Kalman Filter Pairs Trading with Cointegration Screening and Loss Limits

Summary

The project describes a stock pairs strategy that models one price as a time-varying linear function of another. A Kalman filter updates the intercept, hedge ratio, spread, and spread volatility; spread z-scores set entry and exit signals. It screens candidate pairs for entry conditions and tests spread stationarity with the augmented Dickey-Fuller test as evidence of cointegration. The strategy ranks eligible pairs and limits the number held in a sector portfolio.

A test calibrates the filter in one period and trades in a later period, reporting two variants with different maximum trade losses. The tighter loss limit has the stronger reported Sharpe ratio and total P&L in that comparison. The evidence is preliminary: it focuses results on a single pair, chooses thresholds through training-period simulation, and does not establish performance across multiple years. The author identifies screening, lookback selection, recalibration, sector coverage, and broader out-of-sample testing as open work.

Key ideas

• The strategy uses Kalman filtering to estimate a changing hedge relationship and spread between two stocks.
• Spread z-scores provide entry and exit signals for long or short pair positions.
• An augmented Dickey-Fuller test screens estimated spreads for stationarity, though its lookback choice remains a tuning parameter.
• The reported comparison finds better results under the tighter per-trade loss limit.
• Results are preliminary and require broader multi-year out-of-sample evaluation.

Tags

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