Risk Is Not Uncertainty: The Distinction That Changes Your Sizing

Risk is a roulette wheel. You do not know where the ball lands, but you know the wheel: thirty-seven pockets, one zero, and a house edge you can compute to four decimal places. The outcome is unknown. The distribution is not.

Uncertainty is a wheel you have never seen, in a room you cannot enter, and somebody tells you it has about thirty-seven pockets.

Traders treat these as the same problem. They are not, and the difference is measurable.

The same system, two states of knowledge

You win forty percent of the time with 2R winners, risking two percent a trade. Five hundred trades. Ruin means losing half the account.

In the first world, the forty percent is known. Handed to you, verified, fixed for all time.

In the second world, forty percent is your best estimate. It might be thirty-five. It might be forty-five. On average, across all the versions of you, it is exactly forty, and your expectancy is identical.

Your expectancy calculation cannot see this. It takes a point estimate and returns a point estimate, and it will report the two worlds as identical, because on average they are.

You do not live on average. You live along one path, and on the path where your true win rate turned out to be thirty-four percent, the two-percent risk you chose was a sentence.

Ruin is convex in uncertainty

Vary how badly you know your own edge.

Doubling your uncertainty from two points to five multiplies ruin by ten. It is not linear, it is not close to linear, and no amount of accuracy in your expectancy figure compensates for imprecision in the inputs.

Which produces an uncomfortable conclusion. The most important number in your risk model is not your edge. It is your uncertainty about your edge, and it is the one number nobody computes.

Where uncertainty hides

Your sample is small. Two hundred trades sounds like a lot. It gives you a win rate with a confidence interval several points wide in each direction, which is precisely the five-point row.

Your market is not stationary. The win rate you measured last year was measured on a market that no longer exists. This is not estimation error. It is the parameter itself moving, and it moves in the direction of your edge decaying, because edges crowd.

You selected the sample. The trades in your journal are the ones you took, chosen by a process that included your judgement, your mood, and the trades you decided to skip. That is not a random draw from your setup’s population.

You have never traded the regime that is coming. A strategy tested across two years of one market character has a win rate for that character, and a completely unknown one for the next.

Pabrai’s asymmetry

Here is where the distinction stops being a warning and becomes a strategy.

Pabrai’s framing is that the best positions are low risk and high uncertainty. Not low risk and low uncertainty, which is a savings account, and not high risk of any kind.

Low risk means the downside is bounded and known. You can compute what you lose if you are wrong, and it is survivable. High uncertainty means nobody can compute the upside, which is exactly why it is available to you at a price that ignores it.

Markets price risk. They do not price uncertainty, because uncertainty cannot be priced. It can only be avoided, and the crowd avoids it, which is what leaves it lying there.

For a trader, the translation is direct. A setup with a bounded, mechanical stop and an unbounded, unknowable target is the correct shape. A setup with a precisely known reward and an ill-defined loss is the wrong shape, no matter how good the expectancy looks, because the expectancy was computed on the half of the trade you understood.

What follows for sizing

Size for the win rate you might have, not the one you measured. If your estimate is 45% with a five-point interval, size as though it were 40%. The cost is a little growth. The alternative is the 9.2% column.

This is the real argument for half-Kelly. Kelly is optimal under risk. Trading is conducted under uncertainty. Halving the fraction is not timidity, it is the correction term for a parameter you do not possess.

Distrust precision. A trader who states his expectancy as +0.183R has told you something about his software and nothing about his edge. The third decimal is decoration on an estimate whose first decimal is uncertain.

Prefer bounded losses to known probabilities. Between a setup whose probability you know well and a setup whose loss you control absolutely, take the second one. Probabilities drift. A stop does not.

The four-hundred-year point

Bernstein’s history of risk is, read carefully, a history of one mistake made repeatedly by clever people: the belief that having measured something, you have understood it.

Measurement converts uncertainty into risk on paper, and nowhere else. The number in your spreadsheet is not the wheel. It is your sketch of the wheel, drawn from a distance, in a room you were never allowed into.

Trade the sketch, and size for the room.