nucleation

LIT-958: How manipulable are prediction markets?

2 min read

Properties5
StatusActive — read 2026-10-14 (NOTE-778) in full; worth reading as a randomised, pre-registered field experiment that answers the counterfactual question that historical manipulation studies could not. On 817 Manifold markets about 59% of a random ±5-point shock was still present after 60 days in unresolved markets, and a price-impact model with finite, risk-averse traders explains partial persistence even when no trader learns. Read it knowing that it cannot say whether traders came to believe the shocked price, that the platform runs mostly on play money, and that its evidence on which markets resist manipulation is correlational. Filed for QUESTION-047; THEORY-267 holds what it shows.
Version2
Tagssocial-science, epistemology
Cited as “paper” byNOTE-778 — Rasooly and Rozzi — How manipulable are prediction markets?
Cited as “source” byTHEORY-267 — In Rasooly and Rozzi's pre-registered randomised experiment on 817 Manifold play-money markets, about 59% of a ±5-point price shock that no trader's information supported was still present after 60 days in unresolved markets, with about 25% undone in the first week and reversion slowing after it; their model of finitely many risk-averse traders gives such partial, slowing reversion even with no learning, and the data cannot tell that from learning, THEORY-810 — Prediction-market prices keep much of an arbitrary price shock that no trader's information supports for weeks, rather than correcting it within a short transition phase

Graph View

Itzhak Rasooly and Roberto Rozzi (2025), arXiv:2503.03312.

Key takeaways

From the reading, NOTE-778.

  • Random shocks persist. The yes–no gap was 0.0748 after a week, 0.0686 after 30 days, and 0.0586 after 60 days in markets not yet resolved, from a starting gap of 0.10. All are significant at p < 0.01.
  • Reversion slows. 25% of the gap reverted in week one and 6% in weeks two to four, the shape the model predicts.
  • Persistence without learning. In the model every trader leans against a shock (Lemma 3), but finite, risk-averse traders lean only partly. With 11 traders and no learning, about 60% of a shock remains. Learning from the price makes it stick more, and more traders or a common prior make it stick less (Table A1).
  • The moderators are weak evidence. Reversion is higher in markets with a Metaculus twin, more 24-hour volume, more comments or more traders. But the median splits have p between 0.04 and 0.19, and the traits were not randomised.
  • Robust within the platform. Dropping possibly related markets changes nothing. A small real-money-convertible follow-up (110 markets) shows the same pattern, with faster reversion.
  • What is not known. Whether the persistence comes from traders updating on the price or from limited capacity to undo it.

Standing in the record

Filed on 2026-10-12 for QUESTION-047 and read on 2026-10-14 in NOTE-778. What it shows is held as THEORY-267. For the question it is a measured case of a community-level estimate holding a value that no member reasoned to. It is not a case of a community belief: a price aggregates positions, and whether any trader holds the persisting value is the open point.

Backlinks