Parameter optimization¶
Parameter optimization compares pre-defined strategy parameter combinations; it is not a mechanism for finding a number that can be deployed automatically. Build it on reproducible data and backtest configuration, then combine it with out-of-sample and robustness checks.
Principles¶
- Fix strategy version, instrument, timeframe, data range, capital, and costs first.
- Set ranges and steps only for parameters with business meaning; avoid unbounded search.
- Evaluate return, drawdown, trade count, and stability together—not only the highest Sharpe or return.
- Re-test candidates in a period that did not participate in the search and retain failures.
- Store parameters, outputs, and selection rationale in the research workspace.
Common risks¶
| Risk | Mitigation |
|---|---|
| Overfitting | Constrain the search space and use out-of-sample, rolling-window, and robustness validation. |
| Too few trades | Check order lifecycle, data coverage, and a minimum trade count; do not decide on accidental samples. |
| Non-reproducibility | Fix strategy version, source, dates, costs, and random settings. |
| Metric misreading | Review equity curve, drawdown, trade statistics, and risk constraints together. |
Optimization APIs live under /api/v1/optimization; inspect OpenAPI for request models and algorithms enabled in the current environment. See Backtests and validation for result interpretation.