The Case for Legalizing Prediction Markets

In October 2022, Bloomberg published an article with the headline “Forecast for US Recession Within Year Hits 100% in Blow to Biden.” A hundred percent chance of a recession! But here we are, a little over a year since that article was published, and there hasn’t been a recession. 

What will happen to Bloomberg for getting it so wrong? Well, probably nothing. In fact, I’m not even sure they made a bad decision by publishing that article. After all, that over-the-top headline probably got more clicks and generated more advertising revenue than a headline with a more realistic (and boring) forecast would have. Bloomberg, like most media companies, is facing a complex mix of incentives: publishing correct information, getting clicks, maximizing ad revenue, getting new subscribers, retaining current subscribers, etc, and these goals don’t always completely align with each other. Sometimes appealing to the reader’s emotions (especially fear, anger, or political tribalism) might get more clicks and generate more revenue than simply conveying correct-but-boring information.

By contrast, the incentives of a betting market are completely straightforward. Correctly predicting an outcome is financially rewarded, and getting it wrong is financially punished. Or as economist Alex Tabarrok puts it: “A bet is a tax on bullshit.” But it doesn’t take an economist to understand this concept –  schoolkids understand it too: “Oh yeah, wanna bet?” 

Putting something of value on the line can quickly untangle someone’s actual beliefs from their stated beliefs.The Bloomberg editorial board were happy to print a headline predicting 100% chance of a recession, but would they have bet money on a recession happening, given 99:1 odds? Of course not, that would be throwing money away! What about 75:25 odds? 50:50? 

Well, some people were betting on whether or not there would be a recession. At the beginning of this year, the odds-implied probability of a recession occurring in 2023 was around 40-50% on the betting site Kalshi, and gradually drifted towards 0% throughout the year as it became clear that a recession wouldn’t happen. Importantly, the people betting on this market weren’t concerned with getting clicks, ad revenue, or pandering to either political side. Rather, they were actually trying to get it right so they could make money. 

This is what’s called a “prediction market.” Sites like Kalshi allow users to bet on the outcomes of events. You can bet on things like fed interest rate decisions, when GPT-5 will be released, whether Joe Manchin will leave the Democratic Party, the Oscars, and more. The markets involve assets that pay out $1 if an event occurs and $0 if it doesn’t. Traders can buy and sell these assets at different prices, which correspond to implied probabilities that an event will occur. For example, at this writing the movie Oppenheimer is trading at $0.42 to win the Oscar for Best Picture – an implied probability of 42%.

A lot of people have a negative gut reaction to this. I mean, it’s gambling! Right? Isn’t it just like a casino or the lottery or something? Regulatory agencies have had mixed reactions so far. Recently, the CFTC denied a request by Kalshi to operate election markets, and Kalshi is now suing the CFTC in response. The global platform Polymarket is illegal for Americans to bet on, though it can still be interesting to follow the markets and see how people around the world are betting. The New Zealand-based platform PredictIt is currently in a state of legal limbo, temporarily allowed to continue operating while the courts decide its fate. Manifold Markets is an experimental platform that dodges the legal issues by using play-money, with the hope that a reputation-based incentive system can take the place of the financial incentives in real-money markets. Actually, there are currently markets on Manifold about what will happen with PredictIt and whether or not Kalshi will win their lawsuit against the CFTC!

I think putting prediction markets in the same legal category as casinos and lotteries is a mistake. For one thing, these traditional forms of gambling are usually based entirely on luck, with no skill involved – and a negative expected payout for the player. By contrast, success in prediction markets is based on the ability to make accurate probabilistic forecasts, a valuable skill. In the book Superforecasting, Professor Philip Tetlock summarizes years of academic research on this topic. He finds that probabilistic forecasting is a skill that can be learned and honed through practice. It doesn’t require psychic powers or genius-level IQ – rather, it’s a matter of doing careful research and following good epistemic practices, like reasoning from base-rates and being willing to update your predictions based on new information. 

Prediction markets are a good way for participants to practice this skill, and for society to identify people who are good at it. For example, if a media network wants to put together an expert panel on some complex geopolitical issue, they might consider picking someone from the Manifold Markets leaderboard who has a solid track record of correct predictions on geopolitics, rather than simply choosing the person with the fanciest-sounding degree or job title.

Prediction markets also aggregate information in a way that’s useful for outside observers. For example, the website Election Betting Odds is currently tracking odds for the 2024 elections. Rather than trying to keep up with a constant flurry of information – new polls, something outrageous Trump said, some gaffe Biden made, biased commentary in either direction – it’s a lot easier to just check this website once in a while and see how the odds have changed. At the very least, following the markets this way allows you to track the opinions of people who are actually trying to get it right, rather than maximize clicks, pander to either political side, etc.

A best case scenario is that, under the right conditions, the Efficient Market Hypothesis might apply to prediction market prices in the same way it applies to stock prices. That is, with enough trading volume, there is a pressure on prices to reflect all publicly available information about the value of an asset, because if somebody knew some information about its underlying value that wasn’t reflected by the current price, they would have a strong financial incentive to either buy or sell/short the asset to profit from this information. The act of buying/selling an asset changes the market price of the asset (through increased/decreased demand), so the new information gets “priced in” by the trade. 

Of course, this is all theoretical. How accurate are prediction markets in real life? Well, an academic study (Berg et al. 2008) found that prediction markets often outperform polls for predicting election outcomes. Manifold Markets, the play-money platform mentioned previously, has an impressive calibration record according to an analysis posted on their website – events forecast with 20% probability tended to happen about 20% of the time, 80% probability tended to happen about 80% of the time, and so on. Kalshi also boasts an impressive record on predicting inflation, outperforming many economists. Sports betting markets, which operate according to similar fundamental principles as current event prediction markets, are also remarkably well-calibrated.

There’s also some evidence in the other direction. An analysis I did myself and posted on my own blog found that forecasts on PredictIt were slightly worse than Nate Silver’s forecasts for the 2022 US midterm elections. So prediction markets aren’t perfect and can underperform experts sometimes. But even in these cases, there will at least be some cosmic justice in the end, as bettors who get it wrong will lose money to bettors who get it right, and maybe everyone will learn from it and be a little more accurate the next time.

To conclude, while prediction markets might be considered “gambling” in a technical sense, their usefulness – both in creating good epistemic incentives for participants and in aggregating information for observers – ought to put them in a different legal category than other forms of gambling, like casinos and lotteries. 

Further Reading

• Scott Alexander’s Prediction Market FAQ

• Superforecasting: The Art and Science of Prediction by Phillip Tetlock and Dan Gardner.

• Gwern on prediction markets

• Idea Futures by Robin Hanson

• New Yorker Article about PredictIt