Genetic Algorithms and Financial Crises in Emerging Markets

Abstract

indicators (real exchange rates, reserves over M2, imports, short-term debt level...) can be very useful to understand and quantify country risk levels, but are not sufficient. Sudden changes in indicators or combinations of indicators levels can induce a higher risk than what a simple linear model could measure and non-linear models using thresholds, like SETAR or EXPAR, are difficult to use with poor data and require a lot of observations. That's why we decided to create a tool that would help us to make a better diagnosis on country risk without an equation or a thresholds model. We called this indicator the Vulnerability, which is measured as a distance to an optimal combination computed as a risky pattern by a genetic algorithm regularly updated with new official data.

Keywords: Country risk, developing countries, genetic algorithms

JEL Classification: D81, C14, C45, C49