Reading Old Predictions, Including My Own

I went back through twenty quarterly notes and scored every directional claim I had made.

The method

A claim counted if it was specific enough to be checked: a direction, a timeframe, or a conditional that either occurred or did not.

I scored each as right, wrong, or too vague to evaluate. I did the scoring before looking at which ones I remembered being right about, which turned out to matter.

The results

Nine right. Eight wrong. Six too vague.

That is close to a coin flip on the scorable half, which is roughly what should be expected and was still deflating to compute.

The part that was actually useful

I flagged my confidence level at the time of writing, because several entries used words like “clearly” or “almost certainly”.

Five entries were written with high confidence. Four of those five were wrong.

That inverse relationship between confidence and accuracy is documented in the forecasting literature and it is a different experience to find it in your own handwriting.

What I was consistently wrong about

Timing. Every structural prediction I made was directionally reasonable and early by a year or more. Institutional participation, regulatory frameworks, infrastructure maturity: all arrived, all considerably later than I said.

Bottoms. I identified capitulation in four separate quarters. In two of them the price subsequently fell substantially further.

Second-order effects. I was better at predicting that something would happen than at predicting what it would cause.

What I was reasonably good at

Identifying what mattered. The topics I flagged as important were, mostly, the ones that turned out to be important. I simply could not say when or what would follow.

That is a real and limited skill. It is useful for deciding what to pay attention to and worthless for deciding when to act.

What I changed

I stopped putting dates on structural predictions. The record shows I cannot produce them.

I stopped writing confident sentences. Not as a style choice: the record shows confidence is negatively correlated with accuracy in my own notes, so the word is information about my state rather than about the world.

I started scoring every note at the following quarter’s review, which takes five minutes and prevents the selective memory that makes everyone feel like a better forecaster than they are.

Why this matters more than the forecasts

Nothing in my actual strategy depends on prediction. Contributions are scheduled, rebalancing is mechanical, and exits are governed by written invalidation conditions.

The forecasts are a by-product of thinking about the market, and scoring them is the only way to know how much weight my thinking deserves.

The answer, established over five years and twenty notes, is: less than it feels like, and the feeling is strongest exactly when it is least warranted.

The operational side stays deliberately independent of all this. The standing order at a platform I can leave a standing order on executes whether my quarterly note was insightful or embarrassing, which is the entire reason it exists.