A strategy-agnostic ecosystem for rigorous backtesting, robust parameter optimization, and regime analysis.
This repository contains a framework designed to bridge the gap between theoretical trading strategies and market reality. Unlike standard backtesters that prioritize raw returns, this system prioritizes statistical robustness.
It helps researchers answer the critical question: "Is this strategy actually good, or did I just overfit the parameters to historical noise?"
- Strict IS/OOS Separation: Never optimize on the data you validate with.
- Degradation over Profit: The best parameter set is not the one with the highest return, but the one with the least performance degradation between training and testing.
- Regime Awareness: Strategies are stress-tested against specific market conditions (Bull, Bear, Sideways).
- CMA-ES Optimization: Uses the Covariance Matrix Adaptation Evolution Strategy (via Optuna) to navigate non-convex parameter spaces without gradients.
- Vectorized Simulation: High-performance engine handling entry/exit logic, position sizing, and commission modeling.
- Modular Architecture: "Plug-and-play" design allows you to drop new Python strategy files into a folder and immediately start optimizing them.
- Robustness Metrics: Automatically calculates Sharpe, Calmar, and Degradation coefficients.
The framework generates a comprehensive suite of interactive HTML reports for every simulation:
| Report | Description |
|---|---|
| Backtest Performance | Interactive candlestick charts with overlay indicators, trade markers, and equity curves. |
| Robustness Scatter | A correlation view of In-Sample vs. Out-of-Sample metrics to identify overfitting visually. |
| Regime Matrix | A breakdown of strategy performance during specific market regimes (e.g., "Bullish" vs. "Sideways"). |
| Parallel Coordinates | A visualization of the hyperparameter search space, highlighting high-performance clusters. |
| Distribution Boxplots | Visualizes the variance of returns across all trial iterations to distinguish alpha from luck. |
| Stability Analysis | Bar charts comparing In-Sample vs. Out-of-Sample metrics to detect immediate performance decay. |
For detailed instructions on installation, data ingestion (API or CSV), and creating your own strategies, please refer to the technical documentation. (docs/TECHNICAL_GUIDE.md)
git clone https://github.com/strakehyr/Trading-Strategy-Optimizer.git
pip install -r requirements.txt
python main.py --symbols QQQ --strategies all