r (1
2
) o
2
B
_
ro
2
A
(1 ) 1
_
ro
2
B
(1)
=
_
1 r
_
1
2
_
o
2
A
ro
2
B
(1) 1 ro
2
A
[1[
_
1 r (1
2
) o
2
B
ro
2
A
(1 ) 1 ro
2
B
(1 )
,
43
An alternative way of ensuring that o remains non-negative (allowing o
2
A
to become arbitrary large) is of course to
incorporate intrinsic motivation ao into (40), with a 1. The model then nests that of Section 2 as a limiting case
for (o
2
A
, o
2
B
) ! (+1, 0). Alternatively, o < 0 (say) may be interpreted as nefarious or antisocial activities (stealing
coworkers ideas, devising schemes to deceive customers, et.) that require eort but allow the agent to increase his
performance and bonus earned in the 1 dimension.
39
which simplies to (50).
(2) We have a(j
c
H
) < a(j
) < a(j
m
L
) if and only if r
A;c
H
< r
B;c
H
and r
A;m
L
r
B;m
L
, which given
that r
;c
H
0 r
;m
L
for t = , 1, means that j 1. This occurs when
_
1 r
_
1
2
_
o
2
A
^0
B
2
^0
A
_
1 r
_
1
2
_
o
2
B
^0
A
^0
B
==
r
_
1
2
_ _
o
2
A
^0
B
o
2
B
^0
A
(1 )
2
^0
A
== r
_
o
2
A
^0
B
o
2
B
^0
A
^0
A
,
which yields (51). Furthermore, /(j
c
H
) /(j
) if only if r
B
H
r
A
H
, i.e. j , which is implied by
j 1. Note, on the other hand, that competition always increases total gross output above the
ecient level, a(j
)1/(j
) < a(j
c
H
)1/(j
c
H
), if only if 0 <
_
r
A
H
r
B
H
_
1
_
r
B
H
r
A
H
_
, or
equivalently (since r
A
H
0) 0 < (1 j) 1 (j ) = 1j(1), which always holds.
For a monopsonist r
A
L
< 0, so the same condition yields a(j
m
L
) 1/(j
m
L
) < a(j
) 1/(j
).
Appendix B: Main Proofs
Proof of Proposition 1. Only the comparative-statics results remains to prove. First, dieren-
tiating (13) and (15) with respect to ^0 yields
0j
m
L
0^0
=
H
L
1
n
yy
(j
m
L
)
< 0 <
H
n
y
(j
m
L
)
n
yy
(j
m
L
)
=
01
m
0^0
. (B.1)
Turning next to and using (9), we have
0j
m
L
0
=
n
yA
(j
m
L
)
n
yy
(j
m
L
)
=
a
0
(j
m
L
)
n
yy
(j
m
L
)
< 0 (B.2)
1
L
01
m
0
= n
A
(j
; , 1) n
A
(j
m
L
; , 1) n
y
(j
; , 1)
0j
0
n
y
(j
m
L
; , 1)
0j
m
L
0
= a(j
) a(j
m
L
)
H
L
^0
0j
m
L
0
, (B.3)
showing clearly the two opposing eects discussed in the text, and which exactly cancel out in the
quadratic-cost case (see (A.5)). Similarly, for 1 :
0j
m
L
01
=
n
yB
(j
m
L
)
n
yy
(j
m
L
)
=
/
0
(j
m
L
)
n
yy
(j
m
L
)
0 (B.4)
1
L
01
m
01
= n
B
(j
; , 1) n
B
(j
m
L
; , 1) n
y
(j
; , 1)
0j
01
n
y
(j
m
L
; , 1)
0j
m
L
01
= /(j
) /(j
m
L
)
H
L
^0
0j
m
L
01
. (B.5)
Proof of Lemma 1. Denote l
c
L
and l
c
H
the two types utilities in the LCS allocation, and recall
that the former takes the same value as under symmetric information.
Claim 1 The LCS allocation is interim ecient if and only if it solves the following program:
40
(T) : max
(U
L
;U
H
;y
H
;y
L
)
l
H
, subject to:
l
L
_ l
c
L
= n(j
) 0
L
1 (B.6)
l
L
_ l
H
j
H
^0 (B.7)
l
H
_ l
L
j
L
^0 (B.8)
0 _
i=H;L
i
[n(j
i
) 10
i
l
i
[ , (B.9)
Proof. Conditions (B.7) and (B.8) are the incentive constraints for types 1 and H respectively,
and condition (B.9) the employers interim break-even constraint. Now, note that:
(i) If the LCS allocation does not achieve the optimum, interim eciency clearly fails.
(ii) If the LCS allocation solves (T), there can clearly be no incentive-compatible Pareto im-
provement in which l
H
l
c
H
, but neither can there be one in which l
H
= l
c
H
and l
L
l
c
L
.
Otherwise, note rst that one could without loss of generality take such an allocation, to satisfy
j
H
_ j
; otherwise, replacing j
H
by j
while keeping l
H
and l
L
unchanged strictly increases
prots, so the LCS allocation remains (even more) dominated. Starting from such an allocation
with j
H
_ j
L
H
n
0
(j
H
)^0
.
Both types and the rm are now strictly better o than in the LCS allocation, contradicting the
fact that it is a solution to (T).
To study interim eciency and prove Lemma 1, let us therefore analyze the solution(s) to (T).
First, condition (B.9) must be binding, otherwise l
L
and l
H
could be increased by the same small
amount without violating the other constraints. Second, (B.7) must also be binding, otherwise
solving (T) without that constraint and with (B.9) as an equality leads to j
L
= j
= j
H
, l
L
= l
c
L
and l
H
= n(j
) 10
H
; thus l
H
l
L
= 1^0, violating (B.7). Third, (B.8) now reduces to
j
H
_ j
L
. Two cases can then arise:
(i) If (B.6) is binding, the triple (l
L
, l
H
, j
H
) is uniquely given by the same three equality
constraints as the LCS allocation, and thus coincides with it.
(ii) If (B.6) is not binding, the solution to (T) is the same as when that constraint is dropped.
Substituting (B.7) into (B.9), and both being equalities, we have l
H
=
i
[n(j
i
) 10
i
[
L
j
H
^0,
so (T) reduces to
max
y
H
;y
L
H
[n(j
H
) (
L
^0,
H
) j
H
[
L
n(j
L
) [j
H
_ j
L
. (B.10)
For all r _ 0, dene the function j(r) = aig max
y
n(j)rj and let r = n
0
(1). On the interval
[0, r[ the function j is given by n
0
( j(r)) = r, so it is strictly increasing up to j( r) = 1, while
41
for r r, j(r) _ 1. Furthermore, it is clear that j(r) _ j
) 10
L
< l
H
j
H
^0 = l
L
, if and only if
n(j
) 10
L
<
H
[n(j
H
) 10
H
[
L
[n(j
) 10
L
[
H
j
H
^0,
with j
H
= j(
L
^0,
H
). Equivalently, H(
L
^0,
H
) 0, where
H(r) = n( j(r)) n(j
) - 2 (1 j
) ^0,
so that
L
- 1
_
^0, where 1, =
_
2n"(j
) (j
1).
Proof of Proposition 2. To complete the proof of Results (1) and (2), it just remains to show
that: (a) If the LCS allocation is interim ecient, it is a competitive equilibrium; (b) It is then the
unique one. We shall also prove here that: (c) If the LCS allocation is not interim ecient, there
exists no competitive equilibrium in pure strategies.
Claim 2 In any competitive equilibrium, the utilities (l
L
, l
H
) must satisfy
l
L
_ l
c
L
= l
SI
L
= n(j
) 0
L
1, (B.12)
l
H
_ l
c
H
= n(j
c
H
) 0
H
1, (B.13)
Proof. If l
L
< l
c
L
, a rm could oer the single contract (j = j
, . = .
c
L
-) for - small,
attracting and making a prot - on type 0
L
(perhaps also attracting the more protable type 0
H
).
Similarly, if l
H
< l
c
H
, it could oer the incentive-compatible menu (j
, .
c
L
-), (j
c
H
, .
c
H
-)
thereby attracting and making a prot - on type 0
H
(perhaps also attracting and making zero
prot on type 0
L
).
Claim 3 If an allocation Pareto dominates (in the interim-eciency sense) the least-cost separat-
ing one, it must involve a cross-subsidy from high to low types, meaning that
n(j
H
) 10
H
l
H
0 n(j
L
) 10
L
l
L
, (B.14)
Proof. If l
L
_ n(j
L
)0
L
1, then l
L
_ l
c
L
requires that j
L
= j
and l
L
= n(j
L
)0
L
1 = l
c
L
.
Incentive-compatibility and Pareto-dominance then imply that l
L
j
H
^0 _ l
H
l
c
H
= l
L
j
c
H
^0, hence j
H
j
c
H
. This, in turn, leads to n(j
H
) 0
H
1 l
H
< n(j
c
H
) 0
H
1 l
H
=
l
c
H
l
H
< 0, violating the break-even condition. Therefore, it must be that n(j
L
) 10
L
l
L
<
42
0, meaning that low types get more than the total surplus they generate. For the employer to break
even, it must be that high types get strictly less, n(j
H
) 10
H
l
H
0.
We are now ready to establish the properties (a)-(c) listed above, and thereby complete the
proof of Results (1) and (2) in Proposition 2.
(a) Suppose that the LCS allocation, dened by (16)-(20), is oered by all rms. Could another
one come in and oer a dierent set of contracts, leading to new utilities (l
L
, l
H
) and a strictly
positive prot? First, note that we can without loss of generality assume that l
L
_ l
c
L
: if
l
L
< l
c
L
and l
H
is indeed selected (with positive probability) by type H, then (l
c
L
, l
H
) is
incentive-compatible. By oering l
c
L
to 1 types (via their symmetric-information allocation), the
deviating rm does not alter its protability. Second, if l
H
< l
c
H
, the deviating employer does
not attract type H; since it cannot make money on type 1 while providing l
L
_ l
c
L
, the deviation
is not protable. Finally, suppose that l
L
_ l
c
L
and l
H
_ l
c
H
. If at least one inequality is strict,
then interim eciency of the LCS allocation implies that the deviating rm loses money. If both
are equalities, let us specify (for instance) that both types workers, being indierent, do not select
the deviating rm.
(b) By (B.12)-(B.13), in any equilibrium both types must be no worse o than in the LCS allo-
cation, and similarly for the rm, which must make non-negative prots. If any of these inequalities
is strict there is Pareto dominance, so when the LCS allocation is interim ecient, they must all
be equalities, giving the LCS allocation as the unique solution.
(c) Suppose now that LCS allocation is not interim ecient. The contract that solves (T)
is then such that j
H
= j(r), j
L
= j
, l
L
l
c
L
(equation (B.6) is not binding) and l
H
l
c
H
(since the LCS does not solve (T)). A rm can then oer a contract with the same j
H
and j
L
but reducing both l
H
and l
L
by the same small amount, resulting in positive prots; the LCS
allocation is thus not an equilibrium. Suppose now that some other allocation, with utilities l
L
and l
H
, is an equilibrium. As seen in (b), it would have to Pareto-dominate the LCS allocation,
which by Claim 3 implies:
l
L
_ l
H
j
H
^0,
n(j
H
) 10
H
l
H
0.
Consider now a deviating employer oering a single contract, aimed at the high type: j
0
H
= j
H
-
and l
0
H
= l
H
(-^0),2 < n(j
0
H
) 10
H
. The low type does not take it up, as it would yield
l
0
L
= l
L
(-^0),2. The high type clearly does, leading to a positive prot for the deviator.|
The only part of Proposition 2 remaining to prove are the comparative static results. Dieren-
tiating (20) and (22) with respect to ^0 yields
0j
c
H
0^0
=
1 j
c
H
^0 n
y
(j
c
H
; , 1)
0,
01
c
0^0
=
H
n
y
(j
m
H
; , 1)
0j
c
H
0^0
0. (B.15)
Turning next to ,
43
^0
0j
c
H
0
= n
A
(j
; , 1) n
A
(j
c
H
; , 1) n
y
(j
; , 1)
0j
0
n
y
(j
c
H
; , 1)
0j
c
H
0
= a(j
) a(j
c
H
) n
y
(j
c
H
; , 1)
0j
c
H
0
=
0j
c
H
0
=
a(j
c
H
) a(j
)
^0 n
y
(j
c
H
; , 1)
< 0 <
H
^0
0j
c
H
0
=
01
c
0
. (B.16)
Again there is a direct and an indirect eect of on 1
c
, but now the direct one always dominates.
For 1, in contrast, the ambiguity remains:
^0 ^0
0j
c
H
01
= n
B
(j
; , 1) n
B
(j
c
H
; , 1) n
y
(j
; , 1)
0j
01
n
y
(j
c
H
; , 1)
0j
c
H
01
= /(j
) /(j
c
H
) n
y
(j
c
H
; , 1)
0j
c
H
01
= (B.17)
0j
c
H
01
=
^0 /(j
c
H
) /(j
)
^0 n
y
(j
c
H
; , 1)
0, (B.18)
1
H
^0
01
c
01
= 1
0j
c
H
01
=
n
y
(j
c
H
; , 1) /(j
c
H
) /(j
)
^0 n
y
(j
c
H
; , 1)
. (B.19)
In the quadratic case, 1
c
is independent of 1 and the last term thus equal to zero; see (A.9).
Proof of Proposition 4. We solve for the symmetric equilibrium under the assumption that
market shares are always interior, and thus given by (29). In Appendix D we verify that individual
deviations to corner solutions (one rm grabbing the whole market for some worker type, or on the
contrary dropping them altogether) can indeed be excluded.
To characterize the symmetric solution to (30)-(33), we distinguish three regions.
Region I. Suppose rst that the low types individual rationality constraint is not binding,
l
L
l, so that i = 0.
Lemma 3 If i = 0, then j
H
= 0 _ j
L
and j
L
= j
_ j
H:
Proof. (i) If j
H
= j
L
= 0, then j
H
= j
L
= j
L
= t
L
= 0, whereas
H
L
= 1^0 (l
H
l
L
) =
(1 j
)^0 0, a contradiction.
(ii) If j
H
0 = j
L
condition (37) implies n
0
(j
L
) 0, hence j
L
< j
H
L
= n(j
) n(j
L
) (1 j
L
)^0 0,
a contradiction. We are thus left with j
H
= 0 < j
L
, which implies j
L
= j
< j
H
by (36)-(37).
Let us now derive and characterize j
H
as a function of t. We can rewrite (36) as
t
H
n
0
(j
H
) = j
L
^0 =
L
(
L
t)^0. (B.20)
Summing (34)-(35) and recalling that
i
= n(j
i
) 0
i
1 l
i
yields
44
l
L
t =
H
[n(j
H
) 0
H
1 (l
H
l
L
)[
L
[n(j
) 0
L
1[
=
H
[n(j
H
) 0
H
1 j
H
^0[
L
[n(j
) 0
L
1[ , (B.21)
where the second equality reects the fact that (32) is an equality, since j
L
0. Therefore:
L
t = n(j
) 0
L
1 l
L
t
= n(j
) 0
L
1
L
[n(j
) 0
L
1[
H
[n(j
H
) 0
H
1 j
H
^0[
=
H
[n(j
) n(j
H
) (1 j
H
)^0[ (B.22)
Substituting into (B.20) yields
1(j
H
; t) = n(j
H
) n(j
) (1 j
H
)^0
tn
0
(j
H
)
L
^0
= 0. (B.23)
The following lemma characterizes the equilibrium value of j
H
over Region I, denoted j
I
H
(t).
Lemma 4 For any t _ 0 there exists a unique j
I
H
(t) (j
) 0 1(1),
hence existence and uniqueness. Strict monotonicity then follows from the fact that 1 is strictly
decreasing in t, while setting t = 0 in (B.23) shows that j
I
H
(0) must equal j
c
H
, dened in (20) as
the unique solution to n(j
) n(j
c
H
) = (1 j
c
H
)^0. It only remains to verify that the solution
j
I
H
(t) is consistent with the initial assumption that i = 0, or equivalently l
L
l. By (B.21), we
have for all j
H
l
L
t =
H
[n(j
H
) 0
H
1 j
H
^0[
L
[n(j
) 0
L
1[
= n(j
) 0
L
1
H
[(1 j
H
) ^0 n(j
H
) n(j
)[ .
For j
H
= j
I
H
(t), the corresponding value of l
L
is strictly above
l if and only if w(t)
l t,
where we dene for all t:
w(t) = n(j
) 0
L
1
H
__
1 j
I
H
(t)
_
^0 n(j
) n( j
I
H
(t))
. (B.24)
Lemma 5 There exists a unique t
1
0 such that w(t) _
l t if and only if t _ t
1
. On [0, t
1
[, the
low types utility l
L
is strictly decreasing in t, reaching
l at t
1:
Proof. At t = 0 the bracketed term is zero by denition of j
I
H
(0) = j
c
H
, so w(0) = n(j
) 0
L
1
l by (14), which stated that a monopsonist hires both types, and lim
t!+1
_
w(t)
l t
= ,
there exists at least one solution to w(t) =
l t. To show that it is unique and the monotonicity
of l
L
, we establish that, w
0
(t) < 1 for all t 0. From (B.23) and (B.24), this means that
45
H
_
^0 n
0
( j
H
)
_
n
0
( j
H
),
L
^0
^0 n
0
( j
H
) tn
00
( j
H
),
L
^0
_
< 1 ==
H
_
^0 n
0
( j
H
)
_
n
0
( j
H
),
L
^0
_
< ^0 n
0
( j
H
) tn
00
( j
H
),
L
^0 ==
H
_
^0 n
0
( j
H
)
_
n
0
( j
H
)
_
<
L
^0
_
^0 n
0
( j
H
)
tn
00
( j
H
) ==
tn
00
( j
H
) <
_
^0 n
0
( j
H
)
_
H
n
0
( j
H
)
L
^0
where we abbreviated j
I
H
(t) as j
H
. In the last expression, the rst bracketed term is always non-
negative, whereas in the second one j
< j
H
< j
c
H
implies that
H
n
0
( j
H
)
L
^0
H
n
0
( j
c
H
)
L
^0 0, by (21).
In summary, Region I consists of the interval [0, t
1
[, where t
1
is uniquely dened by w(t
1
) =
t
1
l. Over that interval, j
L
= j
while j
H
= j
I
H
(t) is strictly decreasing in t, and therefore so
is the high types relative rent, l
H
l
L
= j
I
H
(t)^0. The low types utility level l
L
need not be
declining, but its starts at a positive value and reaches
l exactly at t
1
.
For t _ t
1
, the constraint l
L
_
l is binding. Recalling that j
H
j
L
must always equal zero,
we distinguish two subregions, depending on whether j
H
= 0 (Region II) or j
L
= 0 (Region III),
and show that these are two intervals, respectively [t
1
, t
2
[ and [t
2
, ), with t
1
< t
2
. Thus, inside
Region II the low types incentive constraint is binding but not the high types (j
L
0 = j
H
for
t (t
1
, t
2
)), whereas inside Region 2 it is the reverse (j
H
0 = j
L
for t t
2
).
Region II. Consider rst the values of t where j
H
= 0 < j
L
. As before, this implies that
j
L
= j
< j
H
and l
H
l
L
= j
H
^0, or l
H
=
l j
H
^0 since l
L
=
l. Therefore:
j
L
=
H
(
H
t) =
H
[n(j
H
) 0
H
1 l
H
t[ =
H
_
n(j
H
) 0
H
1 j
H
^0
l t
. (B.25)
Substituting into condition (36), the latter becomes
I(j
H
; t) = n(j
H
) 0
H
1 j
H
^0
l t
tn
0
(j
H
)
^0
= 0. (B.26)
On the interval [j
) 0
L
1
_
1
1
L
_
tn
0
(j
I
H
(t
1
))
^0
l t
= n(j
) 10
L
t
_
1
H
L
n
0
(j
H
)
^0
_
l. (B.27)
At t = t
1;
substituting (B.23) into (B.24) yields I( j
I
H
(t
1
); t
1
) = 0. Furthermore, as t rises above t
1
,
j
I
H
(t) decreases, so n
0
( j
I
H
(t)) increases. Since
L
^0
H
n
0
( j
H
(t))
L
^0
H
n
0
( j
H
(0)) =
L
^0
H
n
0
(j
c
H
) 0
46
by (21), t
_
L
^0
H
n
0
( j
I
H
(t))
; t) = n(j
)0
H
(1j
)0
L
j
lt.
Dene therefore
t
2
= n(j
) 0
H
(1 j
) 0
L
j
l, (B.28)
and note that
t
1
= n(j
) 0
L
1
H
__
1 j
I
H
(t
1
)
_
^0 n(j
) n( j
I
H
(t
1
))
l
< n(j
) 0
L
1
H
_
1 j
I
H
(t
1
)
_
^0
l
< n(j
) 0
L
1 1 (1 j
) ^0
l = t
2
.
Lemma 6 For all t [t
1
, t
2
[, there exists a unique j
II
H
(t) [j
, j
I
H
(t
1
))[ such that I( j
II
H
(t); t) = 0.
Furthermore, j
II
H
(t) is strictly decreasing in t, starting at j
II
H
(t
1
) = j(t
1
) and reaching j
at t = t
2
.
For all t t
2
, I(j
H
; t) < 0 over all j
H
_ j
.
Proof. For t [t
1
, t
2
[ we have shown that I( j
H
(t
1
); t) _ 0 _ I(j
n
0
L
^0(^0 n
0
) tn
00
<
^0 n
0
^0(^0 n
0
) tn
00
) ==
tn"
^0 n
0
_
H
n
0
L
^0
_
.
with all derivatives evaluated at j
I
H
(t
1
) = j
II
H
(t
1
). Since j
< j
I
H
(t
1
) < j
c
H
the term on the left is
positive and that on the right negative.
As to t
2
, note that it is the only point where j
H
= 0 = j
L
(the only intersection of Regions II
and III). Indeed, this require j
H
= j
= j
L
by (36)-(37) and condition (37) together with l
L
=
l
then implies that t =
L
= n(j
) 0
L
1 j
^0
L
l = t
2:
Region II thus consists of the interval [t
1
, t
2
[. Over that interval, j
L
= j
while j
H
= j
II
H
(t) is
strictly decreasing in t, and therefore so is the high types utility, l
H
=
l j
H
(t)^0, while the
low types utility remains xed at l
L
=
l.
Putting together Regions I and II, we shall dene:
j
H
(t) =
_
j
I
H
(t) for t [0, t
1
[
j
II
H
(t) for t [t
1
, t
2
[
. (B.29)
Region III. Inside this region, namely for t t
2
, we have l
L
=
l but now j
H
j
L
= 0.
This implies that j
H
= j
j
L
by (36)-(37) and l
H
=
l j
L
^0 by (31). Furthermore,
j
H
=
H
(t
H
) =
H
_
t
l j
L
^0 n(j
) 0
H
1
) 0
H
1 j
L
^0
l t
t
L
n
0
(j
L
)
^0
= 0. (B.30)
On the interval [0, j
; t) =
H
_
n(j
) 0
H
1 j
^0
l t
=
H
(t
2
t) < 0.
Recall now that the monopsony price j
m
L
is uniquely dened by n
0
(j
m
L
) = (
H
,
L
) ^0. Therefore:
A(j
m
L
; t) =
H
_
n(j
) 0
L
1
l (1 j
m
L
^)0
0.
Lemma 7 For all t _ t
2
there exists a unique j
L
(t) such that A( j
L
(t); t) = 0, and j
m
L
< j
L
(t) _ j
,
with equality at t = t
2
. Furthermore, j
L
(t) is strictly decreasing in t and lim
t!+1
j
L
(t) = j
m
L
.
Proof. Existence and uniqueness have been established. Next, 0A(j; t),0t =
L
n
0
(j),^0 1.
At j = j
L
(t), this equals 1,t times
H
_
n(j
) 0
H
1 j
L
^0
l t
t =
L
t
H
_
n(j
) 0
H
1
l j
L
^0
< 0,
so the function j
L
(t) is strictly decreasing in t. Taking limits in (B.30) as t , nally, yields
as the unique solution lim
t!+1
j
L
(t) = j
m
L
.
Proof of Proposition 6. The fact that 0l
L
,0t < 0 over Region I was shown in Lemma 5. To
show the last result, note that over Region III, we have
2H =
H
[n(j
) 0
H
1 j
L
^0[
L
[n( j
L
) 0
L
1[
l =
1
L
0H
0 j
L
= n
0
( j
L
)
H
L
^0 n
0
( j
m
L
)
H
L
^0 = 0,
so prots fall as t declines, as was shown to be the case over Regions I and II.
Proof of Proposition 7. Consider rst total pay. Since .
i
= l
i
n(j
i
) 0
i
j
i
, we can write
1
i
= l
i
/(j
i
)j
i
n(j
i
), for i = H, 1. As t declines, l
i
and j
i
increase (at least weakly) and
therefore so does 1
i
, since n
0
(j) = /. Furthermore,
1
H
1
L
= l
H
l
L
/(j
H
)j
H
n(j
H
) n(j
L
) /(j
L
)j
L
.
Over Regions I and II this becomes [^0 /(j
H
)[ j
H
n(j
H
) plus a constant term, with j
H
= j
H
(t);
the result then follow from n
0
(j) = /. Over Region III, 1
H
1
L
= [^0 /(j
L
)[ j
L
n(j
L
) plus a
constant term, with j
L
= j
L
(t); therefore, 0(1
H
1
L
),0t < 0 if and only if /
0
(j
L
)j
L
< ^0, which
need not hold in general. With quadratic costs, /
0
(j
L
) = 1,(1
2
) so it holds on [t
2
, ) if and
only if j
= 1 < (1
2
)^0. Turning now to performance-based pay, we have
0([/(j
H
) 0
H
[ j
H
[/(j
L
) 0
L
[ j
L
)
0t
=
_
/(j
H
) 0
H
j
H
/
0
(j
H
)
0j
H
0t
_
/(j
L
) 0
L
j
L
/
0
(j
L
)
0j
L
0t
.
In Regions I and II the rst term is negative and the second zero; in Region III it is the reverse.
48
Turning nally to xed wages, .
H
.
L
= l
H
l
L
0
H
j
H
0
L
j
L
n(j
H
) n(j
L
). In Regions
I and II, .
H
.
L
= 0
L
(j
H
j
) n(j
H
) n(j
) is decreasing in j
H
, hence increasing in t. In
Region III, .
H
.
L
= (j
L
j
) 0
H
n(j
) n(j
L
), so the opposite holds.
Proof of Proposition 9. For any j [0, j
c
H
), let j
= minj
,
L
= 0 and n( j
) 10
L
=
l
SI
L
,
so in equilibrium they must receive at least that much. Consider therefore the relaxed program
(from which high types incentive-compatibility constraint has been omitted):
(T
r
) : max
f(U
i
;0y
i
y; 0
i
)g
i=H;L
l
H
, subject to:
l
L
_
l
SI
L
(i) (B.31)
l
L
_ l
H
j
H
^0
H
^` (j
L
) (B.32)
0 _
i=H;L
i
[n(j
i
) 10
i
(1 `
i
)
i
l
i
[, () (B.33)
The rst-order conditions in l
H
and l
L
are respectively 1 j
L
H
= 0 and j
L
i
L
= 0;
thus = 1 i 0, (B.33) so that must bind, and j
L
=
L
(1 i) i. The rst-order conditions
in j
H
and
H
then take the form
(1 i)
_
L
^0
H
n
0
(j
H
)
) 10
L
[
H
j
H
^0 =
l
L
l
SI
L
_
H
[n(j
H
) (1 j
H
)^0 n( j
)[ . (B.36)
Let us rst show that j
H
= j. Otherwise, (B.34) implies that i,(1i) =
L
H
n
0
(j
H
),^0 0
(by (21)), so l
L
=
l
SI
L
; meanwhile, j
L
_ 0 requires i,(1 i) _
L
, so j
H
_ j
. But then j
= j
, j
c
H
)), contradicting l
L
=
l
SI
L
.
Next, with j
H
= j, the right-hand side of (B.36) equals
H
(1 j)^0 if j _ j
and
H
[n( j)
(1 j)^0 n(j
)[ if j
^0, it satises
the (omitted) high types incentive constraint. It is separating for j j
,
since (j
H
,
H
) = (j
L
,
L
) = ( j, 0) and l
H
l
L
= j^0, implying .
H
= .
L
. As it solves the
relaxed problem it is interim ecient and therefore (by standard arguments) the unique equilibrium.
Finally, l
H
= n( j) 10
H
and l
L
= n( j
) j^0 10
L
both increase as j declines to j
.
49
Case 2. When (39) holds, (B.31) must bind, otherwise i = 0 and (B.35) fails. From (B.32) and
the binding (B.33) we have l
H
= n(j
H
) 10
H
(1 `
H
)
H
_
l
SI
L
j
H
^0
H
^`, so
(1 `
L
)
H
_ (1 j
H
)^0 n(j
H
) n( j
) 0 (B.37)
since j
H
_ j < j
c
H
. With
H
0, (B.35) must be an equality, which yields i,(1 i) =
L
H
(1 `
H
),^` 0 and j
L
,(1 i) =
L
i,(1 i) 0. Thus (B.32) is binding, (B.37) holds
with equality, and the left-hand side of (B.34) becomes (1 i)
H
[n
0
(j
H
) (1 `
H
)^0,^`[ 0,
by (38); therefore j
H
= j. By (B.37),
H
= [(1 j)^0 n( j) n( j
)[,(1 `
L
) is then strictly
increasing as j decreases from j
c
H
to 0. Since l
H
l
L
= j^0
H
^` j
L
^0, the high types
omitted incentive constraint is also satised. The solution to the relaxed program thus coincides
with the constrained-LCS allocation described in Proposition 9-(2), which is thus interim ecient
and therefore, the unique equilibrium.
Finally, consider how welfare varies with j. Prots always equal zero and low types always
receive
l
SI
L
, which is increasing in j below j, then constant. As to high types, they achieve
l
H
= n( j) 10
H
_
1 `
H
1 `
L
_
[(1 j)^0 n( j) n( j
)[. (B.38)
The right-hand side is strictly concave in j on [j
, j
c
H
[ and decreasing below j
c
H
, by (38). Therefore,
l
H
is strictly increasing in j and maximized at j
c
H
, where the constraint ceases to bind.
Proof of Proposition 10. Let us dene . and j as net compensations. In particular, j is still
the eective power of the incentive scheme. Prot on type i = H, 1 under contract (j, .) is then
H
i
= a(j) 1[0
i
/(j)[
. j[0
i
/(j)[
1 t
, (B.39)
while the expression for l
i
is unchanged. Furthermore,
l
i
(1 t)H
i
= (1 t) [a(j) 1(0
i
/(j))[ [C(a(j), /(j)) a(j)[
= n(j) (1 t)10
i
. (B.40)
Let j
(t) _ j
(t) and j
H
given by n(j
(t)) n(j
H
) = ^0 [(1 t)1 j
H
[ . Welfare \
is equal to
H
n(j
H
)
L
n(j
L
) 10, and so
d\
dt
[
=0
=
L
n
0
(j
L
)
dj
dt
H
n
0
(j
H
)
dj
H
dt
=
H
n
0
(j
H
)
dj
H
dt
. (B.41)
Finally, for small t,
n
0
(j
H
) = n
0
(j
H
) t
d
dj
H
[a(j
H
) 1/(j
H
)[ = n
0
(j
H
) o(t) =
d\
dt
[
=0
=
1^0
^0 n
0
(j
H
)
H
n
0
(j
H
) 0. (B.42)
50
Lemma 8 The rst-best solution dened by (42) satises j
A
< and j
B
< 1.
Proof. The rst-order conditions (42) take the form
(j
A
)(0a,0j
A
) (1 j
B
)
_
0/,0j
A
_
= rj
A
o
2
A
,
(j
A
)(0a,0j
B
) (1 j
B
)
_
0/,0j
B
_
= rj
B
o
2
B
,
with all derivatives evaluated at (j
A
, j
B
). Let 1 = (0a,0j
A
)
_
0/,0j
B
_
(0a,0j
B
)
_
0/,0j
A
_
,
which is easily seen to equal 1,[C
aa
C
bb
(C
ab
)
2
[ 0 (this holds for any (j
A
, j
B
)). We then have
j
A
=
1
1
_
(0/,0j
B
)
_
rj
A
o
2
A
,2
_
(0/,0j
A
)
_
rj
B
o
2
B
,2
_
0,
1 j
B
=
1
1
_
(0a,0j
A
)
_
rj
B
o
2
B
,2
_
(0a,0j
B
)
_
rj
A
o
2
A
,2
_
0.
Proof of Lemma 2. The LCS allocation is interim ecient i it solves the relaxed program
max
f(U
i
; y
i
)g
i=H;L
l
H
, subject to
l
L
_ l
H
j
A
H
^0
A
j
B
H
^0
B
,
i=H;L
i
[n(j
i
) 1
i
l
i
[ _ 0,
l
L
_ l
SI
L
.
The solution to this program must satisfy j
L
= j
H
. (B.43)
Letting o denote the subsidy from the H- to the 1-type, the above program can be rewritten as:
(T
r
) : maxl
H
, subject to
l
H
_ n(j
H
) 1
H
q
L
q
H
o
l
SI
L
o _ l
H
j
H
^0
o _ 0
where j
H
^0 denotes the scalar product of j
H
=
_
j
A
H
, j
B
H
_
and ^0 =
_
^0
A
, ^0
B
_
. Note rst that
the rst two constraints must both be binding. Indeed, denoting `
i
the Lagrange multiplier on the i-
th constraint, the rst-order conditions are 1`
1
`
2
= 0 for l
H
, `
1
\n(j
H
)`
2
^0 = 0 for j
H
and
`
3
`
1
L
,
H
`
2
= 0 for o. The rst two clearly exclude `
1
= 0. If `
2
= 0, then j
H
= j
and `
3
0,
implying o = 0; but then the second constraint becomes n(j
)1
L
= l
SI
L
_ n(j
)1
H
j
^0,
hence 0 _ 1
H
1
L
j
^0 = (j
A
)^0
A
(1 j
B
)^0
B
, a contradiction of Lemma 8.
Next, eliminating o from the binding constraints shows that j
H
solves maxn(j
H
) /^0 j
H
,
51
where / =
L
,
H
(0, ) is the likelihood ratio. Consider any two such ratios / and
/ and the
corresponding optima j
H
and j
H
for this last program; if
/ /, then
n(j
H
) _ n( j
H
) /^0 ( j
H
j
H
) ,
n( j
H
) _ n(j
H
)
/^0 (j
H
j
H
) .
Adding up these inequalities yields ^0 ( j
H
j
H
) _ 0, which in turn implies that n(j
H
) _
n( j
H
). Observe now from (45) that the LCS allocation corresponds to an interior solution to
maxn(j
H
) i
c
^0 j
H
. Consider now any / i
c
and the corresponding solution j
H
. We have
n(j
H
) _ n(j
c
H
) and so n(j
H
) 1
H
/o _ n(j
c
H
) 1
H
= l
c
H
,with strict inequality if o 0.
This last case is impossible, however, since type Hs utility from the relaxed program cannot be
lower than l
c
H
. Therefore, o = 0 and j
H
= j
c
H
: the LCS allocation is interim ecient.
Conversely, let / < i
c
; we then have (as a row-vector equality)
0
0j
H
[n(j
H
) /^0 j
H
[
y
H
=y
c
H
= (| i
c
) ^0,
with ^0
A
_ 0 and ^0
B
0. Since j
H
maximizes (each component of) the expression in brackets,
it must be that j
A
H
_ j
Ac
H
and j
B
H
< j
Ac
H
, hence ^0 (j
c
H
j
H
) 0. By the same properties shown
above, it follows that n(j
H
) n(j
c
H
). If o = 0, the two binding constraints in (T
r
) then imply
l
H
= l
SI
L
j
H
^0 < l
SI
L
j
c
H
^0 = l
c
H
,
l
H
= n(j
H
) 1
H
n(j
c
H
) 1
H
= l
c
H
,
another contradiction. Therefore o must be positive after all, and interim eciency fails.
52
REFERENCES
Acemoglu, D., Kremer, M. and A. Mian (2007) Incentives in Markets, Firms and Governments,
The Journal of Law, Economics, and Organization, 24(2), 273-306.
Acharya, V., Gabarro, M., and P. Volpin (2012) Competition for Managerial Talent, Corporate
Governance and Incentive Compensation, CEPR W.P. 8936.
Acharya, V., Pagano, P. and P. Volpin (2012) Seeking Alpha: Excess Risk Taking and Competition
for Managerial Talent, NYU mimeo, May.
Akerlof, G. and Kranton, R. (2005) Identity and the Economics of Organizations, Journal of
Economic Perspectives, 19(1), 9-32.
Albagli, E., Hellwig, C. and A. Tsyvinski (2014) Risk-Taking, Rent-Seeking, and Investment When
Financial Markets are Noisy, USC mimeo, May.
Armstrong, M. and J. Vickers (2001) Competitive Price Discrimination, Rand Journal of Eco-
nomics, 32(4), 579-605.
Baker, G., R. Gibbons and K. J. Murphy (1994) Subjective Performance Measures in Optimal
Incentive Contracts, Quarterly Journal of Economics, 109, 1125-56.
Bandiera, O., Barankay, I. and I. Rasul (2007) Incentives for Managers and Inequality among
Workers: Evidence from a Firm-Level Experiment, Quarterly Journal of Economics, 122, 729-
773.
Bannier, C., Feess, E. and Packham, N. (2013) Incentive Schemes, Private Information and the
Double-Edged Role of Competition for Agents, Frankfurt University mimeo, January.
Bartling, B., Fehr, E. and K. Schmidt (2012) Screening, Competition and Job Design, American
Economic Review, forthcoming.
Baumol, W.J. (1990): Entrepreneurship: Productive, Unproductive, and Destructive, Journal of
Political Economy, 98, 893-921.
Bebchuk, L. and J. Fried (2004) Pay without Performance: The Unfullled Promise of Executive
Compensation. Cambridge: Harvard University Press, 2004.
Bell, B. and J. Van Reenen (2013) Bankers and Their Bonuses, CEPR Occasional Paper 35,
February.
Bnabou, R., and J. Tirole (2003), Intrinsic and Extrinsic Motivation, Review of Economic
Studies, 70, 489-520.
Bnabou, R., and J. Tirole (2006), Incentives and Prosocial Behavior, American Economic Re-
view, 96(5), 1652-1678.
Bertrand, M. and S. Mullainathan (2001) Are CEOs Rewarded for Luck? The Ones without
Principals Are, Quarterly Journal of Economics, 116, 901-932.
Besley, T., and M. Ghatak (2005) Competition and Incentives with Motivated Agents, American
Economic Review, 95(3), 616-636.
Besley, T., and M. Ghatak (2006) Sorting with Motivated Agents: Implications for School Com-
petition and Teacher Incentives, Journal of the European Economics Association, 4, 404-414.
Besley, T. and M. Ghatak (2011) Taxation and Regulation of Bonus Pay, mimeo, LSE.
Bijlsma, M., Boone, J. and G. Zwart (2013) Competition for Traders and Risk, mimeo, Tilburg
53
University.
Bloom, N. and J. Van Reenen (2010) Human Resource Management and Productivity, LSE-CEP
Discussion Paper No 982, May.
Bolton, P. (2013) The Good Banker, Columbia University mimeo, November.
Bolton, P., Scheinkman, J. and W. Xiong (2006)Executive Compensation and Short-Termist
Behaviour in Speculative Markets, Review of Economic Studies, 73, 577-610.
Brekke, K. A., and K. Nyborg (2008) Attracting Responsible Employees: Green Production as
Labor Market Screening, Resource and Energy Economics, 39, 509-526.
Clrier, C. and B. Valle (2014) Are Bankers Worth Their Pay? Evidence From a Talent Mea-
sure, University of Zurich mimeo.
Chassagnon, A. and P.A. Chiappori (1997) Insurance under Moral Hazard and Adverse Selection:
the Case of Pure Competition, DELTA mimeo, March.
Crocker, K. and A. Snow (1985) The Eciency of Competitive Equilibria in Insurance Markets
with Asymmetric Information, Journal of Public Economics, 26, 207-219.
Cuat, V. and M. Guadalupe (2009) Executive Compensation and Competition in the Banking
and Financial Sectors, Journal of Banking and Finance, 33, 439-474.
Cuat, V. and M. Guadalupe (2009) Globalization and the Provision of Incentives inside the Firm:
The Eect of Foreign Competition, Journal of Labor Economics, 27(2), 179-212.
Dyck, A., Morse, A. and L. Zingales (2013) How Pervasive is Corporate Fraud?, University of
Chicago, mimeo, February.
Desai, M. (2012) The Incentive Bubble, Harvard Business Review, 90(3), 124-132,
Dewatripont, M., Jewitt, I. and J. Tirole (1999a) The Economics of Career Concerns, Part 1:
Comparing Information Structures, Review of Economic Studies, 66, 183-198.
Edmans, A. Gabaix, X. and A. Landier (2009) A Multiplicative Model of Optimal CEO Incentives
in Market Equilibrium, The Review of Financial Studies, 22(12), 4881-4917.
Fabbri, F. and D. Marin (2012) What explains the Rise in CEO Pay in Germany? A Panel Data
Analysis for 1977-2009, CESifo W.P. 3757.
Fahlenbrach, R. and R. Stulz (2011) Bank CEO Incentives and the Credit Crisis, Journal of
Financial Economics, 99(1), 11-26.
Fang, F. Steen, G. and A. Casadevall (2012) Misconduct Accounts for the Majority of Retracted
Scientic Publications, Proceedings of the National Academy of Sciences, (109), October 16,
1702817033.
Fehr, E. and K. Schmidt (2004) Fairness and Incentives in a Multi-Task Principal-Agent Model,
Scandinavian Journal of Economics, 106, 453-474.
Fernandes, N. Ferreira, M., Matos, P. and K. Murphy (2012) Are US CEOs Paid More? New
International Evidence, Review of Financial Studies, 26(2): 323-367.
Frydman, C. (2007) Rising Through the Ranks: The Evolution of the Market for Corporate
Executives, 1936-2003, MIT mimeo, September.
Frydman, C. and D. Jenter (2010) CEO Compensation, Annual Review of Financial Economics,
2(1), 75-102.
54
Frydman, C. and R. E. Saks (2010) Executive Compensation: A New View from a Long-Term
Perspective, 1936-2005, Review of Financial Studies, 23, 2099-2138.
Gabaix, X. and A. Landier (2008) Why has CEO Pay Increased So Much?, Quarterly Journal of
Economics, 123: 49-100.
Gawande, A. (2009) The Cost Conundrum: What a Texas Town Can Teach Us About Health
Care, The New Yorker, June 1.
Guadalupe, M. (2007) Product Market Competition, Returns to Skill, and Wage Inequality,
Journal of Labor Economics, 25(3), 439-474.
Guerrieri, V., Shimer, R. and R. Wright (2010) Adverse Selection in Competitive Search Equilib-
rium, Econometrica, 78, 1823-1862.
Hermalin, B. (2005) Trends in Corporate Governance, The Journal of Finance, 60(5), 23512384.
Holmstrm B. (1982) Moral Hazard in Teams, The Bell Journal of Economics, 13( 2), 324-340.
Holmstrm B., and P. Milgrom (1991) Multi-Task Principal-Agent Analyzes: Incentive Contracts,
Asset Ownership, and Job Design, Journal of Law, Economics and Organization, 7, 24-52.
Inderst, R. and M. Ottaviani (2012) Competition through Commissions and Kickbacks, American
Economic Review, 102(2), 780-809.
Itoh, H. (1991) Incentives to Help in Multi-Agent Situations, Econometrica, 59, 611-636.
Kaplan, S. and J. Rauh (2010) Wall Street and Main Street: What Contributes to the Rise in the
Highest Incomes?, Review of Financial Studies, 23, 1004-1050.
Karuna, C. (2007) Industry Product Market Competition and Managerial Incentives, Journal of
Accounting and Economics, 43, 275-297.
Kosfeld, M. and F. von Siemens (2011) Competition, Cooperation and Corporate Culture, Rand
Journal of Economics, 42, 23-43.
Labaton Sucharow (2012) Wall Street, Fleet Street and Main Street: Corporate Integrity at a
Crossroads. Web link.
Labaton Sucharow (2013) Wall Street in Crisis: A Perfect Storm Looming. U.S. Financial Services
Industry Survey, July 2013. Web link.
Laont, J.J., and J. Tirole (1986), Using Cost Observation to Regulate Firms, Journal of Political
Economy, 94, 614-641.
Lazear, E. (2000) Performance Pay and Productivity, American Economic Review, 90(5), 1346-
1361.
Lemieux, T., MacLeod, B. and D. Parent (2009) Performance Pay and Wage Inequality, Quarterly
Journal of Economics, 104 (1), 1-49.
Lo, D. Ghosh, M. and F. Lafontaine (2011) The Incentive and Selection Roles of Sales Force
Compensation Contracts, Journal of Marketing Research, 48, 781-798.
Marin, D. and T. Verdier (2009) Globalization and the Empowerment of Talent, Journal of
International Economics, 86, 209-223.
Maskin, E., and J. Tirole (1992), The Principal-Agent Relationship with an Informed Principal,
II: Common Values, Econometrica, 60(1), 1-42.
55
Moen, E. and Rosen, A. (2005) Performance Pay and Adverse Selection, Scandinavian Journal
of Economics 107(2), 279298.
Netzer, N. and F. Scheuer (2010) Competitive Markets without Commitment, Journal of Political
Economy, 118,(6), 1079-1109
Philippon, T. and A. Reshef (2012) Wages and Human Capital in the U.S. Financial Industry:
1909-2006, Quarterly Journal of Economics, 127: 1551-1609.
Piketty, T., Saez, E. and S. Stancheva (2014) Optimal Taxation of Top Labor Incomes: A Tale of
Three Elasticities, American Economic Journal: Economic Policy, 6(1): 230-271.
Prendergast, C. (1992) Career Development and Specic Human Capital Collection, Journal of
the Japanese and International Economies, 6, 207-227.
Prendergast, C. (2007) The Motivation and Bias of Bureaucrats, American Economic Review,
97(1), 180196.
Raith, M. (2003) Competition, Risk, and Managerial Incentives, American Economic Review,
93(4), 14251436.
Ramalingam, A. and M. Rauh (2010) The Firm as a Socialization Device, Management Science,
56(12), 2191-2206.
Rochet, J.C. and L. Stole (2002) Nonlinear Pricing with Random Participation, Review of Eco-
nomic Studies, 69, 1, 277-311.
Rosen, S. (1981) The Economics of Superstars, American Economic Review, 71: 845-858.
Rothschild, M., and J. Stiglitz (1976) Equilibrium in Competitive Insurance Markets: An Essay
on the Economics of Imperfect Information, Quarterly Journal of Economics, 90(4), 629649.
Salz Review (2013) An Independent Review of Barclays Business Practices. Web link.
Stantcheva, S. (2014) Optimal Income Taxation with Adverse Selection in the Labor Market,
Review of Economic Studies, forthcoming.
Schmidt, K. (1997) Managerial Incentives and Product Market Competition, Review of Economic
Studies, 64(2), 191-213.
Shearer, B. (2004) Piece Rates, Fixed Wages and Incentives: Evidence from a Field Experiment,
Review of Economic Studies, 71, 513-534.
Steen, G. (2010) Retractions in the Scientic Literature: Is the Incidence of Research Fraud
Increasing?, Journal of Medical Ethics, doi:10.1136/jme.2010.04092
Stewart, J. (1994) The Welfare Implications of Moral Hazard and Adverse Selection in Competitive
Insurance Markets, Economic Inquiry, 32(2), 193-208.
Thanassoulis, J. (2013) Industry Structure, Executive Pay and Short-Termism, Management
Science, 59, 402-419.
Tymula, A. (2012) Competitive Screening of Heterogeneous Labor Force and Corporate Teamwork
Attitude, NYU mimeo, April.
Van Norden, R. (2011) The Trouble with Retractions, Nature, 478(6), October, 26-28.
Veiga, A. and G. Weyl (2012) Multidimensional Product Design, University of Chicago mimeo.
Villas-Boas, M. and U. Schmidt-Mohr (1999) Oligopoly with Asymmetric Information: Dieren-
tiation in Credit Markets, The RAND Journal of Economics, 30(3), 375-396.
56