Why a Backtest Segment May Differ From Its Full-Period Results
Summary
This Chinese-language question and answer explains why a strategy’s apparently strong later years in a long backtest may not reproduce the same pattern when tested over those years alone. It identifies several possible causes rather than prescribing a single fix. For concentrated strategies that invest fully in a small number of stocks, a shift in the initial trade date can change early gains or losses, which then affects subsequent compounded returns.
The answer also points to overfitting, missing stock data on some dates, and factor-feature preparation that consumes additional history before the requested test period. That preprocessing can move the effective start date and alter the returns included in the run. These are diagnostic considerations, not demonstrated findings from a particular strategy: the document gives no data, comparisons, or procedure for measuring each effect. It does not discuss other backtest design issues or specify how to resolve the discrepancies. Its practical lesson is to check entry timing, data completeness, model fit, and the actual dates used after feature preparation when comparing full-period and subperiod results.
Key ideas
- A different first trade date can materially affect returns for concentrated portfolios.
- Compounding can carry early gains or losses through the remainder of a backtest.
- Overfitting is one possible reason a favorable historical segment fails to repeat alone.
- Missing stock observations can change which securities are selected and their measured returns.
- Feature preparation may require extra historical data and shift the effective test start date.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.