ai-trader

Wraps Backtrader's Cerebro event loop to manage the complete backtesting lifecycle, including broker initialization, data feed registration, strategy attachment, and execution sequencing. The AITrader class abstracts Backtrader's complexity by handling calendar-based event dispatch, order management callbacks, and portfolio state tracking across multiple trading days without requiring developers to interact directly with Cerebro's lower-level APIs.

Implements a library of 15+ technical indicators (SMA, RSI, Bollinger Bands, RSRS, ROC, etc.) that inherit from Backtrader's Indicator base class, computing real-time signals during backtesting by processing OHLCV bars sequentially. Each indicator encapsulates its calculation logic and exposes output lines (e.g., signal, upper_band, lower_band) that strategies reference to generate buy/sell decisions without manual formula implementation.

Generates matplotlib-based visualizations of portfolio equity curves with overlaid trade markers (entry/exit points) and indicator signals, allowing traders to visually inspect strategy behavior and identify periods of underperformance. The visualization integrates with Backtrader's plotting module and automatically scales axes, formats dates, and annotates trades without manual matplotlib configuration.

Provides a framework for developers to create custom technical indicators by subclassing Backtrader's Indicator class and defining calculation logic in the __init__ method. Custom indicators integrate seamlessly into the backtesting event loop, compute incrementally on each bar, and expose output lines that strategies can reference for signal generation.

Automatically extracts detailed trade information (entry date, entry price, exit date, exit price, P&L, duration, return percentage) from completed backtests into a pandas DataFrame, enabling post-backtest analysis of trade quality, win rate, average win/loss, and trade duration statistics without manual data extraction.

Provides 10+ pre-built strategy classes (SMA, RSI, Bollinger Bands, ROC, Double Top, Turtle, VCP, Risk Averse, Momentum, Buy and Hold) that inherit from BaseStrategy and implement complete entry/exit logic using technical indicators. Developers instantiate these strategies with parameters (e.g., fast_period=10, slow_period=20) and attach them to the backtester, eliminating the need to write signal generation and order placement code from scratch.

Implements multi-asset portfolio strategies (ROC rotation, RSRS rotation, Triple RSI rotation, Multi Bollinger Bands rotation) that dynamically allocate capital across a basket of stocks based on relative strength or momentum rankings. The framework rebalances the portfolio at fixed intervals (e.g., monthly), selling underperformers and buying outperformers, with position sizing determined by indicator rankings rather than equal weighting.

Provides a StockLoader utility that downloads historical OHLCV data from Yahoo Finance or CSV files, normalizes column names and data types, handles missing values, and converts data into Backtrader-compatible DataFrames. The loader abstracts data source differences, allowing strategies to work with data from multiple providers without custom parsing logic.

Automatically computes portfolio-level performance statistics (total return, annualized return, Sharpe ratio, Sortino ratio, max drawdown, win rate, profit factor) from backtest results using Backtrader's built-in analyzers. Results are aggregated into a structured dictionary and optionally visualized as equity curves, allowing traders to compare strategies quantitatively without manual calculation.

Exposes the AI-Trader backtesting engine as a Model Context Protocol (MCP) server, allowing LLM-based agents (Claude, GPT-4, etc.) to invoke backtesting operations, retrieve strategy results, and modify parameters through standardized tool calls. The MCP server translates natural language requests into Python function calls, enabling non-technical users to run backtests via conversational AI without writing code.

Provides command-line interface (CLI) commands for running backtests, sweeping strategy parameters across ranges, and generating comparison reports without writing Python code. The CLI parses YAML/JSON configuration files defining strategy, data, and parameter ranges, executes backtests in batch mode, and outputs results to CSV or JSON files for further analysis.

Defines an abstract BaseStrategy class that all trading strategies inherit from, providing shared methods for order placement (buy, sell, close), logging, and Backtrader integration. Developers subclass BaseStrategy, override the next() method to implement signal logic, and leverage inherited utilities for order management without duplicating boilerplate code across strategies.

Supports backtesting on both US stocks (via NASDAQ/NYSE tickers) and Taiwan stocks (via TWSE tickers like '2330.TW'), with automatic market calendar handling for trading days and holidays. The framework downloads data from Yahoo Finance, which provides both markets, and applies market-specific trading hours and holiday calendars during backtesting.