Journal check: average R after a win against after a loss, on a made up 20 trade list

[Fairman]

Journal check: average R after a win against after a loss, on a made up 20 trade list

A quick check I like to run on a journal is to split trades by what happened right before them. Here is a made up list of 20 trades in R, not my results, to show the method: +1.5, -1, -1, +1.5, +1.5, -1, +1.5, -1, -1, -1, +1.5, +1.5, -1, +1.5, -1, -1, +1.5, -1, +1.5, -1.

The total is +2.5R from 9 winners, a 45% win rate. Now take only the trades that came directly after a win. There are 9 of them, they sum to minus 4.0R and average minus 0.44R, with only 2 winners. The 10 trades that came directly after a loss sum to plus 5.0R, average 0.50R, with 6 winners.

On the face of it, this list says the trader does worse after a win, maybe from getting loose after a good trade. Before believing it, check the noise. A single trade here has a standard deviation around 1.25R, so the average of 9 trades has an error of about 0.42R and the average of 10 about 0.40R. The gap between the two averages is 0.94R, and the error on the gap is about 0.58R. That's a ratio of 1.6, short of the usual 2.

A gap that size can easily come from luck in 20 trades. To say much you need roughly 50 trades in each bucket, and the gap has to be about 0.5R.

The method is still useful. It's cheap, and it points at one specific habit to watch once the sample grows.

[Fairman]

The arithmetic for the sample size is simple. The error on the difference between two bucket averages is the trade standard deviation times the square root of (1/n1 + 1/n2). With a 1.25R standard deviation and 50 trades in each bucket, that's 1.25 x the square root of 0.04, which is 0.25R. A gap needs to be about twice the error to be taken seriously, so a 0.5R gap is the smallest that 50 per bucket can show. For a 0.25R gap you'd need about 200 per bucket. So the check has most use for large habits, and not for small differences.

[Fairman]

Choosing the buckets is where people go wrong. After a win, after a loss, after two losses, after a news day, before lunch, on Fridays: each cut gives another comparison, and if you try 20 cuts then some will look strong just from chance. With 20 independent comparisons, the chance that at least one differs by more than two standard errors by luck is 1 minus 0.9545 to the power 20, roughly 60%. So decide the two or three buckets before you look at the numbers, and write them down.

[Fairman]

When the sample is too small, the other way to use the idea is as a rule check and not as a statistic. After a win, list the trades that broke a rule, for example size larger than plan or an entry outside the plan's hours. Count them. That's 9 trades after wins in the example, so you can look at each trade on the chart in a few minutes. If four of nine broke a rule, that tells you more than the average R, because rule breaks are visible and can be fixed.

[Fairman]

If a pattern holds up with a bigger sample, the fix can be a mechanical one. For example, after a winning trade, wait for the next full candle or set a minimum gap of 15 minutes before the next entry, and see whether the after-win average moves toward the other bucket. Record the same statistics for the next 50 trades. The comparison is then a test of a change, which is much better than an observation of a habit. Keep the counts in two columns in the journal, one for each bucket, and update after each session.