backtesting-trading-strategies

Run the backtest.py script with a strategy name, symbol, and period, for example: python scripts/backtest.py --strategy sma_crossover --symbol BTC-USD --period 1y. Results include performance metrics, a trade log CSV, equity curve data, and a chart saved to the reports directory.

Use optimize.py with a JSON parameter grid, such as --param-grid '{"fast_period": [10, 20, 30], "slow_period": [50, 100, 200]}'. It runs a backtest per combination and ranks results by Sharpe ratio or another metric you choose with --metric.

It fetches historical data from Yahoo Finance via yfinance by default and supports CoinGecko as an alternative source for crypto symbols. Downloaded data is cached as CSV files in the data directory for reuse across runs.

Unrealistically high metrics usually indicate overfitting or look-ahead bias. Test on out-of-sample data with walk-forward analysis, add realistic commission and slippage, and verify signals do not use future data.

Yes. Create a class extending the Strategy base class in scripts/strategies.py, implement generate_signals() to return a Signal with entry, exit, and direction fields, then register it in the STRATEGIES dictionary.

It uses daily bars by default, so intraday strategies are not modeled, and grid search is single-threaded so large parameter grids are slow. CoinGecko data approximates OHLC values, which reduces accuracy for band-based strategies.