Summary

PerformanceCalculator.calculate_all_metrics() annualized returns by dividing the period by 365.25 in both branches:

if 'datetime' in equity_curve.columns:
    days = (last - first).days        # calendar days
else:
    days = len(equity_curve)          # trading-bar count
years = max(days / 365.25, 1/self.trading_days)

• With a datetime column, days is elapsed calendar days, so /365.25 is correct.
• Without one, days is a count of trading bars, and dividing that by 365.25 over-annualizes the return.

Example: 252 trading bars at +20% total reported ~30% annualized instead of ~20%.

Closes #12.

Fix

Annualize the no-datetime branch by self.trading_days; leave the calendar branch unchanged.

if 'datetime' in equity_curve.columns:
    days = (last - first).days
    years = max(days / 365.25, 1 / self.trading_days)
else:
    bars = len(equity_curve)
    years = max(bars / self.trading_days, 1 / self.trading_days)

trading_days was already used by the Sharpe/Sortino/volatility calculations; this makes annualized_return consistent with it for bar-indexed curves. (The issue's "trading_days is never used" note was slightly off — it was used everywhere except this annualization.)

Verification

252 bars, +20% total -> annualized_return = 0.20   (was ~0.30)

Adds test_annualized_return_without_datetime_uses_trading_days.

Scope note

tests/backtesting/test_performance.py has 3 pre-existing failures unrelated to this change — test_win_rate_calculation, test_sharpe_ratio_calculation, and test_annualization all assert incorrect expected values (wrong oracles). They use the calendar branch this PR does not touch and are tracked by #14. My new test passes; those remain for #14.

🤖 Generated with Claude Code