Time Series Analysis: Application in pair trading

Published on May 14, 2023. How can the process of pair selection be enhanced through the integration of machine learning techniques with statistical methods? In this study, we aim to identify pairs of stationary stocks in the S&P 500 universe by applying statistical tests and clustering methods with machine learning techniques. Our research compares various algorithmic techniques such as NSGA-II, Soft DTW, GMM, and DBSCAN for clustering, with filtering of stationary stocks using ADF Value and Hurst Exponent. Furthermore, we examine whether utilizing these filtering techniques leads to an improved trading performance when employing the Kalman spread strategy and explore the correlation between the obtained results and stationary metrics.