pybroker

The framework provides a vectorized backtesting engine with accurate simulation of transaction costs, slippage, and portfolio constraints, ensuring that backtested performance reflects realistic trading conditions rather than idealized assumptions.

The framework's ML integration supports any scikit-learn-compatible model, XGBoost, LightGBM, PyTorch neural networks, and custom prediction functions as signal generators within strategy logic.

Walk-forward validation utilities prevent look-ahead bias in model training, automatically retraining models on historical windows as the backtest progresses to simulate how a live system would update its models over time.

PyBroker includes a data caching system that stores fetched market data locally, reducing API calls during iterative strategy development.

PyBroker is open-source under the Apache 2.0 license and targets quantitative researchers and systematic traders who want to combine traditional algorithmic trading strategy development with machine learning signal generation in a single Python framework.

It integrates with Alpaca for live paper and live trading execution, and supports custom data feeds for strategies using alternative data sources. The framework's emphasis on realistic simulation and proper ML validation practices reflects the practical challenges that distinguish live trading from research backtests.