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.
