Donald Trump is the presumptive Republican nominee for president. No one else is still running. He’s the nominee. It’s him.

So why do people who bet on this stuff say there’s still a chance he won’t make it?

On PredictIt, a betting exchange that specializes in political contests, thousands of gamblers would still give Trump a seven percent chance of flaming out before officially clinching the nomination in Cleveland. The odds are about the same on Betfair, a British competitor. Astoundingly, you can still make money betting Trump will win—$7 for every $100 wagered, a far better rate than your average savings account.

So what do these people know that the rest of America doesn’t?

Plenty has been written on how betting markets sometimes perform better than polls (and certainly pundits) in forecasting elections. Nate Silver of FiveThirtyEight, the wonk of all prediction wonks, admits “markets are tough to beat in most circumstances.” But Trump seems an especially powerful case. As far back as the New Hampshire primary, the markets gave him at least a 40 percent chance of winning, higher than most (mainstream) political commentators would have wagered at the time.

And at times, the markets have had an uncanny understanding of what’s ahead. Trump’s position among bettors hit its nadir in the days before the Wisconsin primary, just as Cruz’s plans to force a contested convention began to build steam. Cruz won the state handily, beating Trump by 13 points. But at the moment of his defeat, as the punditry debated The End of Trump, trading on the betting markets had hiked the billionaire’s chance of winning by three percentage points. His stock steadily rose for the next two weeks, to 59 “percent” on the eve of the New York primary. He hasn’t lost a primary since.

(I put “percent” in quotes above because that’s not really how the betting markets measure success. On PredictIt, bettors quote the price they’re willing to pay now for the chance at winning $1 later; they’re matched with gamblers willing to make the opposite wager. So if Trump is trading at 93 cents, that means a bettor would make a 7 cent profit if he wins, but would lose their investment otherwise. Betfair, which U.S. residents can’t use, relies on an oddsmaking system that’s a bit more complicated. Both services charge fees, of course.)

What did these bettors see in Wisconsin? David Rothschild, an economist at Microsoft and founder of PredictWise, which tracks betting markets, suspects it was actually John Kasich’s weakness in Wisconsin, not a subsequent rise in Trump’s strength, that made the markets move. “It kind of knocked Kasich back and basically kind of forced this two-person race structure,” he said. “Wisconsin was a signal that Kasich wasn’t a threat in the Northeast.”

Now, the markets are showing reluctance to move Trump fully into the “win” column, even though the candidate himself was confident enough tell his supporters in West Virginia to stay home this week. A few theories why:

Bettors Love Long Shots

Research has long established that humans are horrible at setting odds for unlikely events. As a National Bureau of Economic Research report put it in 2010, “people are systematically poor at discerning between small and tiny probabilities, and hence price both similarly.” The favored example is the racetrack—a horse with a win-loss record that would indicate a 1-in-300 often gets odds around 1-in-100 instead. It works in reverse, too: “People exhibit a strong preference for certainty over extremely likely outcomes, leading highly probable gambles to be under-priced,” according to the NBER paper.

To some degree, all bettors like risk. A wager against Trump is a long shot at this stage in the game, but winning would pay out big. It’s possible that cognitive blinders and the devil-may-care attitude of the average gambler are keeping the markets from reaching a rational conclusion about Trump. But I don’t think this fully explains it. Consider another betting market, which asks gamblers to predict whether Clinton’s margin of victory in the Pennsylvania Democratic primary (which, bizarrely, is still not completely tabulated) will exceed 13 percentage points. This is also a long shot—a very long one— and is priced much more reasonably, at 99 percent against.

There’s Not Enough Liquidity Left in the Market

As part of an agreement with U.S. government, PredictIt has agreed to limit each bettor’s total wager to $850 per question, capping the number of bettors at 5,000. Since this particular question has been open so long, it’s possible everyone has already spent their money. As one participant put it:

The $850 dollar limit in a market makes it hard at times to balance out the people willing to buy Trump No for a few cents. Either it’s not worth risking more to people, or they’re already maxed out so the funds to equalize the market aren’t available.

Fair. But I'd note that Betfair, which doesn’t have the same restrictions, is posting similar odds.

There Actually Is a Slim Chance Trump Won’t Get the Nomination

The rules of the PredictIt market state that bets won’t be paid until the conclusion of the Republican National Convention. The payoff has nothing to do with delegates, of which Trump has plenty, or the billionaire’s position in the polls, which is obviously strong. That 7-percent slack, then, encompasses any number of the unlikely-but-still-possible ways Trump could lose the nomination: a bizarre parliamentary maneuver in Cleveland, a sudden illness, a world-ending asteroid impact.

Given Trump’s unpopularity with Republican leaders, perhaps there’s more cause for concern than usual. Another bettor’s take:

Trump is in the unique position of being his party's presumptive nominee while the party establishment is still actively trying to stop him. That is why we're not seeing Trump at 99. Trump supporters are right to be nervous, as you can trust these Repub party leaders about as far as you throw Chris Christie.

I imagine for them it's like watching your team build a massive lead in the championship, but your stomach doesn't completely unclench until that final buzzer. Except in Trump's case the commissioner, team owner and the refs are having frantic back-room meetings trying to stop the team from winning.

Figuring out the underlying dynamics of a market is hard—maybe impossible. (No one’s quite figured out the one on Wall Street yet.) But at some level, the effect alone is insight enough. A bunch of people who have consistently put their money where their mouths are don’t think Trump has locked down the nomination. I wouldn’t make that wager myself. But neither am I betting against them.