Optimizing Pairs Trading of US Equities in a High Frequency Setting
time series List of disciplines your research might apply to: Statistics
Time Series Analysis
In this paper, we examine how to the performance of high-frequency pairs trading strategies are impacted by the allocation within the pair, opening and closing thresholds, restriction to daily trading, and transaction costs. We generate portfolios by applying high-frequency pairs trading strategies to the pair consisting of Exxon Mobil Corporation (XOM) and Chevron Corporation (CVX) during the year 2005. We find the following results. First, we find that a dynamic model for estimating the spread of a pair is more suitable for high-frequency trading when compared to a static model. Second, we find that allocating investment within the pair based on the ratio of their CAPM β, compared to a 1:1 dollar allocation, and allocation based on the cointegration coefficient, yields the most attractive portfolios. Third, we find that setting the opening threshold to 1.5σ and the closing threshold to 1.0σ, respectively generate portfolios with the highest Sharpe ratios when compared to portfolios constructed using the same strategy, but different threshold values. Finally we find that restricting trading to once-a-day and imposing transaction costs significantly worsens the performance of the strategy.