Pair Trading Strategy Based on Machine Learning and Deep Learning Models

Pair trading, a statistical arbitrage strategy, exploits the cointegration of stock pairs to generate market-neutral returns. We select cointegrated pairs from the 30 Dow Jones Industrial Average (DJIA) constituents (2021–2024) using the Engle-Granger Two-Step Method with a p-value threshold of ≤0.05. Quarterly spread predictions are generated using machine learning (ML) and deep learning (DL) models. Trading signals are based on z-score thresholds derived from the spread’s moving average. Walk-Forward Backtesting evaluates performance, comparing the ML/DL strategy against a baseline OLS model, Buy-and-Hold, and Equal Weight strategies. The results showed that the ML/DL strategy performed well, with XGBoost achieving the highest average Sharpe Ratio and LSTM showing the lowest test set RMSE and MAE, reducing trading risk and improving the efficiency of paired trading.