By Alan Wu • Published April 14, 2026
Originally posted on X.
Play-money prediction markets are weirdly accurate.
Metaculus and Manifold both have strong calibration records. But if real capital is what disciplines markets, why do play-money markets forecast so well?
Prediction markets provide a forecast before an event's resolution. Beyond manipulating event outcomes, actors can also manipulate the forecast before resolution.
The accuracy of play-money prediction markets exists in a regime where there is little incentive to manipulate. No major capital allocator, government, or organization is known to be making decisions based on these numbers.
Because manipulation has little payoff, the participant pool skews toward people trying to forecast rather than game the signal.
The real variable isn't play money versus real money, but the cost of manipulating a signal relative to the value of doing so. This is akin to manipulating the oracle price of a perp market.
Real money raises the cost of manipulation, but it also imports its own distortions: opportunity cost, hedging motives, participation friction, etc.
The question for any market is which tradeoff is more acceptable: exposure to manipulation or the frictions that come with real money.
The strong calibration of play-money markets emerged in a low-stakes environment where few real-world actors base decisions on them. That is not a fixed equilibrium. It holds only so long as these markets remain unimportant.
People treat displayed numbers as inputs. Whether it's DeFi yields, swap quotes, or prediction-market probabilities, they anchor on the number without interrogating the mechanics underneath. Once a number starts guiding decisions, it becomes worth manipulating.
Play-money prediction markets have a ceiling, and that ceiling is their own usefulness.
Right now, their accuracy is real. But it exists in a specific regime: one where no one has a reason to distort the signal.
As prediction markets gain institutional legitimacy (there's been some progress), that regime ends. If prediction markets someday become inputs for policymakers, institutions, or capital allocators, the manipulation value rises accordingly.
But the manipulation cost of play-money markets remains close to zero. So the ratio breaks.
This is the trap. The better these markets perform, the more likely they are to attract institutional attention, which is exactly the thing that will break them. Their accuracy is self-undermining. It holds only as long as nobody important is watching. The moment they become important enough to rely on is the same moment that they become unreliable.
Play money can be good for political markets. But when people or news outlets start citing them, that's when the signal is destroyed.
Real money isn't a pure upgrade to forecast accuracy. It levels up manipulation resistance but at the cost of economic noise.
Consider a market sitting at 95%: you lock up $0.95 to make $0.05 over 2 years. If there exists some meaningful risk-free rate, that's not a great return while still bearing the event's risk. Rational capital exits the heavy side, not because it disagrees with the price, but because the economics don't justify holding it there. The price drift is due to capital efficiency issues, not information.
Real-money markets also filter for willingness to navigate crypto or KYC hoops, not domain knowledge. The most informed person on a niche topic might never bother with those things. Hedging further contaminates the signal, drifting the price to accommodate that demand.
Real money makes a forecast more resistant to attack but less purely about belief. Play money makes a forecast purely about belief but defenseless against attack.
The question is which tradeoff is more acceptable for a given market.
For near-term, consequential markets that institutions reference, manipulation is the bigger threat. The economic noise is worth tolerating, and some of that noise, like hedging, is itself useful.
For longer-horizon, niche, extreme-probability, or regulatory-gray markets where no one has an incentive to distort the signal, there's nothing to defend against. Real money can make the signal worse (In Defense of Play Money Markets). But the markets that get adopted as decision-making tools are the ones where the manipulation incentive arrives first.
As prediction markets become more consequential, they need to be harder to manipulate. The battle-tested way to make manipulation expensive is real capital at risk.
Blockchains went through the same arc. Years of clever reputation-based and alternative consensus mechanisms before the ecosystem converged on proof-of-stake for securing high-value networks. For adversarial environments, capital at risk is ultimately what works.
The prediction markets that matter most will be the ones where manipulation is expensive. That means real money, and markets with enough usefulness and liquidity to make informed participation worth the effort. I wrote about what that looks like in Prediction Markets Grow By Serving The World.
Thank you to @tenad0me, @0xperp, @ryanchern for thinking through this with me!