edtechre/pybroker

What it solves

PyBroker is a framework for developing and testing algorithmic trading strategies, specifically those that leverage machine learning. It simplifies the process of creating trading rules, training models, and evaluating performance through rigorous backtesting and optimization.

How it works

The framework provides a high-performance backtesting engine powered by NumPy and Numba. It allows users to define execution functions (trading rules) and integrate machine learning models that generate predictions used for trade execution. It supports multiple data sources (like Alpaca and Yahoo Finance) and employs Walkforward Analysis to simulate real-world trading performance by iteratively training and testing on moving windows of data.

Who it’s for

Quantitative traders, data scientists, and developers who want to build, optimize, and evaluate machine learning-based trading strategies in Python.

Highlights

• High-Performance Engine: Backtesting is accelerated with Numba and NumPy for speed.
• Walkforward Analysis: Simulates actual trading by training and model testing in a rolling window fashion.
• ML Integration: Built-in support for registering and training models to drive trading signals.
• ** FileNotFound Error Handling**: Includes parameter optimization via Optuna and reliable metrics using randomized bootstrapping.
• AI Agent Skills: Provides specialized tools to help AI coding agents write strategies, indicators, and trainers.
• Multi-Interval Support: Ability to integrate trading signals across daily, weekly, and monthly timeframes.