Performance Metrics

Performance metrics are computed from the equity curve produced by the backtest engine. The calculations live in metrics/performance_calculator and are exposed in reports.

Metrics Diagram

flowchart LR
  A[Equity Curve] --> B[Return Series]
  B --> C[Volatility]
  B --> D[Sharpe/Sortino]
  A --> E[Drawdown]
  E --> F[Calmar]

Common Metrics

• Return: period-over-period change in equity.
• Average return: mean of returns.
• Volatility: standard deviation of returns.
• Downside volatility: downside deviation computed against the full return sample.
• Sharpe ratio: excess return per unit of total volatility.
• Sortino ratio: excess return per unit of downside deviation.
• Max drawdown: worst peak-to-trough decline.
• Calmar ratio: CAGR divided by max drawdown.

Regime-Aware Metrics

Regime-aware reporting now follows the same core conventions as the top-level performance summary:

• regime return is compounded across the intervals assigned to that regime
• regime time share is duration-weighted, not sample-count weighted
• regime Sharpe in the final report is taken from the annualized PerformanceCalculator path
• transition return is the compounded return over the transition window, averaged across occurrences

This matters most for intraday and event-driven runs, where equal sample counts do not imply equal time in regime.

Annualization

Metrics are scaled using periods_per_year. When the engine has timestamped portfolio snapshots, annualization is inferred from the observed sampling interval instead of assuming daily bars. This keeps performance and performance_summary aligned on intraday runs.

Rolling Metrics

Rolling metrics are computed by sliding windows across the equity curve to show time-varying behavior.