Dynamics of predictability and variable influences identified in financial data using sliding window machine learning

In this paper we analyze the dynamics of the predictability and variable interactions in financial data of the years 2007–2014. Using a sliding window approach, we have generated mathematical prediction models for various financial parameters using other available parameters in this data set. For each variable we identify the relevance of other variables with respect to prediction modeling. By applying sliding window machine learning we observe that changes of the predictability of financial variables as well as of influence factors can be identified by comparing modeling results generated for different periods of the last 8 years. We see changes of relationships and the predictability of financial variables over the last years, which corresponds to the fact that relationships and dynamics in the financial sector have changed significantly over the last decade. Still, our results show that the predictability has not decreased for all financial variables, indeed in numerous cases the prediction quality has even improved.