High-Frequency ETF Pairs Trading

Abstract

In this paper we examine the effectiveness of modeling a paris-traded ETF portfolio as an Ornstein-Uhlenbeck process. Using ETF pairs that have similar references indexes, we apply maximum likelihood estimation to historical data in order to optimize trading signals for two strategies. Using this information, we test the optimal trading rules using intraday price observations over a variety of trading periods ranging from 5 days to 42 days. Our results have shown that the sample of ETF pairs-traded portfolios selected exhibit mean-reversion properties that are well modeled as an Ornstein-Uhlenbeck process. We have found that while higher total trading returns were correlated with shorter optimization and trading periods, they also carried considerable risk and as such more stable results were found using longer optimization and trading windows.

Keywords: ETFs, Ornstein-Uhlenbeck, High Frequency Trading, Pairs Trading, Statistical Arbitrage, Mean-reversion