University of Crete4th Conference in Macroeconomic Analysis May 25-28, 2000 The Causes of Banking Crises: What Do We Know? Pierre-Richard Agénor The World Bank
Role of banks in developing countries. • Recent evidence on banking sector problems and why we should care. • Definition of a banking crisis. • Causes of banking crises. • Empirical studies of the determinants of crises. • Some perspectives.
Main Functions of Banks • Transformation of short-term, liquid deposits held by households into illiquid liabilities issued by firms. • Delegated screening and monitoring of borrowers on behalf of depositors (mitigate information asymmetries). • Facilitate transactions by providing payment services.
Role in Industrial Countries • Bank loans in percent of total financial assets: small in the United States, large in Japan and Europe. Merrill Lynch estimates (April 2000): • United States.19.1% (1990), 24.2% (1995), 10.1% (1999). • Japan. 36.8% (1991), 35.2% (1995), 36.5 (1999). • Europe (Germany, France, and Italy). 46.9% in 1999. Germany only: 86.9% in 1995.
Role in Developing Countries • Bank credit: high in proportion of output and total credit allocated by financial institutions. • Highest ratios in Asia and Latin America. • Large share of bank credit goes to firms, to finance short-term working capital needs.
Equity and corporate bond markets: limited role in most developing countries as sources of finance. • Equity finance remains confined to the largest firms; not yet a significant competitor as an alternative to bank loans and retained earnings. • Corporate bonds markets remain quite narrow, concentrated, and relatively illiquid. • Banks often operate with high leverage (limited own capital). • Also low excess liquid reserves; higher volatility than in industrial countries would imply higher liquidity ratios than those actually observed. • Reason: often implicit bailout guarantees.
Banking Sector Problems Recent Evidence • Lindgren et al. (1996): at least two-thirds of IMF member countries experienced significant banking problems over 1980-96. • Caprio and Klingebiel (1996, 1999): evidence differs significantly from IMF estimates. • Nevertheless: incidence of banking crises in the 1980s and 1990s increased relative to the 1970s.
Systemic banking crises Episodes of non-systemic banking crises No crises Insufficient information Banking Problems since Late 1970s Source: Caprio and Klingebiel (1999).
Banking Sector Problems Why we Should Care • Key role in the payments system. • High resolution costs. Caprio and Klingebiel (1999): • Industrial countries. Most severe crises: • Spain, 1977-85 (17% of GDP); Finland, 1991-94 (11%); Sweden, 1991-94 (4%). • U.S. Savings and Loan crisis (1984-91): $175-$225 bn. (2.4-3% of 1990 GDP).
Developing countries. More than a dozen episodes with resolution costs higher than 10% of GDP. • Venezuela, 1994-... (more than 20%), Mexico, 1995-... (20%). • Argentina, 1980-82 (over 55%) Chile, 1982-85 (41%), Côte d'Ivoire, 1988-91 (25%). • East Asia crisis: large fiscal costs induced by bank restructuring (recapitalization and guarantees to depositors). • In Indonesia, bonds totaling some $68 bn. (around 45% of GDP) may need to be issued for the recapitalization of banks and resolution of closed institutions.
In Thailand, total cost of bank restructuring (in terms of public debt issued) is estimated at 32% of GDP; for Korea, 15-16%. • Pressure on fiscal deficits, public debt, and domestic interest rates (default risk premium). • Adverse incentive effects. • Intervention may reduce private incentives to monitor the behavior of banks in the future. • Expectation of future rescues creates incentives for excessive risk taking.
Reduction in bank credit and higher interest rates: adverse supply-side effects (small firms). • During a financial crisis: • Worsening of information and adverse selection problems. • Reason: only the least creditworthy borrowers are prepared to pay higher interest rates. • Adverse effect on the quality of loan portfolios and investment.
Constrains the conduct of monetary policy. • Limits on the possibility to raise interest rates. • Problematic when such response is needed to fend off speculative pressures. • Contraction in output that accompanies financial crises: asymmetric effect on poverty rates.
Definition of a Banking Crisis • Problematic. • Example: (Detragiache and Demirguc-Kunt (1998a)). • A distress episode is a crisis when • Ratio of nonperforming assets to total bank assets exceeded 10%. • Cost of the rescue operation was at least 2% of GDP. • Episode involved a large-scale nationalization of banks. • Extensive bank runs took place or emergency measures (deposit freezes, prolonged bank holidays, or generalized deposit guarantees) were enacted by the government.
Problems • Information on nonperforming loans: often not reliable and timely. Evergreening problem. • Cost of rescue operations is often difficult to measure. Importance of quasi-fiscal costs and contingent liabilities.
Estimating the net costs of banking sector restructuring is difficult; requires assumptions about • amount of liquidity support; • present and future incidence of nonperforming loans and their recovery rate. • Estimates are often calculated on a gross basis; lead to overestimation by excluding • future proceeds from reprivatization; • loan recovery; • repayment of the liquidity assistance provided by the government.
“Run” or “event” criterion: A crisis can indeed, in some cases, be dated that way. Examples: • Massive bank runs in Ecuador, following the currency crisis of February 1999. • The crisis in Indonesia, dated in reference to the closure of 16 banks in late 1997. Problems • Runs are often short lived. • Dramatic “events” rarely represent either the beginning, or the end, of the crisis. • In most cases insolvency problems were already present and worsened over a period of time; event itself is merely the point at which underlying problems are revealed (either to the regulator or the public).
Causes of Banking Crises Microeconomic Distortions and Institutional Failures • Mismatches between assets and liabilities. • Government intervention. • Weaknesses in the regulatory and legal framework. • Government guarantees and incentive failures. • Premature financial liberalization. Macroeconomic Factors • Domestic and exogenous shocks. • Lending booms. • The exchange rate regime. Self-Fulfilling Panics and Information-Based Runs
Macroeconomic Factors A. Macroeconomic Shocks • Both external and domestic. • Example 1: capital outflows induced by an increase in world interest rates or loss of confidence. • If these flows are intermediated, to begin with, via the banking system: • drop in deposits; • may force banks to liquidate long-term assets to raise liquidity or cut lending abruptly. • May entail a recession and a rise in default rates.
Example 2: increase in domestic interest rates (to reduce inflation or defend the currency). • Also weakens the ability of bank customers to service their loans and may lead to an increase in nonperforming assets or a full-blown crisis. • Clearly, the impact of these shocks on the banking system depends on their duration. • But volatility matters also. With highly volatile shocks, it is more difficult for banks to assess project quality and credit risk (distorted price signals). • Example: Jamaica (1994-99).
Macroeconomic Factors B. Lending Booms • Rapid increases in bank credit growth to the economy. • Source of increase in banks' capacity to lend: often large capital inflows. • Often at the expense of credit quality. • Distinguishing between good and bad credit risks is harder when the economy is expanding rapidly because many borrowers are temporarily profitable and liquid (Gavin and Hausman (1996)). • Boom is often accompanied by asset price bubbles.
Banking crisis may occur when the bubble bursts. • Collapse in equity prices: • affects overall confidence. • reduces profitability of bank debtors. • Collapse in real estate prices: • may also affect confidence. • reduces the value of collateral. • Crisis often exacerbated by a high degree of loan concentration (to groups and sectors). • Examples: East Asia, Latin America.
Macroeconomic Factors C. The Exchange Rate Regime • Fixed exchange rate regime with high degree of capital mobility: increases the fragility of the banking system to adverse external shocks. • Example: adverse shock that leads to a balance-of- payments deficit. • Lowers (with no sterilization) the money supply and leads to higher interest rates. • Higher cost of credit: increases the incidence of default and leads to a deterioration in the quality of bank portfolios.
Reserve losses may result from excessive expansion of domestic credit (Krugman-Flood-Garber model). • Rigid exchange rate regime (e.g. currency board): also constrains the lender-of-last-resort function of the central bank; prevents it from reacting quickly to stop a bank run by injecting liquidity. • Example: Argentina, 1994-95 (Tequila crisis).
Self-Fulfilling Panicsand Information-Based Runs • Costly panics may arise from sunspots. • Canonical model: Diamond and Dybvig (1983). • Ingredients for a bank run in the model: • Fractional reserves banking. • Sequential service constraint, that is, deposits can only be withdrawn sequentially. • Changes in expectations can be self-fulfilling.
For instance, depositors think that other depositors think that there will be a significant amount of cash withdrawals in the very near future. • With both fractional reserves and a first-come, first-served rule: depositors understand that if they are at the end of the sequential service line, they may not be able to withdraw their deposits and would suffer losses. • To avoid these losses all depositors try to place themselves at the head of the line, causing a panic in the process. • Extension to an open-economy setting: Chang and Velasco (1999).
Problems • Various analytical limitations; see Dowd (1992) and Freixas and Rochet (1997). • What kind of shocks would cause agents to decide that massive withdrawals are likely? • In practice, panic-induced runs tend to be short-lived and/or do not always have systemic implications for the banking system. • Banks are typically run after they become insolvent; healthy (solvent) banks are generally not run, and when they are, they do not go bankrupt. • No testable restrictions (at least no obvious ones).
Information-Based Runs • Empirical studies: e.g. Gorton (1988), Calomiris and Gorton (1991)) for the United States. • Practical importance of self-fulfilling runs is limited; what often triggers a run is a noisy signal (e.g. a recession) that nonetheless contains useful information about the bank's returns on its assets and its ability to redeem deposits at par. • Models stressing that runs may be triggered by changes in fundamentals: Gorton (1985), Jacklin and Bhattacharya (1988), and Allen and Gale (1998).
Example: Allen and Gale (1998). • Banks hold illiquid assets with risky returns. • A run on a particular bank may occur if depositors expect low returns on the bank's assets. • A run can turn into a crisis as a result of contagion or spillover effects on asset markets. • Reason: the banks that are subject to the initial run may attempt to sell their risky assets in order to meet depositors' demand for liquidity. • “Sunspot” view and information-based view: can be integrated, as in Chari and Jagannathan (1988). • Model that dwells on heterogeneity among depositors.
The signals approach. • Eichengreen and Rose (1998). • Kaminsky and Reinhart (1999); evidence on both banking and currency crises. • Limited dependent regression models. • Demirguc-Kunt and Detragiache (1998a, 1998b, 1999); • Glick and Hutchinson (1999); also evidence on both banking and currency crises.
The Signals Approach Methodology • Starts with the selection of a set of variables based on economic priors and data availability. • For each variable, the average level (or growth) in the period preceding a crisis is compared to that in tranquil periods. • Value that exceeds a threshold before a crisis: provides a warning signal. • Threshold calculated so as to minimize the number of false signals, relative to the number of crises predicted accurately (optimal noise-to-signal ratio).
Threshold level: either the same for all countries, or based on the country-specific empirical distribution of the variable. • Given individual warning signals, a composite leading indicator can be constructed as a weighted average of these individual signals (see Kaminsky (1999)). • In this procedure both the crisis indicator and the explanatory variables are transformed into dummies, larger or smaller than a given threshold. • Should work well if there are sharp changes between crisis episodes and periods of tranquility. • Applications: Eichengreen and Rose (1998), Kaminsky and Reinhart (1999), and Kaminsky (1999).
Kaminsky and Reinhart (1999) • Focus: links between banking and exchange rate crises. • Covers period 1970-95. Based on a group of 20 countries (14 developing countries); total of 26 banking crises. • Incidence of both types of crises increased sharply since the early 1980s. Average number per year of banking crises in the sample rose from 0.3 during 1970-79 to 1.4 in 1980-95. • Banking crises are identified by an event: either a bank run, or in the absence of a run, the closure, merging, takeover, or large-scale government assistance to at least one important financial institution.
Variables used as predictor of banking crises include: • Output and stock prices. • Financial sector variables. • Broad money multiplier, domestic credit-to-GDP ratio, • real deposit interest rates, bank lending rate spread, • broad money-official reserves ratio. • External sector variables. • Exports, imports, terms of trade, • real exchange rate, • changes in net foreign assets, • interest rate differentials.
Main findings • Banking and currency crises appear to share common causes. Before a crisis episode, several of the indicators begin to send stress signals. • Best predictors of banking crises are (in that order) • real exchange rate, broad money multiplier; • stock market prices, output, and real interest rates. • On average, earliest signals provided by the best predictors are between 6 to 18 months before a banking crisis occurs. • Banking crises: often preceded by financial liberalization. • In Latin America: collapse of bank deposits (relative to currency holdings) following a banking crisis.
Eichengreen and Rose (1998) • Sample: 39 crises in developing countries; period: 1975 to 1992. • Measures of banking crises from Caprio-Klingebiel. • Statistical tests for differences in the behavior of various macroeconomic and structural variables at a variety of leads and lags between crisis and non-crisis cases.
Main findings • Domestic macro conditions (slow domestic output growth, real appreciation) are significant but do not entirely explain banking crises. • Domestic credit growth, fiscal deficits, the current account, international reserves, and external debt are not significant. • Measures of financial fragility (ratio of broad money to reserves, the share of bank reserves in total bank assets, and the share of bank lending directed to the public sector) are also not significant.
Large and significant correlation between changes in interest rates in industrial countries and banking crises in developing countries. • Possible reasons: • rising world interest rates worsen access of banks from developing countries to offshore funds; • large capital outflows reduce the deposit base of domestic banks and precipitate a run.
Econometric Studies Methodology • Limited dependent regression models (probit models). • The banking crisis indicator is modeled as a zero-one variable, as in the signals approach. • Explanatory variables are not transformed into dummy variables, however, but are usually included in a linear fashion. • The probit function ensures that the predicted outcome of the model is always between zero and one.
Advantages of the regression approach over the signals approach: • Predictions of the model can be interpreted as measuring the probability of a crisis. • Method considers the significance of all the variables simultaneously; the additional information provided by new variables can easily be checked. • Indicators that are statistically significant are used to calculate the probability of a crisis occurring in a specific period. • Disadvantage of this approach: less easy to detect the impact of an individual variable on the probability of a crisis;
due to the nonlinearity of the probit function, the contribution of a particular variable depends on all the other variables as well. • Other practical problem: the number of crises is usually limited. Consequently, there are only a few ones in the sample, compared to a large number of zeros. • May result in poor estimation results. • To increase the number of ones: studies combine data from industrial and developing economies. • Pooling may not be valid due to significant structural differences among financial systems. • Applications: Detragiache and Demirguc-Kunt (1998a, 1998b, 1999); Glick and Hutchinson (1999).
Detragiache and Demirguc-Kunt (1998a) • Study of 45-65 banking crises for the period 1980-94. • Sample includes both developed and developing countries. • Basic source of banking crisis episodes: Caprio and Klingebiel (1996). • Multivariate probit regressions.
Main findings • Banking crises tend to erupt when growth is low and inflation and real interest rates are high. • Vulnerability to currency crises (e.g. high ratio of broad money to official reserves) also play a role. • Subsequent work (Detragiache and Demirguc-Kunt (1998b)): banking crises are more likely to occur in liberalized financial systems--when the institutional environment is weak (poor rule of law, quality of bureaucracy, contract enforcement). • Deposit insurance: also raises the probability of crisis in a weak institutional environment (Detragiache and Demirguc-Kunt (1999)).
Glick and Hutchinson (1999) • Evidence on both banking and currency crises. • Sample of 90 countries (with at least 72 with a serious banking problem), covering the period 1975-97. • 90 banking crises, of which 37 (41%) are “twin” crises. • Basic source of banking crisis episodes: Caprio and Klingebiel (1996). • Banking and twin crises have occurred mostly in developing countries. • Multivariate probit regressions.
Main findings • Decline in output growth, greater financial liberalization, (more flexible interest rate structure), and higher inflation are highly significant. • Currency crises are not significant in explaining the onset of banking crises (reverse is true). • Note: no out-of-sample test of predictability.