THE JOURNAL OF FINANCE
VOL. LVIII, NO.

2
APRIL 2003

Excessive Dollar Debt: Financial Development and

Underinsurance

RICARDO J. CABALLERO and ARVIND KRISHNAMURTHY n

ABSTRACT
We propose that the limited ¢nancial development of emerging markets is a
signi¢cant factor behind the large share of dollar-denominated external debt
present in these markets. We show that when ¢nancial constraints a¡ect bor-
rowing and lending between domestic agents, agents undervalue insuring
against an exchange rate depreciation. Since more of this insurance is present
when external debt is denominated in domestic currency rather than in dol-
lars, this result implies that domestic agents choose excessive dollar debt.
We also show that limited ¢nancial development reduces the incentives for
foreign lenders to enter emerging markets. The retarded entry reinforces the
underinsurance problem.

ALTHOUGH OBSERVERS STILL DEBATE THE CAUSES underlying recent emerging markets’
crises, one factor they agree on is that domestic ¢rms’ contracting of external
debt in dollars as opposed to domestic currency creates balance sheet mis-
matches that lead to bankruptcies and dislocations.1
The evidence is that most contracts between foreign lenders and borrowers in
emerging markets take the form of dollar debt (see Hausmann, Panizza, and
Stein (2001)). However, although foreign lenders must eventually be repaid in dol-
lars, in principle, there is no reason that these payments cannot be contingent on
the exchange rate. For example, contingencies can be added explicitly by index-
ing debt contracts, or implicitly, by foreign lenders receiving domestic currency
payments that they then convert into dollars. As a result, we are left asking why
the choice of dollar debt is in the best interests of borrowers in emerging mar-
kets. On the one hand, the attraction to dollar debt is that dollar interest rates
are lower than the domestic ones. On the other hand, dollar debt exposes ¢rms to

n
Caballero is from MIT and NBER, and Krishnamurthy is from Northwestern University.
Caballero thanks the NSF for ¢nancial support. We thank Philip Bond, Guillermo Calvo,
Bengt Holmstrom, Hugo Hopenhayn, Nisan Langberg, Adriano Rampini, Jean Tirole, and an
anonymous referee for comments. We also thank seminar participants at Lacea’s Summer
Camp in Buenos Aires, MIT, and the University of Rome conference on Information and Busi-
ness Cycles for comments. We thank Sandra Moore for editorial assistance. All errors are our
own.
1
Much of the analysis of the currency^balance sheet channel assumes that companies
choose dollar-denominated debt. See, for example, Chang and Velasco (1999), Krugman
(1999), Aghion, Bacchetta, and Banerjee (2001), and Caballero and Krishnamurthy (2001a).

867
868 The Journal of Finance

a balance sheet mismatch. Is the low price of dollar debt worth the balance sheet
risk for a ¢rm in an emerging market? Do prices allocate the risk e⁄ciently?
Should a policy maker be concerned that companies underprice the risk of dollar
debt and therefore take on too much of it? We address these questions in this
paper.
Analysis of this issue has, for the most part, centered on the (harmful) incen-
tives of the government.2 In the context of sovereign debt, Calvo and Guidotti
(1990) argue that once foreign lenders purchase domestic-currency-denominated
debt, governments have an incentive to devalue and reduce the real value of their
debt (see also Calvo (1996) and Allen and Gale (2000)). Foreign lenders rationally
anticipate this and avoid purchasing domestic currency debt. First, these expla-
nations seem most compelling for high in£ation countries (Latin America),
rather than the Asian countries where chronic in£ation was not a problem. More-
over, as Calvo (2000) points out, it is hard to extend this argument to private sec-
tor debt if we are interested in the connection between debt choices and ¢nancial
di⁄culties. The problem is that if balance sheet mismatches are indeed costly,
¢rms will prefer to introduce contingencies into their liabilities to avoid them.
In our model, all agents are risk neutral but demand insurance because they
face a risk of liquidation (or production interruptions) in bad states of the world,
and they might need resources at times when the country faces international bor-
rowing constraints. This demand arises from the observation, ¢rst made by
Froot, Scharfstein, and Stein (1993), that the anticipation of borrowing con-
straints in a dynamic setting motivates ¢rms to hedge. Since the exchange rate
also depreciates in bad states of the world, borrowing in domestic currency as
opposed to dollars provides more of this insurance.
We show that when ¢nancial constraints a¡ect borrowing and lending be-
tween domestic agents, their valuation of this insurance is less than its social
value. The undervaluation is because some agents who purchase the insurance
will not need it. In this case, the agent will sell his excess resources to those
who do need it. The ¢nancial constraint places a friction in this transaction. It
limits the amount that agents who need insurance can pay to those who provide
it, and places a wedge between the social valuation of insurance and the equili-
brium return to providing this insurance. In a dynamic setting, agents underva-
lue insurance and take on too much dollar debt.
Our result di¡ers from the sovereign debt literature cited above, because we
show that domestic ¢rms in ¢nancially underdeveloped economies will misvalue
the insurance a¡orded by borrowing in domestic currency. The fault lies with ¢-
nancial constraints in the private sector rather than a misguided government.
The result also explains why the dollar debt problem extends to the private sec-
tor’s debt choices, and why the private sector might undervalue indexing their
debt contracts.

2
Constraints on domestic currency external borrowing may also have a domestic policy
origin. Until recently, the Chilean tax code penalized external borrowing in domestic cur-
rency vis-a'-vis dollar-denominated borrowing.
 Excessive Dollar Debt 869

In the sovereign debt literature, lenders charge higher prices for lending in do-
mestic currency because of the sovereign moral hazard. In our model, foreign
lenders extend loans at actuarially fair prices. However, we show that the same
mechanism responsible for underinsurance can also a¡ect the supply decisions
of foreign lenders. We allow foreign lenders to pay a ¢xed cost to enter domestic
¢nancial markets. In this case, they are able to value more of the collateral of
domestic agents. We show that returns on entry are closely linked to the equili-
brium return on providing insurance to domestic agents against bad states of
the world. As a result, the distortion in the valuation of insurance by the domestic
agents also lowers entry by the foreign lenders.
Although our explanation for dollar liabilities is also driven by an insurance
mispricing mechanism, it is quite distinct from those that point out that ¢xed
exchange rates o¡er free insurance and creates moral hazard that distorts in-
vestment choices (see, e.g., Dooley (1997)). In these models, ¢xing the exchange
rate o¡ers free insurance to ¢rms that borrow in dollars and therefore en-
courages dollar borrowing.3 In our model, on the other hand, it is not government
misbehavior but ¢nancial underdevelopment that creates the private underinsur-
ance problem.This result may explain why the dollar debt problem extends across
emerging markets, regardless of exchange rate systems.4,5
In methodology, our paper relates to a growing literature on aggregate liquid-
ity shortages (Diamond and Dybvig (1983), Allen and Gale (1994), Holmstrom and
Tirole (1998, 2001), Krishnamurthy (2002), and Diamond and Rajan (2001)). Each
of these papers studies di¡erent macroeconomic and asset price consequences
of an aggregate liquidity shortage. The canonical model in this literature is
Diamond and Dybvig, who study banking structure and the e¡ects of runs on
aggregate liquidity. Allen and Gale present a model in which aggregate liquidity
shortages a¡ect asset price volatility, and endogenize the links between market
participation, aggregate liquidity, and asset prices.
Our modeling approach owes most to the Holmstrom and Tirole (1998, 2001)
model of aggregate liquidity in the context of ¢rms. Their papers motivate a role
for the state in the creation of liquid assets when there are aggregate shocks. Our
basic model economy relates to that in Caballero and Krishnamurthy (2001a),
whose central departure from the literature is that they consider two forms of
liquidity, one domestic and one international. In this sense, the paper also relates
to the recent work by Diamond and Rajan (2001) in which bank’s solvency con-
straints play a role similar to our domestic collateral in determining domestic
asset prices. In Caballero and Krishnamurthy, there are two forms of liquidity,

3
Distortions of private sector incentives due to free insurance is also behind the govern-
ment-bailout-type models, such as Burnside, Eichenbaum, and Rebelo (2001).
4
See, for example, the evidence of external dollar debt in ¢xed as well as £exible exchange
rate systems in Hausmann et al. (2001).
5
In most currency crises, governments run out of resources to bail out ¢rms and the ¢rms
that borrow in dollars end up being badly hurt. Thus, we are left with the question as to how
much free insurance a rational ¢rm can expect the government to provide. The free insurance
models require a government with deep pockets. Our explanation has the advantage of relying
only on the resources of the private sector.
870 The Journal of Finance

because foreign and domestic agents have di¡erent technologies to seize collat-
eral on nonrepayment of loans. Aside from the di¡erent substantive issue that
concerns us in this paper, the model of this paper builds the asymmetry between
domestic and foreign lenders from their di¡erent valuation of nontradable goods
rather than from an asymmetry in collateral valuation.
The paper proceeds as follows. Section I presents our model. Section II dis-
cusses the underinsurance result. Section III explores the connection between
underinsurance and domestic ¢nancial development. This section also serves as
a transition to Section IV, where we discuss external supply problems that may
arise in this context. Section Vconcludes.

I. The Model
Our model has two sets of agents, domestic entrepreneurs/¢rms and foreign
investors.There are three periods, which we de¢ne as t 5 0,1,2. All agents are risk
neutral and competitive. Domestic agents borrow from foreign investors, choose
the contingency in their liabilities, and invest in production at date 0. Then at
date 1, there are an idiosyncratic and an aggregate shock that determine the
funds required to continue production.The agents’ability to cope with this shock
depends on the value of their assets minus contracted liabilities. The question of
currency denomination of liabilities turns on whether liabilities are su⁄ciently
contingent to insure against this shock. At date 2, debts are fully repaid and all
agents consume.

A. Technology and Preferences

Domestic agents are ex ante identical and have equal access to the same pro-
duction technology. All production requires foreign (or dollar) goods and pro-
duces domestic (or baht) goods. Domestic agents have no dollars, so they must
borrow from foreigners to ¢nance all production. At date 0, a ¢rm borrows b0
dollars from a foreigner and creates capital of k at a cost of c(k). Thus,
cðkÞ  b0 : ð1Þ
To generate an interior solution, we assume that the function c(k) is convex and
increasing. Once created, the capital is ‘‘baht.’’ It generates domestic goods, and
its value as collateral varies with the exchange rate.
We note in advance that our model is entirely real. Hence, any allusion to the
exchange rate refers to the real exchange rate. We denote the exchange rate as e
(the formal de¢nition is below).
At date 2, if all goes well, capital generates Ak units of baht goods. However, as
part of the normal churn of the economy, production may be interrupted at date 1
by an idiosyncratic shock. If this happens, the ¢rm is required to import an addi-
tional unit of foreign goods per unit of capital to realize output of Ak baht. If a
¢rm chooses not to do so, then its output falls to aoA  a1D on the capital that is
not salvaged.Thus, if a ¢rm chooses to salvage a fraction yr1 of its capital units,
then its date 2 output is (1  y)ak1y Ak and the ¢rm imports yk units of goods.
 Excessive Dollar Debt 871

Firms that are a¡ected by this shock are distressed type, and those that do not
are intact type.
Let oA{l, h} be the aggregate state of the world at date 1. In the h-aggregate
state, no ¢rm su¡ers from a liquidity shock. However, when the aggregate state
is l, half of the ¢rms need to reinvest.The shock is countrywide in the sense that it
a¡ects a positive measure of ¢rms in the l-state, but it is idiosyncratic in that an
individual ¢rm has a probability of 0.5 of being a¡ected by it in the l-state. The
probability of the l-state is p, and that of the h-state is 1  p.
At date 2, the domestic entrepreneurs/¢rms repay the debts accumulated at
date 0 and date 1 out of production proceeds.They consume the excess.Their pre-
ferences are
U d ¼ cB þ cD cB ; cD 0; ð2Þ
B D
where c is consumption of baht goods, and c is consumption of dollar goods.
Unlike domestic agents, foreigners have preferences only over the consump-
tion of dollar goods at date 2,
U f ¼ cD : ð3Þ
Foreigners can lend dollars to domestic agents to ¢nance production at both date
0 and date 1. We assume they have large endowments of dollars at each of these
dates. We also assume that they have access to a storage technology for these en-
dowments, providing a gross rate of return of one. These assumptions pin down
the dollar risk-free interest rate at one.

B. Liability Denomination and Contingency

Domestic agents borrow at date 0 from foreigners, using contracts that are
fully contingent on the aggregate state:

DEFINITION 1 (CONTINGENT LIABILITY CONTRACT): For oA{l, h}, a date 0 contingent lia-
bility contract between a domestic ¢rm and foreign investor speci¢es date 2 repay-
ments, f o dollars, and date 0 funding of b0 dollars. Since foreign investors are risk
neutral, competitive, and the dollar interest rate is one,
b0 ¼ pf l þ ð1  pÞf h : ð4Þ
This de¢nition only allows for aggregate contingencies in the liability structure
of ¢rms. Flexibility in specifying liabilities contingent on the type of ¢rm (dis-
tressed/intact) could provide greater insurance. We assume that the identity of
¢rms that experience the date 1 production shock in the l-state is private informa-
tion of that ¢rm, and is not observable by lenders. (Caballero and Krishnamurthy
(2001b) examine this issue further.)
Although we de¢ne the repayments in units of dollars, this de¢nition does not
automatically mean that all debt is in dollars.The puzzling question in emerging
markets is why ¢rms take on so much noncontingent dollar debt from foreigners.
Since the repayments in De¢nition 1 are contingent on the aggregate state, they
are not the same as the repayments of a noncontingent dollar debt contract.
872 The Journal of Finance

We prove that in equilibrium, el4eh 5 1. Consider how noncontingent debt con-

tracts would be represented under De¢nition 1. A noncontingent dollar debt con-
tract speci¢es dollar repayments of f h 5 f lddollar. A noncontingent baht debt
contract has baht repayments of bh 5 blbbaht. If we convert these to dollar equiva-
lent repayments, we ¢nd that f h 5 bbaht and f l 5 bbaht/el. Since el41, for the same
dollar repayments in the high state (i.e., bbaht 5 ddollar), the baht contract has low-
er repayments in the l-state (i.e., bbaht/eloddollar).
Thus, we look at the liability denomination question as a contingency
question: How high are dollar contracted repayments in the h-state compared to
the l-state?

C. Credit Constraints and Collateral

Liability choices matter because ¢rms face credit constraints. We introduce
credit constraints by requiring ¢rms to post collateral to secure all ¢nancing.
We recall that foreigners do not value any baht goods because they do not
consume these goods. Therefore, eventual repayments to foreigners can
never be in the form of baht goods. We assume that domestic agents have
an exogenously speci¢ed endowment of foreign goods arriving at date 2 given by
w. Following the sovereign debt literature, we de¢ne w as international collat-
eral.
Although all of w is collateral, we assume that not all of the output from pro-
duction is collateral. That is, production returns (1  y)ak1yAk. In a perfect ca-
pital market, ¢rms can pledge all of this output to lenders. However, we assume
that the reinvestment at date 1 is not observable and veri¢able. Thus, courts can-
not verify the extra output of (A  a) k due to reinvestment and can only enforce
repayments up to ak. For y40, (1  y)ak1yAk is clearly larger than ak. This lim-
ited collateral assumption is central to our results.

ASSUMPTION 1 (COLLATERAL): Lenders demand collateral against all loans. Each

domestic ¢rm has international collateral of w dollar goods and domestic collateral
of ak baht goods.Thus, for each o, the total debt capacity of a ¢rm, measured in dollars,
is
ak
fo  w þ : ð5Þ
eo
The collateral value of the ¢rm depends on the exchange rate. Since ak is baht
collateral, as the exchange rate depreciates, the dollar value of this collateral
falls. Our question is: Since collateral is worth less in the l-state, and since ¢rms
will need resources to ¢nance their production shocks in this state, do ¢rms
match this collateral sensitivity by choosing the appropriate amount of contin-
gency in their liabilities?
We note that Assumption 1 rules out the possibility of equilibrium default
in our model. Lenders rationally anticipate the value of a borrower’s collateral
in each state of the world, and never demand repayments above this collateral
value.
 Excessive Dollar Debt 873

The assumption also implies that in equilibrium, foreign debt repayments of

f o will never exceed w. This is because foreigners only value w for consumption,
and if the country has contracted debt above w, it will have to default on this
debt. Since foreign lenders rationally anticipate this and there is never default
in the model, they never demand repayments above w.
Assumption 1 is all that is required to generate our results. However, we can
better explain decisions and equilibrium if we consider a slight variation of this
problem.
If a foreigner is repaid in baht goods at date 2, he will exchange the baht goods
for dollar goods at the exchange rate of e. However, he does not need to wait until
date 2. He could also exchange the claim on the date 2 baht goods for a claim on
date 2 dollar goods at date 0. That is, the foreign lender is indi¡erent between
waiting until date 2 to swap out of claims against domestic collateral and swap-
ping out of them at date 0.
Rather than having foreigners lend against ak and then swap these goods at
date 2 for some of the w dollar goods, we directly impose a constraint under
which foreigners only lend against w.We impose a similar constraint, that domes-
tic lending be only against the collateral of ak. These two assumptions are unne-
cessary for the workings of the model and our main results, but they do simplify
the exposition.

ASSUMPTION 1A n (FOREIGN LENDING): All foreign lending takes the form of liability con-
tracts that are fully secured by the foreign good collateral of w. Lending is default free,
so that
f o  w: ð6Þ

ASSUMPTION 1B n (DOMESTIC LENDING): Domestic ¢rms can lend to each other at

either date 0 or date 1. All domestic lending is fully secured by baht revenues. However,
the domestic ¢nancial market is underdeveloped, so that agents can only use ak of the
date 2 baht revenues to secure ¢nancing from another domestic agent,

f D;o  ak: ð7Þ

It turns out that since domestic agents are identical at date 0, there is no reason
for domestic agents to borrow or lend from each other against the ak at date 0. At
date 0, f D,o is always zero. This is convenient, because it allows us to restrict our
focus to the currency denomination of foreign liabilities. Domestic lending can
occur at date 1, but at this time, uncertainty is resolved and the contingency issue
is moot.

D. Decisions and the Credit Chain

We solve the decision problem of a ¢rm by backward induction. At date 0, the
¢rm borrows b0 funds to create capital of k. At date 1, there are two possible
states of the world. In the h-state, there are no shocks, and all ¢rms continue to
874 The Journal of Finance

produce Ak. Entrepreneurs repay f h and consume

V h ¼ Ak þ w  f h : ð8Þ

In the l-state, ¢rms divide into distressed and intact groups. A distressed ¢rm
raises funds to alleviate its production shock. A choice of yk will result in output
at date 2 of (1  y)ak1y Ak goods.To salvage a fraction y of distressed capital, the
¢rm must borrow and invest y k imported goods.
The ¢rm can do this in two ways. First, it can go to foreigners to raise addi-
tional funds. That is, the ¢rm can always directly raise

b1 ¼ f1l  w  f l ð9Þ

Second, the distressed ¢rms can turn to intact ¢rms for loans.
There is an asymmetry between domestic and foreign agents. Unlike foreign-
ers, domestic agents value the ak of baht output. Thus, the distressed ¢rm can
access foreign funds indirectly by borrowing from the intact ¢rms, who in turn
use their international collateral to borrow from foreign agents. Since the ex-
change rate is el, the distressed ¢rm can borrow a maximum amount of ak/el dol-
lars from intact ¢rms.This credit chain represents the domestic ¢nancial market
in our framework. By using this chain, the distressed ¢rm can aggregate the re-
sources of the economy and pledge this to foreigners, thus raising resources for
date 1 reinvestment.
The decision problem of a distressed ¢rm is

ðPIÞ Vsl  maxy;f l ;f D w þ ak þ ykðA  aÞ  f l  f1l  f1D

1 1
s:t: ðiÞ f1l  w  f l
ðiiÞ f1D  ak ð10Þ
f1D
ðiiiÞ yk ¼ þf1l el
ðivÞ 0  y  1:

Constraints (i) and (ii) are the international and domestic collateral constraints.
Constraint (iii) is that investment must be ¢nanced by the resources raised from
the debt issues of f1l and f1D . Constraint (iv) is purely technological.
An intact ¢rm at date 1 decides how much it will lend to the distressed ¢rm. If
the intact ¢rm lends xD 1 =e1 dollars at date 1 against collateralized baht goods of
xD
1 at date 2, then

xD
ðP2Þ Vil  maxxD w þ Ak þ xD
1 
1
el
 fl
1
xD
ð11Þ
s:t: 1
el
 w  f l:

The constraint is that the intact ¢rm can, at most, lend w  f l dollars to the dis-
tressed ¢rm.
Date 0 problem. At date 0, a ¢rm maximizes its expected pro¢ts over the events
of being either distressed or intact, and in either the low or the high state. Thus,
 Excessive Dollar Debt 875

the decision at date 0 is

ðP3Þ maxk;b0 ;f w ð1  pÞV h þ pðVsl þ Vil Þ=2
f h; f l  w
ð12Þ
s:t: b0 ¼ pf l þ ð1  pÞf h
cðkÞ  b0 :

E. Equilibrium and Exchange Rates

An equilibrium of this economy consists of date 0 and date 1 decisions, (k, b0,
f o) and ðy; f1o ; f1D ; xD o
1 Þ, respectively, and prices e . Decisions are solutions to the
¢rms’problems (P1), (P2), and (P3) given prices of eo. At these prices, the ¢nancial
market clears.
The only equilibrium price is the exchange rate. From the preferences of do-
mestic agents, the following must hold true.

LEMMA 1. Let cB and cD denote the equilibrium consumption of any intact entrepre-
neur in the domestic economy at date 2:

If cB, cD 40, then e 5 1.

If cB40, but cD 5 0, then eZ1.

The case of cB 5 0 and cDZ0 can never occur in our model, since production
always generates at least some baht and domestic agents must consume this baht.
The exchange rate is one as long as the solution is at an interior where domes-
tic agents consume both baht as well as dollar goods. However, if cD 5 0, the econ-
omy runs out of dollar goods and the exchange rate depreciates further to re£ect
this scarcity.
Since this is precisely the case we are interested in, we construct an equili-
brium in which this happens at date 1 in the low state, that is, where el4eh 5 1.
Equilibrium in the l-state.What pins down the exchange rate when the economy
runs out of dollar goods? Intact ¢rms have w  f l of dollar goods that they sell to
distressed ¢rms to use in production. Distressed ¢rms pay for these dollars by
selling f1D of baht to intact ¢rms. The exchange rate is the price in this trade
1 D 1
f ¼ ðw  f l Þel : ð13Þ
2 1 2
This exchange rate is really a date 2 forward exchange rate. Since the interna-
tional interest rate is one and interest parity must hold, the date 1 exchange rate
is the date 2 exchange rate divided by the gross domestic interest rate.The model
has a free parameter in that we need not pin down the domestic interest rate. By
choosing this interest rate to be equal to one, we can call eo the date 1 exchange
rate as well.
A distressed ¢rm that borrows against its international collateral to salvage
its capital generates A  a units of baht goods at date 2 per unit of foreign debt.
876 The Journal of Finance

We let D  A  a be the baht return to salvaging one unit of capital. Since the in-
ternational interest rate is one, as long as DZ1, the distressed ¢rm chooses to
borrow as much as it can against its international collateral (b1 5 w  f l).
If the amount raised from foreign investors, w  f l, is less than the funds
needed for salvaging all of its capital, k, then the ¢rm will have to access the do-
mestic ¢nancial market to make up the shortfall. It can sell up to ak date 2 baht to
another domestic agent at the exchange rate of el. It will choose to do this as long
as the baht return on restructuring exceeds the exchange rate (DZel). The max-
imum amount of funds raised is
f1D ak
 l: ð14Þ
el e
As long as the sum of ak/el and w  f l is more than the borrowing need of k, the
¢rm is unconstrained in its reinvestment at date 1, and all production units will
be salvaged. In this case, the ¢rm will borrow less than ak/el in domestic ¢nancial
markets.
Intact ¢rms will lend dollars to distressed ¢rms as long as the exchange rate
weakly exceeds one (elZ1). The most that intact ¢rms can lend is their excess in-
ternational collateral of w  f l.
If we assume that DZelZ1, then distressed ¢rms will borrow as much as they
can and intact ¢rms lend as much as possible. In total, the economy imports
w  f l goods, which are all lent to the distressed ¢rms. A necessary condition
for all production units to be salvaged is that k2  w  f l. For a given equilibrium
yr1, we refer to the constraint
k
y  w  fl ð15Þ
2
as the international collateral constraint.
As long as the international collateral constraint does not bind, both cB and cD
are positive. From Lemma 1, we note that this will mean that el 5 1. However, if
the constraint does bind, the economy will have sold all of its dollar goods, and
from Lemma 1, we see that the exchange rate exceeds one.
Figure 1 represents the market clearing for the case in which the international
collateral constraint of (15) binds (for yo1).The supply of dollars from intact ¢rms
is elastic at the international interest rate of one, up to 12 ðw  f l Þ. At this point,
the economy has no more international collateral and so the supply of dollars
turns vertical. The ¢gure represents equilibria at points A and B. The points are
distinguished by whether the distressed ¢rms are credit constrained or not.
Given w  f l, in both equilibria y is the same and less than one. However, in the
case where (14) does not bind, the exchange rate is equal to D (this case is
represented by the dashed upper line for demand corresponding to point B). In
the other case (the downward sloping solid curve corresponding to point A), the
exchange rate is
ak
1 oel ¼ oD: ð16Þ
w  fl
 Excessive Dollar Debt 877

Figure 1. Market clearing in the l-state.

Since we are interested in equilibria in which the exchange rate is depreciated in

the l-state, we assume that the international collateral constraint of (15) binds in
this state. We are interested in the distinction in outcomes between the cases
where (14) does and does not bind.

LEMMA 2. (EXCHANGE RATES)

In the h-state, the international collateral constraint does not bind. Therefore,
eh 5 1.

In the l-state, the international collateral constraint binds. Therefore,


l ak
e ¼ min D; 41: ð17Þ
w  fl
If (14) binds, then eloD.

We must also make assumptions such that date 0 investment in capital is

su⁄ciently pro¢table and gives an interior solution (koc  1(w)). We provide
the assumptions on primitives required to generate these equilibria in the
Appendix.

II. Underinsurance: Excessive Dollar Debt

Firms contract to make contingent debt repayments in dollars of f h and f l.
There are two states of the world, and noncontingent dollar and baht debt have
878 The Journal of Finance

linearly independent repayments. Thus, spanning results apply and the contin-
gent repayments of (f h, f l) can be implemented by contracting in a mixture of dol-
lar and baht debt. There is excessive dollar debt when a central planner would
choose a lower fraction of dollar debt than would private agents.

DEFINITION 2: If f h and f l are debt repayment choices in the competitive decentralized

equilibrium, and F h and F l are debt repayment choices of a central planner, then the
economy has excessive dollar debt if
fh Fh
o : ð18Þ
fl Fl

A. Competitive Equilibrium versus Planner’s Choice

To arrive at the program for a ¢rm at date 0, we substitute the value functions
from (P1) and (P2) into (P3). Firms solve their decision problem, given exchange
rates of eh 5 1 and el41 (as in Lemma 2).
If a ¢rm chooses (k, f h, f l), it will make date 2 pro¢ts (net of any contracted
debt) in the h-state of Vh 5 Ak1w  f h (equation (8)). In the l-state, if the ¢rm is
distressed, the date 2 pro¢ts are
ak
Vsl ¼ ðw  f l ÞD þ D: ð19Þ
el
(w  f l) is pledged to foreigners and the proceeds are invested at the project re-
turn of D. The ak of domestic collateral is sold at the exchange rate of el and the
proceeds are invested at D. If the ¢rm is intact, date 2 resources are

Vil ¼ ðw  f l Þel þ Ak: ð20Þ

Combining these results, the date 0 program is

ðP4Þ maxk; f h ; f l ð1  pÞðAk þ w  f h Þ þ p12ððA þ aeDl Þk þ ðD þ el Þðw  f l ÞÞ

s:t: f h ; f l  w
cðkÞ  pf l þ ð1  pÞf h : ð21Þ
In both h and l states, the ¢rm can increase its liabilities up to a maximum of w.
The bene¢t of increasing f h by one dollar is that the ¢rm raises 1  p dollars at
date 0. This dollar is used to increase capital by (1  p)/(c 0 (k). The cost of doing so
is that there is one dollar less in the h-state, which reduces date 2 consumption by
one dollar.The ratio of bene¢t (in units of increased capital) to cost of increasing
f h is:
ð1  pÞ=c0 ðkÞ 1
¼ 0 : ð22Þ
1p c ðkÞ
Now we consider the same exercise in the l-state. Increasing f l by one dollar
raises p dollars at date 0. This dollar is used to increase capital by p/c 0 (k).
However, since in the l-state ¢rms have to ¢nance their production shock, the cost
 Excessive Dollar Debt 879

di¡ers from that in the h-state. From the objective in (P4), we see that the cost is
p(D1el)/2. Thus, the same bene¢t-to-cost ratio in the l-state is:
p=c0 ðkÞ 1
¼ : ð23Þ
pðD þ el Þ c0 ðkÞðD þ el Þ=2
Comparing these last two expressions con¢rms our intuition that since ¢rms will
need resources to ¢nance their production shock in the l-state, it is costlier to
have more liabilities in this state ((D1el)/241).
However, this cost is lower than that which a central planner would compute.
In the l-state, a dollar certainly returns D when used in production. Since eloD,
the cost term for ¢rms is strictly less than the planners ((D1el)/2oD). As a result,
the planner will have ¢rms choose to contract less liabilities in the l-state than in
the competitive equilibrium outcome.
We con¢rm this intuition more formally by constructing the program of a cen-
tral planner who maximizes an equally weighted sum of the utilities of the do-
mestic agents, subject to the domestic and international collateral constraints.
Suppose the central planner makes a date 0 choice of (K, F h, F l) (capital letters
denote the central planner’s aggregate quantities). At date 2, in the high state, all
¢rms earn pro¢ts of Vh 5 AK1w  F h. In the low state, a distressed ¢rm’s pro¢ts
are (w  F l)D1(aK/el)D (equation (19)). For the planning problem, we construct
an objective that is free of prices. Substituting the market clearing condition of
el 5 aK/(w  Fl) into this pro¢t expression, we obtain:

Vsl ¼ 2ðw  F l ÞD: ð24Þ

l l
Similarly, an intact ¢rm’s pro¢ts are (w  F )e 1AK (equation (20)). After substi-
tuting in the market clearing condition, this becomes:

Vil ¼ ða þ AÞK: ð25Þ

The e⁄cient debt choices in this economy are given by the solution to
ðP5Þ maxK;F h ;F l ð1  pÞðAK þ w  F h Þ þ p12ððA þ aÞK þ 2Dðw  F l ÞÞ
s:t: F h ; F l  w
cðKÞ  pF l þ ð1  pÞF h : ð26Þ
We now compare the bene¢ts/costs of increasing liabilities. Starting with the h-
state, since the objectives corresponding to the h-state are the same across (P4)
and (P5), the bene¢t/cost computation for both the planner and ¢rms in increas-
ing f h is the same.

LEMMA 3. In both (P4) and (P5), F h 5 f h 5 w.

Proof: See the Appendix.

Since there is no chance of a liquidity shock in the h-state, there is no reason to

leave a slack in the debt repayment. Optimality requires ¢rms to borrow as much
as possible against w in this state and use the proceeds to increase K at date 0.
880 The Journal of Finance

In the l-state, we con¢rm that the choices over liability diverge:

LEMMA 4. If D4el, then F lof l, or debt repayments are set too high in the l-state in the
decentralized equilibrium.

Proof: See the Appendix.

In the objective in (P4), f l is multiplied by (D1el)/2. In the objective in (P5), f l is

multiplied by D.When eloD, the latter is bigger and we arrive at the lemma.

PROPOSITION 1 (EXCESSIVE DOLLAR DEBT): Suppose that the international collateral con-
straint of (15) binds in the l-state. If the domestic collateral constraint of equation (14)
binds, so that eloD, then ¢rms contract excessive dollar debt. If (14) does not bind so
that el 5 D, then debt choices are e⁄cient.

This proposition follows from Lemma 2, eloD only if the domestic collateral
constraint of (14) binds.

B. The Externality
The planner’s choice di¡ers from the competitive equilibrium because of an
externality that arises when there are domestic collateral constraints.
The market price of a dollar in the l-state at date 1 is given by el. The marginal
value of this dollar in production is D. The di¡erence between these two valua-
tions is responsible for the underinsurance result.
When (14) binds, the demand for dollars is depressed because the ¢rms in need
of dollars are credit constrained. This depressed demand distorts the market
price of a dollar relative to its social value. If (14) does not bind, then the dis-
tressed ¢rms bid up the price of dollars towards their marginal product of D
and there is no distortion.
The distorted price a¡ects the quantity of insurance purchased.The insurance
decision is a date 0 decision to save one dollar into the l-state. If the ¢rm turns out
to be distressed, it uses this dollar in production to return D at date 2. However, if
the ¢rm is intact, the distorted price comes into play: The ¢rm must sell the dollar
at the price of eloD and fetches less than the social marginal product of D. Ex
ante, this translates into underinsurance and the excessive dollar debt result.6

6
We note that there is another factor that reinforces the underinsurance in our model.
When eloD, distressed ¢rms sell their domestic collateral at (ak)/el as opposed to (ak)/D. We
can show that this is an overvaluation, relative to the planner, of the collateral created by k
investment. As a result, ¢rms overborrow and overinvest at date 0. Although, we do not focus
on this aspect of underinsurance because our interest in this paper is in understanding how
liability choices (as opposed to asset/investment choices) are a¡ected by ¢nancial develop-
ment, we can show that when there are more than two aggregate states, the latter e¡ect leads
to overborrowing but does not a¡ect insuring against the l-state.
 Excessive Dollar Debt 881

III. Financial Development and Underinsurance

The main result of the previous section is that the excessive share of dollar
debt in the liabilities of ¢rms in emerging markets arises because credit con-
straints a¡ect borrowing/lending relationships among domestic agents. We now
consider an economy with a mix of ¢rms that face no credit constraints in their
domestic borrowing and the constrained ones of the previous section. We model
¢nancial development as increasing in the fraction of ¢rms that are not credit
constrained.
We simplify the analysis by ruling out domestic insurance markets contingent
on aggregate shocks, which will naturally arise when ¢rms are ex-ante heteroge-
neous. The results in this section are robust to relaxing this simpli¢cation.

A. Constrained and Unconstrained Firms

Assume that a fraction l of the domestic ¢rms face no constraints on domestic
borrowing. For these ¢rms, the date 1 domestic collateral constraint is

f1D  ðð1  yÞa þ yAÞk: ð27Þ

For y40, it is clear that ((1  y)a1yA)k4ak.Thus, these ¢rms are less credit con-
strained than those of (14). In fact, we show in the Appendix that since these ¢rms
are able to pledge all of their baht output as collateral, the domestic collateral
constraint of (27) will never bind for them.

LEMMA 5. (27) will never bind for unconstrained ¢rms.

Proof: See the Appendix.

Next, we consider the date 0 program for these ¢rms. As before in the h state,
Vh 5 Ak1w  f h. In the l state, if the ¢rm is intact, it makes pro¢ts of Vil ¼
Ak þ ðw  f l Þel . If the ¢rm is distressed, it makes pro¢ts of Vsl ¼ akþ
ðD  el Þk þ ðw  f l Þel , because the ¢rm is able to salvage all of its capital units
by borrowing k dollars at the exchange rate of el, and generating D baht at date 2.
Combining these expressions yields the date 0 program of an unconstrained
¢rm:
l
ðP6Þ maxk;f h ;f l ð1  pÞðAk þ w  f h Þ þ pððA  e2 Þk þ el ðw  f l ÞÞ
s:t: f h; f l  w ð28Þ
cðkÞ ¼ pf l þ ð1  pÞf h :

PROPOSITION 2 (FINANCIAL DEVELOPMENT AND EFFICIENCY): If l 5 1 and the interna-

tional collateral constraint binds in the l-state, then el 5 D, eh 5 1, and

f h FH
¼ L: ð29Þ
fl F
882 The Journal of Finance

Proof: Suppose in contradiction that D4el.Then the distressed ¢rm will choose
to borrow the maximum amount and salvage production units. We use Lemma 5
and note that (27) will never bind. As a result, distressed ¢rms will issue debt and
salvage all of their production units (y 5 1). Thus f1D =el ¼ k2. However, since (15)
binds, y k2 ¼ w  f l for yo1. This implies that
f1D
4w  f l ; ð30Þ
el
which violates market clearing.There is excess demand for dollars at date 1. As a
result, it must be that D 5 el.
We substitute el 5 D into the program for a ¢rm at date 0 in (P6).
maxk;f h ;f l ð1  pÞðAk þ w  f h Þ þ pðAþa l
2 k þ Dðw  f ÞÞ
s:t: f h; f l  w ð31Þ
cðkÞ  pf l þ ð1  pÞf h :
This program is identical to that of (P5). Hence, if l 5 1, the economy makes e⁄-
cient debt choices.

This proposition clari¢es the main result of the previous section: Since collateral
is limited to ak in Assumption 1, ¢rms are constrained in their domestic borrow-
ing. This causes the distortion in prices and results in underinsurance.When all
of the baht output of ¢rms can be pledged as collateral, market prices re£ect the
social marginal product and insurance decisions are chosen optimally.
We conclude by showing that for the intermediate cases of lo1, the debt
choices are monotone in l. As ¢nancial development rises and more ¢rms are
unconstrained, the debt choices feature more insurance.

PROPOSITION 3 (FINANCIAL DEVELOPMENT AND UNDERINSURANCE): Consider two econo-

mies indexed by l and l 0, where l4l 0, and in both economies the international colla-
teral constraint binds in the l-state. Then, for both constrained and unconstrained
¢rms,
 
f h  f h 
: ð32Þ
f l l f l l0

Proof: See the Appendix.

IV. Limited Foreign Insurance: Further Costs of Domestic Financial

Underdevelopment
The general principle behind our result is that credit constraints leads to con-
strained demand for funds. Those in need of funds are not credible in transfer-
ring the surplus created by these funds to the lenders. In a dynamic context, the
latter ¢nd that the business of lending to ¢rms with bad collateral is not pro¢t-
able, and so they transfer their resources elsewhere. We apply this principle to
 Excessive Dollar Debt 883

explain the limited entry of specialist foreign lenders into domestic markets (i.e.,
credit line facilities, foreign banks).
So far, we have modeled foreigners as passive lenders who make no pro¢ts and
willingly lend in either currency. This characterization of foreigners (and many
domestic savers) is hardly realistic. This section extends the model to study the
e¡ects of ¢nancial development on foreign lending decisions. We introduce an
active margin whereby foreign lenders may choose to pay a ¢xed cost and specia-
lize in lending to the domestic market.

A. Foreign Specialists
We return to the model of Section III, with 0olo1. We divide foreign lenders
into two classes, specialists and nonspecialists. The specialists value baht goods
as do domestic agents:

U S ¼ cD þ cB : ð33Þ

Nonspecialists are exactly like the foreign lenders of the previous sections. They
value only dollar goods, that is, U 5 cD.
Unlike the nonspecialist, a specialist can invest in loans backed by domestic
baht collateral. This modi¢cation captures the idea that specializing in the do-
mestic market enables a foreign lender to receive higher returns on lending to
domestic agents.We assume that all lenders (both specialists and nonspecialists)
have a date 0 endowment of wf dollars.There is a continuum of measure a of these
specialists, a will shortly be endogenized by positing a cost of specializing.

B. Specialist Lending as Insurance

DEFINITION 3 (SPECIALIST LENDING CONTRACT): A contract between a foreign specialist
and a domestic ¢rm speci¢es repayments of ðfSh ; fSl Þ and initial loan of b0.

b0 q ¼ ð1  pÞfSh þ pfSl ; q el : ð34Þ

The collateral constraints for this lending contract are

fSo  Ak þ w ð35Þ

if the ¢rm is unconstrained (in domestic markets), and

fSo  ak þ w ð36Þ

if the ¢rm is constrained.

In this de¢nition, we have accounted for specialist lending against baht collat-
eral by expanding the collateral constraint to include the baht output.
The required return of the specialist lender is qZel (in (34)).This is because the
specialist has a high return investment opportunity in the l-state. He can lend
one dollar-good and receive el41 baht-goods in return at date 2. If the specialist
884 The Journal of Finance

converts all of his wealth into date 1 dollars in the l-state (e.g., by investing with a
risk-neutral nonspecialist), he will earn the return of el on his wf. Thus, the spe-
cialist lender must receive at least this return on date 0 lending.
Consider the problem of a constrained ¢rm. This ¢rm chooses to borrow from
both specialists and nonspecialists at date 0.We modify (P4) to re£ect this:

ðP7Þ maxk;f h ;f l ;f h ;f l ð1  pÞðAk þ w  f h  fSh Þþ

S S
 
Aþa Dl l Dþel Dþel l
p 2
e
k  f S 2el þ 2 ðw  f Þ

s:t: f h ; f l  w ð37Þ

0  fSh ; fSl  ak
cðkÞ  pf l þ ð1  pÞf h þ q1 ðpfSl þ ð1  pÞfSh Þ:

As in the previous sections, we shorten the collateral constraint for specialist

loans (the second constraint) to being constrained only by ak.This is without loss
of generality for the same reason as in previous sections.
The following lemma describes the insurance features of specialist lending.

LEMMA 6. (LENDING CONTRACT AS INSURANCE)

Consider an economy in which eloD and 0oaoe, where e is positive but small. In this
case, fSh 40 and fSl ¼ 0. In addition, the return to specialists is q 5 (D1el)/2.

Proof: See the Appendix.

In the economy without specialists, the baht collateral of ¢rms in the h-state is
never borrowed against. Since specialists value that collateral, their advantage
vis-a-vis nonspecialists is lending against the h-state collateral. This results in
fSh 40.
Since specialists are limited, they charge the premium of qZel on their lending.
Since nonspecialists lend at the international interest rate of one and f low, a
¢rm prefers to increase borrowing from a nonspecialist before it borrows from
a specialist. This is why ¢rms choose not to borrow against the l-state from non-
specialists ( fSl ¼ 0). Specialists provide more contingency and insurance
than nonspecialists. In this sense, their lending is more domestic currency
denominated.

PROPOSITION 4 (SPECIALIST LENDING AND INSURANCE): Consider two economies indexed

by a and a 0, where a4a 0, eloqoD in both economies. Then
P   P  
fjh þ fS;
h  fjh þ fS;
h 
j2fconst;unconstg lj j  j2fconst;unconstg lj j 
      : ð38Þ
P  P 
j2fconst;unconstg lj fj þ fS; j  j2fconst;unconstg lj fj þ fS; j 
l l l l
a a0
 Excessive Dollar Debt 885

Proof: See the Appendix.

As the mass of specialists increases, constrained ¢rms raise their borrowing

against the h-state. Moreover, these proceeds are used both in increasing k and
insuring against the l-state. This can happen directly by receiving dollars in the
l-state from specialists, or indirectly by reducing f l from the nonspecialists.
There is an indeterminacy in which of the lenders provides the l-state insurance.
However, in total, the liability structure of the ¢rms provide more insurance as
the mass of specialists rises.7

C. Equilibrium, Entry, and Financial Development

We endogenize a and describe how ¢nancial development a¡ects a and lending
premia. Assume that specializing costs C. As a result, entry yields expected uti-
lity of V e 5 (w f  C)q and nonentry yields utility of V ne 5 w f. The free entry con-
dition is that a lenders choose to specialize such that, in equilibrium,
V e ¼ V ne : ð39Þ
The return to specialists of q is a function of both the equilibrium amount of en-
try and the exogenous level of ¢nancial development (l). In the Appendix, we
prove the following comparative statics.

LEMMA 7. (q, a, and l)

As more specialists enter the market, q falls:
@q
o0:
da
As more ¢rms in the economy are unconstrained, q rises:
@q
40:
dl
The ¢rst result is that as more specialists enter the market, the return to the
marginal entrant falls, and q falls.We use previous results to establish the second
comparative static. In Lemma 7 we noted that q was equal to (D1el)/2 for small a.
In the Appendix, we show that this positive relation between q and el holds more
generally for any a. In Section III, we show that el was increasing in l. As a result,
q is also increasing in l. In a more developed ¢nancial market, ¢rms have more
domestic collateral and the return on lending to these ¢rms rises.
7
We note that the proposition applies only to the case in which q4el. This is because when a
is large, q 5 el and unconstrained ¢rms will also be borrowing from specialists in equilibrium.
Since these ¢rms are unconstrained, they are indi¡erent between borrowing against h-state
collateral or l-state collateral. As a result, the liability structure is indeterminate, and we are
unable to make statements about the insurance features of the liabilities. If we assume that
they always borrow against h-state collateral, then the proposition continues to hold. In terms
of welfare, increasing a is always bene¢cial, since it leads to more insurance between the h-
and l-states.
886 The Journal of Finance

Figure 2. Entry by foreign specialists.

Figure 2 represents the equilibrium for the entry decision.8 The solid line d is
the demand for foreign specialist funds as a function of q. The line d 0 represents
demand in an economy which is more ¢nancially developed (i.e., l is higher). In-
spection of the ¢gure leads to Proposition 5.

PROPOSITION 5 (FINANCIAL DEVELOPMENT AND SPECIALIST ENTRY): Financial develop-

ment increases the entry of specialists into the domestic lending market (a is increas-
ing in l).

D. Financial Development and Lending Premia

The limited foreign entry described in the previous section can feed back into
the price charged by foreign specialists through a thin-market externality. Sup-
pose foreign lenders prefer to lend in a market in which there are already other
foreign lenders. Allen and Gale (1994) provide a microeconomic model for this
phenomena. We defer to their results, and take a reduced form approach to this
issue by positing complementarity in foreign entry decisions. Suppose that C is a
function of the amount of entry:


C if a  ^a
CðaÞ ¼
C if a4^a:

8
Foreigners require that,
D þ el 1
¼q¼
2 1  wCf
in order to enter. As a result, a is such that the equilibrium exchange rate is el ¼ 2 11 C  D.
wf
 Excessive Dollar Debt 887

Figure 3. Complementary entry decisions.

Figure 3 illustrates the e¡ect of complementarity. At the low level of ¢nancial

development corresponding to d, there is the possibility that foreign specialists
anticipate limited entry by others and stay out of the market. However, if the mar-
ket is su⁄ciently developed (d 0 ), this possibility disappears, since specialists ex-
pect others to enter and therefore enter themselves. Comparing across these two
cases, we see that the lending premia fall as l rises.

V. Conclusion
We began the paper with the following question: The large share of external
debt in emerging markets that is dollar denominated has played a central role
in most recent crises. However, since this is a private decision, why do ¢rms ex-
pose themselves to the risks of dollar debt?
We answer this question by showing that the choice over liability denomination
was equivalent to a choice over how much insurance to purchase against states of
the world when international collateral is scarce.The central result of our analy-
sis is that when domestic ¢nancial markets are underdeveloped, the private va-
luation of this insurance will be distorted relative to a planner’s valuation. The
distortion leads to underinsurance.
If there is a drop in returns to providing insurance, then the supply of this in-
surance by foreign specialists also falls. Countries with limited ¢nancial devel-
opment also have fewer foreign credit lines and foreign lending in domestic
currency. This situation is exacerbated by complementarities in the lending deci-
sions of foreign specialists.
The primitive result in our analysis is one of underinsurance. Denominating
external liabilities in dollars is just one manifestation of this underinsurance.
888 The Journal of Finance

Other forms of underinsurance include the limited availability of external credit

lines and the large amount of short-term external debt in emerging markets. We
conjecture that the manifestation of underinsurance in a particular country de-
pends on institutional factors, chief among which is the exchange rate system.We
are currently investigating this issue.
Finally, the dollar-debt problem we have discussed is due to an externality. Our
investigations into the role of the government to correct this externality have
proven fruitful. We are able to rationalize some canonical government policies
such as capital in£ow taxation or liquidity requirements as Pareto improving in
some situations, although not without drawbacks (Caballero and Krishnamurthy
(2001b)).We are also able to show that well-designed monetary and international
reserve management policies can be e¡ective (Caballero and Krishnamurthy
(2002)). Without a theory for why governments may undertake these policies, it
has not been possible to study the relative merits of these policies. We are using
our framework to analyze these issues currently.

Appendix
Proof of Lemma 3: We form the Lagrangian for (P5),

1
Ln ¼ð1  pÞðAK þ W  F h Þ þ p ððA þ aÞK þ 2DðW  F l ÞÞ
2 ðA1Þ
lðcðKÞ  pF l  ð1  pÞF h Þ  mh ðF h  WÞ  ml ðF l  WÞ:

First,

@Ln Aþa
¼ ð1  pÞA þ p  lc0 ðKÞ ¼ 0: ðA2Þ
@K 2

Likewise, if mh 5 0,
 
@Ln 1p Aþa 0
¼ ð1  pÞ þ lð1  pÞ ¼ ð1  pÞA þ p  c ðKÞ 40: ðA3Þ
@F h c0 ðKÞ 2

We substitute in l from above and note that the project is su⁄ciently pro¢table at
date 0 to arrive at the inequality. Since @Ln =@F h 40, it must be that F h 5W. The
same proof applies for (P4).

Proof of Lemma 4: We note that,

cðKÞ  ð1  pÞw
Fl ¼ ðA4Þ
p

cðkÞ  ð1  pÞw
fl ¼ : ðA5Þ
p
 Excessive Dollar Debt 889

From the FOC,

1 p 
c0 ðKÞ ¼ ð1  pÞA þ ðA þ aÞ ðA6Þ
D 2
 
0 2 p D
c ðkÞ ¼ l ð1  pÞA þ ðA þ a l Þ : ðA7Þ
e þD 2 e
If eloD then c 0 (k)4c 0 (K) and c(k)4c(K). From the ¢rst set of equations, this also
means that f l4F l.

Proof of Lemma 5: If the unconstrained ¢rms salvage y units of capital by issu-

ing debt that raises yk dollars, then the maximum amount of dollars they can
raise is,
ð1  yÞa þ yA A
l
k4y l k: ðA8Þ
e e
Since
A ¼ a þ D4D el ; ðA9Þ
we conclude that
ð1  yÞa þ yA
k4yk: ðA10Þ
el
Thus, given any y, unconstrained ¢rms will always be able to obtain the funds
required to restructure all of their capital units.

Proof of Proposition 3: From (P6), the FOC for an unconstrained ¢rm is (the
‘‘hat’’denotes choices for unconstrained ¢rms),
 
el ^Þ:
ð1  pÞA þ p A  ¼ el c0 ð k ðA11Þ
2
Note that k^ is strictly decreasing in el. Also, from the budget constraint

^Þ ¼ pw þ ð1  pÞ^f l
cðk ðA12Þ
^l l
f is also strictly decreasing in e . From the same program for constrained ¢rms,
the FOC is
 
A D D þ el 0
ð1  pÞA þ p a l ¼ c ðkÞ: ðA13Þ
2 2e 2
Again we conclude that f l and k are strictly decreasing in el.
We know that if el 5 D, the private sector debt choices coincide with the e⁄cient
choices. If we take the other case where eloD, the market clearing condition in
the l-state is
ak
^þ
lk ð1  lÞ ¼ ð1 þ lÞw  f l ð1  lÞ  2^f l l: ðA14Þ
el
890 The Journal of Finance

We want to prove that f h/f l (for both constrained and unconstrained ¢rms) is
weakly increasing in l. This is obviously true when el 5 D. In the other case, we
construct a proof by contradiction. Suppose not, then for two economies in which
0 0
l4l 0 , we have that eloel . However if this is true, then ^f l 4^f l , k
^ 4k
^0, f l4f l 0 , k4k 0 .
l l0
From the market clearing condition, e 4e , which is a contradiction.

Proof of Lemma 6: For a ¢xed k, we consider the amount of money raised at date
0 by increasing fso in (P7), and compare this to the cost in terms of the objective.
Increasing fSh raises ð1  pÞ q1 more resources at date 0 and incurs costs of (1  p) in
the objective. This gives a gross borrowing cost of q. Increasing f h raises (1  p)
more resources at date 0 and costs (1  p). Increasing fSl raises p q1 more resources
at date 0, but costs p[(el1D)/2el]. The gross borrowing cost is ðq=el Þ½ðD þ el Þ=2.
We compare each of these to the cost/bene¢t ratio of increasing f l.This raises p
resources at date 0 and costs p[(D1el)/2] in the objective for a gross borrowing
cost of (D1el)/2. Since

D þ el q D þ el
o l ;
2 e 2
the constrained ¢rm will always choose fSl ¼ 0. Borrowing from the specialist in
the l-state is dominated by borrowing from the nonspecialist.
However, since specialists lend in equilibrium, fSh 40. For small values of a, the
return to specialist lending is determined by the value of dollars to ¢rms in the l-
state.Thus, q 5 (D1el)/24el. Also, at this price, unconstrained ¢rms choose not to
borrow from specialists.

Proof of Proposition 4: We ¢rst write the program for an unconstrained ¢rm:

maxk; f h ; f l ; f h ; f l ð1  pÞðAk þ w  f h  fSh Þþ

S S  l
 
: p A  e2 k  fSl þ el ðw  f l Þ
s:t: f h; f l  w ðA15Þ
0  fSh ; fSl  Ak
cðkÞ  pf l þ ð1  pÞf h þ q1 pfSl þ ð1  pÞfSh :

In the region where q4el, the unconstrained ¢rms will not borrow from specia-
lists. Thus, all lending by specialists must go to constrained ¢rms.
To show that ratio of h-state to l-state liabilities rises with a, we need to show
that ð1  lÞf l þ l^f l falls. This is because we know from Lemma 6 that for the con-
strained ¢rms, fSh will rise with a, while fSl ¼ 0.
We show the result in two steps. First we show that el falls as a increases, and
second, we show that this implies that ð1  lÞf l þ l^f l falls.
Consider an increase in the mass of specialists of da. Since this increase goes
toward constrained ¢rms altering their date 0 investment and borrowing against
the l-state (from nonspecialists),

ðwf  CÞda ¼ ð1  lÞc0 ðkÞdk  ð1  lÞpdf l 40: ðA16Þ

 Excessive Dollar Debt 891

The market clearing condition in the l-state is

ak
^þ
lk ð1  lÞ ¼ ð1 þ lÞw  f l ð1  lÞ  2^f l l: ðA17Þ
el
Thus,

ð1  lÞak ð1  lÞa ^ þ ð1  lÞdf l þ 2ld^f l :

2
del ¼ 2
dk þ ldk ðA18Þ
el el
If delZ0, then from the FOCs we know that dk  0; dk ^  0; d^f l  0. However,
given (A16), df o0. From (A18), de o0, which is a contradiction. Thus, delo0 for
l l

da40.
Given that delo0, from the FOCs we know that dk40; dk ^40; d^f l 40. From
l ^
(A18), ð1  lÞdf þ ldf o0.
l

Proof of Lemma 7: We ¢rst note that q is proportional to el. This is because in the
region that specialists only lend to constrained ¢rms, q begins at (D1el)/2 and
decreases linearly. When there are su⁄cient specialists, specialists also lend to
unconstrained ¢rms, resulting in q 5 el.
The proof follows the same logic as that of Proposition 4. First, since specia-
lists lend to both constrained and unconstrained ¢rms, we note that

ðwf  CÞda ¼ ð1  lÞc0 ðkÞdk þ lc0 ðk ^  ð1  lÞpdf l  lpd^f l 40:

^Þdk ðA19Þ

If delZ0 for da40, then from the FOCs we know that dk  0; dk ^  0; d^f l  0. How-
ever, given (A19), df o0. From (A18) this means that de o0, which is a contradic-
l l

tion. Thus, delo0 for da40.

For the l comparative static, after a little algebra we can show that

ð1  lÞak ð1  lÞa ^ þ ð1  lÞdf l þ 2ld^f l þ ð1  lÞakðk

^  2ðw  ^f l ÞÞdl:
2
del ¼ 2
dk þ ldk
el el
ðA20Þ

The last term on the right-hand side (RHS) is positive for dl40, because on net,
unconstrained ¢rms are borrowers of dollars in the market at date 1. Thus, we
only need to show that the ¢rst term is nonnegative.The proof is by contradiction.
Suppose delr0 for dl40. Then the ¢rst term on the RHS is positive. However
from the FOCs we know that if delr0, then dk  0; dk ^  0; d^f l  0, and df lo0.
From (A20) this means that de o0, which is a contradiction. Thus, del40 for
l

dl40.
Parameter assumptions We examine the technical assumptions on parameters
that we have used. First, we require that w 5 F h in (P5), or that the return to in-
vesting domestically exceeds that of investing abroad:

Aþa
ð1  pÞA þ p c0 ðwÞ: ðA21Þ
2
892 The Journal of Finance

Second, we require that the solution features some insurance against the l-state,
so that Flow:
Aþa
c0 ðwÞD ð1  pÞA þ p : ðA22Þ
2
Finally, we require that equilibrium has 1oeloD. The FOC for the program in
(P4) is,
 
D þ el 1 D
c0 ðkÞ ¼ ð1  pÞA þ p Aþa l : ðA23Þ
2 2 e
We denote the solution to this equation as k(e). Then the largest value of k is at-
tained when e 5 1, and the smallest value when e 5 D. Using this knowledge as
well as the market clearing condition leads to:
pakð1Þ
oD ðA24Þ
w  cðkð1ÞÞ

pakðDÞ
41: ðA25Þ
w  cðkðDÞÞ

REFERENCES
Aghion, Philippe, Philippe Bacchetta, and Abhijit Banerjee, 2001, Currency crises and monetary
policy in a credit constrained economy, European Economic Review 45, 1121^1150.
Allen, Franklin, and Douglas Gale, 1994, Limited market participation and volatility of asset prices,
American Economic Review 84, 933^955.
Allen, Franklin, and Douglas Gale, 2000, Optimal currency crises, Carnegie-Rochester Conference Ser-
ies on Public Policy 53, 177^230.
Burnside, Craig, Martin Eichenbaum, and Sergio Rebelo, 2001, Hedging and ¢nancial fragility in
¢xed exchange rate regimes, European Economic Review 45, 1151^1193.
Caballero, Ricardo, and Arvind Krishnamurthy, 2001a, International and domestic collateral con-
straints in a model of emerging market crises, Journal of Monetary Economics 48, 513^548.
Caballero, Ricardo, and Arvind Krishnamurthy, 2001b, Smoothing sudden stops, Working paper,
Northwestern University.
Caballero, Ricardo, and Arvind Krishnamurthy, 2002, A‘‘vertical’’analysis of monetary policy in emer-
ging markets,Working paper, Northwestern University.
Calvo, Guillermo, 1996, Money, Exchange Rates, and Output (M.I.T. Press, Cambridge, MA).
Calvo, Guillermo, 2000, Capital markets and the exchange rate: With special reference to the dollari-
zation debate in Latin America,Working paper, University of Maryland.
Calvo, Guillermo, and Pablo Guidotti, 1990, Indexation and maturity of government bonds: An explora-
tory model, in Rudiger Dornbusch and Mario Draghi, eds.: Capital Markets and Debt Management
(Cambridge University Press, Cambridge, U.K.).
Chang, Roberto, and AndresVelasco, 1999, Liquidity crises in emerging markets: Theory and policy, in
Ben Bernanke and Julio Rotemberg, eds.: NBER Macroeconomics Annual 1999 (M.I.T. Press, Cam-
bridge, MA).
Diamond, Douglas, and Philip Dybvig, 1983, Bank runs, deposit insurance, and liquidity, Journal of
Political Economy 91, 401^419.
Diamond, Douglas, and Raghuram Rajan, 2001, Liquidity shortages and banking crises, Working
paper, University of Chicago.
 Excessive Dollar Debt 893

Dooley, Michael, 1997, A model of crises in emerging markets, NBER Working paper 6300.
Froot, Kenneth, David Scharfstein, and Jeremy Stein, 1993, Risk management: Coordinating corpo-
rate investment and ¢nancing policies, Journal of Finance 48, 1629^1658.
Hausmann, Ricardo, Ugo Panizza, and Ernesto Stein, 2001,Why do countries £oat the way they £oat?
Journal of Development Economics 66, 387^414.
Holmstrom, Bengt, and Jean Tirole, 1998, Private and public supply of liquidity, Journal of Political
Economy 106, 1^40.
Holmstrom, Bengt, and Jean Tirole, 2001, LAPM: A liquidity-based asset pricing model, Journal of
Finance 56, 1837^1867.
Krishnamurthy, Arvind, 2002, Collateral constraints and the ampli¢cation mechanism, Journal of
EconomicTheory, forthcoming.
Krugman, Paul, 1999, Balance sheets, the transfer problem, and ¢nancial crises, in Peter Isard, Assaf
Razin, and Andrew Rose, eds.: International Finance and Financial Crises: Essays in Honor of
Robert P. Flood, Jr. (International Monetary Fund,Washington, DC).