Intro
I’ve coined “politically-mediated polarization” to be the process in which unpopular candidates narrowly win elections, anger opposing voters, and in so doing create conditions for the process to repeat. In America, we’re only going to break this cycle by electing broadly popular leaders who engender less negative polarization. To do so, we need more broadly popular candidates to win their primaries. I’ve argued elsewhere that predicting general election outcomes before the primary can do this. This post explores the use of prediction markets in this context.
Money Markets
Let’s say you want to create a prediction market for the general election outcomes before the primary votes are cast.
One way to do that is to create a combinatorial market. For example, one party has candidates Alice, Bob, and Charlie running, while the other party has an incumbent, Dave, who has already advanced to the general election.
Anyone who wants to bet places a $10 wager and selects a winner for each of the three possible matchups.
Payout is simple: once the general election is decided, you check everyone’s bets to see who called the winner in that matchup. Say that Alice wins the primary and runs against Dave. If 100 people bet, and of those, 70 bet on Alice while 30 bet on Dave, and Dave wins, then those 30 people split 1000 dollars and earn 33 dollars each. More candidates expand the matchup matrix that betters must fill out.
Supposing this market were legal in the US. With some promotion, it should garner enough bets to create a thick enough market to interest super-forecasters and pollsters to participate, and thus end up with the information needed to get popular candidates elected. It’s not clear if it is legal, however. The only real-money political prediction market operating in the good graces of the CFTC is Kalshi, and they have significant limits in the kinds of markets they can run. In 2023, they were issued an order prohibiting them from offering a contract that would resolve based on which party would control Congress. Although they successfully filed a countersuit to have that order dismissed, it took two years, highlighting the uncertainty surrounding the legality of complex political bets.
Local Markets
While the above seems worth pursuing, the regulatory uncertainty makes me want to find an alternative that doesn’t require the CFTC’s approval. Let’s say we want to ask thousands of American voters how their neighbors would vote in hypothetical general election contests. Clever polling might work, but at a significant cost. Instead, what if we find a few nerds in every county (there are approximately 3,000 counties in the US), ask them to predict how their county will vote, and offer them non-cash prizes for accuracy? A safe prize is promotional in nature (e.g., clothing with the platform brand, a nice laser-cut of their county voting map, water bottles, notebooks, trophies, etc.). Another safe prize is a donation to a charity of their choice. And, of course, making correct calls earns you an invitation to participate again, resulting in more swag, kudos, and donations. A $300,000 per-election prize pool might be all it takes to build a convincing dataset likely to change how primary voters select their candidates.
You can take it further by training them and giving them better tools. Invite them to seminars on forecasting to help them calibrate their predictions. Give them dashboards and maps to consider. If you have relevant polling data for their county, share it with them. This should both increase their accuracy and help you select those likely to make correct calls.
A nice aspect of using a smaller set of people is that it’s more practical to ask them early and often. Early is helpful because a favorable score from these predictors may convince candidates to run when they otherwise assume they have a poor chance in their party’s primary. Often is helpful because we can check in as new debates (or scandals) occur. It’s not all upside, though, as if this system works and primary outcomes are nudged based on the input of less than 10,000 people who are merely given token compensation, it will become appealing to influence or otherwise buy their votes off. Thus, such a system works as a bootstrap to a future where these markets are seen as useful, and the next step is a market with monetary stakes.
The post will get even less interesting from here, as I am mostly thinking out loud about various details. If you read this far and know of any efforts of groups pointing in this direction, please let me know.
More Thoughts
One complexity of US Presidential elections is that the general election ticket will have a Vice President. Sometimes, these candidates also run in the primary (approximately 30% of the time in recent history), but often they are senators or governors who didn’t run. The choice of VP will influence general election votes, and so the most complete prediction market would consider every possible presidential slate, which also seems unrealistic.
Related, if the VP on a ticket were in the primary, their ranking in the prediction market (as if they were the Presidential candidate) would influence the chances they are selected as VP. This is especially likely if the markets were run regionally and thus could support the idea that a given VP pick is more popular in important states the presidential candidate is not.
Another consideration is how this impacts open primaries. The theory of an open primary is that it will reduce the chance of polarizing candidates advancing, negating the need for these markets in the first place. In California, since 2010, the primary gubernatorial election has been a non-partisan top-two primary. In the three that have run, a D and an R have both advanced. I would imagine a prediction market would be most helpful to Republican voters, as they can try to understand which candidate is most likely to win the general. If this started to occur, D’s would likely start paying attention as well, making sure not to pick someone who is too boring or negatively polarizing, who risks losing to the next Schwarzenegger. Thus, while these markets are mostly a salve for issues created by closed primaries, there is a benefit to open ones.
Could such a system encourage people to change parties when they run? Consider NY, which has closed primaries and a heavy left skew. It’s been over 30 years since a R senator was elected there. Many of the R candidates who advanced past their primaries are far from moderate, being Tea Party members or other firebrands. These have no chance of winning the general, and yes block moderates Rs from running. If there were a credible prediction market indicating that a current R could do well if they switched to D and entered the senatorial primary, might they?
One aspect I like about political prediction markets is that they provide a source of revenue for research & polling firms. Accurate firms are a large societal good, and so it’s important that the correct ones get resources while the incorrect ones fold. Their current desire for accuracy is professional pride and a sense that it impacts their business with corporate clients, which seems insufficient to me.