A WAVELET BASED APPROACH FOR MULTI-SCALE ANALYSIS AND PREDICTABILITY OF STOCK RETURNS.

The aim of this paper is to demonstrate the effectiveness of alternative methods from the time-frequency domain in analyzing the behavior of financial time series. Multiresolution analysis of the discrete wavelet transform class is used to transform the BSE returns series into different sets of coefficients where each set contains coefficients that provide information corresponding to a particular time-scale resolution. The main coefficient series from the transformed time-series is used to make forecasts and the result is then compared with the result when forecasting is done directly using the original returns data. On the analyzed returns data we proved that forecasting using the wavelet transformed algorithm provides much accurate results as compared to forecasting applied directly to the original returns series The effectiveness of a wavelet based cross-correlation technique in analyzing the relation between two markets at different levels of time-frequency resolution is also demonstrated. This approach of multi-scale decomposition of a time-series using wavelet methodology allows us to detect changes in stock market behavior from a time-scale perspective where the data can be analyzed at different time horizons and frequencies simultaneously.