How Do Neural Networks Enhance the Predictability of Central European Stock Returns?

In this paper, the author applies neural networks as nonparametric and nonlinear methods to Central European (Czech, Polish, Hungarian, and German) stock market returns modeling. In the first part, he presents the intuition of neural networks and also discusses statistical methods for comparing predictive accuracy, as well as economic significance measures. In the empirical tests, he uses data on the daily and weekly returns of the PX-50, BUX, WIG, and DAX stock exchange indices for the 2000–2006 period. He finds neural networks to have a significantly lower prediction error than the classical models for the daily DAX series and the weekly PX-50 and BUX series. The author also achieves economic significance of the predictions for both the daily and weekly PX-50, BUX, and DAX, with a 60% prediction accuracy.