REF: Use geometric mean return in Sharpe/Sortino stats computation

See:
https://qoppac.blogspot.com/2017/02/can-you-eat-geometric-returns.html
https://www.investopedia.com/ask/answers/06/geometricmean.asp

Author: kernc

backtesting/backtesting.py

@@ -1285,7 +1285,7 @@ def _round_timedelta(value, _period=_data_period(index)):
         s.loc['Win Rate [%]'] = win_rate = np.nan if not n_trades else (pl > 0).sum() / n_trades * 100  # noqa: E501
         s.loc['Best Trade [%]'] = returns.max() * 100
         s.loc['Worst Trade [%]'] = returns.min() * 100
-        mean_return = returns.mean()
+        mean_return = returns.prod()**(1/(len(returns) or np.nan))
         s.loc['Avg. Trade [%]'] = mean_return * 100
         s.loc['Max. Trade Duration'] = _round_timedelta(durations.max())
         s.loc['Avg. Trade Duration'] = _round_timedelta(durations.mean())