One in Three Prediction Market Moves Reverses Within Four Hours

Prediction markets are supposed to price a headline once and hold it. The news breaks, traders read it, the price adjusts, and the new price stands. If that were the whole story, the first move after a headline would be the move, and holding it would be the default.

It is not the default. We took every reaction of two cents or more that we measured on Polymarket between April 29 and June 25, 2026, more than 60,000 in all, and sorted each one by the shape it traced over the four hours after its headline. They fall into just five recurring trajectories.

The clean move that ramps and sticks, the shape most people assume every reaction has, is barely one in three. The other four do something the folk model does not allow: they grind up late, blow clean past their first peak, fade back toward nothing, or turn and cross to the other side of where they began.

In prediction markets, the shape of the move is the message.

The One in Three Moves Is Fake study · Vera Research

Three of the five hold or extend the move. Snap-and-hold (29.1%) ramps toward its peak inside the first hour and keeps it. Slow-grind (25.3%) barely moves early, then reprices late. Accelerating (16.2%) never settles: by hour four it sits far beyond its early peak. Together these three are 70.6% of measured moves, the reactions that end up meaning what they first appeared to mean.

The other two round-trip. Spike-and-fade (21.7%) pops fast, then slides back to roughly zero by hour four, the same fade equity traders watch after an earnings pop. Reversal (7.7%) does not merely give the move back: it ends hour four on the opposite side of where it began. Together they are 29.4% of everything we measured. Nearly one in three moves hands itself back.

The round-trip share does not depend on our five groups. Raw counting says the same thing. Of every hundred moves we measured, 27 keep less than half of their peak by hour four, and 22.4 finish flipped against their own peak, ending on the far side of where they started. The median move keeps just 0.750 of its peak. Two independent re-derivations of the five groups, run from scratch with a different random seed, match the original at 0.985 out of 1.00, so the taxonomy is not an artifact of one run.

The obvious question is which news makes which shape. The answer is almost none of it. Every news category we track produces all five shapes, in broadly similar proportions. Geopolitics does not spike more than macro; economics does not grind more than politics. Knowing a headline’s topic tells you little about whether its move will stick, and the folklore that “geopolitics spikes and macro grinds” does not survive the measurement.

What survives is simpler, and less comforting. The first price you see after a headline is often the worst information you will get, and its category will not warn you which kind of move it is. The shape is what carries the signal, and the shape only reveals itself with time.

1. More than 60,000 measured Polymarket reactions (every move of two cents or more, April 29 to June 25, 2026) sort into just five recurring shapes over the four hours after each headline.
2. Three shapes hold or extend the move: snap-and-hold (29.1%), slow-grind (25.3%), and accelerating (16.2%), together 70.6% of moves.
3. Two shapes round-trip: spike-and-fade (21.7%) and reversal (7.7%), together 29.4% of moves, nearly one in three.
4. Counted directly, with no clustering: 27% of moves keep less than half their peak by hour four, 22.4% finish flipped against it, and the median move keeps 0.750 of its peak.
5. The five groups reproduce from scratch under an independent audit with a different random seed, matching the original at 0.985 out of 1.00.

The news category does not predict the shape: every category produces all five shapes in broadly similar proportions, so knowing a headline’s topic tells you little about whether its move will hold or hand itself back.

Built on the Vera dataset: more than 60,000 measured Polymarket reactions of two cents or more, April 29 to June 25, 2026, each classified by the shape it traced over the four hours after its headline. Shares come from a five-cluster analysis of peak-normalized trajectories, re-derived from scratch by an independent audit at 0.985 out of 1.00.

Odds shown are each market’s own price at the time Vera flagged the story, stated as fact, not advice. Nothing here is a recommendation to buy, sell, or trade any market or asset. Vera and Crypto Briefing are not registered investment advisers.