A Letter to the CFTC
A few weeks ago, the Commodity Futures Trading Commission decided to withdraw permission for PredictIt to continue operating in the United States, and ordered the exchange to liquidate all contracts by mid-February. Meanwhile, the Commission has been entertaining a proposal from Kalshi to offer contracts referencing congressional control, and has solicited comments. I have posted a supportive comment letter, reproduced below with some minor edits. The comment is less about this specific proposal than about prediction markets more generally, and their role in a complex and polarized world.
Dear Chairman and Commissioners of the Commodity Futures Trading Commission:
I am writing in support of the Commodity Futures Trading Commission approving Kalshi’s proposal for electoral prediction markets.
There are essentially two approaches to predicting the future. One is model-based, and relies on sound scientific understanding of the data generating process. The other is crowd-sourced, and relies on the aggregation of decentralized information and beliefs.
The first approach works well for predicting regularly occurring events that are well understood, such as solar eclipses. But it is much less useful for predicting rare events that have a complex set of determinants, such as global pandemics or financial crises. For example, different research teams have produced widely varying forecasts of Covid-19 cases over the past two years, and even ensemble forecasts that average these predictions “have not reliably predicted rapid changes in the trends of reported cases, hospitalizations, and deaths” over time. In such cases, decentralized approaches that harness the wisdom of crowds can provide useful information.
Electoral outcomes lie somewhere between these two extremes. They arise with regularity, so forecasting models can be developed and estimated. But they also depend on idiosyncratic factors that are unique to each cycle, such as candidate quality or recent court decisions.
Ever since the launch of the pioneering Iowa Electronic Markets in 1988 (operating under a no-action letter by the CFTC), prediction markets have been part of the forecasting landscape for elections. The performance of such markets has matched that of poll aggregates, and is competitive with the best available models.
Prediction market contacts are extremely simple—they have binary payoffs with a fixed resolution date. In addition, the set of traders is relatively stable over short periods of time, and activity is sufficiently frequent to allow researchers to identify trading strategies. As long as the (suitably anonymized) trading data is made available, these markets can serve as experimental laboratories that help us understand precisely how information comes to be absorbed by financial market prices.
Electoral prediction markets reference positive feedback events—beliefs about the success of a campaign can affect the actual probability of success by influencing donations, volunteer effort, turnout, and so on. Campaigns routinely try to manage these beliefs, for example by selectively disclosing internal polls. Prediction market data can help uncover this process of attempted belief manipulation. For instance, in the process of examining trading strategies using prediction market data, David Rothschild and I found that a single trader had placed a sequence of several thousand orders over the course of two years leading up to the 2012 election, with non-negligible price effects, a finding that was covered by several media outlets.
We are living in an age that is characterized by both ideological and affective polarization—people in different political camps don’t just disagree on issues, they despise each other and rarely communicate. Some of this can be attributed to online echo chambers and filter bubbles, although more traditional sources such as cable television broadcasts are also implicated. Under these conditions, prediction markets play an interesting role. They are among the very few online platforms that create strong incentives for people who disagree fundamentally about statements of fact to interact with each other. A prediction market in which only one perspective is represented will attract people who disagree, since they will consider contracts to be mispriced, and will see a profitable trading opportunity. And trading losses can cause even the most stubborn individuals to reconsider their beliefs.
In order to leverage the power of prediction markets, however, the CFTC should allow for a large range of contracts, including those that reference individual races and not just national outcomes such as congressional control. This will allow people with very specific local knowledge to transmit their beliefs, even if they don’t understand the aggregate implications of what they know. In addition, it is important to have competition—multiple exchanges that offer similar contracts so that fees can be kept low and the implications of differences in market design can be investigated.
Thank you for the opportunity to comment.