Rolling Cointegration Tests for Pairs Trading

Summary

The document asks whether a pairs trading strategy should retest cointegration each day using a rolling window, rather than test once during an initial formation period and then trade the selected pair for a fixed holding period. The proposed approach uses daily closing prices and repeatedly assesses whether the relationship still appears cointegrated as the trading period progresses.

The text frames this as a research design question and notes that papers commonly separate formation and trading periods. It does not provide test results, specify a cointegration test, or explain why the cited literature tends to avoid daily retesting. A rolling test could make pair selection responsive to changing relationships, but the document leaves open practical choices such as window length, test thresholds, and how to handle repeated selection or unstable estimates.

Key ideas

• A rolling formation window can be used to reassess cointegration during a pairs trading period.
• The proposed alternative updates the test daily using daily closing prices.
• The document contrasts this with a common design that tests during formation and trades afterward without retesting.
• No empirical results or explanation for the literature’s usual design are provided.

Tags

# Pairs Trading (Cointegration Approach) - Daily Cointegration Test


# Pairs Trading (Cointegration Approach) - Daily Cointegration Test












I have a question regarding the Pairs Trading strategy based on the Cointegration Approach. Most of the papers/literature I found on Pairs Trading using the Cointegration Approach are usually testing a given period for Cointegration (say 6 months) and then consider this pair for their trading strategy if the Cointegration Test suggests that the pair is cointegrated. In the papers they usually trade the next couple months (say 3 months) based on the assumption that the Cointegration from the 6 months testing period still holds.

Here is what I was wondering: (Assuming we use daily closing prices) instead of doing this test only once for the 6 months and then trading for the next 3 months, could we also check for Cointegration on a daily basis (rolling 6 months window) in the 3 months trading period? If so, why did none of the other papers do it in this way before?

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

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