Peterson Institute publications
The Peterson Institute for International Economics is a private, nonprofit, nonpartisan
research institution devoted to the study of international economic policy. More › ›
RSS News Feed Search

News Release

New Study Unveils Early Warning System for Financial Crisis

June 28, 2000

Contact:    Morris Goldstein    (202) 328-9000

Washington, DC—Over the past three decades, there has been a recurring pattern of vulnerability in the run-up to currency and banking crises in emerging economies, and a relatively small set of leading indicators performed well in anticipating the countries most adversely affected during the Asian financial crisis. Implementation of an early warning system by emerging economies themselves, international financial institutions, and private creditors could thus help identify crisis vulnerabilities at an early stage and increase the odds of taking timely corrective action. Such a system should promptly be added to the "international financial architecture."

A sample of 87 currency crises and 29 banking crises that occurred in 25 large emerging economies and small industrial countries over the 1970-95 period, and a comprehensive battery of empirical tests leads Assessing Financial Vulnerability: An Early Warning System for Emerging Economies by Morris Goldstein, Graciela Kaminsky, and Carmen Reinhart (GKR), to the following eight key findings and conclusions:

  1. Contrary to popular belief, currency and banking crises usually do not appear out of the blue. A high proportion—between 50 and 100 percent—were anticipated by the better-performing leading indicators. Only about 15 percent of crises occurred with one-third or fewer of the indicators flashing a danger signal. There is a problem, however, of too many "false alarms" (on the order of one false alarm for every two to five true signals even in the case of the better leading indicators).

  2. Banking crises in emerging economies are more difficult to forecast accurately than are currency crises. This probably reflects difficulties in accurately dating banking crises and the absence of monthly or quarterly data on the institutional characteristics of national banking systems.

  3. For currency crises, the best of the 15 monthly leading indicators were appreciation of the real exchange rate (relative to trend), a banking crisis, a decline in stock prices, a fall in exports, a high ratio of broad money (M2) to international reserves, and a recession. Current-account indicators were the best of the pack for the annual indicators.

  4. For banking crises, the best of the monthly indicators were appreciation of the real exchange rate, a decline in stock prices, a rise in the money multiplier, a recession, a fall in exports, and the rise in the real interest rate. Among the annual leading indicators of banking crises, a high ratio of short-term capital inflows to GDP came out on top. While there is a good deal of overlap between the best-performing leading indicators for banking and currency crises, there is enough of a distinction to treat the two separately.

  5. While the GKR data on sovereign credit ratings cover only a subsample of crises and relate to only two of the major rating firms (Moody's Investor Services and Institutional Investor), 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 a similar vein, interest rate spreads (i.e. foreign-domestic interest rate differentials) are not among the best performing group of leading indicators. Those who look to "market prices" for early warnings of crises would therefore do better to focus on the behavior of real exchange rates and stock prices -- not on credit ratings and interest rate spreads.

  6. The initial out-of-sample results for currency crises were encouraging. Such out-of-sample testing is crucial because some models that fit well within the sample period do not forecast well out of sample. GKR used both the January 1996-June 1997 and January 1996-December 1997 periods to test the performance of the model beyond the estimation period (1970-95). For each period, they concentrated on the ordinal ranking of countries according to their crisis vulnerability. They measured crisis vulnerability according to a "composite vulnerability index," constructed as a weighted average of the better-performing leading indicators, with past forecasting accuracy serving as the weights for the individual indicators.

    The list of the "most vulnerable" countries was quite similar across the two periods and comprised the Czech Republic, South Korea, Thailand, South Africa, Colombia, and the Philippines; if the list were extended to the seven most vulnerable countries, Malaysia would also have been on it. According to the GKR index of currency crises, almost all of the "most vulnerable" countries experienced currency crises either within or very close to the out-of-sample period. Equally relevant, none of the countries (Chile, Venezuela, Mexico, Peru, and Uruguay), classified as "least vulnerable" according to the composite vulnerability index, experienced a currency crisis during the out-of-sample windows.

    The most serious misclassification was Indonesia, which suffered the most severe currency crisis but was not rated as highly vulnerable by the model. The explanation for this error seems to lie in two areas: first, most of the best-performing leading indicators (e.g., a large appreciation of the real exchange rate) were not flashing in the run-up to the Indonesian crisis; and second, several factors important in the Indonesian case (political instability, liquidity/currency mismatches in the corporate sector, and regional cross-country contagion effects) were not included in the list of 25 indicators. Nevertheless, failure to anticipate the Indonesian crisis should not obscure the finding that of the five countries most adversely affected by the Asian crisis (Thailand, South Korea, Indonesia, Malaysia, and the Philippines), the GKR model placed three of them in the "most vulnerable" group and another in the upper third of the country vulnerability rankings. Given the failure of both "market prices" (interest rate spreads and credit ratings) and official-sector surveillance to anticipate these Asian currency crises, this performance is noteworthy.

    The out-of-sample performance for banking crises was respectable but less impressive than that for currency crises. Here, two of the five countries classified as "most vulnerable" experienced banking crises that fell within the prediction window, and none of the five countries in the "least vulnerable" group experienced a crisis. Again, Indonesia emerged as a major misclassification. The "false positives" problem (i.e. predicting a crisis when it does not occur) is more serious for banking crises than for currency crises.

  7. Using a somewhat more limited sample (20 countries), GKR looked for patterns in the cross-country contagion of currency crises where "contagion" is defined as a case where the presence of a crisis elsewhere increases the probability of a crisis at home, even when own-country fundamentals have been taken into account. Cross-country contagion adds significantly to own-country fundamentals in explaining emerging-market vulnerability to financial crises. Such contagion has operated more along regional than global lines. According to the GKR "contagion vulnerability index," Brazil, Argentina, and the Philippines had high vulnerability to the 1994 Mexican peso crisis; Malaysia, South Korea, and Indonesia had high vulnerability to the 1997 Thai crisis; and Argentina, Chile, and Uruguay had high vulnerability to the 1999 Brazilian crisis. Withdrawal of a common bank lender (particularly Japanese banks) and high correlation of asset returns were found to be important in the Asian contagion of 1997-98.

  8. GKR also studied the speed with which emerging economies return to normalcy after a currency or banking crisis. The deleterious effects on economic activity were found to be more lingering for banking crises than for currency crises. Whereas it took about two years for economic growth to return to its average over the two precrisis years after a currency crisis, that recovery was not evident even three years after a banking crisis. Two other dimensions of the protracted nature of banking crises are that it takes about three to four years for a banking crisis to be resolved and that close to a year and a half elapses between the onset of a banking crisis and its peak. The findings highlight the challenges faced by the Asian crisis countries in sustaining their recoveries.