Profit Factor: A Quick Number With Hidden Weaknesses

[Fairman]

Profit Factor: A Quick Number With Hidden Weaknesses

Profit factor is gross profit divided by gross loss. If your winning trades total 3,000 dollars and your losing trades total 2,000, the profit factor is 1.5.

It is popular because it compresses a system into one figure. Above 1 means profitable before considering other factors. Many consider 1.5 to 2 decent, though what is reasonable depends on the strategy and trade count.

But the number has weaknesses. It ignores the sequence of trades, so two systems with the same profit factor can have very different drawdowns. It also depends on sample size. A profit factor of 3 from 20 trades means little, while 1.4 from 500 trades means more.

A single large winner can distort it. Remove the best trade, and a profit factor of 2 can fall to 1.1. Always check the figure without the top one or two trades.

It also hides costs. If spreads and commissions are not included in the data, the real profit factor is lower. And it says nothing about risk per trade or position sizing.

Better practice is to view it with other measures: expectancy in R, maximum drawdown, win rate, and number of trades.

A high profit factor from a martingale system, for instance, hides the tail risk.

Practical step: calculate the profit factor from your journal with and without your best trade and compare.

[Fairman]

A weakness I'd flag: profit factor ignores how many trades it took. A 2.0 profit factor over 15 trades doesn't mean much. A 1.3 over 300 trades is a lot more trustworthy. Always check the trade count alongside it. It also hides how the profit is distributed. A profit factor of 1.5 where one huge winner made most of the gross profit is very different from 1.5 made of many steady wins. Remove the single best trade and recalculate. If the profit factor drops below 1, the strategy depends on rare outliers. That might still be fine for a trend system, but you should know it before trading it, since skipping that one trade would ruin the year.

[Fairman]

Profit factor also hides drawdown. You can have two systems showing an identical 1.5 and wildly different worst losing streaks. I'd always look at max drawdown alongside it before deciding which one to trade. One system might reach 1.5 with steady small losses and a worst drawdown of 8 percent. The other might hit 1.5 after a 30 percent drawdown halfway through the year. Both look the same in the summary, but most people would abandon the second one during that drawdown. A recovery factor, net profit divided by max drawdown, captures this better.

[FXS Prop Desk]

The sample size weakness is the key point. A quick illustration: with 20 trades, one large winner can move the profit factor from 1.2 to 2.5. Looking at profit factor together with maximum drawdown and trade count, or calculating a confidence interval by resampling the trades, gives a much more honest picture.

It's also worth checking profit factor separately by year or quarter. A system that's 1.6 overall but 3.0 in one quarter and 1.1 in the others depends on one period.

[Shadow Trader]

FXS Prop Desk's resampling idea is cheap to run and changed how I read my own figure. My EURUSD London log has 180 trades with a profit factor of 1.45, and 5,000 random resamples of those trades with replacement gave a middle 90 percent range of about 1.13 to 1.87.

img_12961_r1.png (60.21 KiB) Viewed 3 times

The chart shows that spread next to the same test run on only the first 40 trades. With 40 the range ran from 0.83 to 2.48, so a lucky month and a broken system would look almost the same.

The lower bound is the number I plan with now. If it sits near 1.0, I treat the system as unproven however good the headline looks.

Costs move the result more than people expect on a scalp. Taking just 0.1R per trade of extra spread off the same 180 trades drops the profit factor to about 1.17, because every winner shrinks and every loser grows by the same amount.

Fairman's remove the best trade check is still the first thing I run. Resampling just puts a range around what that check hints at.