Endogenous Credit Cycles

Endogenous Credit Cycles
Author(s): Chao Gu, Fabrizio Mattesini, Cyril Monnet, and Randall Wright
Source: Journal of Political Economy, Vol. 121, No. 5 (October 2013), pp. 940-965
Published by: The University of Chicago Press
Stable URL: http://www.jstor.org/stable/10.1086/673472 .
Accessed: 16/09/2014 03:39
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
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].
The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to Journal
of Political Economy.
http://www.jstor.org
This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

All use subject to JSTOR Terms and Conditions
Endogenous Credit Cycles
Chao Gu
University of Missouri–Columbia
Fabrizio Mattesini
University of Rome Tor-Vergata
Cyril Monnet
University of Bern and Study Center Gerzensee
Randall Wright
University of Wisconsin–Madison, Federal Reserve Bank of Minneapolis, Federal Reserve Bank of
Chicago, and National Bureau of Economic Research
This paper studies models of credit with limited commitment and,

therefore, endogenous debt limits. There are multiple stationary equi-
libria plus nonstationary equilibria in which credit conditions change
simply because of beliefs. There can be equilibria in which debt lim-
its display deterministic cyclic or chaotic dynamics, as well as stochastic
ðsunspotÞ equilibria in which they fluctuate randomly, even though
fundamentals are deterministic and time invariant. Examples and ap-
plications are discussed. We also consider different mechanisms for de-
termining the terms of trade and compare the setup to other credit mod-
els in the literature.
We thank two referees and the editor ðRobert ShimerÞ for very useful comments and sug-
gestions. We also thank Huberto Ennis, Karl Shell, Fernando Alvarez, Michael Woodfood, and
Christian Hellwig for their input. For research support, Wright thanks the National Science
Foundation and the Ray Zemon Chair in Liquid Assets at the Wisconsin School of Business.
The usual disclaimers apply.
[ Journal of Political Economy, 2013, vol. 121, no. 5]

© 2013 by The University of Chicago. All rights reserved. 0022-3808/2013/12105-0001$10.00
940

endogenous credit cycles 941
As we have seen, all financial institutions are at the mercy
of our innate inclination to veer from euphoria to despon-
dency. ðNiall Ferguson, The Ascent of Money ½2008Þ
I. Introduction
Are credit markets susceptible to animal spirits, self-fulfilling prophecies,
and endogenous fluctuations? This is different from asking if there are
channels in credit markets that amplify or propagate fluctuations driven
by fundamental factors, as emphasized in the literature surveyed by Gert-
ler and Kiyotaki ð2010Þ. In this paper, we show that even if fundamentals
are deterministic and time invariant, economies with credit market fric-
tions can display cyclic, chaotic, and stochastic dynamics driven purely by
beliefs. The key friction is limited commitment, which leads to endoge-
nous borrowing constraints.1 While most studies of these kinds of models
focus on stationary equilibria, we show that economies with commitment
frictions not only generate multiple steady states, with different credit con-
ditions and allocations, but also display equilibria in which credit condi-
tions and allocations vary over time, even when fundamentals do not.
This is true when the terms of trade are determined by Walrasian pric-
ing or by generalized Nash bargaining, although the economic forces gen-
erating dynamics in economies with the two pricing mechanisms are very
different.
With limited commitment, agents honor their debt obligations to
avoid punishment by exclusion from future credit. If one believes that
one’s debt limit in the future will be 0, one has nothing to lose by re-
neging on current obligations, which makes the equilibrium debt limit 0
today. Hence there is always a no-credit equilibrium. Generally, there is
also a steady-state equilibrium with a positive debt limit. To explain how
we get cycles, we find it useful to first build some intuition by considering
a competitive labor market. Let the unconstrained equilibrium hours
and wages be ð‘*; w * Þ. Suppose that we impose an exogenous restriction
‘ ≤ f, where f 5 ‘* 2 ε with ε > 0. There are two effects on workers: ,
goes down and w goes up. The first effect has a second-order impact on
their utility, by the envelope theorem, if ε is not too big. The increase in
1
By modeling debt limits endogenously using limited commitment, we follow Kehoe and
Levine ð1993, 2001Þ and Alvarez and Jermann ð2000Þ. See Azariadis and Kass ð2007, 2013Þ,
Lorenzoni ð2008Þ, Hellwig and Lorenzoni ð2009Þ, and Sanches and Williamson ð2010Þ for re-
lated work. As discussed in more detail below, our setup differs somewhat from the usual
limited-commitment model and is based on a framework we use elsewhere to study banking
ðGu et al. 2013Þ. While there are no banks here, the setup is used because it is tractable and
flexible. In particular, we have production and investment because we think that it may be
interesting to see how these vary over the credit cycle; but as we show, the model can be re-
interpreted as a Kehoe-Levine pure-endowment economy or as some other standard credit
models.

942 journal of political economy
w, however, generates a first-order increase in utility. This is why unions
want to increase w and lower ,, compared to a competitive market. Of
course, no individual worker can affect w in a competitive market, but an
outside ðe.g., union or policyÞ restriction on , can generate a wage-
increasing effect.2
Now consider a competitive credit market. No individual borrower can
affect the terms of his loan by taking on less debt than his unconstrained
equilibrium choice, say y *. But if all borrowers face a constraint y ≤ f 5
y* 2 ε, as long as ε is not too big, they are all better off because the im-
provement in the terms of trade implies a first-order increase in utility,
while the impact of the reduction in y is second order, again by the enve-
lope theorem. Once this is understood, credit cycles emerge as follows.
If you believe that the debt limit tomorrow will decrease somewhat, but
not too much, your utility will increase tomorrow. This makes you more
reluctant to renege on current debt because you do not want to be shut
out of tomorrow’s market. The reluctance to renege makes today’s en-
dogenous debt limit high. Tomorrow, you believe that the debt limit the
next day will go back up. This makes you worse off the day after tomor-
row, again by the envelope theorem. So tomorrow you are more inclined
to renege, and this makes tomorrow’s debt limit low. There is thus a natu-
ral tendency for credit conditions to oscillate in a two-period cycle. Higher-
order cycles, chaotic dynamics, and sunspot equilibria are more compli-
cated versions of the same idea.
A bilateral bargaining model can generate similar results for completely
different reasons. Generalized Nash bargaining has the property that the
surplus of both parties need not increase with the total surplus: one agent’s
surplus must rise but the other’s may fall ðsee Aruoba, Rocheteau, and
Waller ½2007 for an extended discussionÞ. When a borrower is negoti-
ating over the terms of a loan, he can therefore be worse off if his debt
limit goes up. This is different from the envelope-theoretic reasoning in
Walrasian markets but still implies credit cycles: if your debt limit goes up,
you can be worse off when negotiating tomorrow’s credit contract, justi-
fying a low limit today, and so on. Note that there cannot be cycles in the
model with take-it-or-leave-it offers by the borrower or with Kalai’s pro-
portional bargaining solution, both of which imply that the borrower’s
surplus must increase with his debt limit. But with generalized Nash bar-
gaining, as is the case with Walrasian pricing, borrowers can be better off
when debt limits tighten. In either case, in contrast to the labor market
example, in credit markets we do not need outside forces such as unions
or government to impose the relevant quantity restriction: limited com-
mitment delivers this as an equilibrium outcome.
2
As suggested by a referee, one can also provide related examples using international
trade theory with tariffs; we think that the labor example should suffice.

The rest of the paper involves making this precise and providing the
proofs. Before beginning, we mention that, as is well known, similar dy-
namics emerge in monetary models. In particular, the no-credit equi-
librium is reminiscent of the nonmonetary equilibrium that always ex-
ists with fiat currency. The economic forces are different here. At the
risk of oversimplifying, the basic mechanism at work in those models is a
backward-bending savings function. The key economic mechanism at
work in our model—payoffs decreasing as quantity constraints are re-
laxed—is different.3
II. The Model

Time is discrete and continues forever. Each period has two subperiods.
There are equal measures of two types of agents. Type 1 agents consume
and type 2 agents produce a good x in the first subperiod. Type 1 agents
produce an intermediate good in the first subperiod and invest it, to
generate a gross return y that type 2 agents like to consume in the sec-
ond subperiod. There are gains from trade using the following credit
relationship: type 1 agents ðborrowersÞ get x now from type 2 agents
ðlendersÞ and promise to deliver y later, when their investments come
to fruition. Only type 1 agents have access to the investment opportu-
nity. The utility from this arrangement is U 1 ðx; yÞ for type 1 and U 2 ðy; xÞ
for type 2, where U j is strictly increasing in the first argument, which
is j’s consumption, and strictly decreasing in the second, which is j’s
production. Utility is also concave and twice differentiable and satisfies
U j ð0; 0Þ 5 0 ðwhich can be relaxedÞ. We also assume normal goods for
some results.
The key friction in the model is limited commitment: rather than
honor his obligation to deliver y in the second subperiod, type 1 may ab-
scond with or otherwise divert the investment returns for his own pur-
poses, as in Biais et al. ð2007Þ or DeMarzo and Fishman ð2007Þ. If type 1
behaves opportunistically in this way, he gets a payoff ly over and above
U 1 ðx; yÞ. Therefore, l parameterizes the temptation to renege. We im-
3
See Azariadis ð1993Þ for an overview and references to primary sources on money and
dynamics in the overlapping-generations tradition, see Woodford ð1986, 1988Þ for models
with exogenous cash-in-advance or borrowing constraints, and see Matsuyama ð1990Þ for
money in the utility function. Boldrin and Woodford ð1990Þ provide a survey of related
work. Matsuyama ð2013Þ provides an alternative approach, as does Myerson ð2012, 2013Þ.
There is also work on dynamics in growth models with externalities ðsee, e.g., some of the
papers in the Benhabib ½1992 volumeÞ; there are no such externalities in our model.
Sanches and Williamson ð2010Þ also comment on a similarity between money and credit
models and particularly on the no-credit equilibrium. Although we learned a lot from their
analysis about credit markets, generally, Sanches and Williamson do not study dynamics,
which is the main point of this paper.

pose U 1 ðx; y 1 y 0 Þ 1 ly 0 ≤ U 1 ðx; yÞ for all x, y, y 0 ≥ 0. This says that it is
never efficient ex ante for type 1 to divert y 0 because he is better off not
producing/investing in the first place. However, he may be tempted to
divert resources ex post, after the production cost is sunk. Incentive com-
patibility requires trade to be self-enforcing: U j must be nonnegative for
both agents, and we have to ensure compliance by type 1 in the second
subperiod. The incentive to honor his obligation comes from the threat
to exclude type 1 from future credit, which gives him an autarky payoff
normalized to zero. Because it is useful in applications and examples, we
allow imperfect monitoring: a defaulting type 1 agent gets caught only
with probability p.4
Let Vt j be type j’s value function at t given an allocation ðxt ; yt Þ, which
we call a credit contract, since it specifies that type 1 gets ðtype 2 givesÞ x
now and type 1 repays ðtype 2 receivesÞ y later. Let b ∈ ð0; 1Þ be the dis-
count factor across periods. Discounting across subperiods is subsumed
in the U j notation. Then5
Vt 1 5 U 1 ðxt ; yt Þ 1 bVt11
1
; ð1Þ
Vt 2 5 U 2 ðyt ; xt Þ 1 bVt11
2
: ð2Þ
A feasible contract must satisfy the participation constraints in the first

subperiod,
U 1 ðxt ; yt Þ ≥ 0 and U 2 ðyt ; xt Þ ≥ 0: ð3Þ
The critical condition is the repayment constraint for type 1 in the second
subperiod,
lyt 1 ð1 2 pÞbVt11
1
≤ bVt11
1
: ð4Þ
The left-hand side of ð4Þ is type 1’s instantaneous deviation payoff lyt
plus the continuation value: with probability p, he is caught and ex-
cluded from future markets; with probability 1 2 p, he is not caught and
continues in good standing. It is convenient to rewrite ð4Þ as
4
As mentioned, this environment is taken from Gu et al. ð2013Þ, where we discuss im-
perfect monitoring in more detail. But there we study only the incentive-feasible set in the
spirit of mechanism design, while here we study price-taking and bargaining equilibria. To
be clear, we do not need p < 1 for any general results, but it can be useful in applications
and examples.
5
These equations are sometimes referred to as “promise-keeping constraints” in the
related literature.

yt ≤ ft ; ð5Þ
where ft ; ðbp=lÞVt111
is the debt limit. Using ð1Þ, we can express this re-
cursively to make it clear that the debt limit in one period depends on
the debt limit in the next period:
bp 1
ft21 5 U ðxt ; yt Þ 1 bft : ð6Þ
l
We consider both competitive markets and search-and-bargaining mar-

kets, but to ease the presentation, we present only the former in the text
and relegate the latter to Appendix B. Thus, each period, everyone is as-
signed to one of a large number of spatially distinct Walrasian markets,
where they trade short-term ðacross subperiodÞ credit contracts taking
prices as given.6 Let y be the numeraire. Then type 1 maximizes utility
given his budget and credit constraints. We drop the participation con-
straint because autarky is always feasible:
max U 1 ðxt ; yt Þ subject to pt xt 5 yt and ð5Þ: ð7Þ

xt ;yt
Meanwhile, type 2, with no repayment issues, solves
max U 2 ðyt ; xt Þ subject to pt xt 5 yt : ð8Þ

xt ;yt
Let ðx *; y* Þ denote equilibrium ignoring the repayment constraint, the

solution to
Ux1 ðxt ; yt Þxt 1 Uy1 ðxt ; yt Þyt 5 0; ð9Þ
Ux2 ðyt ; xt Þxt 1 Uy2 ðyt ; xt Þyt 5 0: ð10Þ
If y * ≤ ft , we can implement the unconstrained allocation ðx *; y* Þ. Oth-

erwise equilibrium is constrained: yt 5 ft and xt 5 hðft Þ is the solution to
ð10Þ with yt 5 ft .
6
There are no long-term trades. One way to motivate this is to say that agents meeting
in one market never meet again, as formalized in Aliprantis, Camera, and Puzzello ð2006,
2007Þ, which is why we assume a large number of segmented markets. It is even easier to
motivate the lack of long-term credit contracts in the search-and-bargaining version of the
model, where agents trade bilaterally and never meet again, with probability one. Also, to
avoid issues concerning renegotiation or the incentive compatibility of punishments, when
we say that deviators are excluded from future credit markets, we mean that they literally
lose access to these markets and are simply not allowed in. We also emphasize that we do
not allow entry: the measures of borrowers and lenders ðtypes 1 and 2Þ are fixed.

Since x * 5 hðy* Þ, equilibrium can be represented as follows:
if ft < y *; then xt 5 hðft Þ and yt 5 ft ;

ð11Þ
if ft ≥ y *; then xt 5 hðy* Þ and yt 5 y*:
It will be useful below to note the following: when yt ≤ ft is binding, xt is

not necessarily increasing in f, since the sign of

Uy2
U 1y U 2 2U
2 2 2
yx y
Uxyy xy
5 h 0 ðfÞ 5 2 ð12Þ
yf U2
Ux2 1 x Uxx2 2 x2 Uxy2
Uy
is ambiguous. More importantly, the sign of
yU 1 ½hðfÞ; f
yf

Uy2 U2
Uy1 Ux2 2 Ux1 Uy2 2 yUx1 Uyy2 2 2 Uxy2 1 xUy1 Uxx2 2 x2 Uxy2 ð13Þ
U Uy
5 x
U2
Uy
is ambiguous. As we show below, yU 1 =yf < 0 is not only possible but in-
evitable for some f. Hence, a borrower’s payoff can decrease with his
credit limit.
Following Alvarez and Jermann ð2000Þ, for all t, the equilibrium debt
limit ft is defined as follows: type 1 is indifferent between repaying ft
and defaulting. In any feasible allocation, payoffs, and hence ft , must be
bounded ðso, as in many other models, we rule out explosive bubblesÞ.
We can also bound xt and yt without loss in generality. Hence we have
the following definition.
Definition 1. An equilibrium is given by nonnegative and bounded
` `
sequences of credit limits fft gt51 and credit contracts fxt ; yt gt51 such that
ðiÞ for all t, ðxt ; yt Þ solves ð11Þ given ft and ðiiÞ ft solves the difference
`
equation ð6Þ given fxt ; yt gt51 .
We can collapse the two conditions in definition 1 into one:

ðbp=lÞU 1 ½hðft Þ; ft 1 bft if ft < y *
ft21 5 f ðft Þ ; ð14Þ
ðbp=lÞU 1 ðx *; y * Þ 1 bft otherwise:

The dynamical system ð14Þ describes the evolution of the debt limit in
terms of itself. This system is forward looking, naturally, in the sense that
the debt limit in one period depends on the debt limit in the next
period. Equilibria are characterized by nonnegative bounded solutions
to ð14Þ, from which one can solve for the allocation using ð11Þ. This
completes our description of the baseline model.
III. Results
A stationary equilibrium, or steady state, is a fixed point f ðfÞ 5 f. Ob-
viously f 5 0 is one such point, associated with ðx; yÞ 5 ð0; 0Þ. Intuitively,
if there is to be no credit in the future, you have nothing to lose by re-
neging, so no one will extend you credit today. A nondegenerate steady
state is a solution to f ðfs Þ 5 fs > 0. The mild assumption Ux1 ð0; 0Þh 0 ð0Þ
1 Uy1 ð0; 0Þ > lð1 2 bÞ=bp guarantees the following.
Proposition 1. There exists at least one positive steady state, fs 5
f ðfs Þ > 0. If f ðy * Þ < y *, the repayment constraint is binding at any steady
state.
All proofs are in Appendix A, but the existence result is easy to un-
derstand from figures 1A and B. In figure 1A, the debt constraint in
steady state is slack, fs ≥ y *, and in figure 1B, it is binding, fs < y*. Note
that f ðfÞ is not necessarily monotone for f ∈ ð0; y * Þ, so we cannot guar-
antee uniqueness of a positive steady state fs, in general, although it
was unique in all the examples we tried. Some statements of results be-
low proceed as if fs were unique, but this is only to ease the presenta-
tion, and the results hold more generally if one replaces the clause “the
unique positive” with “the smallest positive” steady state.7
It is clear that there exist equilibria in which ft → 0, as shown in fig-
ures 2A and B. Note that with ft21 on the horizontal and ft on the ver-
tical axis, in these diagrams the curve is f 21. We draw it this way even
though the dynamics are forward looking, with causality running from
future to previous periods. In figure 2A, f is monotone, so f 21 is a
function; in figure 2B, it is not monotone, so f 21 is a correspondence.
When f is monotone, once we pick an initial f0 ∈ ð0; fs Þ, the sequence
fft g is pinned down. When f is not monotone, over some range for f0
there are multiple equilibrium paths for fft g. There are even perfect-
foresight equilibria in which we start at fs and stay there for any number
of periods, until dropping to the lower branch of f 21, and then ft → 0.
This is a serious and escalating credit crunch caused by nothing more
than beliefs—a self-fulfilling prophecy.
7
One can make a few observations about the case with multiple positive steady states;
e.g., they alternate between stable and unstable. We omit this in the interest of space.

F IG . 1.—Steady state with ðAÞ fs > y * and ðBÞ fs < y*
Summarizing this discussion, we have established the following pro-

position.
Proposition 2. Suppose that there is a unique positive steady state
fs, and let f~ 5 arg maxf ðft Þ subject to ft ∈ ½0; fs . Starting from any
~
f0 f, there is a nonstationary equilibrium, and possibly more than one,
<
with ft → 0.

F IG . 2.—Nonstationary equilibria
Now consider two-period cycles, with f1 and f2 > f1 denoting the pe-
riodic points. Following standard methods ðsee Azariadis ½1993 for a
textbook treatmentÞ, we have the following proposition.
0
fs . If f ðfs Þ < 21, there is a two-period cycle with f1 < fs < f2 .

We illustrate the previous result with examples using the following func-
tional forms:
x 12a x 11g
U 1 ðx; yÞ 5 2y and U 2 ðy; xÞ 5 y 2 : ð15Þ
12a 11g
For simplicity, for now we set a 5 0.8 Thus, U 1 is linear, but since U 2 is
nonlinear, interesting dynamics still obtain via x 5 hðfÞ. The parameters
are not calibrated to be empirically reasonable, only to illustrate some
mathematical possibilities ðthis is not to say that the model could not be
calibrated more realistically in future workÞ. Examples 1 and 2 show
cycles in which f1 < f2 < y* and f1 < y * < f2, respectively.
Example 1. Let g 5 2:1, b 5 0:4, and p=l 5 6. Then fs 5 0:7194
and there is a two-cycle with f1 5 0:4777 < y * and f2 5 0:9357 < y *. See
figure 3A.
Example 2. Let g 5 0:5, b 5 0:9, and p=l 5 10. Now fs 5 0:9674
and there is a two-cycle with f1 5 0:9328 < y* and f2 5 1:0365 > y *. See
figure 3B.
The next example has a three-period cycle, shown in figure 4A. The
existence of a three-period cycle implies the existence of cycles of all
orders by the Sarkovskii theorem, as well as chaotic dynamics by the Li-
Yorke theorem ðagain see Azariadis 1993Þ. Chaos is observationally equiv-
alent to a stochastic process for ft. Hence, debt limits, investment, and
consumption can appear random when they are not ðsee fig. 4BÞ. Note
that credit limits are marked by x’s in the figure. As one can see, there
are recurrent episodes with three or four periods of unconstrained credit
followed by a crunch, as well as episodes with three periods of tight credit
followed by easy credit. Hence, compared to equilibria with two-period
cycles, these chaotic equilibria can have much more intricate dynamics.
Example 3. Let g 5 0:8, b 5 0:9, and p=l 5 15. Now fs 5 0:9835,
and there is a three-cycle with f1 5 0:9464 < y*, f2 5 1:0516 > y *, and f3
5 1:1684 > y *.
The next result says that in any cycle we must have ft < y * at some
point over the cycle, but not necessarily at all points. In other words, the
debt limit binds infinitely often.
8
It is not merely a normalization to set a 5 0, although that would be the case in the
model with bargaining. In the bargaining version, it is equivalent to bargain over either x
and y, on the one hand, or x 12a =ð1 2 aÞ and y, on the other. In the Walrasian model, how-
ever, we have already bought into linear pricing. If this is not obvious, consider a competitive
profit-maximizing firm with cost function cðyÞ, for which it obviously matters, for profit,
if cðyÞ is linear or strictly convex. We return to a > 0 below. For now, a 5 0 provides a stark
illustration of the economic mechanism at work: in this case, type 1 agents realize no gains
from trade, like a competitive profit-maximizing firm with linear cost, if the constraint is
slack.

F IG . 3.—Cycle with ðAÞ f1 < f2 < y* and ðBÞ f1 < y* < f2
Proposition 4. In any n-period cycle, at least one periodic point has

the repayment constraint binding, ft < y*.
The model can also generate stochastic ðsunspotÞ cycles, where ft and
ðxt ; yt Þ fluctuate randomly. Consider a Markov sunspot variable st ∈ f1; 2g,
which does not affect fundamentals but may affect equilibrium. Let

F IG . 4.—A, A three-period cycle; B, chaotic dynamics
Prðst11 5 1jst 5 1Þ 5 j1 and Prðst11 5 2jst 5 2Þ 5 j2 . The economy is in

state s at date t if st 5 s. Let Vs;tj be type j’s value function in state s at
date t, and let ðxs;t ; ys;t Þ be the state-contingent credit allocation. Then
Vs;t1 5 U 1 ðxs;t ; ys;t Þ 1 b½js Vs;t11

1
1 ð1 2 js ÞV2s;t11
1
; ð16Þ

Vs;t2 5 U 2 ðys;t ; xs;t Þ 1 b½js Vs;t11
2
1 ð1 2 js ÞV2s;t11
2
: ð17Þ
The generalized repayment constraint is
lys;t ≤ fs;t ; ðbp=lÞ½js Vs;t11

1
1 ð1 2 js ÞV2s;t11
1
: ð18Þ
Equilibrium in state s at date t is given as follows:
if fs;t < y *; then xs;t 5 hðfs;t Þ and ys;t 5 fs;t ;

ð19Þ
if fs;t ≥ y *; then xs;t 5 hðy* Þ and ys;t 5 y *:
The analogue to ð6Þ is
fs;t21 5 js b½ðp=lÞU 1 ðxs;t ; ys;t Þ 1 fs;t

1 ð1 2 js Þb½ðp=lÞU 1 ðx2s;t ; y2s;t Þ 1 f2s;t :
This defines a bivariate dynamical system:

f1;t21 j f ðf1;t Þ 1 ð1 2 j1 Þf ðf2;t Þ
5 1 : ð20Þ
f2;t21 j2 f ðf2;t Þ 1 ð1 2 j2 Þf ðf1;t Þ
Definition 2. A sunspot equilibrium is given by nonnegative and

`
bounded sequences of credit limits ffs;t gt51;s51;2 and contingent contracts
`
fxs;t ; ys;t gt51;s51;2 such that ðiÞ ðxs;t ; ys;t Þ solves ð19Þ given fs;t for all t and s and
`
ðiiÞ ðf1;t ; f2;t Þ solves the difference equation ð20Þ given fxs;t ; ys;t gt51;s51;2 .
A proper sunspot equilibrium has f1;t  f2;t for some t. Consider
proper sunspot equilibria that depend only on the state, not on the date,
given by ðf1 ; f2 Þ with f2 > f1 . Then the repayment constraint is binding
in state 1; otherwise, we have x s 5 x * and ys 5 y* in both states, which
implies f1 5 f2 . Following standard methods ðagain see Azariadis 1993Þ,
one can show that proper sunspot equilibria exist when f 0 ðfs Þ < 21, the
same condition for two-period cycles, and so the examples with cycles
also have sunspot equilibria. Formally, we have the following proposition.
fs . If f 0 ðfs Þ < 21, then there exist ðj1 ; j2 Þ, j1 1 j2 < 1, such that the econ-
omy has a proper sunspot equilibrium with f1 and f2 in the neighbor-
hood of fs and Markov transitions given by Prðst11 5 1jst 5 1Þ 5 j1 and
Prðst11 5 2jst 5 2Þ 5 j2 .

If the conditions in proposition 5 are satisfied, one of the credit re-
gimes can be highly persistent: when j1 is big, for example, the economy
can stay in a regime in which debt limits are tight for a long time before
switching to the other regime. The sunspot equilibria guaranteed by the
proposition satisfy j1 1 j2 < 1, however, so they cannot have both regimes
highly persistent. This does not mean that it is impossible to generate
sunspot equilibria in which j1 and j2 are both big; they are just not
guaranteed by the proposition.
IV. Discussion
The existence of equilibria with deterministic or stochastic cycles relies
on the nonmonotonicity of f ðft Þ. To understand this, recall that
bp 1
ft 5 f ðft11 Þ 5 U ðxt11 ; yt11 Þ 1 bft11 : ð21Þ
l
An increase in the debt limit at t 1 1 influences the economy at t in two

ways. First, it directly raises ft through the linear term bft11 . This effect
in isolation works against cycles; for example, for ft11 > y*, the alloca-
tion ðxt11 ; yt11 Þ does not depend on ft11 , so only the linear term in ð21Þ is
operative, and the system ft11 5 ð1=bÞft is explosive. But there is a sec-
ond, nonlinear, effect coming from U 1 ðxt11 ; yt11 Þ, which is ambiguous in
general and negative when ft11 is near y *.9
Proposition 6. With Walrasian pricing, yU 1 ½hðfÞ; f=yf < 0 at f 5
y 2 ε for some ε > 0 if y is a normal good for type 2.
*
Heuristically, in Walrasian equilibrium, a buyer of good y is always

better off under the restriction y ≤ y* 2 ε for some ε > 0, for the same
reason that monopolists produce less than competitive suppliers. Pursu-
ing the monopolist analogy, we do not suggest that ε has to be small—
that is merely a sufficient condition for the envelope theorem—and the
borrowers may be better off for relatively big ε, just not too big. While
debt limits can make borrowers better off, they cannot make everyone
better off. For small ε, lenders lose, and for sufficiently tight debt limits,
everyone loses ðconsider f 5 0Þ.
When the nonlinear term in ð21Þ is negative and big, f ðft11 Þ can be
decreasing. We now present conditions guaranteeing that the system
satisfies the condition for cycles, f 0 ðfs Þ < 21. We start by defining the
elasticity
9
Looking at ð21Þ, it appears that cycles are more likely when b is small, since this reduces
the impact of the linear term, as long as we keep the coefficient on the nonlinear term big
by increasing p=l. It is not quite that simple, however, since the allocation ðxt11 ; yt11 Þ de-
pends on these parameters.

f yU 1 ½hðfÞ; f
hðfÞ 5 :
U ½hðfÞ; f
1
yf
Corollary 1. If hðy * Þ < 21, then we can always pick b and p=l to
generate cycles.
It is not hard to satisfy this elasticity condition.10 If U 1 5 log ð1 1 xÞ
2 A logð1 2 yÞ and U 2 5 log ð1 1 yÞ 2 A logð1 2 xÞ, for A ∈ ð0; 1Þ, hðy * Þ <
^ where A
21 iff A > A, ^ ≃ 0:132. Basically, hðy * Þ < 21 allows us to choose
bp=ð1 2 bÞl close to y* =U 1 ðx *; y* Þ so that fs is near y * and then pick b
and p=l to guarantee f 0 ðfs Þ < 21. Although this is not a quantitative
paper, we mention that having p=l and b makes it easier to construct in-
teresting examples. This is not surprising but is relevant because in fu-
ture quantitative work it may be useful to allow the monitoring proba-
bility p, which is absent in most related models, to play a role.
To say more about conditions to guarantee cycles, reconsider the func-
tional forms in ð15Þ, where now a ∈ ½0; 1Þ. Also assume ð1 2 aÞg > ð1 1 gÞa.
Then we have the following corollary.
Corollary 2. Given the parameter restrictions listed above, there is
a unique positive steady state fs > 0, and there are cycles around fs if
gð1 2 aÞ 2 að1 1 gÞ
b< and
a1g
ð22Þ
21g2a p 12a 12b
11 < < :
bða 1 gÞ l a b
Corollary 3. As a special case, a 5 0 implies that cycles exist if

p=l > 1 1 ð2 1 gÞ=bg.
While it is hard to identify parameter conditions for a local bifurcation
around fs, in general, given these functional forms, we get f 0 ðfs Þ < 21
when p=l is in the interval defined in corollary 2. When a 5 0, so that
U 1 is linear, all we need is p=l > 1 1 ð2 1 gÞ=bg. In this case, credit cycles
are more likely to emerge when b and p=l are big. It may be surprising
that cycles are more likely when agents are patient, the temptation to
default is low, and the monitoring probability is high, but that is what we
find in this specification.
The last item to discuss is the economic interpretation. In the baseline
model, type 2 produces 1’s consumption good, while type 1 produces
intermediate goods, invests, and repays 2 from the proceeds. For con-
creteness, consider an example suggested by the editor. Agent 1 is a con-
tractor whom agent 2 wants to do some work on his house. Agent 1 needs
10
At the suggestion of a referee and the editor, an alternative elasticity condition is pro-
vided in App. C.

to be paid x in advance; then he invests in supplies, or working capital,
generally. There is a cost to agent 2 of providing x, which can be a pro-
duction cost or an opportunity cost of consuming less himself. Agent 1
promises to deliver home improvements/repairs using his labor y, but as
usual, he is tempted to renege for the opportunistic payoff ly. So we im-
pose the constraint ly 1 ð1 2 pÞbVt11 1
≤ bVt11
1
, saying that the contractor
prefers to do the work rather than renege and risk ruining his reputa-
tion, which occurs with probability p, say because homeowners report bad
behavior only probabilistically. Hence, there is a maximum y 5 f that
agent 1 can credibly commit to deliver. Our analysis implies that this ex-
ample has equilibria in which ft displays complicated dynamics.11
We like having production and investment in the model because it
is interesting to know that there can be endogenous dynamics in these
variables, and in aggregate output, that cannot arise in pure-endowment
specifications. However, similar results apply to endowment economies.
Consider a simple Kehoe-Levine economy, where type 1 agents are en-
dowed with one unit of a generic good in the second subperiod and
type 2 with a unit in the first. There is no production or investment.
Agents all want to consume in both subperiods. If 1 gets xt from 2 in the
first subperiod and gives him yt in the second, payoffs are U 1 5 uðxt Þ
1 vð1 2 yt Þ and U 2 5 uð1 2 xt Þ 1 vðyt Þ, which is a special case of our
general setup. The value functions still satisfy ð1Þ and ð2Þ, while the re-
payment constraint becomes vð1Þ 2 vð1 2 yt Þ ≤ bpV1t11 ; ft , or yt ≤ g ðft Þ
with g ðft Þ 5 1 2 v 21 ½vð1Þ 2 ft . The interpretation of ft here is slightly
different, compared to the baseline model, where ft was the maximum
credible promise, because the deviation payoff is now nonlinear. Still,
following the steps taken above, we get
ft21 5 f ðft Þ

bpu½h ○ g ðft Þ 1 bpv½1 2 g ðft Þ 1 bft if g ðft Þ < y * ð23Þ
;
bpuðx * Þ 1 bpvð1 2 y * Þ 1 bft otherwise:
System ð23Þ is similar to ð14Þ, except now the debt constraint is y ≤

g ðfÞ, while in the baseline model it is y ≤ f. To get cycles, we need
f 0 ðfs Þ < 21, which in this version means that g 0 ðfÞ is big. Since g 0 ðy* Þ
11
While we can imagine a variety of other applications, contracting with contractors is
one that many people have experienced firsthand. In fact, the editor’s suggestion was ac-
tually more along the lines of MacLeod and Malcomson ð1993Þ: A firm needs to hire work-
ers before profits are realized, with only the promise of future wages. The firm may renege
but faces punishment, like any other defaulter. Hence, there is a maximum credible wage
promise. The main difference in these applications concerns who has a first-mover disad-
vantage: the household paying a contractor x up front and hoping to receive y or an indi-
vidual working in advance and hoping to get paid. Again, it is easy to imagine other appli-
cations along these lines.

5 1=v 0 ð1 2 y * Þ, we need v 0 ð1 2 y * Þ small. Given this, all the results hold,
although the algebra was easier when deviation payoffs and hence the
repayment constraints were linear. To be clear, usually pure-credit mod-
els have two types, one endowed with 1 in even periods and 0 in odd pe-
riods and vice versa for the other, while we have them endowed this way
across subperiods. That is not crucial: Kehoe and Levine use U 5 uðct Þ
1 buðct11 Þ 1 , while we use U ðxt ; yt Þ 1 bU ðxt11 ; yt11 Þ 1 in general
and U ðxt ; yt Þ 5 uðxt Þ 1 vðyt Þ as a special case, but one can simply reinter-
pret two periods in their model as one period in ours, with a different b.
We can also map our setup into a model of secured lending, as in
Kiyotaki and Moore ð1997Þ. In the first subperiod, type 1 is endowed with
C units of a good he likes but type 2 does not, while type 2 is endowed
with K units of a good that can be consumed by both types or used as
capital to produce more goods in the second subperiod via technology
F ðK Þ. For simplicity, K and goods fully depreciate across periods. Type 1
considers both goods perfect substitutes but consumes only in the sec-
ond subperiod, while type 2 values goods in both subperiods. If type 1
borrows x from 2 and promises to repay y, U 1 5 F ðxÞ 2 y 1 c 2 C , where
c is the consumption of his endowed goods. We subtract C from his
utility function, so his utility in autarky is normalized to zero. Alterna-
tively, c 2 C is his consumption over and above his endowment ðthink of
him as a profit-maximizing producer/investorÞ. Type 2’s payoff is U 2 5
U 2 ðK 2 x; yÞ ðthink of him as a consumer/saverÞ. To secure a loan, type 1
pledges C as collateral: if he delivers y to the lender, he gets C back;
otherwise he gets f ðxÞ 2 ð1 2 lÞy 2 C , as the lender gets a fraction of
what he was promised, while he takes the rest and forfeits his collateral
plus future credit with probability p.
The relevant constraint becomes
F ðxt Þ 2 yt 1 bVt11
1
≥ F ðxt Þ 2 ð1 2 lÞyt 2 C 1 ð1 2 pÞbVt11
1
;
or the following linear transformation of our baseline constraint,
yt ≤ C =l 1 ðbp=lÞVt11
1
; ft : ð24Þ
If p 5 0, this is a simple model of collateralized lending à la Kiyotaki-

Moore; if C 5 0, this is a special case of our baseline model. Following
the same steps, we again get a system similar to ð14Þ:
ft21 5 f ðft Þ

ðbp=lÞF ½hðft Þ 2 ðbp=lÞft 1 bft 1 ð1 2 bÞC =l if ft < y*
;
ðbp=lÞF ðx * Þ 2 ðbp=lÞy * 1 bft 1 ð1 2 bÞC =l otherwise:
ð25Þ

By virtue of ð24Þ, the equilibrium credit limit is at least C =l. Hence, we
lose the no-credit equilibrium ðas in the original Kehoe-Levine model,
where they endowed agents with assets that can be confiscated in the
event of defaultÞ. Intuitively, even if there were no future credit, you
would repay a debt of y < C =l today to avoid forfeiting valuable collat-
eral. We also lose dynamic equilibria converging to a no-credit equilib-
rium. But that does not affect our main results. As long as p > 0, all the
results on cycles continue to hold as long as C is not too big, simply by
continuity ðpayoffs and behavior over a cycle around the positive steady
state do not depend critically on what happens near the originÞ. We
mention this not only to show the flexibility of the framework but to
make the following point: What generates interesting dynamics is the
forward-looking nature of credit markets. If p 5 0, there is only secured
lending, as in Kiyotaki-Moore, and hence no endogenous dynamics.
However, that is particular to the simplicity of this example, as Kiyotaki-
Moore models more generally can also generate endogenous cycles ðsee,
e.g., He, Wright, and Zhu ½2013 or Rocheteau and Wright ½2013 for re-
cent examplesÞ.
V. Conclusion
This paper has developed a framework to study dynamics in lending,
production, and investment. There always exist multiple steady states and
multiple dynamic equilibria in the baseline model. There can exist deter-
ministic, chaotic, and stochastic cycles, where credit conditions fluctuate
even though fundamentals are deterministic and time invariant. Even if
we add features, such as collateral, that rule out the no-credit equilib-
rium and paths that converge to the no-credit equilibrium, cycles still exist.
The key friction in the theory is limited commitment, although there are
other ingredients, including imperfect monitoring. Still, the model is
simple to use. One reason is that equilibrium allocations reduce to a se-
quence of two-period ðor two-subperiodÞ credit arrangements, although
the economy goes on forever. The setup is also quite flexible and can eas-
ily accommodate both price taking and bargaining. We provided sev-
eral interpretations, including pure-credit economies, as well as those with
production and investment.
Of course, we made some strong assumptions, including our assump-
tion of perfect-foresight or rational expectations. Obviously, if one re-
laxes these assumptions, even more exotic equilibria could arise; the goal
here was to show that even with the discipline of rational expectations,
one can get complicated dynamics. It would be interesting to see if com-
binations of fundamental shocks and the endogenous dynamics empha-
sized here might generate empirically plausible cycles. This and other

quantitative work is left to future research. As a final remark, we want to
emphasize that the mathematics used here is neither new nor difficult and
has been applied in many earlier papers to monetary and other models
ðrecall n. 3Þ. Our contribution was not meant to be technical but to pres-
ent a novel economic application to credit markets. Some further prog-
ress has been made along these lines in recent work by He et al. ð2013Þ
and Rocheteau and Wright ð2013Þ, where similar methods are applied
to markets for financial assets and markets for housing.
Appendix A
Proofs
Proof of Proposition 1
0 0
Define T ðfÞ 5 f ðfÞ 2 f. Our assumptions imply T ð0Þ > 0. Also, T ðfÞ 5 b 2 1
< 0 for f > y *. By the continuity and monotonicity of T ðfÞ for f > y *, it is easy to
see the following: if T ðy * Þ ≥ 0, there exists fs > y* such that T ðfs Þ 5 0; and if
T ðy* Þ < 0, there exists fs in ð0; y * Þ such that T ðfs Þ 5 0. In the latter case, there are
no stationary equilibria in which fs > y* because T ðfÞ is strictly decreasing for
f > y * . QED
Because f ðft Þ is continuous, ft21 covers the interval ½0; f ~ for ft ∈ ½0; fs . Since
there is a unique positive steady state, f ðft Þ ft for ft ∈ ð0; fs Þ and f ðft Þ < ft for
>
ft ∈ ðfs ; `Þ. That is, ft21 > ft for ft ∈ ð0; fs Þ and ft21 < ft for ft ∈ ðfs ; `Þ. Given
~ there is a f1 such that f1 ∈ ð0; fs Þ and f1 < f0 , which implies a f2 ∈ ð0; fs Þ
f0 < f,
with f2 < f1 , and so on. This sequence fft g0` converges to zero. QED
Let f 2 ðfÞ 5 f ○ f ðfÞ. Because fs is the unique positive steady state, f ðfÞ > f for
f < fs and f ðfÞ < f for f > fs. Because f ðfÞ is linearly increasing for f > y *, there
exists a f~ > y* such that f ðfÞ
~ > y*. By the uniqueness of the positive stationary
~ < f ðfÞ
equilibrium, f 2 ðfÞ ~ < f.
~ Note that 0 and fs are two fixed points of f 2 . The
slope of f ðf Þ is
2 s
df 2 ðfs Þ
5 f 0 ½ f ðfs Þ f 0 ðfs Þ 5 f 0 ðfs Þf 0 ðfs Þ 5 ½ f 0 ðfs Þ2 > 1:
dfs
The last inequality uses f 0 ðfs Þ < 21. Similarly, df 2 ð0Þ=df > 1. By continuity, f 2
~ it
must cross the 45-degree line in ð0; fs Þ. Because f 2 lies below the diagonal at f,
~ So there are fixed points f1 and f2 such that
crosses it at least once in ðfs ; fÞ.
0 < f1 < fs < f2 for f 2 ðfÞ. QED

Let f1 ; f2 ; : : : ; fn be the periodic points of an n-cycle. Suppose by way of con-

tradiction that f j > y* for all j 5 1; 2; : : : ; n. As f j > y* for all j, the payoff in
each period of the cycle is U 1 ðx *; y* Þ, and Vt 1 5 U 1 ðx *; y * Þ=ð1 2 bÞ for all t . By the
definition of ft , we have ft 5 bp=½ð1 2 bÞlU 1 ðx *; y* Þ for all t. This is a contra-
diction. QED
0
Since f ðfs Þ < 0, there is an interval ½fs 2 ε1 ; fs 1 ε2 , with ε1 , ε2 > 0, such that
f ðf1 Þ > f ðf2 Þ for f1 ∈ ½fs 2 ε1 ; fs Þ and f2 ∈ ðfs ; fs 1 ε2 . By definition, ðf1 ; f2 Þ is a
proper sunspot equilibrium if there exists ðj1 ; j2 Þ, with j1 , j2 < 1 such that
f1 5 j1 f ðf1 Þ 1 ð1 2 j1 Þf ðf2 Þ; ðA1Þ
f2 5 ð1 2 j2 Þf ðf1 Þ 1 j2 f ðf2 Þ: ðA2Þ
Because f1 and f2 are weighted averages of f ðf1 Þ and f ðf2 Þ, where f ðf1 Þ > f1 and
f ðf2 Þ < f2 , by the uniqueness of the positive steady state, necessary and sufficient
conditions for ðA1Þ and ðA2Þ are
f ðf2 Þ < f1 < f ðf1 Þ and f ðf2 Þ < f2 < f ðf1 Þ: ðA3Þ
Now, because f1 < f2 , we can reduce this to
f2 < f ðf1 Þ and f1 > f ðf2 Þ: ðA4Þ
When we expand f ðf1 Þ and f ðf2 Þ around fs and use f ðfs Þ 5 fs, the above in-
equalities are equivalent to
f 2 2 fs fs 2 f1
< 2f 0 ðfs Þ < :
f 2 f1
s
f2 2 fs
0
Because 2f ðfs Þ > 1, ðf2 2 fs Þ=ðfs 2 f1 Þ < 2f 0 ðfs Þ is redundant if 2f 0 ðfs Þ < ðfs
2 f1 Þ=ðf2 2 fs Þ. Now we have two unknowns ðf1 ; f2 Þ and only one inequality,
2f 0 ðfs Þ < ðfs 2 f1 Þ=ðf2 2 fs Þ, to solve. It is straightforward that multiple solu-
tions exist on ½fs 2 ε1 ; fs 1 ε2 . To show j1 1 j2 < 1, rewrite ðA1Þ and ðA2Þ as
f1 2 f ðf2 Þ 2 f2 1 f ðf1 Þ f 1 2 f2
j1 1 j2 5 5 1 1 < 1;
f ðf1 Þ 2 f ðf2 Þ f ðf1 Þ 2 f ðf2 Þ
which holds because ðf1 2 f2 Þ=½ f ðf1 Þ 2 f ðf2 Þ is negative. QED

If f 5 y*, equilibrium is on the contract curve and 2Ux1 =Uy1 5 2Ux2 =Uy2 5 y=x. A
calculation implies
" #
yU 1 ½hðfÞ; fx Uy1 x 2 Uxx2 1 2xyUxy2 1 y2 Uyy2
5 :
yf x U2
*
f→y2
Uy
The term outside the brackets is negative. The term in brackets is positive as long
as y is normal for type 2. QED
Proof of Corollary 1
To construct cycles, it suffices to show that we can choose b and p=l such
that fs is close to y* and f 0 ðfs Þ < 21. A steady state exists at f 5 y2* if y2* 5
ðbp=lÞU 1 ðx2* ; y2* Þ 1 by2* , or

p 12b f
5 : ðA5Þ
l b U ½hðfÞ; f f→y2*
1
The derivative of f ðfÞ at f 5 y2* is f 0 ðfÞ 5 ðbp=lÞyU 1 ½hðfÞ; f=yfjf→y2* 1 b. Hence,

for f 0 ðfÞ < 21, we need

bp yU 1 ½hðfÞ; f
l yf * 1 b < 21: ðA6Þ
f→y2
Combining ðA5Þ and ðA6Þ, after some algebra, we get

f yU 1 ½hðfÞ; f
b < 21 2 *
U 1 ½hðfÞ; f yf f→y2
ðA7Þ
f yU ½hðfÞ; f
1
12 1 :
U ½hðfÞ; f yf f→y2*
If the elasticity condition holds, we can pick b and p=l to satisfy ðA7Þ. QED
Proof of Corollary 2
Given these utility functions, unconstrained and constrained equilibria satisfy

ðx; yÞ 5 ð1; 1Þ and ðx; yÞ 5 ðf1=11g; fÞ. The unique positive steady state solves

12b l
f2ða1gÞ=ð11gÞ 5 ð1 2 aÞ 1 1 :
b p

The steady state is constrained if and only if p=l < ð1 2 aÞð1 2 bÞ=ab. A calcu-
lation implies f 0 ðfs Þ < 21 if and only if p=l > 1 1 ½ð1 1 gÞ 1 ð1 2 aÞ=bða 1 gÞ.
For these to hold simultaneously, we need b < ½gð1 2 aÞ 2 að1 1 gÞ=ða 1 gÞ with
b > 0 as ð1 2 aÞ=a > ð1 1 gÞ=g. QED
Appendix B
Bargaining
Here we show that a model with bargaining can generate the same type of dy-
namics. For ease of presentation, assume that each type 1 meets a random type 2
at each date, and they negotiate a contract ðxt ; yt Þ, taking as given what happens
in all other meetings. Generalized Nash bargaining determines ðxt ; yt Þ. Note that
the strategic foundations of Nash bargaining are not straightforward in non-
stationary situations ðsee, e.g., Coles and Wright 1998; Ennis 2001, 2004; Coles and
Muthoo 2003Þ. Thus, we are taking the Nash solution as a primitive, and there is
no claim here that it is derived from a strategic bargaining game as in standard
stationary models. That said, there is no presumption that one could not gener-
ate interesting dynamics with some strategic bargaining model, such as the ones
used in the above-mentioned papers, but that is not the goal here.
Let type 1’s bargaining power be v, and let threat points be given by contin-
uation values. Since the continuation values and threat points cancel, the bar-
gaining outcome solves the simple problem
max U 1 ðxt ; yt Þv U 2 ðyt ; xt Þ12v subject to ð5Þ: ðB1Þ

ðxt ;yt Þ
If we ignore the repayment constraint for a moment, the solution ðx *; y* Þ sat-

isfies
vUx1 ðxt ; yt ÞU 2 ðyt ; xt Þ 1 ð1 2 vÞU 1 ðxt ; yt ÞUx2 ðyt ; xt Þ 5 0; ðB2Þ
vUy1 ðxt ; yt ÞU 2 ðyt ; xt Þ 1 ð1 2 vÞU 1 ðxt ; yt ÞUy2 ðyt ; xt Þ 5 0: ðB3Þ
If ft ≥ y *, we can implement the unconstrained contract; if ft < y*, let xt 5 hðft Þ

solve ðB2Þ with yt 5 ft . Equilibrium satisfies ð11Þ, exactly as in the Walrasian model.
When y ≤ f binds, we have
yx 2vðUxy1 U 2 1 Ux1 Uy2 Þ 2 ð1 2 vÞðUy1 Ux2 1 U 1 Uxy2 Þ

5 ; ðB4Þ
yf vðUxx1 U 2 1 Ux1 Ux2 Þ 1 ð1 2 vÞðUx1 Ux2 1 U 1 Uxx2 Þ
which is ambiguous, in general. So is
yU 1 ½hðfÞ; f vU 2 ðUxx1 Uy1 2 Uxy1 Ux1 Þ 1 ð1 2 vÞU 1 ðUy1 Uxx2 2 Ux1 Uxy2 Þ
5 ; ðB5Þ
yf vðUxx1 U 2 1 Ux1 Ux2 Þ 1 ð1 2 vÞðUx1 Ux2 1 U 1 Uxx2 Þ

but it is unambiguous that yU =yf < 0 for f close to y* if v < 1 and y is a nor-
mal good for both types.
It is worth mentioning that bargaining with v 5 1 implies yU 1 =yf > 0, in which
case we cannot get endogenous cycles; the same is true for bargaining with Kalai’s
proportional solution. With generalized Nash and v < 1, however, since yU =yf < 0
for f close to y*, we get results similar to what we found with price taking. In par-
ticular, it is a matter of writing down the appropriate dynamical system and using
the same methods to generate endogenous cycles, even though the economic forces
are different, as discussed in the introduction.
Appendix C
Alternative Elasticity Conditions
Here, at the request of the editor and referee, we present an alternative condi-
tion to guarantee the existence of cycles, in terms of the elasticity of credit sup-
ply. From ð13Þ, we have

Uy2 2 Ux2 2
U 1
U 2
2 U 1
U 2
2 yU 1
U 2
2 U 1 xU 1
U 2
2 U
yU 1 ½hðfÞ; f
y x x y x yy
U 2 xy y xx
Uy2 xy
5 x :
yf U 2
Uy
The left-hand side is yU 1 ½hðfÞ; f=yf 5 Ux1 dh=yf 1 Uy1 . Around ðy*; x * Þ, we have

yU 1 ½hðfÞ; f 1 dh
* 5 Ux df 1 Uy
1
yf f→y2
" #
Uy1 x 2 Uxx2 1 2xyUxy2 1 y 2 Uyy2
5 :
x U2
Uy
As the last term is negative, Ux1 dh=yf 1 Uy1 < 0. Around ðy *; x * Þ, Ux1 =Uy1 5 2f=h,
and the inequality implies

dh f
< 1:
df h f→y2*
That is, around the unconstrained steady state, the elasticity of credit supply is
less than one.
The ðutilityÞ elasticity hðfÞ defined just before corollary 1 can be rewritten

f yU 1 ðhðfÞ; fÞ f 1 dh
5 U 1 U 1
U 1 ðhðfÞ; fÞ yf U 1 ðhðfÞ; fÞ x df y
dh f Ux h1 Uy f
1
5 1 :
df h U 1 U1

Around y *,

dh f Ux1 h Uy1 f Ux1 h dh f
1 5 21 :
df h U 1 U 1 f→y2* U 1 df h
The condition in corollary 1 is equivalent to
dh f U 1 ðx *; y* Þ
<12 :
df h Ux1 ðx *; y* Þx *
We conclude that if the elasticity of credit supply is small, cycles exist. QED
References
Aliprantis, Charalambos, Gabriele Camera, and Daniela Puzzello. 2006. “Matching

and Anonymity.” Econ. Theory 29 ð2Þ: 415–32.
———. 2007. “Anonymous Markets and Monetary Trading.” J. Monetary Econ. 54
ð7Þ: 1905–28.
Alvarez, Fernando, and Urban Jermann. 2000. “Efficiency, Equilibrium, and As-
set Pricing with Risk of Default.” Econometrica 68 ð4Þ: 775–97.
Aruoba, Boragan, Guillaume Rocheteau, and Christopher Waller. 2007. “Bar-
gaining and the Value of Money.” J. Econ. Theory 54 ð8Þ: 2636–55.
Azariadis, Costas. 1993. Intertemporal Macroeconomics. Oxford: Blackwell.
Azariadis, Costas, and Leo Kaas. 2007. “Asset Price Fluctuations without Aggre-
gate Shocks.” J. Econ. Theory 136 ð1Þ: 126–43.
———. 2013. “Endogenous Credit Limits with Small Default Costs.” J. Econ.
Theory 148 ð2Þ: 806–24.
Benhabib, Jess. 1992. Cycles and Chaos in Economic Equilibrium. Princeton, NJ: Prince-
ton Univ. Press.
Biais, Bruno, Thomas Mariotti, Guillaume Plantin, and Jean-Charles Rochet.
2007. “Dynamic Security Design: Convergence to Continuous Time and Asset
Pricing Implications.” Rev. Econ. Studies 74 ð2Þ: 345–90.
Boldrin, Michele, and Michael Woodford. 1990. “Equilibrium Models Displaying
Endogenous Fluctuations and Chaos.” J. Monetary Econ. 25 ð2Þ: 189–222.
Coles, Melvyn, and Abhinay Muthoo. 2003. “Bargaining in a Non-stationary Envi-
ronment.” J. Econ. Theory 109 ð1Þ: 70–89.
Coles, Melvyn, and Randall Wright. 1998. “A Dynamic Equilibrium Model of
Search, Bargaining and Money.” J. Econ. Theory 78 ð1Þ: 32–54.
DeMarzo, Peter, and Michael Fishman. 2007. “Agency and Optimal Investment
Dynamics.” Rev. Financial Studies 20 ð1Þ: 151–88.
Ennis, Huberto. 2001. “On Random Matching, Monetary Equilibria, and Sunspots.”
Macroeconomic Dynamics 5 ð1Þ: 132– 42.
———. 2004. “Macroeconomic Fluctuations and Bargaining.” J. Econ. Theory 115
ð2Þ: 322–40.
Ferguson, Niall. 2008. The Ascent of Money. New York: Penguin.
Gertler, Mark, and Nobuhiro Kiyotaki. 2010. “Financial Intermediation and Credit
Policy in Business Cycle Analysis.” In Handbook of Monetary Economics, vol. 3,
edited by Benjamin M. Friedman and Michael Woodford, 547–99. Amsterdam:
Elsevier.

Gu, Chao, Fabrizio Mattesini, Cyril Monnet, and Randall Wright. 2013. “Banking:
A New Monetarist Approach.” Rev. Econ. Studies 80 ð2Þ: 636–62.
He, Chao, Randall Wright, and Yu Zhu. 2013. “Housing and Liquidity.” Manu-
script, Univ. Wisconsin.
Hellwig, Christian, and Guido Lorenzoni. 2009. “Bubbles and Self-Enforcing
Debt.” Econometrica 77 ð4Þ: 1137–64.
Kehoe, Timothy, and David Levine. 1993. “Debt-Constrained Asset Markets.” Rev.
Econ. Studies 60 ð4Þ: 865–88.
———. 2001. “Liquidity Constrained Markets versus Debt Constrained Markets.”
Econometrica 69 ð3Þ: 575–98.
Kiyotaki, Nohuhiro, and John Moore. 1997. “Credit Cycles.” J.P.E. 105 ð2Þ: 211–48.
Lorenzoni, Guido. 2008. “Inefficient Credit Booms.” Rev. Econ. Studies 75 ð3Þ:
809–33.
MacLeod, W. Bentley, and James Malcomson. 1993. “Investments, Holdup, and
the Form of Market Contracts.” A.E.R. 83 ð4Þ: 811–37.
Matsuyama, Kiminori. 1990. “Sunspot Equilibria ðRational BubblesÞ in a Model
of Money-in-the-Utility-Function.” J. Monetary Econ. 25 ð1Þ: 137–44.
———. 2013. “The Good, the Bad, and the Ugly: An Inquiry into the Causes and
Nature of Credit Cycles.” Theoretical Econ., forthcoming.
Myerson, Roger. 2012. “A Model of Moral-Hazard Credit Cycles.” J.P.E. 120 ð5Þ:
847–78.
———. 2013. “Moral-Hazard Credit Cycles with Risk-Averse Agents.” Manuscript,
Univ. Chicago.
Rocheteau, Guillaume, and Randall Wright. 2013. “Liquidity and Asset-Market
Dynamics.” J. Monetary Econ. 60 ð2Þ: 275–94.
Sanches, Daniel, and Stephen Williamson. 2010. “Money and Credit with Lim-
ited Commitment and Theft.” J. Econ. Theory 145 ð4Þ: 1525–49.
Woodford, Michael. 1986. “Stationary Sunspot Equilibria in a Finance Con-
strained Economy.” J. Econ. Theory 40 ð1Þ: 128–37.
———. 1988. “Expectations, Finance Constraints, and Aggregate Instability.” In
Finance Constraints, Expectations, and Macroeconomics, edited by Meir Kohn and
Sho-Chieh Tsiang. New York: Oxford Univ. Press.


Endogenous Credit Cycles

Uploaded by

Copyright:

Available Formats

Endogenous Credit Cycles

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Endogenous Credit Cycles

Uploaded by

Copyright:

Available Formats

Endogenous Credit Cycles

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

This paper studies models of credit with limited commitment and,

[ Journal of Political Economy, 2013, vol. 121, no. 5]

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

II. The Model

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

A feasible contract must satisfy the participation constraints in the first

U 1 ðxt ; yt Þ ≥ 0 and U 2 ðyt ; xt Þ ≥ 0: ð3Þ

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

We consider both competitive markets and search-and-bargaining mar-

max U 1 ðxt ; yt Þ subject to pt xt 5 yt and ð5Þ: ð7Þ

Meanwhile, type 2, with no repayment issues, solves

max U 2 ðyt ; xt Þ subject to pt xt 5 yt : ð8Þ

Let ðx *; y* Þ denote equilibrium ignoring the repayment constraint, the

Ux1 ðxt ; yt Þxt 1 Uy1 ðxt ; yt Þyt 5 0; ð9Þ

Ux2 ðyt ; xt Þxt 1 Uy2 ðyt ; xt Þyt 5 0: ð10Þ

If y * ≤ ft , we can implement the unconstrained allocation ðx *; y* Þ. Oth-

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

if ft < y *; then xt 5 hðft Þ and yt 5 ft ;

It will be useful below to note the following: when yt ≤ ft is binding, xt is

is ambiguous. More importantly, the sign of

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

F IG . 1.—Steady state with ðAÞ fs > y * and ðBÞ fs < y*

Summarizing this discussion, we have established the following pro-

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

F IG . 3.—Cycle with ðAÞ f1 < f2 < y* and ðBÞ f1 < y* < f2

Proposition 4. In any n-period cycle, at least one periodic point has

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

F IG . 4.—A, A three-period cycle; B, chaotic dynamics

Prðst11 5 1jst 5 1Þ 5 j1 and Prðst11 5 2jst 5 2Þ 5 j2 . The economy is in

Vs;t1 5 U 1 ðxs;t ; ys;t Þ 1 b½js Vs;t11

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

The generalized repayment constraint is

lys;t ≤ fs;t ; ðbp=lÞ½js Vs;t11

Equilibrium in state s at date t is given as follows:

if fs;t < y *; then xs;t 5 hðfs;t Þ and ys;t 5 fs;t ;

The analogue to ð6Þ is

fs;t21 5 js b½ðp=lÞU 1 ðxs;t ; ys;t Þ 1 fs;t

This defines a bivariate dynamical system:

Definition 2. A sunspot equilibrium is given by nonnegative and

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

An increase in the debt limit at t 1 1 influences the economy at t in two

Heuristically, in Walrasian equilibrium, a buyer of good y is always

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

Corollary 3. As a special case, a 5 0 implies that cycles exist if

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

System ð23Þ is similar to ð14Þ, except now the debt constraint is y ≤

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

or the following linear transformation of our baseline constraint,

If p 5 0, this is a simple model of collateralized lending à la Kiyotaki-

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

This content downloaded from 110.33.247.84 on Tue, 16 Sep 2014 03:39:22 AM

Let f1 ; f2 ; : : : ; fn be the periodic points of an n-cycle. Suppose by way of con-

Let ðx ; y Þ denote equilibrium ignoring the repayment constraint, the

If y * ≤ ft , we can implement the unconstrained allocation ðx ; y Þ. Oth-

If we ignore the repayment constraint for a moment, the solution ðx ; y Þ sat-

If ft ≥ y , we can implement the unconstrained contract; if ft < y, let xt 5 hðft Þ