Can prediction markets be manipulated? What the experiments show

A public number invites a thumb on the scale. A price people watch — that journalists quote, that decision-makers glance at — is worth distorting to anyone who benefits from the audience believing it.

So the question deserves better than reassurance. Unusually, it has better. Manipulation of prediction markets has been attempted deliberately, by researchers, repeatedly, for a quarter-century — in laboratories, on live exchanges, and at scale. The attempts were run at the researchers’ own expense and the results published either way, including the results that went against the mechanism.

Prediction-market prices can be pushed, and researchers have pushed them deliberately to find out what happens. Where a market is active and the manipulator is one voice among many, the push has repeatedly failed to hold: documented attempts on live exchanges moved prices sharply and reverted inside a day. Where the manipulator is well funded, the book is thin, or the attack targets a short settlement window rather than the forecast, it holds: one laboratory study degraded accuracy outright, and the largest field experiment left traces detectable two months on. Manipulation-resistance is therefore not a property of the category but of depth, attention, and contract design.

What a push costs

Start with the mechanism, because the mechanism is the defense.

A market price only moves when someone trades (Where a market price comes from). A manipulator cannot argue the number upward; they have to buy it upward, accumulating a position at prices they believe are wrong. That position is the cost of the distortion, and it is also the bait: a price pushed away from the evidence is, by construction, offering better-than-fair terms to everyone who disagrees.

The observation has a formal version, and it is stronger than mere self-correction. Hanson and Oprea (2009) modeled a manipulator with a private preference about where the price should sit. Their result runs the wrong way round. When other traders are uncertain about the manipulator’s target, the average target has no effect on price at all. Raising the uncertainty about it can even raise average accuracy: the manipulator’s noise increases what informed trading is worth, and so pays more people to become informed.

That is a theorem, not a promise. Whether it describes real books is an empirical question — which is what the experiments are for.

The laboratory record

The lab is where manipulation can be studied with the manipulator’s incentive known to the researcher, because the researcher wrote it.

The founding experiment paid people to distort. Hanson, Oprea and Porter (2006) ran a double-auction market with 96 subjects across eight sessions. In half of them it was common knowledge that some traders were being paid in proportion to the closing price — a standing incentive to push the number up, with nobody knowing which of their counterparts held it. The manipulators tried. They failed: prices were no less accurate than in the control sessions, because the traders without the incentive adjusted the terms on which they were willing to trade, and the adjustment absorbed the push. In that setting, a market that knew it might be manipulated defended itself by repricing the offers it would accept.

The counter-result matters just as much, and it arrived from the same laboratory tradition. Deck, Lin and Porter (2013) tested the setting the first experiment did not: a manipulator who is well funded, single-minded, and pushing a price that a decision-maker will actually use. There, manipulation worked. The authors report that such manipulators “can in fact destroy a prediction market’s ability to aggregate informative prices.” One detail of their result matters more than the headline: the damage concentrated in executed trades, while the standing bids and asks remained informative. What the manipulator could buy was the printed price, not the whole book’s opinion.

Read together, the two experiments suggest where the boundary sits. Manipulation struggles against a market with enough informed counterparties to take the other side. It succeeds against one small relative to the manipulator’s budget, and consequential enough to be worth the spend. That is a hypothesis about conditions, and the field record is where it gets tested.

The field record

Field evidence begins with researchers pushing real prices on live exchanges, at their own expense, and publishing what happened.

The most systematic attempt combined both halves of the question. Rhode and Strumpf ran a series of planned, random investments on an academic exchange during the 2000 election, accounting for around two percent of total market volume. Then they set that result beside a century of observational data. In the Wall Street election markets of 1880 to 1944, party operatives traded openly on races, and the record is full of accusations that they traded to move the number. And beside a 2004 episode, in which a single trader made a series of large investments on an offshore exchange in an apparent attempt to make one candidate look stronger.

Their conclusion held across all three. The attacks moved prices. The moves were quickly undone, prices returned close to their previous levels, and they found little evidence that such markets can be manipulated systematically beyond short time periods (Rhode and Strumpf 2008, working paper).

The largest field experiment to date reaches a more qualified verdict. Rasooly and Rozzi’s 817-market shock experiment — described in Where a market price comes from as prices reverting quickly in the first week and more slowly after — found the distortions still detectable two months on (2025, working paper). Those are the same curve, not two results: a fast initial correction that flattens into a long, incomplete tail. Whether it is called reverting or fading, sixty days is a long time for a wrong price to remain visible on a book anyone can look up.

The same paper reports what separated the resistant markets from the rest: more traders, greater volume, and an external source of probability estimates. On books with none of those, a push runs closer to permanent than the earlier literature suggested.

The most uncomfortable field result is the newest, and it concerns a different attack surface. Short-window crypto contracts settle against a spot price at a precise moment. A 2026 study of five-minute contracts on a large on-chain venue found order flow surging in the settlement window, and the price reversing immediately afterward — the signature of pressure applied to the settled price rather than to the forecast. The manipulation was consistently rewarded, and the cost fell mainly on retail participants; the fifteen-minute version of the same contract showed minimal manipulation (Dai, Jia and Yu 2026, preprint). The variable is not honesty. It is how long the settlement window is, and how cheap the underlying is to move inside it — a design choice, made by whoever wrote the contract.

Transaction-level data supply a further field angle. Rothschild and Sethi (2016) reconstructed the behavior of 6,300 unique trader accounts across the entire two-year run of a 2012 presidential market on a real-money exchange of that era. They found a rich ecology of strategies — and within it, evidence suggestive of manipulation by a single large trader, whose possible motives they examine rather than assert. The forensic point stands on its own: on a market with a public trail, a large distorting position is a visible object that researchers can find years later. That is not a property most public numbers have.

The episodes

Outside the experiments, the observational record documents attempts with real motives behind them, and the best-documented cluster comes from the 2004 U.S. election cycle.

On an offshore exchange, several large sales of the presidential re-election contract pushed its price down sharply — on one September day from 63 to 49, on one October day from 53 to 10. In each case the price impact was reversed within 24 hours (Wolfers and Zitzewitz 2006a, working paper). The same survey lists supporters of a minor candidate reportedly bidding his price up on an academic research exchange among the attempts it records as having largely failed, and summarizes the planned-investment experiment described above, whose impact was likewise only temporary.

Read those numbers carefully, because they are the shape of the honest answer. The pushes were real — a contract moving from 53 to 10 is a violent distortion, and for some hours the public number was simply wrong. And they were undone by the next day, by traders who found a badly mispriced contract and took the other side.

What the record does not say

An honest reading keeps the limits in view.

The documented pushes were bounded — hundreds to thousands of dollars against books of a certain era, and research budgets against modern venues both play-money and dollar-redeemable. The record says less about a determined actor with deep funding and a long horizon, and the newest venue-scale evidence is still in working-paper form (the evidence index records the status of each). Thin books remain the soft target, where the manipulation literature converges with the calibration literature on the same warning (Where prediction markets fail).

A market’s resistance is also conditional, not automatic. The mechanism pays defenders only if defenders exist — traders with capital, with attention, and with a way to check the number against reality. Prices with an outside reference recover fastest; questions only the market itself is tracking have less to snap back toward.

And one further limit is structural rather than empirical. Manipulating a price is not the only way to corrupt an outcome. A question can also be settled wrongly, which is why the rules that grade it matter as much as the book that prices it (How an honest venue settles a question).

The honest answer

The answer has two halves and both are in the record. Pushed, yes — anyone with money can move a price, and the record shows exactly that, sometimes dramatically. Held, conditionally. The same record shows pushed prices reverting within hours on active books with informed counterparties. It shows them holding for weeks, or degrading accuracy outright, where books are thin, attention is scarce, a well-funded actor has a decision to sway, or the contract leaves a short settlement window exposed.

The useful version of the answer is therefore structural. Depth, active traders, an outside reference point, and a settlement design that is not cheap to push are what turn a pushable number into an expensive one. Every one of them is something a venue either builds or does not. None of them is a guarantee on its own — depth buys resistance, not accuracy, and on short-horizon contracts the more liquid books have been found no better calibrated than the thin ones (Where prediction markets fail).

Elsewhere, a distortion is free. Words cost nothing, and a confident wrong claim can stand for years unpriced. On a market, a distortion has a running cost, a public trail, and a standing reward for whoever corrects it. That does not make the number incorruptible. It makes corruption a losing trade under conditions a reader can check.

So check them. Before you read the price, find out how deep the book is, whether anything outside the market prices the same question, and how long the settlement window stays open.

Sources

1. Hanson, R., Oprea, R. (2009). “A Manipulator Can Aid Prediction Market Accuracy.” Economica 76(302): 304–314. doi.org
2. Hanson, R., Oprea, R., Porter, D. (2006). “Information aggregation and manipulation in an experimental market.” Journal of Economic Behavior & Organization 60(4): 449–459. ideas.repec.org
3. Deck, C., Lin, S., Porter, D. (2013). “Affecting policy by manipulating prediction markets: Experimental evidence.” Journal of Economic Behavior & Organization 85: 48–62. doi.org
4. Rhode, P. W., Strumpf, K. S. (2008). “Manipulating Political Stock Markets: A Field Experiment and a Century of Observational Data.” Working paper, Natural Field Experiments No. 00325. ideas.repec.org
5. Rasooly, I., Rozzi, R. (2025). “How manipulable are prediction markets?” Working paper, arXiv:2503.03312. arxiv.org
6. Dai, D., Jia, R., Yu, S. (2026). “Settlement Manipulation in Prediction Markets.” Preprint, arXiv:2606.31675. arxiv.org
7. Rothschild, D., Sethi, R. (2016). “Trading Strategies and Market Microstructure: Evidence from a Prediction Market.” The Journal of Prediction Markets 10(1). doi.org
8. Wolfers, J., Zitzewitz, E. (2006a). “Five Open Questions About Prediction Markets.” Working paper, NBER Working Paper 12060. nber.org