Profit Factor and the Best-Trade Test in One Sheet

[Fairman]

Profit Factor and the Best-Trade Test in One Sheet

Goal: compute profit factor, and then test how much it depends on your best trade.

Profit factor: gross profit divided by gross loss, both measured in R, with the loss as a positive number.

=SUMIF(A2:A101,">0")/ABS(SUMIF(A2:A101,"<0"))

Test with the same ten results as before: gross profit 7.5, gross loss 6.0. Profit factor 1.25.

Now the best-trade test. Remove the single best result from the gross profit:

=(SUMIF(A2:A101,">0")-MAX(A2:A101))/ABS(SUMIF(A2:A101,"<0"))

With the best trade, plus 3, removed, gross profit becomes 4.5, and the profit factor falls to 0.75. A system that looked profitable becomes a losing one when a single trade is excluded.

Interpretation. The profit depends heavily on that one outlier. That is typical of trend-following results, but it also means the statistic has little robustness with ten trades.

Add a second variant that removes the top two trades, using LARGE:

=(SUMIF(A2:A101,">0")-LARGE(A2:A101,1)-LARGE(A2:A101,2))/ABS(SUMIF(A2:A101,"<0"))

Compare the three numbers in a small table. A profit factor that stays above 1 after removing the best two trades, with a sample of 100 or more, is more convincing than one that collapses.

Also add the share of total profit coming from the top 10 percent of trades.

Exercise: compute the three profit factors for your journal and write which one you trust.

[Shadow Trader]

Comparing two brokers sounds simple. Trade the same setups on both and see which account ends higher. The trouble is that a lot of versions of this test are unfair before the first trade is taken.

Approach one is the sequential test: a week on broker A, then a week on broker B. It is easy, and it is the weakest design. Market conditions in those two weeks will differ, so any gap in results mixes broker effects with market effects. A quiet week against a news-heavy week says almost nothing about fills.

Approach two is the parallel test: the same signal fired on both accounts at the same moment. This isolates the broker, because the market is identical. The costs are some setup work and the fact that the second order inevitably goes out a fraction later unless both are triggered by software at the same instant.

Where parallel testing goes wrong is in the small details. Different lot size rules, different contract specs, different server time zones and different swap or commission schedules all muddy the comparison. If one account is demo and the other live, the test is void.

What gets compared matters as well. Final profit is a poor metric on a small sample, since a handful of trades can swing it. Better numbers are the difference in entry price on each trade, the difference in exit price, spread at the moment of entry, and the count of orders rejected or requoted. Those are measured per trade, so a few dozen trades already say something.

Order of execution should be randomized. If broker A always gets the order first, it will get the better price a bit too often. Alternating which terminal goes first removes that bias, and it costs nothing.

Finally, a result from one pair at one hour is a result from one pair at one hour. Repeating the test across the sessions that actually get traded is what makes it useful, and it is also what shows whether a difference is real or just a quirk of one liquidity window.

Which metric has anyone found most revealing when comparing two brokers, and how many trades did it take before the difference looked real?

[Fairman]

The collapse from 1.25 to 0.75 after removing a single trade shows how fragile statistics are on small samples. With ten trades, one outlier can dominate everything. With 100 trades, removing the best one usually changes the profit factor much less, unless the strategy genuinely depends on rare large wins. Running the test on increasing sample sizes as the journal grows shows how quickly the result stabilises. If the profit factor without the best trade stays well below the full figure even at 100 trades, the strategy's edge is concentrated in a few outliers, which matters for how you manage it.

[Amanhasnoname]

I ran this exact test on my own journal last spring and it was a humbling afternoon. Over about 140 trades my headline profit factor was 1.34, and I was quietly pleased with it. With the best trade removed it fell to 1.12, and with the top two gone it was 0.98. A single GBPJPY move that ran far past my target had been carrying most of the result.

What I took from it is that the figure to trust is the one after removing the top two, provided the sample is above a hundred. I keep all three numbers in a small table at the top of the sheet now, and I pay more attention to the gap between them than to the first number. A narrow gap means the edge is spread across many trades, and a wide one means I'm leaning on luck.

One thing I'd add is to check whether that best trade came from an exit I planned or from a hold I hadn't intended. If I only stayed in because I froze, I don't count it as evidence for the system.