Can thewavelet-kernelmethodology improve other kernel techniques?
Mónica Gago, Eusebio Juaristi · WIT transactions on information and communication technologies · 2008
Our aim in this paper is to compare different ways of forecasting using the wavelet transform and the kernel regression.We consider that working with block (or segment) of data is richer than working with individual data (as in traditional kernel), as we assume there is some kind of pattern inside each block which will improve the estimation, and therefore the prediction.We choose the wavelet transform because this transform is able to separate components of data in different locations and with different location in time and frequency.To test the performance of the different methodologies we have carried out a Monte Carlo Study in which we have compared the four methodologies: Ordinary Least Square (OLS), Traditional Kernel (TK), Block Kernel (BK) and Wavelet-Kernel (WK).Two real life applications have been realized.On the one hand, volatility smile has been forecast and on the other, the rated temperature of the steel coils' furnace is predicted.Surprisingly contradictory results had been obtained.