Spread Movement Prediction for Pairs Trading with High-Frequency Limit Order Data

Pairs trading is a market-neutral trading strategy that is not easily affected by the market trends. By continuously monitoring the spread value between a pair of stocks whose prices are statistically correlated, traders are able to make profit when the spread deviates from its historical mean. Thus, it is useful to predict the spread movement in order to identify opportunities to buy and sell stocks. Previous pairs trading works mainly develop strategies based on spread values or stock prices. Limit order book (LOB), comprising buy and sell orders over time, is able to reflect the market intention, and thus, is highly related to the formation of stock price as well as spread value. However, to the best of our knowledge, there is no prior work using the LOB data to design pairs trading models. In this paper, we make the first attempt to study the spread movement prediction problem for pairs trading by utilizing not only the spread data but also the high-frequency LOB data. We propose a deep learning model that is capable of extracting representative features from both the spread history and the LOB data, and further exploits the dynamic inter-dependencies between them to sophisticate the spread movement prediction. For evaluation, we collect the historical tick data of the 147 companies from Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX). The experimental results show that our model achieves comparable performance to other baseline models for generating spread movement predictions over different time horizons.