Negative PnL Strategy's Sharpe, Sortino and Calmar ratios are zero

[atakanokan]

Is there any way for Backtest.run() to return values (instead of 0) for strategies with negative PnL? From what I can see, they all return 0.

[kernc]

You could remove np.clip() wrapping in these places:

backtesting.py/backtesting/_stats.py

Line 120 in 7671f04

s.loc['Sharpe Ratio'] = np.clip((s.loc['Return (Ann.) [%]'] - risk_free_rate) / (s.loc['Volatility (Ann.) [%]'] or np.nan), 0, np.inf) # noqa: E501

backtesting.py/backtesting/_stats.py

Line 122 in 7671f04

s.loc['Sortino Ratio'] = np.clip((annualized_return - risk_free_rate) / (np.sqrt(np.mean(day_returns.clip(-np.inf, 0)**2)) * np.sqrt(annual_trading_days)), 0, np.inf) # noqa: E501

backtesting.py/backtesting/_stats.py

Line 124 in 7671f04

s.loc['Calmar Ratio'] = np.clip(annualized_return / (-max_dd or np.nan), 0, np.inf)

Do negative Sharpe, Sortino, Calmar ratios make any sense?

[atakanokan]

It would be good to look at cross-asset robustness via SR distribution which Neg->0 SR clipping skews. Would be great to receive the ratios as is and do clipping if needed.

[atakanokan]

@kernc if you think this would be a good PR, I would be more than happy to do it

[kernc]

Clipping came in in 028f02d with no elaborate justification. I agree, PR welcome. 👍

[kernc]

I remember my thinking: with risk-free rate default assumed 0%, if your strategy's Sharpe is negative, it's better to just divest and keep the money. 😄

[atakanokan]

Opened: #473