Filtering Cointegrated Spreads for Mean-Reversion Pairs Trading

Summary

The document lays out a screening process for spreads used in pairs trading and statistical arbitrage. Candidate constituents are tested for cointegration, mean-reverting behavior, practical reversion speed, and frequent crossings of the spread’s mean. It describes Engle–Granger testing but notes its dependence on which asset is treated as the dependent variable. One mitigation is to test both orderings; an alternative is orthogonal regression, or total least squares, which accounts for variation in both legs when estimating the hedge ratio. Other listed hedge-ratio methods include Johansen eigenvectors, Box–Tiao decomposition, and criteria based on half-life or test statistics.

Further filters use the spread’s Hurst exponent and half-life, with the example framework excluding reversion periods judged too short or too long and requiring at least one mean crossing per month. Thresholds are adjustable, and the approach can extend beyond pairs to larger baskets. These are selection rules, not evidence of profitability: the text supplies no performance results and warns that screening can be computationally heavy.

Key ideas

• The proposed spread screen checks cointegration, mean-reverting behavior, reversion speed, and mean-crossing frequency.
• Engle–Granger results can depend on which constituent is treated as the dependent variable.
• Total least squares estimates the relationship using variation in both spread legs and aims for order-independent hedge ratios.
• The framework uses Hurst exponent and half-life filters to assess mean reversion.
• The rules can be extended to multi-asset spreads, but the document gives no profitability results.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.