Assessing Financial Vulnerability: An Early Warning System for Emerging Markets Morris Goldstein, Graciela L. Kaminsky and Carmen Reinhart Institute for International Economics, Washington, DC, USA. June 2000, 134 pp. Price: $15.95. ISBN 0-88132-237-7

✍ Scribed by Matthieu Bussiere

• Publisher: John Wiley and Sons
• Year: 2001
• Tongue: English
• Weight: 37 KB
• Volume: 6
• Category: Article
• ISSN: 1076-9307
• DOI: 10.1002/ijfe.150

✦ Synopsis

One of the most striking aspects of the literature on currency and banking crises is the wide heterogeneity of the explanations put forward to explain such crises. Clearly, there is a need to empirically assess these competing theories, especially as this would be a major stepping stone towards the prediction of economic crises. The present volume represents an important contribution to this objective by setting up an innovative methodology to test the predictive power of some 25 indicators. In contrast to most estimations that rely on some sort of regression (OLS or probit), the authors' early warning system (EWS) analyses the dynamic effects of dummy variables on a 0/1 crisis index. Chapter 2 provides a clear exposition of this technique: instead of regressing a continuous crisis index on continuous lagged indicators, they let underlying series (for instance, the current account as a proportion of GDP) send a 'signal' when it crosses a given threshold, and observe whether this signal is followed by a crisis. The efficiency of the indicators is then measured by their success in signalling crises while avoiding 'false alarms'. Chapter 2 also presents the 25 countries of the sample, covering the period 1970 -1995 mainly with monthly data. The flexibility of the methodology and the sample size allow the authors to tackle a wide range of topics: Chapter 3 discusses the relevance of each indicator, Chapter 4 assesses the performance of credit rating agencies (which have notoriously underpredicted crises), Chapter 5 presents some out-of-sample tests, Chapter 6 delves into the issue of contagion across countries, and Chapter 7 into the recovery of crisis-affected countries.

Whereas the advantages of EWS over standard regressions are obvious (in particular, the lag at which indicators signal crises can be pinpointed), two important problems arise when the focus moves to the general assessment of a given economy. First, aggregating the information provided by the different indicators is not as straightforward as in a usual regression, and there is no clearly defined significance level to help eliminate irrelevant variables. Second, dummy variables do not provide information beyond their critical threshold. EWS does not let indicators offset each other and relies on the crucial assumption that threshold effects are very large -an assumption rejected by the data (see Berg and Pattillo, 1998, for an interesting discussion). It is in this respect symptomatic that the book, usually very clear, becomes more confused when it comes to the out-of-sample estimation, as Chapter 5 presents no less than five different aggregation procedures. In view of the large number of indicators and weighting schemes, the out-of-sample testing of EWS is rather disappointing. Predictions missed the Indonesian crisis and wrongly identified South Africa as the most vulnerable country in 1997.

Turning now to the economic content of the book properly speaking, one of the most valuable contributions of the authors is to highlight the relation between banking and currency crises, that Kaminsky and Reinhart had christened the 'twin crises' in a series of influential papers published in the 1990s. The two types of crisis differ in important aspects. Chapter 7 underlines that banking crises are more persistent -an unfortunate pattern as they are also more difficult to predict. Some omissions unfortunately counterbalance these informative insights and weaken the economic analysis, particularly when it comes to the selection of the leading indicators. For instance, some indicators are included in levels and others as a 12-month growth rate, the authors providing only a technical motivation for this difference in treatment (in fact, the rule seems to take levels for annual series and 12-month growth rates