Optimal mean-reversion arbitrage with machine learning and advanced hedge ratios
No Thumbnail Available
URL
Journal Title
Journal ISSN
Volume Title
School of Business |
Master's thesis
Authors
Date
2024
Department
Major/Subject
Mcode
Degree programme
Finance
Language
en
Pages
79+13
Series
Abstract
This thesis proposes a statistical mean-reversion arbitrage strategy, commonly known as pairs trading, and evaluates its performance through a robust backtesting framework. The primary objectives are to investigate the efficacy of unsupervised machine learning techniques in identifying suitable pairs of securities and to determine the optimal hedge ratio that maximizes trading performance. The study is confined to equity trading, focusing exclusively on the constituents of the S&P 500 index, with an analysis based on historical daily price data spanning from January 2018 to December 2023. The findings underscore the effectiveness of unsupervised machine learning methodologies in the pair formation process. Specifically, Principal Component Analysis (PCA) and Ordering Points to Identify the Clustering Structure (OPTICS), when applied in conjunction with rigorous cointegration testing and multiple mean-reversion criteria, successfully identify pairs that exhibit stable cointegrating relationships and strong mean-reverting dynamics. Notably, the majority of selected pairs are cross-sector, which underscores the capability of machine learning techniques to uncover latent, non-intuitive cross-sector relationships that are not immediately discernible to market participants. In the context of hedge ratio estimation, this study implements and evaluates six distinct methods, ranging from conventional approaches such as Ordinary Least Squares (OLS) and Total Least Squares (TLS) to more advanced methodologies, including Johansen cointegration, Box-Tiao Canonical Decomposition, Minimum Half-Life, and Minimum Augmented Dickey-Fuller (ADF). Empirical results indicate that hedge ratios derived from the Minimum Half-Life and Box-Tiao methods consistently outperform those estimated by other techniques, generating statistically significant alphas and yielding superior trading performance across various evaluation metrics. To enhance the rigor and validity of future research, this study advocates for the incorporation of stratified sampling techniques to ensure an equitable representation of intra- and cross-sector pairs, thereby facilitating a more comprehensive comparative analysis. Additionally, refinements to the backtesting framework, including the adoption of systematic methodologies for selecting rolling window sizes and the integration of transaction frictions such as slippage, are recommended to bridge the gap between simulated and real-world trading conditions. This research contributes to the field of quantitative finance by advancing the application of machine learning in pairs trading and providing a comparative analysis of hedge ratio estimation techniques. The findings have significant implications for the development of more robust, generalizable, and effective trading strategies.Description
Thesis advisor
Nyberg, PeterKeywords
pairs trading, mean-reversion, machine learning, optimal hedge ratio