New methodological approaches to the construction of currency crashes models

The large financial crises in the last decade have increased the interest of many economists in searching for some indicators, which can predict speculative attacks on currencies. Most of these studies concern on emerging economies, because they are more vulnerable to such speculative attacks. This paper points out main methodological issues of two standard approaches to the construction of the early warning system - the signal approach and the regression probit or logit model approach. Based on avoiding these issues alternative approaches have been evolved. Three of them - two regimes model, VAR model and Markov-switching model - are described in the next part of this paper. The comparison of predictive power shows the importance of the construction of country specific early warning systems.