Detecting Structural Breaks in Trading Strategy Estimates
Summary
The article proposes an automatic procedure for finding changes in historical trading parameters before portfolio optimization. It considers forecast and instrument weights, and describes recursively testing whether an early return segment differs from the data that follows. Candidate splits are evaluated with a statistical test on returns normalized by each segment's standard deviation; the process retains a minimum amount of history for estimation and repeats on the later segment after detecting a break.
The author then compares optimization results using all history against results using data after detected breaks, across multiple in-sample lengths, out-of-sample horizons, and significance thresholds. The reported Sharpe ratios are mixed: some settings favor retaining all data, while others favor break-based selection, with outcomes varying across test configurations. The article gives a procedure and illustrative tests rather than a definitive validation. Its approach uses a particular break test and threshold scheme, and the shown results do not support one universally best threshold or demonstrate that detected breaks reliably improve live portfolio performance.
Key ideas
- The procedure searches sequentially for breaks between an earlier return sample and the remaining history.
- After detecting a break, it recursively analyzes the later segment and preserves a minimum estimation history.
- The described test compares volatility-normalized returns across the two samples.
- Optimization tests compare full-history estimates with estimates based on data after detected breaks.
- Reported performance varies across test settings, so the procedure has no clearly universal best threshold.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.