Correct forecasts get retold. They appear in conversation, in explanations offered to oneself, and in accounts of why a particular decision worked out the way it did. Each retelling deepens the trace, and the frequency of retelling correlates directly with having been right, since nobody repeatedly recounts the occasions on which they were wrong. The mechanism preserving these memories is therefore directional from the outset, and it operates without anyone choosing to make it operate. Nothing about it requires vanity, and it works equally well in people who consider themselves modest.
Incorrect forecasts have no equivalent mechanism working on their behalf. There is no occasion that calls for them, no conversation requiring their citation, and no party with any motive to keep them available for inspection. Nothing is deliberately erased at any point. They merely lack the conditions needed to stay in circulation. After some period they drop out of retrievable memory without leaving any sign of having gone, and the absence of a sign is the part that matters most.
A further piece of processing happens to incorrect forecasts before they fade entirely. They tend to undergo a redescription. The direction was right and only the timing was off. The market behaved irrationally for longer than anyone could reasonably have expected. Some external factor intervened that nobody could have anticipated. Explanations of this kind are occasionally true, and their occasional truth is precisely what makes them effective in the cases where they are not. They convert a failure of judgement into an episode of poor luck.
Taken together, the two mechanisms produce a personal record that has been systematically pruned. Apparent accuracy within that record runs well above the actual figure, and the record is the only one available for consultation when the question arises. Estimates of one's own ability rest entirely on this material. The material has already passed through two rounds of filtering before anybody looks at it, and neither round announced itself while it was happening. What remains presents as a complete history.
The consequences extend past inflated self-assessment. Position sizing depends on an estimate of how reliable one's judgement is, so an inflated estimate feeds directly into how much money is exposed to it. Tolerance for opposing views depends on it as well, since somebody who believes their accuracy is high has correspondingly less reason to take a contrary argument seriously. The error therefore propagates into two places where it does real damage, instead of remaining a matter of self-image. Both places involve money, and neither is visible from inside the belief that produced them.
Externalising the record is the only correction that works. What is required is not elaborate: a simple list holding the forecast, the date, the confidence attached at the time, and what subsequently happened. The confidence column is where the value sits, because the property worth examining is calibration rather than raw accuracy. High-confidence forecasts ought to perform better than low-confidence ones, and whether they actually do is an empirical question with a checkable answer. Almost nobody has checked it on their own record.
After a year or two of keeping such a list, most people arrive at the same finding. Directional accuracy comes out roughly where expected, sometimes slightly better. Calibration comes out far below expectation, and the calls held with the greatest conviction turn out to have performed much like the rest. That discovery usually influences subsequent behaviour more than any individual forecast ever did. It changes the sizing rule itself, and a rule applies to everything that follows it. One corrected forecast improves one decision, while one corrected rule improves every decision made under it.