Full text of Working Papers (Federal Reserve Bank of San Francisco) : Banking and Currency Crises : How Common Are Twins?, Working Paper No. PB99-07

View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

Pacific Basin Working Paper Series

BANKING AND CURRENCY CRISES:
HOW COMMON ARE TWINS?

Reuven Glick
Research Department
Federal Reserve Bank of San Francisco
and

Michael Hutchison

Department of Economics
University of California, Santa Cruz
and
Visiting Scholar
Center for Pacific Basin Monetary and Economic Studies
Federal Reserve Bank of San Francisco

Working Paper No. PB99-07

Center for Pacific Basin Monetary and Economic Studies
Economic Research Department
Federal Reserve Bank of San Francisco

WORKING PAPER PB99-07

BANKING AND CURRENCY CRISES:
HOW COMMON ARE TWINS?
Reuven Glick

Center for Pacific Basin Monetary and Economic Studies
Research Department
Federal Reserve Bank of San Francisco

and
Michael Hutchison
Department of Economics
University of California, Santa Cruz
and
Visiting Scholar
Center for Pacific Basin Monetary and Economic Studies
Research Department
Federal Reserve Bank of San Francisco

December 1999

Center for Pacific Basin Monetary and Economic Studies
Economic Research Department
Federal Reserve Bank of San Francisco
101 Market Street
San Francisco, CA 94105-1579
Tel: (415) 974-3184
Fax: (415) 974-2168
http://www.frbsf.org

Banking and Currency Crises: How Common Are Twins?
December 1999

Reuven Glick
Federal Reserve Bank of
San Francisco
101 Market Street
San Francisco, CA
Email: reuven.glick@sf.frb.org

Michael Hutchison
Department of Economics
University of California, Santa Cruz
Social Sciences 1
Santa Cruz, CA 95064
Email: hutch@cats.ucsc.edu
Research Associate
Economic Policy Research Unit
University of Copenhagen

We thank Mark Peralta, Rasmus Fatum, and Kathleen McDill for research assistance. The
views presented in this paper are those of the authors alone and do not necessarily reflect those of the
Federal Reserve Bank of San Francisco or the Board of Governors of the Federal Reserve System.
Hutchison’s research was supported by The International Centre for the Study of East Asian
Development (ICSEAD) and the University of California Pacific Rim Research Program.
Forthcoming in Glick, (Cambridge University Press)

Abstract
The coincidence of banking and currency crises associated with the Asian financial crisis has
drawn renewed attention to causal and common factors linking the two phenomena. In this paper, we
analyze the incidence and underlying causes of banking and currency crises in 90 industrial and
developing countries over the 1975-97 period. We measure the individual and joint (“twin”)
occurrence of bank and currency crises and assess the extent to which each type of crisis provides
information about the likelihood of the other.

We find that the twin crisis phenomenon is most common in financially liberalized emerging
markets. The strong contemporaneous correlation between currency and bank crises in emerging
markets is robust, even after controlling for a host of macroeconomic and financial structure
variables and possible simultaneity bias. We also find that the occurrence of banking crises provides
a good leading indicator of currency crises in emerging markets. The converse does not hold,
however, as currency crises are not a useful leading indicator of the onset of future banking crises.
We conjecture that the openness of emerging markets to international capital flows, combined with a
liberalized financial structure, make them particularly vulnerable to twin crises.

Introduction
The joint occurrence of banking and currency crises associated with the recent Asian

financial turmoil has drawn renewed attention to the interrelationship between these two phenomena.
Banking and currency crises appeared to arise virtually at the same time in Thailand, Indonesia,
Malaysia, and Korea in 1997-98. In fact, the incidence of “twin” crises has been relatively
widespread, occurring in such diverse parts of the world as in Latin America in the early and mid1980s and in Scandinavia in the early 1990s.
There are good theoretical reasons to expect connections between currency and banking
crises, especially since foreign assets and liabilities are a component in commercial banks’ balance
sheets. In principle, the causality between bank and currency crises may run in either direction. As
we discuss in Section 2, bank crises may lead to currency crises under some circumstances, while
under other conditions currency crises may cause bank crises. Moreover, some recent literature does
not distinguish between the two phenomena and regards them as simultaneous manifestations of
underlying common factors (Chang and Velasco, 1999).
Most of the empirical literature on currency and banking crises has involved analyzing the
determinants of each type of crisis independently of the other. Little empirical work to date has
systematically investigated the association of bank and currency crises. The few exceptions (e.g.
Kaminsky and Reinhart, 1999; Rossi, 1999) typically restrict their data sets to a limited number of
countries experiencing crises. 1

An exception is Eichengreen and Rose (1998) who examine the impact of exchange rate regimes and
variability on the probability of bank crises in a large sample of developing countries.
1

In this paper, we empirically investigate the causal linkages between bank and currency
crises using a broad country and time-series data set. Using a broad control group of countries and
periods that includes observations with and without crises allows us to draw more general
conclusions about the conditions that distinguish crisis from tranquil periods both across countries
and across time.
In our empirical analysis, we first provide a detailed statistical overview of the individual and
joint (“twin”) occurrence of bank and currency crises for 90 industrial and developing countries over
the 1975-97 period. We examine the frequency, regional concentration, association, and relative
timing of the onsets of both bank and currency crises. In addition, we assess the value of banking
crises in helping to predict future currency crises, and vice versa, using signal-to-noise ratio
methodology. We also examine the contemporaneous and lagged relationship of currency and
banking crises more formally by estimating the probabilities of the onset of currency and banking
crises with probit regressions, using bivariate, multivariate, and simultaneous equation
specifications.
We find that the twin crisis phenomenon is concentrated in financially liberalized emerging
market economies and is not a general characteristic of either bank or currency
crises in a broader set of countries. The linkage between the onset of currency and bank crises in
emerging markets is strong, indicating that foreign exchange crises feed into the onset of banking
problems and vice versa. This result is robust to model specification and estimation technique.
Moreover, only in emerging market economies are banking crises a significant leading indicator of
future currency crises. Currency crises do not appear to be a particularly good signal of future
banking problems.

The organization of the paper is as follows: Section 2 describes the relevant literature on the
possible links between bank and currency crises. Section 3 discusses the data used in our empirical
analysis. Section 4 presents the summary statistical features of the data and signal-to-noise ratio
results. Section 5 presents the results of probability model (probit) estimates. Section 6 concludes the
paper.

Linkages Between Currency and Banking Crises
The association of bank and currency crises and the occurrence of “twin” crises may be

attributable to a number of channels of causation: a bank crisis leading to a currency crisis, a
currency crisis leading to a bank crisis, or joint causality. In this section, we provide a brief survey of
the existing literature concerning the linkages between the onset of bank and currency crises.

2.1.

Causality from Banking Sector Distress to Currency Crises
A number of papers discuss the possibility of causality running from banking problems to

currency crises. Obstfeld (1994), for example, argues that a weak banking sector may precipitate a
currency crisis if rational speculators anticipate that policymakers will choose inflation over
exchange rate stability in order to avoid bankruptcies and further strains on the banking sector rather
than endure the costs of defending the domestic currency. Velasco (1987) and Calvo (1997) argue
that a bank run can cause a currency attack if the increased liquidity associated with a government
bailout of the banking system is inconsistent with a stable exchange rate. Miller (1999) explicitly
considers currency devaluation as one of the logical policy options for a government confronted by a
bank run in a fixed exchange rate regime. Gonzalez-Hermosillo (1996) shows that a bank crisis may

lead to a currency crisis in a poorly developed financial system where agents may substitute foreign
assets for domestic assets.
If banking sector unsoundness can contribute to a currency crisis, what causes a banking
crisis? Leading candidate explanations include the well-known “moral hazard” problems in banking
associated with financial liberalization and government deposit insurance, and large macroeconomic
shocks such as a sharp fall in underlying asset values (e.g. “bubble” crash in asset prices). An
alternative, “non-fundamentals,” explanation is that “bank runs” may occur because of the
expectations of individual depositors and creditors (see Diamond and Dybvig, 1983).

2.2.

Causality from Currency Crises to Banking Sector Distress
A possible reverse chain of causality, from currency crises to the onset of banking crises, is

also well recognized. Miller (1996), for example, shows that a speculative attack on a currency can
lead to a bank crisis if deposit money is used to speculate in the foreign exchange market and banks
are “loaned up.” Rojas-Suarez and Weisbrod (1995) and Obstfeld (1994) argue that a currency crisis
may lead to problems in a vulnerable banking sector if policymakers respond to the pressure on the
exchange rate by sharply raising interest rates. A common feature of these mechanisms is that banks
are already “vulnerable” because of large unhedged foreign liabilities and/or a maturity mismatch
between asset and liabilities, and a shock arising from the currency market pushes them “over the
edge.” A currency crisis shock can adversely alter the banking sector directly by causing a
deterioration of bank balance sheets if the currency depreciates, or indirectly by causing the central
bank to raise interest rates to defend the currency.
If currency crises lead to bank crises, what causes currency crises? Candidate explanations
based on fundamentals, usually termed “first generation” models of the collapse of fixed exchange

rates, include overvalued real exchange rates and other macroeconomic factors such as inflation,
budget deficits, and rapid credit expansion (Krugman, 1979). The main alternative explanations,
based on the role of non-fundamentals, are frequently termed “second-generation” models of
exchange rate regime collapse (Obstfeld, 1994). This literature focuses on the existence of multiple
equilibria and self-fulfilling speculative attacks that can arise from the willingness of policymakers
to give up a pegged exchange rate if output and unemployment costs exceed a certain threshold.

2.3.

Joint Causality
The joint occurrence of “twin crises” may also reflect a response to common factors. Chang

and Velasco (1999), for example, emphasize the role of international illiquidity as a common
“fundamental, defined as a situation in which a country’s consolidated financial system has potential
short-term obligations that exceed the amount of foreign currency to which it can have access on
short notice. They argue that an international liquidity shortfall may be a sufficient, though not
necessary, condition to trigger a crisis: “The options left after creditors lose confidence and stop
rolling over and demand immediate payment on existing loans—whether to the private sector in Asia
or to the government in Mexico and Brazil—are painfully few. The collapse of the currency, of the
financial system, or perhaps both is the likely outcome.”
Another common fundamental factor emphasized in this literature is financial liberalization
combined with moral hazard incentives that induce banks to take on particularly risky portfolios,
including unhedged foreign currency liabilities. McKinnon and Pill (1996, 1998), for example,
emphasize the role of financial liberalization in generating dynamics leading to a twin crisis.
Financial liberalization and deposit insurance may fuel a lending boom involving both foreign and
domestic credit expansion that eventually leads to a banking and currency crisis.

More generally, Kaminsky and Reinhart (1999) point out that it is possible that “because the
seeds of the problems are sown at the same time, which event occurs first is a matter of
circumstance.” An example they employ to illustrate a twin crisis, jointly caused by common factors
or events, is the “perverse” dynamics of an exchange rate-based inflation stabilization plan, such as
that of Mexico in 1987 and the Southern Cone countries in the late 1970s. Reinhart and Vegh (1995)
provide empirical evidence that these types of plans have similar dynamics: an early consumption
boom is financed by expansion of bank credit and foreign borrowing. The boom is accompanied by
real exchange rate appreciation because domestic inflation only converges gradually to the
international inflation rate due to inertial effects in wage contracting and price expectations. At some
point, the high level of foreign borrowing, reflected in a current account deficit, may be perceived as
unsustainable and trigger an attack on the currency. As capital inflows turn to outflows and asset
markets crash, the banking sector is affected as well.

Data

3.1.

Defining Currency Crises
Currency crises are typically defined as “large” changes in some indicator of actual or

potential currency value. Some studies focus on episodes of large depreciation alone (e.g. Frankel
and Rose, 1996), while others include episodes of speculative pressure in which the exchange rate
did not always adjust because the authorities successfully defended the currency by intervening in
the foreign exchange market or raising domestic interest rates (e.g. Eichengreen, Rose, and Wyplosz,
1995; Moreno, 1995; Kaminsky and Reinhart, 1999). Alternative criteria have been employed in the
literature for identifying “large” changes in currency value or pressure relative to what is considered
“normal.” Some studies employ an exogenous threshold rate of depreciation common to all

countries in the analysis (e.g., Frankel and Rose, 1996; Kumar, Moorthy, and Penaudin, 1998), while
others define the threshold in terms of country-specific moments (e.g., Kaminsky and Reinhart,
1999; Kaminsky, Lizondo, and Reinhart, 1998; IMF, 1998; Esquivel and Larrain, 1998; Glick and
Moreno, 1998; Moreno, 1999). 2
In this study our indicator of currency crises is constructed from “large” changes in an index
of currency pressure, defined as a weighted average of monthly real exchange rate changes and
monthly (percent) reserve losses. 3 The weights are inversely related to the variance of changes of
each component over the sample for each country. Our measure presumes that any nominal currency
changes associated with exchange rate pressure should affect the purchasing power of the domestic
currency, i.e. result in a change in the real exchange rate (at least in the short run). This condition
excludes some large depreciations that occur during high inflation episodes, but it avoids screening
out sizable depreciation events in more moderate inflation periods for countries that have
occasionally experienced periods of hyperinflation and extreme devaluation. 4 Large changes in

Furman and Stiglitz (1998) and Berg and Patillo (1999) evaluate the predictive power of a range of model
methodologies and definitions for the 1997 Asia crisis.

Our currency pressure measure of crises does not include episodes of defense involving sharp rises in
interest rates. Data for market-determined interest rates are not available for much of the sample period in
many of the developing countries in our dataset.

This approach differs from Kaminsky and Reinhart (1999), for example, who deal with episodes of
hyperinflation by separating the nominal exchange rate depreciation observations for each country
according to whether or not inflation in the previous 6 months was greater than 150 percent, and calculate
for each sub-sample separate standard deviation and mean estimates with which to define exchange rate
crisis episodes.

exchange rate pressure are defined as changes in our pressure index that exceed the mean plus 2
times the country-specific standard deviation. 5, 6

3.2.

Defining Bank Crises
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 involving largescale government intervention.

Kaminsky and Reinhart (1999) use a three standard deviation cut-off. While the choice of cut-off point is
somewhat arbitrary, Frankel and Rose (1996) and Kumar, Moorthy, and Penaudin (1998) suggest that the
results are not very sensitive to the precise cut-off chosen in selecting crisis episodes.

We have also constructed an alternative measure of currency crises following Esquivel and Larrain (1998)
that employs a hybrid condition: the monthly depreciation in the (real) exchange rate either (i) exceeds 15
percent, provided that the depreciation rate is also substantially higher than that in the previous month, or
(ii) exceeds the country-specific mean plus 2 standard deviations of the real exchange rate monthly growth
rate, provided that it also exceeds 5 percent. The first condition insures that any large (real) depreciation is
counted as a currency crisis, while the second condition attempts to capture changes that are sufficiently
large relative to the country-specific monthly change of the (real) exchange rate. The results of our analysis
are unaffected by use of this alternative measure.
8

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 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 (1996, and updated on the
IMF WebPage) and Demirgüç-Kunt and Detragiache (1998a). 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 Demirgüç-Kunt and Detragiache (1998a) since it tends to be
more specific about the precise start and end of each episode. 7

3.3.

Determinants of Currency and Banking Crises
The theoretical and empirical literature has identified a vast array of variables potentially

associated with currency and banking crises (see Kaminsky, Lizondo, and Reinhart, 1998;
Demirgüç-Kunt and Detragiache, 1998a; and Hutchison and McDill, 1999). The choice of
explanatory variables in our analysis was determined by the questions we posed earlier, the
availability of data, and previous results found in the literature. Our objective is to postulate a
“canonical” model of currency and banking crises in order to form a basic starting point to
investigate the linkages between currency and banking crises. We postulate quite simple basic

Demirgüç-Kunt and Detragiache (1998a, 1998b) 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

models with few explanatory variables. The main source of the macro data is the International
Monetary Fund’s International Financial Statistics (CD-ROM). The data series and sources are
described in Appendix B.
The key explanatory variables used in our analysis of currency crises are the degree of real
currency overvaluation, export revenue growth, and the M2/foreign reserves ratio. Prior to episodes
of sharp depreciation, we expect the real trade-weighted exchange rate to be overvalued. We define
overvaluation as deviations from the fitted trend in the real trade weighted exchange rate, created by
taking the trade-weighted sum of the bilateral real exchange rates (defined in terms of CPI indices)
against the U.S. dollar, the deutschemark, and the yen, where the trade-weights are based on the
average bilateral trade with the U.S., Europe, and Japan in 1980.
We also expect export growth (in U.S. dollars) to be sluggish, and the growth rate of
M2/foreign reserves to be higher, prior to a currency crisis. A slowdown in export growth indicates a
decline in foreign exchange earnings that in turn may set up the expectation—and speculative
pressure—of a currency decline. A rise in the M2/foreign reserves ratio implies a decline in the
foreign currency backing of the short-term domestic currency liabilities of the banking system. This
would make it difficult to stabilize the currency if sentiment shifts against it.
Several other variables were considered in this study but were not included in the reported
regressions (for brevity) since they did not increase explanatory power: the current account/GDP
ratio, nominal and real M2 growth, nominal and real domestic credit (net of claims on the public
sector), M2/reserve money multiplier (often used as an indicator of the effects of financial
liberalization, as in Calvo and Mendoza, 1996), as well as the budget surplus/GDP ratio, etc. 8

We also do not consider possible contagion effects during currency crises. See Glick and Rose (1999).

The determinant s of bank crises that we considered in the basic canonical model are real
GDP growth, inflation, and financial liberalization. These are found to be significant determinants
(or associations) of banking crises by Demirgüç-Kunt and Detragiache (1998a) and Hutchison and
McDill (1999). The financial liberalization data is from Demirgüç-Kunt and Detragiache (1998b),
supplemented by national and international sources. It 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.
Several other variables were considered, but not reported since they did not contribute
significantly to the explanatory power of the model. These variables are real credit growth, nominal
(and real) interest rate changes, the budget position of the general government, and explicit deposit
insurance. 9 An index of stock prices was also considered and this entered significantly in
determining the onset of banking crises (see Hutchison and McDill, 1999). However, stock price
data was only available for a small sample of countries and was therefore not included in the base
regressions. 10

3.4.

Data Sample and Windows
Our data sample is determined by the availability of data on currency market movements and

banking sector health, as well as on the determinants of currency and bank crises, discussed above.

Data on the existence of explicit deposit insurance come from the survey by Kyei (1995). We constructed a
dummy variable that took on a value of unity if the country, at the time in question, had a formal system of
deposit guarantee arrangements in place, and zero otherwise. In the Kyei study, 47 explicit arrangements
were identified, as against 55 arrangements implicitly guaranteeing government support for deposits.

External conditions may also matter, but were not considered in our analysis. Eichengreen and Rose (1998)
find evidence that higher interest rates and slower growth in industrial countries contribute to bank crises in
emerging markets.

We do not confine our analysis to countries experiencing banking or currency crises. We also
include developed and developing countries that did not experience either a severe banking problem
or currency crisis/speculative attack during the 1975-97 sample period. Using such a broad control
group allows us to make general statements about the conditions distinguishing between countries
encountering crises and others managing to avoid crises.
The minimum data requirements to be included in our study are that GDP are available for a
minimum of 10 consecutive years over the period 1975-97. This requirement results in a sample of
90 countries. We group the countries into three categories: industrial countries (21), emerging
economies with relatively open capital markets (32), and other developing and transition economies
(37). 11 The particular countries included in our data set are listed in Appendix A. For each countryyear in our sample, we construct binary measures of currency and bank crises, as defined above (1 =
crisis, 0 = no crisis, i.e. tranquil). The dates of currency and bank crises are reported in Appendix B.
Of the 90 countries in our sample, 72 countries had banking problems, and 79 countries
experienced at least one currency crisis at some point during the sample period. Several countries
had multiple occurrences of banking crisis and most had multiple currency crises.
In most of our analysis we are concerned with predicting the onset of currency and banking
crises and their relative timing. To reduce the chances of capturing the continuation of the same
currency or banking episode, we impose windows on our data. In the case of currency crises, after
identifying each “large” change in currency pressure (i.e. two standard deviations above the mean),
we treat any large changes in the following 24-month window as a part of the same currency episode

Our emerging economy sample accords roughly with Furman and Stiglitz’s variant (1998) of that used by
Sachs, Tornell, and Velasco (1996), augmented to include Hong Kong and Uruguay but excluding China,
Israel, the Ivory Coast, and Taiwan. The full developing country sample excludes major oil exporting
countries. The United States is excluded from the sample as well.

and skip it before continuing the identification of new crises. In the case of multi-year banking
crises, we use only the first year in a spell of banking distress, i.e. the year of the banking crisis
“onset.” The duration of banking sector distress was greater than one year in most episodes.
We use annual crisis observations in our study. Attempting to date banking crises by month
(as in Kaminsky and Reinhart, 1999) or by quarter seems arbitrary. We employ monthly data for our
(real) exchange rate pressure index to identify currency crises and date each by the year in which it
occurs. Of course, annual data may obscure or limit some insights about the relative timing of the
onset of currency and banking crises, since it does not enable us to distinguish the lead and lag
timing of crises to the extent that crises occur at different points of the same year. However, we do
not believe that it is possible to date banking crises with such precision as monthly data presumes.
Moreover, using annual data enables inclusion of a relatively large number of countries in the
analysis (Kaminsky and Reinhart focus on a sample of only 20 countries).

The Incidence of Banking and Currency Crises
Table 1 summarizes the number and frequency of bank and currency crises according to our

definitions and disaggregates them by 5-year time intervals and development categories. 12 The table
also reports the incidence of “twin” crises, defined as instances in which a bank crisis is

These figures refer to observations for which data for both bank and currency crises are available; e.g. we
exclude observations where banking crisis data are available while currency crisis data are not, and vice
versa.

accompanied by a currency crisis in either the previous, current, or following year. 13 The data for
the developing countries are also disaggregated by geographic region.
Our sample includes 90 banking crisis episodes and 202 currency crises; thus currency crises
have been twice as common as bank crises. 14 Of the 90 bank crises, 37, i.e. 41 percent, have been
twins.
Observe that (the onset of) banking crises has increased over time: bank crises have risen
steadily both in number and frequency over our sample period and were four times as frequent in the
1990s than in the 1970s. However, the incidence of currency crises has been relatively constant. In
fact, the number and frequency of currency crises were higher in the 1980s than in the 1990s. The
frequency of twin crises appears to have risen in step with that of bank crises: in comparison to the
1975-79 period, they were more than three times as frequent in 1990-94, and more than four times as
frequent in 1995-97.
Table 1 also indicates that individual banking and currency crises as well as twin crises have
been more frequent in developing and emerging markets than in industrial countries. Banking and
twin crises have been particularly evident in emerging markets. Among developing countries, the
frequency of individual and twin crises has been highest in Africa (though the African figure may be
biased upwards because of heavy CFA zone participation and common devaluations by former
French colonies).

A larger window would obviously increase the number of “twins” identified. For example, Kaminsky and
Reinhart (1999), who define twin crises as bank crises followed by a currency crisis within four years,
identify 19 crises over the period 1970-1995 with their sample of 20 countries; we identify 37 crises — less
than twice as many — in a sample roughly four times as large. We implicitly consider a larger window for
classifying twin crises when exploring lag relationships up to two years in length between bank and
currency crises in the probit analysis in Section 5.

With our alternative definition of currency crises [see footnote 6], we identify 94 banking crises and 210
currency crises.
14

Tables 2 and 3 present summary non-parametric indicators of the extent to which the onset of
banking and currency crises are correlated with each other, using frequency statistics and signal-tonoise measures. Following the methodology of Kaminsky and Reinhart (1999) and Berg and Patillo
(1999), consider the association of bank and currency crises in terms of the following matrix:

Currency crisist

No currency crisis t

Bank crisis t

At, t

Bt, t

No bank crisis t

Ct, t

Dt, t

The cell At, t represents the number of instances in which a bank crisis occurring in a
particular year t, was accompanied by a currency crisis in year t (i.e. a bank crisis provides a “good
signal” about the occurrence of currency crises); Bt, t is the number of instances in which a banking
crisis was not accompanied by currency crisis (i.e. a bank crisis provides a “bad signal” or “noise”
about the occurrence of currency crises); Ct, t is the number of instances in which banking
performance failed to provide a good signal about a currency crisis that occurred; and Dt, t is the
number of instances in which neither
a banking or currency crisis occurred. An analogous matrix can be constructed indicating the number
of instances in which a banking crisis in year t was preceded (followed) by a currency crisis in year
t-1 (t+1), denoted by At, t-1 (At, t+1 ), etc.
Table 2 presents information about the association of the onset of banking and currency
crises contemporaneously, one period before, and one period ahead. Table 2a shows the frequency
with which the onset of a bank crisis in year t was accompanied by a currency crisis in either year t1, t, or t+1, i.e. At, t / (At, t + Bt, t ), At, t-1 / (At, t-1 + Bt, t-1 ), At, t+1 / (At, t+1 + Bt, t+1 ). The last column shows
the cumulative frequency with which a bank crisis onset in year t is accompanied by currency crises

in years t-1, t, or t+1, i.e. (At, t-1 + At, t + At, t+1 ) / (At, t + Bt, t ). Table 2b shows the analogous
measures of the frequency with which a currency crisis at time t was accompanied by the onset of a
bank crisis at either t-1, t, or t+1.
We calculate these frequencies for three different country data samples— all available
industrial and developing countries (90 countries), developing countries (69 countries), and
emerging markets only (32 countries). We are concerned here with the onset of either a banking or
currency crisis. We do not use windows in this exercise to exclude observations immediately
following or preceding the onset of a crisis, i.e. the onset of a crisis is coded as unity and all other
observations are coded as zero.
Comparing Tables 2a and 2b, observe that the frequency of banking crises associated with
currency crises is higher than the frequency of currency crises associated with banking crises. The
cumulative frequency with which the onset of a banking crisis is accompanied by a currency crisis
within one year before or after is 40 percent or higher. Correspondingly, the onset of a currency
crisis is accompanied by a banking crisis within one year by less than 20 percent of the time for the
full and developing country samples, though the frequency rises to 29 percent for the emerging
market sample.
Comparing the figures for the frequency of banking crisis accompanied by currency crises in
years t-1 and t+1 in Table 2a provides weak evidence that the frequency of currency crises
accompanying banking crises is higher in year t+1 than in year t-1. This suggests that currency crises
tend to lag banking crises, or equivalently, that banking crises tend to lead currency crises. This
result is strongest for emerging market countries, where 20 percent of banking crises in year t are
accompanied by a currency crisis in year t+1, but only 9 percent are at t-1.

Table 3 calculates the signal-to-noise association of banking and currency crises. Table 3a
reports the signal-to-noise performance of banking crises as a lagging (t-1), contemporaneous (t),
and leading (t+1) indicator of currency crises. For the contemporaneous indicator, this is defined as
the number of times a banking crisis is accompanied by a currency crisis (i.e. banking crises are
good signals of currency crises) as a share of total currency crises (i.e. At, t / (At, t + Ct, t )), all divided
by the number of times a banking crisis is not accompanied by a currency crisis (i.e. banking crises
are “noise” or bad signals of currency crises) as a share of all bank crises (i.e. Bt, t / (At, t + Dt, t )). A
signal-to-noise greater than 1 implies that when banking crises occur currency crises are more likely
than not. Table 3b reports the corresponding signal/noise measures for currency crises as an
indicator of banking crises.
Observe that for the full sample the signal-to-noise ratio of banking crises is higher for
currency crises at time t and t+1 than at time t-1. This is more pronounced for our developing
country and emerging market samples. This suggests that banking crises tend to be a
contemporaneous and/or leading, rather than lagging, indicator of currency crises.

Probit Equation Results
This section presents probit estimates involving currency and banking crises alone as well as

with various macroeconomic and institutional determinants of currency and banking crises. Our use
of probit models allows us to go beyond the bivariate relationship to focus on the joint contribution
of macroeconomic and institutional variables to currency and banking crises.
We estimate the probability of either currency or banking sector crises using a multivariate
probit model on an unbalanced panel data set for both developing and developed countries over the
1975-97 period (or most recent year available). We observe that a country at a particular time

(observation t) is either experiencing onset of a crisis (dummy variable, yt, takes on a value of unity),
or it is not (yt=0). The probability that a crisis will occur, Pr(yt=1), is hypothesized to be a function
of a vector of characteristics associated with observation t, x t , and the parameter vector ß. The
likelihood function of the probit model is constructed across the n observations (the number of
countries times the number of observations for each country) and (the log of the function) is then
maximized with respect to the unknown parameters using non-linear maximum likelihood

[

ln L = ∑t =1 y t ln F ( β ' xt ) + (1 − yt ) ln( 1 − F ( β ' xt ))
n

]

The function F(.) is the standardized normal distribution.
In these equations we employ windows following the onset of either a currency or banking
crisis. In the currency crisis equation, a 24-month window following the onset of a crisis (or episode
of exchange rate pressure) was employed and we eliminated from the data set these observations.
Banking crises are not as frequent as currency crises, so overlapping observations is not a major
problem, but the duration of banking crises is often quite long. We employ a window in these cases
such that every year of a continuing banking crisis, except the initial or onset year, was eliminated
from the data set.

5.1.

Bivariate Probits
We start with a discussion of the probit estimates for the currency and banking crisis onsets

alone, i.e. without controlling for macroeconomic variables. These results are reported in Tables 4a
and 4b. Tables 5a and 5b report results with macroeconomic and other control variables included. 15

All probit equations are estimated by maximum likelihood using LIMDEP windows version 7.0.

In each table we report the effect of a one-unit change in each regressor on the probability of
a crisis (expressed in percentage points so that .01=1%), evaluated at the mean of the data. We
include the associated z-statistics in parentheses; these test the null of no effect. Note that the sample
size of the multivariate probit analysis varies depending on the set of variables considered.
We also report various diagnostic measures. The in-sample probability forecasts are also
evaluated with “pseudo” R2 statistics and analogs of a mean squared error measure, the quadratic
probability score (QPS) and log probability score (LPS), that evaluate the accuracy of probability
forecasts. The QPS ranges from zero to 2, and the LPS ranges from zero to infinity, with a score of
zero corresponding to perfect accuracy for both. 16 For binary dependent variables, it is natural to ask
what fraction of the observations are “correctly called,” where, for example, a crisis episode is
correctly called when the estimated probability of crisis is above a given cut-off level and a crisis
occurs. Such “goodness-of-fit” statistics are shown for two probability cut-offs: 25 percent and 10
percent.
Table 4a shows the simple bivariate link between the onset of currency and banking crises. In
addition to contemporaneous links, we consider a simple one-year lagged effect of bank crisis onsets
as well as a composite lag if a bank crisis began in either of the two previous years. It is apparent
from these tables that currency crises are contemporaneously and significantly correlated with bank
crises for the emerging market and developing country samples, but not for the full sample of

For each of the methods we can generate n probability forecasts where Pt is the probability of a crisis in the
period t, 0 ≤ Pt ≤ 1 . Rt is the actual times series of observations; Rt = 1 if a crisis occurs at time t and
equals zero otherwise. The analog to mean squared error for probability forecasts is the QPS:

1 n
QPS = ∑ 2 Pt − Rt
n t =1

Large errors are penalized more heavily under the LPS, given by:

LPS =

g b

b g

1 n
∑ 1 − Rt ln 1 − Pt + Rt ln Pt
n t =1

countries. Lagged banking crises, occurring within the past two years, also help to predict the onset
of currency crises in emerging markets. Past banking crises, however, do not help predict the onset
of currency crises in either the developing country sample or the full set of countries.
Table 4b reports the corresponding bivariate results for probit regressions of currency crises
on the onset of banking crises. Contemporaneous, but not lagged, currency crises help explain bank
crises in the developing and emerging market samples. The contemporaneous link is weaker for the
full sample of countries, i.e. it is statistically significant at the 10 percent level in only one
formulation of the model. Thus lagged banking crises help predict currency crises in the emerging
markets sample, but not vice versa. This asymmetric result, albeit for a different and smaller sample
of countries, is consistent with the findings of Kaminsky and Reinhart (1999). 17

5.2.

Multivariate Probits
Table 5a reports the results where the onset of currency crises are explained by both the onset

of banking crises and a parsimonious set of macroeconomic variables, i.e. our canonical model. We
find that the macroeconomic variables lead the onset of currency crises and the estimates are
generally consistent with our priors. That is, the probability of a currency crisis generally rises with
greater real overvaluation, higher ratio of (log of) M2/Reserves, and lower export growth.
Overvaluation and M2/Reserves are generally significant for all of our three country samples; export
growth is significant only for the emerging country sample.

In contrast, Eichengreen and Rose (1998) find that neither contemporaneous nor lagged currency “crashes”
are significant in explaining bank crises for a large sample of developing countries.

The bank crisis variable, as an additional explanatory factor, is only significant for the
emerging country sample. As with the bivariate results, lagged as well as contemporaneous bank
crises help to predict future currency crises. 18
Analogous probit equations for the onset of bank crises with contemporaneous macro and
institutional control variables are reported in Table 5b. 19 A decline in output growth and greater
financial liberalization, as measured by a “liberalized” interest rate structure, are each highly
correlated with the onset of banking sector distress. Inflation is only correlated with the onset of
banking sector distress in the full sample, apparently proxying for the developing economies
(developing economies have a higher probability of having a banking crisis and also tend to have
higher inflation than industrialized economies). It is noteworthy that the macroeconomic variables
do not generally help predict the onset of a future banking crisis, i.e. (unreported) results with lagged
values of the macroeconomic variables are insignificant.
It is apparent that the onset of banking sector distress is highly correlated with currency
crises, as indicated by the contemporaneous association reported in Table 5b. In contrast with the
results in the previous table, the significance levels for the contemporaneous correlation between the
onset of banking crises and currency crises range from 1 to 5 percent in all three groups of countries,
i.e. the correlation holds not just in the emerging market sample, but also in the developing country
and full country samples. Once again we find no future predictive power associated with currency
crises—lagged currency crises are not significant in explaining the onset of bank crises onsets in any

These results are robust to excluding all 1997 observations, including the recent Asia crisis episodes, from
the data set.

Fewer observations are available for the bank crisis equations than for the currency crisis equations,
primarily because of limited availability of financial liberalization data.

of our samples. Lagged banking crises help predict currency crises in the emerging markets sample,
but not vice versa.

5.3.

Simultaneous Equation Probits
We have found significant contemporaneous correlation between banking and currency crises

with single equation probit estimation procedures. Table 6 shows the model estimates based on
simultaneous equation estimates of both the banking sector onset and currency crisis equations. 20 As
the table indicates, the basic results for the emerging markets sample are robust. There is clear joint
causality between the onset of currency and banking crises in the emerging markets sample.
However, no contemporaneous association is seen in the developing country sample (in contrast with
Tables 4a, 4b, and 5b) or in the full group of countries (in contrast with Tables 4b and 5b).
In summary, these results suggest a very strong and robust contemporaneous correlation
among the onset of banking and currency crises in emerging market countries, even when
controlling for simultaneity bias and a multitude of other explanatory factors such as financial
liberalization, export growth, real GDP growth, and so on. There is weaker evidence of this
contemporaneous link with a broader sample of developing countries and for the full sample of
countries. The other strong result that emerges is that banking crises are a statistically significant
leading indicator of currency crises in emerging markets.

Our simultaneous equation methodology follows Maddala (1983, pp. 246-7), which describes the procedure
for estimating the structural coefficients and standard errors in a two-equation system where both dependent
binary variables (in a probit context) are endogenous. The two-step procedure involves first estimating the
reduced forms for each endogenous crisis variable as a function of all exogenous and predetermined
variables by probit, then calculating the fitted values of the endogenous variables implied by the reduced
forms, and lastly using these fitted values as independent variables in the structural probit equations. The
covariance matrices are calculated as in Maddala (1983, p. 247). We do not use lags of our endogenouslydetermined crisis variables in these calculations. We assume that all other explanatory variables are
exogenous.
22

5.4.

Predicted Crisis Probabilities
To further illustrate the magnitude of the links between currency and bank crises we examine

how this association affects predicted crisis probabilities. Figure 1 reports
crisis probabilities implied by the single-equation probit estimates in Tables 5a and 5b for four East
Asian emerging market economies—Korea, Malaysia, Indonesia and Thailand—for the period 1989
to 1997. Two graphs are shown for each country: one depicts the probability predictions for the
onset of banking sector distress; the second depicts the onset of currency crises. Two prediction lines
are plotted in each graph: the solid line plots the predicted crisis probabilities implied by the
benchmark “canonical” probit estimates based only on macroeconomic and institutional variables;
the dashed line plots the predicted probabilities for currency (bank) crises implied by augmenting the
benchmark canonical model to include the occurrence of contemporaneous and lagged bank
(currency) crises. Vertical lines indicate the actual occurrence of a crisis.
Observe that the predicted probabilities of both currency and bank crises based on the
benchmark model increase in all four countries at the time of the 1997 Asia crisis. Including
information about the occurrence of other crises causes the predicted probabilities to increase even
more sharply. (The occurrence of a banking crisis in Korea in 1994 causes the predicted probability
of a currency crisis to rise even earlier.)
It should be emphasized that these plots are intended not to show the predictive power of our
model, but rather to illustrate the statistical importance of linkages between banking and currency
crises. 21

It should be noted that these are in-sample probability predictions. An alternative approach is to generate
out-of-sample probabilities for 1997 based on estimates generated from data through 1996.
23

Conclusions
This paper investigates the relative timing of the occurrence of banking and currency crises

over the 1975-97 period. For our sample of 90 countries, 72 had at least one case of a serious
banking problem and 79 experienced at least one currency crisis at some point during the
sample period. Several countries experienced multiple occurrences of banking crisis and most had
multiple currency crises. A total of 90 banking crisis episodes, 202 currency crises, and 37 twin
crises were identified. While the relative frequency of individual banking and twin crises has
increased over time, the frequency of currency crises has been relatively constant. Developing and
emerging market countries suffered both banking and currency crises more often than industrial
countries.
The twin crisis phenomenon, however, is mainly concentrated in a limited set of countries—
financially liberalized emerging-market economies. Summary statistics indicate an association
between crises in broader country groupings (including lesser developed and industrial countries),
but we find a robust link only in emerging markets. In emerging markets, banking crises (currency
crises) have been associated with currency crises (banking crises) almost 50 percent (30 percent) of
the time. This result holds up to a variety of tests—signal-to- noise ratios, bivariate probit
regressions, multivariate probit equations, and simultaneous probit estimates. A strong causal, joint
feedback, link between banking and currency crises appears only in this group of countries.
This result implies that, at least in financially liberalized emerging-market economies, policy
measures taken to help avoid a banking crisis (currency crisis) have the additional benefit of
lowering the probability of a currency (banking) crisis. Thus, measures to limit the exposure of
balance sheets and enhance confidence in the banking sector may reduce the incentives for capital

flight and currency runs. Similarly, policies designed to promote exchange rate stability appear
capable of fostering broader stability in domestic banking institutions.
Our analysis also provides evidence that banking crises provide some leading information
about the possibility of future foreign exchange instability, though again only for our emerging
markets group. Currency crises, by contrast, were not a good leading indicator of impending banking
problems. The power of banking crises to predict future currency instability does not appear to be
due to a common experience with financial liberalization (or other factors) since this is explicitly
taken into account by other variables in our estimation procedure. Instead, it might reflect the
footloose nature of capital flows into emerging markets, where the onset of banking problems can
quickly lead to capital flight and both current and future currency crises.

References

Berg, Andrew and Catherine Patillo (1999). “Are Currency Crises Predictable? A Test” International
Monetary Fund Staff Papers, 46, June, 107-138.
Calvo, Guillermo (1997). “Varieties of Capital-Market Crises” in Guillermo Calvo and Mervyn King
(eds.) The Debt Burden and Its Consequences for Monetary Policy, London: MacMillan Press.
Calvo, Guillermo and Enrique Mendoza (1996). “Mexico’s Balance of Payments Crises. A
Chronicle of a Death Foretold,” Journal of International Economics, 41, 235-264.
Caprio, Gerald and Daniela Klingebiel (1996). “Bank Insolvencies: Cross-country Experiences.”
World Bank Policy Research Paper 1620 (July).
Chang, Roberto and Andres Velasco (1999). “Liquidity Crises in Emerging Markets: Theory and
Policy.” NBER Working Paper 7272 (July).
Demirgüç-Kunt, Asli and Enrica Detragiache (1998a). “Financial Liberalization and Financial
Fragility.” IMF Working Paper WP/98/83 (June).
________ (1998b). “The Determinants of Banking Crises in Developing and Developed Countries.”
IMF Staff Papers, 45, March, 81-109.
Diamond, Douglas and Phillip Dybvig (1983). “Bank Runs, Deposit Insurance, and Liquidity.”
Journal of Political Economy, 91, 401-19.
Eichengreen, Barry and Andrew Rose (1998). “Staying Afloat When the Wind Shifts: External
Factors and Emerging-Market Banking Crises,” NBER Working Paper 6370 (January).
Eichengreen, Barry, Andrew Rose, and Charles Wyplosz (1995). “Exchange Market Mayhem. The
Antecedents and Aftermath of Speculative Attacks,” Economic Policy, 21, October, 249-312.
Esquivel, Gerardo and Felipe Larrain (1998). “Explaining Currency Crises,” Mimeo, Harvard
Institute for International Development (June).
Frankel, Jeffrey and Andrew Rose (1996). “Currency Crashes in Emerging Markets. An Empirical
Treatment,” Journal of International Economics, 41, November, 351-366.
Furman, Jason and Joseph Stiglitz (1998). “Economic Crises: Evidence and Insights from East
Asia,” Brookings Papers on Economic Activity, No. 2, 1-119.
Glick, Reuven and Ramon Moreno (1998), “Money and Credit, Competitiveness, and Currency
Crises in Asia and Latin America,” paper prepared for the 13th Pacific Basin Central Bank
Conference on “Monetary Policy and the Structure of the Capital Account” held in Los Cabos,
Mexico, November 7-11, 1998, and issued as Federal Reserve Bank of San Francisco, Center for
Pacific Basin Studies Working Paper PB98-07.

Glick, Reuven and Andrew Rose (1999). “Contagion and Trade. Why Are Currency Crises
Regional?” Journal of International Money and Finance, 18, August, 603-618. Earlier versions
issued as CEPR Discussion Paper 1947 and NBER Working Paper 6806.
Gonzalez- Hermosillo, Brend (1996). “Banking Sector Fragility and Systemic Sources of Fragility,”
IMF Working Paper WP/96/12 (February).
Hutchison, Michael M. and Kathleen McDill (1999). “Are All Banking Crises Alike? The Japanese
Experience in International Comparison,” Federal Reserve Bank of San Francisco, Center for Pacific
Basin Studies Working Paper PB99-02.
IMF (1995). World Economic Outlook, May. Chapter 4: “Financial Crises: Characteristics and
Indicators of Vulnerability.”
Kaminsky, Graciela, Saul Lizondo, and Carmen Reinhart (1998). “Leading Indicators of Currency
Crises,” International Monetary Fund Staff Papers, 45, March, 1-48.
Kaminsky, Graciela and Carmen Reinhart (1999). “The Twin Crises. The Causes of Banking and
Balance-of-Payments Problems,” American Economic Review, 89, June, 473-500. Earlier version
issued as Board of Governors International Finance Discussion Paper 544 (March).
Krugman, Paul (1979). “A Model of Balance of Payments Crises.” Journal of Money, Credit, and
Banking, 11, 311-25.
Kumar, Manmohan, Uma Moorthy, and William Perraudin (1998). “Determinants of Emerging
Market Currency Crises and Contagion Effects,” paper presented at CEPR/World Bank conference
“Financial Crises: Contagion and Market Volatility,” London, May 8-9.
Kyei, Alexander (1995). “Deposit Protection Arrangements: A Survey,” International Monetary
Fund Working Paper WP/95/134 (December).
Maddala, G.S. (1983). Limited-Dependent and Qualitative Variables in Econometrics. Cambridge,
United Kingdom: Cambridge University Press.
McKinnon, Ronald and Huw Pill (1996). “Credible Liberalizations and International Capital Flows.
The Overborrowing Syndrome,” in Takatoshi Ito and Anne Krueger (eds.), Financial Regulation
and Integration in East Asia, Chicago: University of Chicago Press.
_______ (1998). “The Overborrowing Syndrome: Are East Asian Economies Different?” in Reuven
Glick (ed.), Managing Capital Flows and Exchange Rates” Perspectives from the Pacific Basin,
Cambridge, United Kingdom: Cambridge University Press.
Miller, Victoria (1996). “Speculative Currency Attacks with Endogenously Induced Commercial
Bank Crises,” Journal of International Money and Finance, 15, June, 385-403.

_______ (1999). “The Timing and Size of Bank-Financed Speculative Attacks,” Journal of
International Money and Finance, 18, June, 459-70.
Moreno, Ramon (1995). “Macroeconomic Behavior during Periods of Speculative Pressure or
Realignment. Evidence from Pacific Basin Economies.” Federal Reserve Bank of San Francisco
Economic Review, 3-15.
_______ (1999). “Was There a Boom in Money and Credit Prior to East Asia’s Recent Currency
Crisis?” Federal Reserve Bank of San Francisco Economic Review No. 1.
Obstfeld, Maurice (1994). “The Logic of Currency Crises.” NBER Working Paper 4640,
(September).
Organization for Economic Co-operation and Development (1998). OECD Economic Outlook
(June).
Reinhart, Carmen and Carlos Vegh (1995). “Nominal Interest Rates, Consumption Booms, and Lack
of Credibility—A Quantitative Examination,” Journal of Development Economics, 46, April, 357378.
Rojas-Suarez, Liliana and Steven Weisbrod (1995). “Financial Fragilities in Latin America: The
1980s and 1990s.” International Monetary Fund Occasional Paper 132.
Rossi, Marco (1999). “Financial Fragility and Economic Performance in Developing Countries.”
IMF Working Paper WP/99/66 (May).
Sachs, Jeffrey, Aaron Tornell, and Andres Velasco (1996). “Financial Crises in Emerging Markets.
The Lessons from 1995,” Brookings Papers on Economic Activity, No. 1, 147-215.
Velasco, Andres (1987). “Financial Crises and Balance of Payments Crises: A Simple Model of the
Southern Cone Experience,” Journal of Development Economics, 27, October, 263-283.

Table 1. Bank and currency crises
Time distribution
19751997

19751979

19801984

19851989

19901994

19951997

Number

Frequency a

5.0

1.6

4.2

5.3

7.2

6.8

Number

202

Frequency a

11.3

11.0

12.0

12.6

11.6

8.0

Number

Frequency a

2.1

0.8

1.3

2.0

2.6

4.0

Bank Crises

Currency Crises

“Twin” Crises

Developmental and Geographic distribution
Developing
Industrial Developing Emerging

Africa

Asia

Latin
America

Otherb

Bank Crises
Number

Frequency a

4.4

5.2

6.6

5.8

5.0

5.1

4.8

Number

160

Frequency a

9.6

11.8

11.2

16.5

9.6

10.4

10.2

1.6

2.2

3.3

3.1

2.3

1.6

2.2

Currency Crises

“Twin” Crises
Number
Frequency a
Note:

“Twin” crises are defined as banking crises accompanied by a currency crisis in previous, current, or following year.
a Number of crises divided by total sum of country-years.
b Includes Eastern Europe and the Middle East.

Table 2a.

Bank crises and frequency of currency crises (percent)

Number of
bank crises

Frequency of accompanying
currency crisisa
t-1

t+1

Cumulative frequency
of accompanying
currency crisisb

All Countries

Developing Countries

Emerging Markets

Table 2b. Currency crises and frequency of bank crises (percent)

Number of
currency crises

Frequency of accompanying
bank crisisc
t-1

t+1

Cumulative frequency
of accompanying
bank crisisd

All Countries

202

Developing Countries

160

Emerging Markets

Note:
a
b
c
d

Frequency with which onset of bank crisis in year t is accompanied by currency crisis in year t-1, t, or t+1.
Total of currency crises in years t-1, t, and t+1 divided by banking crises in year t.
Frequency with which currency crisis in year t is accompanied by onset of bank crisis in year t-1, t, or t+1.
Total of bank crisis onsets in years t-1, t, and t+1 divided by currency crises in year t.

Table 3a.

Performance of bank crises as signal of currency crises

Good signal/noise ratio of currency crisesa
t-1

t+1

All Countries

.98

1.44

1.42

Developing Countries

.82

1.66

1.35

Emerging Markets

.77

2.46

1.96

Table 3b.

Performance of currency crises as a signal of bank crises

Good signal/noise ratio of bank crisesb

Note:

t-1

t+1

All Countries

1.38

1.40

0.98

Developing Countries

1.32

1.59

0.82

Emerging Markets

1.87

2.30

0.78

Number of years in which the onset of a bank crisis in year t is accompanied by a currency crisis in year t-1, t, or t+1 (i.e.
bank crises are good signals) as a proportion of possible instances in which a currency crisis could have occurred,
divided by the number of years a bank crisis in year t is not accompanied by a currency crisis in year t-1, t, or t+1 (i.e.
banking crises are “bad” signals) as a proportion of all bank crises.

Number of years a currency crisis in year t is accompanied by a bank crisis onset in year t-1, t, or t+1 (i.e. currency crises
are good signals) as a proportion of possible instances in which a bank crisis could have occurred, divided by the number
of years a currency crisis in year t is not accompanied by a bank crisis in year t-1, t, or t+1 (i.e. currency crises are “bad”
signals) as a proportion of all currency crises.

Table 4a.

Probit regression estimates for currency crises
All Countries

Developing Countries

Emerging Markets

Variable
Bank Crisis t

4.89

5.38

5.60

6.64 *

7.00 *

7.16 *

11.35 **

12.26 ***

12.98 ***

(1.38)

(1.51)

(1.56)

(1.67)

(1.77)

(1.81)

(2.52)

(2.78)

(2.96)

Bank Crisis t-1
Bank Crisis t-1 or

4.71

4.58

10.58 **

(1.29)

(1.06)

(2.14)

t-2

4.48

3.86

11.03 ***

(1.63)

(1.19)

(2.98)

Summary statistics
No. of Crises

202

193

160

152

No. of Observations

1587

1520

1196

1147

615

589

Log likelihood

-604.0

-576.7

-576.2

-469.3

-446.7

-446.6

-230.9

-215.3

-213.3

Pseudo-R2

0.28

0.29

0.30

Quadratic Probability Score

0.22

0.23

0.22

0.21

Log Probability Score

0.38

0.39

0.38

0.37

0.36

Goodness-of-fit (25% cutoff)a
% of observations correctly called

% of crises correctly called

100

% of non-crises correctly called

Goodness-of-fit (10% cutoff)a
% of observations correctly called

% of crises correctly called

100

% of non-crises correctly called
Note:

The table reports the change in the probability of a crisis in response to a 1 unit change in the variable evaluated at the mean of
all variables (x 100, to convert into percentages) with associated z-statistic (for hypothesis of no effect) in parentheses below.
Significance at 10 percent level is denoted by *; at the 5 percent level by **; at the 1 percent level by ***. Constant included, but
not reported.
Goodness-of-fit statistics defined respectively as (A + D) / (A + B + C + D), A / (A + C), and D / (B + D), where A (C) denote
number of crises with predictions of crises above (below) probability cutoff and B (D) denote number of corresponding noncrises with predictions of crises above (below) the cutoff.

Table 4b.

Probit regression estimates for bank crises onsets
All Countries

Developing Countries

Emerging Markets

Variable
Currency Crisis t

2.70

2.85

3.21 *

3.80 *

3.88 *

4.31 **

9.72 ***

10.97 ***

11.26 ***

(1.54)

(1.52)

(1.78)

(1.94)

(1.82)

(2.10)

(3.15)

(3.29)

(3.40)

Currency Crisis t-1
Currency Crisis t-1 or

1.06

0.28

1.44

(0.53)

(0.11)

(0.34)

t-2

2.16

1.61

2.71

(1.49)

(0.92)

(0.89)

Summary statistics
90

No. of Observations

1537

1443

1470

1152

1079

1103

562

530

536

Log likelihood

-341.6

-327.5

-333.5

-264.8

-254.9

-261.1

-154.5

-151.3

-151.4

Pseudo-R2

0.20

0.21

0.25

0.26

Quadratic Probability Score

0.11

0.12

0.15

Log Probability Score

0.22

0.23

0.24

0.27

0.29

0.28

No. of Crises

Goodness-of-fit (25% cutoff)a
% of observations correctly called

% of crises correctly called

100

% of non-crises correctly called

Goodness-of-fit (10% cutoff)a
% of observations correctly called

% of crises correctly called

100

% of non-crises correctly called

Note:

See Table 4a.

Table 5a.

Probit regression estimates for currency crises
All Countries

Developing Countries

Emerging Markets

Variable
Overvaluation t-1
Ln (M2/Reserves) t-1
Export Growth t-1
Bank Crisis Onset

Bank Crisis Onset

t-1

0.26 ***
(6.83)

0.25 ***
(6.76)

0.24 ***
(6.26)

0.23 ***
(5.81)

0.22 ***
(5.74)

0.21 ***
(5.31)

0.22 ***
(4.23)

0.21 ***
(4.08)

0.18 ***
(3.54)

0.96

1.11

1.58 *

1.59 *

1.62 *

3.19 ***

3.11 ***

(1.23)

(1.26)

(1.42)

(1.80)

(1.81)

(1.82)

(2.64)

(2.68)

(2.61)

-0.048

-0.050

-0.046

-0.05

-0.052

-0.056

-0.16 **

-0.17 **

(1.16)

(1.20)

(1.06)

(1.14)

(1.19)

(1.22)

(2.03)

(2.00)

(2.11)

4.26
(1.22)

4.76
(1.35)

5.01
(1.30)

5.72
(1.48)

8.82 **
(2.10)

10.51 **
(2.54)

or t-2

2.60

3.65

8.69 **

(0.92)

(1.16)

(2.40)

Summary statistics
No. of Crises

183

174

151

143

No. of Observations

1471

1408

1145

1097

601

575

Log likelihood

-522.5

-521.8

-499.0

-421.3

-420.5

-400.8

-213.1

-211.0

-196.9

Pseudo-R2

0.32

0.31

0.32

0.34

0.35

Quadratic Probability Score

0.21

0.22

0.21

0.20

Log Probability Score

0.36

0.35

0.37

0.35

0.34

Goodness-of-fit (25% cutoff) a
% of observations correctly called

% of crises correctly called

% of non-crises correctly called

Goodness-of-fit (10% cutoff) a
% of observations correctly called

% of crises correctly called

% of non-crises correctly called

Note:

See Table 4a.

Table 5b.

Probit regression estimates for bank crisis onsets
All Countries

Developing Countries

Emerging Markets

Variable
Inflation t
Output Growth t
Fin. Liberalization t

0.023 *
(1.88)

0.021 *
(1.68)

0.023 *
(1.74)

0.009
(0.61)

0.006
(0.41)

0.008
(0.56)

0.006
(0.23)

0.002
(0.07)

0.006
(0.26)

-0.56 ***

-0.54 ***

-0.58 ***

-0.65 ***

-0.60 ***

-0.68 ***

-1.42 ***

-1.20 ***

-1.43 ***
(3.80)

(3.64)

(3.30)

(3.40)

(3.56)

(3.22)

(3.40)

(4.08)

(3.53)

7.74 ***

7.96 ***

7.99 ***

9.82 ***

10.11 ***

6.13 *

6.96 **

5.68

(5.28)

(5.26)

(4.91)

(5.18)

(4.97)

(1.84)

(2.16)

(1.63)

4.26 **
(2.26)

4.41 **
(2.21)

6.04 **
(2.53)

6.09 **
(2.38)

11.26 ***
(3.06)

11.03 ***
(2.77)

Currency Crisis t
Currency Crisis t-1 or t-2

0.081

-1.12

-2.22

(0.04)

(0.47)

(0.54)

Summary statistics
No. of Crises

No. of Observations

960

903

862

560

545

521

336

335

320

-200.8

-190.4

-186.3

-131.1

-124.4

-123.2

-92.9

-87.9

-85.7

Pseudo-R2

0.32

0.33

0.36

0.37

0.38

0.35

0.38

0.39

Quadratic Probability Score

0.11

0.13

0.12

0.13

0.16

0.15

Log Probability Score

0.21

0.22

0.23

0.24

0.28

0.26

0.27

Log likelihood

Goodness-of-fit (25% cutoff) a
% of observations correctly called

% of crises correctly called

% of non-crises correctly called

Goodness-of-fit (10% cutoff) a
% of observations correctly called

% of crises correctly called

% of non-crises correctly called

Note:

See Table 4a.

Table 6. Simultaneous probit regression estimates
All Countries
Variable

Currency crisis

Developing Countries

Bank crisis

Currency crisis

Bank crisis

Emerging Markets
Currency crisis

0.24 ***
(4.46)

0.16 ***
(2.58)

0.16 *
(1.84)

Ln (M2/Reserves) t-1

1.88
(1.51)

4.11 **
(2.28)

4.08 *
(1.84)

Export Growth t-1

-0.048
(0.68)

-0.062
(0.76)

-0.18
(1.52)

1.82
(0.74)

4.16
(1.53)

7.44 ***
(2.64)

Overvaluation t-1

Bank Crisis Onset

Bank crisis

Inflationt

0.02
(1.44)

0.0022
(0.14)

-0.0042
(0.18)

Output Growth t

-0.38 **
(2.09)

-0.48 **
(2.02)

-0.74 *
(1.66)

Fin. Liberalization t

7.98 ***
(3.54)

11.18 ***
(4.00)

9.61 **
(2.18)

3.48
(1.26)

5.04
(1.44)

8.43 **
(2.3)

Currency Crisis t

Summary statistics
No. of Crises

No. of Observations

730

463

303

-242.3

-158.0

-160.4

-116.4

-92.6

-84.8

Pseudo-R2

0.31

0.30

0.34

0.36

0.38

0.40

Quadratic Probability Score

0.18

0.20

0.21

0.18

0.19

Log Probability Score

0.32

0.38

0.34

0.41

0.31

0.33

Log Likelihood

Goodness-of-fit (25% cutoff)a
% of observations correctly called

% of crises correctly called

% of non-crises correctly called

Goodness-of-fit (10% cutoff) a
% of observations correctly called

% of crises correctly called

% of non-crises correctly called

Note:

The table reports the change in the probability of a crisis in response to a 1 unit change in the variable evaluated at the mean of
all variables (x 100, to convert into percentages) with associated z-statistic (for hypothesis of no effect) in parentheses below.
Significance at 10 percent level is denoted by *; at the 5 percent level by **; at the 1 percent level by ***. Constant included, but
not reported. Coefficients and standard errors are adjusted for simultaneous equations bias, as discussed in text.
Goodness-of-fit statistics defined respectively as (A + D) / (A + B + C + D), A / (A + C), and D / (B + D), where A (C) denote
number of crises with predictions of crises above (below) probability cutoff and B (D) denote the corresponding number of noncrises with predictions of crises above (below) the cutoff.

Figure 1. Crisis Probability Predictions
Indonesia
Predicted Onset of Banking Crisis

0.5

0.4

0.3

0.2

0.1

0
1989

1990

1991

1992

1993

1994

1995

1996

1997

Korea
Predicted Onset of Bank Crisis

0.5

0
1989

0.4

0.3

0.2

0.1

1990

1991

1992

1993

1994

1995

1996

1997

Malaysia
Predicted Onset of Bank Crisis

0.5

0
1989

0.4

0.3

0.2

0.1

1990

1991

1992

1993

1994

1995

1996

1997

0
1989

1991

1992

1990

1991

1992

1990

0.3

0.2

0.1

Note:

1991

1992

1993

1994

1996

1997

1993

1994

1995

1996

1997

1991

1992

1993

1994

1995

1996

1997

1996

1997

Thailand

0.4

1990

1995

Predicted Onset of Currency Crisis

0.5

0.4

0
1989

1994

Malaysia
Predicted Onset of Currency Crisis

Thailand
Predicted Onset of Bank Crisis

0.5

1993

Korea
Predicted Onset of Currency Crisis

0.5

0.4

0
1989

1990

0.5

0.4

0
1989

Indonesia
Predicted Onset of Currency Crisis

1996

1997

0
1989

1990

1991

1992

1993

1994

1995

Solid lines indicate currency (bank) crisis probabilities implied by benchmark probit equations. Dashed lines indicate currency
(bank) crisis probabilities implied by probit equations augmented to include the contemporaneous and composite lagged
occurrence of bank (currency) crises. Vertical lines denote the actual occurrence of a crisis.

Appendix A

Note:

Industrial Countries

Emerging Markets

Other Developing

Austria
Belgium
Canada
Denmark
Finland
France
Germany
Greece
Iceland
Ireland
Italy
Japan
Luxembourg
Netherlands
New Zealand
Norway
Portugal
Spain
Sweden
Switzerland
United Kingdom

Argentina
Bangladesh
Botswana
Brazil
Chile
Colombia
Ecuador
Egypt
Hong Kong
Ghana
India
Indonesia
Jordan
Kenya
Korea
Malaysia
Mauritius
Mexico
Morocco
Pakistan
Peru
Philippines
Singapore
South Africa
Sri Lanka
Thailand
Trinidad and Tobago
Tunisia
Turkey
Uruguay
Venezuela
Zimbabwe

Belize
Bolivia
Burundi
Cameroon
Costa Rica
Cyprus
Dominican Republic
El Salvador
Equatorial Guinea
Ethiopia
Fiji
Grenada
Guatemala
Guinea-Bissau
Guyana
Haiti
Honduras
Hungary
Jamaica
Lao P.D. Rep.
Madagascar
Malawi
Mali
Malta
Mozambique
Myanmar
Nepal
Nicaragua
Nigeria
Panama
Paraguay
Romania
Sierra Leone
Swaziland
Syrian Arab Rep.
Uganda
Zambia

The "All Country" sample includes "Industrial Countries", "Emerging Markets", and "Other Developing Countries"; the
"Developing Country" sample includes "Emerging markets" and "Other Developing".

Appendix B
Occurrences of Banking and Currency Crises

United Kingdom

Banking Crisis

Currency Crisis

Financial Liberalization

1975-1976, 1984

1976, 1979, 1981-1982, 1986, 1992

1974

Austria

1975

Belgium

1982

Denmark

1987-1992

France

1994-1995

Germany

1978-1979

Italy

1990-1995

1986
1981

1982

1975
1975

1976, 1992, 1995

1975

Luxembourg

Netherlands

1975

Norway

1987-1993

1978, 1986, 1992

1984

Sweden

1990-1993

1977, 1981-1982, 1992-1993

1980

1978

1989

Switzerland
Canada

1983-1985

1976, 1992

1975

Japan

1992-1997

1979, 1989-1990

1985

Finland

1991-1994

1977-1978, 1982, 1991-1993

1986

Greece

1991-1995

1980, 1982-1983, 1985

1975

Iceland

1985-1986, 1993

1983-1984, 1988, 1992-1993

Ireland

1985

Malta

1992, 1997

Portugal

1986-1989

1976-1978, 1982-1983, 1993, 1995

1984

Spain

1977-1985

1976-1977, 1982, 1992-1993

1974

Turkey

1982-1985, 1991, 19941995

1978-1980, 1994

1980-1982, 1984

New Zealand

1987-1990

1975, 1983-1988, 1991

1980, 1984

South Africa

1977, 1985, 1989

1975, 1978, 1984-1986, 1996

Argentina

1980-1982, 1989-1990,
1995-1997

1975-1976, 1982-1983, 1989-1991

1977

Bolivia

1986-1987, 1994-1997

1981-1985, 1988, 1990-1991

1985

Brazil

1990, 1994-1997

1982-1983, 1987, 1990-1991, 1995

1975

Chile

1976, 1981-1983

1985

1975

Columbia

1982-1987

1985

1980

Costa Rica

1987, 1994-1997

1981

1985, 1987, 1990

1982-1983, 1985-1986, 1988

1986-1987, 1992

Dominican Republic
Ecuador

1980-1982, 1996-1997

Banking Crisis

Currency Crisis

Financial Liberalization

El Salvador

1989

1986, 1990

1991

Guatemala

1991-1992

1986, 1989-1990

1989

Haiti

1977, 1991

Honduras

1990

Mexico

1981-1991, 1995-1997

1976, 1982, 1985, 1994-1995

1989

Nicaragua

1988-1996

1993

Panama

1988-1989

Paraguay

1995-1997

1984-1986, 1988-1989, 1992

1990

Peru

1983-1990

1976, 1979, 1978-1988,

1980-1984, 1990

Uruguay

1981-1984

1982-1983

1976

Venezuela

1978-86, 1994-1997

1984, 1986, 1994-1996

1981-1983, 1989

1978

1978, 1989-1991

1991

Grenada
Guyana

1993-1995

Belize

Jamaica

1994-1997

1978, 1983-1984, 1990-1992

1991

Trinidad & Tobago

1982-1993

1985, 1988, 1993

Cyprus
Jordan

NA
1989-1990

Syrian Arab Republic

1983, 1987-1989, 1992

1988

1977, 1982, 1988

No Liberalization

Egypt

1980-1985, 1991-1995

1979, 1989-1991

1991

Bangladesh

1987-1996

1975-1976

Myanmar

1996-1997

1975-1977

Sri Lanka

1989-1993

1977

1980

China, P.R.: Hong Kong

1982-1986

India

1993-1997

1976, 1991, 1993, 1995

1991

Indonesia

1994, 1997

1978, 1983, 1986, 1997

1983

Korea

1997

1980, 1997

1984

Lao People’s D. R.

1991-1994, 1997

1995

Malaysia

1985-1988, 1997

1986, 1997

1978

Nepal

1988-1994

1975, 1981-1982, 1984-1986, 1991, 1993,
1995

Pakistan

Philippines

1981-1987, 1997

1983-1984, 1986, 1997

1981

Singapore

1982

1975

Thailand

1983-1987, 1997

1981, 1984, 1997

1989

Botswana

1994-1995

1984-1986, 1996

Burundi

1994-1997

1976, 1983,1986, 1988-1989, 1991, 1997

Banking Crisis

Currency Crisis

Financial Liberalization

Cameroon

1987-1993, 1995-1997

1982, 1984, 1994

Equatorial Guinea

1983-1985

1991, 1994

Ethiopia

1994-1995

1992

Ghana

1982-1989, 1997

1978, 1983, 1986-1987

Guinea-Bissau

1995-1997

1991, 1996

Kenya

1985-1989, 1992-1997

1975, 1981-1982, 1985, 1993-1995, 1997

1991

Madagascar

1988

1984, 1986-1987, 1991, 1994, 1996

1982, 1985-1987, 1992, 1994

Malawi
Mali

1987-1989

1993

No Liberalization

Mauritius

1996

1979, 1981

1983-1985, 1990

Morocco
Mozambique

1987-1997

1993, 1995

Nigeria

1993-1997

1986-1987, 1989, 1992

1990-1993

Zimbabwe

1995-1997

1982, 1991, 1993-1994, 1997

Sierra Leone

1990-1997

1988-1990, 1997

Swaziland

1995

1975, 1979, 1982, 1984-1986

Tunisia

1991-1995

1993

Uganda

1994-1997

1981, 1987-1989

1991

Zambia

1995

1985, 1987, 1994

1992

1986-1987

Fiji
Hungary

1991-1995

1989, 1994-1995

Romania

1990-1997

1990-1991

Full text of Working Papers (Federal Reserve Bank of San Francisco) : Banking and Currency Crises : How Common Are Twins?, Working Paper No. PB99-07

FRASER