FORECASTING SMOOTHED NON-STATIONARY TIME SERIES USING GENETIC ALGORITHMS

• DOI: 10.1142/s0129183107011133
• Publication Date: 2007-08-27T09:59:35Z
• Authors: P. NOROUZZADEH; B. RAHMANI; M. S. NOROUZZADEH
• Keywords: Smoothing; Kernel (algebra); Kernel smoother

We introduce kernel smoothing method to extract the global trend of a time series and remove short scales variations fluctuations from it. A multifractal detrended fluctuation analysis (MF-DFA) shows that multifractality nature TEPIX returns is due both fatness probability density function long range correlations between them. MF-DFA results help us understand how genetic algorithm methods act. Then we utilize recently developed for carrying out successful forecasts in financial deriving functional form Tehran price index (TEPIX) best approximates variability The final model mainly dominated by linear relationship with most recent past value, while contributions nonlinear terms total forecasting performance are rather small.