Why there are only a few thousand prediction markets

There are more questions worth pricing than there are prediction markets, by several orders of magnitude. Every sports fixture in every league, every token's weekly range, every product launch date, every local election, every earnings print, every "will this ship by Q3" inside every company. Millions of them, easily.

The actual count is in the low thousands.

That gap is usually explained as a demand problem — people do not want to bet on obscure things, the audience is small, regulation is hard. Those are all real and none of them is the binding constraint. The binding constraint is on the supply side, and it is precise:

Prediction markets number in the low thousands because each one must be worth a market maker's attention or an operator's subsidy.

The per-market cost, in every structure

Go through them. Each one has a per-market cost with a name.

Order book. The cost is a professional's attention. A market maker must decide this market is worth pricing, monitor it, and quote both sides. There is a floor on how much attention costs, and it does not fall with the market's size — pricing a niche market takes more effort than pricing a liquid one, for less spread income. Venues extend their reach by paying liquidity incentives, which is the same cost paid in cash instead of attention.

CFMM. The cost is an LP's expected loss. On an expiring event contract the pool mechanically rebalances into the losing side as information arrives, and the LP absorbs it. That is a real, funded, negative-expectation position, per market.

Scoring rule. The cost is the liquidity subsidy, bounded at b · ln n. Elegant, known in advance, and still b · ln n times your market count.

Classic parimutuel. The cost looks like zero, and this is the interesting case. Pools genuinely work from the first dollar. But the venue pays at the other end of the market's life: it must close entry before the event, because late money free-rides at a measured +25.27% expected value. And it pays continuously through dilution — the last decile of entrants captures 13.20% of the losing pool, transferred from the people who were early.

So four structures, four per-market costs, and every one of them scales linearly with catalogue size. That is a ceiling, and it is roughly where the industry sits.

What "the ten-thousandth market" actually means

The paper's framing is worth quoting because it locates the problem precisely:

In a deep market, being early already pays: you buy at 20¢ and sell at 80¢, and that is what a price is for. The problem this paper addresses is narrower and it is a problem of the long tail: in the ten-thousandth market there is no book to sell into, no market maker willing to quote, and no operator able to subsidize.

Note the concession in the first clause. Where a book works, the mechanism is solving a problem that does not exist. Nobody needs an engineered early-entry gradient in a market that already has a price you can trade against.

The failure at the tail is not gradual. Market ten works. Market ten thousand has a thirty-point spread or does not exist. There is no partial version of "a market maker will quote this."

What removing the constraint would look like

The claim is not that the vested rule creates demand. It is that it removes one specific gate, and that gate is currently binding.

The per-market capital does not disappear — the paper is careful about this and repeats it in three places. It stops being spent:

That is the arithmetic difference between a ceiling and a working capital line.

And the mechanism's own crowding measurement pushes in the same direction: growing the seed from $50 to $5,000 a leg — about three times the organic pool — compresses ordinary early winners from 1.70× to 1.09×. So the venue that seeds thinly across many markets does better by the mechanism's own lights than the venue that seeds heavily across few. The policy that scales is the policy the mechanism prefers.

Who arrives at market ten thousand

Removing the capital gate is necessary and not sufficient. Someone still has to find the market, price it, and post the seed. That is R7 — the requirement the paper adds to Melee's five:

Floors make early capital safer; they do not summon it. A venue hosting millions of markets needs a participant class for which discovering, pricing, and seeding a brand-new market is cheap and systematic.

Humans are not that class. Nobody is going to manually evaluate the ten-thousandth market, and the reason has nothing to do with capital — it is attention again, in a different costume.

The paper's answer is that seeding is the first market-making job that is legible to software: a creation floor, a closed-form position value from two scalars, and economics that depend on forecastable quantities — flow volume and balance — rather than on out-quoting a professional. That argument, with its limits, is here.

Three honest limits on the claim

Nothing here creates demand. A market nobody wants to trade is a market nobody wants to trade, seeded or not. What changes is that its non-existence stops being a supply decision. The paper's version: a market nobody will seed at any tilt "has no believer willing to stand on any side of it, and the mechanism declines to pretend otherwise."

The mechanism trades volume for placement. In the behavioural study, total volume falls from $2,845 to $2,254 a market — about 21%. A venue with ten times as many markets at 79% of the volume each is ahead; a venue with the same markets is not.

Thin markets are where the entry rule is hardest to use. In the sparse arm — arrival rate 0.03, roughly 27 stakes a market, exactly the long-tail regime — agents using the naive growth estimate realize −11.5% per entry post-fee while the same beliefs under classic rules earn +11.2%. The naive estimator wildly overstates growth precisely where flow is thin. The gate is removed; the pricing problem behind it is not.

The claim, stated at its actual size

The paper's closing paragraph is unusually restrained for a whitepaper, and the restraint is the point:

A market settled by arithmetic and seeded by its own creator carries no such per-market cost. We do not know what the ceiling is. We know it stops being the number of market makers.

That is the whole argument. Not that a million markets will exist. Not that demand is waiting. Only that the count is currently a function of how many markets are worth a professional's attention or an operator's budget — and that a settlement rule with a floored seed makes it a function of something else.

What the new ceiling is, nobody knows. Finding out requires someone to build the thing and count.

Next: Agents as market makers.