V The Implementation of an Early Warning System

Catherine Pattillo, Andrew Berg, Gian Milesi-Ferretti, and Eduardo Borensztein
Published Date:
January 2000
  • ShareShare
Show Summary Details

While there has been a recent flurry of research on the prediction of currency crises, it is clear that more work needs to be done to better understand their causes and more successfully predict them. The empirical literature on early warning systems is relatively young, at least for models with an explicit forecasting objective, especially when compared with, say, leading indicators of business cycles or largescale macroeconomic models that have been in development for several decades. Many ideas have not yet been explored satisfactorily, including basic issues such as the empirical definition of a crisis. It may also be possible to incorporate new variables that seem to be relevant but have not previously been available in the desired format or frequency, such as measures of nonperforming loans in the banking sector, more refined measures of short-term liabilities (including valuations of the forward and contingent positions of the monetary authorities, for example), and even the budget deficit. An important feature of currency crises that deserves further study is contagion, which is difficult to incorporate in a satisfactory way but which undoubtably plays a role at least in the timing of crises. Other issues that should be studied include institutional factors such as the strength of regulatory frameworks, corporate governance, the degree of openness of the capital account, and variables accounting for political developments. Finally, various aspects of the specification of models could be further studied to improve their reliability, including by reconsidering the most appropriate forecast horizon.

The work done to date and reviewed in Section IV nonetheless suggests that the implementation of a quantitative early warning system might be a fruitful endeavor. Early warning systems contain significant useful information and can produce estimates of the probability of crisis and rankings of the susceptibility to crisis of different countries. In particular, Section IV demonstrated that some of the models can forecast with at least a modest degree of success the timing of crises, and that they do better still at predicting the cross-country vulnerability to crisis. That is, they may help predict which country is more likely to suffer a crisis during a given period, even if they do less well at predicting how many crises there will be in that period.

A central fact relating to the use of early warning systems is that their accuracy is of necessity highly imperfect. Sections II and IV showed that while early warning systems help identify patterns common to many crises, they cannot be expected to predict with complete, or even high, accuracy. While further research can be expected to improve performance, many of the reasons for inaccuracy, detailed above, will remain. In particular, the timing of crises may be impossible to forecast as it may depend on changes in market sentiment and other inherently unpredictable events. Moreover, even a system that is highly successful from a statistical point of view would only address part of the relevant information regarding the prospects for a crisis. The analysis of the vulnerability of an economy involves many more aspects than can be summarized in a handful of generally available variables.

It should also be recognized that such an early warning system framework presumes that differences in economic structure, institutions, and so on across countries are not so important as to negate the effectiveness of a cross-country approach. However, key determinants of the likelihood of crisis may not be the same across countries, or their relative impact may differ. And the relevant factors may change over time as the level of economic development and market structures change. This suggests the use of a fairly homogeneous group of countries and sample period in the design of an early warning system to minimize such problems. For instance, these models are likely to apply better to the group of “emerging” economies that have been more active in international capital markets over the past few years.

To conclude, early warning systems should be used but not abused. This means that the primary use of such a system should be as only one tool among many others for analysis of external risks. Thus, any utilization of early warning systems must be embedded in broader analysis that takes into account all the important complexities, some of which a cross-country statistical model inevitably must ignore.

    Other Resources Citing This Publication