Which process metric predicts next month better than P&L?

[LondonScalper]

I've been testing this question on my own records for the past year, because monthly P&L is a terrible predictor of the next month. A good month is followed by a bad one about as often as by another good one.

What I tracked monthly, alongside P&L:

Percentage of trades that followed the written plan, judged from screenshots.

Number of trades taken outside my allowed hours.

Average time between a losing exit and the next entry.

Percentage of days on which I stopped at my daily limit.

Number of days with no trade taken (valid "no setup" days).

Then I looked at which of these, in one month, best predicted the next month's result in R.

With only 18 months of data, I have to be careful. But the clearest relationship was with plan adherence. Months in which I followed the plan on more than 90% of trades were usually followed by positive months. When adherence fell below 80%, the following month was negative more often than not.

The time between a loss and the next entry was also useful. When that average fell below ten minutes, the next month tended to be poor. I read that as a sign that I'm getting impatient, which usually gets worse before it gets better.

P&L itself predicted almost nothing.

I don't want to overclaim from 18 data points. But I now look at plan adherence first on the monthly review, and P&L second.

[Shadow Trader]

Your finding on time between a loss and the next entry matches mine, with one twist. In my records the average hides a lot. What predicted a bad following month was the number of re-entries within three minutes of a loss, not the mean gap.

A month with two or three of those was normal. A month with eight or more was almost always followed by a worse one, even when the month itself was profitable.

On plan adherence, I'd be careful with the 90% threshold given how it's measured. Judging from screenshots after the fact, I tend to be generous with myself on the trades that won. I started having the screenshot reviewed before looking at the outcome, with the P&L column hidden.

That changed my adherence figures by about 5 points in some months, which would have moved several of them across your 80 and 90 lines.

One metric I'd add: trades taken in the first five minutes after a big win. That's where I see the opposite of revenge trading, the overconfidence trade, and it's been as costly for me.

Did the days with no trade show any relationship at all? I expected them to, and in my data they didn't.

[Fairman]

With only 18 months of data, I have to be careful.

That caution is warranted, and there are two specific traps worth naming. First, with five metrics tested against the next month, one of them looking predictive across 17 month pairs is fairly likely by chance alone, so plan adherence would ideally be confirmed on the next 12 months before treating it as settled. Second, the direction might partly run the other way. Choppy, difficult markets make it harder to follow a plan and can also make the following month harder if the conditions persist. One way to separate those is to check whether adherence still predicts the next month after allowing for something like average daily range. If it does, it's a stronger case that the behaviour itself is what matters.

[Shadow Trader]

Following up on the 18 data points problem. One way I've tried to get more out of limited history is to use weeks instead of months. You get about four times the observations, and while weekly P&L is noisier, weekly plan adherence is surprisingly stable.

When I did that, the relationship held, but it was weaker. Weeks above 90% adherence were followed by a positive week about 60% of the time, against roughly 50% for the rest. Useful, but not the strong signal the monthly view suggested.

Fairman's point about chance with five metrics over seventeen pairs is the one that stuck with me. I split my own record into the first half and the second half, and only kept a metric if it still pointed the same way in both. Adherence survived that cut. Time between losses did not.

I'd still put adherence first in the review, just for a different reason. It's the only metric on the list I can actually control on a given day.

Has anyone else tried a split sample check like that, or do you wait for a full extra year before trusting a process metric?