Chapter 26 Early Warning Indicators of Financial Crises
- International Monetary Fund
- Published Date:
- January 2001
My aim in this paper is to discuss early warning indicators of financial crises. I plan to do that in three steps.
First, I will suggest why one might be interested in early warning indicators.
Second, I will highlight a number of the results that come out of existing empirical studies (drawing in part on some ongoing joint work with Carmen Reinhart).1
And third, I will outline a few avenues for further work in the future.
Because Graciela Kaminsky is also contributing to this volume, and because she has done a great deal of work in this area—much of it with the same coauthor—I will try to avoid undue overlap by emphasizing different points.
2. WHY DO EARLY WARNING STUDIES MATTER?
Why might one be willing to spend resources on finding early warning indicators of banking and currency crises?
I think there are two key reasons. First, there is strong evidence that banking and currency crises are extremely costly to the countries in which they originate (as well as posing significant spillover risks to other countries). According to the IMF’s tally, there have been over sixty-five developing-country episodes during the 1980-95 period when the banking system’s capital was completely or nearly exhausted.2 The public-sector bailout costs of resolving banking crises in developing countries during this period has been estimated to be at least $250 billion.3 In more than a dozen of these banking crises, the public-sector resolution costs amounted to ten percent or more of GDP.4 In the latest additions to the list of severe banking crises, the cost of bank recapitalization for the countries most affected in the ongoing Asian financial crisis is expected to be huge—on the order of thirty percent of GDP for both Thailand and South Korea and twenty percent of GDP for Indonesia and Malaysia.5
In addition to the enormous fiscal costs, banking crises also exacerbate declines in economic activity. Illustrative of the magnitude of output losses, an IMF (1998) study, drawing on a sample of thirty-one developing countries, reports that it typically takes almost three years for output growth to return to trend after the outbreak of a banking crisis and that the cumulative output loss averages twelve percent.
The costs of currency crises have likewise been shown to be significant. Mexico’s peso crisis was accompanied in 1995 by a decline in real GDP of six percent—its deepest recession in sixty years. During the ERM crises of the fall of 1992 and summer of 1993, on the order of $150 billion was spent on official exchange market intervention in the fruitless effort to stave off the forced devaluation and/or floating of ERM currencies. In emerging Asia, consensus forecasts for 1998 growth issued just prior to the crisis (that is, in May/June 1997) generally stood in the six to eight percent range. These forecasts have now been subject to unprecedented downward revisions in the midst of the currency, banking, and debt crises enveloping these economies. The consensus 1998 growth forecasts are now about –15 percent for Indonesia, –5 to –6 percent for South Korea, Thailand, and Malaysia, –3 percent for Hong Kong, and about zero for East Asia as a whole.6 And like banking crises, currency crises too seem to exhibit contagious behavior. One recent study found that a currency crisis elsewhere in the world increases the probability of a speculative attack by about eighth percent, after controlling for economic and political fundamentals in the country concerned.7
The more costly it is to clean up after a financial crisis has already occurred, the greater the returns to identifying reliable early warning indicators that can prompt corrective policy action.
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.
Empirical studies of the 1992–93 ERM crisis have typically concluded that market measures of currency risk did not point to the specter of significant devaluations of the weaker ERM currencies before the fact.8 In the run-up to the Mexican crisis, market signals were again muted or inconsistent. More specifically, measures of default risk on tesebonos (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.9 From April 1994 on, market measures of currency depreciation on the peso usually were beyond the government’s announced rate; nevertheless, this measure of currency risk fluctuated markedly and the gap between market expectations and the official rate was widest in the summer of 1994 when the attack came with most ferocity only in late December.10
The preliminary evidence now available similarly suggests that the performance of interest rate 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.11 A recent analysis of spreads using local interest rates for South Korea, Thailand, and Malaysia found similarly little indicator or growing crisis vulnerability.
Sovereign credit ratings (on long-term, foreign currency debt) issues by the two largest international ratings firms were even less prescient in the Asian crisis. There were almost no downgrades for the most severely affected countries in the eighteen-month run-up to the crisis. As the Economist [1997, p. 68] put it, “in country after country, it has often been the case of too little, too late.” Looking at a larger sample of cases, a recent OECD study was unable to find consistent support for the proposition that sovereign credit ratings act more like a leading than a lagging indicator of market prices (i.e. of interest rate spreads).12
If interest rate spreads and credit ratings only blow the whistle on financial crises once in a while, then there should be strong interest in any other indicators that the data suggest would do a better job.
3. SOME RECENT FINDING
Enough for motivation. What do the early warning exercises usually show?
First, within the sample period, and much like the leading-indicator analyses of business cycles—you do find that there are recurring patterns in the run-ups to banking and currency crises. Crises don’t just come out of the blue. The better leading indicators seem to anticipate correctly somewhere between 50–100 percent of banking and currency crises.13 Of course, they also send false signals, with even the best ones sending one false signal for every three to five correct signals.
Second, when monthly data are examined, banking crises typically turn out (again, within the sample period) to be harder to predict than currency crises. Part of that, I think, reflects the fact that it’s harder to identify the duration of banking crises, particularly as regards the end of the crisis. Also, the indicators exercise is pretty much restricted to macro variables and to proxies for the onset of financial liberalization. You can’t pick up institutional weaknesses in banking systems with available time-series data (see section 4).
Third, there is wide variation in performance across leading indicators, with the best performing indicators displaying noise-to-signal ratios and conditional probabilities that are two to three times (or more) better than those for the worst-performing ones. In addition, the group of indicators that shows the best (in sample) explanatory power also seem, on average, to send the most persistent and earliest signals. Lead times are usually somewhere between ten to eighteen months.
Fourth, when the “signals approach” to forecasting crises is employed and when the in-sample tests are conducted on monthly data, what works well in anticipating banking crises are appreciation of the real exchange rate (relative to trend), a decline in equity prices, a rise in the (M2) money multiplier, a decline in real output, a fall in exports, and a rise in the real interest rate14. When annual data are used as a supplement to the monthly tests, the best of the pack seem to be a high ratio of short-term capital flows to GDP and a large current account deficit relative to investment. When a “regression approach” to explaining past banking crises is utilized (instead of the signals approach), (an increase in) world or “northern” interest rates usually winds up at or close to the top of the class.
Fifth, turning to currency crises, the best of the monthly indicators (from the signals approach) are: appreciation of the real exchange rate (relative to trend), a decline in equity prices, a fall in exports, a high ratio of broad money (M2) to international reserves, a low ratio of international reserves by itself, and excess narrow-money (Ml) balances. A recession just misses the top group. Among the annual indicators, the two best performers are both current-account indicators, namely, a large current-account deficit relative to both GDP and investment. The regression format identifies many of the same leading indicators, as well as finding some support for the inflation rate, world commodity prices, and a decline in (incoming) foreign direct investment.
Sixth, we found that changes in sovereign credit ratings (at least those produced by Moody’s and Institutional Investor magazine) performed considerably worse than the better leading indicators of economic fundamentals in anticipating both currency and banking crises in emerging economies. In addition, we could find no support, in our sample, for the view that rating changes have led financial crises rather than reacting to these crises. In a similar vein, we found that interest rate spreads (i.e. foreign/domestic real interest rate differentials) were not among the best-performing group of leading indicators. While we had data only for a sub-sample of countries and while we need to test this proposition further, our results suggest that if you are looking to “market prices” for early warning of crises in emerging economies, you should focus on the behavior of real exchange rates and of equity prices—not on credit ratings and interest rate spreads.
Seventh, like Kaminsky (1998), we have performed some out-of-sample tests looking at the January 1996–June 1997 period and the January 1996–December 1997 period. For the Asian crisis countries, the model does quite well (with the exception of Indonesia).
For example, suppose we look at currency crises over the 1996 to mid-1997 period. The ordinal vulnerability rankings among our group of twenty-five sample countries were as follows (with “1” being the most vulnerable):
- 1. Czech Republic
- 2. Thailand
- 3. South Korea
- 4. Greece
- 5. South Africa
- 6. Colombia
- 7. Turkey
- 8. Philippines
- 9. Malaysia.
The countries estimated to be the least vulnerable were Mexico, Venezuela, Argentina, Peru, and Indonesia.
For the somewhat longer January 1996 to December 1997 period, Czech Republic, Thailand, Korea, Greece, Philippines, South Africa, Colombia, Turkey, and Malaysia again emerged as highly vulnerable to currency crises.
Overall, I would regard these ordinal country rankings as encouraging. Most of the countries actually experiencing crises were estimated to be highly vulnerable and there were few misclassifications as well at the low vulnerability end of the spectrum. The one Asian crisis country where the model does poorly is Indonesia, which is ranked twenth-fifth and twenty-third place in the two out-of-sample tests. This may reflect the fact that (owing to data limitations) our set of indicators does not include a variable that picks up liquidity and currency mismatches among nonfinancial corporations. In addition, some of the best performing indicators overall—namely real exchange rate overvaluation, equity prices, and low international reserves, were not flashing in Indonesia’s case.
The out-of-sample results for banking crises are decent, but not as impressive as the currency predictions.
My conclusion is that these kinds of forecasting models are useful for making coarse distinctions between, say, the six or seven countries ranked as most vulnerable and the six or seven countries ranked as least vulnerable. In other words, these models can serve as a useful first screen.
4. SUGGESTIONS FOR FURTHER WORK
Finally, let me offer some suggestions for how we could improve these kinds of models.
(a) An obvious extension would be to bring cross-country contagion into the model. At the simplest level, you could include as an independent indicator the number of crises that had taken place in the region or in the world over the past “x” months.
Now, you might say, I could improve on that by accounting for the fact that a crisis in country A is more likely to make country B vulnerable than a crisis in country C.
The difficulty is that there are many channels of contagion.
One is straightforward bilateral trade and investment flows. The more I export to you and the more my banks lend to you, the more vulnerable I am when you have a crisis.
A second channel is the dynamics of competitive devaluation. As one country after another in the region devalues, the countries that have not devalued suffer a loss of competitiveness that makes them more vulnerable. Here we also want to compare competition in third-country markets.
Channel number three for contagion is perceived similarities in country vulnerabilities—what I’ve labeled (in the context of the Asian financial crisis) the “wake-up call.”15 When Thailand falls, it wakes up investors to the reality that many other Asian emerging economies have similar vulnerabilities, e.g. weak financial sectors, large current account imbalances, appreciated real exchange rates etc. They then reassess creditworthiness in the region, and write down other Asian economies with similar vulnerabilities.
Yet another contagion channel operates via the induced effects of a crisis on demand and prices for primary commodities. For example, the Asian crisis has exacerbated the weakness in oil prices and as such has increased the vulnerability of oil producers (Russia, Ecuador, Mexico, Venezuela, and Norway).
The liquidity and margin call channel also should be taken into account. As financial firms mark to market their losses on some emerging market securities, their efforts to meet margin calls and to stay within value-at-risk targets induce sales of other emerging market economy securities—including those with relatively good creditworthiness and relatively high liquidity.
Finally, a perceived regime shift can be an instrument of contagion. For example, the unilateral and discriminatory nature of the recent Russian debt rescheduling/default, cum the relatively mild criticism of it by the G-7 countries and the IMF, may have convinced investors that more debt rescheduling by other emerging economies would be in the offing, and that the risks of investing in emerging economies as an asset class had undergone a sharp upward shift. This, in turn, would increase interest rate spreads for all emerging market borrowers.
Modeling cross-country contagion is therefore going to be a real challenge—but one that will need to be addressed.
(b) A second improvement would be to bring institutional characteristics of weak banking systems into the forecasts of banking crises. There is a strong presumption that such institutional features as weak accounting, provisioning, and legal frameworks, policy-directed lending, the ownership structure (government ownership, foreign ownership, etc.) of the banking system, high levels of connected lending, and incentive-incompatible official safety nets, matter for vulnerability. Yet it is only very recently that these factors have begun to enter the empirical literature.16
The main constraint on making better use of these institutional banking variables is that one can’t of course get high frequency measurements of them; indeed, for some of them (e.g., the share of government ownership or the share of foreign banks in total banking assets), it’s proven difficult to get annual data that is less than two or three years old. This means that such institutional variables have to be introduced as zero-one dummy variables in a time-series context. There should be more scope to take advantage of such factors in cross-section work, that is, in explaining cross-country differences in the incidence of banking crises over long-time periods. To my mind, such work should be encouraged as a companion to the high frequency time-series studies since the former would help us to obtain a better picture of which institutional characteristics of emerging market banking systems seem to increase crisis vulnerability and which do not.
The third and final set of improvements goes under the heading of methodological refinements. Here I would emphasize three suggestions.
First, we need to learn more about the relative performance of the signals approach and the regression approach. In a very useful recent paper, Berg and Patillo  do out-of-sample tests on currency crises, where they compare the signals approach to two regression-based models by Frankel and Rose , and Sachs et al . The signals approach wins, but they also find that (i) you could obtain better results by dropping the threshold restrictions from the signals approach, that is, just letting the indicators enter linearly; and (ii) that while the signals approach does better than the others, the signals model does not do well in forecasting currency crisis in 1997. In my work with Reinhart, the signals model does better for 1996 and 1997 than Berg and Patillo find. We need further work on these out-of-sample tests.
Second, the work Kamisnky  is now doing on the best way to construct a “composite indicator” is important and should be expanded. In a similar vein, it would be useful to test whether the interaction of indicators (e.g. the real interest rate in conjunction with the debt stock, rather than just the real interest rate alone) shows any promise.
Finally, we need to find out whether the different forecasting models of crises now available in the official and private sectors produce robust ordinal rankings of country vulnerabilities, that is, can we get from the models a reasonable consensus about “whose next.”
All of this should yield fertile ground for future work on early warning of banking and currency crisis.
Leiderman and Thorne (1996).
See Goldstein (1998).
- Search Google Scholar
- Export Citation
- Search Google Scholar
- Export Citation
Berg, Andrew, and CatherinePatillo. 1998, “Are Currency Crises Predictable? A Test,” International Monetary Fund, working paper, July.
Caprio, Gerard, and DanielaKlingebiel. 1996, “Bank Insolvencies: Cross-Country Experience,” World Bank, unpublished.
Demirguc-Kunt, Ashli, and EnricaDetragiache. 1997, “The Determinants of Banking Crises: The Evidence from Developed and Developing Countries,” International Monetary Fund and the World Bank, unpublished, May.
Eichengreen, Barry, AndrewRose, and CharlesWyplosz. 1996, “Contagious Currency Crises,” NBER, working paper, No. 5681, July.
Eschweiler, Bernhard.1997, “Did the Market See the Asian Crisis Coming?,” paper presented at World Bank Conference on The Asian Crisis, Washington, DC, (October4).
Eschweiler, Bernard.1998, “Asian Financial Markets,”Singapore: JP Morgan, (July17).
Goldstein, Morris.1986, “The Asian Financial Crisis: Causes, Cures and Systemic Implications. Policy Analyses in International Economics,”No. 55, June.
Goldstein, Morris, and CarmenReinhart.1999, Forecasting Financial Crises: Early Warning Signals for Emerging Markets Policy Analyses in International Economics. Washington: Institute for International Economics forthcoming.
Honohan, Patrick.1997, “Banking System Failures in Developing and Transition Economies: Diagnosis and Prediction.”Dublin: University College, unpublished.
International Monetary Fund, 1996, World Economic Outlook, Washington: International Monetary Fund, October.
Kaminsky, Graciela, and CarmenReinhart. 1996, “The Twin Crises: The Causes of Banking and Balance of Payments Problems,” Board of Governors of the Federal Reserve System and the International Monetary Fund, International Finance, discussion paper, No. 544,
Larrain, Guilermo, HelmutReisen, and Julia vonMaltzan. 1997, “Emerging Market Risk and Sovereign Credit Ratings.” OECD, technical papers, No. 124, April.
Leiderman, Leonardo, and Alfredo E.Thorne. 1991, “The Mexican Crisis and Its Aftermath: What are the Main Lessons?” in Private Capital Flows to Emerging Markets After the Mexican CrisisGuillermoCalvo, MorrisGoldstein, and EduardHochreiter (eds), Washington: Institute for International Economics.
Lindgren, Carl-Johan, GillianGarcia, and MathewSaal. 1996, Bank Soundness and Macroeconomic Policy, Washington: International Monetary Fund.
Obstfeld, Maurice, and KenRogoff. 1995, “The Mirage of Fixed Exchange Rates,” Journal of Economic Perspectives, Vol. 9, pp. 73–96.
Rose, Andrew, and LarsSvensson. 1994, “European Exchange Rate Credibility Before the Fall,” European Economic Review, pp. 1185–1216.
Sachs, Jeffrey, AaronTornell, and AndresVelasco. 1996, “Financial Crises in Emerging Markets: The Lessons from 1995,” Brookings Papers on Economic Activity, Vol. 1, Washington: Brookings Institution.