Journal of Monetary Economics 30 (1992) 439-465.

North-Holland

The liquidity premium in average

interest rates*

Wilbur John Coleman II

Duke Universir!. Durham. ,L’C .?7706. C’SA

Christian Gilles and Pamela Labadie

Board o/Governors, Federal Reserve SJsrem, Washingron, DC 20551. 6.24

Received May 1992, final version received August 1992

We extend recent models of liquidity to study how a systematic relationship between monetary
shocks and output affects the average real short-term interest rate.

1. Introduction

In this paper, we study the determination of interest rates in a model in which

money is positively correlated with output and negatively correlated with
interest rates. Such a model might account for the empirical regularities
documented by Cagan (1966), Cagan and Gandolfi (1969), and others, and
recently reexamined by several authors, including Christian0 and Eichenbaum
(1991) and Leeper and Gordon (1992). We build on earlier work by Lucas (1990)
and Fuerst (1992) - who were themselves building on the work of Grossman and
Weiss (1983) and Rotemberg (1984). In this type of model, a liquidity effect of
money on interest rates stems from imposing separate cash-in-advance con-
straints in the goods and asset markets and delaying the flow of money between
these markets. Thus, this flow of money cannot immediately respond to a variety
of shocks in each market, and, as a result, one-period interest rates respond
negatively to open-market purchases of securities,
In this setting, the decision to allocate money between the goods and asset
markets is made before interest rates are known. Thus, with a predetermined

Correspondence IO: Pamela Labadie, Mail Stop 70, Board of Governors, Federal Reserve System,
Washington, DC 20551, USA.
*This paper represents the views of the authors and should not be interpreted as reflecting those of
the Board of Governors of the Federal Reserve System or other members of its staff.

0304-3932/92/505.00 % 1992-Elsevier Science Publishers B.V. All rights reserved

450 W. Coleman er al.. The liquidir! premium in merage interest rates

demand for bonds in the asset market. bond prices respond negatively to the size
of the bond issue - a non-Fisherian relation called the liquidity effect. When
open-market operations are correlated with output. the liquidity effect opens up
the possibility of a systematic premium in one-period interest rates - which we
call the Iiquidir~ premium - that is due to the uncertain return on money
allocated to the asset market. In developing his model Lucas focused on a simple
stochastic structure in which output was held constant, and consequently he
found no such premium. By inducing a negative correlation between output and
interest rates through monetary shocks, we find that short-term interest rates
are systematically lower than forward interest rates. This model may help
explain why short-term interest rates are low relative to the return on the
other assets, a result that has been difficult to explain with models that have
Arrow-Debreu markets.
We start by describing the model for a rather general monetary policy
conducted through open-market operations in which bonds of various maturi-
ties are traded for money. Then, we prove a Modigliani-Miller-Ricardian
Equivalence Theorem to justify focusing the remainder of the paper on a mone-
tary policy that calls for issuing only one-period bonds and lump-sum monetary
transfers. We then discuss how we solve this model. In the final section we report
how, for various values of the parameters, average interest rates differ from their
Fisherian fundamentals.

2. The model
The model we study is that of Lucas (1990) with two changes: The endow-
ment varies stochastically over time, and, within a period, information
(but not money) travels from the asset to the goods market. Households are
identical and have preferences over stochastic consumption streams {c,} that are
given by

The discount factor satisfies 0 c p < 1, and the single-period utility function u is
continuously differentiable, strictly increasing, and strictly concave. A represent-
ative household begins period t with a portfolio of money (I?,) and discount
bonds of maturity r (A,,) for r 2 1. There are two markets, a goods market and
an asset market, and each household consists of three members who have
particular roles in these markets. One member, the seller, receives an endow-
ment 5, of perishable goods, which he sells in the goods market in exchange for
money at the price P,. We assume that no household can consume any part of its
own endowment. Another member, the buyer, takes a portion 2, of the house-
hold’s money balances to the goods market and purchases consumption goods
 W. Cokvnan er al.. The liyuidiry premium in arerage inreresr rates 151

from other households. The third member, the portfolio manager, takes the
remaining money H, - Z, and all the bonds to the asset market and rebalances
the portfolio, purchasing W,, new r-period bonds from the government at price
qtr. Holdings of r-period bonds at time t are redeemed for money at the
beginning of the (t + r)th period. At the end of the period, the household
consumes its purchased goods and consolidates its assets; at the beginning of the
next period, it receives a lump-sum monetary transfer of X,, 1. The size of the
endowment is known at the beginning of the period. but new issues of govern-
ment securities and the prices of goods and bonds are revealed only after the
household divides its money between the goods and asset markets. There is no
transfer of assets between the goods and asset markets within a period, but
information flows freely, so that the price of goods may depend on the new issue
of government securities.
The government begins each period with outstanding stocks of money, M,,
and r-period bonds, B,,, for r 2 1. In the asset market the government issues an
additional supply, D,,, of r-period bonds at prices qtr. At the beginning of the
next period, the government issues a lump-sum monetary transfer of X,, I. The
new issue of bonds relative to the supply of money is denoted d,, = D,,/M,,
and the lump-sum monetary transfer relative to the supply of money is
denoted ,Y,+, = X,, ,/M,. Assume that (<,, d,_ 1,.u,) follows a first-order, finite-
state Markov process with transition matrix Il in which < is always strictly
positive.
Under the assumptions of a Markov process and time-separable utility, it
makes sense to seek an equilibrium in which all nominal variables scaled by the
stock of money are stationary. We follow the conventions of (1) denoting with a
corresponding lower-case letter the resealed variables, (2) denoting without
a prime variables that are known at the beginning of a period, and (3) denoting
with a prime variables that are first known later in the period, or at the
beginning of the next period. With these conventions, information at the
beginning of a period is summarized by

s = (5, d .y,b),

where < is this period’s endowment of goods, d = (d,) is the additional supply of
bonds relative to the stock of money that was issued in last period’s asset
market, x is the beginning of period’s monetary transfer relative to last period’s
stock of money, and b = (6,) is the outstanding stocks of government debt of
various maturities relative to the current stock of money. Since participants in
both markets know the current open-market operation, we seek an equilibrium
in which the vector of bond prices q = (qr) and the (resealed) price of goods p can
be represented as q(s, d’) and p(s, d’), respectively.
The evolution of the exogenous state variables (<, d, x) is given by a transition
matrix l7. If we denote the next period’s stock of money relative to this period’s
452 W. Coleman et al., The liquidity premium in acerage inrererr rates

by g’, the stocks of money and bonds evolve according to

g’= 1 -q(s,d’).d’+bl +d; +x’, (1)

b: = (brcl + di,,)lg’ for T 2 1, (2)

where

4(%d’).d’ = c qr(s, d’)di.

Denote the representative household’s initial stocks of money and bonds
relative to the aggregate stock of money by h and the vector a, respectively. At
the beginning of the period, the household allocates a portion, z, of its initial
money to the goods market and the rest, h - z, to the asset market. Money
brought into the goods market constrains purchases: p(s, d’)c I z. Purchases of
bonds, We, in the asset market are also subject to a finance constraint:
q* w < Jo- z; clearly, households hold excess cash in the asset market only if
one-period interest rates are iero. Given the price functions, p and q, the
transition matrix, II, and the laws of motion (lH2) for (g’, b’), the household’s
problem can be solved via the dynamic programming problem

c(h, a, s) = max
: c.
w max {Erd,[u(c) + &$I’, a’, s’)] ) ,
11
subject to

z 2 p(s, d’)c,

h - z 2 q(s, d’). w,

h’ = (h - pc - q’w + a, + WI + p< -I- x’)/g’,

4 = (a,,, + bv,*r)/g’ for T 2 1.

In the objective function, E, denotes expectation conditional on information

available at the beginning of the period (that is, knowing the state s); similarly,
Es,,. stands for the expectation given the information (s,d’) available when
trading occurs. For given functions p and q, the proof of the existence of
a solution to this problem is standard [see, for example, Stokey et al. (1989)].
 W. Coleman eI al.. The liquidity premium in average interest mes 353

An equilibrium consists of goods and bond prices, p(s, d’) and q(s, d’), a cash
allocation function !(/I, a,~), a consumption function c(h, a, s, d’), and bond
purchase function w(h, Q, s, d’) such that: (i) p(s, d’) I=-0 and 0 c q(s, d’) < 1,
(ii) 2, c, and w are the optimal policy functions for the dynamic program-
ming problem of the household facing prices p(s, d’) and q(s, d’), and
(iii) at the equilibrium values h = 1 and a = b, these policy functions satisfy
a feasibility condition, 0 < 2( 1, b, s) cz 1, and two market-clearing conditions,
~(1, b, s, d’) = 5 and ~(1, b, s, d’) = d’.
The first-order condition corresponding to the optimal choice of )v,, for T 2 1,
is

The function u, is the partial derivative of c with respect to its rth agument [so
that r1 is the partial derivative with respect to money holdings and c’~,for T > 1,
is the partial derivative with respect to holdings of (T - I)-period bonds]. The
function 2 is the Kuhn-Tucker multiplier corresponding to the asset market
finance constraint, which satisfies

q(s, d’). w I h - I(h, a, s) with equality if :(h, a, s, d’) > 0. (4)

The first-order condition for the optimal choice of consumption c is

(5)

The function u1 is the marginal utility of consumption, and the function ji is the
Kuhn-Tucker multiplier corresponding to the finance constraint in the goods
market, which satisfies

P(S, d’)c I f(h, a, s) with equality if fi(h, a, s, d’) > 0. (6)

The first-order condition for the optimal cash allocation function 2 is

E,[fi(h, a, s, d’)] = E,[;(h, a, s, d’)]. (7)

Finally, the envelope conditions with respect to holdings of money and bonds
are

~1 W, a, s, d’))
ol(h, 0,s) = Es (8)
P(S, 0 1
454 W’. Colemun et al., The liyuidit,v premium in acerage interest rues

(9)

Eqs. (3) through (9) can be used to solve for the optimal policy functions and
multipliers, given the functions p and q. To find the equilibrium, we impose the
market-clearing conditions II = I, a = b, ~(1, b, s, d’) = <, and w,(l, b, s. d’) = d;,
and try to solve for prices. To simplify the notation, define z(s) = t(l, b, s),
i.(s, d’) = i(l, b, s, d’), p(s, d’) = fi(1, b, s, d’), w(s, d’) = u,(<)/p(s, d’), and
~Js) = c,(l, b, s). With these definitions, the envelope conditions (8) and (9)
evaluated at the equilibrium, become

41(s) = E,C4s, d’)] (10)

and

c&(s) = /Es y for r>l, (11)

[ 1

while the first-order condition (3) becomes

ilL[~] = qAs.d’)(i(s,d’) + bE,,,[y]). (12)

Multiplying this equation by d:, summing over r, and using the Kuhn-Tucker
condition (4), the value of E. can be expressed as

(13)

A corresponding expression for n can be derived from (5) and (6),

p(s, d’) = max 0, .. (14)

Substituting this equation for p into (5) evaluated at the equilibrium, we get

o(s, d’) = max %$!, PEsd. [?I}, (15)

 W. Coleman rr al.. The liquidity premium in acerage interest races 455

where we used Esd.[41(s’)/g’] = Esd,[~(s’, d”)/g’], a direct implication of (IO).

The above expressions for i. and p can be substituted into (7) and rearranged to
obtain

E,[o(s,d’)] = E,[max{DESd,[y]. E5,,,[g~,(s’J:~))]~]. (16)

Eqs. (10) to (16), along with (1) and (2), determine the equilibrium functions o,
:, 4, and 4, along with the equilibrium evolutions of g and b. The equilibrium
goods price is then p(s, d’) = n,(<)/w(s, d’).

3. A Modigliani-Miller-Ricardian equivalence theorem

As Lucas (1990) has noted, this is a Modigliani-Miller-Ricardian world

despite the separation of the goods and asset markets, and thus we should expect
many monetary policies, as reflected in the behavior of open-market operations
and transfers, to result in the same equilibrium. In particular, the only constraint
households face in the asset market is that the total value of their net bond
purchases cannot exceed their available money holdings, and so vve should
expect the composition by maturity of the bonds the government issues to be
largely irrelevant, In this section we prove a version of the ;IlodigIiani-
Miller-Ricardian Equivalence Theorem for this economy, and we use this
theorem to justify focusing the remainder of this paper on the case in uhich the
government issues only one-period bonds.
Consider, then, equilibrium p and 4 corresponding to a particular monetary
policy. Conduct the following experiment: Perturb the monetary policy while
maintaining the same inflows and outflows of funds to the government (at the
original prices). This experiment is meant to correspond to the standard applica-
tion of the Modigliani-Miller-Ricardian Equivalence Theorem in which gov-
ernment policy is changed, but expenditures, taxes, and seignorage are kept
constant. In our model there is no fiscal policy, and all flows between the
government and the private sector are money flows. In period r, funds flow to
the government in the asset market in an amount equal to qr.d,. At the
beginning of the following period, funds flow to the private sector in an amount
equal to b,, + d,, + xl+ 1. Note that maintaining the same flows fixes the rate of
growth of money.
Now select any other monetary policy (.i, d^)under the restriction that gI = gr
and ql. (d, - d,) = 0, for any possible realization (x, d) under the original policy.
We want to check whether the correspondin g realizations P, and qr are also
equilibrium paths under the perturbed monetary policy. A word of caution is in
order. With the new monetary policy replacing the original one, there is no
guarantee that the recursive structure of the equilibrium can be preserved under
456 u’. Coleman rr ~1.) The hyuidity premium in acrrage interest rates

the new policy. But the only reason households care about the state and the size
of the new bond issues is to forecast future prices. Since the distribution of prices
is completely unchanged, households can make the same forecasts in the new
economy. To prove that the original realizations for pr and qt remain equilib-
rium paths under the new policy, we check that with these prices the same
Kuhn-Tucker multipliers L and ii and the same functions z, CD,and $, continue
to satisfy all the equations determining the equilibrium. In eqs. (10) through (16),
since s’ only enters through g’, and d’ only enters through Erd.[4(s’). d’], we
need to show only that E,,.[$(s’).(d^’ - d’)/g’] = 0 to prove the desired invari-
ance property. After multiplying (12) by 6: and summing over r, we see that the
hypothesis q(s, d’).(n^’ - d’) = 0 establishes this result, which completes the
proof.
Perhaps an example would help to clarify the situation. Suppose that the
government wishes to replace quantity n of one-period bonds, sold at a unit
price of q1 in the financial market, with two-period bonds, sold at q2 each. For
the swap to be a zero-value transaction, the new issue of two-period bonds must
equal (ql/q2)d. If this were the only change in the policy, then the equilibrium
would be affected because the path of the money stock would change. Under the
new policy, the money stock is reduced by d the following period (when the
one-period bonds would have matured) and is increased by (ql/q2)d in the
subsequent period (when the two-period bonds mature). To prevent this change
in the path of the money stock, the government can increase transfers by d next
period and reduce them by (q1/q2)d in the subsequent period. Then the new
policy is equivalent to the old one. Note that the present value of the change in
transfers is zero at the time the policy is changed. Note also that the new policy
induces no change in a households wealth (bonds plus transfers) at any time,
regardless of how asset prices evolve.
One interpretation of this theorem is that the government has only two
instruments to affect the economy: the value of open-market operations and the
growth rate of the money supply. Any policy shift that does not alter these two
quantities is irrelevant. In particular, bond swaps intended to take advantage of
an upward-sloping yield curve will not change this yield curve, so long as the
effect of the swap on the rate of growth of the money stock is sterilized (for
example, through transfers). Such a swap will in general change the value of
government debt, but only because this measure of debt ignores the transfers;
with proper accounting, total government debt (bonds plus money plus
present value of transfers) is unchanged at any time and in any state.
Clearly, policy shifts may affect the value of open-market operations or the
growth rate of money. Such shifts will in general change the equilibrium,
even when they are designed to keep the value of government debt (not
including money balances and transfers) unchanged at the original prices.
Indeed, Grilli and Roubini (1992) have noticed that such policy shifts disturb
the equilibrium.
4. Computing the ~~ili#~i~rn

For the remainder of this paper we focus on the special case in which the
government issues only one-period bonds. but where eq. (I?) permits us to
compute the eq~ilibri~m prices of bonds of arbitrary maturity. In the previous
section we proved that in general there is a Modigliani-Miller-Ricardian
equivalence between such an economy and one in which multi-period bonds are
ac~uaIly issued. Also, we assume that transfers are set to obtain particutar money
growth rates. so that CJreplaces .Yas an exogenous state variable, and ($ d, s)
evolves according to a Markov transition matrix n.
To ease the notational burden we will write d; as simply tf’. Finding the
equilibrium reduces to finding functions e$s. cl’} and z(s) that satisfy (15) and
(16), which become

and

To compute a solution (z, CO),we found it sufticient to impose the following

assumption.

In the appendix we prove that with Assumption 1 there exists one and
(essentiaIIy) only one solution fz*w) to eqs. f 17) and (18), and that it satisfies
0 < z ==cI. Furthermore, we construct a sequence of functions [z,,, e~,j that
converges to this equilibrium.

5, The premium in one-period interest rates

When the government issues one-period bonds only, eqs, (!Z) and t 15) imply
that the one-period gross rate of interest is

Rjs, a*) = ----J--z

q*(6 a’)
maxjl,&j.
458 Ct. Colemun el trl.. The lup4idtry premium in acerage inreresr rules

As can be seen from this equation. and as discussed by Lucas, movements in the
one-period interest rate are not explained completely by the Fisherian funda-
mentals of the marginal rate of substitution and expected inflation. Extracting
information about the state of expectations from short rates of interest thus
requires an understanding of how interest rates deviate from their Fisherian
fundamentals. Since this deviation is the result of an excess supply or demand for
money in the asset market, in the sense that households would like to transfer
money between the goods and asset markets, Lucas referred to this non-
Fisherian factor as a liquidity effect. Understanding the liquidity effect on
interest rates is important when considering average values of short rates
because it may lead to a systematic, and even time-varying, premium to holding
short bonds. In this section, we study the determination of such a premium when
households perceive a systematic relationship between open-market operations
and subsequent output shocks.
If one-period bonds were traded at the beginning of the period, before
knowledge of d’. their price would be determined by the Fisherian rate of
interest,

EsCu,(Sh'~l EsCo1
F(s)
=/?E,[u,(;')/p'g'] =,OE,[w’/g’] ’
(19)

where CO’stands for ~(s’, d”) and p’ stands for p(s’, d”). Define the liquidity
premim in the average interest rate as

E,[Rb, d’)]
i’(s) =
F(s) ’
so that y(s) is the proportional deviation of the average short rate from funda-
mentals. If y(s) equals unity, then there is no premium.
To evaluate y(s), divide eq. (12) (for r = 1) by q1 and apply the operator E, to
both sides. Eq. (7) implies that E,[i.] can be replaced by E,[p]. Use the
first-order condition (5) in the resulting equation, which can then be manipu-
lated to obtain

Substituting the value of F(s) into the definition of s(s), using the last equation,
and rearranging, we get

Pw(s’,d”)/g’
R(s,d’), E,Go(S, d,), .

1
Note that if d’ is known at the beginning of a period, a situation that
corresponds to the standard cash-in-advance model with integrated markets,
 W’. Colemun er al., The liquidirx premium in ctrerage inreresr rares 459

then the liquidity premium ;.(s) is always one.’ As can be seen from (ZO), in
general this premium depends on the correlation between the liquidity shocks to
interest rates and shocks to the marginal value of money in the next goods
market. Such a correlation is the result of a correlation between the rate of
interest and either output or the price level: in the next section we simulate this
model to study the effect of an interest-output correlation on the premium.
Consider the following two examples to help clarify this dependence.

E.uanrple 1. Suppose the vector (<‘, ti’) is iid (although 5’ may not be indepen-
dent of d’). Suppose also that lump-sum transfers keep the stock of money
constant and that households exhibit constant relative risk aversion utility,
Ui(C) = c-u, d > 0. In the appendix we prove that w(<, if’) is a nonincreasing
function of < if G > 1, independent of 5 if c = 1. and a nondecreasing function of
< if 0 < 1. If we assume that a high d’, which leads to high interest rates, is also
associated with low output next period (so that 5’ and ~1’ are negatively
correlated, although t and d’ are independent). then for CJ> 1 it follows that Ii’
and o(i’, d”) are positively correlated. From the expression (70), it follows that
the liquidity premium y(s) is less than unity for cr > 1. Short-term interest rates
are then systematically lower than their Fisherian counterparts. Note that for
log utility (a = 1) there is no liquidity premium in short-term interest rates, and
for less risk-averse households (a < 1) the liquidity premium is greater than
unity.
Fundamentally, the liquidity premium is an insurance premium that is driven
by the correlation between the nominal return on a bond and the future
marginal value of money [refer to eq. (20) and recall that w(s, d’) =
u,(<)/p(s, d’)]. When B > 1, this correlation is positive, giving rise to a liquidity
premium of less than unity. Note that the correlation is also positive when goods
prices are constant. When u < 1, the correlation is negative due to how goods
prices respond to output and open-market operations, a behavior that leads to
a liquidity premium greater than unity.

In Example 1 the fact that the letlel of output is stationary complicates the
analysis because movements in the cash allocation function 3 induce a correla-
tion between prices and output. Although not strictly covered by Theorem 1 in
the appendix, consider the case when open-market operations n’ and the growth
rate of output p’ = <‘/< follow a finite-state Markov process. With households
exhibiting constant relative risk aversion, ~r (c) = C-O, eqs. (17) and (18) become

tj(s, d’) = maxi-&, BE,,~[(p)L-~(s” “)]I

‘Note that the liquidity premium is not only a risk premium. as it may be different from one when
utility is linear or output is constant.
 W. Coleman er al.. The l&dir! premium in arerage inrrresr rafes

and

E,[rj/(s, d’)] = E,[max { 1, &I fiE,,,pP’)l-y(r” ““)I],

where $(s, d’) stands for I/(<&, d’)).

Example 2. Suppose the vector (p’, d’) is iid (although p’ may not be indepen-
dent of d’) and the transfer policy is used to achieve a constant money growth
rate Q. The solution to the above system of equations is then z(s) = Z (for some
constant Z) and II/(s, d’) = max ( l/f, zEd. [( p’)’ -“I ), where z is a constant equal
to fiEsdPIII/(s’,d”)]/d (note that $ does not depend on s). The liquidity premium
can be written as

note that the function II/ disappears from this equation. Hence, if we assume that
a high d’, which leads to a high rate of interest, is associated with a low output
growth p’, this premium is less than unity when CJ> 1.

Despite a possible correlation between open-market operations and sub-

sequent output, the result of Lucas that forward rates are unaffected by the
liquidity shocks still obtains. To see this for the one-period forward rate one
period ahead,f(s, d’) = ql(s, d’),iqz(s, d’), note that

EstrcW’/G”l
fh 0 = BEsd,Co,,,g,g<,, .

The form of this equation is the same whether or not d’ is known at the
beginning of the period; in other words, f(s, d’) is equal to its Fisherian
counterpart, and thus not subject to a liquidity premium. This is because the
forward price l/Jis the price set in the current asset market for the purchase of
a one-period bond in the next asset market. Money to purchase this bond is
withdrawn from the next goods market, and the bond pays off in money that is
available in the following goods market. This transaction thus amounts to
a transfer of money from the next goods market to the following one; by
construction (the economy is fully integrated at the beginning of each period)
such a transaction is unaffected by liquidity shocks. Therefore, the presence of
a correlation between monetary shocks and output shocks, which directly
affects average short rates but not forward rates. may alter the average shape of
the yield curve.
6. Numerical illustrations

To get a sense of the quantitative importance of the liquidity premium in this

model, we study explicit solutions that are computed along the lines suggested
by Theorem I. In selecting the parameters of the model, we choose a period
length that is short enough to reflect lags in the adjustment of portfolios in
response to open-market operations, and yet long enough to capture some
perceived relationship between money and output. We choose p = 0.9999,
which with a quarterly frequency corresponds to an annual rate of time prefer-
ence of0.04 percent, and we choose a utility function exhibiting constant relative
risk aversion, u,(c) = cmb. Throughout this exercise, we model d as following
a two-state Markov process with states (ofL, cfz) = (0.050. 0.0.52), and transition
matrix

e l-8
i7d =
( l-0 8 >

We choose cl1 and d2 so that d2/d, = 1.04, a value that corresponds to an

interest rate differential of four percentage points between the two states if
interest rates are always positive. We also choose lump-sum monetary transfers
so that money grows at the constant gross rate 1.005.
We solve three versions of the model. In all versions open-market sales of
securities and subsequent output shocks are perfectly negatively correlated. In
the first version we assume that the level of output takes on two values,
(cl, c2) = (1.0, 1.05), so that there are only the two states ((I,, rz) and (d2, <,). In
the second version we assume that the growth rate of output takes on two
values, (pl, pz) = (0.98, 1.03), so that the two states are (d,, pz) and (d2, pl). In
the third version, which we use as a benchmark, we use the same stochastic
process as in the second version except that we allow households to know the
value of d’ at the beginning of the period. Since markets are integrated, interest
rates are then determined exclusively by Fisher effects. Each of these models is
solved for three values of the parameters (e, c), and the results are reported in
table 1.’ The numbers in table 1 reflect expectations with respect to the ergodic
distribution of the state variables, and are expressed at annual rates assuming
that the period is a quarter. The variable F refers to the average one-period
Fisher nominal interest rate F(s) given by eq. (19), R refers to the model’s
one-period nominal interest rate, andfrefers to the forward rate.3

‘We also solved the model for the case Lucas considered in which information does not Row
across markets within a period, that is. goods prices do not immediately respond to the open-market
operation. Except for Model 2 when 0 = 0.75, interest rates are virtually identical in both versions of
the model. When 0 = 0.75 with Lucas’s Row of information, R = 5.21 percent, F = 5.62 percent. and
J’= 5.62 percent.
‘In the text all rates were expressed as gross rates per period; in table 1 rates are expressed as net
rates in percent per year.
 462 IV. Colrmun cr (11.. The liquidir! pwmrum in awruge inrerrsi rates

Table 1
Money, output. and the liquidity premium.”

Model 1 Model 2 Model 3

s = (d,i) s = (d.P) s = (d.p)

U = 0.5. c = 1

;’ 0.9999 0.9955 1.0000

R 2.05 6.77 7.31
2.06 6.75 7.3 I
r’ 2.06 6.15 6.75

0 = 0.5, G = 0.3
.. 10001 1.0001 1.0000
k 2.07 0.57 0.57
2.06 0.56 0.57
/F 2.06 0.56 0.56

U = 0.75 G = -t
., 0.9999 0.9950 1.0000
k 2.03 -k-t8 5.62
2.05 5.01 5.62
f 2.05 5.01 5.04

“In Model I, the level of output < is stationary and perfectly negatively correlated with the
open-market operation d; in Model 2 and Model 3. the growth rate of output p is stationary and
perfectly negatively correlated with ti; in Model 3. markets are integrated.
Definirions oj” carinbles:

Q = diagonal elements in the 2 x 2 symmetric transition matrix for d;

CT = coefficient of relative risk aversion:
.. = liquidity premium;
k = one-period nominal rate of interest:
F = Fisherian one-period nominal rate of interest:
f = one-period forward rate one period ahead.

All values are expectations taken with respect to the ergodic distribution. All rates of interest are
expressed in percent per year.

In general we find an insignificant liquidity premium in Model 1, in which the

level of output and open-market operations evolve according to the Markov
matrix determined by 8. In Model 2, where the growth rate (rather than the
level) of output and open-market operations evolve according to the same
Markov matrix, however, we find that the liquidity premium begins to be
quantitatively significant. In the iid case (6 = 0.5) when 0 = 4, the nominal
one-period interest rate is on average 48 basis points below the corresponding
Fisher nominal interest rate (6.27 versus 6.75), and 104 basis points below the
nominal interest rate in Model 3 where d’ is known (6.27 versus 7.31). For log
utility, the liquidity premium vanishes, and lowering risk aversion even further
(a = 0.25) leads to an insignificant premium (one basis point higher). Keeping
(r = 4, for serially correlated shocks (0 = 0.75x the liquidity premium, at minus
53 basis points, remains significant; the interest rate is now on average 114 basis
points below the one-period interest rate determined in Model 3. In Model 3, of
course, all interest rates are determined by Fisherian factors and there is no
liquidity premium C;:(s) = I]. A comparison of Model 2 with Model 3 shows,
however, that the presence of a liquidity effect can lower the whole yield curve
and not only the short rate relative to the long rate.

7. Concluding remarks

The essential point of this exercise is to show that, in an economy constructed

to display a liquidity effect, monetary shocks that are correlated with output
shocks induce a premium, positive or negative, in short-term interest rates
relative to long-term interest rates. The source of this correlation. or the
direction of causation, is irrelevant in determining the premium; what is impor-
tant is the size of the correlation. In particular, monetary shocks could either
lead to output shocks or reflect an immediate response by monetary authorities
to such shocks. It is widely accepted that there exists a positive money-output
correlation due to one or both of these factors [see, for example, Sims (1972).
among many others]; and it is also true that short-term interest rates are on
average lower than long-term interest rates (from 1953 to 1991 the yield on
three-month Treasury bills averaged 5.66 percent per year while the yield on
ten-year Treasury bonds averaged 6.87 percent). This paper explored the possi-
bility of a link between the positive money-output correlation and the slope of
the yield curve.

Appendix

Theorem 1. Under Assumption 1 there exists one, and only one, solution pair
(0, z) to eqs. (17) and (18) satisfying 0 < z(s) < I and z(s) 2 1 - z(s), where z(s) is
the largest possible calue of d’, gicen s.

Proof: We first prove existence. Let Y be the set of functions { y : 0 < J(S) < 1
for all s>. For a fixed function ye Y replacing z in (17), define the operator T,, on
the space X of real-valued functions with domain the finite set ((s, d’)) so that
this equation reads o = T,.w. Endowing X with the sup norm and the usual
partial ordering, Assumption 1(i) guarantees the satisfaction of Blackwell’s
sufficient conditions for T, to be a contraction on X with modulus not exceeding
/I max {Esd. [l/g’] 1 -=z1 [see Stokey et al. (1989, p. 54)]. Denote by RJ the fixed
point of T,. and consider the following equation in the unknown XE(O, l), for
464

a fixed YE Y and a fixed state s:

= E,[max {I,+&]flE,,.[n,i~:d”]]. (21)

For .Yin the (perhaps empty) interval (0, 1 - z(s)] the right side of this equation
is constant at k(s) = /?E,[Ry(s’,tl”)/g’]; elsewhere in (0,l) it is strictly increasing
and tends to infinity as x goes to one. The left side is unbounded in any
neighborhood of zero and is strictly decreasing as long as its value exceeds k(s):
if its value reaches k(s) at some 3 < 1, it stays constant at k(s) for .YE [.U, I).
Therefore, there exists a unique solution 0 < X*(S) < 1 to (21) satisfying
.X*(S) 2 1 - z(s). Define a function Ay by AL’(~) = -U*(S). Then, A is a well-
defined operator on Y into itself; moreover, its range AY is contained in
Yn {y: 1 - (T(s) < J(S) < 1). By construction, a fixed point Z* = As* corre-
sponds to an equilibrium (w, Z) = (Qz*, -_*).
It can be verified that A is monotone in the usual partial ordering. Also. there
exists Z E Y such that A(5) 2 5. To see this, choose q such that max (<u,(:)) < ‘1
and 1 2 fiE,[max { 1, (I’/(1 - <u~(<)/v))] [by Assumption 1(ii), this can be
done]. Then, let Z(s) = <I~~(<)/P/ and note that R:(s, n’) = q for all s and tf’; using
this value in (21) and evaluating at .X = f(s), the left side exceeds the right side.
The monotonicity of each side now implies that AT 2 t, as was to be shown.
Therefore, A”(f), an increasing sequence of functions in Y, has a limit Z* that
belongs to Y and is a fixed point of A. Finally, z* E AY implies that
1 - J(s) I Z*(S) < 1 for all s.
We now prove uniqueness [following Krasnosel’skii and Zabrenko (1984.
theorem 46.1)]. Suppose that A possesses two fixed points, Z, and z2. Choose
0 < t < 1 such that :I 2 tz2 and zl(s) = CZ~(S)for some s (hence Z, > r:? does
not hold). Note that tR(ry) = Ry, from which it follows that A(ty) > ~A_vfor any
0 < c < 1 and y E Y. Then, z1 = Azl 2 A(tz2) > tAzz = tzZ, which is a contra-
diction.” H

Theorem 2. In Example 1, the equilibrium frtnction w(<, d’) is noniucreasirlg in

5 if o > 1, independent of i if CJ= 1, and nondecreasing in < if CJ-c 1.

“If the constant portions of both sides of eq. (21) overlap, it has an infinity of solutions. of which
A.v(s) is the largest. Therefore. there may be more than one equilibrium function :. but all such
functions support the same equilibrium. (w. p. q). The unique fixed point of A selects the supremum
of the functions : supporting this unique equilibrium.
 Cc’. Colemnn et ul.. The fiquidit!, premium in areroge inreresr rates 465

Proof: Since (5, d) is iid, the second argument in the max operator in (17) is
independent of <, so we need only establish that the first argument. <t-“/z(<),
has the properties stated for o. If c = I. then the equilibrium : is constant, which
establishes the result for this case. If ~7> 1, for any J E Y the solution .-l~(<) to
(11) and 2 ‘-“/AJ(<) are nonincreasing in 5, so this must also be true at the
equilibrium 1. A similar argument holds for CT< 1. n

References

Cagan, P.. 1966. Changes in the cyclical behavior of interest rates, Review of Economics and
Statistics -IS. 219-250.
Cagan. P. and A. Gandolfi. 1969, The lag in monetary policy as implied by the time pattern of
monetary effects on interest rates. American Economic Review Papers and Proceedings 59,
277-284.
Christiano. L.J. and M. Eichenbaum, 1991, Identification and the liquidity effect of a monetary
policy shock, Unpublished manuscript, Nov. (Federal Reserve Bank of Minneapolis. MN).
Fuerst, T.S.. 1992. Liquidity. loanable funds, and real activity, Journal of Monetary Economic 29.
3-24.
Grilli, V. and N. Roubini. 1992, Liquidity and exchange rates, Journal of International Economics
33, 339-352.
Grossman. S.J. and L. Weiss, 1983, A transaction-based model of the monetary transmission
mechanism. American Economic Review 73. 871-850.
Krasnosel’skit. M.A. and P.P. Zabretko. 1984. Geometrical methods of nonlinear analysis
(Springer-Verlag, Berlin).
Leeper. EM. and D.B. Gordon. 1992. In search of the liquidity eflect, Journal of Monetary
Economics 29. 34 l-369.
Lucas, R.E., Jr., 1990, Liquidity and interest rates, Journal of Economic Theory 50. 237-264.
Rotemberg, J.J.. 1984, A monetary equilibrium model with transaction costs, Journal of Political
Economy 92. W-58.
Sims. C.A., 1972, Money. income and causality. American Economic Review 62, 540-552.
Stokey, N.L. and R.E. Lucas. Jr.. with E.C. Prescott. l9S9. Recursive methods in economic dynamics
(Harvard University Press, Cambridge, MA).