A Poll-emic: Betting on Prediction Markets
My thinking on polls vs prediction markets
Polls are brittle, and prediction markets are robust.
For the 2024 Presidential Election, there is a stark contrast between the polls and prediction markets. The polls suggest the race is in a “photo finish” or a “dead heat,” while the markets—especially Polymarket—are predicting a landslide victory for Trump.
Polls ask a tiny group of voters which candidate it’s going to vote for. In a perfect world, we could confidently combine results of all these polls. But pollsters (defined as the entity publishing results not necessarily the entity conducting the poll) care more about their reputation than the truth, which is why sometimes they don’t publish results that dramatically vary from the consensus. This is called "poll herding” — or put more bluntly, a pollster is hedging its bets (hey, we all got it wrong). Another hedge is to forecast a tie which basically renders the poll pointless, but protects a pollster’s reputation. Polls got it wrong big time both in 2016 and 2020. Pollsters also use deceptive statistical methods to modify the numbers that get published.
Your belief in polls should go as far as you think the pollsters have integrity. If they’re willing to not publish non-consensus results, or apply statistical methods to yield more “favorable” results, they’re literally publishing misinformation. Ironically, these reputation-saving tactics erode their integrity.
I think one of biggest arguments against polls is that they produce lagging indicators in a world where candidates are dodging literal bullets or blundering with embarrassing gaffes. These sorts of events can evoke strong emotional responses in people. Elon Musk’s level of support for Trump dramatically spiked after the assassination attempt in July; the same is true of millions of other Americans. Polls don’t have a real-time pulse on that “level of support”. There’s not enough time update poll data to measure the pulse after each of Trump’s or JD Vance’s appearances on Joe Rogan’s podcast. Or if Kamala has an embarrassing gaffe on stage, that won’t be rolled into the poll data. Prediction markets measure this and give us a leading indicator.
Prediction markets are robust because millions of people (a subset of which are voters) are investing in what they think will happen—they’re betting on outcomes. Polymarket successfully called JD Vance as Trump’s choice for VP well before it happened. That’s remarkable because it’s not super obvious—JD Vance was once a “never-Trumper”, and it would follow that Trump would see him as an adversarial. Polymarket also successfully predicted Joe Biden’s departure from the party, even after Biden reiterated he’d never step down. Not all prediction markets are equally accurate: I think accuracy increases with betting volume.
The biggest arguments against prediction markets in no particular order:
Markets reflect biases of the bettors / people bet tribally.
Men are over-indexed in the population of bettors.
Bettors have more discretionary income than the median voter.
Bettors skew younger because crypto is the primary I/O for bets.
Markets are banned in America, so its forecasts are moot in America.
Markets perpetuate false likelihoods with momentum.
1 - there is some of this, sure. But again, with enough betting volume, it’s not a big factor. If there’s $100,000 in total bets placed that Taylor Swift is releasing a new album in November, you could chalk that up to fervent and hopeful fans. If there’s a staggering $2,000,000,000 in total bets, she’s dropping the album in November.
2 - 4 aren’t actually flaws. These would be fatal flaws for polls, which depend on representative sample populations, but not for markets. Newsflash: representation is irrelevant for markets. Bettors have a financial incentive to be right. It doesn’t matter if you’re a man or woman, or young or old—you will seek out high quality data and make informed bets.
5 - the internet does not have political boundaries, nor does crypto. Getting around these virtual geographical limitations is a walk in the park for crypto bettors. But more importantly—to reiterate my previous point—betting is about the outcome. It doesn’t matter who is betting because they’re financially incentivized to seek out the best data.
6 - hmm, yeah ok. But again, with enough betting volume, it’s not a big factor.
Coda
Predicting the future is a losing game. If you’re wrong, you’re a loser. If you’re right, it was obvious all along. Polls play this game; prediction markets play this game. But it’s clear that one of them plays it better. You could argue that prediction markets are a superset of polls because bettors have at least glanced at poll data before placing their bets. And if it’s a superset with three billion dollars in bets signaling which way the scales will tip, you ought to pay attention.