Good Bets Lose

A position lost money. That fact, on its own, tells you almost nothing about whether buying it was right.

The mistake has a name. Annie Duke, who spent two decades playing poker professionally before writing about decision making, calls it resulting: judging the quality of a decision by the quality of its outcome. Poker teaches it out of you quickly, because the game deals you a losing river card often enough that you would go broke playing only the hands that worked. Markets teach the same lesson far more slowly and charge more for the tuition.

So the question we ask after a loss is not whether it worked. It is whether, knowing only what we knew at the time, we would make the same bet again. If the answer is that we would do it ten times out of ten, the loss was the price of a bet worth making. Nothing about the outcome makes that untrue.

This is what naming the risk factors in advance is actually for. A named risk that then happens is not a surprise and not an indictment — it is the distribution doing what distributions do. The error would have been failing to name it, or sizing the position as though it could not occur. Both of those are visible in the written thesis, which is why the thesis gets written before the position exists.

The harder half of this is the winner that was a bad decision. A stock goes up and the P&L says you were right, but the reason it went up appears nowhere in your thesis — a bid arrived, a commodity moved, rates fell, someone else’s thesis paid you. You got the money. You did not get the information.

That is the genuinely dangerous box, because it teaches you something false and pays you for the lesson. Do it enough and you build conviction in a process that has never actually worked, and you find out at the worst possible moment, in the largest possible size.

Which is why the post-mortem runs on winners too. Most firms autopsy losses only, and it feels rigorous — but it examines half the sample and leaves out the half most likely to be teaching bad habits. If a position worked for a reason that was never in the thesis, we would rather record that as luck than let it accumulate as evidence of skill.

None of this is operable without a written record. Memory does not hold decisions; it holds outcomes, and then quietly reconstructs the reasoning to fit them. You cannot separate we were right from we got paid two years later from recollection alone. You can from a document written before you knew.

There is a sample-size problem underneath all of this. Over a small number of decisions, luck swamps everything — a good process can look broken for years and a bad one can look brilliant for just as long. Only over enough repetitions does the process show through the noise, which is another reason a structure with no fund clock matters: you need enough draws for skill to become visible, and a forced exit date can arrive before it does.

It also changes what a reversal means. New information that genuinely breaks a thesis is a reason to sell without embarrassment, not evidence that buying was a mistake. The original decision is judged against the information available when it was made. The current decision is judged against today’s. Confusing the two produces people who defend dead theses to protect a record that was never at stake.

We would rather make good decisions and occasionally lose than make bad ones and occasionally get paid. Over enough decisions, only one of those compounds.