Transcript No Slide Title

University of Crete
4th Conference in Macroeconomic Analysis
May 25-28, 2000
The Causes of Banking Crises:
What Do We Know?
Pierre-Richard Agénor
The World Bank
1






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.
2
The Role of Banks
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.
4
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.

5
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.
6




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.
7
Banking Sector Problems:
Recent Evidence
and why we Should Care
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.
9
Banking Problems since Late 1970s
Systemic banking crises
Source: Caprio and Klingebiel (1999).
Episodes of non-systemic banking crises
No crises
Insufficient information
10
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).
11

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.
12
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.



13


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.
14


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.
15
Definition of a Banking Crisis
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
17
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.
18


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.
19
“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
20
(either to the regulator or the public).

Causes of Banking Crises
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
22
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.
23
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).




24
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.
25





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.
26
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-ofpayments 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.
27



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).
28
Self-Fulfilling Panics
and 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.
29




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, firstserved 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).
30
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).
31
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).
32



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.
33
Determinants of Banking Crises:
Empirical Evidence


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.
35
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).
36





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).
37
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.
38
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.
39
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.
40
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.
41
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.
42


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.
43
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.
44


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;
45






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).
46
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.
47
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)).
48
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.
49
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.
50
Perspectives



How financial sector inefficiencies can magnify the
incidence and cost of banking crises (see Agénor, Miller,
Vines, and Weber (1999)).
Specific example: Agénor and Aizenman (2000).
Limitations of the econometric evidence.
52
Financial Sector Inefficiencies
and Information-Based Bank Runs




See Agénor and Aizenman (2000).
Focuses on a particular type of financial sector
inefficiencies: high verification and enforcement costs
associated with loan contracts.
Combines costly state verification approach (Townsend
(1979)) and the model of limited enforceability of
contracts used in the sovereign debt literature (e.g.
Helpman (1989)).
Townsend’s approach: creditors can observe a debtor's
performance only by bearing a (fixed) monitoring cost.
53
In good states of nature, borrower honors his contractual
commitments, and creditors incur no costs.
 In bad states of nature, borrower cannot honor its
commitments, creditors must verify at a cost the debtor's
performance.
Features of the model
 Risk-neutral producers, banks, and consumers.
 Producers demand bank credit to finance investment and
cannot issue claims on future output (no collateral).
 Productivity shocks (both idiosyncratic and aggregate)
are random.
 Producer repays in good states of nature, and defaults in
bad states.

54




In case of default, creditors can confiscate a fraction of
the realized value of output; involves costly recourse to
the legal system.
Main result: the juxtaposition of macroeconomic
volatility and costly financial intermediation magnifies
the incidence of “fundamentals-based” bank runs.
More costly intermediation also compounds the losses
associated with greater aggregate volatility.
Argument: a mean-preserving spread of the distribution
of the aggregate shock raises the probability of default,
expected monitoring and enforcement costs, and the
lending rate; this lowers investment today and the
expected rate of return on bank assets--which may be
large enough to trigger a run.
55


If consumers are risk averse, the bank's equilibrium
deposit contract provides partial insurance against
adverse macroeconomic shocks, thereby mitigating the
incidence of fundamentals-based bank runs.
Framework can be extended in various directions; e.g.
effectiveness of policies aimed at reducing the incidence
of bank runs:
 Policies aimed at improving the efficiency of the
banking system;
 entry by foreign banks; more efficient (lower
verification and enforcement costs) and better be able
to diversify away domestic macroeconomic shocks.
56
Econometric Evidence: Limitations
Existing studies are subject to serious caveats.
 “Basic” datasets in many studies: IMF and World Bank
estimates; results are sensitive to sample chosen. Partly a
result of the lack of reliable banking data (e.g. on nonperforming loans).
 Variables are often incorrectly measured (exemple: real
exchange rate “overvaluation”: captured by the deviation
of the actual rate from a deterministic trend).
 Most studies use annual data--limited usefulness for
understanding the dynamics leading up to crises.
57




“Pooling” problem: cross-county studies typically assume
that the parameters characterizing the behavior of
(potential) indicators or explanatory variables in the
periods preceding crises are similar across time and
across countries.
Paucity of data on crisis episodes: remains a major
problem in the refinement of current models.
Both the signals and regression approaches define a crisis
as a discrete event; this does not account for the depth
of the crisis.
Studies do not capture the possibility of nonlinearities.
58


Existing models: do not reliably predict the timing of
banking crises. Example: out-of-sample probabilties for
the East Asia crisis generated by Demirguc-Kunt and
Detragiache (1998a).
However, some indicators do affect significantly the
probability of a crisis; they should alert policymakers.
Studies therefore have some value.
59
References






Agénor, Pierre-Richard, Banking Crises in Developing Countries:
Causes, Effects, and Prevention, unpublished, the World Bank
(Washington DC: May 2000).
Agénor, Pierre-Richard, and Joshua Aizenman, “Costly Financial
Intermediation and Information-Based Bank Runs,” unpublished,
the World Bank (May 2000).
Agénor, Pierre-Richard, Marcus Miller, David Vines, and Axel
Weber, eds., The Asian Financial Crisis: Contagion and Market
Volatility (Cambridge University Press: 1999).
Allen, Franklin, and Douglas Gale, ‘Optimal Financial Crises,’
Journal of Finance, 53 (August 1998), 1245-84.
Alonso, Irasema, ‘On Avoiding Bank Runs,’ Journal of Monetary
Economics, 37 (February 1996), 73-87.
Caprio, Gerard, Jr., and Daniela Klingebiel, “Bank Insolvencies:
Cross Country Experience,” Policy Research Working Paper No.
1620, the World Bank (July 1996).
61






------, ‘Episodes of Systemic and Borderline Financial Crises,”
unpublished, the World Bank (October 1999).
Chang, Roberto, and Andrés Velasco, “Liquidity Crises in Emerging
Markets: Theory and Policy,” Working Paper No. 7272, National
Bureau of Economic Research (July 1999).
Demirguc-Kunt, Asli, and Enrica Detragiache, “The Determinants of
Banking Crises in Developing Countries,” IMF Staff Papers, 45
(March 1998a), 81-109.
------, “Financial Liberalization and Financial Fragility,” Policy
Research Working Paper No. 1917, the World Bank (May 1998b).
------, “Does Deposit Insurance Increase Banking System Stability?
An Empirical Investigation, P olicy Research Working Paper No.
2247, the World Bank (November 1999).
Diamond, Douglas W., and Philip H. Dybvig, “Bank Runs, Deposit
Insurance, and Liquidity,” Journal of Political Economy, 91 (June
1983), 401-19.
62






Dowd, Kevin, ‘Models of Banking Instability: A Partial Review of
the Literature,’ Journal of Economic Surveys, 6 (- 1992), 107-32.
Eichengreen, Barry, and Andrew K. Rose, “Staying Afloat when the
Wind Shifts: External Factors and Emerging-Market Banking
Crises,” Discussion Paper No. 1828, Centre for Economic Policy
Research (April 1998).
Chari, V. V., and R. Jagannathan, ‘Banking Panics, Information, and
Rational Expectations Equilibrium,’ Journal of Finance, 43 (1988), 749-61.
Diamond, Douglas W., and Philip H. Dybvig, ‘Bank Runs, Deposit
Insurance, and Liquidity,’ Journal of Political Economy, 91 (June
1983), 401-19.
Freixas, Xavier, and Jean-Charles Rochet, Microeconomics of
Banking, MIT Press (Cambridge, Mass.: 1997).
Glick, Reuven, and Michael Hutchison, ‘Banking and Currency
Crises: How Common are Twins?,’ FRBSF (Sept. 1999).
63





Jacklin, Charles J., and Sudipto Bhattacharya, ‘Distinguishing
Panics and Information Based Bank Runs: Welfare and Policy
Implications,’ Journal of Political Economy,96 (June 1988), 568-91.
Kaminsky, Graciela, ‘Currency and Banking Crises: The Early
Warnings of Distress,’ Working Paper No. 99/178, International
Monetary Fund (December 1999).
Kaminsky, Graciela, and Carmen M. Reinhart, ‘The Twin Crises:
The Causes of Banking and Balance-of-Payments Problems,’
American Economic Review, 89 (June 1999), 473-500.
Lindgren, Carl-Johan, Gillian Garcia, and Matthew I. Saal, Bank
Soundness and Macroeconomic Policy, by International Monetary
Fund (Washington D.C.: 1996).
Townsend, Robert M., ‘Optimal Contracts and Competitive Markets
with Costly State Verification, ’ Journal of Economic Theory, 21
(October 1979), 265-93.
64