Betting on Elections
For a few hours last Thursday the regulated prediction market Kalshi listed contracts referencing outcomes of the 2024 congressional elections in the United States—contracts that had previously been prohibited by order of the Commodity Futures Trading Commission. Trading began shortly after a federal district court vacated the order, and was halted when an appeals court granted a temporary stay.
In its emergency motion requesting the stay, the CFTC argued that “there is an acute risk of short-term manipulation of election markets and threats to election integrity.” As evidence for this assertion the agency cited some of my writing—a recent post in this newsletter and a paper written a decade ago with David Rothschild.1 Both pieces describe attempts at market manipulation, one for financial gain and the other seemingly for maintaining optimism about the prospects of a candidate.
However, I feel that the agency is drawing the wrong conclusions from this work, and a proper understanding of it undermines rather than bolsters the case for prohibition.
Some people believe that attempts at manipulating prediction markets are doomed to failure—that such attempts can have no more than a modest and short-lived effect on prices before other traders see a significant profit opportunity and pounce. I do not subscribe to this view. But when there exist prediction markets that lie outside the reach of our regulators, such as crypto-based Polymarket or the British exchange Betfair, the best defense against market manipulation is not prohibition but greater competition and transparency.
That is, I favor allowing Kalshi to proceed with the listing of contracts that reference election outcomes, not because I dismiss concerns about market manipulation or election integrity, but because I take them very seriously.2
There are some events—solar eclipses for example—that occur regardless of what we believe about them. Election outcomes are different. If we come to believe that a candidate is losing viability, morale will collapse, donors will close their wallets, volunteer effort will dry up, and turnout will decline. Such pessimism can feed on itself and become a self-fulfilling prophecy.
Similarly, beliefs that a campaign has momentum, even if initially erroneous, can generate actual momentum for a candidate.
This is why campaigns release some internal polls with great fanfare, while suppressing others entirely. They are trying to manipulate beliefs, of course, but we are not so easily fooled. We tend to discount such data, understanding that there are strategic and self-serving motives for its release.
Something similar applies to markets. It is true that affluent and committed supporters of a candidate may be tempted to try and shift prediction market prices in the hope that this will move beliefs in the electorate. But this can only succeed if the attempt remains invisible to us.
And visibility is sharpened by the existence of multiple competing markets, especially if they have limited participant overlap. Kalshi is a regulated exchange restricted to verified domestic accounts funded with cash. Polymarket is crypto-based, does not accept cash deposits from US residents, and lies largely outside the purview of our regulatory apparatus. Prices on the two exchanges may well diverge from time to time, but a sharp movement on one that is not reflected in the other and that cannot be traced to new information is likely to arouse suspicion. And manipulation of beliefs only works on the unsuspecting.
Prediction markets are characterized by an inescapable paradox. If they are taken seriously as unbiased aggregators of distributed information, beliefs will shift when prices change, and incentives for manipulation will be significant. But if they are seen as vulnerable to manipulation and frequently biased, price changes will be largely ignored, and they will not be worth manipulating. This logic places bounds on the extent of manipulation in markets—it cannot be absent altogether, but cannot be so large as to undermine their credibility.
Can the forecasting accuracy of markets exceed that of statistical models? This is an empirical question that I am in the process of exploring. One thing is clear—markets can take into account novel sources of information and handle historically unprecedented events to a degree that models simply cannot. This has been especially apparent in an election cycle that saw a major party nominee replaced at the eleventh hour, leading conventional models to lose continuity and reliability.
Perhaps election prediction in general adds no informational value, as some political scientists have recently claimed. But people will always seek out forecasts for events that matter to them. If they don’t have access to markets or models, they will turn to pundits and prognosticators, and leave themselves vulnerable to much greater degrees of manipulation. This can’t possibly be good for election integrity.
Update: Andrew Gelman (always worth reading) follows up.
I suppose it needs to be said that I have no financial stake in Kalshi, and nothing to gain from the removal of the stay.