Michael Hutchison & Kathleen Mcdill - Determinants, Costs, And Duration Of Bank Sector DistressPdf
This paper examines episodes of banking sector distress for a large sample of countries,
highlighting the experience of Japan. We estimate a model that links the onset of banking
problems to a set of macroeconomic variables and institutional characteristics. The model
predicts a high probability of banking sector distress in Japan in the early 1990s, matching actual
developments closely, and suggests that the Japanese episode fits a well-established pattern
characterizing banking sector problems elsewhere. An empirical model explaining the output
cost of banking sector distress is also investigated. The results indicate that output loss is
smaller the more quickly banking sector problems are resolved and when exchange...
Determinants, Costs, and Duration of Banking Sector Distress:
The Japanese Experience in International Comparison
October 8, 1998
Michael Hutchison and Kathleen McDill*
Department of Economics
Social Sciences 1
University of California, Santa Cruz
Santa Cruz, CA 95064 USA
email: [email protected]
Abstract
This paper examines episodes of banking sector distress for a large sample of countries,
highlighting the experience of Japan. We estimate a model that links the onset of banking
problems to a set of macroeconomic variables and institutional characteristics. The model
predicts a high probability of banking sector distress in Japan in the early 1990s, matching actual
developments closely, and suggests that the Japanese episode fits a well-established pattern
characterizing banking sector problems elsewhere. An empirical model explaining the output
cost of banking sector distress is also investigated. The results indicate that output loss is
smaller the more quickly banking sector problems are resolved and when exchange rate stability
is maintained. Explicit deposit insurance also appears to lessen the output cost of banking sector
distress. The real output loss to Japan of not resolving banking sector problems is estimated at
almost 1 percent of GDP annually.
The authors thank the UC Pacific Rim Research Program, the International Centre for the Study
of East Asian Development and the UCSC Committee on Research and Division of Social
Sciences for financial support. This paper was prepared for presentation at the NBER-TCER
Japan Project Meeting in Tokyo, October 29-30, 1998.
1. Introduction
Recent events in Japan and East Asia draw renewed attention to the many problems
associated with financial sector distress— how quickly and unexpectedly crisis situations arise,
disruption in credit channels, economic contraction, and the difficulty in designing effective
policy responses. Japan’ banking problem emerged gradually in the early 1990s and has since
s
attracted increasing attention, evidenced most recently (October 1998) when representatives of
the Bank of Japan had a public row with the Ministry of Finance over measures of bank capital
and the extent of the non-performing loan problem.
The general features of banking sector distress in Japan are by now well recognized (e.g.
Cargill, Hutchison and Ito, 1997, 1998; Hutchison, 1997; OECD, 1998). In important ways the
Japanese case resembles episodes of banking sector distress seen in many countries-- booming
economies and sharply rising asset prices, followed by recession, severe asset price decline and
the emergence of banking problems. The international character of the asset price boom, and
subsequent collapse, suggests common explanatory factors.
This paper investigates the causes and consequences of banking sector distress in a large
sample of countries, highlighting the special circumstances of the Japanese case. We review
some of the basic statistical characteristics of countries experiencing banking sector distress and
test several empirical propositions about the factors that affect the probability of having banking
problems. In particular, we estimate a model that links the onset of 65 episodes of banking
sector distress to a set of macroeconomic variables and institutional characteristics. We employ
this model to predict the likelihood of banking sector distress emerging in Japan, and investigate
whether this episode fits an internationally recognized pattern.
We also seek to identify factors that influence the way economies respond to banking
sector distress. We focus on the “output cost” associated with episodes of banking sector
distress, i.e. the (present discounted value) loss in output that may be attributed to banking
problems. Beyond controlling for the state of the business cycle, potential determinants of the
output cost that we consider include institutional features, such as the existence of deposit
insurance, and policy measures such as the speed at which the banking problem resolved.
2
Our major finding is that Japan’ banking crisis follows a pattern found in many other
s
countries, and formal tests do not distinguish Japan as a special case. Our model predicts that
Japan was particularly “vulnerable” to banking sector distress in the early 1990s. That is, the
model indicated that there was almost a 20 percent probability of banking sector distress in
Japan in 1992 given the configuration of asset prices, credit conditions and other economic
factors prevailing at the time. The results also indicate that output loss associated with an
episode of banking sector distress is smaller the more quickly banking sector problems are
resolved and when exchange rate stability is maintained. Explicit deposit insurance also appears
to lessen the output cost of banking sector distress. The real output loss to Japan of not
resolving banking sector problems is estimated at almost 1 percent of GDP annually. The main
factor distinguishing Japan from other countries is the slow and poorly designed policy response
by the Japanese government to resolve the country’ financial crisis.
s
In the next section, Section 2, we briefly review the theoretical and empirical literature on
financial and banking sector distress. In section 3 we discuss the data for the study. In section 4
we present summary statistics, comparing the economic and institutional characteristics
distinguishing those countries that have experienced episodes of banking sector distress and the
characteristics of economies in the lead-up to and aftermath of episodes of banking sector
distress. This section also presents estimates of the probit model, and considers the predictions
of the model for Japan. Section 5 presents the estimates of the model linking the output cost of
banking problems to observable characteristics prior to the onset of the crisis and policy actions
taken following the crisis. Section 6 concludes the paper.
2. Analytical issues and empirical literature
Much of the theory on banking crises focuses on the special characteristics of banks, such
as maturity and currency transformation and asymmetric information, which make the industry
particularly vulnerable to collapse following adverse shocks (e.g. Jacklin and Bhattacharya,
1988 and Diamond and Dybvig, 1986). Institutional features of economies, such as the
existence of deposit insurance and market-determined interest rate structure, are also
emphasized in the literature as impacting the profitability of banks and the incentives of bank
3
managers to take on risk in lending operations. The special features of banks, combined with
particular institutional characteristics of economies, frequently lead to the emergence of banking
problems when adverse macroeconomic shocks such as a fall in asset prices (impacting bank
capital and/or collateral underlying loans) or economic activity (more delinquent loans) occurs.
Empirical regularities
Several common features of countries experiencing banking problems emerge from
numerous case studies. A recent IMF (1998) report summarizes this literature and identifies
several general categories of problems frequently associated with financial crises: unsustainable
macroeconomic policies, weaknesses in financial structure, global financial conditions, exchange
rate misalignments, and political instability. Macroeconomic instability, particularly
expansionary monetary and fiscal policies spurring lending booms and asset price bubbles, has
been a factor in many episodes of banking sector distress, including most experienced by the
industrial countries in the postwar period. External conditions, such as large shifts in the terms
of trade and world interest rates, have played a large role in financial crises in emerging-market
economies. By affecting the profitability of domestic firms, sudden external changes can
adversely impact banks’balance sheets.
Weakness in financial structure refers to a variety of circumstances ranging from the
maturity structure and currency composition of international portfolio investment flows to the
allocation and pricing of domestic credit through banking institutions. These weaknesses
oftentimes arise in times of rapid financial liberalization and greater market competition, when
banks are taking on new and unfamiliar risks on both the asset and liability side of balance
sheets. Weak supervisory and regulatory policies under these circumstances have also increased
moral hazard by giving an incentive for financial institutions with low capital ratios to increase
their risk positions in newly competitive environments, and allowing them to avoid full
responsibility for mistakes in monitoring and evaluating risk. Further, deficiencies in accounting,
disclosure, and legal frameworks contribute to the problem because they allow financial
institutions (or financial regulators) to disguise the extent of their difficulties. Governments have
4
frequently failed to quickly identify problem institutions, or to take prompt correct action when
a problem arises, resulting in larger and more difficult crisis situation.
Formal Studies on the Probability of Banking Sector Distress
Several recent econometric studies, moving beyond individual case studies and
descriptive international comparisons, have tested the statistical significance of a number of
variables in predicting banking sector distress. In terms of institutional characteristics, Kaminsky
and Reinhart (1996), estimating a probit model, find that financial liberalization increases the
probability of a banking crisis. DemirghH-Kunt and Detragiache (1998a) also use probit analysis
and find that financial liberalization is an important determinant of banking crises. In a related
study, DemirghH-Kunt and Detragiache (1998b) find that explicit deposit insurance (increasing
moral hazard) and low values of a “law and order” index (a proxy for a weaker regulatory and
supervisory structure) also appear to be important institutional characteristics increasing the
likelihood of a banking problem.
In terms of macroeconomic disturbances, DemirghH-Kunt and Detragiache (1998b) find
that low real GDP growth is contemporaneously associated with banking crises but is not a
useful leading indicator. They also find that (i) high real interest rates, (ii) high inflation, and (iii)
external vulnerability (measured by a high ratio of broad money to international reserves) help
predict the onset of a banking problem. Calvo (1996) argues that the M2/Reserves ratio is a
good predictor of a country’ vulnerability to balance-of-payments crises. Eichengreen and Rose
s
(1998), focusing on developing economies, find that increases in world interest rates,
overvalued real exchange rates and slowing domestic output growth increase the probability of
a banking problem. However, neither DemirghH-Kunt and Detragiache (1998b) nor
Eichengreen and Rose (1998) find that rapid credit growth help predict exchange rate crises.
Other variables commonly found in empirical work include the government budget surplus, e.g.
governments in a strong financial position may be likely to quickly re-capitalize problem banks,
and the terms-of-trade.
Output Cost of Banking Crises
5
The theory explaining the wide variation in economic outcomes following the emergence
of banking sector distress is much less well developed than that explaining why crises may arise.
Nor is there a consensus over the appropriate design of policy when confronted with a banking
problem. The literature guides empirical work to some extent, however, in that several
institutional characteristics and macroeconomic policy variables have been suggested as
potentially important factors influencing how economies develop after the emergence of banking
sector distress.
Three characteristics of economies, evident prior to the development of banking sector
problems, are noted as influencing the way economies respond: the state of the business cycle
(“prior state of the economy”), the rate of inflation (“prior inflation”), and whether a system of
explicit deposit insurance was in place (“explicit deposit insurance”). Controlling for the state of
the business cycle prior to the onset of banking problems is obviously important, as output
developments around trend are highly path dependent. The rate of inflation prior to the onset of
a banking problem may also influence the ultimate output cost. Financial sectors in countries
with relatively high inflation over a sustained period (defined here as average inflation over a
five-year period prior to the onset of banking sector problems) may already have adjusted to
financial sector dislocation, investing the resources to make them less vulnerable to a banking
sector problem. This is related to the “New Keynesian” view of how economies and financial
sectors adjust to the historical experience of sustained inflation (Ball, Mankiw and Romer,
1988)1.
Deposit insurance, combined with financial liberalization, creates conditions (moral
hazard) under which banking problems are more likely to occur. However, given that banking
sector distress is already evident, an economy with explicit deposit insurance may be able to
maintain greater confidence in the banking sector and limit contagion (e.g. depositor panic and
1
We also add an intercept dummy variable for very high inflation countries (“high inflation slope dummy”).
This is defined as countries with 30 percent average annual inflation or greater in the five-year period preceding
banking sector distress. Empirical evidence suggests that the response of real variables to inflation is often quite
different in those economies where high inflation is the norm.
6
bank runs). Most severe banking problems in recent decades were associated with deterioration
in asset quality, not with contagion associated with bank runs. Deposit insurance may play a
role in maintaining confidence in the banking system. Maintaining confidence, in turn, would
presumably be associated with continuation of banking relationships and credit flows, and less
output loss when a banking problem arises.
Three factors which are related to the policy response taken after banking sector distress
are also frequently discussed in the literature: duration of the banking problem (“duration”), the
change in policy interest rates and the rate of exchange rate depreciation. There is no single
quantitative measure of the speed, effectiveness and extent of policy actions taken to contain
and resolve banking sector distress. However, the duration of banking sector distress, measured
simply in years during which severe banking problems are observed, serves as a rough indication
of the adequacy or effectiveness of the policy response. Numerous episodes demonstrate that
effective regulatory policy can contain and resolve even severe banking problems quite quickly.
But there are other cases that demonstrate how effective policies may be stymied by institutional
or political factors. Regulatory authorities, for example, may follow a “forbearance” policy
(inaction) or the political process may not be able to reach a consensus over the appropriate
design and source of funds to re-capitalize financial institutions facing severe problems. The
duration of Sweden’ banking sector distress, for example, was four years and that of Japan is
s
seven years (1992-98) to date. This suggests a sharp contrast in the speed and effectiveness by
which the two countries responded to their respective banking problems.
The optimal policy response when confronted with banking problems is controversial. A
significant decline in policy interest rates (money market rate or discount rate) at the onset of
banking sector distress help offset the adverse effects on aggregate demand arising from
banking sector distress2. Declining interest rates, however, may exacerbate banking distress by
encouraging foreign capital outflows, weakening the exchange rate and generally lowering
investor confidence that prudent macroeconomic policies are being followed. These two
channels are likely to have offsetting effects on GDP, and the net impact is ambiguous.
2
The empirical results are not qualitatively different if short-term real interest rates are used in the regressions
rather than nominal rates.
7
The role of exchange rates in the stability of the banking sector is also debatable and is
only partially under the control of policymakers. Although exchange rate depreciation may be
expected to exert a positive demand stimulus, it also incurs losses to a financial sector holding
net foreign liabilities denominated in foreign currency. If the latter effect dominates, currency
depreciation may worsen the banking problems and lead to a larger output loss. This effect has
recently been observed in Korea, Indonesia and elsewhere in East Asia where domestic financial
institutions held large net (short-term) foreign liabilities denominated in foreign currency.
3. Data
We follow previous studies in guiding our selection of a number of the explanatory
variables. However, several variables, such as the definition of banking sector distress and
measurement of the output cost of the banking crisis, require explanation.
Defining Banking Distress
Banking problems are usually difficult to identify empirically because of data limitations.
The potential for a bank run is not directly observable and, once either a bank run or large-scale
government intervention has occurred, the situation most likely will have been preceded by a
protracted deterioration in the quality of assets held by banks. Identifying banking sector
distress by the deterioration of bank asset quality is also difficult since direct market indicators
of asset value are usually lacking. This is an important limitation since most banking problems in
recent years are not associated with bank runs (liability side of the balance sheet) but with
deterioration in asset quality and subsequent government intervention. Moreover, it is often
laxity in government analysis of banking fragility, and slow follow-up action once a problem is
recognized, that allows the situation to deteriorate to the point of a major bank crisis involved
large-scale government intervention.
Given these conceptual and data limitations, most studies have employed a combination
of events to identify and date the occurrence of a bank crisis. Institutional events usually include
forced closure, merger, or government intervention in the operations of financial institutions,
runs on banks, or the extension of large-scale government assistance. Other indicators
8
frequently include measures of non-performing assets, problem loans, and so on. We have
identified and dated episodes of banking sector distress following the criteria of Caprio and
Klingebiel (1997) and DemirghH-Kunt and Detragiache (1998). If an episode of banking
distress is identified in either study, it is included in our sample. If there is ambiguity over the
timing of the episode, we use the dating scheme of DemirghH-Kunt and Detragiache (1998)
since it tends to be more specific about the precise start and end of each episode3. We have
updated these samples using data from the Bank for International Settlements (1998).
Although the dating of banking sector distress is somewhat arbitrary, we nonetheless
follow these studies closely to avoid “data mining”, i.e. identifying the date of the banking crisis
after observing developments in macroeconomic and other variables thought to be determinants
of the crises. Japan’ banking crisis, for example, is dated by Caprio and Klingebiel (1997) as
s
“the 1990s” and by the DemirghH-Kunt and Detragiache criteria as starting in 1992. 1992 was
the first year of substantial government attention to the problem. However, the first substantial
plans for restructuring a significant part of the financial sector wasn’ until 1993 and many
t
observers wouldn’ characterize the Japanese banking problem as a full-blown “crisis” until
t
1995. On the other hand, using realistic estimates of non-performing loans as an indicator might
date the beginning of Japan’ banking distress already in 1991.
s
We initially consider data for 132 countries over the 1975-97 period, of which 67 had one
or multiple episodes of banking distress. We identify 82 episodes of banking distress of different
magnitudes. The minimum data requirements to be considered in this study is that GDP and
inflation are available, which in turn limits the sample to 98 countries. Of this group, 44
countries had no episodes of banking distress and 53 countries (65 episodes) had severe banking
problems at some time during the sample.
3
DemirghH-Kunt and Detragiache (1998) identify banking sector distress as a situation where one of the
following conditions hold: ratio of non-performing assets to total assets is greater than 2 percent of GDP; cost of
the rescue operation was at least 2 percent of GDP; banking sector problems resulted in a large scale
nationalization of banks; and extensive bank runs took place or emergency measures such as deposit freezes,
prolonged bank holidays, or generalized deposit guarantees were enacted by the government in response to the
crisis.
9
Institutional Variables
Our source on the existence of explicit deposit insurance is the recent survey on the issue
by Kyei (1995). We construct a dummy variable that takes a value of unity at times when a
particular country had a formal system of deposit guarantee arrangements in place, and zero
otherwise. In the Kyei study, forty-seven explicit arrangments were identified, as against fifty-
five arrangments implicitly guaranteeing government support for deposits. Our main source for
the financial liberalization data is DemirghH-Kunt and Detragiache (1998a), supplemented by
national and international sources. The variable is constructed on the basis of the beginning of
observed policy changes to liberalize interest rates, taking on a value of unity during the
liberalized period of market-determined rates and zero otherwise. Our central bank
independence variable is taken from Cukierman (1992). It takes on a value from zero (no
independence) to unity (complete independence), and represents a weighted average of legal
and institutional characteristics reflecting the central bank’ relationship with the government.
s
Macroeconomic Variables
The macroeconomic variables employed are standard in this literature: real GDP growth,
the rate of exchange rate depreciation, real credit growth, nominal (and real) interest rate
increase, inflation (and a dummy variable taking on a value of unity for countries with sample
average inflation rates over 30 percent, zero otherwise), the change in an index of stock prices,
and the budget position of the general government. We also consider the M2/Reserve ratio
(ratio of a broad money aggregate to international reserves) as an indicator of the extent to
which the country is vulnerable to balances-of-payments crises.
Output cost of banking sector distress
We measure the output cost associated with each episode of banking sector distress as
the present discounted value (at a 3 percent discount rate) of the cumulative output loss starting
from the year after which banking sector distress is identified. The Hodrick-Prescott filter is
used to estimate the “normal” or trend output level, and the output gap is measured as the
percentage difference of actual from trend output. When the output gap is negative following a
10
banking problem (49 episodes), the (discounted) sum of the output gap loss is calculated up to
the point to where the output returns to within 0.5 percent of trend. When the output gap is
positive following the onset of a banking problem (12 episodes), no recession had occurred, and
the single year’ (positive) output gap is considered an output gain.4
s
Section 4. Predicting Banking Crises
This section presents a model linking economic developments and institutional structures
to banking crises. We investigate whether economic and institutional characteristics of countries
are associated with the onset of banking crises, and use the model to see if Japan's banking
problems fit a pattern seen in other countries. Our objectives are both to investigate the general
characteristics associated with episodes of banking sector distress, and to determine whether
Japan's experience (or circumstances surrounding the banking crisis) is idiosyncratic. Using a
panel data encompassing 97 countries (65 episodes of banking distress) over the 1975-97
period, we use a multivariate probit analysis to estimate how a particular variable changes the
probability of the occurrence of banking sector distress holding constant the other explanatory
factors.
Summary Statistics
Table 1 shows the differences in economic characteristics between the group of
countries experiencing banking distress and the group that avoided severe problems. The
average values of these variables are calculated over the full sample period for those countries
which have not experienced banking sector distress and over the period leading up to the
banking crisis for the other countries. The objective is to identify different movements in these
variables that distinguish the crisis and non-crisis countries during the periods of relative
tranquillity, i.e. before banking problems become critical.
4
An alternative would be to drop from the sample those episodes in which an output did not
decline below trend following the onset of banking distress. This would lose potentially
revealing information, however, about those factors which contributing to such a favorable
outcome.
11
[Table 1 about here]
The first column of statistics show the mean values for the countries not experiencing
banking sector distress and the second column shows the mean values for the bank crisis
countries. The third column shows the mean difference (t-statistic) tests, and the fourth column
presents the corresponding value for Japan over the period prior to the banking crisis. The
standard deviations are shown in parentheses below the mean values.
The mean difference tests indicate that the rate of currency depreciation, average rate of
inflation, and the M2/Reserve ratio are significantly higher in countries struck by severe banking
problems. Real interest rates also appear to be marginally lower and stock price increases
somewhat higher (but not significant at the 10 percent level). Average values of real GDP
growth, real credit growth and budget deficits-— contrary to conventional accounts-- are not
clearly distinguishable between the two groups.
Where does Japan fit into this general pattern distinguishing economic developments in
the crisis countries from the non-crisis countries? By most indicators, Japan was in a very strong
position relative to most countries before banking problems arose. Similar to others
experiencing banking problems, Japan’ M2/reserve ratio was substantially higher, and rise in
s
stock prices faster, during the tranquil period than the control group. However, Japan
experienced more rapid GDP growth, less exchange rate depreciation (indeed, strong
appreciation), lower inflation, lower budget deficits and somewhat higher real interest rates than
even the non-crisis countries. Credit growth was also slower in Japan than either the crisis or
non-crisis countries.
Economic Developments Before and After Episodes of Banking Sector Distress
Table 2 shows the economic characteristics of the countries experiencing episodes of
banking sector distress at different periods: prior to the banking crisis, the first year of the
onset of the crisis, during the banking crisis, and after the crisis. The number in parenthesis
below the mean value is the standard deviation of the variable. The number in brackets below
12
the mean value is the probability that the sample mean for that period is different from the
immediately preceding period. For example, real GDP growth during the first year of banking
distress averaged 0.76 percent per annum (standard deviation of 5.55), a significant decline
[0.00 confidence level] from the 3.50 percent average (4.39 standard deviation) recorded during
the “tranquil” period before the crisis episodes.
[Table 2 about here]
The basic time-series statistics support the “asymmetric view”. Four variables indicate a
distinct shift over crisis episodes: real GDP growth, exchange rate depreciation, credit growth
and stock prices. The most striking feature for the full sample is the development of real GDP
and credit growth: strong growth rates prior to the crisis, recession and a contraction of credit
during the first year of the crisis, and moderate rebounds during the latter phase of the episode.
Exchange rate depreciation also jumps significantly at the onset of the crisis, and the rate of
stock price increase drops markedly.
In several respects, Japan experienced a similar pattern over time to other countries
experiencing bank crises: a booming economy (rapid real GDP and credit growth and rising
inflation) and strong asset markets (rapid stock price increase) prior to the bank crisis, followed
by a sharp slowdown and falling asset prices. All of these indicators suggest that recessionary
conditions and asset price deflation typically characterize banking sector distress, and that the
Japanese episode clearly fits this pattern. However, one striking feature of this table is evident:
Japanese GDP growth, credit growth and stock prices failed to recover at the same pace as
most other countries following the onset of the banking crisis (after the first year in the “during
crisis” period).
Basic Results
Table 3 reports the probit equation estimation results for four alternative model
specifications. The estimated coefficients (standard errors) for each model are reported in the
columns, as are the number of observations, pseudo R-square, and the percentage of crisis
13
episodes that are correctly predicted by the model (using a 10 percent probability as an indicator
of prediction success). Three groups of variables are considered: institutional characteristics,
macroeconomic developments and an “other” category. The number of observations ranges
from 391 to 1317, depending on data availability of the included regressors. Columns (1) - (3)
report the results using contemporaneous values of the regressors, and column (4) reports
results with the explanatory variables lagged one year. The “pseudo” R 2 ranges from 0.05 to
0.22, suggesting a moderate degree of explanatory power for the extended model presented in
column 3. Episodes of banking distress which are correctly predicted range from 15 percent to
57 percent, depending on the specification of the model.
The results indicate that, for the contemporaneous values of the explanatory variables
reported in column (1) – (3), the institutional characteristics (central bank independence, explicit
deposit insurance and financial liberalization), two of the macroeconomic variables (real GDP
growth and change in stock prices), and the OECD intercept dummy variable are statistically
significant (95 percent level) in most model specifications. In addition, real credit growth is
significant in one specification of the model.
[Table 3 about here]
In terms of the institutional characteristics, the coefficient estimates indicate that a high
degree of central bank independence decreases the probability of banking sector distress.
Explicit deposit insurance and a liberalized financial system, by contrast, increase the likelihood
of banking sector distress. In terms of macroeconomic developments, a sharp fall in real GDP
growth is a very good indicator that banking problems may emerge. A fall in stock prices is also
associated with an increased likelihood of banking sector distress.
By contrast with the conclusions of a number of individual case studies and descriptive
international comparison studies, we find that exchange rates, real credit growth, real interest
rates and inflation do not affect the likelihood of banking sector distress. That is, these variables
14
do not help predict the occurrence of banking sector distress after controlling for the movement
in stock prices, real GDP growth and institutional factors.5
Predictions for Japan
The model estimates reported in Table 3 cover all the episodes of banking sector distress
in the sample and may or may not do well in predicting the likelihood of banking sector distress
in any given country at a particular point in time. That is, the model could have relatively high
predictive accuracy in general but still not predict the occurrence of any particular banking
crisis. Our question is whether the general statistical characteristics of banking sector distress
identified by the model help to explain the timing and likelihood of the bank crisis that occurred
in Japan.
To this end, Figure 1 reports the predicted probability of banking sector distress
occurring in Japan during the 1980-97 period. The line labeled “predicted in-sample” uses the
coefficient estimates from column 3 of Table 3 (specifications including stock price changes) to
predict the probability of banking sector distress in Japan for each year. The probability was
below 5 percent until 1990, at which time the probability jumps to almost 10 percent. The
probability climbs further and peaks at about 20 percent in 1992. The estimated probability then
declines, returning to below the 5 percent level again by 1996. These results indicate that the
model does quite well in predicting the occurrence of the Japanese banking crisis. Since the
institutional variables are quite stable, the results are driven by the collapse in stock prices in
1990-92 and shift from strong to weak output growth.
[Figure 1 about here]
It is noteworthy that the model’ predictive accuracy is not driven by the fact that Japan is
s
included in the data sample from which the model coefficients are estimated, and then used to
5
We also investigated using alternative measures of economic activity (real GDP as a deviation from
potential GDP) and credit growth (nominal credit growth less nominal GDP growth) as explanatory variables.
The explanatory power of the output measure does not increase further, however, and credit remains
insignificant.
15
predict the probability of banking sector distress (an “in-sample prediction”). The line labeled
“predicted out-of-sample” in Figure 1 shows the predicted values for Japan when Japanese data
are excluded from the estimation equation6. The coefficients from the model (excluding Japan)
are then matched up with actual macroeconomic developments in Japan to predict the
probability of banking sector distress. Although somewhat lower in magnitude, the general
pattern of a sharply rising probability of banking sector distress is again clearly evident.
We also report the predicted values for Japan from the model specification excluding
stock price developments (column 2 of Table 3). This model has more observations than the
specification including stock prices but, of course, excludes an important explanatory variable.
Nonetheless, the model accurately predicts 56 percent of the banking sector distress episodes.
These predicted values, labeled “without stock prices” in Figure 1, also indicate an increasing
likelihood of banking problems in Japan in the early 1990s. The peak probability value is again
reached in 1992-93, but at a lower level (around 12 percent) than previously.
Leading Indicators of Banking Sector Distress
The results reported in columns (1) – (3) of Table 3 and Figure 1 are for macroeconomic
variables that are contemporaneously associated with banking crises and, hence, caution should
be exercised in interpreting these as causal relationships. It is possible that the onset of banking
sector distress, for example, may in turn trigger a fall in stock prices or currency depreciation.
In Japan’ case, however, we know that the fall in the stock market (which peak in December
s
31, 1989) preceded the banking crisis by a full two years, indicating a causal link running from a
collapsing stock market to a the onset of banking sector distress.
A number of previous studies have found that some variables may be useful leading
indicators (as opposed to contemporaneous) of banking sector distress. The descriptive
statistics reported in Table 2 also suggest that a discernable pattern to macroeconomic
developments may be evident prior to the onset of banking sector distress. We also estimated a
simple leading indicator model— a probit equation using lagged macroeconomic explanatory
6
The coefficient estimates are not shown for brevity, but are available from the author upon
request.
16
variables (one-year lag). These estimates are presented in column (4) of Table 3 and the
predicted values for Japan are presented in Figure 2. The model has moderate explanatory
power and correctly forecasts a high probability of a future banking crisis in 40 percent of the
actual occurrences.
[Figure 2 about here]
Surprisingly, most of the institutional variables (which do not change markedly over the
sample) are no longer statistically significant in the leading indicator model. The two
macroeconomic variables which appear to be good leading indicators of future banking
problems are high real interest rates and declining stock prices. None of the other
macroeconomic variables, including real GDP, appear to reliable indicators of future banking
distress. The predictions of the model for Japan, shown in Figure 2, follow a similar pattern to
the other results— a sharp rise in the likelihood of a banking problem in 1991 and a peak
reached in 1992 (at close to 15 percent probability).
5. The Output Cost of Banking Sector Distress
In this section we investigate the determinants of the output cost of banking crises. Given
the occurrence of a banking problem, we are concerned with the factors that play a role in
exacerbating the extent of the economic downturn that usually follows. Why is it that some
episodes of banking distress, such as those experienced by Mexico, Finland and Spain, are
associated with sharp recessions while economic activity in others, such as Paraguay and the
United States, appeared not to be affected?
The output cost of banking sector distress
Table 4 reports summary statistics on a number of features associated with banking sector
distress. The average duration of an episode of banking sector distress in the full sample was 3.9
years. Japan’ banking problem is substantially longer than the average duration. Prior to the
s
onset of banking sector distress, most countries were in an advanced state of the business cycle
17
(average 2.5 percent above trend GDP) and the Japanese economy, in particular, was in very
strong position (4.1 percent above trend in 1991).
[Table 4 about here]
The output loss of banking crises, taken as an average over the full sample (61 episodes),
was –7.3 percent of GDP. Limiting the sample to those countries experiencing recessions
following the onset of banking problems (49 countries), the output loss was –9.7 percent of
GDP and output was below trend for about 3 years. Japan’ output loss was less, estimated at
s
about –4.0 percent of GDP though 1997, but a weak economy has been evident for the past 5
or 6 years (output growth slowed substantially in 1992 and economic activity fell below trend in
1993) and continued into 19987.
Nominal interest rates tend to rise (1.3 percentage points) and exchange rates sharply
depreciate (20 percent) the first year of banking sector distress. Japan, by contrast, saw nominal
(and real) interest rates drop sharply in 1992 and the exchange rate remained fairly stable.
Regression results
Table 5 reports the results from the output cost regressions. Four alternative model
estimations are reported. Not surprisingly, the prior state of the economy is an important
control variable-- highly significant and negative in all of the regressions. Explicit deposit
insurance is also significantly positive, indicating that the existence of deposit insurance tends to
mitigate output declines following the onset of banking problems8. It appears that, once the
banking sector is already experiencing a problem, deposit insurance tends to dampen the
adverse output effects. This supports the view that, while explicit deposit insurance tends to
increase the probability of a banking problem emerging, it also seems to limit the serious of the
7
Our cyclical measure of output, upon which the output loss measures are based, indicate that
economic activity in Japan has continued to be below “normal” since 1993.
8
As noted in the previous section, however, the existence of deposit insurance raises the
probability of having some form of banking sector distress.
18
banking crisis. The mechanism at work is not identified in our empirical work, but deposit
insurance may limit bank runs and contagion once a banking problem arises9.
[Table 5 about here]
Prior inflation is also statistically significant and positive, indicating that countries with a
history of moderately high inflation (prior to the emergence of banking sector problems) tend to
experience less output disruption. However, the high inflation slope dummy variable is
significant and negative. For those countries with a history of high inflation (above 30 percent
annual average), there is no link between the inflation rate and the output cost of banking sector
distress (the sum of the two inflation coefficients is not significantly different from zero).
The duration variable is also highly significant and negative. The longer the banking crisis
continues, the greater is the output cost. The point estimates in Table 5 indicate that each
additional year that the banking crisis continues costs the economy about 0.8-0.9 percent of
GDP. By not resolving the banking crisis in 1995, for example, Japan has lost almost 3 percent
of (1996-98).
Increased interest rates appear to worsen the output cost associated with banking
problems and exchange rate stability (or appreciation) tends to mitigate the output cost. Only
the exchange rate effect is robust in the full regression, however, shown in column 5 of Table 5.
This may indicate that exchange rate depreciation when banking problems emerge tends to
weaken the financial system, create greater uncertainty, and lead to greater output costs.
6. Conclusion
The analysis of macroeconomic developments before and after banking sector crises
indicates that Japan followed a similar pattern to other countries experiencing banking
problems. The factors leading up to the Japanese banking crisis were not unique: financial
liberalization, expansionary credit growth, rapid real GDP growth and an asset price bubble.
9
With public capital at risk, deposit insurance may also provide an incentive to resolve banking
sector problems quickly.
19
The immediate aftermath also followed a pattern seen in other industrial economies
experiencing severe banking sector distress: recession and deflation, falling asset prices, and a
credit "crunch." Our empirical model predicts that Japan was particularly “vulnerable” to
banking sector distress in the early 1990s. That is, the probit model estimates indicate that there
was almost a 20 percent probability of banking sector distress in Japan in 1992 given the
configuration of asset prices, credit conditions and other economic factors prevailing at the time
Moreover, the adverse effects of the Japanese banking crisis appear no worse than
experienced by many countries. By comparison with episodes in other industrial countries,
however, the general malaise over the Japanese economy associated with banking sector
distress appears much greater. That is, the intensity of the crisis was felt longer and more
sharply in Japan than most other industrial countries. This is illustrated by the failure of asset
prices to recover and the weak GDP growth/ near-recession that prevailed in Japan over much
of the 1990s.
If the determinants and timing of the bank crisis in Japan seemed to conform to
experiences elsewhere, why then was the Japanese problem so severe and the sense of a crisis
situation so prolonged? We investigate this issue empirically by identifying factors that influence
the way economies respond to banking sector distress. We focus on the “output cost”
associated with episodes of banking sector distress, i.e. the (present discounted value) loss in
output that may be attributed to banking problems. Beyond controlling for the state of the
business cycle, potential determinants of the output cost that we consider include institutional
features, such as the existence of deposit insurance, and policy measures such as the speed at
which the banking problem resolved.
The results indicate that output loss associated with an episode of banking sector distress
is smaller the more quickly banking sector problems are resolved and when exchange rate
stability is maintained. The output cost of not resolving banking sector problems in Japan is
estimated at almost 1 percent of GDP annually. Our empirical results suggest that the low
interest rate policy adopted by Japan is the appropriate macroeconomic response to the severe
banking problem. The slow regulatory policy response and numerous half-measures to resolve
20