A brief survey of the theoretical literature on early warning systems may be found in the annex to this chapter. Three of the recent and more comprehensive empirical studies are summarized below; other recent empirical studies are listed and described briefly in the annex.

In a recent example of the use of multiple indicators, eight indicators of a financial crisis were proposed.¹¹ It is argued that these would have signalled the Asian crisis. These eight indicators, their critical regions (as determined by the author's rules of thumb), and their values in 1996 for the five Asian countries most affected by the crisis are presented in table IV.1. (Unfortunately, the author does not explain the basis for his rules of thumb for the critical range of values.)

It is clear from the table that not all of the countries were experiencing values for the variables that were critical in 1996, except for foreign debt as a percentage of GDP. However, each country (except Malaysia) had at least two variables in the critical range. Such an approach has the advantage of being fairly simple to understand. Nevertheless, the data requirements in terms of timeliness and accuracy are not easy to meet.

In another paper,¹² 28 selected empirical studies on leading indicators for currency crises by various authors were surveyed covering a number of countries, both industrial and developing, and spanning the period from the 1950s to the 1990s. The approach involves monitoring the evolution of several indicators that tend to exhibit unusual behaviour in the period preceding a crisis. When an indicator exceeds or goes below a certain threshold value, this is interpreted as a warning signal that a currency crisis may take place within the following 24 months. The variables (either levels or growth rates) reported as having the best track record within this approach include exports, deviations of the real exchange rate from the trend, the ratio of broad money to gross international reserves, output and equity prices.

This research suggests that: (a) an effective warning system should consider a broad variety of indicators - a currency crisis usually seems to be preceded by economic and sometimes political problems; (b) those variables that receive ample support as useful indicators of currency crises include international reserves, the real exchange rate, credit growth, money supply as a ratio of international reserves, real GDP growth and the fiscal deficit; (c) several foreign, political, institutional and financial variables also have some predictive power in anticipating a currency crisis - banking-sector problems stand out in this regard; and (d) (surprisingly) variables associated with the external debt profile did not fare well. Also contrary to expectations, the current account balance did not receive much support as a useful indicator of crisis. This may be because information provided by the behaviour of the current account balance may already have been reflected to some extent in the evolution of the real exchange rate.

There are limits to the usefulness of this approach for a particular country precisely because the set of indicators proposed are those that perform well in warning of crises over a large sample of countries. If each country is peculiar to itself, as this research suggests, then the best set of leading indicators for a particular country is unique and not necessarily the same as the general set. Nevertheless, the proposed set could be useful as a starting point for those countries or economies that have not yet found a set of leading indicators.

However, if it is accepted that the more important ability of an indicator is to predict a crisis when one in fact does occur later¹³ (in other words, not miss a true crisis), then the results are not very promising. The literature would refer to this as avoiding a type I error that is, failing to signal a crisis that in fact occurs; a type II error corresponds to a false alarm that is, signalling a crisis when none in fact follows. The following simplified version of the table reported in the IMF Staff Papers highlights this aspect of the results. On the average, the indicators signalled a crisis correctly only 68 per cent of the time; that is to say, they missed 32 per cent of the crises.

Berg and Patillo¹⁴ undertook a comparison and evaluation of three early warning models of currency crisis. The models they picked represented the three classes of early warning models enumerated earlier. Specifically, they evaluated the Kaminsky-Lizondo-Reinhart model discussed earlier as a representative of the signals approach, the Frankel and Rose probit model¹⁵ as a representative of the probit/logit approach, and the Sachs, Tornell and Velasco model¹⁶ for the cross-country regression approach. They attempted to replicate the results of these three models, introducing and experimenting with slightly different specifications along the way. Unfortunately, their conclusion is not very promising from the perspective of those who hope for a viable early warning system. While they find that the models are better than naive guesswork, they would not have predicted the 1997 crisis very well.

All the models surveyed tried to come up with leading indicators that performed well over a cross section of countries and crises. As pointed out earlier, precisely because crises and economies can vary in nature, "one size may not fit all" when it comes to leading indicators. In most developed economies, work has been done to identify leading indicators of business cycles (see box IV.2 for a discussion of those used in the United States). In terms of the elements of the composite indexes of leading indicators of business cycles as compared with those for currency crises, it is noticeable that the former tend to be more micro in nature. The only variable common to both is stock prices; other variables which measure similar economic phenomena are the monetary aggregates (M1 and M2), interest rates and GDP. This finding is consistent with the work of Estrella and Mishkin,¹⁷ who found that financial variables (specifically stock prices) are useful for forecasting recessions in the United States economy.

In general, all the models attempting to produce early warning systems or leading indicators of currency crises focused only on macro variables. Some even avoided using micro variables (specifically banking industry variables) in favour of macro variables whose forecasts are more readily avaiable.¹⁸ It is therefore not surprising that there is very little overlap in the composition of both types of indicators. However, even if these types of leading indicators were effective as early warnings of currency crises, their usefulness outside of OECD countries may be limited, as not many other countries collect data on similar leading indicators.

Next

Footnotes:

¹¹ Dominick Salvatore, "Could the financial crisis in East Asia have been predicted?", Journal of Policy Modeling, vol. 21, No. 3 (May 1999), pp. 341-347.

¹² Graciela Kaminsky, Saul Lizondo and Carmen Reinhart, "Leading indicators of currency crises", IMF Staff Papers, vol. 45, No. 1 (March 1998), pp. 1-48.

¹³ This might be the case if, whatever early warning system is used, it is not disseminated to the public. From a policy maker's perspective, it might be more tolerant of a system issuing false alarms rather than one which fails to signal a crisis that in fact occurs later.

¹⁴ Andrew Berg and Catherine Pattillo, "Are currency crises predictable? A test", IMF Staff Papers, vol. 46, No. 2 (June 1999), pp. 107-138.

¹⁵ Jeffrey Frankel and Andrew Rose, "Currency crashes in emerging markets: an empirical treatment", Journal of International Economics, vol. 41, Nos. 3/4 (November 1996), pp. 351-366.

¹⁶ Jeffrey Sachs, Aaron Tornell and Andres Velasco, "Financial crises in emerging markets: the lessons from 1995", Brookings Papers on Economic Activity, vol. 1 (Washington DC, Brookings Institution, 1996), pp. 147-215 with summary of the article available at (25 January 2000).

¹⁷ Arturo Estrella and Frederic S. Mishkin, "Predicting U.S. recessions: financial variables as leading indicators", Review of Economics and Statistics, vol. 80, No. 1 (February 1998), pp. 45-61, available at (25 January 2000).

¹⁸ See, for example, Asli Demirguc-Kunt and Enrica Detragiache, "Monitoring banking sector fragility: a multivariate logit approach with an application to the 1996-97 banking crises", Policy Research Working Paper, No. WPS 2085 (Washington DC, World Bank, 1999).