Statistical Arbitrage and Robust Tests for Cointegration

Abstract

One application of cointegration tests is screening candidate stocks for the investment strategy known as statistical arbitrage. This paper develops two robust tests for cointegration by using rank-based and least absolute deviation regression to modify the seminal Engle-Granger test. Critical values are generated and power calculated for various error distributions. Finally, the tests are utilized in a simple pairs trading strategy and backtested on daily data from 2001 to 2010. The rank-based cointegration test has superior qualities in that context, suggesting one application of this new statistical test.