High Frequency Statistical Arbitrage with Kalman Filter and Markov Chain Monte Carlo

Abstract

Statistical arbitrage, or sometimes called pairs trading, is an investment strategy which exploits the historical price relationships between two or several assets and profits from relative mispricing. It has a long history in hedge fund industry and variates of this kind of strategies are still profitable nowadays. The idea is simple and the source of the profit has support from fundamentals in economics and pricing theories. However, there are still many difficulties in implementing and testing such strategies in real life, which include how to select pairs, how to estimate hedge ratio, when to enter, when to exit and etc. Due to its proprietary nature, there is very few literature on this subject. This thesis is an attempt to demystify statistical arbitrage in high-frequency settings, using freely available data of Chinese commodity futures. This thesis introduces and discusses the existing research done on this subject. Also, with the help of advanced statistical inference approaches for treating time series, this thesis proposed a new model which generalizes the entire process of creating a profitable statistical arbitrage trading strategy for a given market. Several different approaches are implemented and their simulated performances in the Chinese commodity future market are compared horizontally. Unlike much other existing literature, transaction costs and market frictions have been considered thoroughly in order to make the research result more meaningful. Empirical results show that our new model delivers very competitive performance in online hedge ratio estimation.