PF can already be used, as is, in "the domain of non-constant position sizes" as long as it's calculated correctly. PF will change for the same system with and without position sizing but given what it measures (how many $ your system has made for each $ lost) , it should. Faith's argument for normalizing it is to remove "the skew of recent history dominating the test." But I think it's a bell/whistle to help him sell Veritrader and personally prefer the original PF, in conjunction with other metrics.Quote from horribilicus:
Curtis Faith and Chuck Branscomb came up with a fairly slick way to extend "Profit Factor" into the domain of non-constant position sizes. Rather than take the ratio of (Total_Dollars_Won) / (Total_Dollars_Lost), they invented "Percent Profit Factor" which takes the ratio of (Total_Percent_of_Account_Won) / (Total_Percent_of_Account_Lost). The attached image, swiped off Curtis's website, shows the math more precisely.
