سیستم های هشدار دهنده زود هنگام برای بحران ارز: رویکرد ارزش شدید چندمتغیره

We apply extreme value theory to assess the tail dependence between three currency crisis measures and 18 economic indicators commonly used for predicting crises. In our pooled sample of 46 countries in the period 1974–2008, we find that nearly all pairs of variables are asymptotically independent: in the limit, extreme values of economic indicators are not followed by severe currency crashes. Our findings may explain the poor performance of existing early warning systems for currency crises. However, we do find that economic variables with stronger extremal association with the exchange rate have better crisis prediction performance, both in-sample and out-of-sample.

Currency crises can have a devastating impact on the real economy, as shown by the Mexican peso crisis in 1994–1995, the Asian crisis in 1997–1998, and the Argentinean crisis in 2002.1 To prevent future crises researchers have since tried to identify common factors underlying exchange rate instability,2 and to build early warning systems.3 However, existing early warning models fail to predict crises out of sample (Furman and Stiglitz, 1998; Berg and Pattillo, 1999; Berg et al., 2005), despite impressive in-sample results. In this paper we test whether currency crises and extreme movements in lagged economic and financial variables are truly linked in the tail, using multivariate extreme value theory (EVT). Further, we investigate whether EVT measures can provide useful information for building an early warning system. Our motivation for applying EVT is that currency crises are by definition infrequent extreme events that are more suited for specialized techniques that focus exclusively on rare tail events. The poor out-of-sample results of existing early warning systems also justify an investigation of alternative methodologies such as EVT.4 Our dataset consists of monthly exchange rates and 18 economic indicators, commonly used in the crisis prediction literature, for 46 countries during the period 1974–2008. We assess the asymptotic dependence of three currency crisis measures and the lagged indicators, using non-parametric measures developed by Poon et al. (2004). Asymptotic dependence means that in the limit, as the condition of the economic fundamental deteriorates and the variable moves deeper into the tail, more extreme currency depreciations (or devaluations) tend to follow with positive probability. Out of 54 pairs of variables, we find only two pairs that are asymptotically dependent: increases in the real interest rate and the real interest rate differential are asymptotically dependent with currency crashes. The other pairs are asymptotically independent, which means that the relation disappears as we move deeper into the tail. Apart from testing asymptotic dependence, we also estimate an EVT measure that can be interpreted as the extremal association in the tail between two variables. We find that the association in the tail between currency crisis measures and lagged economic fundamentals is stronger than suggested by a standard Pearson correlation measure. Further, we find a strong positive link between our extremal association measure and the in-sample crisis prediction success of an economic indicator. In an out-of-sample period (emerging markets, 1995–2008), we compare crisis prediction performance with standard approaches from the literature, such as a probit model and the signaling approach. Out-of-sample crisis prediction performance is poor for all methods. However, in contrast to other methods, the EVT approach foreshadowed this poor outcome based on the lack of asymptotic dependence between fundamentals and currency crises measures. In the currency crisis literature, there is no empirical work yet that applies EVT to estimate the relation between fundamentals and currency crisis measures. Koedijk et al. (1992) use EVT to study the univariate distribution of exchange rate returns. Pozo and Amuedo-Dorantes (2003) apply EVT to set thresholds for the identification of currency crises, instead of the more common approach that defines a crisis when a measure of foreign exchange market pressure is two or three standard deviations beyond its mean. Pozo and Amuedo-Dorantes (2003) find that the EVT method more accurately identifies actual crises than the conventional method. Hartmann et al. (2010) demonstrate how joint currency crises can occur through shared fundamental linkages, and they estimate the tail dependence of exchange rate pairs. Several studies apply EVT to analyze extreme dependence among financial markets, including Longin and Solnik (2001), Embrechts et al. (2002), Poon et al. (2004), Hartmann et al. (2004, 2010), Jondeau and Rockinger (2006) and Ning (2010). The literature on extreme events in financial markets confirms that most economic and financial variables are non-normally distributed and that the dependency between variables in the tail area and in the centre range can be drastically different, providing a clear motivation for our study.

In this paper we apply extreme value theory to test whether currency crises are linked with extremes in lagged economic and financial variables. Using monthly data for a cross-section of 46 countries in the period 1974–2008, we first show that nearly all economic fundamentals are asymptotically independent with currency crisis measures. Asymptotic independence means that as the values of the economic variable become more extreme, the conditional probability of more extreme currency events approaches zero in the limit, casting doubt on whether a causal link exists. There are only two variables in our dataset for which we cannot reject tail dependence with the exchange rate return: the domestic real interest rate, and the real interest rate differential relative to a reference country. Asymptotic dependence is a concept defined in limit only, and it could be too strict for crisis prediction in practice. We therefore also estimate a measure that can be interpreted as the association between two variables in the tail. We assess the value of this EVT measure for predicting currency crises. We find that in most cases the extremal association between currency crashes and lagged fundamentals is higher than a simple correlation measure suggests. Moreover, the extremal association measure has a strong link (r = 0.92) with the in-sample crisis prediction performance of economic indicators. We also evaluate the crisis prediction performance of indicators and competing models in an out-of-sample assessment period (1995–2008) in 33 emerging markets. All indicators and models perform poorly out of sample, and much worse than in sample. This especially holds true for approaches that maximize in-sample model fit, like a probit model. In contrast, our non-parametric EVT approach clearly indicates that currency crisis measures and most economic fundamentals are asymptotically independent in sample, in line with the poor out-of-sample results. Moreover, the two real interest rate indicators selected by the EVT method perform better out of sample than other models. Our results show that positive extremes of the two real interest rate variables are potentially useful as warning signals for authorities and risk managers. In the literature, high real interest rates are considered as symptoms of weak economic and financial conditions. A high real interest rate on deposits is associated with issues such as overlending cycles, financial sector problems, liquidity crunch, and a potential cause of future recessions. A high domestic-foreign real interest rate differential captures a heightened risk premium for holding domestic currency assets and is a potential cause of economic slowdown, bank fragility and the burst of asset price bubbles.