شاخص های پیشرو اقتصادی بحران ارز برای کشورهای در حال ظهور

This study identifies common features of currency crises in 15 emerging countries using quarterly data over the period 1980–1998. We use Fisher's linear discriminant analysis in order to build an early-warning system. Capital control and contagion dummies, as well as an indicator for problems in the banking sector, are included in the set of explanatory variables; the overvaluation of currencies is assessed by real effective exchange rates. We propose a ‘balancing’ approach to deal with the trade-off between good classifications and false alarms. The model yields a relatively good—and unbiased—ratio of correct predictions: four out of five crises are correctly predicted and only one out of five non-crises is predicted as a crisis.

Recent years have seen an increase in currency crises affecting a large number of countries, either directly or indirectly. Some similarities are generally observed in the way these crises unfold: a loss of foreign exchange reserves, capital outflow and a sudden depreciation of the currency. If similarities between these economies also exist just before currency crises occur, they could be used to predict these crises. It is certainly true that each crisis is specific: the debt crises in Latin America of the 1980s are different from the Mexican crisis in 1994, which differs from the crises of Southeast Asia in 1997, and even within each of these crisis episodes, every crisis presents some specificity. However, this does not preclude the existence of common features. Theoretical studies have identified several types of crises Žfor a recent survey, see Jeanne, 1999.. ‘Speculative attack models’ of currency crises stress the importance of fundamental factors in the triggering of crises Že.g. Krugman, 1979.. In this type of model, speculators’ behavior is not the source of the crisis; they only accelerate the process. ‘Escape clause models’ introduce strategic considerations based on the consequences of the defense of a peg in speculative attacks Že.g. Obstfeld, 1994.. Here, expectations of crises may be self-fulfilling and there is a possibility of multiple equilibria. Therefore, crises may occur without a worsening of fundamentals and without inconsistent policies. Other recent papers have emphasized the role of Žreal. contagion in the triggering of crises Že.g. Gerlach and Smets, 1995., the role of moral hazard, especially after the Asian crisis Že.g. Krugman, 1998; Chinn et al., 1999; Corsetti et al., 2000., or insurance motives ŽDooley, 2000.. Theoretical work provides some guidance when choosing potential leading indicators, which should reflect fundamentals as well as any variable able to influence the market expectations. However, it does not allow discrimination between competing indicators, nor does it enable their respective weights to be determined. That is why an abundant empirical literature has developed in parallel. The existing studies Že.g. Frankel and Rose, 1996; Sachs et al., 1996; Kaminsky et al., 1997; Radelet and Sachs, 1998; Esquivel and Larra´ın, 1998; Milesi-Ferretti and Razin, 1998; Schnatz 1998; see Goldstein et al., 2000 for an overview. can be classified according to different criteria: the countries analyzed Ždeveloping, emerg- ing, industrialized., the periodicity of the time series Žmonthly, annual., the crisis indicator used Žexchange rate pressure, actual devaluation., the time horizon of the prediction Žseveral months, 1 year, 2 years. and the methods implemented Žsignal approach, LogitProbit, structural VAR2 .. These criteria are not always independent. For example, in the case of developing and emerging countries, an actual devaluation seems to be a reasonable concept, whereas it is more difficult to justify using this criterion for industrialized countries.This paper constructs leading indicators of currency crises for a set of 15 emerging countries in the period 1980:Q1
1998:Q4 with the aim of predicting looming currency crises in probabilistic terms. The rest of the paper is organized as follows. Section 2 describes the main specific features of this work, compared to other early-warning systems ŽEWSs. in the literature. Section 3 is devoted to selecting a proper definition of a crisis, which implies the construction of several simultaneous crisis indicators. Section 4 deals with methodological issues concerning discriminant analysis and presents the data. Results are given in Section 5. Several scores are proposed, according to the sample retained. Constructing score functions by region improves upon the ‘global score’ obtained on the basis of the whole population and allows a ‘synthetic score’ to be obtained. Section 6 compares our results to those of other studies, and Section 7 concludes.

This study has aimed at constructing early warning indicators able to predict looming currency crises in probabilistic terms for a set of 15 emerging countries over the period 1980:Q1
1998:Q4. As in other EWSs, we had to face a trade-off between a high ratio of good classifications of crises and a low ratio of false alarms. We characterized this trade-off by Bayes’ formula and proposed a ‘balancing’ approach to deal with it. We also addressed the issue of the behavior of leading indicators as they approach crises. The score function calculated here has satisfactory properties in this sense. A short prediction horizon Žone quarter. gives neither different nor better results than a four-quarter horizon. Contagion, overvaluation of currencies and problems in the banking sector have been taken into account in the model. The currency overvaluation has been measured by the real effective exchange rate, which seems a more accurate proxy than the bilateral rates usually observed in the literature. Finally, the study is carried out using Fisher’s linear discriminant analysis, whereas the existing literature relies on the signal approach or on LogitProbit analysis. Our results seem to show that there are some fundamental determinants of crises. Overvaluation caused by long periods of pegs increases the probability of crisis; the same is true for short-term debt and inflation. We also evidenced that losses in foreign reserves may constitute leading indicators, for they are detectable some quarters before the crisis. This result is consistent with fundamental models following Krugman, as well as with ‘speculative attack models’ featuring self-fulfilling crises, the contingent liabilitymoral hazard argument or a model relying on insurance motives. Our regional scores exhibit specific leading indicators Žopenness for Latin American countries, banking problems for Asian countries.. It is sometimes argued that the use of EWSs can create feedback effects. A statistical regularity observed may tend to collapse once pressure is placed on it for control purposes. This is, however, likely to be less the case when the predictions remain private information. We might also argue that if crises predictions of EWSs are invalidated because they are published, that might not necessarily be one of their worst achievements.