Regression to the mean in monthly results
Posted: Wed Oct 28, 2026 6:10 am
A trader has a spectacular month, up 12%. Next month, up 2%. Disappointment. Another has a terrible month, down 8%, followed by a modest recovery. Relief. Both are experiencing regression to the mean.
When results combine skill and luck, extreme outcomes tend to be followed by more average ones, because the luck component doesn't repeat. A great month probably included favourable randomness. A bad one probably included unfavourable randomness.
Misreading the effect leads to errors. After a great month, the trader may attribute the result to a change made at the time, and keep making that change. After a bad month, a random tweak is followed by a recovery that would have happened anyway, and the tweak gets the credit.
Superstitions form this way. A "lucky" routine, a new indicator or a new schedule appears to work because it coincided with regression.
The antidote is looking at longer windows and at the statistical distribution of monthly results. Knowing the typical spread of outcomes makes the extremes less surprising.
A practical technique is comparing each month with a "shrunk" expectation: if the long-run average is 2% and this month produced 12%, a reasonable forecast for next month is something like 2% plus a small fraction of the surprise, not 12%. The same applies on the downside, where a bad month shouldn't push expectations to zero.
Writing the shrunk forecast in the journal each month, and checking it afterwards, builds a feel for how much of each result was repeatable.
What's the typical range of monthly results, and how often have extremes been followed by averages?
When results combine skill and luck, extreme outcomes tend to be followed by more average ones, because the luck component doesn't repeat. A great month probably included favourable randomness. A bad one probably included unfavourable randomness.
Misreading the effect leads to errors. After a great month, the trader may attribute the result to a change made at the time, and keep making that change. After a bad month, a random tweak is followed by a recovery that would have happened anyway, and the tweak gets the credit.
Superstitions form this way. A "lucky" routine, a new indicator or a new schedule appears to work because it coincided with regression.
The antidote is looking at longer windows and at the statistical distribution of monthly results. Knowing the typical spread of outcomes makes the extremes less surprising.
A practical technique is comparing each month with a "shrunk" expectation: if the long-run average is 2% and this month produced 12%, a reasonable forecast for next month is something like 2% plus a small fraction of the surprise, not 12%. The same applies on the downside, where a bad month shouldn't push expectations to zero.
Writing the shrunk forecast in the journal each month, and checking it afterwards, builds a feel for how much of each result was repeatable.
What's the typical range of monthly results, and how often have extremes been followed by averages?