Calculation of profit factor

[JensGBG]

It seems like the calculation of profit factor using .trades.profit_factor() is incorrect.

[JensGBG]

Profit factor = Gross wins/Gross losses

I suspect what it currently does it it takes (Sums of all wins from winning trades)/(Sums of all losses from losing trades). It might sound like they do the same but they don't.

[polakowo]

Isn't gross profit the sum of profits from all winning trades, and gross loss the absolute sum of losses from all losing trades?

[JensGBG]

Not exactly, here is a code snippet from Timothy Masters from his book Testing and Tuning market trading systems, its in C++ though.

double profit_factor ( int n , double *prices )
{
  int i ;
  double ret, win_sum, lose_sum ;
  
  win_sum = lose_sum = 1.e-60 ;
  
  for (i=1 ; i<n ; i++) {
    ret = prices[i] - prices[i-1] ;
    if (ret > 0.0)
    win_sum += ret ;
    else
    lose_sum -= ret ;
    }
  return win_sum / lose_sum ;
}

Its way simpler than you would think

EDIT: indentations in code snippet

[polakowo]

This seems like a bar-based or tick-based profit factor and not the classic trade-based profit factor. I may include is an additional metric, but it shouldn't replace the original formula.

[JensGBG]

No its definitely the right formula. I think profit factor is often misinterpreted though. But if you think about it, Masters formula makes way more sense since Profit Factor is supposed to be a risk-adjusted metric.

Imagine you have a portfolio that starts with 100 cash, enters a position and then exits the position at 120. Then it takes another position and exits at 110. So the portfolio has done: 100 ---> 120 ---> 110. Using your formula it would be 20/10, not taking into account what happens in between the price change of (100, 120) and (120, 110). However if you use Masters formula, the Profit factor will go towards 1 as volatility of the portfolio goes towards infinity. Imagine for example the equity of the portfolio throughout the backtest is suuuuper volatile, then the ratio between gross wins and gross losses would go towards 1.

[polakowo]

The term profit factor has a well-established standard usage in trading literature where it's computed over discrete trades, not price bars. This definition intentionally ignores intra-trade volatility to focus on net outcome per trade, which makes it strategy-centric rather than price-centric, and easier to interpret alongside other trade-based metrics like win rate, expectancy, Sharpe, etc. What Masters most likely proposes is a different metric, still valuable, perhaps even more risk-sensitive, but not what most traders or software refer to as profit factor.

If needed, I can consider adding a separate barwise_profit_factor method to expose this alternative logic, since it's a valid and informative angle, just under a different name.