Monitoring Cointegration Weight Stability with Rolling Eigenvector Windows
Summary
The article develops Rolling Windows Eigenvector Comparison (RWEC) to monitor whether the portfolio weights of a cointegrated stock basket remain stable. It contrasts RWEC with In-Sample/Out-of-Sample ADF validation, which estimates a cointegration vector on training data and tests the resulting spread for stationarity on held-out data. The article identifies limits of that validation for live monitoring: it needs relatively long out-of-sample periods, depends on the split point, and does not show when instability begins.
RWEC compares eigenvectors estimated over rolling, overlapping windows to identify changes in basket weights sooner. The proposed workflow combines it with IS/OOS ADF and walk-forward analysis, using the ADF test to assess spread stationarity and RWEC as an early warning for weight instability, possible rebalancing, or halting a basket. An ETF pair and Python examples illustrate the approach. The article argues that the methods complement each other, but the supplied discussion does not establish universal thresholds or guarantee future stability; the results depend on the data and test settings.
Key ideas
- IS/OOS ADF estimates weights in-sample and tests the resulting spread for stationarity out-of-sample.
- Long validation periods, split sensitivity, and delayed break awareness limit IS/OOS ADF for live monitoring.
- RWEC compares eigenvectors estimated in rolling windows to track portfolio-weight changes.
- Combining RWEC, IS/OOS ADF, and walk-forward analysis can inform monitoring and rebalancing decisions.
- Weight and signal stability are probabilistic and cannot be guaranteed from historical data.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.