Evaluating the Effects of Incomplete Markets on Risk Sharing and Asset Pricing

Author(s): John Heaton and Deborah J. Lucas

Source: Journal of Political Economy , Jun., 1996, Vol. 104, No. 3 (Jun., 1996), pp. 443-
487
Published by: The University of Chicago Press

Stable URL: https://www.jstor.org/stable/2138860

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend
access to Journal of Political Economy

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 Evaluating the Effects of Incomplete Markets
on Risk Sharing and Asset Pricing

John Heaton and Deborah J. Lucas

Northwestern University and National Bureau of Economic Research

We examine an economy in which agents cannot write contracts

contingent on future labor income. The agents face aggregate un-
certainty in the form of dividend and systematic labor income risk,
and also idiosyncratic labor income risk, which is calibrated using
the PSID. The agents trade in financial securities to buffer their
idiosyncratic income shocks, but the extent of trade is limited by
borrowing constraints, short-sales constraints, and transactions costs.
By simultaneously considering aggregate and idiosyncratic shocks,
we decompose the effect of transactions costs on the equity premium
into two components. The direct effect occurs because individuals
equate the net-of-cost margins. A second, indirect effect occurs be-
cause transactions costs result in individual consumption that more
closely tracks individual income. In the simulations we find that the
direct effect dominates and that the model can produce a sizable
equity premium only if transactions costs are large or the assumed
quantity of tradable assets is limited.

We thank George Constantinides, Darrell Duffie, Mark Gertler, Narayana Kocherla-

kota, Thomas Lemieux, Jose Scheinkman, Danny Quah, Jean-Luc Vila, and two anony-
mous referees for helpful comments and suggestions. We also thank participants at
the Canadian Macroeconomics Study Group, the NBER Asset Pricing meeting, the
Northwestern Summer Workshop in Economics, and seminar participants at Berkeley,
Cornell, Duke, the Federal Reserve, McGill, the Minneapolis Federal Reserve Bank,
Massachusetts Institute of Technology, New York University, Princeton, Queen's,
Rutgers, Stanford, University of California, Los Angeles, Western Ontario, and Whar-
ton. Part of this research was conducted at the Institute for Empirical Macroeconomics
in the summer of 1991. We would also like to thank Makoto Saito for research assis-
tance and the International Financial Services Research Center at MIT, the National
Science Foundation, and the Sloan Foundation (Heaton) for financial assistance.

[Journal of Political Economy, 1996, vol. 104, no. 3]

? 1996 by The University of Chicago. All rights reserved. 0022-3808/96/0403-0005$01.50

443

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
444 JOURNAL OF POLITICAL ECONOMY

I. Introduction

Incomplete markets in the form of an inability to borrow against risky

future income have been proposed as an explanation for the poor
predictive power of the standard consumption-based asset pricing
model.' With complete markets, individuals fully insure against idio-
syncratic income shocks, and individual consumption (under some
preference assumptions) is proportional to aggregate consumption.
With limited insurance markets, however, individual consumption
variability may exceed that of the aggregate, and the implied asset
prices may differ significantly from those predicted by a representa-
tive consumer model. In this paper we study an economy in which
agents cannot write contracts contingent on future labor income real-
izations. They face aggregate uncertainty in the form of dividend
and systematic labor income risk and also idiosyncratic labor income
risk. Idiosyncratic income shocks can be buffered by trading in fi-
nancial securities, but the extent of trade is limited by borrowing
constraints, short-sales constraints, and transactions costs.
The motivation for considering the interaction between trading
frictions and asset prices in this environment is best understood by
reviewing the findings of a number of recent papers. Telmer (1993)
and Lucas (1994) examine a similar model with transitory idiosyn-
cratic shocks and without trading costs. Surprisingly, they find that
even though agents cannot insure against idiosyncratic shocks, pre-
dicted asset prices are similar to those with complete markets.2 This
occurs because when idiosyncratic shocks are transitory, consumption
can be effectively smoothed by accumulating financial assets after
good shocks and selling assets after bad shocks. Aiyagari and Gertler
(1991) consider a related model with no aggregate uncertainty and
with transactions costs in which agents trade to offset transitory idio-
syncratic shocks. In this case, differential transactions costs affect the
relative returns on stocks and bonds and reduce the total volume of
trade. Finally, Constantinides and Duffie (1994) study the case of
permanent idiosyncratic shocks.3 Although agents have unrestricted
access to financial markets, no trades occur, resulting in increased
equilibrium consumption volatility. When the conditional variance of
idiosyncratic shocks is assumed to increase during economic down-

l For discussions of problems with the standard model, see, e.g., Hansen and Single-
ton (1983), Mehra and Prescott (1985), and Hansen and Jagannathan (1991).
2 Using a more volatile aggregate income process, Marcet and Singleton (1991) also
calibrate this model. They find that the equity premium rises in the presence of fre-
quently binding short-sales constraints.
3 Related two-period models can be found in Mankiw (1986), Scheinkman (1989),
and Weil (1992).

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 INCOMPLETE MARKETS 445

turns, the risk-free rate falls and the equity premium rises relative to
the complete markets case.
These results suggest that the quantitative asset price predictions
from this class of models will depend critically on several factors: (i)
the extent of trading frictions in securities markets, (ii) the size and
persistence of idiosyncratic shocks, and (iii) the correlation structure
of idiosyncratic and aggregate shocks. To address points ii and iii, we
develop an empirical model of individual income that captures both
the size of the idiosyncratic shocks and the persistence of these shocks
over time, on the basis of evidence from the Panel Study of Income
Dynamics (PSID). The time-series properties of aggregate income
and dividends are estimated using the National Income and Product
Accounts nipaA). This process is then used to calibrate the theoretical
model.
Our theoretical model differs substantively from those discussed
above by considering the effects of transactions costs in an environ-
ment with both aggregate and idiosyncratic shocks. Transactions costs
play an important role because agents choose to trade frequently
in order to buffer shocks to their individual income. As a result,
transactions costs can have two effects on asset prices.
First, (gross) rates of return on securities may be altered because
lenders require higher rates and borrowers require lower rates to
compensate for transactions costs. This direct effect of transactions
costs was emphasized by Amihud and Mendelson (1986), Aiyagari
and Gertler (1991), and Vayanos and Vila (1995).4
A second, indirect, effect of transactions costs is that they limit the
ability of agents to use asset markets to self-insure against transitory
shocks. Consequently, individual consumption is more volatile than
aggregate consumption. This affects each individual's attitude toward
aggregate uncertainty. Since preferences are assumed to be proper in
the sense discussed by Pratt and Zeckhauser (1987) and Weil (1992),
an increase in individual consumption volatility increases the amount
agents are willing to pay to avoid the aggregate uncertainty reflected
in dividends. In equilibrium, the implied equity premium could rise
in response to increases in transactions costs for this reason alone.
This paper appears to be the first to evaluate the importance of this
mechanism.

4 In contrast, Constantinides (1986) argued that transactions costs should have only
a small effect on asset returns. In his model, in which there is no idiosyncratic risk
and agents trade only to rebalance their portfolios, agents avoid most transactions costs
by reducing the frequency of trades. As a result, asset returns are not much affected
by the presence of transactions costs. However, because of the idiosyncratic shocks that
individuals face in our model, it is more costly for them to change their asset trading
patterns in response to trading costs.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
446 JOURNAL OF POLITICAL ECONOMY

We calibrate the model under a variety of assumptions about the

size and incidence of trading costs and the supply of tradable securi-
ties. When trading costs differ across markets, we find that agents
readily substitute toward transacting in the lower-cost market. For
example, if transactions costs are introduced only in the stock market,
then agents trade primarily in the bond market and by this means
effectively smooth transitory income shocks. In this case transactions
costs have little effect on required rates of return.
When transactions costs are also introduced in the bond market in
the form of a wedge between the borrowing and lending rate, then
the equilibrium lending rate falls. With a binding borrowing con-
straint or a large wedge between the borrowing and lending rate, a
transactions cost in the stock market can produce an equity premium
of about half of the observed value. In this case approximately 20
percent of the predicted equity premium can be attributed to the
indirect effect, so that the direct effect of transactions costs domi-
nates. These results, however, are quite sensitive to the structure of
transactions costs and the supply of tradable assets. For example, with
a sufficiently large outside supply of bonds, the impact of transactions
costs is reduced dramatically since the new bonds allow the agents to
avoid the borrowing constraint and borrowing costs.
The remainder of the paper is organized as follows: In Section II
we describe the model economy. In Section III we present the state
variables, the calibrated model of income and dividends, and the
parameterizations for trading costs, borrowing constraints, and short-
sales constraints. Simulation results are reported in Section IV, and
Section V presents concluding remarks.

II. The Model

A. The Environment

The economy contains two (classes of) agents who are distinguished
by their labor income realizations.5 At each time t, agent i receives
stochastic labor income Y'. By assumption, agents are not allowed to
write contracts contingent on future labor income. We shall refer to
the share of individual i's labor income in aggregate labor income as
idiosyncratic income because the innovations to the share of income
received by the first agent are perfectly negatively correlated with the
share of income received by the second agent. Notice that these

5 Scheinkman and Weiss (1986) consider one of the first two-agent equilibrium m
els in which idiosyncratic employment shocks are uninsurable.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 INCOMPLETE MARKETS 447

shocks would be completely diversified away in a complete market

setting.6
Agents also receive income from investments in stocks and bonds.
At time t, a share of stock, with price ps, provides a claim to a flow
of stochastic dividends from time t + 1 forward, {d2} =t+ I. The bond,
with price pt', provides a risk-free claim to one unit of consumption
at time t + 1. The agents trade these two securities to smooth their
consumption over time. Trading is costly, with transactions cost func-
tion K(b) in the stock market and w() in the bond market. Trade is
also limited by short-sales and borrowing constraints.
At time t, each agent's preferences over consumption are given by

U 0E1JZ i 1 ;()} y>, 1

U - E { at+T) | > ? (l

where 9;(t) is the time t information set that is common across agents.
This information is generated by the state, Zt, which is specified be-
low. In principle, -y could be allowed to differ across agents. Since we
want to interpret the two groups as similar except for realizations of
idiosyncratic shocks, however, it seems appropriate to equate y across
the two groups.7
At each date t, agent i maximizes (1) via the choice of consumption
ct. stock share holdings s'+ , and bond holdings b+ I subject to the
flow wealth constraint

c, + ptst+ 1 + ptbb+ 1 + K(s'+ 1' St; Zt) + w(b'+ 1' b'; Zt) (2)
'(ps + dt) + bt + Y(

and short-sales or borrowing constraints

sI 2 KSts t = U 1, 2, ... ., (3)

bt 2 K;', t = O. 1,2 (4)

The components of initial wealth dog s , bM

are taken as given.

6 The structure of our model differs from that of models with an infinite number
of agents who each receive independent shocks that sum to a constant by the law of
large numbers, such as in Bewley (1986), Clarida (1990), Aiyagari and Gertler (1991),
Huggett (1993), and Aiyagari (1994). For a comparison of these models and the model
considered here, see Heaton and Lucas (1995).
7Dumas (1989) considers the implications of different risk aversion parameters in
a complete markets setting.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
448 JOURNAL OF POLITICAL ECONOMY

B. Trading Frictions

The extent to which individuals will use asset markets to buffer idio-
syncratic income shocks depends on the size and incidence of trading
costs and the severity of borrowing and short-sales constraints. Since
the assumed form of these frictions qualitatively affects predicted
asset prices and since there is little agreement about the exact form
of these costs, we consider several alternative cost structures.

Transactions Costs in the Stock Market

In most of the simulations we assume that both buyers and sellers

face a quadratic transactions cost function:

K(S+ 1; Zt) = kt[(st? 1 i-s)ps]2. (5)

The parameter kt is used to control the assumed magnitude of th
cost. Because the realized cost is endogenous, the range of attainable
costs is bounded; an increase in the cost parameter eventually leads
to an offsetting reduction in trade. Dividing the cost function by
(Is"+I - s|I)ps gives the trading cost as a percentage of the value
shares traded: ktIs'+1 - s'Ips. The average of this percentage cos
reported in the simulation results.
We use a quadratic cost function primarily for computational sim-
plicity. However, it also captures the idea that as more assets are
sold, agents must sell increasingly illiquid assets. The fact that many
individuals hold no stock at all suggests that there may be significant
fixed costs to entering this market as well.8 To partially address this
issue, we also estimate the income process conditioning on data from
families that own nonnegligible amounts of stock.
Another possible objection to the quadratic form is that small
changes in stock holdings are likely to be as costly (proportionally) as
large changes. A quadratic cost function makes small transactions
relatively inexpensive, which encourages small transactions. How-
ever, in the discretized state space model below, agents trade every
period to smooth large shocks so that the strategy of using very small
transactions to avoid transactions costs is not relevant. As a robustness
check, we also report results for the case in which costs are propor-
tional to the size of the trade.

Bond Market Transactions Costs

The inside bonds in this model represent private borrowing and lend-
ing. While it seems sensible to treat transactions costs symmetrically

8 The effects of transactions costs of this form have been considered by Saito (1995).

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 INCOMPLETE MARKETS 449

for sales and purchases of stock or outside bonds, this is less true for
consumption loans. Typically consumers pay a substantial spread
over the lending rate to borrow. Although part of the observed
spread is a default premium that does not apply to the risk-free bonds
of the model, a portion of the spread can be attributed to costs of
financial intermediation or monitoring that must be incurred even if
the debt is ex post risk-free.
To capture the asymmetry between effective borrowing and lend-
ing rates, in some of the simulations the bond transactions cost func-
tion is assumed to have the form

(bt+ 1s b =; Zt) = fl min(O, b'+ Ipt)2 (6)

The parameter ft controls the magnitude of the spread between the
borrowing and lending rates. By convention, borrowing at time t is
represented by a negative value for bl+ l, so only the agent who bor
rows pays the transactions cost: As for stocks, the cost is reported as
a percentage of the per capita amount transacted: ftI bt+ 1 pUbl2.
We also consider the implications of a symmetric quadratic cost
function in the bond market of the form

w(bt+ 1, bt; Zt) = fQt(b+ pl )2l (7)

As demonstrated below, the choice of (6) versus (7) will have a signifi-
cant effect on the predicted equity premium.
Notice that with the one-period bonds of the model, the transac-
tions cost is paid every period even when a loan is rolled over for
several periods. The reason is that we interpret the transactions cost
not as a trading cost, but rather as a wedge between the borrowing
and lending rates due to monitoring and other costs incurred each
period. For comparison, in some of the simulations we assume that
the transactions cost in the bond market depends only on the change
in the amount of bonds outstanding. We also consider the case in
which the incurred cost is proportional to the amount borrowed.
To match the observed aggregate income and dividend process,
the economy is assumed to be stationary in the aggregate income
growth rate and in the dividend's share of income. As a result, the
price of the stock grows over time. The borrowing constraint, K b, is
assumed to be linear in aggregate income, ya, which accommodates
growth in the face value of bonds issued as the economy grows. The
parameters kt and ft are also chosen to induce stationarity. This re-
quires setting kt = kIYya and ft = flYya, where k and ft are constants.
Finally, we refer to the case in which K(v) 0 and w() 0 as the
frictionless model. The frictionless model is similar to the models in
Telmer (1993) and Lucas (1994) except that the income process is
calibrated using income data from the PSID.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
450 JOURNAL OF POLITICAL ECONOMY

Borrowing and Short-Sales Constraints

Consumption smoothing may also be curtailed by institutional limits

on the amount of borrowing. This type of credit rationing is repre-
sented by (4). Alternatively, these limits can be viewed as representing
the effects of very high transactions costs.
How to calibrate the upper bound on borrowing is not obvious.
For instance, the value of household collateral in the form of housing
and other assets is a plausible limit on debt but is not easily measured.
To bracket a plausible range, we consider an upper bound of 10
percent of average per capita income and a lower bound of 0 percent
(no borrowing). In the simulations the agents rarely hit the assumed
10 percent upper bound, so a less restrictive limit would have little
effect on the analysis. We also consider the effect of adding an outside
supply of bonds.
No short sales are permitted in the stock market: Ks = 0 in (3).
This is motivated in part by the observation that it is costly for individ
uals to take short positions, for instance, because of margin require-
ments. Allowing short sales in principle increases the effective quan-
tity of tradable assets, but in the simulations, this constraint rarely
binds.

C. Equilibrium

At time t aggregate output consists of the aggregate dividend, dt,

the sum of individuals' labor income, YtJ + Y'. Market clearing
quires

bl + b 2 = 0, t =0, 1, 2,..., (8)

sIl + St2 = 1, t =0, 1, 2, . . ,(9)

and

Z [rC + K(S'+ 1,st; Zt) + xo(bt+ 1, bt; Zt)]

i= 1,2(I0
=dt+Yt +Y, t=0,1,2 (10)

Notice that (8) implies that bonds are in zero net supply. In Section
IVF, we relax this assumption.
Let

St+ 1 St t)
K I (St+ I 9 St; Zt) =
dt+ I

dK(t+ 1,9 St; t)

K2 (S't+ I 9 S't; Zt)- ds

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 INCOMPLETE MARKETS 451

(bt~l, bt; Zt) _ )(b

(b'+~ ~ 8()(t I, bt; Z

(R)2(b+1 bt;Zt)- dt+

When the short-sales and borrowing constraints are not binding, the
first-order necessary conditions from the agent's optimization prob-
lem imply that, for all i and t,

[Ps + KI(S +1,st;Zt)]u'(ct)

= '(ct+1)[Pt+ + dt+1 - K2(st+2,St+ 1; Zt+1)]II;(t)}
and

[pb + 1(b+ 1, b'; Zt)] u'(ct)

= E{u'(c+01)[1 - w2(b'+2,bt+i;Zt+1)]I;(t)}. (12)

If an agent is constrained by the short-sales constraint (3), then
(11) is replaced by

S=Kt. (11')

Similarly, if the
then (12) is replaced by

bt = Kt. (12')

At time t the unknowns are Ps; pt; c', i = 1, 2; s', i = 1, 2; and

bt, i = 1, 2. The equations defining an equilibrium are the budget
constraints (2), i = 1, 2; the market-clearing conditions (8), (9), and
(10); and the asset pricing equations (11) or (11'), i = 1, 2, and (12)
or (12'), i = 1, 2. By Walras's law, one of the market-clearing condi-
tions or a budget constraint is redundant. We restrict our attention to
stationary equilibria in which the consumption growth rate, portfolio
rules, and equilibrium prices are functions of the time t state, Zt.

III. State Variables

The exogenous state is described by a Markov chain that gives the

dynamics of aggregate income, dividend income, and individual in-
come over time. The state, Zt, includes these exogenous variables and
the endogenous distribution of asset holdings.

A. The Exogenous State Variables

The exogenous state of the economy is divided into two sets of vari-
ables: the aggregate state of the economy and the idiosyncratic in-

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
452 JOURNAL OF POLITICAL ECONOMY

come state. To specify the

income and Da be aggregate dividend income at time t. Total aggre-
gate income is given by Y' Y1 + D?. Further, let - MY'IY I and
at-D/IY' be the (gross) growth rate in aggregate income and divi-
dend's share in aggregate income, respectively. The aggregate state of
the economy at time t is given by [ya aty].
Individual i's labor income as a fraction of aggregate labor income
is given by t-9= YtIY'. In our two-person economy, the law of mo
for al implies a law of motion for '2 since q1 + -2 = 1. The entire
exogenous state of the economy is given by [ya at 51]

Calibrating the Markov Chain

In calibrating the Markov chain, we require that it capture the mo-

ments of aggregate labor income, dividend income, and individual
income that are most likely to affect the predictions of the theoretical
model. The dynamics are first summarized by an autoregression and
then discretized using the method of Tauchen and Hussey (1991).
The moments of the aggregate economy are summarized by the
coefficients of a bivariate autoregression for Xa [log(-ya) log(8t)]':

Xa = Ra + AaXa? + OaFa, t = 1 2, 3 ... . (13)

where Fa is a vector of white-noise disturbances with covariance ma-
trix I, the matrix Oa is assumed to be lower triangular, and pa is a
vector of constants. The parameters of the vector autoregression
(VAR) are estimated using annual aggregate labor income and divi-
dend data from NIPA. We use aggregate labor income data rather
than aggregate consumption data because they more closely match
our individual-level labor income data. As shown in Section IVA, the
implications of this income process under complete markets are simi-
lar to those using consumption data, so this choice is unlikely to sub-
stantially affect the results. Appendix A describes the data in more
detail and reports the results of estimating the parameters of (13).
We also require that the Markov chain approximate the persistence
and volatility of individual income. The individual income process is
estimated from a sample of 860 households in the PSID that have
annual income from 1969 to 1984. For each household the income
process is summarized by a first-order autoregression of the form9

log(^') = `j + ploghtq1) + t. (14)

9 Appendix A describes the data in more detail and examines the results of estimat-
ing the model with several subsamples from the PSID and several alternative specifica-
tions of the income process.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 INCOMPLETE MARKETS 453

For the entire sample, the average value of p across households is

0.529 and the average standard deviation of E' is 0.251.
As the base case for investigation, we choose an eight-state Markov
chain for [yt bt nrl]' that closely approximates the estimated VAR'"
given by (13) and (14). The parameters of this Markov chain are
given in table 1. The Markov chain appears to do a reasonable job
of fitting the moments that we have chosen (App. B reports the ap-
proximation error, and table B1 summarizes the coefficients).
In the base case model, idiosyncratic labor income shocks are as-
sumed to be independent of the aggregate growth rate and dividend
realization. As discussed in Appendix A, we find little statistical evi-
dence of a significant effect of the aggregate state on the conditional
mean of individual income share. However, Mankiw (1986) and Con-
stantinides and Duffie (1994) show that if the distribution of labor
income widens in a downturn, then individuals will demand a large
equity premium to hold stocks. To examine this potential effect, we
also consider the cyclical distribution case (CDC) model.
The CDC model is calibrated by adding the following equation to
the system, which accommodates heteroskedasticity in the individual
shocks:

Std{e logQ ) = +to I logQYt) (15)

The parameters of (15) are again estimated using the PSID and ag-
gregate income data. For the subsample of households that own stock,
the point estimate of a I is - 1.064, which is consistent with the conjec-
ture that the conditional volatility of individual income is lower in
high-growth states. However, this value of al and the Markov chain
for aggregate income of table 1 imply a standard deviation of 0.28
for it in the low-income growth state and 0.22 in the high-income
growth state. This difference across aggregate growth states is too
small to produce a significant departure from the base case model
and hence has no detectable effect on predicted asset prices.
Since data limitations may be the reason for the small estimated
change in conditional variance over the cycle,"1 we calibrate the CDC
model under the assumption that the standard deviation of shocks to
individual income is twice as large in the low-aggregate growth state
as in the high-aggregate growth state. The parameters of the re-
sulting eight-state Markov chain, estimated as above, are reported in

10 As discussed below, we adjusted the mean of the process for the share of dividends
in income to be 15 percent.
" In particular, the short time series of individual data provide few observations of
cyclical variations.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 454 JOURNAL OF POLITICAL ECONOMY

TABLE 1

A. MARKOV CHAIN MODEL FOR EXOGENOUS STATE VARIABLES: BASE CASE

STATES
STATE
NUMBER ya 8

1 .9904 .1402 .3772

2 1.0470 .1437 .3772
3 .9904 .1561 .3772
4 1.0470 .1599 .3772
5 .9904 .1402 .6228
6 1.0470 .1437 .6228
7 .9904 .1561 .6228
8 1.0470 .1599 .6228

B. TRANSITION

.3932 .2245 .0793 .0453 .1365 .0779 .0275 .0157

.3044 .3470 .0425 .0484 .1057 .1205 .0147 .0168
.0484 .0425 .3470 .3044 .0168 .0147 .1205 .1057
.0453 .0793 .2245 .3932 .0157 .0275 .0779 .1365
.1365 .0779 .0275 .0157 .3932 .2245 .0793 .0453
.1057 .1205 .0147 .0168 .3044 .3470 .0425 .0484
.0168 .0147 .1205 .1057 .0484 .0425 .3470 .3044
.0157 .0275 .0779 .1365 .0453 .0793 .2245 .3932

* pij = Prostate j at time t + I I state i at time t}.

table 2. Appendix B reports on the ability of the model to approxi-

mate several moments of the data (see table B2 for a summary). Of
note is the fact that the Markov chain implies that the aggregate state
does not affect the conditional mean of income shares and that the
implied value of al, in (15) is -4.5.
So far we have assumed that the only tradable assets in positive net
supply are claims to the dividend stream. Since dividends average
only 3.9 percent of total income in the NIPA, this clearly understates
the share of income from tradable assets. Assets such as government
securities, corporate bonds, and so forth may also be sold or used as
collateral for loans. To roughly approximate these additional sources
of wealth, in most of the simulations we increase tradable asset hold-
ings by grossing up the assumed fraction of dividend income so that
nonlabor income is 15 percent of total income.'2 By comparison, capi-
tal's share of income in the NIPA averages about 30 percent. In

12 An alternative would be to use a debt measure such as net interest paym

the corporate sector to the household sector (as measured by "net interest" from the
NIPA). This is complicated, however, by the fact that these payments are not stationary,
increasing from 1.4 percent of total income in 1946 and rising to 12.1 percent of total
income in 1990.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 INCOMPLETE MARKETS 455

TABLE 2

A. MARKOV CHAIN MODEL FOR EXOGENOUS STATE VARIABLES:

CYCLICAL DISTRIBUTION CASE

STATES
STATE
NUMBER ya 8 1I

1 .9904 .1403 .3279

2 1.0470 .1437 .4405
3 .9904 .1562 .3279
4 1.0470 .1600 .4405
5 .9904 .1403 .6721
6 1.0470 .1437 .5595
7 .9904 .1562 .6721
8 1.0470 .1600 .5595

B. TRANSITION

.4365 .2343 .0881 .0473 .1515 .0098 .0306 .0020

.2416 .3461 .0337 .0483 .1698 .1201 .0237 .0168
.0555 .0458 .3977 .3281 .0193 .0019 .1381 .0137
.0360 .0792 .1783 .3924 .0253 .0275 .1253 .1362
.1515 .0098 .0306 .0020 .4365 .2343 .0881 .0473
.1698 .1201 .0237 .0168 .2416 .3461 .0337 .0483
.0193 .0019 .1381 .0137 .0555 .0458 .3977 .3281
.0253 .0275 .1253 .1362 .0360 .0792 .1783 .3924

Section IVF we also examine several cases in which bonds are as-
sumed to be in positive net supply and the level of dividend income
is reduced accordingly.
To summarize, the exogenous state variables include wt. 5t, and
tq. They evolve according to the Markov process specified above. An
endogenous component of the state is portfolio composition. In our
two-person economy, this is summarized by agent 1's holdings of
stocks and bonds, since agent 2's holdings can be derived using the
market-clearing conditions (8) and (9). We define the state vector of
the economy by Zt = {ty, bt 9 1, Sig bl}. Asset prices, consumption
policies, and trading policies are found as a function of Zt.

IV. Simulation Results

In this section we report the results of Monte Carlo simulations of

the economy of Section II using the exogenous driving processes
described in Section III. The equilibrium is solved numerically using
a modified version of the "auctioneer algorithm" described in Lucas
(1994).13

13 The numerical solution of the model introduces approximation error discussed

in App. C.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 456 JOURNAL OF POLITICAL ECONOMY

A number of summary statistics are reported. First, the volatility

of individual consumption implied by the model is compared to the
volatility of aggregate consumption. This statistic is of interest be-
cause risk sharing is not complete, so the extent to which individual
consumption behaves like aggregate consumption indicates the de-
gree to which risk sharing is being accomplished by trading securities.
We also report the mean and standard deviation of implied asset
returns in order to evaluate whether the model with incomplete mar-
kets and transactions costs can explain the equity premium and risk-
free rate puzzles discussed by Mehra and Prescott (1985) and Weil
(1989). Finally, we report the mean and standard deviation of trading
volume in order to gauge the sensitivity of trade to the level of trans-
actions costs.
These statistics are computed under the assumption that initially
each agent holds half the shares of stock and no debt. The economy
evolves for 1,000 years, driven by realizations of the exogenous in-
come process. Asset returns, trading volume, and consumption
growth between years 999 and 1,000 are recorded for each history.
The reported statistics are based on averages across 1,500 of these
experiments. 14

A. Representative Agent Baselines

Before turning to the implications of the incomplete markets model,

we briefly examine the implications of the complete markets case.
This experiment is very similar to the one undertaken by Mehra
and Prescott (1985), except that they equate dividends to aggregate
consumption whereas we treat consumption as the sum of measured
dividends and labor income.
Table 3 contrasts the results of the complete markets case with the
sample moments in the data. Columns 2 and 3 report the model
predictions for -y = 1.5 and ,B = 0.95, where the representative agent
is assumed to consume aggregate income. The representative agent
results for the base and cyclical distribution cases are slightly different
because the two laws of motion approximate the moments of aggre-
gate data differently. Notice that although the consumption process
differs from those used in previous studies, the implied asset prices
are similar.'5 For example, the predicted average return on risk-free

14 The long time horizon was chosen to eliminate the effect of initial conditions.
15 The sample moments of stock returns were calculated using annual returns
the 1947-90 value-weighted return from the Center for Research in Security Prices.
The moments of the bond returns were calculated using annual Treasury bill returns
for 1947-90. Both are converted to real returns using the Consumer Price Index
(CPI).

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 INCOMPLETE MARKETS 457

TABLE 3

MOMENTS IMPLIED BY THE COMPLETE MARKETS CASE

AGGREGATE INDIVIDUAL

Cyclical Cyclical
Base Distribution Base Distribution
DATA Case Case Case Case
MOMENT (1) (2) (3) (4) (5)

Consumption growth:
Average .020 .018 .015 .018 .015
Standard deviation .030 .028 .028 .217 .259
Bond return:
Average .008 .080 .077 .055 .041
Standard deviation .026 .009 .012 .175 .213
Stock return:
Average .089 .082 .078 .137 .152
Standard deviation .173 .029 .028 .375 .441

bonds is high and close to the stock return in each case, whereas the
observed average bond return is quite low.'6 Further, the standard
deviation of the stock return is low relative to historical levels.
To assess whether uninsurable labor income shocks have the poten-
tial to explain the poor performance of the representative agent
model, consider columns 4 and 5 in table 3. These statistics were
calculated in a representative agent model using one of the agent's
labor income dynamics as though they were the aggregate labor in-
come dynamics. These results demonstrate that when consumption is
equated to an individual's labor income process, the predicted equity
premium becomes much larger. For example, in the base case model,
the premium is predicted to be 8.2 percent. Consistent with the data,
the predicted standard deviation of the stock return is also quite large
(37.5 percent). Notice that in the CDC model, the equity premium is
even higher because the idiosyncratic shocks are concentrated in pe-
riods of low aggregate growth. The predicted level of both the bond
and stock returns is too high, but this can be corrected by assuming
a larger value of p.17
In general, these results indicate that the model with uninsurable
labor income has the potential to explain several of the observed
moments of asset returns. Of interest, however, is how much these
results change once trading is allowed in the stock and bond markets.

16 This finding is sometimes referred to as the risk-free rate puzzle rather than the
equity premium puzzle to draw attention to the fact that the standard model can
match the mean return on stocks but has difficulty simultaneously producing a realistic
average risk-free rate. For a discussion, see Weil (1989).
17 We keep 13 at a relatively low value in order to improve the performance of the
solution algorithm.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
458 JOURNAL OF POLITICAL ECONOMY

B. Frictionless Trading

Table 4 summarizes the case in which individuals can trade costlessly

in both the stock and bond markets, subject to the conditions that
stocks cannot be sold short and individuals can borrow only up to 10
percent of per capita income. Notice that for either income process,
individual consumption is only slightly more volatile than aggregate
consumption and is much less volatile than consumption under au-
tarky (see table 3).
The result that idiosyncratic shocks are offset by asset trades when
trading is costless strengthens the findings of Telmer (1993) and Lu-
cas (1994), who perform a similar experiment under the assumption
that labor income shocks are independent over time. In our model
the labor income shares are correlated over time, and hence the idio-
syncratic shocks are relatively persistent. This persistence, via a wealth
effect, should make it more difficult to self-insure through the asset
markets. A striking example of this is given by Constantinides and
Duffie (1994), who construct cases in which labor income shocks fol-
low a random walk and no smoothing occurs. However, the results
of table 4 indicate that with the persistence estimated using the PSID,
agents can still effectively smooth their idiosyncratic income shocks
by trading.
These findings suggest that increasing the predicted consumption

TABLE 4

MOMENTS IMPLIED BY THE FRICTIONLESS MODEL

Cyclical
Base Distribution
Moment Case Case

Consumption growth:
Average .018 .016
Standard deviation .044 .045
Bond return:
Average .077 .073
Standard deviation .012 .017
Stock return:
Average .079 .073
Standard deviation .032 .030
Bond trades (percent-
age of con-
sumption):
Average .045 .042
Standard deviation .060 .052
Stock trades (percentage
of consumption):
Average .131 .146
Standard deviation .066 .082

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 INCOMPLETE MARKETS 459

volatility in this model requires introducing some type of asset market

frictions. This observation motivates the experiments with transac-
tions costs and borrowing constraints that follow.

C. Transactions Costs in Both Markets and Asymmetric

Bond Market Costs

In simulations not reported here, we find that imposing a friction

in one market alone has a negligible impact on asset prices. As in
Constantinides (1986), portfolio balance is a second-order consider-
ation relative to intertemporal consumption smoothing. In the ab-
sence of binding borrowing or short-sales constraints, this causes the
agents to trade almost exclusively in the frictionless market. Hence
equilibrium returns are largely unaffected by the presence of trading
costs in only one market. This motivates our focus on specifications
in which either it is costly to trade in both markets or there are severe
borrowing constraints.
We first examine the case in which transactions costs are given by
(5) and (6). Bonds are in zero net supply, outstanding debt cannot
exceed 10 percent of per capita income, and only the borrower pays
the transactions costs. To examine the effect of increasing transac-
tions costs in a balanced way across markets, fl is varied from zero to
2.0 and k is set to W/2. The results are summarized in figures 1-4
for the base case model and 5-8 for the CDC model.'8 In each of
the figures the x-axis gives the value of f.
Figures 1 and 5 show the average stock return, bond return, equity
premium, and net premium (described below) as a function of ft for
each of the models. Because of the cost of borrowing, there is a wedge
between the lending rate and the borrowing rate. In the figures we
plot the average return to lending, since the equity premium is typi-
cally measured using the average return on Treasury bills, and indi-
viduals cannot borrow at this rate. Notice that as ft increases, the
average bond return falls and the equity premium widens. As shown
in figures 2 and 6, the average per capita transactions costs paid, as
a percentage of the value of trade, increase in f. For example, in the
base case model the average cost paid in the stock market rises from
0 percent to 5 percent of the value of trade.
Figures 3 and 7 report the average volume of trading in the two
markets as a percentage of per capita income. As the transactions

18 In constructing these results, we used a grid of values for fi between zero

2.0. To smooth across these grid points, we fit a third-order polynomial through t
points. This curve is reported in the figures. This smoothing removes some very small
variability in the figures that can be attributed to approximation error.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 0.09 l l l l

0.08 Stock Return

0.07
" Bond Return

0.06 -

E 0.5-

0.04 -

0.03 -

Equity Premium
0.02 -

0.01 -
Net Premium + +

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

Omega (k= Omega/2)

FIG. 1.-Base case, returns

0.06

0.05

0.04-

o Stock Market
X 0.03 -

a)

0.02 -

Bond Market _ - - -

0.01 /

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

Omega (k= Omega/2)

FIG. 2.-Base case, average costs

460

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 0.12 __~~~~~~Soc Market

m 0.1
Cc._
12

0 0.08
0)

<0.06

0.04
Bond Market

0.02 -

C , , , L I , I I I
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Omega (k= Omega/2)

FIG. 3.-Base case, average trading

0.11

0.1

0.09

0.08 -

V .
0

a 0 0.06 0
CD

0.05 -

0.04-

0.03-

0.02 f | |
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Omega (k= Omega/2)

FIG. 4.-Base case, STD of consumption

461

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 0.09

0.08 Stock Return

0.07

0.06 - s Bond Return

E 0.05 -

0.04 -

0.03 -

Equity Premium.
0.02 -

0.01 , Net Premium + + +

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

Omega (k = Omega/2)

FIG. 5.-CDC, returns

0.06

0.05

0.04 -

0
0
0

hi 0. 03 - Stock a

CL

0.02-

Bond Market -

0.01 /

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

Omega (k = Omega/2)

FIG. 6.-CDC, average costs

462

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 m 0.16
0. _

o0.08

o 0.06

0.04 - _ Bond Market

0.02

C
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Omega (k = Omega/2)

FIG. 7.-CDC, average trading

0.11

0.1

0.09

2i 0.08 /

0ro 0.07-

lba 0.06 0

0.05/

0.04-

0.03-

0.02 , . . . . . . .
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Omega (k = Omega/2)

FIG. 8.-CDC, STD of consumption

463

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
464 JOURNAL OF POLITICAL ECONOMY

costs increase, the level of trading drops off. This is reflected in a

higher level of consumption volatility, measured by the standard devi-
ation of the consumption growth of the first agent, shown in figures
4 and 8. The increased consumption volatility suggests that the model
can potentially explain the volatility bounds of Hansen and Jaganna-
than (1991) better than a representative agent model.
Notice that an increase in transactions costs results in a decrease in
the average bond return but no noticeable change in the average
stock return. This result is best understood by considering the direct
and indirect effects of transactions costs.

The Direct Effect of Transactions Costs

With moderate transactions costs in each market and fairly unrestric-

tive borrowing and short-sales constraints, agents trade almost every
period to buffer the idiosyncratic income shocks. The recipient of
the adverse idiosyncratic shock borrows and sells stock to partially
offset the shock, whereas the recipient of the favorable shock buys
both bonds and stocks to create a buffer against future adverse
shocks. For shocks of the size found in the PSID, the borrower is
willing to pay a relatively high borrowing rate including transactions
costs. For example, consider figure 1 at fl = 2. The average return
on the bond is 2.8 percent and the average transactions cost in the
bond market (for a single agent over time) is 2.1 percent. Since the
cost is paid only when borrowing and the cost function is quadratic,
the borrower pays a marginal transactions cost of 8.4 percent (4 x
2.1 percent), which implies a net marginal borrowing rate of 10.5
percent. The lender, however, receives only a 2.8 percent rate of
return on average. The lender is satisfied with this relatively low
average rate of return because the high consumption variability cre-
ates a precautionary demand for assets.
In contrast, the average stock return is near 8 percent for all levels
of fl in each model. For stocks, the direct effect of transactions costs
on the equilibrium stock return is difficult to predict since the buyer
demands a lower price to compensate for transactions costs whereas
the seller demands a higher price. Notice that the net return from
buying stock is lower than 8 percent because of transactions costs,
although in equilibrium the observed return is not greatly affected.
The lower net return also reflects a precautionary demand for assets
due to higher equilibrium consumption volatility.
In sum, for this cost specification, there is a direct effect of transac-
tions costs that depresses the observed lending rate and hence in-
creases the observed equity premium. Both stocks and bonds have a

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 INCOMPLETE MARKETS 465

lower net return due to a precautionary demand induced by higher

consumption variability.
A similar effect of differential stock and bond market transactions
costs on relative observed returns is found by Aiyagari and Gertler
(1991) and Vayanos and Vila (1995). However, in those models there
is no aggregate risk, so that differences in rates of return are gener-
ated solely by the differences in transactions costs across asset mar-
kets.

The Indirect Effect of Transactions Costs

To the extent that the increase in consumption volatility increases the

covariance between individual consumption and the net returns on
stock, the net-of-transactions-costs equity premium that agents demand
widens. There are several ways to construct this net premium. First
consider the payoff to an individual investor from investing in an
incremental unit of the stock at time t. This investment affects the
transactions costs paid at time t and at time t + 1. The sign and
absolute magnitude of these effects depend on the portfolio positions
at time t and at time t + 1. From (5), the marginal transactions cost
that agent i pays at time t in purchasing a unit of the stock is
2k,(stil- si)(ps)2. The marginal effect of this purchase on tr
tions costs at time t + 1 is - 2kt+ 1 (si+2 - s+ )(p+l)2. As a result
net (of transactions costs) one-period rate of return from an incre-
mental unit of investment in the stock is given by

rstn+eI+l + dP+s + 2k(s+2-s+i)(Ps )221. (16)

Pt 2k(st+1 - Stt

Notice that this rate of return satisfies

E{n u( )1 + rs',j) n it(t)j = 1, (17)

which is just the Euler equation for individual i (see [ 1 1]).

Similarly, the net-of-transactions-costs rate of return from an incre-
mental unit of investment in the bond is given by

1 iflb' O
rb net = (18)

.Pb + 2fltbt

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 466 JOURNAL OF POLITICAL ECONOMY

The indirect effect of transactions costs occurs to the extent that

the covariance between individual consumption and the net return
on the stock changes. To assess this effect, we plot the "net premium"
in figures 1 and 5. This measure of the equity premium is given by
E{rit net - rnet'~j}, which we calculate for individual 1. The net p
mium reflects the changing conditional covariance between r n~tj and
consumption growth as transactions costs increase. In the base case
model at fl = 2, the equity premium is 5 percent but the net premium
is only 1 percent. Also in the CDC model when Q1 = 2, the equity
premium is 5.5 percent and the net premium is 1.2 percent. It ap-
pears that the indirect effect does affect the observed equity pre-
mium, although it accounts for only 20 percent of the premium in
this case.'9
Another way to measure the net-of-transactions-costs premium is
to look at the difference between average stock and bond returns,
subtracting out the average realized transactions costs in each market.
Notice that a borrower must pay transactions costs on the entire
amount borrowed each period. In contrast, the stock market transac-
tions costs are a small percentage of the total return on stock holdings,
since only a portion of the stock portfolio typically turns over each
period. For instance, the average transactions costs in the stock mar-
ket as a percentage of the value of the typical stock portfolio rise to
a maximum of 0.4 percent when fl = 2 and k = 1 in the base case
model. As a result, the average net return on stocks is about 7.6
percent and the average return to a lender is 2.8 percent. This implies
a large net-of-transactions-costs equity premium of 4.8 percent.
Which of these two measures is a more appropriate measure of
the net-of-transactions-costs premium depends on the question being
asked. The first measure directly reflects the relation between risk
and return in the model since it is based on marginal rates of substitu-
tion, and it is the measure we focus on in the analysis. The second
measure answers the empirical question of how much investors earn
on stocks relative to bonds net of transactions costs. This latter mea-
sure can be calculated using only information on returns and average
transactions costs, whereas the former would also require detailed
information on consumption.

Size of the Costs and Robustness to

Cost Specifications

We have seen that transactions costs can be chosen so that the pre-
dicted stock and bond returns are similar to their observed values, but

19 It is possible that the impact of the indirect effect would be greater with higher
assumed risk aversion. An investigation of this issue presents significant computational
difficulties, which we plan to address in future research.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 INCOMPLETE MARKETS 467

the question remains whether observed transactions costs are large

enough to provide a plausible explanation. A full accounting of costs
should include brokerage fees, bid-ask spreads, and "price impact."20
Aiyagari and Gertler (1991) discuss the magnitude of brokerage fees
and argue (on the basis of Sharpe [1985]) that transactions costs as
high as 5 percent are reasonable. Further, the average level of broker-
age fees has fallen significantly over time, so a 5 percent transactions
cost in the stock market may not be large by historical standards.
Using the data base from the Institute for the Study of Security Mar-
kets, He and Modest (1995) estimate that the bid-ask spread implies
transactions costs of 1.5 percent for an equally weighted portfolio of
stocks and 0.75 percent for a value-weighted portfolio of stocks. This
increases the cost of trading over and above brokerage fees. Finally,
the price impact of a trade is a potentially important source of trans-
actions costs, but it is difficult to obtain a simple measure of the size
of this cost.
If marginal stock market transactions costs of 6 percent are taken
as a reasonable estimate, the model still predicts a substantial equity
premium. For example, when fl = 1 in the base case model, the
average costs are 1.4 percent in the bond market and 2.9 percent in
the stock market, and the equity premium is 3 percent. The net pre-
mium as measured using (17) and (18) is 0.5 percent, which implies
that the majority of the impact of the transactions costs comes from
the direct effect. Notice that to obtain an equity premium as large as
5 percent requires a marginal stock market transactions cost of 10
percent, so that without strict borrowing constraints, very large costs
are needed to produce a premium close to its observed average level.
As discussed in Section III, to determine whether the qualitative
results are sensitive to the quadratic form of the cost function, we
consider an alternative that is close to being proportional to the value
of trade. In the stock market the alternative cost function is given by

kpS E
k(Ist+l -St| - E)Ps + 2 if Ist+1 -StJ>E
K(St+ ,st;Zt) = 1k(st+,-St)2Pt i (19)

2> ~~~~~if JSt+ 1- Stl E

The cost function (19) is quadratic for small transactions and becomes
linear afterward.2' For E small, the cost function is essentially propor-

20 Price impact refers to the fact that large trades tend to move the price at which
the trade occurs.
21 The differentiability induced by smoothing the function at zero is convenient fo
computational purposes.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
468 JOURNAL OF POLITICAL ECONOMY

tional to the value of trade and k gives the marginal cost of a trade.
The cost function in the bond market is also pseudoproportional and
one-sided as in (6).
Figure 9 summarizes average predicted returns for this cost speci-
fication. Notice that the results are qualitatively similar to those for
the quadratic cost function reported in figure 1. For marginal transac-
tions costs of 2.5 percent in the stock market (Q = 0.05 and k =
0.025), the model with proportional costs predicts a 3 percent equity
premium and an average stock return of 8 percent.

D. Transactions Costs in Both Markets and Symmetric

Bond Market Costs

To investigate the effect of the incidence of transactions costs on

predicted returns, we assume that borrowing costs are split equally
by the lender and the borrower; the bond cost function (6) is replaced
with (7). In this case, any difference between the average bond and
stock return should reflect differences in risk between the two secu-
rities.
The results for expected rates of return are reported in figure 10
for the base case model and figure 11 for the CDC model. Limiting
fl to the range [0, 1] and setting k equal to fi produces average
transactions costs that are similar to those in figures 1 and 5, so that
the results of the two cost structures are directly comparable. In sharp
contrast to the case of one-sided borrowing costs, the equity premium
rises only slightly as a result of the increase in consumption volatility.
In this case the drop in the lending rate is not due to the incidence
of transactions costs but instead to an increased precautionary de-
mand for savings. The effect of the precautionary demand can be
seen most clearly by considering the net-of-transactions-costs bond
return. For example, in the CDC model with Q = 1, the average
bond return is 6.6 percent and the average bond market transactions
cost is 2.3 percent. This is not reflected in the observed bond return
because of the form of the cost function.
Again to assess the indirect effect of transactions costs on the equity
premium, it is useful to construct the net premium as measured by
E{rt+ -I tt+I}, where rst t +'I is given by (16) and

rtt+ E {b + 2flb'+ (pb)2} (20)

Notice that consistent with the idea that in this case the net premium
reflects only risk differentials, the measured equity premium and the
net premium track each other closely.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
0.09

Stock Return

0.07 - Bond Return

0.06 -

0.04

0.03

0.02 -.
Equity P
0.01 _
22. _+ -+ + + ~ + ~ ++ Net Premium

+++t + ++++++, ++.+.+ ,+++++ ,+.++

0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Omega (k= Omega/2)

FIG. 9.-Linear costs, base case, returns

0.09

0.08 _ Stock Return

0.07 - Bond Return

0.06

E0.05

0.04

0.03

0.02

0.01 Equity Premium (-.) Net Premium (+)

so
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Omega (k= Omega)

FIG. 10.-Base case, symmetric costs, returns

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
470 JOURNAL OF POLITICAL ECONOMY

0.09 , I I l

0.08 Stock Return

0.07 ?
Bond Return

0.06

E 0.05

M 0.04

0.03

0.02

0.01 Equity Premium (-.) Net Premium (+)

0.1 qutyPemum(-) + + + + +-

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Omega (k = Omega)

FIG. 11 .-CDC, symmetric costs, returns

E. Transactions Costs in Stocks and No Borrowing

In the preceding analysis the borrowing constraint is set so that it

rarely binds. There is some evidence, however, that agents may face
much more severe borrowing constraints. To examine this possibility
we consider a market structure that permits costly trading in stocks
but precludes borrowing altogether. In this case, the (shadow) price
of bonds is calculated using the marginal rate of substitution of the
agent with a good idiosyncratic shock.22 Hence, the results of this
section can be interpreted as approximating a situation in which there
is extreme credit rationing or there is a very large wedge between
the borrowing and lending rates.
Figures 12 and 13 report average returns and costs for the base
case model without borrowing,23 in which the only cost parameter
that varies is k. For any level of the costs, the effect of transactions
costs on the predicted bond return and the equity premium is larger
than in the case in which costly borrowing is allowed. For example,

22 A second way to close the bond market would be to impose a very high cos
the borrower in the bond market, which from a pricing perspective is equivalent to the
case in which borrowing is not allowed. See Heaton and Lucas (1992) for a discussion of
this issue.
23 We do not report the net premium in this case since the marginal borrowing rate
is effectively infinite.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 0.09 l l l l

Stock Return
0.08

0.07 -

0.06 - Bond Return

E 0.05 -

0.04 -

0.03 -

Equity Premirrm -
0.02 -

0.01 - _-

C
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
k

FIG. 12.-Base case, no bonds, returns

0.035

0.03

0.025

o0 0.02 - Stock Market

Cu

0 /
2 0.015 -

0.01

0.005 /

0 / --
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
k

FIG. 13.-Base case, no bonds, average costs

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
472 JOURNAL OF POLITICAL ECONOMY

when marginal costs in the stock market are 6 percent, the predicted
equity premium is 4 percent compared to 3 percent in the case with
costly borrowing (figs. 1 and 2). This occurs because agents cannot
substitute toward the bond market to avoid transactions costs. The
results for the CDC model are similar and are not reported.

F. The Effect of an Outside Supply of Bonds

In the preceding analysis, debt was assumed to be in zero net supply,

and the assumed level of stock holdings was grossed up to reflect
the income and liquidity provided by assets such as government and
corporate debt. This approach has the shortcoming that it does not
allow examination of the effect of differential transactions costs be-
tween stocks and outside bonds. However, we can reinterpret some
of our previous analysis to capture the presence of an outside supply
of government bonds.24
Suppose that the outside supply of bonds at time t is 2B, and, as
in Aiyagari and Gertler (1991) and Huggett (1993), that this debt
is financed by per capita lump-sum taxation in the amount of 1t
Bt- p+ Bt + . Assume further that there are no costs of transacti
in the bond market. The budget constraint of agent i becomes

ci + pSsi+1 + pbb'+ + K(s'+1,s'; Zt) + Tt sS(Ps + dt) + b' + Y', (21)

and the condition for equilibrium in the bond market is b1 + b2 -
2BV. Let a' --B' . Then (21) can be written as
C' + ps s+ I + t a'+ 1 + K(S+ 1, s'; Zt) ' s'(ps + dt) + a' + Y', (22)
and in equilibrium a' + a2 = 0. Further, if we assume that no bor-
rowing is allowed (b' ? 0), then each individual faces the restriction
at >- -Bt. Notice that this setup is equivalent to the case in which
there is no outside supply of bonds and some borrowing is allowed,
with a' in place of b'.
In Section IVC we showed that when the borrowing constraint is
not too restrictive (e.g., when it is set at 10 percent of per capita
income) and trading in at least one market is costless, agents substitute
almost completely toward trading in the market without transactions
costs. Consumption is effectively smoothed, and as a result, transac-
tions costs have little impact on the equity premium. In light of the
discussion in the previous paragraph, this result also will obtain when
there is a sufficiently large outside supply of bonds (e.g., equal to 10
percent of per capita income) and trading these outside bonds is

24 This interpretation was suggested to us by an anonymous referee.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 INCOMPLETE MARKETS 473

costless. Aiyagari and Gertler (1991) report similar results from a

model without aggregate shocks but with an infinite number of
agents.
To examine the sensitivity of our results to the supply and composi-
tion of assets in the presence of transactions costs, in the simulations
the sum of bond holdings across the two agents is set to a positive
value. This treats bonds as another "tree" technology, but with a
dividend equal to the equilibrium interest rate.25 The quadratic cost
function for bonds is also modified so that the buyer and seller face
symmetric transactions costs based only on changes in the value of
their bond holdings, so that the cost depends on lpb,, - btI rath
than b +1pb
To calibrate the average quantity of bonds held by households, we
use median household liquid assets as measured in the PSID, which
represent about 3.3 months of income for our sample.26 To bracket
this value, we set maximum per capita debt holdings to either 20
percent or 39 percent of per capita income and assume no inside
debt.27 Recall that the level of dividends in the preceding analysis was
grossed up relative to measured dividends to proxy for other tradable
assets. Here the assumed level of dividends is scaled back slightly to
keep the total capital stock approximately constant. This is referred
to below as the case of "high stock holdings." We also experiment
with reducing dividends to their measured level in the data, which
significantly lowers the total available liquidity in the economy.
The results can be summarized as follows: For average stock trad-
ing costs of 2.6 percent, average bond trading costs of 0.5 percent,
maximum per capita debt holdings of 20 percent, and high stock
holdings, the implied equity premium is 1.9 percent and the standard
deviation of consumption growth is 6 percent. For the same specifica-
tion except with no trading costs in the bond market, the equity pre-
mium and consumption volatility are almost identical. The reason is
that the level of debt is insufficient to entirely buffer the idiosyncratic
consumption shocks; debt holding is frequently at a corner, and most

25 This approach abstracts from the taxes the government would levy to support this
debt policy, which if included would be only a small fraction of income. As modeled,
these bonds more closely resemble corporate debt.
26 An alternative would be to use the per capita amount of government debt held
by the public, which represents a much larger fraction of income. This would probably
overestimate available household liquidity, however, since a substantial fraction of gov-
ernment debt is held by foreigners, pension funds, corporations, and other institutions.
27 The assumption of no inside debt was made primarily for computational tractabil-
ity. Whether it would represent a significant additional source of liquidity would de-
pend on the associated transactions costs assumed and the severity of the borrowing
constraint.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
474 JOURNAL OF POLITICAL ECONOMY

smoothing takes place via the stock market. When the quantity of
debt is increased to 38 percent of per capita income with no change
in other parameters, the equity premium falls to 1.2 percent and
the standard deviation of consumption to 5.5 percent, reflecting the
increased opportunities to smooth costlessly in the bond market. As
in our comparison of the results of figures 10 and 12, the premium
is quite sensitive to the amount of debt available and is reduced when
an outside supply of bonds is introduced.
Finally, we consider the low-liquidity case in which dividends aver-
age only 5 percent of income (consistent with NIPA data) and maxi-
mum bond holdings are 20 percent of per capita income. For an
average transactions cost of 2.6 percent in the stock market and no
transactions costs in the bond market, the predicted equity premium
increases to 2.7 percent, and the standard deviation of consumption
is 8 percent. Notice that this result is similar to the base case (fig. 1),
where the level of dividends was grossed up from its observed value
to more closely reflect the quantity of tradable assets. As a result,
even with an outside supply of bonds the model predicts a substantial
equity premium for a reasonable level of transactions costs in the
stock market. However, consistent with the base case, very large trans-
actions costs in the stock market are needed to obtain an equity pre-
mium close to its historical value.

V. Concluding Remarks

In this paper we have examined asset prices and consumption pat-

terns in a model in which agents face both aggregate and idiosyncratic
income shocks, and insurance markets are incomplete. Agents reduce
consumption variability by trading in a stock and bond market to
offset idiosyncratic shocks, but frictions in both markets limit the
extent of trade. The theoretical model is calibrated under a variety
of assumptions about the form and size of frictions using an empirical
model of labor and dividend income estimated using data from the
PSID and the NIPA. Transactions costs in the stock and bond markets
generate an equity premium and lower the risk-free rate. By simulta-
neously considering aggregate and idiosyncratic shocks, we can de-
compose this effect of transactions costs on the equity premium into
two components, a direct and an indirect effect.
The direct effect occurs because individuals equate net-of-cost mar-
gins, so an asset with a lower associated transactions cost will have a
lower market rate of return. The size of the direct effect varies widely
with the structure of transactions costs. When the cost structures in
the stock and bond markets are assumed to be similar, the direct
effect is negligible. However, when we assume that the transactions

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 INCOMPLETE MARKETS 475

cost in the bond market takes the form of a wedge between the bor-
rowing and lending rates or that there is a binding borrowing con-
straint, the direct effect can account for close to half of the observed
premium. This is related to the findings of Aiyagari and Gertler
(1991), which led them to conjecture that realistic transactions costs
might account for about 50 percent of the observed equity premium.
A second, indirect effect occurs because transactions costs result in
individual consumption that more closely tracks individual income
than aggregate consumption. The higher variability of individual
consumption increases the covariance between consumption and the
dividend process and, hence, increases the systematic risk of the stock.
While the size of the indirect effect increases with the assumed level
of transactions costs, it is relatively insensitive to the incidence of
transactions costs. In the base case analysis, the indirect effect ac-
counts for about 20 percent of the premium.
Although the model can generate an equity premium and a risk-
free rate that match the historical means, to do this the necessary
level of transactions costs is very large. Also the results of our model
are quite sensitive to the assumed quantity of tradable securities. For
example, the addition of a large supply of government securities (out-
side bonds) dampens the effect of transactions costs substantially. All
this suggests the sensitivity of asset price predictions to assumed mar-
ket structure.
A further difficulty with the model is that it does not explain the
observed second-moment differentials. It shares the feature of many
consumption-based models that an increase in the equity premium is
not accompanied by a substantial change in the relative volatility of
bond and stock returns. Typically in models that fit the equity pre-
mium, the resulting volatility of the bond return is too high. In con-
trast, when this model is parameterized to match the equity premium,
there is little increase in observed return volatilities, resulting in too
little stock return volatility. It remains an open question whether
there is a realistic assumption about transactions costs that can simul-
taneously explain the low volatility of short bond rates and the high
volatility of stock returns.

Appendix A

Data Description and Estimation Results from the PSID

In this Appendix we describe the estimation (summarized in Sec. IIIA) in

more detail and describe a more general model of labor income that is also
investigated.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
476 JOURNAL OF POLITICAL ECONOMY

A. Specification of the Model o

Individual i's labor income as a fraction of aggregate labor income is given
by -t, where { ..}.., is assumed to be a stationary process for each i. To
generalize (14), we estimate the process for -t using the specification

log(-qr) = ji, + clog( + p log(-q'1) + et, (Al)

where the {E'}1-=1,n are individual shocks that have mean zero, are in
dent over time and across individuals, and are independent of the lagged
aggregate shock Etal for all t; E{(Ei)2}"12 = vi; and {jiji=1,n, {qi}i=,
{pZ}Z ln are parameters.
The parameters { 1j}Z=1,n capture permanent differences in relative labor
income. The parameter {JC} reflects the degree to which individual i's relative
income can be predicted by lagged aggregate shocks. Differences in h across
individuals allow the aggregate shock to differentially affect the conditional
mean of each individual's income. The parameter pi captures the persistence
in shocks to individual i's labor income.
To capture the differences in the distribution of labor income shocks over
the business cycle, we also consider a linear model of the standard deviation
of individual income shares as a function of the growth rate in aggregate
income, which is given by (15) above.

B. Data

The PSID provides a panel of annual observations of individual and family

income and other variables. In using the PSID, we take family income per
family member as a measure of Yi for our model. Because of the changing
character of many of the families, we take a subsample from the PSID con-
sisting of those families in which the head of the household was male and
neither the head nor his spouse changed over the sample. This selection is
used to avoid having to keep track of new families, family split-offs, and
other dramatic changes in the family. The complete PSID oversamples
poorer members of the U.S. population because of an inclusion of a sample
of poor individuals from the Survey of Economic Opportunity. To make the
sample closer to a random sample of the U.S. population, we excluded those
families that were originally part of the Survey of Economic Opportunity.
Total labor income of the head of the household and his wife along with
total transfers to the family are used for total family income. The transfers
included unemployment compensation, workers' compensation, pension in-
come, welfare, child support, and so on. If a family had zero income from
all sources in any year, it was excluded from the sample.28 This sample in-
cludes 860 families with income data spanning the years 1969-84. Because

28 Once we conditioned on families with neither a change in the head of the hou
hold nor a change in his spouse, there were only 15 out of 875 families that had zero
income in any of the sample years. As a result, the exclusion of families with zero
income has little effect on the results.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 INCOMPLETE MARKETS 477

families differ in size, family income per family member is created by dividing
each family's income by the total number of family members in each year.29
This measure of nominal family labor income is weighted by the CPI to
obtain a measure of real labor income per family member.
The model of individual income dynamics is estimated using this full sam-
ple and also using a subsample restricted to households owning stock. The
sample is split because a large segment of the population does not hold
financial assets. For analysis of an asset pricing model, it is clearly more
appropriate to consider the income dynamics of those individuals who partici-
pate in securities markets. Following Mankiw and Zeldes (1991), we split the
sample on the basis of individual holdings of securities using questions about
stock holdings in the 1984 PSID. If a family reported some holdings of stocks
in 1984, it is included in the group called stockholders.
For both the complete sample and the stockholder sample, -qc is constructe
as Yt/(l(Y Y'), where n is 860 for the complete sample and 327 for the
stockholder sample.
Along with observations of individual labor income from the PSID, mea-
sures of annual aggregate labor income and dividends are taken from the
NIPA for the years 1947-92, obtained from Citibase. For labor income we
use "total compensation of employees." The aggregate series are weighted
by the total U.S. population and the CPI in each year to obtain real per capita
labor income and dividends.

C. Empirical Results

Aggregate Dynamics

A natural alternative to using the NIPA to measure aggregate labor income

would be to use total income from the PSID. However, because of the limited
time dimension in the PSID, the aggregate dynamics could not be estimated
with much precision. Although the NIPA measure of labor income does not
exactly correspond to the measure of labor income obtained from the PSID,
the two measures of aggregate income are similar. For example, figure Al
gives a plot of the NIPA (from Citibase) measure of aggregate labor income
growth, log(Ya/Ya 1), along with the corresponding measure constructed
from the PSID from 1970 to 1984. The correlation between the two series
is .88. As a result, measuring aggregate labor income from the NIPA series
should do a reasonable job of capturing the aggregate labor income that
corresponds to the sample from the PSID. Table Al reports ordinary least
squares estimates of the law of motion for aggregate labor income and aggre-
gate dividends (see eq. [12]).

Estimation of (Al)

For each household in both samples from the PSID, we estimate (Al) using
ordinary least squares. A summary of these findings is given in table A2,

29 We also conducted the analysis in which family income was weighted only by the
number of adults in the family. The results were similar and hence are not reported.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 0.1 I I I I I

0.08 - Citibase PSID

0.06

0.04 -

0.02-

-0.02-

-0.04-

-0.06l
1970 1972 1974 1976 1978 1980 1982 1984
Date

FIG. Al. -Labor income growth rates

TABLE Al

AGGREGATE DYNAMICS

'Yt t y 1 t-tDta/Yta, and Xa [log(ya)log(8t)]'

x a = AaXa1 + 0aEa + la

Ordinary Least Squares Estimates

.1487 .0557 -
(.1645) (.0378) -.0278 0
A= [ 1a =
-.5016 .9168 .0121 .0536
L (.2532) (.0696)1
[ .1961 1
(.1228)
IL a =

- .2607
(.2260)1

NOTE.-Standard errors are in parentheses.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 INCOMPLETE MARKETS 479

TABLE A2

INDIVIDUAL INCOME DYNAMICS WITH AGGREGATES:

CROSS-SECTIONAL MEANS AND STANDARD ERRORS
OF COEFFICIENT ESTIMATES

log(qi) = Ni + pi log(Nq1) + (ilog(YIY 1) + et

i = 21/2

CROSS-SECTIONAL MEAN

(Standard Deviation)

COEFFICIENT All Individuals Stockholders

-3.564 (2.534) -4.222 (1.919)

.081 (2.542) - 1.482 (.314)
p. .527 (-349) .299 (.155)
9r .241 (.127) .365 (3.638)

which reports sample averages of the parameter estimates along with the
cross-sectional standard deviations of the estimates. The estimates reported
in table A2 are consistent with results reported in MaCurdy (1982) and
Abowd and Card (1989). In particular, the estimated parameter values im-
ply that individual income growth is negatively correlated over time. Abowd
and Card also argue that aggregate variation in income has little effect on
the autocorrelation structure of individual income. The average parameters
reported in table A2 along with the point estimates of the aggregate dynamics
reported in table Al imply that the aggregate shocks account for only 1
percent and 2 percent of the variance in individual income shares for the
entire sample and the stockholder sample, respectively. For this reason, we
also estimate the law of motion for the W's with ( 0 for each i. A summary
of the results of this estimation is reported in table A3. Notice that for each
sample the average estimated value of pi and a' is very close to the average
value reported in table A2. We conclude that aggregate shocks are not very
important in explaining the conditional mean and unconditional variance of
the average individual's income share.
In comparing the results for the entire sample with the results for the
stockholder subsample, notice that the average autocorrelation coefficient for
the stockholders is smaller than for the entire sample. Also, the variance
of the idiosyncratic shock, ,i, for the stockholders is slightly larger. Howev
the results generally indicate that there is significant autocorrelation in indi-
vidual labor income shocks, the variance of these shocks is quite large, and
there is little role for the aggregates in explaining the unconditional variance
of the shocks. The base case used in experimenting with the asset pricing
model is based on an approximation to the dynamics implied by the point
estimates of tables Al and A3 for the complete sample. As described below,
this requires matching the first-order autocorrelation and variances in indi-
vidual income shares, along with the first-order dynamics in aggregate in-
come and dividends.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 TABLE A3

INDIVIDUAL INCOME DYNAMICS WITHOUT AGGREGATES:

CROSS-SECTIONAL MEANS AND STANDARD DEVIATIONS
OF COEFFICIENT ESTIMATES

St--Y/t

log~r) = P j + p log(9i, ) + E,
a =~~t
r = [E (E i)2]2

CROSS-SECTIONAL MEAN
(Standard Deviation)

COEFFICIENT All Individuals Stockholders

-3.354 (2.413) -4.241 (.299)

p .529 (.332) .299 (.304)
9i .251 (.131) .383 (.159)

TABLE A4

CROSS-SECTIONAL STANDARD DEVIATION AND AGGREGATES

Std[Iog(Et)] = oao + aot log(YI/Yt 1) + it

ESTIMATED PARAMETER
(Standard Error)

COEFFICIENT All Individuals Stockholders

at0 .272 (.001) .360 (.014)

tl .052 (.039) - 1.064 (.4 19)

NOTE.--.-tkI1jOg(Et)l is the estimated cr

individual income.

TABLE A5

CROSS-SECTIONAL STANDARD DEVIATION AND AGGREGATES

St-i[log(-T)] = ao + aI log(YI/Yt 1) + it

ESTIMATED PARAMETER
(Standard Error)

COEFFICIENT All Individuals Stockholders

ao .716 (.002) .682 (.007)

(X1 -.226 (.058) .894 (.226)

NOTE.--SidjIog(1t)j is the esti

income shares.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 INCOMPLETE MARKETS 481

Conditional Heteroskedasticity

To capture conditional variation in the standard deviation of individual

shocks, we estimate (15) on the basis of the results of tables A2 and A3 and
the residuals implied by the estimates of table A3. For each time period, the
cross-sectional standard deviation of the estimated residuals, given by (1 IN)
E1 (i)2, is used as an estimate of the variable on the left side of (15).
The resulting parameter estimates from ordinary least squares are re-
ported in table A4. Notice that for the entire sample of individuals the aggre-
gate state has a positive effect on the cross-sectional standard deviation, which
contradicts the presumption that individual income variance increases in eco-
nomic downturns. The reason may be that our selection criterion eliminates
those households with substantial changes in family composition, which may
bias our estimate of the conditional heteroskedasticity of income shocks. Also
the short time-series dimension of the PSID may limit our ability to recover
a countercyclical pattern in the variance of shocks to individual income. In
contrast, the results from the stockholder subsample provide evidence in
favor of a model in which the aggregate state has a small negative effect on
the cross-sectional standard deviation of labor income. The estimates are
sensitive to specification, however. In table A5 we report regressions of the
cross-sectional standard deviation of log(^q) on aggregate income. Notice that
in this case, when the complete sample from the PSID is used, there is evi-
dence of a widening in the distribution of labor income during aggregate
downturns. These results indicate that with the short sample period of the
PSID, it is difficult to precisely quantify the cyclical movement in the distribu-
tion of labor income.

Appendix B

Markov Chain Approximation

We construct Markov chain approximations to the laws of motion (12), (14),

and (15) using the method of Tauchen and Hussey (1991), with two states
for each of the exogenous states. Here we report measures of how well the
Markov chain approximates important aspects of the dynamics of income as
summarized by the estimates reported in tables Al, A3, and A4.

Base Case

Table B 1 reports the inputs for the base case model, which are the parameter
vectors A, 0, and FL of the trivariate VAR for aggregate income growth,
the dividend share, and individual I's (demeaned) income share. They are
reported in table B 1 as "Input Values" and are given by the point estimates
of tables Al and A3 for the complete sample. As discussed by Tauchen and
Hussey (1991), one way to assess the Markov chain approximation error is
to examine the parameter matrices A, 0, and FL implied by the Mar
chain. They are reported in table B1 as "Implied Values." Notice that the
persistence and variability of individual l's labor income are reasonably well

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
482 JOURNAL OF POLITICAL ECONOMY

TABLE B1

PARAMETERS IMPLIED BY MARKOV CHAIN APPROXIMATION: BASE CASE

aY Ya- 1, St Dt IY,, and X= [Iog(-y) 1og(bt)[1og(' ) - E 1og(jt)]]',

Xt = AX, I + 0Et + p

Input Values

[ .1487 .0557 0 1 -.0278 0 0 1

A= -.5016 .9168 0 , Et= .0121 .0536 0
[ 0 .52901 L 0 0 .25081

[ .19611
-.2607
0

Implied Values

[ .1459 .0539 0 1 [.0272 0 0 1

A= -.3330 .7331 0 , Ot = .0118 .0379 0
[ 0 0 .48461 L 0 0 .21941

[ .1902-
-.8594
0

approximated. However, the aggregate dynamics are not as well approxi-

mated.
Figure B1 plots the dividend's share of aggregate income from 1948 to
1992. The most substantial movement in dividends occurs in the early 1970s.
The Markov chain does not capture the slow movement down and then up
in the dividend share over this period. However, the overall variation in
dividends is not large. Since we focus on the effects of individual income
shocks, the fact that we do not completely fit the dynamics of aggregate
dividends appears to be unimportant. Essentially the dividend stream in our
model is given by a constant fraction of the aggregate endowment. We miss
some of the slow movement in the dividend share, but this is not substantial.
The Markov chain reported in table 1 is a transformation of the Markov
chain approximation to the VAR of table B1. First, as described in Section
III, the dividend share is grossed up so that, on average, it accounts for 15
percent of aggregate income. This is done by multiplying the share variable
by a scale factor. Second, the individual income share is scaled so that it
averages 50 percent. As a result, each individual's labor income share is, on
average, 50 percent. This is also accomplished by multiplying the share value
by a scale factor.

Cyclical Distribution Case

We deviate from the estimated parameters of the conditional heteroskedastic-

ity model in order to obtain a case that is significantly different from the

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 0.06

0.055

0.05

0
C0.045 -

0.045

0.045

0.035
1950 1955 1960 1965 1970 1975 1980 1985 1990
Date

FIG. B1.-Dividend income share

TABLE B2

PARAMETERS IMPLIED BY MARKOV CHAIN APPROXIMATION:

CYCLICAL DISTRIBUTION CASE

-Y1 Yt1Y,, Bt DYt, and X1 [log(,y') log(&1)[log(rt) - E log(-q)]]',

Xt = AXt-I + 0tet + L

Input Values

[ .1487 .0557 0 1 .0278 0 01

A= -.5016 .9168 0 , 1= .0121 .0536 0
[ 0 0 .529 L 0 0 urj

F .19611
= -.2607

Implied Values

[ .2194 .0517 0 1 -.0267 0 01

A= -.3010 .7322 0 , 0'= .0116 .0379 0
0 0 .4307 0 0 ort

[ .1797-
.8640, at = .2898 - 4.4504 log(,y')
0

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 484 JOURNAL OF POLITICAL ECONOMY

base case and provides a countercyclical distribution of income. As in the

base case, there are eight states in total, with two states for aggregate income.
The parameters of tables Al and A3 are used to calibrate the conditional
mean of the variables. The volatility of the individual shock is chosen to be
twice as large in the low-aggregate growth state as in the high-aggregate
growth state, with an average volatility chosen to match the average volatility
of the shocks reported in table A3. The model maintains the feature of the
base case that the conditional mean of individual income shares (as measured
by a linear regression) is not affected by the aggregate state.
Table B2 reports the inputs for the Markov chain. The matrix 0e is time
dependent because of the conditional heteroskedasticity of the shocks. The
parameters of the approximating Markov chain are given in table 2 (trans-
formed as in the base case). To check the approximations, table B2 also
reports the VAR parameters implied by the approximating Markov chain.
Notice that the model fits the persistence in individual income reasonably
well, and the aggregate dynamics are fit about as well as in the base case. As
a way of evaluating the heteroskedasticity implied by the model, table B2
also reports the implied parameters of the heteroskedasticity regression. No-
tice that the coefficient on aggregate income is four times larger than the
estimated value reported in table A4 for the stockholder subsample.

Appendix C

Numerical Accuracy

Model equilibria are computed using the "auctioneer" algorithm of Lucas

(1994). This algorithm searches for an allocation of bond and stock holdings
for the two agents that clears markets in every state of the world and implies
agreement between the two agents on the prices of the two securities.30 For
a given allocation, the Euler equations of the unconstrained agents define
functional equations in the prices of the two securities as functions of the
exogenous and endogenous state variables. These equations are approxi-
mated over a grid of 30 equally spaced points for stock and bond holdings.
The resulting discrete state space is 30 x 30 x 8 since the exogenous vari-
ables take on eight different values. A fixed point to the equations is found
using the fact that the Euler equations form a contraction mapping. Iterations
continue until there is less than a 0.5 percent difference in the (uniformly
weighted) average solution across the grid space.
Owing to the discrete approximation, it is not possible to set the prices
quoted by the two agents exactly equal to one another. We report the average
of the difference between the prices quoted by the two agents as a percentage
of the price quoted by the first agent in tables C I and C2. The approximation
error is reported for the case of no outside bonds and asymmetric and sym-
metric cost functions in the bond market. For settings of the cost parameters
that produce plausible marginal transactions costs, the algorithm produces

30 For a complete description of the algorithm, see Lucas (1994).

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 INCOMPLETE MARKETS 485

TABLE C1

APPROXIMATION ERRORS: ASYMMETRIC COSTS (Average Percentage Difference in

Prices between the Two Agents)

BASE CASE CYCLICAL DISTRIBUTION CASE

VALUE OF il Stock Price Bond Price Stock Price Bond Price

.0 .19 .10 .23 .09

.2 .19 .36 .28 .37
.4 .14 .24 .29 .22
.6 .15 .23 .33 .22
.8 .17 (.23 .39 .27
1.0 .17 .21 .53 .39
1.2 .18 .18 .59 .49
1.4 .20 .21 .65 .50
1.6 .22 .23 .71 .49
1.8 .25 .27 .86 .62
2.0 .23 .27 .90 .63

TABLE C2

APPROXIMATION ERRORS: SYMMETRIC COSTS (Average Percentage Difference in

Prices between the Two Agents)

BASE CASE CYCLICAL DISTRIBUTION CASE

VALUE OF Q Stock Price Bond Price Stock Price Bond Price

.0 .26 .14 .21 .08

.2 .30 .07 .33 .17
.4 .17 .33 .34 .86
.6 .13 .28 .37 .42
.8 .12 .34 .39 .43
1.0 .12 .40 .45 .51

reasonably accurate solutions. For example, when il is restricted to be less

than one in the asymmetric cost cases, the average errors are less than 0.5
percent.

References

Abowd, John M., and Card, David. "On the Covariance Structure of Earnings
and Hours Changes." Econometrica 57 (March 1989): 411-45.
Aiyagari, S. Rao. "Uninsured Idiosyncratic Risk and Aggregate Saving."
Q.J.E. 109 (August 1994): 659-84.
Aiyagari, S. Rao, and Gertler, Mark. "Asset Returns with Transactions Costs
and Uninsured Individual Risk."J. Monetary Econ. 27 (June 1991): 311-3
Amihud, Yakov, and Mendelson, Haim. "Asset Pricing and the Bid-Ask
Spread." J. Financial Econ. 17 (December 1986): 223-49.
Bewley, Truman F. "Stationary Monetary Equilibrium with a Continuum
of Independently Fluctuating Consumers." In Contributions to Mathematical

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 486 JOURNAL OF POLITICAL ECONOMY

Economics in Honor of Gerard Debreu, edited by Werner Hildenbrand and

Andreu Mas-Colell. Amsterdam: North-Holland, 1986.
Clarida, Richard H. "International Lending and Borrowing in a Stochastic
Stationary Equilibrium." Internat. Econ. Rev. 31 (August 1990): 543-58.
Constantinides, George M. "Capital Market Equilibrium with Transaction
Costs."J.P.E. 94 (August 1986): 842-62.
Constantinides, George M., and Duffie, Darrell. "Asset Pricing with Hetero-
geneous Consumers." Manuscript. Chicago: Univ. Chicago, Grad. School
Bus., 1994.
Dumas, Bernard. "Two-Person Dynamic Equilibrium in the Capital Market."
Rev. Financial Studies 2, no. 2 (1989): 157-88.
Hansen, Lars Peter, and Jagannathan, Ravi. "Implications of Security Market
Data for Models of Dynamic Economies."J.P.E. 99 (April 1991): 225-62.
Hansen, Lars Peter, and Singleton, Kenneth J. "Stochastic Consumption,
Risk Aversion, and the Temporal Behavior of Asset Returns."J.P.E. 91
(April 1983): 249-65.
He, Hua, and Modest, David M. "Market Frictions and Consumption-Based
Asset Pricing."J.P.E. 103 (February 1995): 94-117.
Heaton, John, and Lucas, Deborah J. "The Effects of Incomplete Insurance
Markets and Trading Costs in a Consumption-Based Asset Pricing Model."
J. Econ. Dynamics and Control 16 (July/October 1992): 601-20.
. "The Importance of Investor Heterogeneity and Financial Market
Imperfections for the Behavior of Asset Prices." Carnegie-Rochester Conf.
Ser. Public Policy 42 (June 1995): 1-32.
Huggett, Mark. "The Risk-Free Rate in Heterogeneous Agent Incomplete-
Insurance Economies." J. Econ. Dynamics and Control 17 (September/No-
vember 1993): 953-69.
Lucas, Deborah J. "Asset Pricing with Undiversifiable Income Risk and Short
Sales Constraints: Deepening the Equity Premium Puzzle." J. Monetary
Econ. 34 (December 1994): 325-41.
MaCurdy, Thomas E. "The Use of Time Series Processes to Model the Error
Structure of Earnings in a Longitudinal Data Analysis."J. Econometrics 18
(January 1982): 83-114.
Mankiw, N. Gregory. "The Equity Premium and the Concentration of Aggre-
gate Shocks." J. Financial Econ. 17 (September 1986): 211-19.
Mankiw, N. Gregory, and Zeldes, Stephen P. "The Consumption of Stock-
holders and Nonstockholders."J. Financial Econ. 29 (March 1991): 97-112.
Marcet, Albert, and Singleton, Kenneth J. "Equilibrium Asset Prices and
Savings of Heterogeneous Agents in the Presence of Incomplete Markets
and Portfolio Constraints." Manuscript. Stanford, Calif.: Stanford Univ.,
Grad. School Bus., 1991.
Mehra, Rajnish, and Prescott, Edward C. "The Equity Premium: A Puzzle."
J. Monetary Econ. 15 (March 1985): 145-61.
Pratt, John W., and Zeckhauser, Richard J. "Proper Risk Aversion." Economet-
rica 55 (January 1987): 143-54.
Saito, Makoto. "Limited Market Participation and Asset Pricing." Manuscript.
Vancouver: Univ. British Columbia, Dept. Econ., 1995.
Scheinkman, Jose A. "Discussion: Market Incompleteness and the Equilib-
rium Valuation of Assets." In Theory of Valuation: Frontiers of Modern Finan-
cial Theory, vol. 1, edited by Sudipto Bhattacharya and George M. Con-
stantinides. Totowa, N.J.: Rowman and Littlefield, 1989.
Scheinkman, Jose A., and Weiss, Laurence. "Borrowing Constraints and Ag-
gregate Economic Activity." Econometrica 54 (January 1986): 23-45.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms
 INCOMPLETE MARKETS 487

Sharpe, William F. Investments. 3d ed. Englewood Cliffs, N.J.: Prentice-Hall,

1985.
Tauchen, George, and Hussey, Robert. "Quadrature-Based Methods for Ob-
taining Approximate Solutions to Nonlinear Asset Pricing Models." Econo-
metrica 59 (March 1991): 371-96.
Telmer, Chris I. "Asset-Pricing Puzzles and Incomplete Markets."J. Finance
48 (December 1993): 1803-32.
Vayanos, Dimitri, and Vila, Jean-Luc. "Equilibrium Interest Rate and Liquid-
ity Premium under Proportional Transactions Costs." Manuscript. Stan-
ford, Calif.: Stanford Univ., Grad. School Bus., 1995.
Weil, Philippe. "The Equity Premium Puzzle and the Risk-Free Rate Puzzle."
J. Monetary Econ. 24 (November 1989): 401-21.
. "Equilibrium Asset Prices with Undiversifiable Labor Income Risk."
J. Econ. Dynamics and Control 16 (July/October 1992): 769-90.

This content downloaded from

3.104.43.49 on Fri, 27 Sep 2024 13:08:55 UTC
All use subject to https://about.jstor.org/terms