Quality of forecasting based on compressed high frequency time series

Abstract

In this paper the general compression method of time series will be presented and adapted to financial time series analysis where dimensionality reduction is crucial. It will be shown that a double compression using Daubechies 4 wavelet does not significantly affect the quality of information carried by a time series. The reduction of dimensionality significantly affects the algorithmic complexity and improves its quality of prediction. In order to verify this hypothesis the highly frequent time series will be evaluated in terms of forecasting quality where future value is predicted only on the basis of the past quotations. In this project as a predictive algorithm ARAR will be applied due to its good results in forecasting of the real financial time series. © 2012 Springer-Verlag.