Journal of International Money and Finance 27 (2008) 695706

Contents lists available at ScienceDirect

Journal of International Money and Finance

journal homepage: www.elsevier.com/locate/jimf

A model of the interactions between banking crises and currency crises

Michael Bleaney a, Spiros Bougheas a, *, Ilias Skamnelos b
a b

School of Economics, University of Nottingham, Nottingham NG7 2RD, UK World Bank, Washington, DC 20433, USA

a b s t r a c t
JEL classication: F41 G21 Keywords: Banks Currency crisis Suspension of convertibility

A second-generation model of currency crises is combined with a standard banking model. In a pegged exchange rate regime, after funds have been committed to the banks, news arrives about the quality of the banks assets and about the exchange rate fundamentals. A run on the banks may cause a currency crisis, or vice versa. There are multiple equilibria (with either twin crises or no crisis), depending on depositors expectations of other depositors actions. Suspension of deposit convertibility can prevent a speculative attack on the currency, but last resort lending to solvent banks can induce one. 2008 Elsevier Ltd. All rights reserved.

1. Introduction Recently there has been a great deal of interest in the interaction between banking crises and currency crises. Empirical evidence of a correlation between the two, particularly for emerging markets, is reported by Burkart and Coudert (2002), Kaminsky and Reinhart (1999) and Komulainen and Lukkarila (2003), amongst others. Theoretical work on the subject may be divided into two strands. In one strand, there is assumed to be a currency mismatch associated with foreign loans to (or deposits in) the domestic banking system, whose assets consist of loans to domestic agents (Chang and Velasco, 2000; Takeda, 2001; Goldstein, 2005). In the other strand, associated with Flood and Marion (2004) and a series of papers by Miller (1996, 1998, 2000), the banking system is not assumed to be subject to such a currency mismatch. Our paper belongs to the second strand.

* Corresponding author. E-mail address: spiros.bougheas@nottingham.ac.uk (S. Bougheas). 0261-5606/$ see front matter 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.jimonn.2008.04.004

696

M. Bleaney et al. / Journal of International Money and Finance 27 (2008) 695706

We contribute to the literature in three signicant ways. First, we demonstrate that, even in an environment where the fundamentals of the banking and the foreign sector exhibit no correlation, fragility in either sector can cause a panic in the other. Put differently, contagion can spread in both directions. Second, we focus on the role of the domestic depositor in twin crises, and on bank deposits as a source of funds for currency speculation, rather than on foreign deposits or the exposure of bank assets to currency movements. Finally, we consider the impact of a policy of suspension of convertibility of bank deposits on the outcome in the currency market. Compared with the papers by Flood and Marion (2004) and Miller (1996, 1998, 2000), our contribution differs in its approach to currency crises (our model is of a second-generation rather than a rst-generation type) and in the explicit modelling of the banking contract.1 The seminal model of Diamond and Dybvig (1983) shows how banks may be subject to sunspot panics because of the maturity mismatch between their assets and their liabilities. With rational expectations, such sunspot panics can be eliminated by a commitment to suspend convertibility of deposits in the event of a run. In practice, however, bank runs occur, and are usually associated with bad news about the banks solvency, with less solvent banks experiencing larger runs (e.g. Schumacher, 2000). To allow for this possibility we develop the model of Jacklin and Bhattacharya (1988). Whilst theoretically banks might offer depositors run-proof contracts, such contracts are often dominated by alternatives in which there are some probabilities of a run induced by bad news (Alonso, 1996). We combine a second-generation model of currency crises2 with a standard model of the banking contract. There are three assets available to consumers: bank deposits, domestic currency or foreign currency. The exchange rate between domestic currency and foreign currency is initially pegged, but might be signicantly devalued at some future date. Ex ante, returns on bank deposits exceed those on domestic or foreign currency. Subsequent bad news in the currency market (such as an adverse terms of trade shock) may cause depositors to seek to withdraw their funds from the bank in order to convert them into foreign currency, even when expected returns to bank deposits exceed those on domestic currency. Alternatively, bad news about banks solvency may cause an information-based bank run. The likelihood of devaluation is assumed to be decreasing in the quantity of foreign exchange reserves. If banks assets are liquidated and the funds distributed to depositors, this increases the funds available for currency speculation and may induce a currency crisis that would not otherwise have occurred. For certain values of the parameters there are multiple equilibria, with either twin crises or no crisis. The work of Goldstein (2005) is most closely related to ours. His model differs in assuming that (1) the banks creditors are foreign (so deposit withdrawals are always associated with reserve losses), (2) the banks suffer from a currency mismatch (so the expected returns to the banking contract are negatively correlated with the expected returns on foreign currency), and (3) there are increasing returns to scale on the long-term investment (so withdrawals reduce the expected returns of those who remain in the bank). We show that there is an interaction between banking and currency crises even in the absence of these assumptions. We also investigate a variety of possible policy responses, including suspension of convertibility of bank deposits. The empirical relevance of our model is demonstrated by the behavior of bank deposits in Argentina over the course of 2001. The spread between domestic 30-day interest rates denominated in pesos and in dollars increased steadily and markedly after February, and yet there were large withdrawals of bank deposits that were concentrated on peso-denominated deposits and affected all types of bank. Between December 2000 and November 2001, peso deposits fell by over 35% and dollar deposits by about 10%, which indicates that depositors were responding to perceived currency risk (De la Torre et al., 2004). 2. The model The foundation of the model is the modern view of banks as offering consumers insurance against liquidity risk, so that they can exploit the higher returns available on longer-term investments, even

1 Buch and Heinrich (1999), Burnside et al. (2000) and Velasco (1987) also consider banking and currency crises within a single framework, but they focus on different issues. 2 Our modelling approach to currency crises follows Morris and Shin (1998) and Obstfeld (1996).

M. Bleaney et al. / Journal of International Money and Finance 27 (2008) 695706

697

though they may wish to withdraw funds earlier. When long-term investments are risky, however, even long-term investors may want to get out of the bank early if they revise their expectations about the returns on these investments. The same thing can happen if assets outside the banking system appear to offer particularly high returns. This is the possibility which we explore here by introducing a currency market. There are three periods (t 0, 1, 2) and a continuum of agents whose measure is normalised to one. Each agent is endowed with one unit of the single divisible good that can be stored, invested or consumed. All agents are ex ante identical, but each agent faces a privately observed uninsurable risk of being either of type 1 or of type 2. At the beginning of period-1 agents learn their type, which remains private information. Type-1 agents care only about consumption in period 1 and type-2 agents care only about consumption in period 2. Let p and 1 p denote the probabilities of each agent being type 1 and type 2, respectively. These probabilities are public knowledge. Each types ex post preferences are described by a continuous, twice differentiable, concave utility function, u(c). Endowments can be stored at no cost. Thus in each period the safe asset offers a gross return of one. ~ In addition, there is a risky technology available at t 0 that yields a random gross return R at t 2 for ~ each unit invested at t 0. Agents have the following prior beliefs about the distribution of R: with probability 1 q the return is equal to the low value RL(<1), and with probability q the return is equal ~ to the high value RH(>1). The true value of R only becomes known in period 2. The risky technology is irreversible, in the sense that liquidation in period 1 yields only s (0 < s < 1) units of the good. There are two types of risk involved here: the uncertain return to the risky project at t 2, and the risk to agents of discovering, after committing funds to the risky project in period 0, that they need to consume in period 1 rather than period 2 (which we term the liquidity risk). 2.1. Banks In the above environment banks play an important role by designing demand deposit contracts that insure depositors against the liquidity risk. We assume that the deposit contract is denominated in domestic currency (we consider the possibility of foreign-currency contracts later). We also assume that the optimal contract does not take account of interim information arriving at time t 1, and so leaves open the possibility of bank runs. The optimal contract (c* , c* , c* ) species consumption allocations 1 2H 2L contingent on the time of withdrawal and (for t 2) contingent on the return of the risky technology. Let I denote the fraction of resources that the bank invests in the risky technology. The following program solves for the optimal contract: Maximise puc1 1 pquc2H 1 quc2L ; subject to (1)

pc1 1 I;
1 pc2H RH I; 1 pc2L RL I; and uc1

(2) (3) (4) (5)

quc2H 1 quc2L hEuc2 q:

Eqs. (2)(4) are the resource constraints. Eq. (2) states that the total resources available to satisfy the period-1 demands of type-1 agents can be no more than the amount that the bank has invested in storage (it is possible to liquidate some of the investments in the risky technology early, but it is not optimal to plan to do so). Eqs. (3) and (4) state that the resources available in period 2 are equal to the return of the risky technology. Inequality (5) ensures that type-2 agents prefer their prospective return from the banking contract in period 2 to that offered to those withdrawing their deposits in period 1. In what follows, we assume that constraint (5) is not binding.3 In addition q has to be sufciently high that the banking contract is preferable to storage.
p c and RH is sufciently high.

For example, as we demonstrate in the Appendix this will be the case if uc

698

M. Bleaney et al. / Journal of International Money and Finance 27 (2008) 695706

2.2. Foreign exchange market The exchange rate e (domestic currency units per unit of foreign currency) depends in the absence of government intervention on the state of fundamentals, z, so that a poor state of fundamentals (a high value of z) corresponds to a weak currency (i.e. e0 (z) < 0). The distribution of the state of fundamentals has support on the interval [zL, zH], where zL > 0. Instead of allowing the currency to oat in this way, the government may peg the exchange rate. 2.3. Government At the beginning of period 0 the government pegs the exchange rate at e*(e* e(zL)), so the peg never represents an undervaluation, and maintains the peg through the beginning of period 1 at which time it re-evaluates its commitment to the peg. We assume that the peg is credible in the sense that the a priori expectation of the currency shock is such that the peg is expected to be maintained indenitely. Let V(>0) denote the value that the government derives from pegging the exchange rate. The government also faces costs, C, in defending the exchange rate that depends on both the state of fundamentals and the speculative demand for foreign currency, X. Let C aX bZ a; b > 0: (6)

Thus the costs of defence depend positively on both the fundamentals and the amount of speculative pressure.4 The inclusion of speculative pressure captures the fact that the resources available to the government in defending the peg are nite. The government only maintains the peg if V exceeds C. If V < C (or equivalently if z > ((V/b) (aX/b))) the currency is oated. 2.4. Interim information When agents learn their types at the beginning of period 1, they also receive a signal, s, about the state of the banking system that leads them to update their posterior beliefs about the return 0 distribution of the risky technology. Let qs denote their posterior belief about the probability that the above return will be equal to RH conditional on s. The consistency of posterior beliefs with P 0 the priors requires that probsqs q. Alonso (1996) demonstrated that although at t 0 the bank could condition the deposit contract on the state of the banking system, such a conditional contract does not necessarily maximise depositors ex ante utility. For relatively high values of q a non-conditional contract that allows for the possibility of runs dominates a run-proof conditional contract. We assume that no such conditional contract is offered. However, the possibility of runs requires that we modify constraint (5) since, when a run takes place, the expected utility of a late depositor who does not run, captured by the right-hand side, is lower as a result of the bank being forced to liquidate the long-term technology. Nevertheless, we have assumed that the original constraint (5) does not bind and this will still be the case under the modied constraint as long as the probability of a run is relatively low.5 At t 1 all agents also learn the state of fundamentals, which remain unchanged thereafter. The order of events is as follows. Still during period 1, type-2 agents decide whether to try to withdraw their funds from the banking system (imitating type-1 agents), and, if so, whether to try to convert these funds into foreign currency. At the last stage of period 1, conditional on agents reactions to the new information about the state of both the banking system and fundamentals, the government decides whether to keep defending the peg. If the pressure of speculation is high enough to trigger a collapse

4 This is similar to the specication in Morris and Shin (1998). In their paper, an increase in the fundamentals represents a strengthening of the currency and thus their cost function is negatively related to the fundamentals. 5 Notice that this is consistent with our assumption that it is not optimal for the bank to offer a run-proof contract.

M. Bleaney et al. / Journal of International Money and Finance 27 (2008) 695706

699

of the peg, not all the type-2 agents who have withdrawn funds from the banking system succeed in converting their funds into foreign currency before the collapse.6 At t 2 type-2 agents consume all their resources, converting bank deposits or foreign currency into domestic currency in order to do so. Their level of consumption depends on the returns to the risky technology (if they retained funds in the bank) or the exchange rate (if they moved funds abroad in period 1). 3. Bank panics and foreign exchange market crises We begin the analysis of the model by examining the reaction of depositors to news about the state of the banking system. This rst step will determine the potential excess demand for foreign currency, X, by type-2 agents. 3.1. Information runs in the banking system For the purposes of this section we assume that agents know that the currency peg will be maintained. We relax this assumption in the next section. 0 When type-2 agents learn the state of the banking system qs , they update their prior beliefs about the returns of the risky technology. Then their posterior expected utility from the second-period 0 0 0 deposit contract allocation is equal to qs uc 1 qs uc hEuc qs . For sufciently low values 2H 2L 2 0 of qs the deposit contact will cease to be incentive compatible (i.e. type-2 agents will run to the bank to extract funds in period 1, like type-1 depositors; we term this situation an information-based run). Let q* be such that uc1 Euc q ; then if q0s q the posterior beliefs are such that type-2 depositors 2 will be indifferent between withdrawing their deposits from the bank in period 1 and leaving them there until period 2. At t 0, the banks total investment in the risky technology, I, was equal to 1 pc* . Multiplying this 1 amount by s yields the liquidation value of the risky technology. Adding to this product the amount that the bank has kept in storage in order to satisfy its contractual obligation to type-1 agents, p1c* , 1 yields the total amount of funds that at t 1 the bank has available for distribution when an informa0 tion run takes place. If qs < q then type-2 agents prefer to attempt to withdraw c* in period 1 and to 1 store the liquidation allocation for one period (all depositors receive the type-1 allocation since types are private information) rather than wait for the period-2 allocation specied in their contract. Notice that not all depositors can be successful in this. Because s < 1, there are insufcient resources to satisfy the demands of all depositors in period 1, even if the risky technology is liquidated. The proportion of depositors receiving c* is equal to D/c* , where D h pc* s(1 pc* ) is equal to the total amount (units of 1 1 1 1 consumption) distributed to depositors. In essence, q* represents the minimum value of q for which the banking system is stable in the absence of currency speculation. Although some depositors do not receive anything in the event of an information run, and would therefore have been better off ex post had no type-2 depositors tried to withdraw their funds in period 1, it is a dominant strategy for each individual type-2 depositor to 0 attempt to withdraw in period 1 when qs is sufciently low, whatever the other type-2 depositors decide to do. Next, we turn our attention to the foreign exchange market. 3.2. Speculation and crisis in the foreign exchange market When the state of fundamentals is revealed, type-2 agents have three options whose payoffs depend on the state of the banking system, their expectations about the actions of other depositors

6 This ordering of events, also followed in Morris and Shin (1998), entails that private agents react to the information before the government. In our chosen formulation multiple equilibria arise as a result of uncertainty about the actions of other type-2 depositors. Alternatively we could assume that the fundamentals become known only at the beginning of period 2, after type-2 agents have made their period-1 decisions on the basis of their expectations of the period-2 fundamentals. Then the government chooses (at the beginning of period 2) whether to abandon the peg having observed the true value of the fundamentals. The basic results would still emerge under the alternative assumptions.

700

M. Bleaney et al. / Journal of International Money and Finance 27 (2008) 695706

and their calculations of the future exchange rate. The rst option is not to withdraw their deposits, in 0 which case their expected consumption allocation at t 2 will be equal to Euc qs . Their second op2 tion is to withdraw their deposits in period 1 but simply to store the resulting funds until period 2. (Remember that only a proportion D/c* of type-2 depositors will receive c* when there is a run.) The third 1 1 option is to withdraw their deposits in period 1 and immediately to convert any funds available into foreign currency at the pegged rate e*, planning to convert them back into domestic currency in period 2 at the rate e* (if they expect the peg to be maintained) or at the rate e(z) (if they expect the peg to be abandoned). We assume that e(z) > e* which implies that the return to a successful attack is positive. Next, we derive the excess (above fundamental) demand for foreign currency, X. Obviously X 0 if type-2 agents choose the rst or second option, i.e. they decide not to attack the currency. If they choose the third option and attack the currency then X (1 p)D (total deposits withdrawn in period 1 multiplied by the proportion of type-2 individuals, since type-1 individuals simply consume their allocation in period 1). 0 The equilibrium of the model depends on (a) the state of the banking system qs , (b) the state of fundamentals (z), and (c) the beliefs that type-2 agents have about the actions of other depositors and the governments utility function. For simplicity we assume that the governments utility function is known to all agents. This implies that type-2 depositors know exactly what the returns to foreign exchange speculation will be, conditional on the weight of speculation. The weight of speculation will depend on the actions of other type-2 depositors. In general type-2 agents will choose the rst option (no withdrawals) if the expected return on the banking contract is superior to the returns on domestic or foreign currency. A necessary but not suf0 cient condition for this is that qs > q , as we shall see below. Agents will choose the second option (withdraw and hold cash) only if there is bad news about the banking system and good news about the exchange rate. They will choose the third (withdraw and speculate) if the news about the exchange rate is sufciently bad relative to the news about the banking system. Multiple equilibria can arise, however, because the individual agent is uncertain what the others will do. For some values of z, the decision whether to abandon the peg depends on the weight of speculation, in which case the peg will be abandoned if all type-2 agents withdraw and speculate, but not if they do not. We make the following assumptions about the extreme values of z7: A1 : zH > V=b; in the worst state of fundamentals even if type-2 depositors do not attack the currency the governments payoff from defending the peg is negative. This means that the peg can be abandoned even if all type-2 deposits remain in the bank. A2 : zL < V=b a1 pD=b; in the best state of fundamentals the peg is maintained even under the maximum amount of speculative pressure. We are ready now to fully describe the equilibria of our model: Case 1 : z > V=b; the government abandons the peg with certainty. If Euc qs > uc1 ez=e 2 there is no bank run. Otherwise, there is a twin crisis. 0 Case 2 : z < V=b a1 pD=b; the government maintains the peg. If Euc qs < uc1 there is 2 a bank run. 0 Case 3 : V=b a1 pD=b < z < V=b; if Euc qs > uc1 ez=e there is no bank run and 2 0 the peg is maintained given that there is no speculative pressure. If Euc qs < uc1 there is 2 a bank run and therefore attacking the currency is a weakly dominant strategy. Because of the ensuing speculative pressure the government abandons the peg. In contrast, if 0 uc1 < Euc qs < uc1 ez=e there are multiple equilibria. If none of the type-2 depos2 itors attacks the currency, then the value of defending the currency exceeds the cost and as a result the government will maintain the peg. Of course, the action of the government justies the decisions of the depositors who in that case are better off by leaving their funds at the bank. In contrast, if all type-2 depositors attack the currency, then the cost of defending the currency
0

With other assumptions, some of the possible solutions to the model as described below would be an empty set.

M. Bleaney et al. / Journal of International Money and Finance 27 (2008) 695706

701

exceeds the critical value and in this case the government will let the currency oat. Once more, the decisions of type-2 depositors are justied by the actions of the government. Thus in this case the optimum strategy for an individual type-2 depositor depends on her expectations of other depositors actions.8 The model shows that the stability of the banking system does not just depend on the signal, qs , and likewise that the maintenance of the exchange rate peg does not just depend on the fundamen0 0 tals, z. For example, in case 3, when qs < q or Euc qs < uc1 the banking crisis causes a currency 2 crisis that would not otherwise have occurred. Put differently, if the expected returns to the banking contract were high enough, the corresponding speculative pressure would have been reduced to a level where the peg will be maintained. Conversely, the attractions of currency speculation may be such as to induce a run on the bank even though the expected returns on the bank contract are superior to those on domestic currency. This is exactly what takes place in case 1 when 0 uc1 < Euc qs < uc1 ez=e . If the government were to maintain the peg, type-2 depositors 2 would not have an incentive to run as indicated by the rst inequality. However, the second inequality implies that given that the peg is abandoned these depositors prefer to run. 4. Policy implications 4.1. Suspension of deposit convertibility One option, in the event of a bank run, is to suspend the convertibility of bank deposits into cash, once bank reserves are exhausted (Diamond and Dybvig, 1983; Engineer, 1989; Gorton, 1985). In the absence of exchange rate considerations, suspension may be attractive because, when deciding to participate in a run, type-2 depositors ignore the liquidation costs, each hoping to arrive at the bank before resources are exhausted and therefore not to suffer any of these costs. In other words, the returns to the banking contract may be higher than the liquidation value, but insufciently high to stop type-2 depositors from running. In that case suspension of convertibility would be socially optimal (and would raise the utility of type-2 depositors as a whole) if it could be guaranteed that the depositors who did succeed in withdrawing money in period 1 were all of type 1. Unfortunately that is not possible, because individual types cannot be identied by the bank, so suspension traps the funds of some type-1 individuals until period 2, reducing their utility to zero. This loss has to be set against the gains to type-2 individuals from ensuring that the long-term technology is not liquidated. As Miller (1996) has pointed out, however, where bank deposits are used to nance currency speculation, suspension of convertibility may make the difference between an exchange rate peg surviving a speculative attack and not surviving it, by reducing the weight of speculative funds behind the attack. In this section, we analyse this possibility in the context of our model. We assume that depositors arrive at the bank to withdraw funds in a random order. Therefore, if all depositors attempt to withdraw in period 1, and an amount W is paid out, pW will be paid out to type-1 depositors to nance immediate consumption and (1 p)W to type-2 depositors. Provided that W < D (the total liquidation value of the bank), suspension reduces speculative pressure in the event of a run from (1 p)D to (1 p)W. Replacing D with W has the following effect on the models equilibria: suspension of convertibility increases the range of fundamentals where the government successfully defends the peg, and reduces the range of fundamentals where the currency is unstable (multiple equilibria) when the expected returns to the banking system are high. In other words, suspension of convertibility shifts the boundary between cases 2 and 3. Suppose now that there is a banking authority that is charged with maximising the welfare of bank depositors, ignoring any gains which type-2 depositors may make in the currency market. That is to say, the banking authority is concerned only with the utility value of the depositors on
0

8 Morris and Shin (1998) have demonstrated that introducing a small amount of noise into the signals that depositors receive about the fundamentals would be sufcient to eliminate multiple equilibria. It is clear that such considerations would not alter the implications of our model.

702

M. Bleaney et al. / Journal of International Money and Finance 27 (2008) 695706

the assumption that (if they are type-2 depositors) they store what they receive from the bank in period 1 until period 2. In a standard bank contract model, runs would not occur (ruling out sunspot 0 0 panics) if qs > q , and would occur with certainty if qs < q , in which case suspension is a policy option. The banking authority would then have to consider whether the potential gains to type-2 depositors from suspension (which arise so long as the banking contract yields higher expected returns in period 2 than its liquidation value) outweigh the utility losses of type-1 depositors whose funds are trapped in the bank, and who are therefore unable to consume in period 1. Without specifying the precise weights attached to the utilities of the two types of depositors, without loss of generality we can denote the value of q at which the authority is just indifferent between suspension and nonsuspension in these circumstances as q**. Then the banking authority will suspend in all cases where there is a run and q > q**. If q** > q*, suspension will never happen in the standard model as there is no reason for any depositors to run. Now consider the banking authoritys problem in the model developed in this paper. Bank runs do 0 0 not occur only at low values of qs . They can take place even at very high values of qs , provided that z is sufciently low. As in the standard model, the authority will suspend payments whenever q > q**. Below we list all possible equilibria outcomes for the case q** < q*: Case Case (a) (b) 1 : if q < q**, there is never any suspension, so nothing is changed. 2 : if q > q**, then there are various sub-cases: z > V=b; convertibility of deposits is suspended but the currency still oats; z < V=b a1 pW=b; the peg is maintained with certainty but convertibility of deposits is suspended if at least one type-2 depositor runs (q** < q < q*); and (c) V=b a1 pD=b < z < V=b; convertibility of deposits is suspended provided that at least one type-2 depositor runs; whether the currency oats or pegs depends on how many (if any) type-2 depositors withdraw their funds in period 1 before convertibility is suspended; when q is sufciently high type-2 depositors prefer not to run and the peg is maintained.

In case 2(c) there are still effectively multiple equilibria. If a proportion k of type-2 depositors participates in a run, and suspension takes place as soon as it is clear that at least one type-2 depositor is running (with (1 k) of type-2 depositors staying behind to reap the higher payoff in period 2) then the fraction of funds available for speculation is k(1 p)pc1/[p k(1 p)], assuming that everyone who attempts to do so has an equal chance of successfully withdrawing before suspension. Given k, the higher the value of z, the more likely that the currency will oat. When q is high and z is low there is more risk in running, because the gains from running (conditional on the peg being abandoned) are relatively small compared to the potential losses (conditional on it being maintained). The probability that a run is large enough to precipitate abandonment of the peg thus has to be viewed as high for running to seem attractive. The opposite situation obtains when q is low and z is high. Consequently it seems likely that k will be a decreasing function of q and an increasing function of z, as will be the probability of the peg being abandoned. We have assumed that the banking authority is not concerned with the exchange rate. Nevertheless consider what happens when V=b a1 pD=b < z < V=b a1 pW=b, a sub-case of case (2b). None of the type-2 depositors run because they anticipate that the demand for foreign currency will not be sufciently high (given that the banking authority will suspend convertibility) for the government to abandon the peg. Put differently, the suspension of convertibility of bank deposits will be sufcient to maintain a peg that might otherwise have been abandoned. Since suspension of convertibility of bank deposits is equivalent to a reduction in D, a lower proportion of early consumers (p) or lower payments to them (c1) will have a similar effect. 4.2. Lender of last resort Suppose that the central bank acts as a lender of last resort, lending on demand to banks that are 0 solvent (i.e. for which uc1 < Euc qs ). Since in our model a run would occur in such a situation 2 only because of the attractions of currency speculation, such a policy would in fact fuel such speculation, because it effectively relieves running bank depositors of the liquidation costs. In other words it

M. Bleaney et al. / Journal of International Money and Finance 27 (2008) 695706

703

raises D from pc* s(1 pc* ) to c* . This shifts the boundary between cases 2 and 3 so as to induce runs 1 1 1 in some situations where they would not otherwise have occurred, i.e. it has exactly the opposite effect to that of the suspension of deposit convertibility. 4.3. Suspension of currency convertibility A third policy option is to suspend convertibility of domestic currency into foreign currency. The obvious consequence of such a policy is to eliminate any speculative pressure and thus allow the government to maintain the peg. In addition, when the fundamentals are weak but the expected return of the banking contract is relatively high, it might prevent a bank crisis that would have taken place had the government allowed the exchange rate to oat (case 1). 4.4. Bankruptcy laws The government might introduce tougher bankruptcy laws in order to reduce the ex ante expected returns of bank liquidation and hence the incentives for bank runs and speculative attacks on the currency.9 Clearly such policy measures could not be used as a response to an ensuing crisis, but they might be successful in reducing the vulnerability to crises in the rst place. We can capture this type of policy in our framework by setting a lower liquidation value s for the irreversible technology. The direct effect of the policy would be to lower the amount of funds, D, available for distribution to depositors in the case of a bank run. In the absence of a foreign exchange rate this policy would not have any effects on the probability of a banking crisis. The reason is that the liquidation value does not appear in the no-run constraint, (5), as each type-2 depositor compares the expected second-period return of the deposit contract with the allocation promised to type-1 depositors. Things are different when we introduce the foreign exchange market. The reason is that a reduction in D lowers the likelihood of an exchange rate crisis because it reduces potential speculative pressure. As a consequence it can also eliminate the likelihood that an exchange rate crisis will lead to a bank crisis that otherwise would have never occurred. 4.5. Financial dollarization What would happen if banks offered depositors the option of a dollar contract as well as a peso contract at time t 0? We assume that the bank accepts pesos from depositors, and redeems their deposits in pesos, but uses the current exchange rate to redeem dollar contracts. The bank has two long-term investments available one that produces tradable goods and one that produces non-tradables and, in order to avoid a currency mismatch, invests in tradable goods production in proportion to the number of depositors choosing the dollar contract. There are some heterogeneities about agents exchange rate expectations, so depositors who are relatively optimistic about the peg choose the peso contract, and those who are relatively pessimistic choose the dollar contract. At time t 1, suppose that there is a bad signal in the currency market. Dollar depositors have no incentive to run, because they have effectively taken out exchange rate insurance. Peso depositors still have an incentive to run if investing in dollars offers higher expected returns than keeping the money in the bank. The bank might offer peso depositors the option of converting into dollar deposits, but at what price? In order to hedge the resulting currency exposure and avoid a currency mismatch, the price of dollar deposits would have to reect the shadow exchange rate at t 1, so switchers would get less in t 2 than those who chose dollar deposits in period t 0. Switching to dollar deposits may yield either higher or lower expected returns than running and buying dollars in the foreign exchange market. Effectively the dollars cost more, but in exchange agents still benet from the higher returns from the long-term technology. Note, however, that under nancial dollarization D is effectively reduced (a) by the dollar depositors not running and (b) by the possibility that some

We are grateful to a referee for this suggestion.

704

M. Bleaney et al. / Journal of International Money and Finance 27 (2008) 695706

peso depositors would choose to switch to dollar deposits. That shifts the boundary between cases 2 and 3 in a favorable direction and makes a currency crisis less likely. Thus one can argue that nancial dollarization in Argentina helped to reduce deposit withdrawals and to protect banks from runs once the currency came under pressure. In fact, though, banks in Argentina did not take care to match dollar liabilities with dollar-earning assets, and were not required by the authorities to do so (probably in large part because this would have undermined the authorities repeated claim that the peg was permanent). This meant that peso deposits could be switched into dollar deposits at the pegged exchange rate, worsening the currency mismatch and ultimately threatening the returns to all depositors. Hence, in late 2001, dollar depositors rationally began to withdraw their deposits as well. 5. Conclusions If bank deposits are potentially available for currency speculation, then there is an interaction between the stability of the banking system and that of a currency peg. Bad news about the banking system, causing a run, may provide the liquidity necessary for a successful speculative attack on the currency. Alternatively, high expected returns from currency speculation may destabilise an otherwise sound banking system. This appears to have happened in Argentina during 2001, when there were large withdrawals of (especially) peso-denominated bank deposits. This paper has provided a detailed theoretical analysis of the issue. Uncertainty about whether other bank depositors are going to withdraw and participate in a speculative attack means that there is a region of multiple equilibria, in which there are either no crises (nobody runs because they do not expect others to do so, and therefore they expect the peg to hold) or twin crises (everybody runs because they expect others to do so, forcing the exchange rate to oat). Suspension of convertibility of bank deposits can rescue a sound banking system threatened by the temptations of currency speculation. If the fundamentals are sufciently bad, suspension will not prevent the currency from oating. If the fundamentals are sufciently good, suspension can ensure the successful defence of the peg. In the intermediate range of the fundamentals, the outcome depends on agents expectations of other depositors actions, even in the case of suspension, because the weight of currency speculation depends on type-2 depositors share of the funds withdrawn from banks in period 1, which in turn depends on the proportion of them participating in the run. On the other hand, if the central bank acts as a lender of last resort to solvent banks irrespective of the causes of a run, then it can actually induce bank runs that would not otherwise have occurred. This happens because the potential weight of speculative pressure on the currency is increased by relieving running bank depositors of liquidation costs, and this makes a currency depreciation more likely. Disclaimer The views expressed here are those of the authors and should not be taken to represent those of the World Bank or of its Executive Board or of its member countries. Appendix. The solution for the banking contract when uc [ The following program solves for the optimal contract: p p p Max p c1 1 pq c2H 1 q c2L ; subject to p c

(A1)

pc1 pc1

1p c 1; RH 2H 1p c 1; RL 2L

(A2) (A3)

M. Bleaney et al. / Journal of International Money and Finance 27 (2008) 695706

705

and

p c1

p p q c2H 1 q c2L :

(A4)

Next, we are going to derive the optimal contract under the supposition that (A4) does not bind and then we will show that this is the case for sufciently high values of RH . Let l1 and l2 denote the Lagrange multipliers associated with constraints (A1) and (A2), respectively. The F.O.C.s are given by 1 p l1 l2 ; 2 c1 p 1 ; 2 c2H RH and (A5) (A6)

l 1q p 2 : 2 c2L RL
From (A2) and (A3) we have c2H R H: c2L RL From (A6) and (A7) we have p q l R c2L p 1 L : l2 RH 1 q c2H Using (A8) and rearranging we get s RH q l : l1 RL 1 q 2 Substituting (A10) in (A5) we get s ! 1 RH q 1 l2 : p 2 c1 RL 1 q From (A7) we have

(A7)

(A8)

(A9)

(A10)

(A11)

l2 1 qRL p:
Substituting (A12) in (A11) and rearranging we get c1 1 c2L ;

1 2 c2L

(A12)

(A13)

p where G q RH =RL 1 qRL 2 . Substituting (A13) in (A3) and rearranging we get c2L RL G : pRL 1 pG (A14)

From (A13) and (A14) we get c1 RL ; (A15)

pRL 1 pG

and from (A8) and (A14) we get

706

M. Bleaney et al. / Journal of International Money and Finance 27 (2008) 695706

c2H

RH G : pRL 1 pG

(A16)

Eqs. (A14)(A16) describe the optimal contract under the supposition that (A4) does not bind. Substituting the above p solution in Eq. (A14) and rearranging we nd that this will be the case if p p RL < q RH 1 q RL G. Notice that for sufciently high values of RH and/or high values of q the constraint will be satised and for low values of RL we also get that c2L < c1 < c2H . References
Alonso, I., 1996. On avoiding bank runs. Journal of Monetary Economics 37, 7387. Buch, C., Heinrich, R., 1999. Twin crises and the intermediary role of banks. International Journal of Finance and Economics 4, 313324. Burkart, O., Coudert, V., 2002. Leading indicators of currency crises for emerging countries. Emerging Markets Review 3, 107 133. Burnside, C., Eichenbaum, M., Rebelo, S., 2000. Government Guarantees and the Self-Fullling Speculative Attacks. Kellogg School of Management, Chicago, IL. Chang, R., Velasco, A., 2000. Banks, debt maturity and nancial crises. Journal of International Economics 51, 169194. De la Torre, A., Yeyati, E., Schmukler, L., 2004. Living and dying with hard pegs: the rise and fall of Argentinas currency board. In: Monetary Unions and Hard Pegs. Oxford Scholarship Online Monographs, pp. 183231. Diamond, D., Dybvig, P., 1983. Bank runs, deposit insurance, and liquidity. Journal of Political Economy 91, 401419. Engineer, M., 1989. Bank runs and the suspension of deposit convertibility. Journal of Monetary Economics 24, 443454. Flood, R., Marion, N., 2004. A model of the joint distribution of banking and currency crises. Journal of International Money and Finance 23, 841865. Goldstein, I., 2005. Strategic complementarities and the twin crises. Economic Journal 115, 368390. Gorton, G., 1985. Bank suspension of convertibility. Journal of Monetary Economics 15, 177193. Jacklin, C., Bhattacharya, S., 1988. Distinguishing panics and information-based bank runs: welfare and policy implications. Journal of Political Economy 96, 568592. Kaminsky, G., Reinhart, C., 1999. The twin crises: the causes of banking and balance-of-payments problems. American Economic Review 89, 473500. Komulainen, T., Lukkarila, J., 2003. What drives nancial crises in emerging markets? Emerging Markets Review 4, 248272. Miller, V., 1996. Speculative currency attacks with endogenously induced commercial bank crises. Journal of International Money and Finance 15, 383403. Miller, V., 1998. Domestic bank runs and speculative attacks on foreign currencies. Journal of International Money and Finance 17, 331338. Miller, V., 2000. Central bank reactions to banking crises in xed exchange rate regimes. Journal of Development Economics 63, 451472. Morris, S., Shin, H., 1998. Unique equilibrium in a model of self-fullling currency attacks. American Economic Review 88, 587 597. Obstfeld, M., 1996. Models of currency crices with self-fullling features. European Economic Review 40, 10371047. Schumacher, L., 2000. Bank runs and currency runs in a system without a safety net: Argentina and the tequila shock. Journal of Monetary Economics 46, 257277. Takeda, F., 2001. Bank runs and international nancial instability revisited. Economics Letters 73, 187194. Velasco, A., 1987. Financial crises and balance of payments crises: a simple model of the southern cone experience. Journal of Development Economics 27, 263284.