Smooth Component Analysis and MSE Decomposition for Ensemble Methods

This paper is addressed to economic problems for which many different solutions (models) can be proposed. In such situation the ensemble approach is a natural way to improve the final prediction results. In particular, we present the method for the prediction improvement in multi-agent environment based on the multivariate decompositions. As a decomposition method we present the smooth component analysis. The resulting components are classified as destructive and then removed or as constructive and then recomposed to create final forecast. The classification of the components is based on the theoretical analysis of MSE error measure. The robustness of the method was validated based on energy load data from Polish power system.