Efficient local techniques in financial time series forecasting
D.A. Karras, B.G. Mertzios, R. C. Papademetriou · 2000
Several improvements for the local reconstruction method in time series prediction are developed and presented in this paper. More specifically, it is suggested that by augmenting the state vectors with informative features coming from second order information involving the topology of their neighbouring state vectors then, significantly better results could be obtained with respect to time series reconstruction. Also, the weighting methodology in the least square fitting procedure has been modified aiming at eliminating outliers. The validity of these new ideas is investigated by applying them to the reconstruction task of a time series representing stock price evolution.