Leveraging the explainability of associative classifiers to support quantitative stock trading

Abstract

Forecasting the stock market is particularly challenging due to the presence of a variety of inter-related economic and political factors. In recent years, the application of Machine Learning algorithms in quantitative stock trading systems has become established, as it enables a data-driven approach to investing in the financial markets. However, most professional traders still look for an explanation of automatically generated signals to verify their adherence to technical and fundamental rules. This paper presents an explainable approach to stock trading. It investigates the use of classification rules, which represent reliable associations between a set of discrete indicator values and the target class, to address next-day stock price prediction. Adopting associative classifiers in short-term stock trading not only provides reliable signals but also allows domain experts to understand the rationale behind signal generation. The backtesting of a state-of-the-art associative classifier, relying on a lazy pruning strategy, has shown promising performance in terms of equity appreciation and robustness of the trading system to market drawdowns.