the most successful prediction market ever built is not polymarket, augur, or kalshi. it is the perpetual futures contract, which crypto retail has used to aggregate beliefs about billions of dollars of underlying every day for the last five years without noticing that it was a prediction market at all.
the trick is the funding rate. a perp has no expiry, so there is no natural mechanism to pull the contract price back to spot — the thing that makes a quarterly future anchor to reality. perps fix this with a periodic payment between longs and shorts proportional to the premium of the perp price over the index. longs pay shorts when the perp trades above spot; shorts pay longs when it trades below. if you ignore the market microstructure details, the funding rate is doing exactly what a prediction market's price does: it is a continuously-cleared statement about where the consensus expects the spot price to move, in the short interval over which funding rebalances.
read it that way and a bunch of confusing facts about perps suddenly make sense. why does funding go positive when people are "bullish"? because long demand exceeds short demand, which is another way of saying the perp price prices a higher probability of upward moves, which in a prediction-market frame is just the market's belief moving. why is a spike in funding often a contrarian indicator? because the funding spike is the market's belief, and at extreme readings you have the same mean-reversion pressure you would have on a polymarket contract trading at 95% on a coin flip. none of this requires invoking sentiment or flow-of-funds mysticism. it falls out of treating the contract as what it is: a continuous scoring rule that pays traders for correctly predicting the next funding interval's spot path.
the obvious next question is why perps work so much better than anything the prediction-market community has built. the answer, i think, is that perps removed three frictions that every standard prediction market still fights:
resolution risk is zero. a polymarket contract on "will x happen by y date" depends on an oracle making a correct binary call. disputes are expensive and chill participation. a perp resolves continuously against a publicly-indexable price. there is nothing to dispute. this is underrated and probably the single biggest reason perps have 100x the volume of every prediction-market contract combined.
holding period can be anything. to bet on a normal prediction market you commit capital until resolution, which for multi-month markets is a real drag on retail engagement. a perp lets you express a belief for five minutes or five months and exit when you change your mind, because the payoff accrues incrementally via funding rather than in one lump at resolution. this is the same advantage an amm has over an orderbook for long-tail assets, applied to time.
the payoff is linear in a continuous variable. binary prediction markets quantize the question into "will x happen, yes/no." perps quantize nothing — your payoff is proportional to how right you are. this is a much better scoring rule when the underlying is inherently continuous, which most "predictions" about the world actually are.
if perps are the prediction markets that actually work, the question is what you can build by dragging perps into domains that currently run on binary contracts. a perp on the us unemployment rate, marking-to-index off the bls release, and paying funding against a smoothed estimator between releases. a perp on a benchmark score (say, a perp on a public ai eval) where the index is the latest published number and the funding rate continuously clears the market's belief about the next update. a perp on a climate index. a perp on electricity load. all of these exist as futures in some dusty corner of the world, and almost none of them have the retail-accessible continuous structure that crypto perps do.
the rest of the design work is not free. you need a reliable on-chain index, which is the oracle problem in a different costume. you need enough two-sided interest to keep funding from blowing out to absurd levels in one direction, which means either a large directional user base or a subsidized market-maker. you need a margin model that does not liquidate half the book during normal volatility. these are the same problems crypto perps have already more-or-less solved, which is why i expect the next generation of "prediction markets" — things like conditional markets on policy outcomes, scorecards on ai models, macroeconomic-index perps — to look much more like hyperliquid or dydx than like polymarket.
the prediction-market community has spent a decade building discrete, legal-text-resolution, winner-take-all binary contracts. the crypto community has spent the same decade, almost accidentally, building the better mechanism and using it to speculate on dog coins. nothing stops us from pointing the mechanism at more useful questions.
the piece of this that still needs work is the scoring rule for non-price indices. a perp on btc is clean because the underlying is a price anyone can read. a perp on "gdp growth next quarter" has to decide how to smooth between discrete releases, how to handle revisions, and how to settle a funding payment when the index itself is an estimate. these are tractable design choices, but they are the real intellectual content of moving perps out of crypto-native domains. if you are going to write one post about perps this year, it should be about this.