Predicting financial time series: An approach with Support Vector Machines
A classical statistical model for inference and prediction of financial time series is the VARMA model. This model requires that the order of the process is known beforehand or can be estimated from the data. Furthermore an assumption for the distribution of the residuals has to be made. To reduce the number of assumptions one has to make, one can use Support Vector Regression (SVR) instead.
In this project a parallel version of the SVR is implemented. Furthermore an empirical study to compare the goodness of prediction for the two above mentioned methods is carried out. The data used is the stock prices for the stocks that are currently in the DAX30, their respective returns and their respective squared returns. We find that the SVR gives a more accurate prediction in terms of the mean square error.
- KONWIHR funding: two months during Multicore-Software-Initiative 2009/2010
- Prof. Dr. Ingo Klein, Lehrstuhl für Statistik und Ökonometrie, Uni-Erlangen/Nürnberg