Genetic programming for combining directional changes indicators in international stock markets

The majority of algorithmic trading studies use data under fixed physical time intervals, such as daily closing prices, which makes the flow of time discontinuous. An alternative approach, namely directional changes (DC), is able to convert physical time interval series into event-based series and allows traders to analyse price movement in a novel way. Previous work on DC has focused on proposing new DC-based indicators, similar to indicators derived from technical analysis. However, very little work has been done in combining these indicators under a trading strategy. Meanwhile, genetic programming (GP) has also demonstrated competitiveness in algorithmic trading, but the performance of GP under the DC framework remains largely unexplored. In this paper, we present a novel GP that uses DC-based indicators to form trading strategies, namely GP-DC. We evaluate the cumulative return, rate of return, risk, and Sharpe ratio of the GP-DC trading strategies under 33 datasets from 3 international stock markets, and we compare the GP's performance to strategies derived under physical time, namely GP-PT, and also to a buy and hold trading strategy. Our results show that the GP-DC is able to outperform both GP-PT and the buy and hold strategy, making DC-based trading strategies a powerful complementary approach for algorithmic trading.