طراحی یک سیستم هشدار دهنده برای بحران بدهی

This paper develops an early warning system for debt crises broadly defined as episodes of outright default or failure of a country to be current on external obligations. A multinomial model is applied, that allows to differentiate between three regimes: a ‘tranquil’ period, a ‘crisis’ state and an adjustment or ‘post-crisis’ phase. The model, estimated using a large set of macroeconomic variables, is able to predict 76% of entries into crisis while sending 36% of false alarms and has rather good out-of-sample performance. Finally the paper tries to integrate the analysis based on macroeconomic variables with the approach based on risky market instruments.

Emerging market economies have experienced crises, driven by their poorly developed financial systems, volatility of macroeconomic policies, weak banking sector, high dependence on external capital flows and uncertain growth prospects; such crises had disruptive effects on these economies. For this reason, both the academic and the official sectors have felt the importance of developing models that cannot only identify weaknesses and vulnerabilities in emerging market economies, but also send timely and correct signals about the onset of a financial crisis, the so-called early warning systems (EWS). Most of the EWS models developed so far have tried to signal the onset of currency and banking crises, both individually or jointly determined (so-called twin-crises). The seminal papers in the field are those written by Kaminsky et al. (1998) and by Frankel and Rose (1996), whereas a thorough review of such category of models can be found in Berg et al. (2004). Until now, however, little work has been done on ‘debt’ crises, which can be broadly defined as episodes of default or failure to be current on external obligations. The time span from 1994 onwards has been characterized by large sovereign and corporate defaults, or debt servicing difficulties on foreign currency-denominated bonds. Currency and debt crises might be generated by common factors, such as unfavorable macroeconomic developments, a deterioration in external financing conditions (e.g. a sudden reduction in capital flows or a sharp upsurge in their cost) or an increase in the extent of international investors' risk aversion. Notwithstanding the previous considerations, currency and debt crises do remain quite distinct events, in fact: 1) the two types of crises are not perfectly correlated, as shown in Sy (2003), in the sense that it is possible to have a currency crisis which is not associated with a debt crisis, as it is conceivable that a country may fall into arrears or default on its external debt without any major disruption in the exchange rate, as happened in Pakistan in 1999; 2) it is not clear what the causal relationship should be: in fact, one could expect a sharp depreciation of the exchange rate as a response to an excessively high growth of the external debt or a rapidly worsening scenario for the country's financing needs; under such a scenario, investors might not trust the government's ability to face its external obligations and therefore start selling off assets denominated in that particular currency. The literature on the empirical determinants of a debt crisis is quite small compared with the large body of theoretical and empirical work on currency and banking crises. A broad classification is between models which are based on the evolution of a particular set of macroeconomic variables leading to the build-up of a crisis, and models that extract information from financial data and market prices for widely traded financial instruments such as sovereign bonds or, more recently, credit default swaps (CDSs). As recognized by the IMF (2002) itself, any macro-based model should be complemented with information on market expectations extracted from bond spreads as well as from CDSs. A detailed description of the aforementioned categories of EWS for debt crises can be found in Ciarlone and Trebeschi (2004). The rest of this paper is organized as follows: Section 2 reports the definition of debt crisis, the data set and the event study analysis; Section 3 presents the two econometric specifications, the binomial and the multinomial logit, both based on macro variables; Section 4 is devoted to the analysis of CDSs spreads with the aim to extract early warning signals. Section 5 concludes.

The paper focuses on the occurrence of a particular kind of financial crisis, determined by increasingly unsustainable levels of public and/or external debt, which have grown in importance in the very recent past, taking over currency crises as a major source of concern in emerging countries. After reviewing the relatively small academic literature on early warning systems for debt crises, we have drawn the conclusion that these episodes may take a variety of different forms, which range from outright default to more general debt-servicing difficulties. This consideration has led us to construct a debt crisis indicator able to account for such different forms. This indicator has been used to evaluate a series of debt crisis episodes experienced, in the period 1980–2002, by a relatively large sample of emerging market economies, characterised by significant access to international capital markets. The next step was to discover which macroeconomic factors might be at the roots of a debt crisis. We have first applied a classic binomial logit approach, according to which the variables that are significant in explaining debt crisis episodes are mainly those that measure the burden of external indebtedness and foreign-currency generating capabilities of a country. We then improve upon the out-of-sample predictive ability, the most important aspect one has to take into account in order to evaluate an early warning model, performing a multinomial logit analysis, characterised by the existence of three, rather than two, regimes. The main flaw of both the previous specifications is the high number of false alarms launched. As a further step, we try to integrate the analysis based on macroeconomic variables with the approach based on risky market instruments. A model that combines various technical aspects from the existing theoretical literature has been developed in order to estimate the one year forward probability of default and the recovery rate from CDSs quotes for Argentina, Brazil, Colombia and Mexico. We subsequently have worked out a simple rule to obtain crisis signals from the default parameters that produced better results, in terms of the percentage of episodes correctly identified and in terms of false alarms, with respect to the binomial and multinomial logit approach. We concluded suggesting that an ideal EWS should be able to integrate the two aspects. In this respect, timely information on the relevant macroeconomic variables are needed along with reliable and robust market data. With more frequent macroeconomic data, and a wider set of countries for which financial data are available, further research could be carried out with the aim of better integrating the two aspects of an EWS for debt crises.