Dynamic Pair Selection with Stationarity Tests and Bollinger Bands

Summary

This project describes a stock pairs strategy that periodically selects candidate pairs based on whether their price spread appears stationary. It estimates a hedge ratio with linear regression, forms a spread, and applies an Augmented Dickey-Fuller test to decide whether to track the pair. The premise is that correlations can change, so candidate pairs should be reassessed using recent data rather than fixed indefinitely.

For selected pairs, the strategy uses Bollinger Bands on the spread to define entries when it moves beyond a band, exits when it returns to the mean, and stops when it moves beyond the band again. The project also describes reviewing historical trade accuracy and returns, with an example of discarding pairs that perform poorly. It presents no complete performance record or robust out-of-sample evidence, and the stated spread formulas are inconsistent in places. Short-term stationarity and historical trade results may not persist, so the approach depends on careful estimation and validation.

Key ideas

• The project periodically selects pairs based on the stationarity of their price spread.
• Linear regression supplies a hedge ratio used to construct the spread.
• The Augmented Dickey-Fuller test is used to screen candidate pairs for stationarity.
• Bollinger Bands define spread entries, mean exits, and band-based stop conditions.
• Historical accuracy and return filters do not guarantee that a pair will remain cointegrated.

Tags

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