Incomplete Market Dynamics

in a Neoclassical Production Economy

George-Marios Angeletos and Laurent-Emmanuel Calvet∗

This version: May 2004

We investigate a Ramsey growth model with incomplete markets, decentralized production, and idiosyncratic
technological risk. The combination of uninsurable shocks with the precautionary motive can generate
endogenous poverty traps even when agents are very patient and technology is strictly convex. In contrast
to earlier research, the multiplicity of equilibria arises from the endogeneity of the interest rate instead of
the usual wealth effect. Depending on the economy’s parameters, the local dynamics around a steady state
are locally unique, totally unstable or locally undetermined, and the equilibrium path can be attracted to
a limit cycle. The model generates closed-form expressions for the equilibrium dynamics under finite and
infinite horizon, and easily extends to a variety of environments, including multiple sectors and endogenous
labor supply.

Keywords: Idiosyncratic Risk, Precautionary Motive, Endogenous Fluctuations, Poverty Traps.

∗
Correspondence: Department of Economics, MIT, 50 Memorial Drive, Cambridge, MA 02142, and NBER, an-
gelet@mit.edu; Department of Economics, Littauer Center, Harvard University, Cambridge, MA 02138, and NBER,
laurent_calvet@harvard.edu. We received helpful comments from P. Aghion, A. Banerjee, R. Barro, J. Benhabib,
O. Galor, J. Geanakoplos, J.M. Grandmont, O. Hart, N. Kiyotaki, D. Laibson, H. Polemarchakis, K. Shell, R.
Townsend, and seminar participants at Brown, Harvard, Maryland, MIT, NYU and Yale. Ricardo Reis provided
excellent research assistance.
1 Introduction

How does incomplete risk-sharing affect the level and volatility of macroeconomic activity? We
investigate this question in a Ramsey growth model with incomplete markets, decentralized pro-
duction, and idiosyncratic technological risk. The combination of uninsurable risks with the pre-
cautionary motive can generate poverty traps and endogenous fluctuations, even when agents are
very patient, technology is strictly convex, and the wealth distribution has no effect on endogenous
aggregates.
We conduct the analysis in the GEI growth model with heterogeneous agents introduced in
Angeletos and Calvet (2003). Agents are both consumers and producers, and can invest in a private
neoclassical technology with diminishing returns to scale. In contrast with the traditional Ramsey
model, agents are exposed to idiosyncratic shocks in their productive investment, and possibly in
some exogenous income. Agents can also trade in financial markets. They can borrow or lend a
risk-free bond, and partially hedge their idiosyncratic risks by exchanging a finite number of risky
assets. All securities are real and there are no constraints on short sales. The economy cannot be
described by a representative agent, but explicit aggregation is possible in the CARA-normal case.
When markets are complete, the model reduces to the Ramsey growth model with identical agents,
as in Cass (1965), Koopmans (1965), and Brock and Mirman (1972). On the other hand, with
missing markets, inefficiencies arise because individual producers must bear idiosyncratic risks that
cancel out at the aggregate level.
Two main results are derived in this framework by Angeletos and Calvet (2003). First, idiosyn-
cratic production risk can generate a higher steady state than under complete markets, in contrast
to the results of Aiyagari (1994). These results illustrate the fact that market incompleteness gen-
erally has an ambiguous effect on aggregate wealth in a steady state. Limited risk-sharing increases
the riskiness of returns on productive activities as compared to complete markets, thus discouraging
investment. Incompleteness also increases the precautionary demand for savings, thereby decreas-
ing the interest rate and encouraging productive investment. In calibrated examples, the level of
aggregate wealth is found to be larger in economies with higher levels of financial sophistication.1
Second, market incompleteness increase the magnitude and persistence of the business cycle. In the

1
The literature has often suggested that financial innovation reduces the precautionary motive and may therefore
be detrimental to growth (Mauro, 1995; Devereux and Smith, 1994). This type of argument ignores the adverse
effect of production risk on the accumulation of physical or human capital. Angeletos and Calvet (2003) analyzes
the trade-off between the two effects in a Ramsey framework with diminishing returns to capital accumulation, while
other pertinent work (e.g., Greenwood and Jovanovic, 1990; Obstfeld, 1994; Devereux and Smith, 1994) address these
issues in the context of endogenous growth.

1
complete market Ramsey framework, a negative wealth shock has some persistence because agents
seek to smooth consumption across time. When markets are incomplete, productive investment
is risky and becomes even less attractive relative to current consumption. This can slow down
convergence to the steady state, and thus increase the persistence of aggregate shocks.
The present paper extends these results in a number of useful directions. First, we investigate
finite horizon economies. Because shocks received at the terminal date cannot be smoothed through
time by borrowing and lending, agents have a strong precautionary motive and accumulate large
aggregate wealth in later periods. The anticipation of these movements generate fluctuations along
the entire equilibrium path in economies with large uninsurable shocks.
Second, under an infinite horizon, multiple steady states can arise from the endogeneity of the
interest rate. In a rich equilibrium, a low interest rate supports a high investment in the risky
technology and a low level of marginal productivity. Conversely in a poor steady state or poverty
trap, a high interest rate motivates agents to choose small amounts of the risky investment. Because
the interest is endogenous, agents may thus prefer to remain poor in equilibrium rather than bear
the large investment risks that are required to increase their standards of living. Poverty traps thus
originate from the capital markets, a source of multiplicity that, to the best of our knowledge, is new
to the literature. In contrast, earlier work has obtained poverty traps from wealth effects, building
on the idea that poorer agents take less risks and can thus be trapped at low wealth levels (e.g.,
Banerjee and Newman, 1991). Our model assumes CARA preferences and therefore eliminates the
influence of wealth on risky investment. The paper thus highlights that, in addition to the wealth
effect, there is another channel through which financial incompleteness may affect the development
of an economy: the impact of the interest-rate on risk taking.
Third, the local dynamics around the steady state have a rich structure under incomplete
markets. Depending on the economy’s parameters, a steady state is locally unique, totally unstable
or locally undetermined, and the equilibrium path can be attracted to a limit cycle. The complicated
dynamics arise even when agents are very patient, suggesting that financial incompleteness is a
useful substitute for technological nonconvexities in models of endogenous business cycles (Boldrin
and Montrucchio, 1986; Benhabib and Farmer, 1994; Mitra, 1996).
Fourth, we show that tractability is preserved under a number of extensions. For instance, we
introduce a storage technology with fixed positive returns, as well as randomness in the depreciation
of productive capital, as in Krebs (2003). We also investigate how the model is affected by multiple
sectors and the structure of the labor market.
Section 2 presents the GEI Ramsey economy. We solve the individual decision problem and
calculate the equilibrium path under a finite horizon in Section 3. In Section 4 we introduce the

2
infinite-horizon economy and examine the comparative statics and local dynamics of the steady
state. Extensions to multiple sectors and labor markets are considered in section 5. We conclude
in Section 6. Unless stated otherwise, all proofs are given in the Appendix.

Literature Review

The paper brings together several branches of the literature. First, it considers the effect of idio-
syncratic technological shocks in the Ramsey growth model developed by Cass (1965), Koopmans
(1965) and Brock and Mirman (1972). When the technology has decreasing returns to scale, the
one-sector model is known to exhibit a unique saddle path in which future wealth is an increasing
function of current wealth. This precludes nonmonotonicities or endogenous fluctuations along the
equilibrium path. For this reason, researchers have considered the optimal growth problem with
nonconvexities in production, and shown that complicated dynamics or endogenous cycles can arise
(Boldrin and Montrucchio, 1986; Deneckere and Pelikan, 1986; Sorger, 1992). These fluctuations
only appear when agents are very impatient.2 For instance, Mitra (1996) proves that a period-three
√
cycle only exists if the psychological discount factor β is less than the constant [( 5−1)/2]2 ≈ 0.38.
By contrast, our GEI growth economy generates deterministic fluctuations with a Cobb-Douglas
technology for large values of the discount factor, such as β = 0.99.
The macro fluctuations exhibited in this paper are also reminiscent of the equilibria observed
in deterministic economies with overlapping generations (Benhabib and Day, 1982; Grandmont,
1985) or credit market imperfections (Galor and Zeira, 1993; Kiyotaki and Moore, 1997; Aghion,
Bacchetta and Banerjee, 1998; Aghion, Banerjee and Piketty, 1999). We contribute to this literature
by showing that incomplete risk-sharing is an alternative source of fluctuations in aggregate output.
This paper thus extends the results obtained in the incomplete-markets exchange economy of Calvet
(2001). While in this earlier model aggregate wealth is fixed by construction, the present paper
shows that idiosyncratic technological shocks can generate endogenous cycles in the productive
sector of the economy.
The recent numerical literature has explored the robustness of the Ramsey growth model to
agent heterogeneity and idiosyncratic uncertainty.3 Aiyagari (1994) explores the effect of individual
exogenous income risk and precautionary savings on the steady-state capital and interest rate,
assuming neither aggregate, nor idiosyncratic technological risk.4 Krusell and Smith (1998) extend
2
Note that it is possible to observe fluctuations in a multi-sector growth model for more patient agents (Benhabib
and Nishimura, 1979).
3
See Ljungqvist and Sargent (2000) for an excellent discussion of these developments.
4
In related work, Scheinkman and Weiss (1986) consider an incomplete market economy with endogenous labor
supply but no capital accumulation. Because agents have a disutility for labor, the poor are more willing to work

3
Aiyagari’s framework to the case of aggregate productivity shocks, but retain the assumption of no
undiversifiable idiosyncratic technological risks. In their model, the behavior of the macroeconomic
aggregates is almost perfectly described by the first moment of the wealth distribution, and the
transitional dynamics are strikingly similar to those of the single-agent Ramsey economy.
Our model differs from these approaches in several respects. First, in earlier work, the aggregate
capital stock and possibly a risk-free bond are the only available assets and cannot be sold short.
By contrast, we assume that agents can trade a riskless asset and an arbitrary number of risky
securities. There are no short-sales constraints, which allows us to disentangle the impact of
borrowing constraints from missing assets or insurance markets. Second, and more importantly, we
consider idiosyncratic technological risk instead of the usual endowment shocks. The introduction
of idiosyncratic rather than aggregate investment risk generates important market inefficiencies.
Third, our model implies closed-form expressions for the aggregate dynamics. Unlike earlier models,
we can calculate the growth paths when the economy has a finite horizon. Finally, our model
generates rich dynamics in financial prices, capital investment and aggregate output.
In general incomplete market formulations, wealth effects typically prevent the existence of an-
alytical solutions and approximate numerical simulations are the only way to proceed. The CARA-
normal specification permits us to overcome this analytical obstacle, but does not seem to invalidate
our result that incomplete markets can generate complicated dynamics. With CRRA preferences
or borrowing constraints, the wealth distribution would be an additional channel through which
idiosyncratic uncertainty affects aggregate behavior, as has been emphasized in the literature (Ga-
lor, 1993; Aiyagari, 1994). Perhaps surprisingly, we show that rich dynamics can arise even in the
absence of such effects.

2 The Model

We consider a neoclassical growth economy with a finite number of heterogeneous agents. Time
is discrete, indexed by t ∈ {0, 1, ...T }. The horizon is either finite (T < ∞) or infinite (T = ∞).
The economy is stochastic and all random variables are defined on a probability space (Ω, F, P).
Individuals are indexed by h ∈ {1, ...H}. They are born at date t = 0, and live and consume a
single consumption good in dates t = 0, ..., T.

than the rich, and aggregate production is higher when the more productive agents have little wealth. Fluctuations
in our model have a profoundly different origin, and arise even though macro variables are always independent of the
distribution of wealth.

4
2.1 Production and Idiosyncratic Risks
Each individual is an entrepreneur who owns his own stock of capital and operates his own pro-
duction scheme. The technology is standard neoclassical, convex, and requires neither adjustment
costs nor indivisibilities in investment. These assumptions would lead to the standard neoclassical
growth model, if it were not for the following: production is subject to (partially undiversifiable)
idiosyncratic uncertainty. An investment of kth units of capital at date t yields
h
yt+1 = Aht+1 F (kth ) + (1 − δ ht+1 )kth

units of the consumption good at date t + 1. The production function F is increasing, strictly
concave, and satisfies the Inada conditions.5 The total factor productivity Aht+1 and the depreciation
rate δ ht+1 are random shocks specific to individual h, which introduce investment risk. They are
the key ingredients of the model.
We find it useful to introduce two additional sources of income. First, producers have access
to a storage technology with gross rate of return ρ ∈ [0, 1]. An investment of sht units of the good
in the storage technology yields ρsht with certainty at date t + 1. Second, entrepreneurs receive
an exogenous stochastic endowment stream {eht }Tt=0 . The additive shock eht+1 models risks that are
outside the control of individuals and do not affect production capabilities. The endowment risk
eht+1 and the productivity shocks (Aht+1 , δ ht+1 ) are all idiosyncratic, but only the latter affect the
return on investment. All shocks are unknown at t and revealed at t + 1.
The idiosyncratic productivity risks (Aht+1 , δ ht+1 ) allow us to capture the impact on growth of a
wide range of technological risks. The uncertainty of an entrepreneurial project obviously influences
specific investment in capital or R&D. Similarly, the riskiness of a worker’s human capital affects
a wide range of decisions, such as the supply of labor,6 education, learning by doing, job search
and career choices. The model can thus be used to analyze how a large class of idiosyncratic risks
influences the accumulation of physical, human or intangible capital, as will be further discussed
in Section 5.

2.2 Asset Structure and Preferences

Individual risks can be partially hedged by trading a limited set of short-lived securities n = 0, .., N .
Each asset n is worth π n,t units of the good at date t, and yields a random amount of consumption
dn,t+1 at date t+1. Security n = 0 is riskless, in the sense that d0,t+1 ≡ 1 with certainty in date t+1.
5
The function F satisfies F 0 > 0, F 00 < 0, F (0) = 0, F 0 (0) = +∞, F (+∞) = +∞, and F 0 (+∞) = 0.
6
See Marcet, Obiols-Homs and Weil (2000) for a recent discussion of the labor supply in the neoclassical growth
context.

5
The quantity Rt ≡ 1/π 0,t denotes the gross interest rate between t and t + 1, and rt = Rt − 1 is the
corresponding net rate. Assets are in zero net supply, there are no short-sale constraints, and default
is not allowed. It is convenient to stack asset prices and payoffs into the vectors π t = (π n,t )N n=0 and
N N
dt+1 = (dn,t+1 )n=0 . Without loss of generality, we assume that (dn,t+1 )n=0 is an orthonormal family
of L2 (Ω), implying that risky assets have zero expected payoffs and are mutually uncorrelated. At
the outset of every period t, investors are informed of the realization of the asset payoffs dt and
idiosyncratic shocks {(Aht , δ ht , eht )}H 7
h=1 . Information is thus symmetric across agents and generates
a filtration {Ft }Tt=0 . Conditional on the information available to her, agent h selects a portfolio
θht = (θhn,t )N
n=0 in period t.
We assume by construction that all the assets traded in one period are short-lived, in the sense
that they only deliver payoffs in the next period. In the next sections, we will show that there is
no risk premium in equilibrium, and that the interest rate sequence {Rt }0≤t<T is deterministic and
known in advance to all agents. For this reason, equilibrium allocations and prices do not change
if we introduce a perpetuity, namely a security delivering one unit of the good every period.8 The
P −1
perpetuity is worth Πt = Ts=t 1/(Rt ...Rs ) at date t after delivery of the period’s coupon.
P
Individual preferences are given by U0h = E0 Tt=0 β t u(cht ), where β ∈ (0, 1) is the discount
factor and u is the utility function. The agent maximizes expected lifetime utility subject in all
date-events to the budget constraints

cht + kth + sht + π t · θht = wth

(1)
h
wt+1 = eht+1 + Aht+1 F (kth ) + (1 − δ ht+1 )kth + ρsht + dt+1 · θht

The variable wth , called wealth, represents the trader’s net credit or debt position at date t. It is
independent of the income expected in later periods. When time is finite, we use the convention
that shT = 0 and θhT = 0. Under an infinite horizon, we impose the transversality condition
limt→∞ E0 β t u(wth ) = 0.

2.3 CARA-Normal Specification

Preferences are assumed to exhibit constant absolute risk aversion:
1
u(c) = − exp(−Γc).
Γ
7
The results of this paper would not be modified under the weaker assumptions that income shocks are privately
observed and the structure of the economy is common knowledge. We use a stronger assumption in the text to
emphasize that the endogenous fluctuations and multiple equilibria observed in equilibrium do not originate from
information asymmetries.
8
More generally, traders can dynamically replicate a large class of long-lived risky assets.

6
 ¡ ¢
The asset payoffs dt+1 and the idiosyncratic shocks { Aht+1 , δ ht+1 , eht+1 }h are jointly normal and
independent of past information Ft . There is thus no distinction between the conditional and un-
conditional distribution of these variables. The (unconditional) distribution of payoffs and shocks
can vary through time, which in applications may prove useful to capture the dynamic effect of
financial innovation or business cycle variations in uninsurable risk (e.g. Mankiw, 1986; Constanti-
nides and Duffie, 1996). We obtain a tractable model under the following restrictions. Idiosyncratic
shocks have a constant mean:

A ≡ EAht+1 > 0, δ ≡ Eδ ht+1 , e ≡ Eeht+1 .

For all h and t, we project the idiosyncratic risks on the asset span available at date t:
P eh
Aht+1 = A + N h
n=1 β A,n dn,t+1 + At+1 ,
P eh
δ ht+1 = δ + N h
n=1 β δ,n dn,t+1 + δ t+1 ,
P
eht+1 = e + N h
n=1 β e,n dn,t+1 + eeht+1 .

We observe that for all x ∈ {A, δ, e}, the coefficient β hx,n coincides with Cov(xht+1 , dn,t+1 ). The resid-
h
eh , e
uals (A δ , eeh ) represent the undiversifiable component of the investment and endowment
t+1
t+1 t+1
shocks to individual h. They are assumed to be mutually uncorrelated, implying
 h    
Ae σ 2A,t+1 0 0
t+1
 eh    
 δ t+1  ∼ N 0,  0 σ 2δ,t+1 0  .
eeht+1 0 0 σ 2e,t+1

The quantities σ t+1 = (σ A,t+1 , σ δ,t+1 , σ e,t+1 ) denote the (conditional and unconditional) standard
deviations of uninsurable risks. They are useful measures of financial incompleteness, which can
deterministically vary through time. We assume that the vector σ t+1 is the same for all agents.
The CARA-normal specification will ensure that, given a deterministic sequence of price, aggregate
quantities are independent of the wealth distribution. This will overcome the curse of dimensionality
that arises when the distribution of wealth — an infinite-dimensional object — enters the state space
of the economy.
We rule out aggregate risk by imposing in every period the cross-sectional restriction:
 h   
H A A
1 X h   t+1

 δ t+1  =  δ  .
H
h=1 eht+1 e

This assumption provides tractability, and illustrates that the endogenous fluctuations observed in
equilibrium do not arise from aggregate shocks.

7
2.4 Equilibrium
For any quotation {π t }Tt=0 of asset prices, each agent chooses a contingent plan {cht , kth , sht , θht , wth }Tt=0
adapted to the filtration {Ft }Tt=0 .

Definition 1 An equilibrium consists of a price sequence {π t }Tt=0 and a collection of individual

plans ({cht , kth , sht , θht , wth }Tt=0 )1≤h≤H such that:
(i) Given prices, each agent’s plan is optimal.
P
(ii) Asset markets clear in every date-event: H h
h=1 θ t = 0.

We make two observations. First, the absence of arbitrage implies that Rt ≥ ρ along any equilibrium
path. This is because if Rt < ρ, agents want to borrow an infinite amount at rate Rt and invest it
in the storage technology. Second, the absence of aggregate risk in the CARA-normal framework
implies that deterministic price sequences {π t }Tt=0 can be obtained in equilibrium. We henceforth
focus on price sequences satisfying these two conditions.

3 Equilibrium under a Finite Horizon

3.1 Decision Theory
Consider an agent h facing an exogenous, deterministic path of asset prices {π t }Tt=0 satisfying the
no-arbitrage condition Rt ≥ ρ in every period. We denote by J h (w, t) the indirect utility of wealth
along the price path as of date t. The sequence {J h (., t)}Tt=0 satisfies the Bellman equation:

J h (wth , t) = max u (c) + βEt J h (wt+1

h
, t + 1)
{ch h h h
t ,kt ,st ,θ t }

subject to the budget constraints (1). Since J h (w, T ) ≡ u(w), the value functions J h (., t) can be
solved by a backward recursion.
The results are conveniently presented using

Φ(k) ≡ AF (k) + (1 − δ)k + e,

G(k, a, σ) ≡ Φ(k) − Γa[F (k)2 σ 2A + k 2 σ 2δ + σ 2e ]/2.

The function Φ(k) denotes the expected non-financial wealth when investing k units of capital.
In contrast, G(k, a, σ) represents the risk-adjusted level of non-financial wealth when productive
investment is k, the marginal propensity to consume next period is a, and the residual risks are
σ = (σ A , σ δ , σ e ) .

8
Lemma 1 The value function belongs to the CARA class in every period and satisfies

Jwh (w, t) = exp[−Γ(at w + bht )].

The coefficients at and bht are defined recursively by the terminal conditions aT = 1, bhT = 0, and
the equations
1
at = , (2)
1 + (at+1 Rt )−1
· ¸
h at h h h ln(βRt )
bt = b + at+1 G(kt , at+1 , σ t+1 ) − at+1 Rt kt −
at+1 Rt t+1 Γ
XN · ¸
Rt π n,t
+ at π n,t β he,n + β hA,n F (kth ) − β hδ,n kth + .
n=1
2Γat+1

The optimal capital stock satisfies

( N
)
X
kth ∈ arg max G(k, at+1 , σ t+1 ) − Rt k + Rt π n,t [β hA,n F (k) − β hδ,n k] . (3)
k≥0
n=1
PN h
It is strictly positive if and only if A + Rt n=1 π n,t β A,n > 0. Individual consumption and portfolio
choices are determined by

cht = at wth + bht ,

h
θn,t = −β he,n − β hA,n F (kth ) + β hδ,n kth − Rt π n,t /(Γat+1 ), ∀n ∈ {1, ..., N }, (4)
P
θh0,t + ρsht = (at /at+1 )wth − Rt bht − Rt kth − Rt N h
n=1 π n,t θ n,t .

The agent does not use the storage technology if Rt > ρ, and is indifferent between the bond and
storage if Rt = ρ

The slope of the consumption function at increases with the future slope at+1 and the current
interest rate Rt . Forward iteration of (2) implies that at = 1/(1 + Πt ). The intercept bht decreases
with the standard deviations of uninsurable risks σ t+1 , an implication of the precautionary motive
for savings (Leland, 1968; Sandmo, 1970; Caballero, 1990).

3.2 Equilibrium Characterization

Let Ct , Wt , Kt , and St denote the population average of consumption, wealth, capital and storage
P
investment in a given date t. The initial wealth H h
h=1 w0 /H is an exogenous parameter of the
economy, which is denoted by W 0 .

9
 The economy is assumed to have no aggregate risk. We therefore focus on equilibrium paths in
which the risk premium is zero.9 Risky assets, which are normalized to have zero expected payoffs,
are therefore costless: π n,t = 0 for all n ∈ {1, ..., N } and t ∈ {0, ..., T }. Individual productive
investment then maximizes
G(k, at+1 , σ t+1 ) − Rt k. (5)
All agents have the same marginal propensity to consume at = 1/(1 + Πt ) and the same objective
function (5). We can therefore choose the same capital stock for all agents: kth = Kt for all h.
Aggregating across the population, we can derive the equilibrium dynamics in closed form.

Proposition 1 The aggregate equilibrium dynamics are deterministic and the risk premium on
financial assets is zero. All agents have identical marginal propensities to consume and choose
identical levels of investment. Macroeconomic aggregates satisfy the recursion

Kt ∈ arg max {G(k, at+1 , σ t+1 ) − Rt k} (6)

k≥0

Wt+1 = Φ(Kt ) + ρSt (7)

at = 1/[1 + (at+1 Rt )−1 ] (8)
£ ¤
Ct+1 − Ct = ln(βRt )/Γ + Γa2t+1 F (Kt )2 σ 2A,t+1 + Kt2 σ 2δ,t+1 + σ 2e,t+1 /2 (9)
Wt = Ct + Kt + St (10)

for all t ∈ {0, ..., T }, with the boundary conditions W0 = W 0 , aT = 1, and KT = 0.

Consumption growth Ct+1 − Ct increases with the variance of future consumption V art (cht+1 ) =
a2t+1 [F (Kt )2 σ 2A,t+1 + Kt2 σ 2δ,t+1 + σ 2e,t+1 ]. This is a well-known consequence of the precautionary
motive.
Since the mean productivity A is positive, the optimal capital stock is always positive and
satisfies the FOC:
Rt = Gk (Kt , at+1 , σ t+1 ). (11)
In the absence of undiversifiable productivity risk (σ A,t+1 = σ δ,t+1 = 0), the FOC implies that the
marginal product of capital is equated with the interest rate: Rt = 1 − δ + AF 0 (Kt ).
When technological risks cannot be fully hedged (σ A,t+1 > 0 or σ δ,t+1 > 0), the interest rate is
equal to the risk-adjusted return to capital:

Rt = 1 − δ + AF 0 (Kt ) − Γat+1 [F (Kt )F 0 (Kt )σ 2A,t+1 + Kt σ 2δ,t+1 ]. (12)

9
We show in the Appendix that the risk premium is zero in any equilibrium when the economy contains a unique
type of idiosyncratic investment risk (i.e. only productivity risk or only depreciation risk).

10
We interpret the additional term Γat+1 [F (Kt )F 0 (Kt )σ 2A +Kt σ 2δ ] as the risk premium on investment.
Under incomplete markets, risk aversion creates an interaction between the current investment Kt
and the future marginal propensity at+1 .
Although the production function F (k) is strictly concave, the risk-adjusted function G(k, a, σ)
need not be concave when σ A > 0. As a result, the FOC is necessary but not sufficient in general —
indeed, equation (11) may admit multiple solutions. The optimal capital stock is easily determined
under

Assumption 1 One of the following conditions holds: (i) The function F (k)2 is convex; (ii) There
is no idiosyncratic depreciation risk and the production function is Cobb-Douglas: F (k) = k α ,
0 < α < 1; (iii) There is no idiosyncratic depreciation risk and ρ ≥ 1 − δ.

We can then show

Lemma 2 Under Assumption [1], the optimal capital stock kth is the smallest positive solution to
FOC condition (11).

We henceforth assume that Assumption [1] holds. In general, the optimal capital stock need not
be the first solution to (11), and switches between solutions may be a potential source of rich
dynamics. We find it more striking to exhibit fluctuations in the absence of such effects.

3.3 Equilibrium Analysis

Equilibrium paths can be calculated by a backward recursion. To simplify the analysis, we
now specialize to a stationary economy. Consider the complete macroeconomic vector xt =
(at , Ct , Wt , Kt , St , Rt ) and the reduced vector zt = (at , Ct , Wt ). We easily show

Lemma 3 There exist unique mappings H and H e such that the sequence {at , Ct , Wt , Kt , St , Rt }T
t=0
is an equilibrium if and only if, for all t,

zt = H (zt+1 ) and e t+1 )

xt = H(z (13)

where zt = (at , Ct , Wt ) ∈ (0, 1] × R × [e, +∞) and xt = (Kt , St , Rt ) ∈ R2+ .

Proof. Let Kt∗ = min {k : Gk (k, at+1 , σ t+1 ) = ρ} and Wt+1

∗ = Φ(K ∗ ). Any solution x can be cal-
t t
culated by following procedure. If Wt+1 < Wt+1 ∗ , equations (6) and (7) imply that K = Φ−1 (W
t t+1 ),
∗
St = 0, and Rt = Gk (Kt , at+1 ) > ρ. On the other hand if Wt+1 ≥ Wt+1 , we infer from (6) and (7)
that Rt = ρ, Kt = Kt∗ , and St = [Wt+1 − Φ(Kt )]/ρ. Equations (8) − (10) then assign unique values

11
to at , Ct and Wt . Finally, we observe that (at , Kt , St , Rt ) ∈ (0, 1) × R∗+ × R+ × [ρ, +∞), while no
simple restrictions seem to be imposed on current consumption and wealth. ¥

The equilibrium dynamics are thus fully characterized by the three-dimensional vector zt and
the equilibrium calculation is similar to the method used for the standard complete-markets Ramsey
model.10 At terminal date T , individuals consume all their wealth (chT = wTh ), and the state variables
satisfy the relations aT = 1 and CT = WT . For any choice of WT ∈ [e, +∞), we let zT = (1, WT , WT )
and recursively calculate the implied path zt−1 = H(zt ). Note that the condition Wt+1 ∈ [e, +∞)
does not guarantee that Wt ∈ [e, +∞). Some terminal wealth levels WT imply that Wt < e and
the algorithm stops at some instant t > 0. Other values WT generate an entire path {zt }Tt=0 such
that Wt ≥ e for all t, leading to some initial wealth level W0 . The recursion thus defines a function
W0 (WT ), whose domain is a subset of [e, +∞). Since the initial wealth W 0 ∈ R is exogenously fixed,
an equilibrium corresponds to a terminal wealth level WT such that W0 (WT ) = W 0 . We show in
the Appendix

Proposition 2 There exists an equilibrium in any finite-horizon economy.

The intuition of the proof is to extend the function W0 (·) to a continuous mapping defined on
[e, ∞). The behavior of the function W0 (·) when WT → e and WT → +∞ helps to conclude.
We now investigate the properties of the equilibrium growth path. When uninsurable risks are
relatively small, we may expect that the GEI equilibrium remains relatively close to the complete-
market growth path. In fact, the GEI growth path combines two features: a turnpike property
and a strong precautionary effect around the terminal date. Starting from a low level of initial
wealth, the economy accumulates wealth and remains many periods in the neighborhood of the
GEI steady state.11 This is evident in Figure 1, which plots log(Wt /W∞ ), the natural logarithm
of the ratio between equilibrium wealth Wt and the steady-state level W∞ . We choose an economy
with T = 1, 000 periods and parameters Γ = 10, A = 1, β = 0.95, α = 0.40, δ = 0.05, ρ = 1 − δ,
e = 0, σ δ = σ e = 0, and σ A = 0.50. In the traditional Ramsey model, aggregate wealth progressively
decreases around the terminal date. Under incomplete markets, however, the last shocks cannot
be easily smoothed by borrowing and lending, and individuals have a very strong precautionary
motive before the terminal date. As a result, aggregate wealth overshoots the steady state before
10
We note that the dimensionality of our system is influenced by the financial structure and the aggregate produc-
tivity of capital. More specifically, our three-dimensional model reduces to: a two-dimensional system in (Ct , Wt )
when A > 0 and markets are complete; a one-dimensional system in at when agents do not produce (A = 0, δ = 1,
ρ = 0).
11
We discuss the steady state of the infinite-horizon economy in the next sections.

12
being consumed in the very last periods. These results thus contrast with finite-horizon complete
market dynamics, in which the capital stock never exceeds its steady-state level.
With a large degree of market incompleteness, the growth paths of the economy can be start-
ingly different from the traditional complete market model. The differences originate in the non-
monotonicities embedded in the equilibrium recursion. Consider for instance the impact on current
wealth of a higher future wealth level Wt+1 in the absence of storage (Wt+1 < Wt+1 ∗ ). Higher wealth

next period requires higher investment Kt = Φ−1 (Wt+1 ) in the current period. The interest rate
Rt is therefore lower by equation (11), which leads to an increase in the component − ln(βRt )/Γ of
current consumption:
£ ¤
Ct = Ct+1 − ln(βRt )/Γ − Γa2t+1 F (Kt )2 σ 2A + Kt2 σ 2δ + σ 2e /2. (14)

Under complete markets, this is the only effect. Current consumption Ct and wealth Wt = Ct + Kt
are increasing functions of future wealth Wt+1 . The function W0 (·) monotonically increases in WT
and there exists a unique equilibrium growth path. More generally, we show in the Appendix that
the monotonity of W0 (·) and equilibrium uniqueness hold when there are only endowment shocks
(σ e ≥ 0, σ A = σ δ = 0).
Under uninsurable production risks, however, a high level of output next period corresponds
to a high level of capital investment and individual risk-taking in the current period. Agents thus
have a strong precautionary motive, which can lead them to reduce current consumption (14). As
a result, current consumption Ct and wealth Wt = Ct + Kt can be decreasing functions of future
wealth Wt+1 . The anticipation of high wealth in the future implies a precautionary reduction
in current consumption and possibly lower wealth today. The recursion mapping may thus be
non-monotonic in future wealth.
These non-monotonicities have two possible implications: multiple equilibria and complicated
dynamics, which we succesively examine. For instance, Figure 2 illustrates the function W0 (WT )
for appropriate values of the economy’s parameters. We observe that W0 (WT ) is non-monotonic
and thus infer

Proposition 3 There robustly exist multiple equilibrium paths in some finite-horizon economies.

Given an initial wealth level W 0 , agents can coordinate on several paths. For instance, they can
choose a high level of risky investment, which reduces current consumption for precautionary rea-
sons. Conversely, they can coordinate on a low investment level, which is matched by a reduction
in the precautionary motive and a higher level of current consumption. Under uninsurable produc-
tion risks, two economies with identical technologies, preferences and financial structures can thus
exhibit different paths of capital accumulation, savings and interest rates.

13
 Complicated dynamics can also arise. To simplify the discussion, we consider an economy with
a single equilibrium, and calculate the unique growth path using the equilibrium recursion. Figure 3
displays an economy with large endogenous fluctuations. This graph illustrates the kind of complex
dynamics that the introduction of incomplete markets may generate in an otherwise standard
neoclassical economy. Such behaviors do not necessarily require a high degree of impatience. For
instance in Figure 3, we obtain endogenous fluctuations along the unique equilibrium path for a high
discount factor (β = 0.99) and a low rate of capital depreciation (δ = 0.1). Note that endogenous
cycles and non-monotonicities require substantial undiversifiable idiosyncratic risks in these two
examples.
Endogenous fluctuations can be usefully understood in terms of interest rate dynamics. Agents
coordinate their capital investment through the bond market. Assume that the interest rate next
period Rt+1 is anticipated to be high. Agents have a high marginal propensity to consume at+1
and are thus more sensitive to investment risk (“risk aversion effect”). They thus choose a low level
of current investment Kt . Low investment Kt and high marginal propensity at+1 have conflicting
effects on the current interest rate Rt = Gk (Kt , at+1 , σ). When uninsurable production risks are
large, the risk aversion effect dominates and the interest rate is low next period. There can thus be
a negative feedback between future and current rates. As can be seen in Figure 3, the presence of
large uninsurable risks can imply nonlinear comovements between productive investment, interest
rates, consumption and wealth along the equilibrium path.
These results show that market incompleteness helps obtain endogenous fluctuations with more
patient agents than in earlier models considered in the literature. For instance, large investor
impatience is required to obtain cycles in one-sector growth models with non-convex technologies
(Boldrin and Montrucchio, 1986; Deneckere and Pelikan, 1986; Sorger, 1992; Mitra, 1996). The
results of Figure 3 and Section 4 indicate that financial incompleteness may be a useful substitute for
investor impatience in the calibration of these stationary economies. Similarly, incomplete markets
could help generate fluctuations in endogenous growth models (Benhabib and Farmer, 1994).

4 Equilibrium with Infinite Horizon

We now extend the analysis to infinite horizon economies.

4.1 Decision Theory and Equilibrium

Given a deterministic sequence of asset prices {π t }+∞
t=0 , we calculate the optimal decision of an indi-
vidual investor by taking the pointwise limit of the finite horizon problem. In every period, let Πt =

14
P+∞
s=t+1 1/(Rt ..Rs−1 ) denote the exogenous price of a perpetuity, at = (1 + Πt )−1 the implied mar-
ginal propensity to consume, and kth the smallest solution to equation (11). We also consider Mth =
P £ h ¤
G(kth , at+1 , σ t+1 )−kth −(Γat+1 Rt )−1 ln(βRt )+ N h h h h
n=1 π n,t β e,n + β A,n F (kt ) − β δ,n kt + Rt π n,t /2Γat+1 ,
and à !
+∞
X h
M s
bht = at Mth + .
s=t+1
R t ...R s−1

The convergence of the series at and bht is guaranteed by the following sufficient conditions:

Assumption 2. The sequences {π t }+∞ +∞ +∞

t=0 , {Rt }t=0 and {Πt }t=0 are bounded.

We can the show

Lemma 4 Under Assumption [2], the value function belongs to the CARA class every period:
J(W, t) = −(Γat )−1 exp[−Γ(at W + bht )], and the consumption-portfolio decision satisfies (4).

Since the optimal decision is the limit of the finite horizon problem, Ponzi schemes are ruled out
in the strongest conceivable way, along any possible path.
We now let σ t be invariant over time. As in the finite-horizon case, we focus on GEI equilibria
in which the prices are deterministic, there is no risk premium, capital investment is the same
across agents, and assumption [2] holds. It is then straightforward to characterize the dynamics of
the vector zt = (at , Ct , Wt ).

Proposition 4 The sequence {zt }∞

t=0 satisfies the recursion zt = H(zt+1 ) in every period.

Furhermore, Assumption [2] implies that the sequence at = 1/(1 + Πt ) is bounded away from 0.
Conversely, consider a sequence {zt }∞
t=0 satisfying equilibrium recursion (13) and the condition
inf t at > 0. The perpetuity price Πt = 1/at − 1, the corresponding interest rate Rt and the price
sequence π t = (1/Rt , 0, ..0) are bounded across t. Lemma 4 implies that each agent has a unique
optimal plan. We then easily check that markets are clearing in every date-event. Thus, any
sequence {zt } satisfying the condition inf t at > 0 and the recursion zt = H(zt+1 ) defines a GEI
equilibrium. We henceforth focus on infinite horizon paths with these properties.

4.2 Steady State

A steady state is a fixed point of the dynamical system (6) − (10), or equivalently a fixed point of
the recursion mapping H. Assumption [2] implies that the gross interest rate is larger than unity:
R∞ > 1. Storage yields by assumption a lower rate of return (ρ ≤ 1 < R∞ ), and agents never find

15
it profitable to allocate savings to this low-yield technology. We infer that storage is not used in
the steady state (S∞ = 0). It is then easy to show

Proposition 5 In a steady state, the interest rate and capital stock satisfy
£ ¤
ln(R∞ β) = −Γ2 (1 − R∞ ) F (K∞ )2 σ 2A + K∞ 2 σ 2δ + σ 2e /2
−1 2
(15)
£ ¤
R∞ = 1 − δ + AF 0 (K∞ ) − Γ(1 − R∞ −1
) F 0 (K∞ )F (K∞ )σ 2A + K∞ σ 2δ (16)

The first equation corresponds to the stationarity of consumption, and the second to the optimality
of the capital investment K∞ . We note that R∞ ≤ 1/β, and that the upper bound 1/β is reached
when markets are complete. This result is a possible solution to the low risk-free rate puzzle, as
has been extensively discussed in the literature (e.g. Weil, 1992; Aiyagari, 1994; Constantinides
and Duffie, 1996; Heaton and Lucas, 1996).
The system (15) − (16) allows us to analyze existence, multiplicity and comparative statics of
the steady state. We show in the Appendix:

Proposition 6 There exists a steady state in every economy.

System (15)−(16) implicitly defines two decreasing functions K1 (R) and K2 (R). A simple continuity
argument provides existence.
The system can be used to derive the comparative statics of the steady state with respect to
the economy’s parameters. As can be seen in equation (15), the idiosyncratic risks σ e , σ A and σ δ all
increase the precautionary demand for savings, which tends to reduce the interest rate and increase
capital investment. This implies that the capital stock K∞ increases with the endowment risk σ e ,
as in Aiyagari (1994). We note, however, that uninsurable technological shocks σ A and σ δ also
reduce the risk-adjusted return to investment and can thus discourage capital accumulation. These
conflicting effects lead to a non-monotonic relation between capital and idiosyncratic technological
shocks σ A or σ δ in some economies. While better insurance against endowment risk always reduces
output through the precautionary effect, aggregate activity may actually be increased by new
hedging instruments for technological risks.
We easily check that the steady state is unique when markets are complete, or more generally
when agents are exposed only to uninsurable endowment shocks (σ e ≥ 0, σ A = σ δ = 0). In the
presence of productivity shocks, however, multiple steady states can arise.12

Proposition 7 There robustly exists a multiplicity of steady states in some economies.

12
It is easy to show that the number of steady states is generically odd.

16
A steady state with a high interest rate has low capital and aggregate wealth, and can be interpreted
as a poverty trap. In earlier work (e.g. Banerjee and Newman, 1991), poverty traps originate in a
wealth effect: poor agents are unwilling to take risk and make large investments in the productive
technology. In our model, investment is independent of individual wealth and the poverty trap
arises from the capital markets. When the endogenous interest rate is high, investors are unwilling
to invest large amounts in the high-yield risky technology and the economy is stuck at a low wealth.
The endogeneity of the interest rate thus leads to a new type of poverty trap.
The multiplicity of steady states appears to be a natural implication of the dynamic com-
plementarity in investment developed in Angeletos and Calvet (2003). The anticipation of large
investment in the future can feed back into low effective risk aversion and large investment in
the present. This type of complementarity is likely to produce multiple equilibria, as has been
emphasized in the literature on macroeconomic complementarities. While we did not explore this
possibility in our earlier work, Proposition 7 establishes that this effect can impact the number of
steady states. They also affect the local dynamics around the steady state, as is now discussed.

4.3 Local Dynamics

The local dynamics around the steady state can be examined by linearizing the recursion mapping
H defined by (6) − (10). The calculation, which is derived in the Appendix, is simplified by the fact
that there is no storage at the steady state. We observe that wealth Wt is the only predetermined
variable in the model, while consumption and marginal propensity are free to adjust. When markets
are complete, the stable manifold has dimension 1, the system is determined, there is a unique steady
state, and convergence to the steady state is monotonic. None of these results need hold under
incomplete markets.
Let #ST denote the number of steady states. We begin by considering economies with #ST =
1.

Proposition 8 When #ST = 1, the steady state is either locally determined or locally undeter-
mined depending on the parameters of the economy.

We show this property by computing numerically the eigenvalues of the local dynamics in some
selected examples. Indeterminacy is obtained in an economy with patients agents (e.g. β = 0.95)
and large idiosyncratic production risk. We also note that it is not easy to rule out this form of
indeterminacy, because there is no obvious focal equilibrium on which agents can coordinate.
We provide in the Appendix an example of a supercritical flip bifurcation as σ A increases. This
result is obtained in a class of economies with fixed psyschological discount factor β = 0.95. We

17
infer that cycles of period 2 exist on an open set of economies.

Proposition 9 Some economies robustly contain attracting cycles.

Note that these cycles can arise in economies with reasonably patient investors (β = 0.95). Con-
sistent with the results of Section 3, this suggests that financial incompleteness may help to obtain
cycles under less heavy discounting than is required in nonconvex complete-market production
economies (Boldrin and Montrucchio, 1986; Deneckere and Pelikan, 1986; Sorger, 1992). More
generally, these results illustrate the kind of complicated equilibrium dynamics that financial in-
completeness generates in the neoclassical Ramsey framework.
We now examine economies with multiple steady states. For the case #ST = 3, the Appendix
provides examples in which the equilibria with the lowest and highest interest rate exhibit local
uniqueness, while the middle steady state is totally unstable.

Proposition 10 When #ST > 1, some steady states can be totally unstable.

No growth path leads the economy to such a steady state.

Several mechanisms help to explain the rich local properties uncovered in this subsection. First,
we discussed in the finite-horizon section that endogenous fluctuations can arise from a negative
feedback between future and current interest rates. Agents can then coordinate on multiple interest
rate paths, as in Calvet (2001). Second, the dynamic complementarity between future and current
investment (Angeletos and Calvet, 2003) is also a powerful source of multiple equilibria. While a
precise decomposition of these effects is unavailable for our nonlinear system, we conjecture that
both effects contribute to indeterminacy in infinite horizon economies. Intuition also suggests that
the negative feedback helps to generate endogenous cycles, while the complentarities in investment
induce multiple steady states. More generally, all these results emphasize the role of the bond mar-
ket for the coordination of expectations and investment in economies with decentralized production
and uninsurable technological risks.

5 Extensions
5.1 Endogenous Labor Supply
We previously abtracted from the possibility that entrepreneurs may adjust their productive effort
as a function of realized productivity. We now endogenize the labor supply and show that the
CARA-normal framework remains tractable under appropriate specifications of the production
function and the disutility of labor.

18
 Let lth denote the effort agent h exerts during period t. Output is given by

yth = Aht F (kt−1

h
, lth )

where F is a neoclassical production function. Preferences are given by

1 n h io
u(cht , lth ) = exp −Γ cht − v(lth ) ,
Γ
where v(l) is a strictly convex and increasing function representing the disutility of effort. Agents
choose investment before the realization of the idiosyncratic risks, but can adjust their effort after
the fact. It follows that
n o
lth = arg max Aht F (kt−1
h
, l) − v(l)
l

That is, the optimal effort maximizes output net of the disutility of effort.
£ ¤
h )α 1/(2−η) and therefore
Assume F (k, l) = kα lη and v (l) = l2 /2. It follows that lth = ηAht (kt−1

yth − v(lth ) = Zth f (kt−1

h
)

where Zth ≡ (Aht )ζ , f (k) ≡ ξk αζ , ξ ≡ η η/(2−η) /ζ and ζ ≡ 2/(2 − η). The present model is thus
isomorphic to the benchmark model, with Zth and f (.) replacing Aht and F (.). We conclude that the
CARA-normal structure is preserved if we assume that Zth is normal. The equilibrium dynamics
are then the same as in the benchmark model when output and consumption are interpreted as net
of the disutility of labor.

5.2 Physical and Human Capital

Let us now abstract again from endogenous labor supply but allow agents to invest in two types
of capital, which we interpret as physical and human capital. The question of interest is how
idiosyncratic risks affect the intensity of human capital and what are the implications for production
efficiency.
Let k be the amount of physical capital invested and x the amount of human capital. The
output for agent h is now given by

yth = Aht Fj (kt−1

h
, xht−1 ) + (1 − δ hkt )kt−1
h
+ (1 − δ hxt )xht−1

where δ hkt and δ hxt are the depreciation rates of physical and human capital, respectively. To sim-
plify notation, and without any serious loss of generality, we assume that there is no exogenous

19
endowment and there are no financial assets other than the riskless bond. The budget constraint
for agent h is thus given by

cht + kth + xht + θht /Rt = wth

, (17)
h
wt+1 = Aht+1 F (kth , xht ) + (1 − δ hk,t+1 )kth + (1 − δ hx,t+1 )xht + θht

where θht denotes bond holding in period t. Preferences remain as in the benchmark model. We
finally let
(Aht , δ hkt , δ hxt ) ∼ N (1, Σ)
where  
σ 2A 0 0
 
Σ =  0 σ 2k 0  .
0 0 σ 2x
σ A measures risk common to both types of capital, whereas σ k and σ h measure risks specific to
physical and human capital, respectively.
Define the risk-adjusted output as follows:
Γa h i
G (k, x, a) ≡ F (k, x) − F (k, x)2 σ 2A + k 2 σ 2k + x2 σ 2x
2
We can easily show that, in equilibrium,13

(Kt , Xt ) ∈ arg max {G (k, x, at+1 ) s.t. k + x = Kt + Xt } .

(k,x)

That is, the allocation of savings between physical and human capital maximizes the risk-adjusted
wealth of the agents. In general, this is different than the allocation that maximizes output, which is
what happens in the first best (complete markets). That is, incomplete markets lead to inefficiency
in the allocation of savings in the sense that

(Kt , Xt ) ∈
/ arg max {F (k, x) s.t. k + x = Kt + Xt } .

In other words, agents equate the risk-adjusted returns of different investment options, implying a
gap between the marginal products of physical and human capital:

Fx (Kt , Xt ) − Fk (Kt , Xt ) = Γat+1 (σ 2x Xt − σ 2k Kt ) (18)

Note that, unlike the type-specific risks σ k and σ x , the production risk σ A does not affect the
allocation of savings between physical and human capital, precisely because this risk is common to
both types of capital.
13
This follows from the fact that (Kt , Xt ) maximize G (k, x, at+1 ) − Rt k − Rt x.

20
 To see most clearly the effect of incomplete markets on the intensity of human capital, let the
production function be Cobb-Douglas:

F (k, x) = kα xη ,

where α > 0, η > 0 and α + η < 1. We can then show

Proposition 11 The equilibrium allocation satisfies Xt /Kt 6= η/α and

F (Kt , Xt ) < max {F (k, x) s.t. k + x = Kt + Xt }

(k,x)
p p
if and only if σ k /σ x 6= η/α. In particular, Xt /Kt < η/α if σ k /σ x < η/α and Xt /Kt > η/α if
p
σ k /σ x > η/α.

The ratio α/η is the ratio Kt /Xt that characterizes the first-best (complete-markets) allocation
of savings between physical and human capital. The production risk σ A , being common for both
types of capital, does not affect the allocation of savings. But the type-specific risk σ k and σ x do.
Ceteris paribus, an increase in σ k shifts savings from physical to human capital, whereas the reverse
is true for an increase in σ x . The effect of such a shift on productivity depends on the ratio σ k /σ x .
The allocation of savings remains efficient under incomplete markets if and only if the type-specific
risks are “balanced” in that σ k /σ x takes a critical value. If σ k /σ x is below this value, an increase
in σ k will actually increase output for any level of savings, whereas an increase in σ x will decrease
productivity.
The model of human capital we have used here is very simplistic. The implications of labor-
income risk and other shorts of human-capital risk for productivity and growth is an issue that
deserves a separate paper on its own.

5.3 Multiple Sectors

We now consider the possibility that agents have access to multiple technologies. Like the previous
example, agents have multiple investment opportunities; but unlike the previous example, there
are many technologies and each technology uses its own specific capital. We interpret this set up
as a multi-sector economy and examine how production risks affect the cross-sectoral allocation of
savings.
There are J technologies indexed by j ∈ {1, ..., J}. The output of technology j for agent h in
period t is given by
h
yj,t = Ahj,t Fj (kj,t−1
h
)

21
 h
where kj,t−1 is the amount of capital agent h invests in technology j in period t − 1, Ahj,t is an
idiosyncratic and technology-specific shock, and Fj has positive and diminishing marginal product,
for all j. To simplify notation, we assume that depreciation is full, that the endowment shock is
zero, that there is no storage, and that there are no financial assets other than the riskless bond.
The budget constraint is thus given by
Pj h
cht + h
j=1 kt + θ t /Rt = wt
h

h
P j h h h
wt+1 = j=1 Ajt+1 Fj (kjt ) + θ t

Finally, we let the shocks Aht = (Ahjt )j=1,...,J be jointly normal, with
¡ ¢
Aht ∼ N A, Σt ,

where A = 1 and  
σ2 ... 0
 .1t .. .. 
Σt = 
 .
. . .  .
0 ... σ 2Jt
σ jt ≥ 0 thus measures the risk in sector j. For example, if J = 2 and σ 1t > 0 = σ 2t , we can interpret
sector 1 as private equity and sector 2 as public equity.
Define the risk-adjusted output of technology j as follows:
Γa
Gj (kj , a) = Fj (kj ) − Fj (kj )2 σ 2j
2
We can easily show that, in equilibrium,
h
kj,t = Kj,t ∈ arg max {Gj (k, at+1 ) − Rt k} .
k

for all j, h, and t. The first-order condition for the above gives14
∂
Gj (Kj,t , at+1 ) = Rt .
∂k
That is, agents equate the interest rate with the risk-adjusted return of capital in every sector.
By implication, the cross-sectoral allocation of capital is inefficient in the sense that it does not
maximize aggregate output:

Proposition 12 For almost every {Rt , Σt }∞

t=0 , the allocation of capital satisfies
 
XJ X J XJ 
Fj (Kj,t ) < max Fj (kj ) s.t. kj = Kt
(kj )  
j=1 j=1 j=1
PJ
where Kt = j=1 Kj,t .
14
This FOC is always necessary. If ∂ 2 Gj /∂kj2 < 0, it is also sufficient.

22
 As an example, let F1 = F2 = ... = FJ . Aggregate output is then maximized if and only if
capital is allocated equally across sectors. This is obtained under incomplete markets if and only if
σ 1 = σ 2 = ... = σ J . If instead risks are asymmetric across sectors, the allocation of capital is biased
towards the least risky sectors.
In our benchmark one-sector model, incomplete market could result to a lower level of income
and savings only because production risk discourages capital accumulation. With multiple sectors,
the allocation of any given amount of capital is inefficient, which gives an additional reason for
incomplete markets to lead to lower levels of income and savings than complete markets. The more
“unbalanced” the risks across different sectors, the more severe the distortion of the cross-sectoral
allocations, and the larger the reduction in aggregate output and savings generated by incomplete
markets. What is more, cyclical variation in either the sectoral risks or the effective risk aversion
of the agents may lead to cyclical variation in the distortion of the cross-sectoral allocation of
resources, thus amplifying the business cycle. We leave the further examination of these issues for
future research.

6 Conclusion

This paper investigates an incomplete market economy with decentralized production and idio-
syncratic technological risk. We assume that there are no short sales constraints, and that the
absence of certain assets is the only source of market imperfection in the economy. To simplify the
analysis, we do not consider the possibility of aggregate uncertainty. Macroeconomic aggregates
and financial prices are then deterministic but can possibly follow complicated dynamics and cy-
cles. Because it influences risk-taking, market incompleteness plays a critical role on both the level
and the volatility of aggregate economic activity, even though the distribution of wealth plays no
role in the equilibrium dynamics of our model. The interaction between the precautionary motive
and investment risk is thus a powerful source of poverty traps and volatility in incomplete-market
growth economies.
This work suggests several directions for further research. Work in progress attempts to intro-
duce aggregate uncertainty in our model. This will allow us to embed our results in a stochastic
RBC framework and characterize the impulse response of the economy to aggregate shocks. Other
projects could further consider the role of market incompleteness in human capital accumulation
and the labor supply. This research may help us further understand the interaction between finan-
cial markets and the real sector of the economy.

23
7 Appendix

Proof of Lemma 1

An entrepreneur chooses consumption cht , capital expenditure kth , storage sht and asset holdings θht
that maximize
h i
u(cht ) + β Et J eht+1 + Aht+1 F (kth ) + (1 − δ ht+1 )kth + ρsht + dt+1 · θht , t + 1

subject to the budget equality cht + kth + sht + π t · θht = wth , and the non-negativity constraints:
kth ≥ 0, sht ≥ 0. The agent never uses storage if Rt > ρ, and is indifferent between investing in
h
storage and investing in the bond if Rt = ρ. It is then convenient to consider b θ0,t = ρsht + θh0,t .
The solution to the Bellman equation is simplified by the following observations. Since wt+1 h is
normal, we rewrite the objective function as
· ¸
h h b h Γat+1 h
u(ct ) + β J Φ(kt ) + θ0,t − V art (wt+1 ), t + 1 .
2
h P
The period-t budget constraint implies that b θ0,t = Rt (wth − cht − kth − N h
n=1 π n,t θ n,t ). We infer that
the decision problem reduces to maximizing
" N
#
X Γat+1
u(cht ) + β J Φ(kth ) + Rt (wth − cht − kth − π n,t θhn,t ) − h
V art (wt+1 ); t + 1 .
n=1
2

over the unconstrained variables (cht , kth , (θhn,t )N

n=1 ).
We begin by considering the demand for risky assets. Conditional on productive investment
and consumption, the optimal portfolio maximizes
N
" N
#
X Γat+1 X
h h h h h h h
−Rt π n,t θn,t − V art et+1 + At+1 F (kt ) + (1 − δ t+1 )kt + θn,t dn,t+1 .
n=1
2 n=1

We write the FOC of this mean-variance problem and immediately obtain

θhn,t = −β he,n − β hA,n F (kth ) + β hδ,n kth − Rt π n,t /(Γat+1 ). (19)

Individual wealth in period t + 1 then reduces to

N
h h Rt X
h
wt+1 = Φ(kth ) + b
θ0,t + F (kth )Ãht+1 − δ̃ t+1 kth + ẽht+1 − π n,t dn,t+1 . (20)
Γat+1 n=1

24
The agent thus fully hedges the endowment and productivity shocks, and takes a net position in
risky assets to take advantage of a possible risk premium.
We now consider the optimal level of productive investment. Substitute next-period wealth
(20) into the objective function:
" Ã N
! N
#
X Rt2 X 2
u(cht ) + β J G(kth , at+1 , σ t+1 ) + Rt wth − cht − kth − π n,t θhn,t − π , t+1 .
n=1
2Γat+1 n=1 n,t

The optimal portfolio rule (19) implies that the objective function can be rewritten as
" #
h
G(kth , at+1 , σ t+1 ) + Rt (wth − cht − kth )
u(ct ) + β J P Rt2 PN ;t + 1 . (21)
+Rt N h h h h h
n=1 π n,t [β e,n + β A,n F (kt ) − β δ,n kt ] + 2Γat+1
2
n=1 π n,t

The optimal level of productive investment therefore maximizes the criterion (3). Since F 0 (0) =
P
+∞, we infer that kth > 0 if A + Rt N h
n=1 β A,n π n,t > 0.
We finally turn to optimal consumption. We differentiate equation (21), take logs and infer

ln βRt
cht = − + bht+1 + at+1 [G(kth , at+1 , σ t+1 ) + Rt (wth − cht − kth )]
Γ
N
X N
h h h h h Rt2 X 2
at+1 Rt π n,t [β e,n + β A,n F (kt ) − β δ,n kt ] + π n,t .
2Γat+1
n=1 n=1

We infer that cht is a linear function of wealth wth , with slope at = 1/[1 + (at+1 Rt )−1 ] and intercept
½ h i¾
at ln βRt
bht = bht+1 − + +at+1 G(kth , at+1 , σ t+1 ) − Rt kth
at+1 Rt Γ
XN · ¸
h h h h h Rt π n,t
+at π n,t β e,n + β A,n F (kt ) − β δ,n kt + .
n=1
2Γat+1

By the envelope theorem Jw (w, t) = exp(−Γcht ), the Recursive Condition is satisfied at date t.

Equilibrium under a Finite Horizon (Proposition 1)

We show the theorem by a backward recursion. At date t = T , agents have identical propensities
aT = 1 and make no risky investment: kTh = 0. Assume now that the properties of the theorem
hold at instants t + 1, .., T. Step 1 shows that when there is only one type of technological risk,
the risk premium is necessarily equal to zero in any finite-horizon equilibrium. Step 2 derives the
equilibrium conditions for all economies.

25
Step 1. We first want to show that there is no risk premium in equilibrium when agents are exposed
to shocks to total productivity Aht+1 but not to the depreciation rate δ ht+1 . It is convenient to define
h h
the truncated price vector πbt = (π n,t )N N
n=1 and the individual vectors β A = (β A,n )n=1 . Since there
PH P
is no aggregate multiplicative shock, we know that h=1 β hA = 0 and therefore H bt · β hA = 0.
h=1 π
Without loss of generality, we index households so that

π bt · β L
bt · β 1A ≥ ... ≥ π bt · β L+1
A ≥0≥π A bt · β H
≥ ... ≥ π A.

The corresponding capital stocks provided by equation (11) then satisfy kt1 ≥ ... ≥ ktH . For any
n ∈ {1, .., N }, the market clearing of security n implies
H H
1 X h 1 X Rt π n,t
θn,t = − F (kth )β hA,n − = 0.
H H Γat+1
h=1 h=1
PH h h
bt = −Γat+1
This yields the vector equality π h=1 F (kt )β A /(HRt ). Since
" L H
#
2 Γat+1 X h h
X
h h
kb
πt k = − bt · β A +
F (kt ) π bt · β A ,
F (kt ) π
HRt
h=1 h=L+1
hP PH i
π t k2 ≤ −Γat+1
we infer kb L L b · βh +
h=1 F (kt ) π t A h=L+1 F (kt
L+1
) b
π t · β h
A /(HRt ), or

XL
Γat+1
π t k2 ≤ −
kb [F (ktL ) − F (ktL+1 )] bt · β hA ≤ 0.
π
HRt
h=1

We conclude that πbt = 0 in any equilibrium. A similar argument holds if agents are only exposed
to depreciation shocks δ ht+1 , but not to total productivity Aht+1 .

Step 2. In equilibrium agents make the same investment kth = Kt > 0 in the risky technology. Each
agent purchases θhn,t = −β he,n − β hA,n F (Kt ) + β hδ,n Kt units of each risky asset; she thus sells at no
cost the tradable components of her production and endowment risks. Equations (6) and (8) are
implied by individual decision, and equations (7) and (10) are obtained by aggregating the budget
constraints. We note that equations (6) and (7) uniquely determine Rt , Kt and St . The first-order
condition u0 (cht ) = βRt Et u0 (cht+1 ) implies

1 £ ¤
0 = Γcht + ln(βRt ) + Γ2 a2t+1 F (Kt )2 σ 2A + σ 2E − Γcht+1 .
2
Aggregation of this relation then yields equation (9). We finally note that the macro aggregates
are the same in all states of date t.

26
Proof of Lemma 2

Under condition (i), the function G(k, at+1 , σ t+1 ) − Rt k is concave in k, and FOC (11) has a unique
solution.
We now consider a Cobb-Douglas technology F (k) = kα in an economy without depreciation
risk. Since Gk (k, at+1 , σ t+1 ) ≡ F 0 (k)[A − Γat+1 F (k)σ 2A ] + 1 − δ, we infer that

Gkk (k, at+1 , σ t+1 ) = F 00 (k)[A∗ − Γat+1 σ 2A F (k)] − Γat+1 σ 2A [F 0 (k)]2

£ ¤
= αkα−2 (1 − 2α)Γat+1 σ 2A F (k) − (1 − α)A∗ .

When α > 1/2, the function Gk is strictly decreasing in the capital stock, and satisfies limk→0 Gk (k, at+1 ) =
+∞, limk→+∞ Gk (k, at+1 ) = −∞; the equation Gk (k, at+1 ) = Rt thus has a unique solution for all
Rt > 0. On the other hand when α < 1/2, the function Gk (k, at+1 ) is decreasing for low values of
k, reaches a minimum Rmin (at+1 ), and converges to 1 − δ as k → +∞. Equation (11) has therefore
two solutions when Rmin (at+1 ) < Rt < 1 − δ, and no solution if Rt < Rmin (at+1 ). When equation
(11) has two solutions k1 and k2 (k2 > k1 ), we observe that
Z k2
G(k2 , at+1 ) = G(k1 , at+1 ) + Gk (k, at+1 )dk
k1

< G(k1 , at+1 ) + (k2 − k1 )Rt .

Any admissible plan of the form (c, k2 , s, θ) is therefore suboptimal.

More generally, consider an arbitrary production function and let k0 = F −1 [A/(Γat+1 σ 2A )]. The
function Gk (k, at+1 ) = F 0 (k)[A − Γat+1 F (k)σ 2A ] + 1 − δ is decreasing and larger than 1 − δ when
k ∈ (0, k0 ], and is strictly smaller than 1 − δ on [k0 , +∞). The equation Gk (k, at+1 ) = Rt has thus
a unique solution when Rt ≥ 1 − δ, as is always the case under condition (iii).

Proof of Proposition 2

Consider a ∈ R and the mapping

(
W0 (WT ) if W0 (WT ) is defined and larger than a,
ϕ(WT ) =
a otherwise.

Since ϕ is continuous on its domain [e, +∞), its image is an interval of the real line. When WT → e,
productive investment KT −1 and sT −1 decline to zero, implying that RT −1 → +∞, aT −1 → 1,
cT −1 → −∞, WT −1 → −∞, and ϕ(e) = a. On the other hand when WT → +∞, we observe that

27
Rt = ρ for all t and ϕ(WT ) → +∞. This establishes that ϕ[e, +∞) = [a, +∞). Since the argument
applies to any choice of a, we conclude that ϕ(WT ) = W0 has a solution for any W0 ∈ R.

Proof of Lemma 4
We consider the decomposition Mth = Xth + Yth , where

Xth = G(kth , at+1 , σ) − kth − ln(βRt )/(Γat+1 Rt )

N
X
Yth = π n,t [β he,n + β hA,n F (kth ) − β hδ,n kth + Rt π n,t /(2Γat+1 )].
n=1

We want to check the convergence of the series

à +∞ +∞
!
X Xsh X Ysh
h h
bt = at Mt + ++ .
R ...Rs−1
s=t+1 t
R ...Rs−1
s=t+1 t

By the assumption that the sequences {π t , Rt π t }∞ ∞

t=0 and {Πt }t=0 are bounded, the sequences {at }t=0
∞

and {Rt }∞
t=0 are contained in compact intervals of the form [a, 1] and [R, R], where a > 0 and
R ≥ R > 1. This implies that the infinite sequences {Kt } and {Xth }∞ t=0 are bounded. We also infer
the boundedness of {Yt }t=0 and {Mt }t=0 . The deterministic sequence {bht }∞
h ∞ h ∞
t=0 is thus well-defined
and bounded.
£ ¤
We now check that J(W, Z) ≡ −(Γa0 )−1 exp −Γ(a0 W + bh0 ) coincides with the value function.
For any real number W , any admissible plan {c0t , kt0 , s0t , θ0t , Wt0 }∞ 0
t=0 such that W0 = W satisfies

J(W, Z) = M ax [u(c) + βEJ(W1 , SZ)]

{c,θ,W1 }

≥ u(c00 ) + βE0 J(W10 , SZ)

≥ u(c00 ) + βE0 u(c01 ) + β 2 E0 J(W20 , S 2 Z),

and by repetition "T −1 #

X
t
J(W, Z) ≥ E0 β u(c0t ) + β T E0 J(WT0 , S T Z). (22)
t=0

Since {exp(−ΓbhT )/aT }∞ 0 0

T =0 is bounded and exp(−ΓaT WT ) ≤ 1 + exp(−ΓWT ), Assumption [1] im-
£ ¤
plies that β T E0 J(WT0 , S T Z) = −(ΓaT )−1 β T E0 exp −Γ(aT WT0 + bhT ) converges to zero as T goes
to infinity. Letting T go to infinity in (22), we obtain
"+∞ #
X
t 0
J(W, Z) ≥ E0 β u(ct ) . (23)
t=0

28
This proves that J(W, Z) is an upper bound to the value function. It is then easy to check that
the consumption plan defined by (4) is admissible and reaches J(W, Z).

Proof of Proposition 6

Letting ϕ(R) ≡ Γ−2 (1 − 1/R)−2 ln( Rβ

1
), we rewrite system (15) − (16) as:

£ ¤
ϕ(R) = F (K)2 σ 2A + K 2 σ 2δ + σ 2e /2 (24)

· µ ¶ ¸ µ ¶
0 2 1 1
F (K) A − Γσ A 1 − F (K) − Γ 1 − Kσ 2δ + 1 − δ − R = 0. (25)
R R

The function ϕ(R) decreases on (1, 1/β] and satisfies ϕ(1, 1/β] = [0, +∞). Equation (24) thus
defines the decreasing function K1 (R),which maps (1, ϕ−1 (σ 2E /2)] onto [0, +∞). Similarly, equation
(25) implicitly defines the decreasing function K2 (R) that maps [1, +∞) onto (0, K ∗ ], where K ∗ =
(F 0 )−1 (δ/A).
Consider the function ∆(R) = K1 (R) − K2 (R). When R → 1, we observe that K1 (R) → +∞,
K2 (R) → K ∗ , and therefore ∆(R) → +∞. We also note that ∆[ϕ−1 (σ 2E /2)] = −K2 [ϕ−1 (σ 2E /2)] <
0. The graphs of the functions K1 and K2 therefore intersect and there exists at least one steady
state.
Finally, we analyze the monotonicity of the steady state with respect to the economy’s para-
meters. We consider the case |K20 (R)| < |K10 (R)|. An increase in σ e or β pushes down the function
K1 (R) and leaves the function K2 (R) unchanged. The steady state is therefore characterized by
a lower interest rate and a higher capital stock. Similarly, an increase in 1 − δ and A pushes up
K2 (R), also leading to a lower interest rate and a higher capital stock. We note that an increase
in Γ, σ A or σ δ pushes down both K1 (R) and K2 (R) and can have ambiguous effects, as is verified
in simulations.

Proof of Proposition 7

It is enough to provide an example of multiplicity. We numerically check that there exist three
equilibria when Γ = 1.2, A = 20, α = 0.95, β = 0.05, δ = 0.5, ρ = 1 − δ, σ 2A = 85, σ E = 0, e = 0.

29
Local Stability of the Steady State

Around the steady state, we know that there is no storage, and the iterated function H is implicitly
defined by:
Kt = Φ−1 (Wt+1 ),
£ ¤
Rt = F 0 (Kt ) A − Γat+1 F (Kt )σ 2A + 1 − δ,
at = 1/[1 + (at+1 Rt )−1 ],
£ ¤
Ct = Ct+1 − ln(βRt )/Γ − Γa2t+1 F (Kt )2 σ 2A + σ 2e /2,
Wt = Ct + Kt .
The function H has Jacobian  
∂at ∂at
∂at+1 0 ∂Wt+1
 
J =

∂Ct
∂at+1 1 ∂Ct
∂Wt+1
.

∂Ct ∂(Ct +Kt )
∂at+1 1 ∂Wt+1

We observe that ∂Kt /∂Wt+1 = 1/Φ0 (Kt ) > 0, and15

∂Rt
= −ΓF (Kt )F 0 (Kt )σ 2A ≤ 0,
∂at+1
∂Rt 1 © 00 £ ¤ ª
= 0 F (Kt ) A − Γat+1 F (Kt )σ 2A − at+1 [F 0 (Kt )]2 Γσ 2A < 0.
∂Wt+1 Φ (Kt )

Let ϕ(v) = 1/(1+v −1 ). We note that ϕ(v) = v/(1+v) and therefore ϕ0 (v) = 1/(1+v)2 = [ϕ(v)/v]2 .
Since at = ϕ(at+1 Rt ), we infer that
µ ¶2 µ ¯ ¯¶
∂at at ¯ ∂Rt ¯
= ¯
Rt − at+1 ¯ ¯ ,
∂at+1 at+1 Rt ∂at+1 ¯
µ ¶2 ¯ ¯
∂at at ¯ ∂Rt ¯
=− ¯
at+1 ¯ ¯ < 0.
∂Wt+1 at+1 Rt ∂Wt+1 ¯

Future propensity at+1 has an ambiguous effect on current propensity at . There is both a positive
direct effect (due to the complementarity of future and current consumption) and a negative indirect
effect (the precautionary motive causes a decline in the current interest rate Rt ). Since Ct =
£ ¤
Ct+1 − ln(βRt )/Γ − Γa2t+1 F (Kt )2 σ 2A + σ 2e /2, we infer that
¯ ¯
∂Ct 1 ¯¯ ∂Rt ¯¯
= − Γat+1 [F (Kt )2 σ 2A + σ 2e ]
∂at+1 ΓRt ¯ ∂at+1 ¯
¯ ¯
∂Ct 1 ¯¯ ∂Rt ¯¯ 2
0
2 F (Kt )F (Kt )
= − Γa σ
∂Wt+1 ΓRt ¯ ∂Wt+1 ¯ t+1 A
Φ0 (Kt )
15
Since R∞ > 1, we infer that A − Γat+1 F (Kt )σ2A > 0 on a neighborhood of the steady state.

30
An increase in at+1 and Wt+1 leads to a decline in the current interest rate, which has a positive
effect on current consumption. On the other hand, the increase in at+1 and Wt+1 implies that the
agent bears more risk between time t and time t + 1. The precautionary motive can then lead to a
decrease in current consumption and current wealth, which may generate endogenous fluctuations.
The characteristic polynomial P (x) = det(J − xI) can be rewritten as
µ ¶½ · ¸ ¾
∂at 2 ∂(Ct + Kt ) ∂Kt ∂Ct ∂at
P (x) = −x x − 1+ x+ +x .
∂at+1 ∂Wt+1 ∂Wt+1 ∂at+1 ∂Wt+1

The roots of P are the eigenvalues of the backward dynamical system. Since P (−∞) = +∞ and
P (+∞) = −∞, there always exists a real eigenvalue. Simple calculation shows that P (1) > 0 if
and only if |K20 (R∞ )| < |K10 (R∞ )| . Thus when there is a unique steady state, the Jacobian matrix
J has an eigenvalue in (1, +∞), and the dimension of the stable manifold is at least 1.
When markets are complete (σ A = σ E = 0), we know that

∂at ∂Ct ∂Ct

= β, = 0, > 0,
∂at+1 ∂at+1 ∂Wt+1

and the characteristic polynomial is of the form P (x) = (β − x) Q(x), where

· ¸
2 ∂(Ct + Kt ) ∂Kt
Q(x) = x − 1 + x+ .
∂Wt+1 ∂Wt+1

This implies that x = β is an eigenvalue, which is contained in the interval (0, 1). We observe that
Q(0) > 0 and Q(1) < 0, and infer the polynomial Q has one root in the interval (0, 1) and one
root in (1, +∞). Overall, the Jacobian matrix J has two eigenvalues in the interval (0, 1), and one
eigenvalue larger than 1. The stable manifold has thus dimension 1.
Consider the parameter values: β = 0.95, Γ = 2, A = 10, α = 0.35, δ = ρ = 0.1, σ e = 0,
σ δ = 0. We check numerically that the characteristic polynomial has one eigenvalue λ1 > 1 and
two eigenvalues in the unit circle when σ A ≤ 30.5. On the other hand when σ A = 31.5, the steady
state is unique and there is one eigenvalue in each of the interval (−∞, −1), (−1, 1) and (1, +∞).
This establishes Theorem 9. We also note that the system undergoes a flip bifurcation as σ A varies
between 30.5 and 31.5, which proves Proposition 9.
Finally, consider the economy Γ = 1.2, A = 20, α = 0.95, β = 0.05, δ = 0.5, ρ = 1 − δ, σ 2A = 85,
σ e = 0, σ δ = 0, e = 0 examined in the proof of Proposition 7. This economy has three steady
states, and it can be checked numerically that the steady state with the intermediate interest rate
is totally unstable, thus establishing Proposition 10.

31
Proof of Proposition 11

From (18), we have µ ¶

η α
Ktα Xtη − = Γat+1 (σ 2x Xt − σ 2k Kt )
Xt Kt
or equivalently µ ¶ µ ¶
Xt η σ 2x Γat+1 σ 2k Xt
− = − .
Kt α αKtα−1 Xtη−1 σ 2x Kt
The result is then immediate.

Proof of Proposition 12

Suppose that the marginal product of capital happens to be equal across sectors. Construct a
perturbation Σ0t of Σt by letting σ 0jt = σ jt for all j 6= 1 and σ 01t = σ 1t + ε, for ε > 0. Given Rt , this
perturbation leaves Ktj unaffected for all j 6= 1 but reduces Kt1 . The marginal product in sector 1
thus increases and is no more equal to the marginal product of other sectors. This is true for ε > 0
arbitrarily small.

32
References
[1] Aghion, P., Bacchetta, P., and Banerjee, A. (1998), “Capital Markets and the Instability of
Open Economies,” manuscript, Harvard University.

[2] Aghion, P., Banerjee, A., and Piketty, T. (1999), Dualism and Macroeconomic Activity, Quar-
terly Journal of Economics 114, 1359-1397.

[3] Aiyagari, S. R. (1994), Uninsured Idiosyncratic Risk and Aggregate Saving, Quarterly Journal
of Economics 109, 659-684.

[4] Angeletos, G. M., and Calvet, L. E. (2003), “Idiosyncratic Production Risk, Growth and the
Business Cycle,” NBER Working Paper 9764.

[5] Banerjee, A., and Newman, A. (1991), Risk-Bearing and the Theory of Income Distribution,
Review of Economic Studies 58, 211-235.

[6] Bewley, T. (1977), The Permanent Income Hypothesis: A Theoretical Formulation, Journal
of Economic Theory 16, 252-292.

[7] Benhabib, J., and Day, R. (1982), A Characterization of Erratic Dynamics in the Overlapping
Generations Model, Journal of Economic Dynamics and Control 4, 37-55.

[8] Benhabib, J., and Farmer, R. (1994), Indeterminacy and Aggregate Returns, Journal of Eco-
nomic Theory 63, 19-41.

[9] Benhabib, J., and Nishimura, K. (1979), The Hopf Bifurcation and the Existence and Stability
of Closed Orbits in Multisector Models of Optimal Economic Growth, Journal of Economic
Theory 21, 421-444.

[10] Boldrin, M., and Montrucchio, L. (1986), On the Indeterminacy of Capital Accumulation
Paths, Journal of Economic Theory 40, 26-39

[11] Brock, W., and Mirman, L. (1972), Optimal Economic Growth and Uncertainty: The Dis-
counted Case, Journal of Economic Theory 4, 497-513.

[12] Caballero, R. (1990), Consumption Puzzles and Precautionary Saving, Journal of Monetary
Economics 25, 113-136.

[13] Calvet, L. E. (2001), Incomplete Markets and Volatility, Journal of Economic Theory 98,
295-338.

33
[14] Cass, D. (1965), Optimum Growth in an Aggregative Model of Capital Accumulation, Review
of Economic Studies 32, 233-240.

[15] Chamberlain, G., and Wilson, C. A. (2000), Optimal Intertemporal Consumption under Un-
certainty, Review of Economic Dynamics 3, 365-395.

[16] Constantinides, G., and Duffie, D. (1996), Asset Pricing with Heterogeneous Consumers, Jour-
nal of Political Economy 104, 219-240.

[17] Deneckere, R., and Pelikan, S. (1986), Competitive Chaos, Journal of Economic Theory 40,
13-25.

[18] Devereux, M. B., and Smith, G. W. (1994), International Risk-Sharing and Economic Growth,
International Economic Review 35, 535-550.

[19] Galor, O., and Zeira, J. (1993), Income Distribution and Macroeconomics, Review of Economic
Studies 60, 35-52

[20] Grandmont, J. M. (1985), On Endogenous Competitive Business Cycles, Econometrica 53,

995-1045.

[21] Green, E. J. (1987), “Lending and the Smoothing of Uninsurable Income,” in Prescott, E., and
Wallace, N. (eds.), Contractual Arrangements for Intertemporal Trade, University of Minnesota
Press, 3-25.

[22] Greenwood, J., and Jovanovic, B. (1990), Financial Development, Growth, and the Distribu-
tion of Income, Journal of Political Economy 98, 219-240.

[23] Grossman, S., and Hart, O. (1979), A Theory of Competitive Equilibrium in Stock Market
Economies, Econometrica 47, 293-330.

[24] Heaton, J., and Lucas, D. J. (1996), Evaluating the Effects of Incomplete Markets on Risk
Sharing and Asset Pricing, Journal of Political Economy 104, 443-487.

[25] Kimball, M. S. (1993), Standard Risk Aversion, Econometrica 63, 589-611.

[26] King, R. G., and Levine, R. (1993), Finance and Growth: Schumpeter Might Be Right, Quar-
terly Journal of Economics, 717-737.

[27] Kiyotaki, N., and Moore, J. (1997), Credit Cycles, Journal of Political Economy 105, 211-248.

34
[28] Koopmans, T. C. (1965), “On the Concept of Optimal Economic Growth,” in The Economic
Approach to Development Planning, Amsterdam: North-Holland.

[29] Krebs, T. (2003), Human Capital Risk and Economic Growth, Quarterly Journal of Economics
118, 709-744.

[30] Krusell, P., and Smith, A. (1998), Income and Wealth Heterogeneity in the Macroeconomy,
Journal of Political Economy 106, 867-896.

[31] Kydland, F., and Prescott, E. C. (1982), Time to Build and Aggregate Fluctuations, Econo-
metrica 50, 1345-1370.

[32] Leland, H. (1968), Saving and Uncertainty: The Precautionary Demand for Saving, Quarterly
Journal of Economics 82, 465-473.

[33] Levhari, D., Mirman, L. J., and Zilcha, I. (1980), Capital Accumulation under Uncertainty,
International Economic Review 21, 661-671.

[34] Ljungqvist, L., and Sargent, T. J. (2000), Recursive Macroeconomic Theory, MIT Press.

[35] Mankiw, N. G. (1986), The Equity Premium and the Concentration of Aggregate Shocks,
Journal of Financial Economics 17, 211-219.

[36] Marcet, A., Obiols-Hom, F., and Weil, P. (2000), “Incomplete Markets, Labor Supply and
Capital Accumulation”, Working Paper, Universitat Pompeu-Fabra.

[37] Mauro, P. (1995), Stock Markets and Growth: A Brief Caveat on Precautionary Savings,
Economics Letters 47, 111-116.

[38] Mas-Colell, A., and Monteiro, P. K. (1996), Self-Fulfilling Equilibria: An Existence Theorem
for a General State Space, Journal of Mathematical Economics 26, 51-62.

[39] Mehra, R., and Prescott, E. C. (1985), The Equity Premium: A Puzzle, Journal of Monetary
Economics 15, 145-161.

[40] Merton, R. C. (1969), Lifetime Portfolio Selection under Uncertainty: The Continuous Time
Case, Review of Economics and Statistics 51, 247-257.

[41] Mitra, T. (1996), An Exact Discount Factor Restriction for Period-Three Cycles in Dynamic
Optimization Models, Journal of Economic Theory 69, 281-305.

35
[42] Obstfeld, M. (1994), Risk-Taking, Global Diversification, and Growth, American Economic
Review 84, 1310-1329.

[43] Ramsey, F. P. (1928), A Mathematical Theory of Saving, Economic Journal 38, 543-559.

[44] Saint-Paul, G. (1992), Technological Choice, Financial Markets and Economic Development,
European Economic Review 36, 763-781.

[45] Sandmo, A. (1970), The Effect of Uncertainty on Saving Decisions, Review of Economic Studies
37, 353-360.

[46] Scheinkman, J., and Weiss, L. (1985), Borrowing Constraints and Aggregate Economic Activ-
ity, Econometrica 54, 23-45.

[47] Shell, K. (1987), “Sunspot Equilibrium,” in The New Palgrave: A Dictionary of Economics,
J. Eatwell, M. Milgate and P. Newman, eds., New York: Macmillan & Co.

[48] Shiller, R. J. (1981), Do Stock Prices Move Too Much to be Justified by Subsequent Changes
in Dividends?, American Economic Review 71, 421-436.

[49] Shiller, R. J. (1989), Market Volatility, Cambridge, Mass.: MIT Press.

[50] Sibley, D. S. (1975), Permanent and Transitory Income Effects in a Model of Optimal Con-
sumption with Wage Income Uncertainty, Journal of Economic Theory 11, 68-82.

[51] Sorger, G. (1992), On the Minimum Rate of Impatience for Complicated Optimal Growth
Paths, Journal of Economic Theory 56, 160-179.

[52] Yaari, M. E. (1976), A Law of Large Numbers in the Theory of Consumer’s Choice Under
Uncertainty, Journal of Economic Theory 12, 202-217.

[53] Weil, P. (1992), Equilibrium Asset Prices in Economies with Undiversifiable Labor Income
Risk, Journal of Economic Dynamics and Control 16, 769-790.

36
 Wt Wealth Wt Rt Interest Rate Rt =1+rt
22.5 1.06
20 1.04
17.5
15 1.02
12.5 t
200 400 600 800 1000
10 0.98
7.5
t 0.96
200 400 600 800 1000

Kt Capital Kt St Storage St
20 10
8
15
6
10
4
5 2
t t
200 400 600 800 1000 200 400 600 800 1000

Ct Consumption Ct at M.P.C. at

0.024
2.25
0.022
2.2
t
2.15 200 400 600 800 1000
0.018
t
200 400 600 800 1000

Figure 1A. Finite-Horizon Growth Path

We plot the time paths of wealth, interest rate, capital investment, storage, consumption and
marginal propensity to consume for an economy with parameters Γ=10, A=1, α = 0.4, β = 0.95, δ
= 0.05, ρ = 0.95, e = 0, σe = 0, σδ = 0, σA = 0.5, T = 1000.
 0.3
0.2
0.1
t
200 400 600 800 1000
-0.1
-0.2
-0.3
-0.4

Figure 1B. Turnpike Property

The figure plots ln(Wt / W∞), which quantifies the distance between the wealth Wt in period t of a
finite-horizon economy (T=1000) and the steady-state level of wealth W∞ of the infinite-horizon
economy (T=∞). The economy is the same as in Figure 1A

WT
10 20 30 40 50 60 70

Figure 2. Nonmonotonicity of W0(WT)

Multiple equilibria in a finite-horizon economy. The figure illustrates the non-monotonic relation
between initial and final wealth. For any given W0 ∈ [1,3] there are multiple WT corresponding to
different equilibrium paths. The parameters are Γ=10, A=10, α = 0.35, β = 0.9, δ = 0.5, ρ = 0.5, e
= 0, σe = 0, σδ = 0, σA = 100, T = 40.
 Wt Wealth Wt Rt Interest Rate Rt =1+rt
14
12
12
10 10
8 8
6
6
4
4 2
t t
10 20 30 40 50 10 20 30 40 50

Kt Capital Kt St Storage St
10
1.5
1.25 8
1 6
0.75
4
0.5
0.25 2
t t
10 20 30 40 50 10 20 30 40 50

Ct Consumption Ct at M.P.C. at
5
0.06
4 0.05
0.04
3
0.03
2 0.02
0.01
t t
10 20 30 40 50 10 20 30 40 50

Figure 3. Endogenous Fluctuations in Finite Horizon

We plot the time paths of aggregate wealth, interest rate, capital investment, storage,
consumption, and marginal propensity to consume. The parameters are Γ=1.625, A=10, α = 0.35,
β = 0.99, δ = 0.1, ρ = 0.3, e = 0, σe = 0, σδ = 0, σA = 60, T = 50.