A. Financial Institutions and Informational Problems

In this subsection we explain how financial institutions can cause informational problems in an interbank market.² We assume that an entrepreneur always prefers to have his project completed regardless of its type; but when completion is not possible, he prefers to quit the project as soon as possible. To express this assumption in a formal way, we assume that an entrepreneur gets a private benefit b_t from working on a project, where t denotes the date when the project is either completed or terminated at t = 1,2,3. Specifically, if an entrepreneur quits the project at date 1, he gets a low private benefit, b₁ > 0. If a bad project is liquidated at date 2, an entrepreneur gets an even lower private benefit b_2b, where 0 ≤ b_2b < b₁. At date 3, if a bad project is reorganized and completed, it will generate a private benefit b_3b > b₁ to an entrepreneur; in the case of a good project, it will generate a private benefit, b_3g > b_3b, to an entrepreneur. To summarize, we have b_3g > b_3b > b₁ > b_2b ≥ 0.

With respect to financing, a project can be financed by one bank alone, or can be co-financed by two (or more banks) jointly.

The timing of the game related to project financing is as follows:

• Date 0: All parties know the distribution of the projects and the depositors, but no one knows the type of each project and the type of each depositor. The bank(s) offer a take-it-or-leave-it contract to an entrepreneur. If the contract is signed, the bank(s) will invest I₁ units of money into the project during period 1.

• Date 1: The entrepreneur learns the type of the project. If the entrepreneur stops the project (liquidation), he gets a private benefit b₁ > 0 and all the banks observe the liquidation of the project. However, unless a project is stopped by the entrepreneur the bank(s) still does (do) not know the type of the project and further I₂ units of money are invested. Moreover, the bank(s) will know the distribution of their own project better than before as its private information, i.e., λ_m, is more accurate than the prior λ.

• Date 2: The type of a project becomes public knowledge:

  • If a project is of a good type, a further I₃ will be invested.

  • If a project is of a bad type, a decision whether to liquidate or to reorganize has to be made.

• Date 3: All projects are completed,

  • for a good project, return Y goes to the bank(s), the entrepreneur gets b_3g;

  • for a bad project, return X goes to the bank(s), the entrepreneur gets b_3b.

If a project is a good one, it generates a high return Y no matter how it is financed. For a bad project, we suppose that there are several strategies to reorganize it during the third period, but only one of these strategies can generate X, which is ex post profitable. However this strategy can only be selected and implemented when all the involved bank(s) is (are) in agreement.

Under single-bank financing, given that the earlier investments are sunk, the bank will choose an ex-post efficient strategy to reorganize the project such that the payoff is greater than the ex-post cost of refinancing, I₃. As a result, the bank is unable to commit to terminating a bad project ex post.

Moreover, the fact that the bank is not able to commit to terminating a bad project affects the entrepreneur’s ex-ante incentives to reveal information. When the entrepreneur at date 1 discovers that his project is a bad one, he anticipates that the project will still be continued and refinanced by the bank at date 2 as long as it lasts until then. Consequently, if he decides to quit the project, he gets private benefit b₁; if he decides to continue the project, the bad project will always be refinanced by the bank and will generate a private benefit b_3b > b₁ to the entrepreneur. Therefore, the entrepreneur will always choose to continue a bad project after he privately discovers its type. This implies that in an economy where every project is financed by one bank, the information to separate the good projects from the bad ones is available neither to the financier nor to the interbank market at date 1.

However, in the case of multi-bank financing, the asymmetric information and conflicts of interest among the co-financiers related to reorganizing the project incur a cost, F for ex-post negotiations. When this cost, F, is high, the gain from reorganization is less than the total costs, i.e., X < I₃ + F. Therefore reorganization is not worthwhile and liquidation will follow.³

The commitment to liquidate a bad project at date 2 has a deterrent effect on entrepreneurs who have bad projects. Fearing further losses of his private benefit later, an entrepreneur will choose to quit a bad project as soon as he discovers it is bad. Assuming the observability of liquidation, this result implies that if all projects in an economy are financed by two banks, at date 1 information is available in the interbank market to separate the good projects from the bad projects.

The following lemma summarizes the above results.

To simplify our language in the above lemma, in the reminder of the paper we call an economy under multi-bank financing an economy with hard-budget constraints (HBC); and an economy under single-bank financing an economy with soft-budget constraints (SBC), a term coined by Kornai (1980).

B. Deposit Contracts

We consider a one-good economy. Each depositor’s $1 endowment can be stored from one period to the next, without any cost, or it can be invested in a bank which then invests in a project with stochastic technology, yielding a positive expected return in the future as described in the above section.⁴

Each depositor’s preference is defined as

where C_j is the consumption of type j depositor; j = 1 being early consumers who consume at t = 1 and j = 2 being late consumers who consume at t = 3; π_j is the probability that a depositor is a type 1 or a type 2 consumer, and π₁ + π₂ = 1; ρ < 1 is the discount factor and ρ (R + 1) > 1, where R is the return from investment, which is to be determined in later sections; and u′ > 0, u″ < 0, and (Cu′)′ = u′ + Cu″ < 0.

At date 0, consumers make a deposit decision by solving⁵

In general, the first order condition of this problem is $u^{\prime }(C_1^{*})=\rho (1+R)u^{\prime }(C_2^{*})$. Given that u′ + Cu″ < 0, ρ < 1 and ρ(1+ R) > 1, we have u′(1) > ρu′(1) > ρ(1 + R)u′(R). Consequently, an ex-ante optimal market equilibrium can only be achieved by increasing C₁ and decreasing C₂, that is, $C_1^{*}>1\text{ and }{C}_2^{*}<1+R$.

As in Diamond and Dybvig, a market equilibrium, in which a bank implements deposit contracts with consumers, can Pareto dominate autarchy; that is for $1 deposit at t = 0, a depositor receives either $C_1^{*}$ at t = 1, or $C_2^{*}$ at the end of the exercise. The bank holds ${\pi }_1{C}_1^{*}$ (as cash) at no extra cost, and invests the rest in an illiquid project which yields a higher return. At equilibrium, an early consumer always wants to consume at t = 1, but a late consumer has no incentive to withdraw early. This is because when ρ(1 + R) > 1, the first order condition $u^{\prime }(C_1^{*})=\rho (1+R)u^{\prime }(C_2^{*})$ implies $C_1^{*}<C_2^{*}$. Thus for a late consumer a deviation does not pay as long as other late consumers do not deviate.

However, as there is no perfect insurance against liquidity shocks, there may be a bank run equilibrium, that is, a simultaneous deviation of all late consumers. With the presence of an interbank market, the key for the existence of a bank run equilibrium is the possibility that the banks cannot solve their liquidity shortage problems by borrowing from the market. Conditional on this, all late consumers will withdraw at t = 1. This paper will focus on a mechanism whereby idiosyncratic shocks can trigger a bank run.

In our multi-bank economy there are Ν depositors in each bank and the realized numbers of type 1 and type 2 depositors in each bank are random draws from a binomial distribution of π₁ and π₂ = 1 − π₁ respectively.

C. Timing of the Model

Combining the outcome of the banks’ investments in projects and the consumers’ deposit decisions, the evolved timing of the game can be summarized as follows:

• Date 0: Depositors make a savings decision; the banks make an investment decision regarding how much and in which project to invest.

• Date 1: Early consumers withdraw their money from the banks; late consumers make decisions about whether to withdraw or to keep their deposits in the banks. A bank facing too many early withdrawals either has to borrow from other banks, liquidate premature assets, or face a bank run. All bad projects financed by multi-banks will be liquidated.

• Date 2: All single-bank financed bad projects will be reorganized by the banks.

• Date 3: All projects are completed; banks pay back the interbank loans if they borrowed at date 1; and late consumers collect their rewards.

A. Pooling Equilibrium in the Interbank Market and Financial Crisis

In an economy with single bank financing at equilibrium all projects will last for three periods (an SBC economy), requiring a total investment of (I₁ + I₂ + I₃), that is, at date 1 banks are not able to distinguish between good and bad projects. Moreover, each project’s expected rate of return is $R^S(\lambda )=\frac{\lambda Y+(1-\lambda )X}{I_1+I_2+I_3}-1>0$. Thus, with an endowment of Ν and anticipating the expected withdrawal at date 1, a bank’s optimal investment decision is to hold $N{\pi }_1{C}_1^{*}$ in cash and invest $N(1-{\pi }_1{C}_1^{*})$ in a project at t = 0.⁶

We suppose that at date 1 banks in an SBC economy have better information than at date 0 because of their monitoring over one period of time, and this is their private information. This private information is difficult to convey since all banks have an incentive to falsely report if there is a benefit of doing so. Moreover, we suppose that the only public information in the interbank market is the average quality of all the projects financed by all the banks in the market.

Formally, we suppose that at date 1 the manager of bank m (m = 1, …, M) learns privately that the probability of his project being good is λ_m. Moreover, the rank of the qualities of all the banks’ projects is λ₁ < λ₂ < λ₃ < … < λ_M. and the average quality is $\overline{\lambda }=\frac{1}{M}{\displaystyle {\sum }_{m=1}^M{\lambda }_m}$.

If the total number of early withdrawals at date 1 is smaller than the expected number π₁Ν, a bank will have excess liquidity reserves to lend; if the total number of early withdrawals is more than π₁Ν, however, the bank will face a liquidity shortage; we call this bank illiquid. An illiquid bank may solve its liquidity problem either by borrowing in the interbank market or by liquidating some of its investment prematurely.

In the case of borrowing, an illiquid bank issues a bond in the interbank market. With limited liability, a borrowing bank can only repay the bond if it has a good project. However, given that the market knows only $\overline{\lambda }$, all illiquid banks will be treated in the same way when they borrow. Therefore, all bonds issued by the banks have the same structure: contingent on the realization of the project at date 3, the bond pays,

To highlight our points, we assume that there is a Bertrand competition among lenders such that these banks break even in lending. Hence, given the lenders’ belief that the probability that a bank will pay back 1 is $\overline{\lambda }$, the equilibrium bond price is $p^S=\overline{\lambda }$.

For an illiquid bank to raise $1, it needs to issue $\frac{1}{p^S}$ shares of a bond in the interbank market. Thus, in order to deal with n excessive early withdrawal consumers for an amount of $n{C}_1^{*}$, a total of $\frac{n{C}_1^{*}}{p^S}$ shares of bonds should be issued. While the bond structure is the same for all illiquid banks, with the private information about the quality of each bank’s project, the borrowing cost for each bank is different. For bank m, with a probability of being able to repay the bond as λ_m, the cost of raising $n{C}_1^{*}$ dollars is $\frac{{\lambda }_m}{\overline{\lambda }}\frac{n{C}_1^{*}}{p^S}$, that is, the marginal borrowing cost is $C_B({\lambda }_m)=\frac{{\lambda }_m}{\lambda p^S}=\frac{{\lambda }_m}{{\overline{\lambda }}^2}$. Therefore, the higher the quality of a bank the higher the borrowing cost. Not surprisingly, $\overline{\lambda }$ has to be relatively high to λ_m to make the expected profit of bank m non negative.

In addition to borrowing, an illiquid bank can also liquidate part of its assets prematurely to solve its liquidity problem. However, liquidating assets prematurely is costly. We denote the exogenously given marginal cost of raising cash through liquidating premature assets as C_L.⁷ Given the cost C_L, only the banks that have a non-negative net return after liquidating some of their assets, i.e.,

In general, at date 1 a solvent illiquid bank will compare C_B with C_L to decide how to raise cash. When C_B ≤ C_L then it will borrow; otherwise, it will liquidate some of the premature assets.

We suppose that there are $\overline{m}$ banks facing liquidity shocks. In order to highlight our points in a simple way, we assume that λ_m = λ_m−1 + μ for all $m=1,2\dots ,\overline{m}$; and that C_L is the same for every bank. Denoting ${\overline{\lambda }}_{\overline{m}}=({\displaystyle {\sum }_{i=1}^{\overline{m}}{\lambda }_i})/\overline{m}$, the following Lemma provides a condition for the proof of a contagious bank run.

The ratio of $\frac{{\lambda }_{\overline{m}}}{{\lambda }_{\overline{m}}}$ reflects how much a private evaluation of a project differs from a public evaluation, since ${\lambda }_{\overline{m}}$ is private information of bank $\overline{m}\text{ while}{\lambda }_{\overline{m}}$ is public information. It is also a measurement of the heterogeneity of the projects’ quality, that is, the more heterogeneous the projects’ quality, the larger the difference can be between a private evaluation and a public evaluation. This proposition says that if projects financed by the banks in an SBC economy have a high enough heterogeneity such that the borrowing cost for the best illiquid bank is higher than the liquidation cost, when there are many banks facing liquidity shocks, then a contagious bank run may be an equilibrium. A basic intuition to explain the result is that the market breaks down due to the asymmetric information generated by the SBCs. When the quality of a good bank is private information and the bank is treated as an average bank in the interbank market, it may find borrowing to be too costly and thus it will withdraw from the market. But this will generate externalities such that the average quality of the borrowers in the interbank market will decrease – which will induce more banks to withdraw from the market until there is a total collapse of the interbank market and a contagious bank run.

A low ratio $\frac{{\lambda }_{\overline{m}}}{{\lambda }_{\overline{m}}}$ of implies a low heterogeneity of the projects’ quality. The condition of ${\overline{\lambda }}^2\ge \frac{{\lambda }_m(I_1+I_2+I_3)}{\lambda Y+(1-\lambda )X}$ means a high average quality of the projects. Specifically, it expresses that the average probability that the projects will be successful is high, and the expected return is high. This proposition says that if projects financed by the banks are more homogenous, such that the borrowing cost for the best illiquid bank (thus for all illiquid banks) is lower than the liquidation cost, and the average quality of the projects is high, then at equilibrium the problems of all illiquid banks can be solved in the interbank market in an SBC economy. Thus, idiosyncratic shocks do not cause a contagious bank run in an SBC economy. The intuition is that when projects are more homogeneous, the asymmetric informational problem is reduced (when projects are perfectly homogeneous, there will be no asymmetric informational problem), thus the best illiquid bank will not face too high of a borrowing cost; with high average quality projects, lenders can afford to lend at more favorable terms to all illiquid banks which reduces their borrowing costs.

These results have implications for the timing of a financial crisis in an SBC economy. When an economy is technically less developed such that most investment projects are characterized by imitations, the uncertainty of projects is low and the bank run contagion condition will not be satisfied. When an SBC economy is more developed such that a large proportion of investment projects consists of high-tech or R&D-intensive projects which are more uncertain, the bank run contagion condition is satisfied.

B. Separating Equilibrium in the Interbank Market and Bank Runs

In an economy with multi-bank financing (an HBC economy), at equilibrium all bad projects are stopped at date 1, and good projects will be completed.

To meet an expected number of early withdrawals a bank’s optimal investment decision is to store cash in the amount of $N{\pi }_1{C}_1^{*}$, and to invest all the rest — in the amount of $N(1-{\pi }_1{C}_1^{*})$ — into projects; each project is jointly invested with another bank, which will generate an HBC mechanism that liquidates bad projects at date 1.

Without a loss of generality, we suppose that each bank invests in k projects and invests (I₁ + l)/2 in each project, where I₁ is a new project’s initial investment; l is the liquidity stored for each project; and k and l are to be determined endogenousìy later. At t = 1, a bank can sell its extra liquidity in the interbank market when uncertainties are realized; or a bank needs to borrow if it has more than expected good projects or it faces liquidity shocks. Thus, with an endowment of N, a bank can invest up to $N(1-{\pi }_1{C}_1^{*})$ in k real projects at t = 0, where

Each bank optimally chooses l to maximize its expected returns. That is,

where, p is the price for liquidity. Banks trade liquidities in the interbank market at a price p for each share of bonds, where each share will be paid $1 at date 3.

At (l*,p*) the ex-ante (at date 0) expected rate of return of a project, R^H, is,

In an efficient interbank market, where there is no extra cost of trading liquidity, R^H should be the same as the rate of return in trading liquidity at price p*,

Thus, no bank has an incentive to deviate from (l*,p*): ex ante holding more than l* liquidity for a later selling (at t = 1) in the interbank market will not generate a better expected return; nor will holding less than l* liquidity and investing more in projects generate a better return. The following Lemma records these results.

In the event that a project is bad and aborted at date 1, liquidity ½l* is saved, which the bank may sell in the interbank market to earn a higher return. If a project is good and to be continued at date i, it will need ½(I₂ + I₃ − l*) further liquidity to be completed. When a bank faces j good projects and k — j bad projects, its balance of liquidity is,

Obviously, without a liquidity shock a bank is a net liquidity provider to the interbank market if it has more bad projects, i.e. kλ > j, than the average in the banking system. Otherwise, it is a net liquidity borrower.

When there are excessive early withdrawals, i.e., the number of early withdrawals is greater than $N{\pi }_1{C}_1^{*}$, given that the demand deposit contract requires that $C_1^{*}>1$ and $C_2^{*}<1+R^H$, which implies that

it costs the bank

at t = 1 to finance each excess early withdrawal. The expected return rate of a bank at date 1 when there are j good projects and n excess early withdrawals is

A negative excess early withdrawal number, i.e., n < 0, represents less than the expected early withdrawals, which implies an extra liquidity provision. It is intuitive and straightforward to see that $R_1^H$ increases with the number of good projects, j, and decreases with the number of excess early withdrawals, n.

To highlight our points, we restrict our interests to cases where the total number of depositors in each bank, N, is large, the quality of projects, λ, is not too low, and the proportion of early type consumers, π₁, is not so large that $2{\lambda }^{2}N(1-{\pi }_1{C}_1^{*})\ge 1;\text{ and }{\pi }_1{C}_1^{*}$ is different enough from 1 such that $\frac{\lambda Y-I_1-\lambda (I_2+I_3)}{\lambda Y{C}_1^{*}-C_2^{*}(I_1+\lambda (I_2+I_3))}\ge \frac{(1-{\pi }_1)}{(1-{\pi }_1{C}_1^{*})}$.⁸ Then we have the following results.⁹

The first result says that when a bank’s projects are not worse than average, the present value of the projects’ returns will be enough to cover the cost of borrowing to deal with excessive early withdrawals even in a case where all depositors withdraw at date 1. Therefore, given symmetric information about the quality of a bank assets in the market, this bank will be able to borrow at the market rate to solve its liquidity problem. The second result says that when there are no excess early withdrawals, even in the worst case where a bank has no single realized good project, the return from selling liquidity at the market rate will keep the bank solvent.

The above results are based on the existence of an efficient interbank market which redistributes liquidities among banks according to the ex-post realized shocks. The efficiency of an interbank market is the result of an HBC mechanism in the banking sector that generates symmetric information among banks.

Under a separating equilibrium solvency is public information in the interbank market; an insolvent bank will not be able to borrow in the market and thus faces a bank run. In general, given $R_1^H$ decreases in n and increases in j, banks with a large enough n and small enough j will be unable to solve the liquidity shortage problem in the interbank market and thus will be subject to bank runs. However, due to the symmetric information in the interbank market, their bank runs do not have externalities, i.e. they will not affect the borrowing of the solvent banks in the market. Thus, bank runs are not avoidable but they are contained. The following proposition summarizes these results.

An HBC mechanism generates symmetric information among banks. With the symmetric information the interbank market should be able to provide liquidity to all illiquid banks that are not hit too badly by technological shocks. As a result, although a bank run cannot be completely avoided, a contagious bank run does not occur in an HBC economy. In sharp contrast, in an SBC economy where information about the quality of bank investments is private to each bank alone, the interbank market is a lemon market. As a result, a bank run can be avoided completely when the projects are homogeneous; or there can be a contagious bank run equilibrium when the projects are heterogeneous and there are liquidity shocks.

A. East Asian “Miracle” vs. East Asian Financial Crisis

Although our model is very stylized, it sheds some new light on our understanding of financial institutions and financial crises (e.g., phenomena documented by Kindleberger, 1996, and Delhasise, 1999), particularly the recent East Asian financial crisis.¹⁴ The Korean, and Taiwan Province of China (POC), economies are good examples to illustrate that our theory, which links financial crises to financial institutions, is relevant.

Korea and Taiwan (POC) are at similar development stages, geographically close, and they also have similar technologies, labor inputs, and high savings (e.g., high shares of trade in GNP; recent transformations from a traditional economy to one that is oriented toward high-tech). However, while Korea was at the center of the East Asian crisis, Taiwan (POC) had been much less affected. Our explanation for this difference rests on the substantially different financial institutions in the two economies.

It is well documented that Korean development has been characterized by the establishment of large conglomerates (chaebols) through government-coordinated bank loans. In a typical case, financing decisions for projects in Korea are made by the government or by the principal bank among a group of investing banks.¹⁵ Using the language of our model, this institution corresponds to the single-bank financing mode.

It is also well documented that bankruptcies rarely occurred in Korea prior to the recent financial crisis (particularly in the chaebols); and information about investment quality was often unavailable.¹⁶ Our theory says that single-bank financing leads to a pooling equilibrium in the interbank market, which can result in a financial crisis with moderate idiosyncratic shocks.

With respect to the cause of the financial crisis, it has been claimed by many economists, policy makers, and businessmen that the bankruptcy of several insolvent top chaebols in early 1997 triggered the crisis (e.g., Park, 1997). This observation is consistent with our prediction if the chaebols are viewed as banks and the investors (e.g., creditors) are viewed as depositors in our model. In fact, the chaebols are conglomerates with the functions of financial institutions.

In sharp contrast, Taiwan (POC) firms relied on diversified financial sources and there were frequent bankruptcies in the corporate sector. Inefficient firms were bankrupt, and information about investment quality was available.¹⁷ Our theory predicts that in this economy the interbank market will function well and there should be no financial crisis when there are idiosyncratic shocks (regardless of the strength of the shocks).

Our theory also helps to reconcile the paradox between the East Asian “miracle” in the three decades prior to the mid-1990s and the East Asian “financial crisis” in 1997. In the period of early development, that is, during the catching-up period of the 1960s to the early 1990s, the projects were more homogeneous in terms of uncertainty due to the nature of technological imitation. In this case, our theory predicts that there are no project liquidations and no bank runs in an economy with a pooling equilibrium; that is, an SBC economy appears even to outperform an HBC economy and may attract a large number of investments. However, when an economy is on technological frontiers and attempts major innovations (e.g., South Korea since the early 1990s), the heterogeneity of the projects rises precipitously and the negative effects of an SBC economy are dominant, thus incubating trouble in the financial system.

B. Transparency and Financial Liberalization

At a more general level, our results are closely related to another important policy issue: the transparency of the financial sector to the market and to the government. A widely held view emphasizes that policies can improve the transparency of financial institutions. However, the lessons of centralized economies and the long debate over the viability of centralized economies since the late 1930s (Lange vs. Hayek) both tell us that it is impossible to have transparency in all aspects of any economy. In light of such an impossibility, a key issue is which aspects are to be made transparent and how. Moreover, not only is it impossible to have transparency in all aspects of an economy, our theory suggests that targeting the wrong aspect of an economy to improve transparency can make things even worse.

Hayek (1945) outlined a principle according to which it is the market, not the government, that provides the right information for the economy to run efficiently. But what this means in the context of financial crisis is unclear. One of our major contributions is that we provide a model to illustrate that an HBC mechanism makes solvency information transparent to the market and to the government. Thus, it makes the financial market more stable and helps the government to intervene correctly when necessary.

Moreover, our theory implies that wrongly targeting transparency can result in disaster. This is because an HBC mechanism relies on the ‘non-transparency’ of the co-financiers’ information regarding the reorganizing of a bad project. If this condition is changed but all other conditions remain the same due to a wrong policy, then the HBC mechanism is destroyed such that all the bad projects will be reorganized even when they are financed jointly.

To summarize, the essential message of our theory regarding this issue is that a key policy to improve transparency correctly and effectively may not be a policy which directly targets information, but a policy which hardens budget constraints or a policy which lowers the institutional costs of multi-financier financing.

Another important policy issue concerns the liberalization of financial institutions. To analyze this, we can change our model from an M-bank economy to a one-bank economy. According to our theory, a one-bank economy must be an SBC economy. Moreover, because all the deposits in the economy are pooled in one bank, there are no idiosyncratic liquidity shocks or technological shocks. Thus, although inefficient, this one-bank economy is almost immune from bank runs or financial crisis if there is no systemic shock.

Unlike the one-bank economy, an M-bank SBC economy is very sensitive to a contagious bank run due to the lemon problem in the interbank market. This comparison has important implications for policies/reforms of the banking system. The basic message is that the liberalization of financial institutions must be contingent on measures to harden budget constraints. If liberalized banks are operating under an SBC, a liberalization policy alone may greatly destabilize the financial system! This simple analysis captures some characteristics of the reform/liberalization of banking systems. For instance, a major reform measure in the transition from a centralized economy to a market economy is to change the banking system from a one-bank system (at least conceptually one can regard all state banks as branches of one bank — the state bank) to a multi-bank system. Many of the banking system liberalization reforms in East Asia prior to 1997 were carried out in this spirit. According to our theory, a banking system reform designed to enhance competition as described above can induce huge contagious risks to the system if simultaneously the system is not designed to harden budget constraints.

C. Corruption and Financial Crisis

To highlight our points, in our basic model we assume that there is no corruption in the economy. All the problems are generated from corruption-free and ‘pure’ economic institutions. Now we relax this assumption to look at how corruption affects financial institutions and the likelihood of a financial crisis.

There are two aspects of corruption that can be introduced into our model to generate relevant results. The first aspect is that corruption itself can be a mechanism of an SBC (Shleifer and Vishny, 1993); that is, when it is discovered that a project is a bad one at date 2, in a corrupted economy firms and/or financial institutions have the option of bribing the government to bail out the project regardless of profitability (the option will be even stronger if bailing out a project is ex post profitable). Thus, with the presence of corruption there will be a more serious SBC syndrome. Moreover, with serious corruption it may be very difficult for a multi-financier institution, e.g., an equity market, to operate due to the lack of contract enforcement and the lack of transparency in the market. As a result, HBC mechanisms may be destroyed or HBC banks may be out-competed by corrupted banks.

If the impact of corruption is restricted to the above aspect, all of our theoretical results will be qualitatively unchanged. However, there is another aspect of corruption that can enter into our model; that is, corruption can affect the selection of projects. It turns out that with this aspect in our model the timing and likelihood of a financial crisis can be changed significantly.

To illustrate this, suppose that at date 0 there is asymmetric information such that entrepreneurs know the distribution of the projects better than the banks; moreover, some risky projects may be beneficial to some entrepreneurs. In a corrupted economy an entrepreneur may bribe the bank so that the project will be financed. In that case, the selected projects will be more heterogeneous than the corruption-free projects. Thus, many high-risk projects may be selected even in a less developed economy. When corruption not only results in more serious SBC problems but also results in the selection of more risky projects, then our theory predicts that the economy will be more likely to encounter a financial crisis than a corruption-free economy regardless of the stage of development. This may help explain the financial crisis in some economies where corruption is regarded as a serious problem, such as Thailand.