Prediction markets are the future. They’re a type of trustless, decentralized expertise that often equals or outperforms official sources.

But they’re not quite the present. Right now I only know of three prediction markets, and none of them live up to their potential. As usual, it’s the government’s fault: betting on prediction markets is technically gambling, which makes it mostly illegal (of course, you can still buy all the Gamestop stock you want).

Each of the three big prediction markets tries to solve this in a different way.

PredictIt struck a special deal with the government where they can run a limited number of predictions as long as nobody is allowed to bet very much. They probably have the most users, but most of them are dumb money, and the restrictions prevent smart money from coming in and replacing them. Also, the market fritters away its limited number of predictions on dumb political horse race stuff. I think of this as a missed opportunity.

Augur is a decentralized Ethereum-based prediction market. Since it’s crypto, it ought to be able to violate laws with impunity, and in theory this should make it the leader of the pack. In practice it’s either nonfunctional or so minimally functional as to be useless. The team involved seems pretty dedicated and competent, and I assume they have some good reason for pretending their product currently exists instead of replacing the whole thing with a COMING SOON banner. I hope this will one day be our savior. But like the real Messiah, it’s taking its sweet time.

Metaculus solves the regulatory problem by using fake Internet points instead of money. This is a disappointing solution; it limits the user base to Internet obsessives instead of (say) investment bankers. Still, there are a lot of Internet obsessives. And the team running it is really top-notch, interested in pushing the limits of what prediction markets can do, and trying to focus on some of the most important questions.

I want to raise awareness of prediction markets, and right now Metaculus seem like the best people to raise awareness of. So welcome to Metaculus Mondays, where I make you listen to reports of how the prediction markets did this week and what they’re predicting for later.

This week: the coronavirus.

Late last year, when coronavirus had already killed 285,000 Americans, Metaculus asked users to predict how many would be dead by the end of 2021. The guesses started at about 500,000. But as cases rose further through December and January, the guesses rose too, until now they’re averaging almost 690,000 people.

Right now 440,000 people have died, so Metaculus expects another 250,000. Right now 3,500 people die of COVID daily. If that death rate continued, we’d be past their estimate for December by April. But it’s reasonable to expect deaths to decrease as winter turns to spring, more people get vaccinated, and the usual control system causing people to lock down tighter as deaths rise kicks in. Still, I notice that every day their estimate rises slightly. Seems suspicious. I’m betting my fake Internet points on something a little higher, maybe 750,000, with a long tail in case of variants that can reinfect the already infected, or if so many people refuse vaccination that we don’t have herd immunity for the danger period in winter 2021.

Moving on:

When will some official like the Director of the CDC announce that a coronavirus vaccine is available to any adult who wants it (as opposed to just front-line workers, only seniors, etc)? The question as asked is a little odd, since it will probably be available in some areas before others, but this is apparently about some kind of nationwide announcement.

Apparently Anthony Fauci said “as early as April”, former HHS secretary Alex Azar said “March or early April”, and former FDA chief Scott Gottleib says “heading into fall 2021”. Metaculus users fall in the middle at May 19. This doesn’t seem to have changed too drastically since the question opened in December. I don’t think I disagree too strongly here.

Next up:

When will 50% of the US population be vaccinated? Metaculus defines this as getting both doses of a two-dose vaccine; right now 1.5% of Americans are vaccinated by this definition. Metaculus seems to think we’ll reach 50% by June. My main reason for pessimism is that only about 50% of Americans really want the COVID vaccine immediately - the rest say they’ll “wait and see”. Some of those people probably will be satisfied once there are no major side effects. Others will see COVID rates go down (as they inevitably will during the spring) and decide they don’t need it. I don’t think the average Metaculan truly understands the difficulty involved in getting random people to accept public health interventions in any situation other than “you will die IMMEDIATELY if you don’t do this”. And there’s a wild card scenario where the virus mutates, the vaccine doesn’t work as well, and people decide it’s no longer worth it.

Next up:

The specific question is “Will a new variant of SARS-COV-2 that, due to a mutation, can infect people who had immunity to a previous variant, cause more than 10M infections globally by the end of 2021?”

But I’m not happy with how it’s phrased. My understanding is that probably some new strains will have a 1% or 10% or whatever chance of infecting already-immune people - lower than infecting coronavirus-naive people, but still more than zero. If a strain that can infect immune people at a 1% rate infects 10 million (mostly non-immune) people, does that count? Since this question doesn’t say, I don’t think it’s that meaningful. Which is too bad, because it’s probably the most important question of 2021 and the one that should be getting the most attention right now. Anyway, 260 people say an average of 37% chance, which sounds vaguely horrifying, even if we can’t quantify exactly how horrified we should feel.

Okay, now we’re getting into things we can be properly horrified about. The exact question is “Before 2023, will the United States CDC recommend that those who have already been vaccinated for SARS-CoV-2 (COVID-19) be vaccinated again due to a mutation in the virus?”

The average is 60% yes! This is probably going to happen! I might write more on this later, but we haven’t prepared the regulatory framework we need to deal with this, gotten the public ready for this, or anything. Right now regulators’ default assumption should be that we need to plan what to do about COVID-21, COVID-22, et cetera, and get to work figuring out how we’re going to screw this up less badly than we screwed up the last one.

This one is some creepy deep magic. It’s Metaculus really pushing the limit of what you can do with prediction markets. On September 1, before the US had really decided whether or not to hold human challenge trials for vaccines, Metaculus asked its users to predict how many Americans would die of COVID in 2021 if they (did/didn’t) allow the challenge trials (ie experiments where they test vaccines by deliberately infecting volunteers). The market concluded that if they did, 295,000 people would die; if they didn’t, 343,000 people would die.

You don’t have to believe them if you don’t want to. But this is a pretty impressive example of decentralized expertise weighing in on important inputs to a policy question. Should we hold human challenge trials or not? Well, if we don’t, an extra 50,000 people or so will die. You can do this with any question. How much will global temperatures increase if we (do/don’t) rejoin the Paris Agreement? How much will British GDP be in 2025 if there (is/isn’t) Brexit? How many people will get shot by cops next year if we (do/don’t) go ahead with some defund-the-police plan?

We shouldn’t be the sort of straw-man technocrats who unthinkingly choose the lowest number because Our Metrics Perfectly Capture Reality. But in this case I’m pretty okay with being the sort of technocrat who thinks about it for a second, remembers that the lower number represents 50,000 real breathing human beings who get to be alive instead of dead, and then goes with it.

(in the real world, we didn’t do human challenge trials, 100K people have already died this year, and it’s only February)

And finally, since you’ve been bearing with me so nicely, here’s something a little on the lighter side, from PredictIt: