A three cent contract does something to a trader's brain. The outcome feels at least vaguely possible, the payoff is better than thirty to one, and the ticket costs less than lunch. I buy one myself now and then, which is an awkward way to open a post about why those contracts are, on average, the worst-priced things on any prediction market. But the evidence on this is old, boring, and unusually consistent, and once you see the shape of it you can trade the other side.
The pattern is called the favorite longshot bias. Contracts priced very low, roughly below ten cents, tend to resolve YES less often than their price implies. Contracts priced very high, roughly above ninety cents, tend to resolve YES more often than their price implies. Buyers of longshots systematically overpay, buyers of near-certainties get a small discount, and the middle of the price range is comparatively fair. The one takeaway up front: the band below ten cents has historically carried negative expected value for buyers, and the worst of it sits below five.
This predates prediction markets by decades
Racetrack economists documented the same shape in horse betting going back to the middle of the last century. Longshots at the track return far less per dollar wagered than favorites, and the gap widens as the odds get longer. The same lean shows up in sports books and in deep out-of-the-money options, which are longshot tickets with more paperwork. Prediction markets inherited the shape wholesale, and studies that group resolved contracts by price bucket keep finding the same lean at both ends.
The practical translation is blunt: if you have been buying five cent contracts on the theory that they only need to hit one time in twenty, historically they hit less often than that. The price is set by a crowd that includes a lot of people who simply enjoy holding lottery tickets, and their enjoyment is baked into what you pay.
Why the mispricing never gets arbitraged away
Three forces hold it in place, and none of them are temporary.
First, lottery preference. People overweight small probabilities. A one percent chance feels like five, and the feeling gets stronger when the story attached to the outcome is vivid. Prediction markets are full of vivid stories, so the cheap tail of the book gets steady, price-insensitive buying from people who are paying for entertainment as much as for expected value.
Second, correcting it is capital-ugly. Shorting a three cent contract means buying the other side at ninety-seven. You lock up ninety-seven cents of collateral to win three, and when you are wrong you lose the whole stake. The trader who would push the price back to fair value needs deep pockets, wide diversification, and patience measured in months. The trader pushing it away from fair value needs a fun opinion and a few dollars, and that asymmetry never resolves in favor of the arbitrageur.
Third, costs and lockup. A ninety-seven cent contract that pays out in a year returns roughly three percent before fees, which is not obviously better than leaving the money somewhere boring and liquid. So favorites stay slightly underpriced because owning them ties up capital for a thin return, longshots stay overpriced because shorting them ties up even more capital for the same thin return, and fees and spreads eat a meaningful slice of whatever edge is left at the extremes. The bias lives in that gap, and the gap is wide enough to trade but narrow enough that sloppy execution erases it.
Building a short-longshot book
The trade that falls out of all this is selling longshots systematically. On a binary market that just means buying the expensive complement, and I would frame it that way on purpose. Buying NO at ninety-four keeps your actual risk in front of you in a way that shorting YES at six does not.
Run as a book, it behaves like selling insurance: small premiums most of the time, an occasional large hit, and results that come down almost entirely to construction. Here is the checklist I would use.
- Only short longshots that are weeks or months from resolution and have a stable information environment. Skip anything with a scheduled binary catalyst, a court ruling, a data release, an announcement date. Those are the markets where four cents can be a fair price.
- Size so that one full loss costs less than the premium from a couple dozen winners, because that is roughly what a blowup at ninety-six takes back when you are collecting four cents a contract. Decide the position count and unit size before opening anything.
- Check correlation before every add. Ten political longshots that all pay off on the same surprise are one position wearing ten costumes.
- Annualize the expected return net of fees and spread, then compare it to what the collateral would earn parked somewhere safe. A thin gross return over eleven months usually fails this test.
- Write down the path by which each longshot could actually hit. If you cannot describe it in a sentence or two, skip the market. A short position you cannot explain is hope with collateral attached.
The classic failure mode is how most people exit this strategy. The book pays quietly for months, nothing hits, and you conclude you are underexposed. You size up, relax the catalyst filter, and then one correlated event takes back a year of premium in an afternoon. Insurance sellers have failed this exact way for as long as insurance has existed, and the only fix I know of is a sizing rule written down before the winning streak starts, which you then refuse to renegotiate with yourself mid-streak.
When the bias flips near resolution
Everything above describes the long middle of a market's life. Close to resolution the geometry changes, and a trade that was structurally in your favor can turn against you.
Take the favorite side first. Near-certainties trade at a discount mostly because of capital lockup, and lockup stops mattering when payout is hours away. The discount on a ninety-nine cent contract largely evaporates in the endgame, and what remains is often thinner than the spread. Grinding out the last cent of a resolving market rarely compensates you for the occasional freak reversal.
The longshot side is where people actually get hurt. In live, path-dependent endgames, a recount, an overtime, a last-day announcement, prices update slower than the underlying situation, and a contract sitting at four cents can be genuinely cheap for stretches because holders of the other side are slow to accept what is happening. Shorting longshots into that kind of endgame is picking up pennies in front of a steamroller that has already left the depot. The structural lean pays the people who harvest it slowly, across many markets, far from resolution.
My own rule is to stop opening new short-longshot positions once a market enters its final days, and to let existing ones ride only when the resolution path is mechanical rather than newsy. Beyond that, verify the price bands yourself instead of taking my word for it. We pull prediction market prices into Blockcircle alongside the rest of our market data, but any venue that publishes resolved-market history will do. Group past contracts by price bucket, measure how often each bucket actually resolved YES, and see whether the cheap end underperforms its implied probability the way it historically has. If your sample says otherwise, trade what your sample says.