Quants & Researchers¶
For quantitative analysts and researchers who prioritize statistical rigor and want systematic optimization and walk-forward validation.
Quantitative Evaluation Features¶
| Feature | Details |
|---|---|
| Walk-Forward Validation | Automatic IS/OOS splits to detect overfitting |
| Optuna Optimization | Bayesian search over parameter space (200–1000 trials) |
| Multiple Objectives | Choose from Sharpe ratio, max drawdown, Calmar ratio, etc. |
| Reproducibility | Lock seeds and config in JSON for fully reproducible experiments |
| Journal Logging | All experiment results automatically recorded as JSON/CSV |
Typical Research Workflow¶
# 1. Declare hypothesis in JSON
alpha-forge strategy create --template hmm_bb_pipeline_v1 --out regime_test.json
# 2. Grid search over the parameters declared in the strategy JSON
# (the grid is defined by optimizer_config.param_ranges in regime_test's JSON)
alpha-forge optimize grid QQQ --strategy regime_test --top-k 10
# 3. Walk-forward validation (5 folds)
alpha-forge optimize walk-forward QQQ --strategy regime_test --windows 5
# 4. Save experiment to journal
alpha-forge journal note regime_test "HMM period sensitivity analysis"
Evaluating Overfitting Risk¶
AlphaForge provides walk-forward testing (WFT) as a standard feature. The output is a table listing each window's IS Score and OOS Score; a large drop from IS to OOS suggests overfitting (compute the degradation yourself from IS/OOS).
Window IS Score OOS Score Best Params
1 1.8000 1.4000 {...} ← small drop, acceptable
2 2.5000 0.3000 {...} ← large drop, likely overfit
Related Docs¶
- Strategy Templates — Full JSON for HMM, regime-switching, multi-timeframe
- Strategy Gallery — Cross-market strategy comparison with result interpretation
- End-to-End Strategy Development Workflow — From optimization to WFT validation
- optimize command — Full optimization options