Professional Documents
Culture Documents
Reinhart, Carmen M. - Goldstein, Morris Arthur - Forecasting Financial Crises - Early Warning Signals For Emerging Markets
Reinhart, Carmen M. - Goldstein, Morris Arthur - Forecasting Financial Crises - Early Warning Signals For Emerging Markets
FINANCIAL
VULNERABILITY
An Early Warning System
for Emerging Markets
ASSESSING
FINANCIAL
VULNERABILITY
An Early Warning System
for Emerging Markets
Morris Goldstein
Graciela L. Kaminsky
Carmen M. Reinhart
Preface ix
Acknowledgments xv
1 Introduction 1
Organization of the book 9
2 Methodology 11
General Guidelines 11
Putting the Signals Approach to Work 18
3 Empirical Results 33
The Monthly Indicators: Robustness Check 33
The Annual Indicators: What Works? 38
Do the Indicators Flash Early Enough? 40
Microeconomic Indicators: Selective Evidence 42
6 Contagion 73
Defining Contagion 74
Theories of Contagion and Their Implications 75
Empirical Studies 76
References 115
Index 121
Tables
Table 1.1 Emerging Asia: real GDP growth forecasts, 1996-98 3
Table 1.2 Rating agencies’ performance before the Asian crisis:
Moody’s and Standard & Poor’s long-term debt ratings,
1996-97 4
Table 2.1 Currency crisis starting dates 22
Table 2.2 Banking crisis starting dates 24
Table 2.3 Selected leading indicators of banking and currency crises 26
Table 2.4 Optimal thresholds 29
Table 2.5 Examples of country-specific thresholds: currency crises 30
Table 3.1 Ranking the monthly indicators: banking crises 34
Table 3.2 Ranking the monthly indicators: currency crises 35
Table 3.3 Annual indicators: banking crises 38
Table 3.4 Annual indicators: currency crises 39
Table 3.5 Short-term debt: selected countries, June 1997 40
Table 3.6 How leading are the signals? 41
Table 3.7 Microeconomic indicators: banking crises 42
Table 4.1 Comparison of Institutional Investor sovereign ratings with
indicators of economic fundamentals 46
Table 4.2 Do ratings predict banking crises? 48
Table 4.3 Do ratings predict currency crises? 48
Table 4.4 Do ratings predict banking crises for emerging markets? 50
Table 4.5 Do ratings predict currency crises for emerging markets? 50
Table 4.6 Rating agencies’ actions on the eve and aftermath of the Asian
crisis, June-December 1997 51
vi
Figures
Figure 2.1 Mexico: real exchange rate, 1970-96 31
Figure 5.1 Probability of currency crises for four Southeast Asian
countries, 1990-97 69
vii
The last decade has been disrupted by a series of currency, banking, and
debt crises. Hence the research agenda of the Institute for International
Economics over this period has given high priority to the analysis of,
and policy prescriptions for, crisis prevention and management. Wendy
Dobson, in Economic Policy Coordination (1991), and C. Fred Bergsten and
C. Randall Henning, in Global Economic Leadership and the Group of Seven
(1996), explored ways to improve G-7 policy coordination. Morris Gold-
stein, in The Exchange Rate System and the IMF (1995), examined the
appropriate design of G-3 currency arrangements and the role of the
International Monetary Fund. John Williamson, in What Role for Currency
Boards? (1995), analyzed the pros and cons of currency boards. Guillermo
Calvo, et al., in Private Capital Flows to Emerging Economies after the Mexican
Crisis (1996), drew lessons from the Mexican peso crisis. Banking crises
in emerging economies—and how to reduce their incidence and severity
by adopting a voluntary international banking standard—were the objects
of Morris Goldstein’s 1997 study, The Case for an International Banking
Standard, which led to subsequent official adoption of the Basel Core
Principles of Effective Banking Supervision.
The Asian financial crisis provided a powerful additional stimulus to the
Institute’s work on crisis prevention and management. Morris Goldstein
outlined the origins of the Asian crisis as well as the key policy responses
needed to overcome it in The Asian Financial Crisis: Causes, Cures and
Systemic Implications (1998). Marc Noland, et al., in Global Economic Effects
of the Asian Currency Devaluations (1998), used a general equilibrium model
to assess the implications for the global economy of devaluations in the
ix
C. Fred Bergsten
Director
June 2000
xi
The authors wish to thank Richard Cantor for sharing his database on
credit ratings and Andrew Berg, Michael Bordo, Stijn Claessens, Richard
Cooper, Susan Collins, Hali Edison, Barry Eichengreen, Cathy Mann, Peter
Montiel, Adam Posen, Vincent R. Reinhart, Ewoud Schuitemaker, Harry
Stordel, Holger Wolf, and participants at the Institute for International
Economics seminar for useful comments and suggestions. We are particu-
larly indebted to C. Fred Bergsten, not only for his extensive suggestions
on earlier drafts, but also for his unwavering confidence in the project.
Ian Anderson, Trond Augdal, Mark Giancola, and Neil Luna provided
superb research assistance.
xv
This study analyzes and provides empirical tests of early warning indica-
tors of banking and currency crises in emerging economies. The aim is
to identify key empirical regularities in the run-up to banking and cur-
rency crises that would enable officials and private market participants
to recognize vulnerability to financial crises at an earlier stage. This, in
turn, should make it easier to motivate the corrective policy actions that
would prevent such crises from actually taking place. Interest in identify-
ing early warning indicators of financial crises has soared of late, stoked
primarily by two factors: the high cost to countries in the throes of crisis
and an increasing awareness of the insufficiency of the most closely
watched market indicators.
There is increasing recognition that banking and currency crises can
be extremely costly to the countries in which they originate. In addition,
these crises often spill over via a variety of channels to increase the
vulnerability of other countries to financial crisis.
According to one recent study, there have been more than 65 develop-
ing-country episodes during 1980-95, when the banking system’s capital
was completely or nearly exhausted;1 the public-sector bailout costs of
resolving banking crises in developing countries during this period have
1. See Caprio and Klingebiel (1996b). Other identifications of banking crises over this period
can be found in Demirgüç-Kunt and Detragiache (1998), Eichengreen and Rose (1998), IMF
(1998c), Kaminsky and Reinhart (1999), and Lindgren et al. (1996).
2. This figure is net of the estimated amount of loans that were eventually repaid. See
Honohan (1997).
3. See Goldstein (1997) for a list of these severe banking crises. For comparison, the public-
sector tab for the US saving and loan crisis is typically estimated at about 2 to 3 percent of
US GDP.
4. In chapter 7, we present our own estimates of how long it takes growth rates of real
output to recover after banking and/or currency crises.
5. In chapter 3, we provide further evidence that the presence of a banking crisis is one of
the better leading indicators of a currency crisis in emerging economies. At the same time,
the evidence also suggests that a currency crash aggravates the problems in the banking
sector, as the peak of a banking crisis most often occurs following the collapse of the
currency. The dating of currency and banking crises is discussed in detail in chapter 2.
6. See Calvo and Reinhart (1996) and Goldstein (1998a). Kaminsky and Reinhart (2000)
provide an analysis of contagion in the Asian crisis that stresses the financial links among
these countries—including the sudden withdrawal of funds by a common commercial bank
lender or mutual fund investor. See also chapter 6.
INTRODUCTION 3
Table 1.2 Rating agencies’ performance before the Asian crisis: Moody’s and Standard & Poor’s long-term debt
ratings,a 1996-97
15 January 1996 2 December 1996 24 June 1997 12 December 1997
Rating Outlook Rating Outlook Rating Outlook Rating Outlook
Moody’s foreign currency debt
Indonesia Baa3 Baa3 Baa3 Baa3
Malaysia A1 A1 A1 A1
|
5
The second reason for the increased interest in early warning indicators
of financial crises is that there is accumulating evidence that two of the
most closely watched ‘‘market indicators’’ of default and currency risks—
namely, interest rate spreads and changes in credit ratings—frequently
do not provide much advance warning of currency and banking crises
(see chapter 4).
Empirical studies of the 1992-93 ERM crisis have typically concluded
that market measures of currency risk did not raise the specter of signifi-
cant devaluations of the weaker ERM currencies before the fact (Rose and
Svensson 1994). Another study, encompassing a larger number of crisis
episodes, similarly concluded that the currency forecasts culled from
survey data are useless in anticipating the crises (Goldfajn and Valdés
1998). In the run-up to the Mexican crisis, market signals were again
muted or inconsistent. More specifically, measures of default risk on
tesobonos (dollar indexed, Mexican government securities) jumped up
sharply in April 1994 (after the Colosio assassination) but stayed roughly
constant between then and the outbreak of the crisis (Leiderman and
Thorne 1996; Obstfeld and Rogoff 1995). From April 1994 on, market
expectations of currency depreciation on the peso usually were beyond
the government’s announced rate; nevertheless, this measure of currency
risk fluctuated markedly. The gap between market expectations and the
official rate was widest in summer of 1994, but the attack came with most
ferocity only in late December (Obstfeld and Rogoff 1995; Leiderman and
Thorne 1996; Rosenberg 1998).
The evidence now available suggests that the performance of interest
spreads and credit ratings was likewise disappointing in the run-up to
the Asian financial crisis. Examining interest rate spreads on three-month
offshore securities, one study found that these spreads gave no warning
of impending difficulties (i.e., were either flat or declining) for Indonesia,
Malaysia, and the Philippines and produced only intermittent signals for
Thailand (Eschweiler 1997b). A recent analysis of spreads using local
interest rates for South Korea, Thailand, and Malaysia similarly found
little indication of growing crisis vulnerability (Rosenberg 1998).
Sovereign credit ratings (on long-term, foreign-currency debt) issued
by the two largest international ratings firms were even less prescient in
the Asian crisis (see chapter 4, as well as Radelet and Sachs 1998; Goldstein
1998c).7 As shown in table 1.2, there were almost no downgrades for the
7. In a recent report, Moody’s (1998) argues that its rating record in the East Asian crisis
was better than it appears at first sight from ratings changes alone. More specifically, the
report argues, inter alia, that Moody’s went into the crisis with lower ratings for the crisis
countries than the other major ratings agencies (i.e., Standard & Poor’s and Fitch-IBCA),
that it took ratings actions before its main competitors, that its low bank financial strength
ratings identified many of the banks that subsequently experienced stress in the crisis
countries, that changes in sovereign credit ratings led to a widening of yield spreads in the
crisis countries, and that one should examine the sovereign research reports—not just the
ratings—in looking for early warning signals. At the same time, the report acknowledges
that the firm is studying several potential enhancements to its analytical methodology to
help improve the predictive power of its sovereign ratings.
8. It is sometimes also argued that even when credit rating agencies or international financial
organizations (such as the IMF) conclude that crisis vulnerability is high, they will be
reluctant to go ‘‘public’’ with a downgrade or a warning for fear of being accused of
precipitating the crisis.
INTRODUCTION 7
9. Michael P. Dooley has stressed this point in several papers (see Dooley 1997, for instance).
10. See Krugman (1998), Dooley (1997), and Calomiris (1997) on the role of expected national
and international bailouts in motivating capital flows and/or banking crises. Zhang (1999),
on the other hand, tests for such ‘‘moral hazard’’ effects in private capital flows to emerging
markets and finds no evidence for it. Claessens and Glaessner (1997) highlight the link
between fiscal positions and the wherewithal to honor explicit and implicit guarantees in
the financial sector. The Council on Foreign Relations (1999) offers a set of proposals on how
the moral hazard associated with international financial rescue packages might be reduced.
INTRODUCTION 9
General Guidelines
First, finding a systematic pattern in the origin of financial crises means
looking beyond the last prominent crisis (or group of crises) to a larger
sample. Otherwise there is a risk either that there will be too many
potential explanations to discriminate between important and less impor-
tant factors or that generalizations and lessons will be drawn that do not
necessarily apply across a wider body of experience.1 We try to guard
against these risks by looking at a sample of 87 currency crises and 29
banking crises that occurred in a sample of 25 emerging economies and
smaller industrial countries over 1970-95.2
Several examples help to illustrate the point. Consider the last two
major financial crises of the 1990s: the 1994-95 Mexican peso crisis and
1. One can also view ‘‘early warning indicators’’ as a way to discipline or check more
‘‘subjective’’ and ‘‘idiosyncratic’’ assessments of crisis probabilities for particular econo-
mies—just as more comprehensive, subjective assessments can act as a check on the quality
of early warning indicator projections.
2. Our out-of-sample analysis spans 1996-97. Our criteria for defining a currency and a
banking crisis is described later in this chapter.
11
3. See Leiderman and Thorne (1996) and Calvo and Goldstein (1996) for an analysis of the
Mexican crisis.
4. These alternative explanations of the Asian crisis are discussed in BIS (1998), Corsetti,
Pesenti, and Roubini (1998), Goldstein (1998a), Radelet and Sachs (1998), IMF (1997), and
World Bank (1998).
5. Some of these explanations, of course, are not mutually exclusive. For example, large
current account deficits may be the outcome of financial liberalization and its attendant
credit booms.
6. See Kaminsky, Lizondo, and Reinhart (1998) for a review of this literature. Among the
relatively few studies that include or concentrate on banking crises in emerging economies,
we would highlight Caprio and Klingebiel (1996a, 1996b), Demirgüç-Kunt and Detragiache
(1998), Eichengreen and Rose (1998), Furnam and Stiglitz (1998), Honohan (1997), Gavin
and Hausman (1996), Goldstein (1997), Goldstein and Turner (1996), Kaminsky (1998),
Kaminsky and Reinhart (1998, 2000), Rojas-Suarez (1998), Rojas-Suarez and Weisbrod (1995),
and Sundararajan and Baliño (1991).
7. Both Kaminsky and Reinhart (1998) and the IMF (1998c) conclude that the output costs
of banking crises in emerging economies typically exceed those for currency crises and that
these costs are greater still during what Kaminsky and Reinhart (1999) dubbed ‘‘twin crises’’
(that is, episodes when the country is undergoing simultaneous banking and currency
crises). We provide further empirical evidence on this issue in chapter 7.
8. See Calvo and Reinhart (2000) and Reinhart (2000) for a fuller discussion of this issue.
METHODOLOGY 13
9. For example, the studies of banking crises in emerging markets by Caprio and Klingebiel
(1996a, 1996b), Goldstein and Turner (1996), Honohan (1995), and Sundararajan and Baliño
(1991) are primarily qualitative, while the studies by Demirgüç-Kunt and Detragiache (1997),
Eichengreen and Rose (1998), and the IMF (1998c) use annual data for their quantitative
investigation of the determinants of banking crises.
10. Private-sector ‘‘early warning’’ analyses likewise seem to be moving in the direction of
using monthly data. See Ades, Masih, and Tenegauzer (1998) and Kumar, Perraudin, and
Zinni (1998).
11. First-generation models stress poor fundamentals as the cause of the currency crises,
while second-generation models focus on shifts in market expectations and self-fulfilling
speculative attacks. See Flood and Marion (1999) for a recent survey of this literature.
12. This approach is described in detail in Kaminsky, Lizondo, and Reinhart (1998).
METHODOLOGY 15
13. While this is one of many potential ‘‘composite’’ indicators (i.e., ways of combining the
information in the individual indicators), Kaminsky (1998) provides evidence that this
weighting scheme shows better in-sample and out-of-sample performance than three alterna-
tives. Also, see chapter 5. One can equivalently evaluate the performance of individual
indicators by comparing their conditional probabilities of signaling a crisis.
14. Of course, ease of application is only one of many criteria for choosing among competing
crisis-forecasting methodologies. For example, the signals approach also carries the disad-
vantage that is less amenable to statistical tests of significance. In addition, some of the
restrictions it imposes (e.g., that indicators send a signal only when they reach a threshold)
may leave out valuable information.
15. Indeed, for many countries, detailed data on the state of the banks may not even be
available annually.
METHODOLOGY 17
The signals approach described above was first used to analyze the
performance of macroeconomic and financial indicators around ‘‘twin
crises’’ (i.e., the joint occurrences of currency and banking crises) in Kamin-
sky and Reinhart (1999). We focus on a sample of 25 countries over 1970
to 1995. The out-of-sample performance of the signals approach will be
assessed using data for January 1996 through December 1997. These are
the countries in our sample:
The basic premise of the signals approach is that the economy behaves
differently on the eve of financial crises and that this aberrant behavior
has a recurrent systematic pattern. This ‘‘anomalous’’ pattern, in turn, is
manifested in the evolution of a broad array of economic and financial
indicators. The empirical evidence provides ample support for this prem-
Currency Crisis
16. See Kaminsky, Lizondo, and Reinhart (1998) for a survey of this literature.
17. This index is in the spirit of that used by Eichengreen, Rose, and Wyplosz (1996), who
also included interest rate increases in their measure of turbulence.
18. Of course, for a study of market turbulence as well as crisis, one may wish to consider
readings in this index that are two standard deviations away from the mean.
METHODOLOGY 19
Banking Crises
Our dating of banking crises stresses events. This is because on the banking
side there are no time series comparable to international reserves and the
exchange rate. For instance, in the banking panics of an earlier era large
withdrawals of bank deposits could be used to date the crisis. In the wake
of deposit insurance, however, bank deposits ceased to be useful for
dating banking crises. As Japan’s banking crisis highlights, many modern
financial crises stem from the asset side of the balance sheet, not from
deposit withdrawals. Hence the performance of bank stocks relative to the
overall equity market could be an indicator. Yet in many of the developing
countries an important share of the banks are not traded publicly. Large
increases in bankruptcies or nonperforming loans could also be used to
mark the onset of the crisis. Indicators of business failures and nonper-
forming loans are, however, usually available only at low frequencies, if
at all; the latter are also made less informative by banks’ desire to hide
their problems for as long as possible.
Given these data limitations, we mark the beginning of a banking crisis
by two types of events: bank runs that lead to the closure, merging, or
takeover by the public sector of one or more financial institutions (as in
Venezuela in 1993); and if there are no runs, the closure, merging, takeover,
or large-scale government assistance of an important financial institution
(or group of institutions) that marks the start of a string of similar out-
comes for other financial institutions (as in Thailand in 1997). We rely on
existing studies of banking crises and on the financial press; according
to these studies the fragility of the banking sector was widespread during
these periods.
Our approach to dating the onset of the banking crises is not without
drawbacks. It could date the crises ‘‘too late’’ because the financial prob-
lems usually begin well before a bank is finally closed or merged. It could
also date crises ‘‘too early’’ because the worst of crisis may come later.
19. Similar results are obtained by looking at significant departures in inflation from a 6-
and 12-month moving average.
The Indicators
20. The real exchange rate is defined on a bilateral basis with respect to the German mark
for the European countries in the sample and with respect to the US dollar for all other
countries. The real exchange rate index is defined such that an increase in the index denotes
a real depreciation.
METHODOLOGY 21
METHODOLOGY 23
21. This definition of the spread between lending and deposit rates is preferable to using
merely the difference between nominal lending and deposit rates because inflation affects
this difference and thus the measure would be distorted in the periods of high inflation.
An alternative would have been to use the difference between real lending and deposit rates.
METHODOLOGY 25
22. Since there are two readings of this index per year, in a typical year, say 1995, we would
have the percentage change in the rating from September 1994 to March 1995, from March
1995 to September 1995, and the change from September 1995 to March 1996.
23. Of course, normal behavior may change over time, hence, this approach, like other
commonly used alternatives (such as logit or probit) is not free from Lucas-critique limita-
tions. For further discussion of this issue, see Kaminsky and Reinhart (1999).
METHODOLOGY 27
Suppose we wish to test the null or maintained hypothesis that the econ-
omy is in a ‘‘state of tranquility’’ versus the alternative hypothesis that
a crisis will occur sometime in the next 24 months. Suppose that we
wish to test this hypothesis on an indicator-by-indicator basis. As in any
hypothesis test, this calls for selecting a threshold or critical value that
divides the probability distribution of that indicator into a region that is
considered normal or probable under the null hypothesis and a region
that is considered aberrant or unlikely under the null hypothesis—the
rejection region. If the observed outcome for a particular variable falls
into the rejection region, that variable is said to be sending a signal.
To select the optimal threshold for each indicator, we allowed the size
of the rejection region to oscillate between 1 percent and 20 percent. For
each choice, the noise-to-signal ratio was tabulated and the ‘‘optimal’’ set
of thresholds was defined as the one that minimized the noise-to-signal
ratio—that is, the ratio of false signals to good signals.24
Table 2.4 lists the thresholds for all the indicators for both currency
and banking crises. For instance, the threshold for short-term capital
flows as a percentage of GDP is 85 percent. This conveys two kinds of
information. First, it indicates that 15 percent of all the observations in
our sample (for this variable) are considered signals. Second, it highlights
that the rejection region is located at the upper tail of the frequency
distribution, meaning that a high ratio of short-term capital inflows to
GDP will lead to a rejection of the null hypothesis of tranquility in favor
of the alternative hypothesis that a crisis is brewing.
While the threshold or percentile that defines the size of the rejection
region is uniform across countries for each indicator, the corresponding
country-specific values are allowed to differ. Consider the following illus-
tration. There are two countries, one which has received little or no short-
term capital inflow (as a percentage of GDP) during the entire sample,
while the second received substantially larger amounts (also as a share
of GDP). The 85th percentile of the frequency distribution for the low
capital importer may be as small as a half a percent of GDP and any
increase beyond that would be considered a signal. Meanwhile, the coun-
try where the norm was a higher volume of capital inflows is likely to
have a higher critical value; hence only values above, say 3 percent of
GDP, would be considered signals.
24. For variables such as international reserves, exports, the terms of trade, deviations of
the real exchange rate from trend, commercial bank deposits, output, and the stock market
index, for which a decline in the indicator increases the probability of a crisis, the threshold
is below the mean of the indicator. For the other variables, the threshold is above the mean
of the indicator.
25. Indeed, as shown in Kaminsky and Reinhart (1998), the volatility pattern for these
three countries is representative of the broader historical regional pattern. The wild gyrations
in financial markets in Asia in 1997-99, however, may be unraveling those historic patterns.
METHODOLOGY 29
A perfect indicator would only have entries in cells A and D. Hence, with
this matrix we can define several useful concepts that we will use to
evaluate the performance of each indicator.
If one lacked any information on the performance of the indicators, it
is still possible to calculate, for a given sample, the unconditional probability
of crisis,
If an indicator sends a signal and that indicator has a reliable track record,
then it can be expected that the probability of a crisis, conditional on a signal,
P(C/S), is greater than the unconditional probability. Where
Formally,
31
We can also define the noise-to-signal ratio, N/S, as
It may be the case that an indicator has relatively few false alarms in its
track record. This could be the result of the indicator issuing signals
relatively rarely. In this case, there is also the danger that the indicator
misses the crisis altogether (it does not signal and there is a crisis). In this
case, we also wish to calculate for each indicator the proportion of crises
accurately called,
The signals approach was applied to the indicators around the dates of
the 29 banking and the 87 currency crises. In what follows, we first
compare our results for the 15 monthly indicators to those presented in
Kaminsky and Reinhart (1999) and reproduced in table 3.1. In addition
to presenting our in-sample findings, this exercise allows us to gauge
robustness of the signals approach, since the results reported here are
derived from a larger sample of countries (25 versus 20.)1 Moreover, in
this chapter we report results for many of the indicators that have been
stressed in the financial press surrounding the coverage of the Asian crisis.
Tables 3.1 and 3.2 summarize the in-sample performance of the monthly
indicators along the lines described in chapter 2 and presented in Kamin-
sky, Lizondo, and Reinhart (1998) and Kaminsky (1998). Table 3.1 covers
banking crises, and table 3.2 presents the results for currency crises. The
variables are shown in descending order based on their marginal predict-
ive power. For banking crises, for instance, the real exchange rate has the
greatest predictive power and imports the least. For each indicator, the
first column of the tables shows the noise-to-signal ratio. An indicator
with a noise-to-signal ratio of unity, such as those in the bottom of the
1. The five countries included here that were not a part of the Kaminsky and Reinhart
(1999) sample are the Czech Republic, Egypt, Greece, South Africa, and South Korea.
33
tables, issues as many false alarms as good signals. The second column
shows the percent of crises (for which there were data for that indica-
tor) accurately called, while the third column lists the probability of a
crisis conditional on a signal from the indicator, P(C 兩 S). The fourth column
shows the difference between the conditional and unconditional probabili-
ties, P(C 兩 S) ⳮ P(C), the fifth column shows the ranking that the indicator
received in the previous signals approach analysis, and the last column
calculates the difference between its current and previous rank. Hence,
a Ⳮ3 in the last column would mean that the indicator moved up three
notches as the sample was enlarged, while a ⳮ2 would reflect a decline
in its ranking.
The indicators’ rankings based on their marginal predictive power are
shown under the heading P(C 兩 S) ⳮ P(C). The better the indicator, the
higher the probability of crisis conditioned on its signaling—that is, the
higher the P(C 兩 S)—and the bigger the gap between the conditional proba-
bility (P(C 兩 S) and the unconditional probability P(C). The unconditional
probability of a banking crisis (not shown) varies slightly from indicator to
2. As shown in Kaminsky, Lizondo, and Reinhart (1998), the bigger the gap between the
conditional probability (P(C兩S) and the unconditional probability P(C), the lower the noise-
to-signal ratio.
EMPIRICAL RESULTS 35
3. We did not included the larger industrial countries (particularly the G-7 countries) in
our sample because they have characteristics (such as the ability to borrow in their own
currency, a relatively good external-debt servicing history, and high access to private capital
markets) that on a priori grounds would seem to make their crisis vulnerability different
from that of most emerging economies. In addition, data constraints, extremely large struc-
tural shifts over time, and difficulties associated with identifying a ‘‘normal’’ period led to
the decision to exclude China, Russia, and most of the transitional economies from the
sample. Finally, we excluded low-income developing countries from the sample because
we wanted to concentrate on emerging economies that had (in addition to the requisite
data availability) significant involvement with private international capital markets. In the
end, however, one can only tell whether our sample selection results in certain biases by
doing further robustness checks on alternative samples of countries.
4. For instance, lending-deposit interest rate spreads could widen in advance of a crisis
due to a deterioration in loan quality or a worsening in adverse selection problems. Alterna-
tively, it could be persuasively argued that ahead of financial crises, banks may be forced
to offer higher deposit rates, so as to stem capital flight.
5. However, some recent models (Goldfajn and Valdés 1995) have highlighted the role of
bank runs in precipitating currency crises.
EMPIRICAL RESULTS 37
EMPIRICAL RESULTS 39
the currency crises (as opposed to the banking crises) took place in the
1970s in an environment of highly regulated internal and external financial
markets, where portfolio flows were negligible.
While our list of indicators is comprehensive, it is by no means exhaus-
tive. The Asian crisis in particular highlighted the importance of currency
and maturity mismatches in increasing vulnerability to currency and
banking crises. Table 3.5 presents an indicator of the imbalance between
liquid liabilities and liquid assets: namely, the ratio of short-term debt to
international reserves. All the emerging economies in this group with
debt-to-reserves levels in excess of 100 percent in mid-1997 have been
casualties of financial turmoil in recent years (even if not all the speculative
attacks ultimately succeeded, as in the case of Argentina.) This suggests
that variables such as short-term debt to reserves could be a valuable
addition to our list of leading indicators of crisis vulnerability.6
6. See Calvo and Mendoza (1996) for an early discussion of this issue. We did not use the
ratio of short-term debt to reserves as an indicator in our tests because its relevance was
highlighted mainly by the Asian crisis and we did not want the out-of-sample tests to be
biased by its inclusion. In addition, the data were not available for the early part of our sample.
EMPIRICAL RESULTS 41
longest lead time—namely, 19 months. The average lead time for these
early signals is 15 months for currency crises. All the indicators considered
are therefore best regarded as leading rather than coincident, which is
consistent with the spirit of an ‘‘early warning system.’’ For banking
crises, there is a greater dispersion in the lead time across indicators, and
the average lead time is also lower (about 11 months). Furthermore, most
of the indicators signal at about the same time, thus the signaling is
cumulative and all the more compelling. Thus, on the basis of these
preliminary results, there does appear to be adequate lead time for pre-
emptive policy actions to avert crises.
EMPIRICAL RESULTS 43
45
1. The 20 countries are those in the Kaminsky and Reinhart (1999) sample: Argentina,
Bolivia, Brazil, Chile, Colombia, Denmark, Finland, Indonesia, Israel, Malaysia, Mexico,
Norway, Peru, the Philippines, Spain, Sweden, Thailand, Turkey, Uruguay, and Venezuela.
2. An unbalanced panel, in this case, refers to the fact that we do not have the same number
of observations for all the countries.
3. These results are not reported here but are available from the authors.
4. The IMF’s World Economic Outlook of April 2000 notes that sovereign risk and devaluation
tend to move together in the case of emerging economies.
5. We do not place much weight on the Moody’s results, as the number of banking crises
is very small. In future work, it would be interesting to test whether Moody’s bank financial
strength ratings (which are meant to capture the health of banks independent of the likelihood
of a government bailout) are better predictors of banking crises. However, as these ratings
were only introduced in 1995, the time series is not yet long enough to encompass many
banking crises.
6. We want to examine whether the rating changes follow immediately after the crisis, but
as the index is only published twice a year this ability to discriminate is not possible.
55
Mexico 11 2 0 2 2 0 17
http://www.iie.com
Norway 9 3 0 1 1 0 17
Peru 16 2 0 1 1 0 13
The Philippines 59 8 1 2 2 0 43
South Africa 42 8 3 1 1 0 39
South Korea 32 8 3 3 3 0 48
Spain 44 6 2 1 1 0 30
Sweden 55 5 1 1 1 0 26
Thailand 50 6 3 1 1 1 30
Turkey 22 4 1 3 3 0 30
Uruguay 58 5 0 1 1 0 26
Venezuela 18 5 2 1 1 0 26
57
from the Czech Republic, which indeed floated following a speculative
attack and substantial reserve losses in May 1997.
Table 5.2 repeats the same accounting exercise, but here we include
‘‘borderline’’ signals. Specifically, we enlarged the size of the rejection
region by 5 percent for all the indicators. For instance, instead of having
a 10 percent threshold for stock prices, we now have a 15 percent threshold.
This sensitivity analysis increases the likelihood of making a Type II error
(rejecting the null hypothesis of tranquility when you should not) while
reducing the probability of a Type I error (not rejecting when you should).
Including borderline signals does not seem to generate large shifts in the
most and least vulnerable groups. As shown in the last column in table
5.2, borderline signals do not alter the picture at all for some countries
(such as Argentina), but they do markedly increase the proportion of
indicators signaling, as well as the number of signals, for countries such
as South Korea (from 48 to 65 percent) and South Africa (from 39 to
52 percent).
Tables 5.3 and 5.4 report the results for banking crises using the original
thresholds and the ‘‘borderline’’ scenario, respectively. The country pro-
files that emerge are similar to those for currency crises; this may reflect
the fact that several of the indicators have common thresholds for currency
and banking crises.
While conveying useful information on vulnerability, the preceding
analysis does not fully discriminate between the more and less reliable
indicators. Kaminsky (1998) shows how to construct a ‘‘composite index’’
to gauge the probability of a crisis conditioned on multiple signals from
various indicators; the more reliable indicators receive a higher weight
in this composite index. This methodology and its out-of-sample results
are described in the remainder of this chapter.
In weighting individual indicators, a good argument can be made for
eliminating from our list of potential leading indicators those variables
that had a noise-to-signal ratio above unity; this is tantamount to stating
that their marginal forecasting ability, P(C 兩 S) ⳮ P(C), is zero or less.
Applying this criterion to banking crises, the lending-deposit ratio, the
terms of trade, government consumption growth, and FDI as a share of
GDP should be dropped. For currency crises, the excluded indicators are
the domestic-foreign interest rate differential, the lending-deposit ratio,
bank deposits, central bank credit to the public sector, and FDI as a share
of GDP. For the remaining indicators with noise-to-signal ratios below
unity, we weighed the signals by the inverse of the noise-to-signal ratios
reported in tables 3.1 through 3.4. For a currency crisis, suppose that both
the real exchange rate and imports are issuing a signal. Because the real
exchange rate has a very low noise-to-signal ratio (0.22), it would receive
a weight of 4.55 (i.e., 1/0.22); in contrast, with a relatively high noise-to-
signal ratio (0.87), imports would receive a weight of only 1.49 (i.e., 1/0.87).
Mexico 24 2 0 2 2 0 17
http://www.iie.com
Norway 31 7 2 1 1 0 35
Peru 26 5 1 2 2 0 30
The Philippines 68 8 3 3 3 0 48
South Africa 63 10 3 2 2 0 52
South Korea 63 11 3 5 4 1 65
Spain 55 7 2 1 1 0 35
Sweden 60 6 1 1 1 0 30
Thailand 54 6 3 1 1 1 30
Turkey 33 5 3 3 3 0 35
Uruguay 71 5 1 1 1 0 26
Venezuela 29 5 2 1 1 0 26
59
60
Mexico 16 4 1 2 2 0 26
http://www.iie.com
Norway 30 8 2 1 1 0 39
Peru 19 5 1 1 1 0 26
The Philippines 59 8 3 2 2 0 43
South Africa 55 10 3 1 1 0 43
South Korea 42 10 4 3 3 1 57
Spain 51 7 1 1 1 0 35
Sweden 59 5 1 1 1 0 26
Thailand 53 6 2 1 1 1 30
Turkey 27 5 3 2 2 0 30
Uruguay 74 5 1 1 1 0 26
Venezuela 18 4 1 2 2 0 26
Table 5.4 Borderline signals of banking crises, June 1996-June 1997
Monthly indicators Annual indicators Total
Institute for International Economics
Mexico 30 5 1 2 2 0 30
http://www.iie.com
Norway 37 8 2 1 1 0 39
Peru 27 6 2 1 1 0 30
The Philippines 73 8 3 2 2 0 43
South Africa 68 10 3 1 1 0 48
South Korea 74 13 5 3 3 1 61
Spain 58 9 2 1 1 0 43
Sweden 66 8 2 1 1 0 39
Thailand 58 9 3 1 1 1 43
Turkey 35 6 3 2 2 0 35
Uruguay 86 6 1 1 1 0 30
Venezuela 34 6 2 2 2 0 35
61
Table 5.5 Weighting the signals for currency and banking crises in
emerging markets, June 1996-June 1997
Currency crises Banking crises
Country Weighted signals Rank Weighted signals Rank
Argentina 5.41 16 7.98 10
Bolivia 6.59 12 7.30 13
Brazil 7.57 10 6.08 14
Chile 5.90 15 5.74 16
Colombia 10.59 8 11.87 6
Czech Republic* 15.42 2 17.24 1
Egypt 6.02 14 8.33 9
Greece 14.27 6 14.15 3
Indonesia* 7.54 11 8.33 9
Israel 6.30 13 10.38 8
Malaysia* 12.46 7 7.74 12
Mexico 2.82 19 2.59 19
Peru 2.82 19 5.33 17
The Philippines* 14.40 5 11.52 7
South Africa 16.52 1 12.74 4
South Korea* 14.57 4 14.55 2
Thailand* 14.63 3 12.09 5
Turkey 8.21 9 7.87 11
Uruguay 4.40 18 4.88 18
Venezuela 5.28 17 6.02 15
Note: An asterisk (*) denotes the country had a currency crisis, a banking crisis, or both in
1997-98.
Note: An asterisk (*) denotes the country had a currency crisis, a banking crisis, or both in
1997-98.
economies in our sample; currency and banking crises are treated sepa-
rately. The first data column provides the relevant value of the index for
a currency crisis. The next column shows the country’s ordinal ranking
for the vulnerability index relative to the remaining 19 countries. South
Africa, the Czech Republic, and Thailand emerge as the most vulnerable
on the basis of the signals issued and the quality of those signals during
January 1996-June 1997.
For banking crises, the comparable exercise ranks the Czech Republic,
South Korea, and Greece as the most vulnerable. Perhaps not surprisingly,
near the bottom of the list are countries such as Mexico and Venezuela,
which are still recovering from their 1994-95 crises.
Thus far, we have treated banking and currency crises separately in
our vulnerability rankings. If one wanted to assess the ‘‘average’’ vulnera-
bility to both banking and currency crises, one may want to combine
the information contained in these two measures. Table 5.6 provides
information on the average proportion of indicators signaling banking
OUT-OF-SAMPLE RESULTS 63
These probabilities will be estimated using all the information from all
the countries in the sample. Once we estimate these probabilities and use
the information on the number of signals being issued at any moment in
time, we can construct time-series probabilities of crisis for every country.
P tm denotes the probability of a crisis for country m in period t.
Once we construct these time series of crisis probabilities, we can also
evaluate the forecasting ability of the composite indicator and compare
its track record with that of other indicators, such as our top-ranked
univariate indicator, the real exchange rate. To conduct this horse race,
we follow Diebold and Rudebusch (1989) and employ the Quadratic
Probability Score (QPS) as our metric of goodness of fit. In particular, the
QPS evaluates the average closeness of the predicted probabilities and
OUT-OF-SAMPLE RESULTS 65
Memorandum:
Unconditional Unconditional
probability of a probability of a
currency crisis banking crisis
0.29 0.10
n.a. ⳱ not applicable
Source: Kaminsky (1998).
where k ⳱ 1,2,3 refers to the indicator Pk, refers to the probability associ-
ated with that indicator, and Rt refers to the zero-one realizations. The
QPS ranges from zero to two, with a score of zero corresponding to
perfect accuracy.
Empirical Results
Table 5.7 reports the conditional probabilities of both currency and bank-
ing crises using the composite indicator. One column reports the likeli-
hood of currency crises. When almost none of the indicators are signaling
a future crisis, the composite indicator takes on values between zero and
two, and the probability of a currency crisis is only about 10 percent. The
probability of a currency crisis increases sharply and nonlinearly as signs
2. This approach has also been used to assess the ability of various indicators to anticipate
turning points in the business cycle (Diebold and Rudebusch 1989).
OUT-OF-SAMPLE RESULTS 67
4. For a more detailed exposition of the incidence of flashing indicators in the run-up to
the Asian crisis, see Kaminsky and Reinhart (1999).
5. This is at odds with the interpretation of these crises provided in Radelet and Sachs
(1998), who argue these crises are the byproduct of a financial panic.
6. It is noteworthy that finance companies had been receiving substantial assistance from
the central bank during this period.
0.8
0.6
0.4
0.2
0.0
1990 1991 1992 1993 1994 1995 1996 1997 1998
Malaysia
1.0
0.8
0.6
0.4
0.2
0.0
1990 1991 1992 1993 1994 1995 1996 1997 1998
The Philippines
1.0
0.8
0.6
0.4
0.2
0.0
1990 1991 1992 1993 1994 1995 1996 1997 1998
Thailand
1.0
0.8
0.6
0.4
0.2
0.0
1990 1991 1992 1993 1994 1995 1996 1997 1998
OUT-OF-SAMPLE RESULTS 69
7. The Philippines was classified as a managed float in the IMF’s exchange rate arrangements
classification. Yet even a relatively uninformed bystander could see the large-scale extent
of foreign exchange intervention before mid-1997, which kept the Philippine peso’s value
virtually unchanged against the dollar.
8. The beginning of the banking crisis in Indonesia can be dated to November 1992, when
a large bank (Bank Summa) collapsed and triggered runs on three smaller banks. Most
state-owned banks also experienced serious difficulties.
9. The reversal was, in fact, quite pronounced, from capital inflows in the region of $50
billion in 1996 to an outflow of $21 billion in 1997. See Kaminsky and Reinhart (2000) and
the next chapter for a discussion on world and regional financial links and their effects on
the probability of currency crises.
OUT-OF-SAMPLE RESULTS 71
1. Of course, the political turmoil at this time in Indonesia is likely to have contributed to
the meltdown of the currency and the economy.
73
Defining Contagion
Only one study that we are aware of examined the issue of contagion in
the context of Latin America’s debt crisis of the 1980s. Doukas (1989)
interprets contagion as the influence of ‘‘news’’ about the creditworthiness
of a sovereign borrower on the spreads charged to the other sovereign
borrowers, after controlling for country-specific macroeconomic funda-
mentals. Most other studies, such as Valdés (1997), define contagion as
excess comovement in asset returns across countries, be it for debt or
equity. This comovement is said to be excessive if it persists even after
common fundamentals, as well as idiosyncratic fundamental factors, have
been controlled for. A recent variant to this approach (as in Forbes and
Rigobon 1998) defines ‘‘shift-contagion’’ as an increase in excess comove-
ment of asset returns during crisis periods.
Eichengreen, Rose, and Wyplosz (1996) define contagion as a case where
knowing that there is a crisis elsewhere increases the probability of a
crisis at home, even when fundamentals have been properly taken into
account. This is the definition of contagion that we will explore in the
remainder of this chapter. These fundamentals could be country-specific,
along the lines analyzed in the preceding chapters, or they could be
external and common to all countries or a group of countries. Changes
2. See Gerlach and Smets (1995) for a model that emphasizes bilateral trade and Corsetti
et al. (1998) for one in which emerging markets compete in a common third market.
3. As a story of fundamentals-based contagion, of course, this explanation does not speak
to the fact that central banks often go to great lengths to avoid the devaluation in the
first place.
CONTAGION 75
Empirical Studies
Very few studies have attempted to run ‘‘horse races’’ among alternative
models of contagion. Eichengreen, Rose, and Wyplosz (1996) tested the
influence of bilateral trade links against similarities to the crisis country
in macroeconomic fundamentals. Glick and Rose (1998) examined the
trade issue further within a much broader country sample, while Wolf
(1997) attempted to explain pairwise correlations in stock returns by bilat-
eral trade and by common macroeconomic fundamentals. All studies
CONTAGION 77
4. The countries are classified by bank clusters according to which financial center they
depend on the most (on the basis of the Bank for International Settlements data). For the
high-correlation asset returns cluster, we include countries that have a correlation that is
0.35 percent or higher in their daily stock returns. For the bilateral trade cluster, we include
countries for which either imports or exports to the second country are 15 percent or higher.
For the third-party trade cluster, we require countries to have a common third market and
similar commodity export structure. We focus on the top 10 to 15 goods that account for
40 percent or more of exports in the initial crisis country; we then see if those same goods
account for a significant share (20 percent or higher) of exports of the remaining countries.
For example, the top 14 Thai exports account for 46 percent of total exports; these same
goods account for 44 percent of Malay exports; hence Malaysia is in the same third-party
trade cluster. By contrast, those goods only account for 15 percent of Indonesia’s exports,
leaving Indonesia outside the third-party trade cluster.
We now consider, on the basis of the trade and financial sector linkages
discussed here, which countries would have been classified as vulnerable
to contagion during three recent episodes of currency crises in emerg-
ing markets.
The first of these episodes began with the devaluation of the Mexican
peso in December 1994. On the heels of the Mexican devaluation, Argen-
tina and Brazil were the countries to come under the greatest speculative
pressure. In a matter of a few weeks in early 1995, the central bank of
Argentina lost about 20 percent of its foreign exchange reserves and bank
deposits fell by about 18 percent as capital fled the country. Such a severe
outcome could hardly be attributed to trade linkages and competitive
devaluation pressures, as Argentina does not trade with Mexico on a
bilateral basis, nor does it compete with Mexican exports in a common
third market. 5 In the case of Brazil, the speculative attack was brief,
although the equity market sustained sharp losses. Both of these countries
record high vulnerability index scores following the Mexican devaluation.
While the effects on Asia of the Mexican crisis were relatively mild,
the country that encountered the most turbulence in the region was the
Philippines, which also registers a relatively high vulnerability score.
In the case of the Thai crisis, Malaysia shares both trade and finance
links with Thailand. For the other Asian countries, the potential channels
of transmission are fewer. As noted earlier, the Philippines is a part of
the same third-party trade cluster as Thailand, which receives a weight
of 1.75 (i.e., 1/0.57) in the composite index; it is also part of the Asian
high-correlation cluster, which receives a weight of 2.57 (i.e., 1/0.39) in
the index. Indonesia shares the same high-correlation cluster with Thai-
land, and it is a part of the Japanese bank cluster, which receives a weight
of 14.08 (i.e., 1/0.07). Hence, as shown in table 6.4, Indonesia and the
Philippines’ contagion vulnerability index scores are 16.65 and 4.32,
5. See Kaminsky and Reinhart (2000) for details on the pattern of trade.
CONTAGION 79
Table 6.3 Countries sharing financial and trade clusters with original crisis country or region
High-correlation Third-party trade Bilateral trade
Institute for International Economics
Peru 1
http://www.iie.com
The Philippines 1 1 1
South Korea 1 1
Spain
Sweden
Thailand 1 1 1
Turkey
Uruguay 1 1
Venezuela 1 1
CONTAGION 81
Table 6.5 Characteristics of affected countries in Asian and Mexican episodes of contagion
Level of trade with
Institute for International Economics
Malaysia (July) Managed Crisis Yes High, 0.60 Moderate, 5.88 Moderate, 4.1 High, 44.4
http://www.iie.com
float
The Philippines Managed Crisis Yes High, 0.68 Low, 2.40 Moderate, 3.8 Low, 19.2
(July) float
Indonesia Narrow Crisis Yes High, 0.54 Moderate, 4.35 Low, 1.8 Low, 15.5
(August) band
Hong Kong Currency Turbulence No High, 15.33 Low, 1.0 Low
(October) board
South Korea Crawling Crisis Yes Low, 0.24 Moderate, 6.16 Low, 2.0 Moderate, 27.9
(November) band
Table 6.6 Asia and Latin America: added power of Thai crisis in
explaining probability of contagion in bank cluster,
July 1997
Probability of a crisis conditioned
on crises elsewhere in the cluster minus
Country unconditional probability of crisis
Asia
Indonesia 0.60
Malaysia 0.35
The Philippines 0.02
Latin America
Argentina 0.02
Chile 0.02
Mexico 0.02
CONTAGION 83
85
a. We note in parentheses whether the variable remained below or above the tranquil-
period norm.
b. Domestic credit as a share of GDP remains above normal levels largely as a result of
the decline in GDP following the crisis.
c. The disparity between the postcrisis behavior of real interest rates lies in the fact that a
large share of the currency crises occurred in the 1970s, when interest rates were controlled
and not very informative about market conditions.
Table 7.1 summarizes the results of that aftermath exercise for currency
and banking crises. The number given after each indicator is the average
number of months that it takes for that variable to reach its norm during
tranquil periods. In parentheses, we note whether the level or growth
rate of the variable remains above or below its norm in the postcrisis
period. Several findings merit special attention.
First, the deleterious effects of banking crises do linger longer than
currency crises’ effects. This is evident in several of the indicators. While
the 12-month change in output remains below its tranquil-period norm
for (on average) 10 months following the currency crash, it takes nearly
twice that amount of time to recover following the banking crisis. This
more sluggish recovery pattern is also evident in imports, which take
about 21⁄2 years to return to their norm. The weakness in asset prices,
captured here by stock prices that are below the norm, persist for 30
months on average for banking crises—more than twice the time it takes
to recover from a currency crash.
There are several explanations for banking crises’ more protracted
recovery periods. One concerns the special nature of the ‘‘twin’’ crises.
The bulk of the banking crises in this sample were accompanied by cur-
rency crises, and twin crises ought to have more severe effects on the
economy, as argued in Kaminsky and Reinhart (1999).
t ⳱ year of crisis
a. Moderate-inflation countries are those with inflation rates below 100 percent in all years
surrounding the crisis; high-inflation countries are those in which inflation exceeded 100
percent in at least one year.
1. For a comparison of the recent crises with the historic norm, see Kaminsky and Rein-
hart (1998).
Some Caveats
The preceding discussion has suggested a ‘‘representative time profile’’
for the recovery process in the wake of currency and banking crises. This
2. See also Kamin and Rogers (1997) for an interesting analysis of the case of Mexico.
Edwards (1989) 1962-82, 39 devaluations Inflation, GDP, current Inflation doubles, on average from about 8 to 16.7 percent one
greater than 15 percent in account as a share of year after the crisis; net foreign assets/money fall by about 5
24 countries GDP, change in net percent in the three years following the crisis.
foreign assets/money
Kamin (1988) 1953-83, 50 to 90 Inflation, GDP, exports, The trade balance does not change much the year following the
devaluations in excess of imports, export prices, devaluation; import and export growth increase. Capital inflows
15 percent import prices, capital and reserves are about the same at tⳭ1 as in the year of the
inflows, trade balance, devaluation. Inflation increases the year of the devaluation then
reserves declines. GDP growth falls the year of the devaluation then
recovers the following year.
Kiguel and Ghei 1950-90, 33 devaluations Real exchange rate, About 60 percent of the devaluation is not eroded by increases
(1993) in excess of 20 percent in inflation, GDP growth, in domestic prices. Inflation increases, on average by about
low-inflation countries exports/GDP, reserves/ 11⁄2 percentage points, between tⳭ3 and ⳮ1; growth
imports, parallel pre- increases by 1 percent in that same period; exports and
Institute for International Economics
91
Table 7.5 Comparison of severity of crises by region and period,
1970-97
Banking crises (bailout cost as
Currency crises (indexa) share of GDP)
Latin Latin
Period America East Asia Other America East Asia Other
1970-94 48.1 14.0 9.0 21.6 2.8 7.3
1995-97 25.4 40.0 n.a. 8.3 15.0 n.a.
profile suggests that growth will return to normal within about two years
of the crisis and that the inflationary consequences of the devaluation
will abate within about three years. Yet this pattern would hardly describe
the protracted recovery process of many Latin American economies dur-
ing the 1980s, not even the relatively rapid recovery experienced by Chile.3
The speed at which the economy recovers from a financial crisis will
be heavily influenced by how policymakers respond to the crisis as well
as by external conditions. The high level of international real interest rates
in the 1980s (the highest levels since the 1930s) were hardly conducive
to speedy recovery. Moreover, as suggested in Kaminsky and Reinhart
(1998a), severe currency and banking crises are apt to be associated with
more delayed recoveries. This latter point is particularly relevant to the
recovery from the 1997-98 crises in Asian countries, which are significantly
more severe that the earlier crises in that region.
To analyze this issue formally, we measure the severity of currency
and banking crises, as in Kaminsky and Reinhart (1998). For banking
crises, the measure of severity is the cost of the banking bailout expressed
as a share of GDP. For currency crises, we construct an index that gives
equal weights to reserve losses and currency depreciation. This index is
centered on the month of the currency crisis, and it combines the percent-
age decline in foreign exchange reserves in the six months before the crisis,
since reserve losses typically occur before the central bank capitulates,
and the depreciation of the currency in the six months following the
abandonment of the existing exchange rate arrangement. This latter com-
ponent captures the magnitude of the currency meltdown.
Table 7.5 presents these measures of severity for the 76 currency crises
and 26 banking crises in the Kaminsky-Reinhart sample. For the 1970-94
sample, currency and banking crises were far more severe in Latin
America than elsewhere. The 1970-94 crises in East Asia, by contrast, were
relatively mild and not that different by these metrics from the crises in
3. Chile’s inflation rate was in single digits when it abandoned its crawling peg policy in 1982.
Summary of Findings
95
Low-frequency indicators
Current account balance/GDP Short-term capital inflows/GDP
Current account balance/investment Current account balance/investment
the pack were a high ratio of short-term capital inflows to GDP and a
large current account deficit relative to investment (table 8.1).
Sixth, while there is a good deal of overlap between the best-perform-
ing leading indicators for banking and currency crises, there is enough
of a distinction to warrant treating the two separately. To highlight
two noteworthy differences, the two indicators that serve as proxies for
financial liberalization—namely, a rise in the real interest rate and an
increase in the money multiplier—turned out to be more important for
banking crises than for currency crises, whereas the opposite proved true
for the two indicators designed to capture currency/maturity mismatches
and excessively expansionary monetary policy—namely, a high ratio of
broad (M2) money balances to international reserves and excess M1 money
balances, respectively.
Seventh, while our data on sovereign credit ratings cover only a subsam-
ple of crises and relate to only two of the major rating firms (Moody’s
Investor Services and Institutional Investor), we find that changes in
sovereign credit ratings have performed considerably worse than the
better leading indicators of economic fundamentals in anticipating both
currency and banking crises in emerging economies. In addition, we
find no empirical support for the view that sovereign rating changes have
led financial crises in our sample countries rather than reacting to these
crises. In a similar vein, we have found that interest rate spreads (i.e.,
foreign-domestic real interest rate differentials) are not among the best-
performing group of leading indicators. More empirical work needs to
be done to determine whether these results are robust to other rating
agencies and other samples. Nevertheless, our findings suggest that those
who are looking to ‘‘market prices’’ for early warning of crises in emerging
economies would therefore be advised to focus on the behavior of real
exchange rates and stock prices—not on credit ratings and interest rate
spreads.
3. Our preferred measure of vulnerability was an index equal to the weighted average of
‘‘good’’ indicators issuing signals in the out-of-sample period. By ‘‘good’’ indicators, we
mean those that had noise-to-signal ratios less than unity during the 1970-95 period. Taking
the monthly and annual indicators as a group, there were 18 ‘‘good’’ indicators. We used
the inverse of the noise-to-signal ratios as weights for the better indicators. We then ranked
each of the 25 countries in the sample according to the computed value of this index. The
index is meant to capture the probability of a crisis—not necessarily its severity.
crisis arrives later, in the summer of 1998. Moreover, while South Africa
did not formally make the cut, it could reasonably be classified as a
near miss since it experienced a quasi-crisis in June 1998 (a 14 percent
devaluation cum a 13 percent decline in reserves that pushed the exchange
market pressure index 2.7 standard deviations above its mean). Malaysia,
which just makes it into the group of the seven most vulnerable, did have
a currency crisis in 1997.
Further information on the out-of-sample performance of the leading
indicators of currency crisis can be gleaned by looking for episodes in
which, to borrow from Sherlock Holmes, the ‘‘dogs were not barking’’—
that is, by looking to see how often crises occurred among those countries
estimated to have relatively low vulnerability. The lower panel of table
8.2 indicates the five countries that were estimated to have relatively low
vulnerability to currency crises in 1996-97. As with the high vulnerability
group, the ordinal rankings of countries are very similar across the two
out-of-sample periods, with Venezuela, Peru, and Uruguay slightly shift-
ing their relative positions in the least vulnerable list. Perhaps an explana-
tion as to why the index of vulnerability is relatively low for some of
these countries can be found in the fact that some of these countries were
still recovering from earlier crises (Mexico and Venezuela).
But what about Indonesia, which after all suffered the most severe
currency crisis (beginning in the summer of 1997) among the sample
4. It should be recognized that none of the existing early warning models—including the
regression-based models—anticipated the Indonesian crisis.
5. Indonesia’s equity prices did suffer a severe decline, but it did not begin until August 1997.
6. Using a very similar approach but a slightly different set of indicators, Kaminsky (1998),
who presents a time series of calculated crisis probabilities for the Asian economies, finds
results that are in line with those shown in tables 1.6 and 1.7—namely, that estimated
currency-crisis vulnerability increased markedly before the 1997 event in Thailand and
moderately in Malaysia and the Philippines. Again, no such increase in estimated vulnerabil-
ity was present for Indonesia. South Korea was not in her sample. Radelet and Sachs (1998)
take the opposing view that the crisis in Asia was mainly attributable to investor panic. As
discussed in chapter 6, we only find that argument to be convincing in the case of Indonesia.
7. The Czech banking crisis was not included in our in-sample test, and hence the model
is not calibrated to account for this crisis.
8. The Malaysian crisis would probably best be regarded as beginning in March 1998,
when the central bank announced losses at Sime Bank and elsewhere and when Malaysian
President Datuk Seri Mahathir bin Mohamad pledged state funds to prop up weak
institutions.
9. Malaysia was ranked fourteenth (out of 25 countries) in the shorter period and tenth in
the longer one.
10. This is consistent with the results of a recent IMF study (Berg and Pattillo 1999), which
found that the signals model of Kaminsky, Lizondo, and Reinhart (1998) did a better job
of predicting the Asian crisis than the models of Frankel and Rose (1996) and of Sachs,
Tornell, and Velasco (1996).
11. See Berg and Pattillo (1999), The Economist (1998), Furman and Stiglitz (1998), IMF
(1998c), and Wyplosz (1997, 1998).
12. More specifically, the ‘‘tranquil’’ period excludes the 24 months before and after currency
crises and the 24 months before and 36 months after banking crises.
13. In the cases where currency and banking crises coincide, the postcrisis performance
would show up in both the averages for currency and banking crises. See chapter 7 for details.
14. See, for example, Demirgüç-Kunt and Detragiache (1998), who introduce law enforce-
ment and deposit-insurance variables into their banking crisis model.
15. Calvo, Reinhart and Végh (1995) present empirical evidence on this issue.
The Indicators
Sources include the IMF’s International Financial Statistics (IFS), the Interna-
tional Finance Corporation’s (IFC) Emerging Market Indicators, and the
111
Ades, Alberto, Rumi Masih, and Daniel Tenegauzer. 1998. A New Framework for Predicting
Financial Crises in Emerging Markets. New York: Goldman Sachs. December.
Agenor, Pierre-Richard, Jagdeep S. Bhandari, and Robert P. Flood. 1992. Speculative Attacks
and Models of Balance of Payments Crises. IMF Staff Papers 39 (June): 357-94.
Bank for International Settlements (BIS). 1998. The Maturity, Sectoral, and Nationality Distribu-
tion of International Bank Lending. Basel: Bank of International Settlements.
Berg, Andrew, and Catherine Pattillo. 1999. Are Currency Crises Predictable? A Test. IMF
Staff Papers, no. 2 (June): 107-38.
Blanco, Herminio, and Peter M. Garber. 1986. Recurrent Devaluation and Speculative Attacks
on the Mexican Peso. Journal of Political Economy 94 (February): 148-66.
Borensztein, Eduardo, and José de Gregorio. 1998. Devaluations and Inflation after Currency
Crises. Photocopy. Washington: International Monetary Fund.
Calomiris, Charles. 1997. The Post-Modern Bank Safety Net: Lessons from Developed and Develop-
ing Countries. Washington: American Enterprise Institute.
Calvo, Guillermo A. 1998. Varieties of Capital-Market Crises. In Guillermo A. Calvo and
M. King, eds., The Debt Burden and Its Consequences for Monetary Policy. New York:
MacMillan Press.
Calvo, Guillermo A., and Morris Goldstein. 1996. What Role for the Official Sector? In
Guillermo A. Calvo, Morris Goldstein, and Eduard Hochreiter, eds., Private Capital
Flows to Emerging Markets after the Mexican Crisis. Washington: Institute for Interna-
tional Economics.
Calvo, Guillermo A., and Enrique Mendoza. 1996. Mexico’s Balance-of-Payments Crisis: A
Chronicle of a Death Foretold. Journal of International Economics 41, no. 3/4 (November):
235-64.
Calvo, Guillermo A., and Enrique Mendoza. 2000. Rational Contagion and the Globalization
of Securities Markets. Journal of International Economics, volume 51:79-113.
Calvo, Guillermo, Carmen M. Reinhart, and Carlos Vegh. 1995. Targeting the Real Exchange
Rates: Theory and Evidence. Journal of International Economics, volume 47:97-133.
Calvo, Guillermo A., and Carmen M. Reinhart. 2000. Fear of Floating. Photocopy. University
of Maryland.
115
REFERENCES 117
REFERENCES 119
121
INDEX 123
INDEX 125
INDEX 127
INDEX 129
INDEX 131
INDEX 133