Optimized Pair Trading Strategy using Unsupervised Machine Learning

Pair trading is a popular trading strategy that involves taking advantage of the correlation between two stocks to generate profits. However, traditional pair trading strategies have limitations, such as the difficulty in identifying the most suitable pairs of stocks, determining the optimal trading parameters and large computational cost. To address these challenges, this paper proposes an approach that leverages unsupervised machine learning techniques to improve computation time and identify the best pairs of stocks. The presented model uses PCA and DBSCAN algorithm for dimensionality reduction and clustering. Furthermore, the paper attempts to optimize the pair selection criteria using Engle-Granger test and the trading algorithm using moving averages. The proposed strategy is then tested on historical National Stock Exchange data and the results demonstrate great performance of the proposed approach with respect to the NIFTY-50 Global benchmark.