Wavelet analysis of intraday share prices

Abstract

This research tested whether wavelet based algorithms can improve the performance of intraday share trading algorithms. The trading algorithms investigated, each consisted of two parts: the first part performed share price prediction and the second part traded based on the prediction. All the trades in the shares BTI, MTN, NPN and SBK through 2013 on the JSE with the associated time stamps, transaction share prices and volumes, served as the basic sample. The sample was further reduced by using end-of-interval transaction share prices at intervals of one, two, five and ten minutes throughout the trade days. Three types of prediction algorithms were employed: auto regressive moving average (ARMA), wavelet-ARMA and wavelet regressive algorithms. The wavelet based algorithms were further broken down by using up to six different levels of scales in each of the algorithms. These algorithms were fitted using the first half year of data while the tests were conducted on the second half year of data. Two trade algorithms were created by the researcher: One algorithm for buyand- sell and another for short-and-close. Both algorithms used the predicted share price one and two intervals ahead as input and took transaction cost into account. The trade algorithms entered the market daily after opening time and exited the market before closing time. The wavelet based algorithms were not found to improve the accuracy of share price prediction. However, in agreement with previous research, wavelet based algorithms were found to improve the accuracy of predicting the direction of the share prices. The wavelet based algorithms were also found to improve trading performance. Short-and-close algorithms outperformed buy-and-sell. None of the intraday trade algorithms were found to outperform buy-and-hold over the test period. This study contributes to academic research regarding the manner in which wavelet based and ARMA algorithms were combined, the application of a wavelet-regressive prediction method to financial time series and the application of wavelet based trading algorithms on an intraday time scale.