Forecasting financial time series with support vector machines based on dynamic kernels

The technical analysis of financial time series and in particular the prediction of future developments is a challenging problem that has been addressed by many researchers and practitioners due to the possible profit. We provide a forecasting technique based on a certain machine learning paradigm, namely support vector machines (SVM). SVM gained more and more importance for practical applications in the past years as they have excellent generalization abilities due to the principle of structural risk minimization. However, standard kernel functions for SVM are not able to compare time series of variable length appropriately, i.e., when we assume that these time series must be scaled in a non-linear way. Therefore, we use the dynamic time warping (DTW) technique as a kernel function. We demonstrate for two financial time series (FDAX and FGBL futures) that excellent results can be obtained with this approach.