A market can read the race, not run it

On Monday the odds that Graham Platner drops out of Maine's Democratic Senate primary jumped from 9 percent to 96 percent on a single scandal. In the same hours, the odds that Democrats hold the seat rose about nine points. Alex Tabarrok ran the arithmetic: if an 87-point swing in Platner's chance of quitting buys nine points of winning, his leaving outright is worth roughly ten. The market's read is that Platner is costing his party about ten percentage points — and, the argument goes, the party should use that to decide.

That last clause is where to slow down. Reading a race and running one are different jobs, and a price that is good at the first is not automatically fit for the second.

Tabarrok knows the ten-point figure is soft — it is reverse-engineered from one unconditional price that happened to lurch 87 points in a day. His fix, from Robin Hanson, is to stop inferring the conditionals and trade them directly:

The refund is the clever part. It makes the price a clean conditional probability — how the race goes if Platner leaves, held apart from whether he actually does — and posts it continuously, no scandal required. As measurement, that is a real gain. You would know what a candidate costs before the news hits, which is exactly when the number is worth most.

A forecast you watch and a forecast you obey are different instruments.

Concede all of it. The measurement problem is solved. But measurement was never why we don't let markets pick our candidates. The reason is what happens to a number the moment it is load-bearing. A forecast you watch and a forecast you obey are different instruments. Once the conditional price decides whether Platner stays, the price becomes the prize — and a contract keyed to one named person's political survival, thin by construction, is among the cheapest things in politics to move. You don't have to be right about the race. You have to nudge a lightly traded conditional a few points on the afternoon the call gets made. The 87-point lurch Tabarrok reads as signal is the same sensitivity a motivated party would rent.

And a forecast cannot stay neutral about a decision it is now making. "Democrats do better without Platner" stops being a prediction and becomes an instruction — and once the instruction is followed, it can never be checked. He drops out, the other branch never runs, and the market is never scored on the only question that mattered. A prediction market is disciplined by reality catching up to it. A decision market spends that discipline down: it kills the outcome it was predicting.

The ten-point number is worth having. Read the world with it. The mistake is thinking that because you can price a decision, you should let the price make it.