Every long-dated prediction market I have looked at has the same shape. Contracts on outcomes that feel close to certain trade a few cents lower than they should, and they stay there for months. The first few times I noticed this I blamed thin books and lazy money. The better explanation is more boring and more useful. A dollar locked in a prediction market usually earns nothing, a dollar in a Treasury bill earns the risk-free rate, and the market charges you for that gap whether or not you ever think about it.
Take a contract that pays a dollar on an outcome you honestly believe is 95 percent likely, resolving in twelve months, trading at 90 cents. On the surface you are buying five points of edge. Now run the alternative. Put the same 90 cents in a T-bill at, say, 4.5 percent and you finish the year with roughly 94 cents, no resolution risk, no platform risk, no chance a rules committee reads the question against you. The contract's expected value at your own probability is 95 cents, so your expected return is about 5.6 percent over the year. You are barely a point over cash, and you are taking real risk to earn it. Most of the edge was an illusion created by ignoring what the money could do elsewhere.
Where the discount comes from
Binary prediction markets are usually fully collateralized. Buy YES at 90 cents and your 90 cents sits in the pot until resolution, with 10 cents locked on the other side. Every dollar of open interest is a dollar of idle cash, and idle cash has an opportunity cost equal to whatever short-term rates are doing. On a contract that resolves next week the cost rounds to a tenth of a cent, invisible inside the spread. On a contract that resolves in a year it is several cents, and on near-certain contracts it becomes the dominant term in the price.
You can read the financing rate straight off the market. On most venues YES plus NO redeems for exactly one dollar at resolution, so if the pair can be bought for a combined 96 cents a year out, it is a synthetic zero-coupon bond yielding roughly 4 percent. Arbitrage capital will only compress that gap down to what it could earn elsewhere, adjusted for platform risk, so the combined discount tends to hover near short-term rates. When I want to know how much of a single contract's cheapness is carry, I check the pair first.
Computing the hurdle rate
The screen takes about a minute. Call the price p, the time to resolution T in years, and your own probability q.
- Annualize the payout. If the contract pays, your gross return is 1/p, so the annualized return is (1/p)^(1/T) minus 1. A 90-cent contract with a year left annualizes to about 11 percent if it hits. The same price with two years left annualizes to about 5.4 percent, which should already slow you down.
- Weight by your probability. Expected annualized return is roughly (q/p)^(1/T) minus 1. This slightly abuses how binary payoffs work, but it is fine for screening.
- Subtract the risk-free rate for the same horizon. What remains is your compensation for resolution risk, platform risk, and the chance you cannot exit early at a fair price.
- Subtract any yield your collateral earns while parked. On many venues that is zero, but it is the variable that changes everything.
- Act only if the remainder clears a premium you would genuinely accept for tying up money in something you may not be able to sell. For me that is a few points at minimum, more when the book is thin.
The same arithmetic runs in reverse when you use prices as forecasts. If the marginal holder demands full risk-free carry, the price sits near q divided by (1+r)^T, so the implied probability is p times (1+r)^T. A 40-cent contract a year out at 5 percent rates is really pricing something like 42 percent. On mid-range prices the adjustment is small. At the extremes it wrecks naive reads, because a 3-cent contract has no room to fall by its own carry discount, so long-dated tails look expensive relative to favorites even when nobody involved is being irrational.
The failure mode I see most lives at those extremes. Someone finds an outcome they consider absurd priced at 4 cents a year out and buys NO at 96, reasoning that free money is still free money. Held to resolution, the position returns about 4.2 percent annualized when it works, at or below what a T-bill pays, with tail risk, platform risk, and a lockup attached. You can be completely right about the world and still lose to cash. I have made this exact trade, felt clever for a year, and underperformed the boring alternative by a point.
Yield-bearing collateral changes the answer
Everything above assumes locked collateral earns nothing, and that assumption does all the work. If a venue parks the collateral pool in Treasuries or an interest-bearing stable asset and passes the yield to position holders, the carry cost drops toward zero and long-dated prices are free to sit near true probabilities. Some newer designs accept yield-bearing tokens as collateral directly, so a position keeps compounding while it waits. In that regime the hurdle becomes the risk-free rate minus the collateral yield, which can land near zero or even negative.
The catch is who captures the float. Plenty of venues earn interest on the pool and keep it, which means the platform is quietly running a money-market business on your margin while your side of the trade bears the full carry drag. Two venues can list the identical contract and rationally trade it at different prices for this reason alone, so before calling a cross-venue gap an arbitrage, check the collateral treatment on both legs. It also matters when you study a market's price history, because a persistent long-dated discount means something different under each regime.
I will admit the clean version of this story is too clean. Even with full yield pass-through, extreme prices probably stay a little distorted, since capital tied against a 4-cent tail carries sizing problems no formula captures, and I have not seen anyone convincingly separate how much of the long-dated longshot pattern is carry and how much is people liking lottery tickets. But the carry component is the part you can compute in a minute, and it turns a seductive discount into a number you can hold up against a T-bill. Mostly this arithmetic has killed trades before I made them, which is duller than finding mispricings but has been worth more.