Backtest Tearsheets

Generate a standalone HTML report from normalized metrics, trades, and returns. The minimal supported input is a metrics dictionary or a return series. The example uses synthetic daily returns and does not require a backtest engine. Install the viz extra for Plotly rendering. Use real returns at the frequency assumed by the metrics and statistical corrections.

Generate an HTML report

from pathlib import Path

import numpy as np

from ml4t.diagnostic.visualization.backtest import generate_backtest_tearsheet

rng = np.random.default_rng(42)
daily_returns = rng.normal(loc=0.0005, scale=0.01, size=252)
metrics = {
    "n_trades": 80,
    "total_pnl": 12_500.0,
    "win_rate": 0.54,
    "profit_factor": 1.6,
    "sharpe_ratio": 1.3,
    "max_drawdown": -0.12,
}

output = Path("backtest_report.html")
html = generate_backtest_tearsheet(
    metrics=metrics,
    returns=daily_returns,
    template="quant_trader",
    theme="default",
    output_path=output,
    n_trials=25,
)

assert output.exists()
assert "plotly" in html.lower()
print(f"Wrote {output}")

The file-existence and Plotly assertions check that rendering completed. Inspect reported Sharpe, drawdown, and trial count against your source backtest; HTML generation does not validate supplied metrics.

Choose a template

Template	Primary content
quant_trader	Trades, performance, and model diagnostics
hedge_fund	Performance, costs, and reporting context
risk_manager	Risk and statistical validation
full	Every available section

Use BacktestProfile when you have normalized trades, returns, positions, costs, predictions, and factor results. Use generate_tearsheet_from_result for an ml4t-backtest result. The integration normalizes the backtest object before rendering.

PDF export also depends on the browser and rendering packages documented in the installation guide. A headless environment can verify HTML generation without claiming that browser-dependent PDF export was exercised.

See the visualization API and the book's performance reporting notebook, which directly calls Diagnostic in a longer backtest workflow. For Pyfolio inputs, use the migration guide.