[galafis]

A zero gross return (a -100% return) currently makes geometric_mean return 0 instead of -1. Consequently, a backtest whose equity falls to zero reports Return (Ann.) as 0% even though Return is -100%.

Handle zero gross returns explicitly as a geometric mean return of -1. Preserve the existing behavior for negative gross returns, NaNs and empty inputs.

Tests cover total loss before and after positive returns, missing observations, ordinary positive growth factors, empty data, and an end-to-end leveraged strategy whose equity is exhausted. The independent product-root formula, and scipy.stats.gmean for zero gross returns, give -1. Reference: https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.gmean.html .

Before: four total-loss helper subcases and the end-to-end annualized-return assertion fail. After: all 26 TestBacktest tests pass. flake8 on both changed files and git diff --check pass. Python 3.12.14, pandas 3.1.0, NumPy 2.5.4.

Full suite after installing test extras: 84 tests OK, 1 skipped. Existing pandas deprecation warnings remain. Mypy reports the same 10 errors in unchanged _plotting.py as the clean baseline (error lines compared identical).