Experiments in predicting the German stock index DAX with density estimating neural networks

We compare the performance of multilayer perceptrons and density estimating neural networks in the task of forecasting the return and the volatility of the DAX index. We claim that for nontrivial target distributions, density estimating networks should lead to improved predictions. The reason is that the latter are capable of embodying more complex probability models for the target noise. We discuss appropriate distribution assumptions for the important cases of outliers and non constant variances, and give interpretations of the new estimates in regression theory.