Why Confidence Is the Wrong Thing to Track
Ask someone how a trade is going and they will usually answer with a feeling. High conviction. Strong view. I really like this one. The feeling is doing a lot of work in that sentence, and almost none of it is the work you need. Confidence is a report on your internal state. It tells you how sure you are. It does not tell you whether being that sure was warranted, and those are different questions with different answers.
The thing worth tracking is calibration: when you say seventy percent, do the seventy-percent things actually happen about seven times in ten. A person can be supremely confident and badly calibrated, which is the dangerous combination, because the confidence funds the position and the miscalibration pays for it. The reverse is fine. Low confidence and good calibration just means you size small and stay solvent. You can build on that. You cannot build on a loud voice that is wrong at a stable rate.
Part of the problem is that we store a view as a point. A single number, a single direction, a thing that will either be right or wrong. But a view is never a point. It is a distribution. When I say I think something is more likely than not, I am not predicting one outcome, I am describing the shape of many. The honest version of a forecast has a width to it. The dishonest version collapses that width into a story, and a story has no error bars.
A forecast is a distribution. A position is a point drawn from it. That gap is where most of the damage lives, because we feel the conviction of the distribution’s center and then take the risk of its tails.
This is easiest to see if you stop predicting and start repeating. Suppose you have a real but small edge, the kind that makes you right a little more than half the time. One belief, held once, tells you almost nothing about that edge. Held a hundred times, it tells you everything. The single bet is a coin you cannot read. The thousand bets are a distribution you can.
Every line above is the same belief, the same edge, run forward again and again. Notice what conviction would have told you and what actually happens. The edge is real, so most of the runs drift up and the median ends in the green. But the edge is small, so the fan is wide, and a meaningful share of the paths spend time underwater. Some never recover inside the window. Drag the edge down and watch the median sag toward the line while the ruin paths thicken. Drag it up and the whole fan lifts, but the spread never disappears. It cannot. The spread is the truth about a small edge. Confidence is what you feel looking at the median; the rest of the fan is what you are actually exposed to.
[McKinley: insert the specific trade or race where the conviction was high and the calibration was off, and what the realized distribution looked like versus the point belief.]
What changes when you track calibration instead of confidence is mostly boring, which is the point. You start writing down probabilities before you know the result, so you cannot launder a loss into a near-miss after the fact. You start sizing to the width of the distribution rather than to the strength of the feeling. You stop being impressed by a single good call, because one draw from a wide distribution is noise wearing a costume. And you get slightly suspicious of your own certainty, since certainty is exactly the signal that tends to be miscalibrated. The feeling of being sure is not evidence. It is weather.
None of this makes the bet safer. The fan stays wide because the world stays uncertain. What it changes is which number you steer by. Conviction asks how much you believe. Calibration asks whether your belief has been right at the rate you claim. Over one bet, the first question feels more important. Over a thousand, only the second one survives.