Skip to content

README.md

Latest commit

 

History

History
111 lines (87 loc) · 4.78 KB

README.md

File metadata and controls

111 lines (87 loc) · 4.78 KB
Raw

Python Backtest Engine

A compact, event-driven backtesting engine designed to demonstrate production-minded quantitative software design: clear domain models, deterministic execution assumptions, transaction costs, portfolio accounting, pluggable strategies, research parameterisation, and testable analytics.

Educational software only. It is not investment advice and should not be used for live trading without substantially more validation, controls, and market-data handling.

What it demonstrates

  • Explicit MarketEvent, SignalEvent, OrderEvent, and FillEvent models
  • Long-only portfolio accounting with cash, positions, equity and realised costs
  • Pluggable strategy interface
  • Moving-average crossover strategy
  • Three-moving-average trend filter inspired by the EPAT backtesting exercise
  • RSI mean-reversion strategy with overbought/oversold, take-profit and stop-loss exits
  • Donchian breakout strategy using prior high/low channels
  • Deterministic next-bar execution with configurable commission and slippage
  • Trade logs with entry/exit prices, commissions and net profit
  • Strategy and buy-and-hold benchmark equity curves
  • Extended metrics including CAGR, Sortino, Calmar, win rate and profit factor
  • Parameter sweeps and rolling out-of-sample walk-forward evaluation
  • A research-oriented run_strategy helper for isolated strategy runs

Strategy suite

The feature/epat-strategy-suite branch extends the original moving-average example with four reusable strategy implementations:

Strategy Core idea Parameters
MovingAverageCrossStrategy Fast/slow trend crossover fast/slow windows
ThreeMovingAverageStrategy Price above short/medium/long averages 3 windows
RSIMeanReversionStrategy Buy oversold, exit on overbought/target/stop RSI period, thresholds, TP, SL
DonchianBreakoutStrategy Break previous high channel, exit below low channel entry/exit windows

The strategy layer is deliberately separated from the engine. A strategy emits a target position; the engine turns that into an order, applies the simulated execution model, and updates the portfolio.

EPAT alignment

The strategy suite captures the main reusable ideas from the supplied backtesting material without copying the notebook structure directly:

  • multi-moving-average signal generation
  • RSI entry/exit logic
  • percentage take-profit and stop-loss controls
  • breakout channels based on prior observations
  • strategy return evaluation and parameter research

The implementation uses pure Python for the indicators so that the core package does not require TA-Lib. Yahoo Finance remains an optional research-data source through the existing data adapter.

Architecture

market-data adapter
        |
        v
     Strategy
        |
        v
      Signal
        |
        v
      Order
        |
        v
 simulated broker
        |
        v
       Fill
        |
        v
    Portfolio
        |
        v
    Analytics

The engine uses next-bar execution assumptions. Strategy signals are generated from information available on the current bar and are passed to the broker using the engine's deterministic execution model. This keeps the research design explicit and provides a foundation for later work on order types, intrabar execution and more realistic market microstructure.

Quick start

python -m venv .venv
source .venv/bin/activate
pip install -e '.[dev]'
python examples/run_moving_average.py TSLA
python examples/run_sp500_moving_average.py
python examples/plot_summary.py
pytest

The S&P 500 research runner uses the configured 10-stock universe and writes its results to research/sp500_ma/. For each ticker it produces a combined strategy/benchmark equity curve, a trade log, a parameter-sweep result, and walk-forward results, plus the aggregate summary.csv. The period and universe are configurable:

python examples/run_sp500_moving_average.py --start 2024-01-01 --end 2025-01-01 --tickers TSLA AAPL MSFT

After the research run, generate five PNG charts from the summary CSV:

python examples/plot_summary.py
python examples/plot_summary.py research/sp500_ma/summary.csv --output-dir research/sp500_ma/charts

The default output directory is research/sp500_ma/charts/. It contains total-return, maximum-drawdown, Sharpe-ratio, risk/return, and combined dashboard charts. Failed rows are excluded, and the input must contain both strategy and benchmark return fields.

Roadmap

  1. Add strategy-specific tests and deterministic fixtures.
  2. Add visual parameter surfaces and portfolio-level aggregation across symbols.
  3. Add richer execution models: spread, market/limit orders and partial fills.
  4. Add risk controls and position sizing.
  5. Add multi-asset portfolios.
  6. Add event-driven research and performance reporting.