Recipe: Interpret Backtest Metrics¶
Goal¶
Read backtest outputs consistently when comparing strategy changes.
Generate a run¶
stacksats strategy backtest \
--strategy my_strategy.py:MyStrategy \
--start-date 2020-01-01 \
--end-date 2025-01-01 \
--output-dir output
Key metrics to inspect¶
win_rate: windows where dynamic beats uniform.exp_decay_percentile: recency-weighted percentile.score: blended ranking metric.
Artifact checklist¶
backtest_result.jsonmetrics.json- plot
.svgfiles
Comparison rules¶
- Keep date range and allocation span identical.
- Compare against same baseline assumptions.
- Use strict validation outputs to flag overfitting risks.