Exploring Statistical Arbitrage Opportunities Using Machine Learning Strategy

Abstract

Arbitrage opportunity exploration is important to ensure the profitability of statistical arbitrage. Prior studies that concentrate on cointegration model and other predictive models suffer from various problems in both prediction and transaction. To prevent these problems, we propose a novel strategy based on machine learning to explore arbitrage opportunities and further predict whether they will make a profit or not. The experiment is conducted in the context of Chinese financial markets with high-frequency data of CSI 300 exchange traded fund (ETF) and CSI 300 index futures (IF) from 2012 to 2020. We find that machine learning strategy can explore more arbitrage opportunities with lower risks, which outperforms cointegration strategy in different aspects. Besides, we compare different algorithms and find that LSTM achieve better performance in predicting the positive arbitrage samples and obtaining higher ROI and Sharpe ratio. The profitability of machine learning strategy validate the mean reversion and price discovery function of asset price between spot market and futures market, which further substantiate the market efficiency. Our empirical results provide practical significance to the development of quantitative finance.