Pairs trading with deep learning

抄録

Pairs trading is a trading strategy that aims to achieve stable earnings by holding two shares in combination. Due to the fact that the trading group of Morgan Stanley earned over 50 million dollars a year, this method remains to be one of the most popular trading methods. Indeed, if you could find stocks showing "similar movements", you would be able to perform stable earnings. So far, it has been mainly studied how to find appropriate pairs using classical statistical techniques. However, we have not yet attained the robust method to find appropriate pairs by the classical technique. This study uses deep learning to construct a network of stocks which enable to find the appropriate pairs of a stock. More precisely, by giving a shock on the price of a stock, we can search a stock moving most similarly to the stock given the price shock and we select it as the partner of the stock. This leads to a more stable method for the pair selection than conventional methods. In order to show the validity of this method, we conducted a trading simulation using stocks of S&P500, which is the representative index of the US stock market. Then, we confirmed the strong stability of the method proposed in this study by comparing it with the distance method, which is one of the most famous pairs trading strategies.