SQN calculation not using expectancy?

[souresia]

Hi,

It's good to see the stats include SQN. According to https://evilspeculator.com/tools-section/concepts/:

SQN = root(n) * expectancy / stdev(R)

but I suspect it's not correctly calculated:

backtesting.py/backtesting/_stats.py

Line 139 in 77f4c46

s.loc['SQN'] = np.sqrt(n_trades) * pl.mean() / (pl.std() or np.nan)

as it's not using the result of expectancy:

backtesting.py/backtesting/_stats.py

Line 138 in 77f4c46

s.loc['Expectancy [%]'] = returns.mean() * 100

Unless I am missing something?

Thanks!

[kernc]

If we look slightly higher up:

backtesting.py/backtesting/_stats.py

Lines 72 to 73 in 77f4c46

	pl = trades_df['PnL']
	returns = trades_df['ReturnPct']

returns for expectancy are in percent whereas pl are in absolute currency units.

I see Van Tharp mentions expectancy and R-multiples (reward/risk), which are percent-based, but I modeled it after backtrader and thought those pnl were absolute values:
https://github.com/mementum/backtrader/blob/e2674b1690f6366e08646d8cfd44af7bb71b3970/backtrader/analyzers/sqn.py#L75-L78

I guess we might be using the wrong values then. 😬