Currency crisis : Early warning systems

Abstract

Recent currency turmoil has brought currency crisis models back into spotlight. Unlike Asian Crisis, this time around the interest is not limited to explaining but rather in predicting crisis. Early warning systems serve this need to observe behavior of certain macroeconomic variables and determine the probability or chances of a currency crisis occurring in near future. In this research study, we have primarily looked at developing EWS for Indian Economy. This is particularly relevant because though there has been lot of academic interest in Latin American and South Asian economies, very few studies have been conducted for Indian Economy. One of the main reasons seems to be paucity of data. Our research was also affected by this and we have suggest-ed certain structural changes that can be incorporated by Indian agencies to increase transparency as well as initiate more academic interest in India. Developing EWS for any economy involves two stages, defining crisis and detecting the same. We have looked at several approaches in use for both these stages. For Indian context, we de-cided to use extreme value theory (EVT) for defining crisis. This is based on Hill estimate and is more relevant here because other measures were too strict for an economy like India. Specifically, we focus on defining crisis as successful attacks only because otherwise, too much noise enters the system. Since our data is limited we have tried to minimize noise. We have reduced the set of possible macroeconomic indicators to five main categories – cur-rent account, capital account, financial sector, domestic sector and global sector. Between these five categories most of the variables are covered. We use two different approaches for determining the appropriate function of these variables. First is based on determining thresholds for each variables beyond which the variable is presumed to give a warning signal. These thresholds are based on mini-mizing NSR of the index. Second, we determine the function of variables using regression, separately as well as in combination with other categories. From each category we have found acceptable indica-tors and even with the data restrictions, the model seems to predict crisis with acceptable accuracy. We conclude by summarizing the indicator variables chosen and their relative importance in the mod-el. We also look at possible future direction to this research and areas of improvement in model as well as implementation.