An In-Depth Investigation of Genetic Programming Under Physical Time and Directional Change Frameworks for Algorithmic Trading

Abstract

The majority of algorithmic trading studies rely on datasets with fixed physical time intervals, such as hourly or daily data, resulting in a discontinuous representation of time. Directional Changes (DC) is an alternative method, which transforms these datasets from physical time (PT) intervals into event-driven sequences, allowing for unique price analysis. We introduce two novel algorithms based on Genetic Programming (GP): GP-DC, which only uses DC-based indicators as features, and GP-DC-PT, which combines DC-based indicators with physical-time indicators from technical analysis to create trading strategies. We extensively evaluate the return and risk of DC-based trading strategies in 220 datasets from ten international financial markets for periods up to ten years. We compare the performance of these trading algorithms against a non-DC-based GP, technical analysis derived strategies, as well as the well-known benchmark of buy-and-hold. Our results demonstrate that the proposed DC-based algorithm GP-DC creates profitable trading strategies with low risk, and statistically and significantly outperforms the other benchmarks. We also find that the feature set that includes both DC and physical time indicators (GP-DC-PT) leads to further improvements in algorithmic trading performance and is very competitive, as it achieves an average total return of over 18%. These findings suggest that including directional changes in trading strategies significantly enhances their return and risk performance.