A wavelet neural network application to predict financial market crises

Forecasting currency crisis is an important financial problem that has received much attention especially because of its intrinsic difficulty and practical applications. Some non-parameter models, such as KLR's single analysis method, have been proposed for obtaining accurate prediction results, in an attempt to remedy the performance of parameter models, such as Probit/Logit method. This paper develops a new approach for predicting currency crisis, namely Wavelet Neural Networks (WNN) model. Using same indicators and a set of data, we analyze the probability of crisis by this approach. According to result, both models are able to signal currency crises reasonably well in-sample, and that the forecasting power of WNN out-of-sample has better performance than ANN.