Statistical Arbitrage Using Cointegration and Principal Component Analysis Approach

Abstract

Two approaches to model-driven statistical arbitrage in the most liquid equities tradable on the NYSE and NASDAQ are studied in the article. Cointegration and PCA analysis were used in the research. In both strategies, we are developing contrarian trading signals, which then are back-tested through period of 2016–2020 with 4-h frequency and 2019–2020 with 1-h frequency data. Back-testing of the strategies through growing market of 2016–2019, and global sell-off at beginning of 2020 is the main contribution of the research. Both strategies demonstrated consistent returns, with lower than the broad market drawdowns. While Sharpe ratio of the strategies may be considered quite low at 0.49 for PCA-based approach, it outperformed both long- and short-only proxies. Lower than broad market downturn through sell-off period, and consistent median returns for both strategies were observed.