Statistical arbitrage using pairs trading with support vector machine learning

Abstract

The purpose of this study is to analyze the performance of dynamic PCA (Principal Component Analysis) Statistical Arbitrage, and to validate the results with the help of a novel Machine Learning approach known as Support Vector Machines using the “Pairs trading” strategy. The paper starts by explaining the fundamental concepts behind our analysis e.g. Linear Regression, Auto-Regressive processes and Orstein Uhlenback modeling of residuals. Research focus will be on two things: how the principal components are obtained and how the portfolio of systematic risk factors is formed. Stock data of 20 stocks from the XLF financial sector is chosen for the principal components analysis. The data includes each stock’s daily opening price, high, low, adjusted close price and daily volume from the year 1998 to 2012. There are total of 69,920 observations. The paper concludes by demonstrating the scenario when SVM gave better results compared to the basic Mean-reversion strategy and future enhancements possible with this mixed approach.