James Surowiecki’s The Wisdom of Crowds begins with a story of British scientist Francis Galton, who on a brisk autumn day in 1906 went to a fair where attendees could guess the weight of an ox. At the time, Galton believed that the average guess would be quite far from the actual weight, and thought that the experts on livestock — the farmers and butchers — would be more accurate. Instead he left with a profound truth: under the right circumstances, groups are often smarter than the smartest people among them. The average of all 787 guesses ended up being 1,197 pounds. The ox weighed 1,198 pounds. In other words, the crowd’s guess was nearly perfect.
What made the average guesses of a county fair accurate in guessing the weight of an ox? According to Surowiecki there are four conditions that characterize wise crowds:
- Diversity of opinion: each person should have some private information, even if it’s just an eccentric interpretation of the known facts
- Independence: people’s opinions are not determined by the opinions of those around them
- Decentralization: people are able to specializing and draw upon local knowledge
- Aggregation: some mechanism exists for turning private judgements into a collective decision.
Can prediction markets be more accurate than polling?
First, it must be noted that prediction markets are not a recent invention in the US. In The Long History of Political Betting Markets, Paul Rhode and Koleman Strumpf describe Civil War era Wall Street betting markets, and that out of fifteen presidential elections between 1884 and 1940, those betting markets only were incorrect once. A University of Iowa study showed the Iowa Election Market’s (IEM) outperformed major national polls. In 49 different elections between 1988 and 2000, election-eve prices in the IEM were off by just 1.37 percent in presidential elections, on average. Although prediction markets can work well, they don’t always. IEM, PredictIt, another popular election market, didn’t get Brexit write, nor Trump’s 2016 win. On the eve of the 2016 election, PredictWise, a prediction market aggregator, gave Clinton an 89% chance of being the next president and assigning just an 11% chance for Republican Donald Trump, which was lower than Nate Silver’s FiveThirtyEight odds.
There is a record interest in prediction markets for today’s U.S. election. Betfair said that their “next President market is now the biggest betting event of any kind in history, including sports.” It’s also proven to be a popular draw to nascent prediction markets built on Ethereum, bringing in thousands of different bets and nearly two million $USDC and $DAI in volume. But do these Ethereum-based markets satisfy the conditions that Surowecki described for a wise crowd: diversity of opinion, independence, decentralization, and aggregation? I tried out three new Ethereum-based prediction markets — Polymarket, Omen, and Catnip — to see how wise they might end up being. (Disclaimer: Different jurisdictions have different laws around prediction markets, so do your own research to make sure it is legal to participate. Several of these markets are in beta, so don’t bet more than you’re willing to lose. )
The first market I checked out was Gnosis’s Omen Market, which has had the longest historical basis for predictions, starting June 12th. The user interface is quite intuitive, made by DXdao, and connecting MetaMask just took one approval which cost about $2 in gas.
Users can choose between “yes” shares if they think Joe Biden will win, and “no” shares if they think Trump will. The market uses $DAI, so putting a 100 $DAI bet will yield 169.17 shares. If Joe Biden were to win, the user would net about 53 $DAI in profit, and if Donald Trump were to win, the users would get nothing in return. If users were to choose “no” shares, and Donald Trump were to win, the user would net about 172 $DAI in profit.
The “trade history” displays the aggregation of voting changes over the lifetime of the market. On August 23rd a notable shift had 59% predicting Joe Biden would lose, but by September 8th, the collective market knowledge switched in favor of Joe Biden, giving him a 64% chance of winning. Since, the market has remained somewhat steady with 60% betting that Biden will win, and 40% betting Trump will win.
One cool feature is a 3Box comment section for participants in the market to ask questions like, “What if the result isn’t known (due to recounts, tied election going to the House, etc.) by Dec. 20th,” or lob unintelligible insults at each other’s preferred candidate. This is useful for getting information on the arbitration process, and the choice to use a Web3 native chat function is part of the composability that makes Ethereum great. However, if you need independence to create wise crowds prediction, it’s hard to say if people’s opinions — diverse as they may be with their own local knowledge — might not be influenced by the opinions of participants in the market.
Polymarket’s user interface is also very intuitive and walks users through a tutorial to show how to make a prediction. One gripe I had is that to join a prediction market, users must buy or deposit USDC into the Polymarket wallet, instead of simply connecting to your MetaMask wallet. Polymarket writes that the prediction market is “for informational and educational purposes only,” and does not take a profit, host the actual markets or “custody of anyone’s money or cryptocurrency.” It was not initially clear to me in their FAQ which existing prediction markets they pull from.
While this market has only been since October 9th, users have already deposited $335,453 in liquidity, and like other prediction markets, it’s giving Joe Biden about 60-40 odds to win.
I originally sought out to try Augur’s V2 prediction market. Launched in 2015, Augur was one of the first Ethereum dapps to gain notable attention for its prediction market, and also its enormous ICO. With V2, it’s the real deal in its Web3 stack, integrating with the Interplanetary File Systems (IPFS) for decentralized client storage, and using Uniswap for its AMM. Augur has no fewer than 4 current markets to bet on the outcome of the US presidential election, but the marketplace is much more intimidating. To get started you have to make three separate token approvals, each which costs about $0.50 in Eth.
Instead, I was lured to Catnip instead, a very beta-looking prediction market that was built on Augur and Balancer. The cute cat logo of Catnip promises “ultra-low fees” and that it is “100% non-custodial and decentralized.” The only market available right now is, “Will Trump win the 2020 U.S. presidential election,” and the market doesn’t settle until after January 22nd, but presumably participants can sell winning shares on Catnip or on Balancer after the outcome is known. You can actually check out the pool in Balancer to investigate the total number of swaps, liquidity, and volume. A whopping 1.3k swaps into the prediction market have already taken place, with about $1.1M in total liquidity.
It was easy to connect my MetaMask wallet, and from there, choose which shares I wanted to buy, either yTrump or nTrump, which represent “yes” and “no” respectively. Unlike Omen and Polymarket, these yTrump or nTrump tokens are added directly to your MetaMask wallet.
Will Ethereum-prediction hold up on election day?
Of the four considerations for gathering useful crowd wisdom, Ethereum-based prediction markets are advantageous in aggregating decentralized inputs, and trustlessly determining winning and losing bets. There is a case to be made that these same prediction markets could represent a diversity of opinion, not least due to MetaMask’s global user base of pseudonymous wallet addresses more interested in maximizing their bet than the political or material persuasions they may have for a particular candidate. But in a race so reliant on the U.S.’s peculiar electoral college system, decided by a few small regions of voters, is it possible for these nascent prediction markets to match up against the margins of error rigorous polling data provides at a county level?
What’s especially interesting about the Iowa Election Market mentioned earlier is that the typical sample size isn’t more than a few thousand traders, and definitely doesn’t reflect the makeup of the electorate as a whole. The majority of traders are white men from Iowa (it’s hard to imagine that Ethereum-based prediction markets are overrepresented by males as well). Yet participants in the market aren’t necessarily predicting their own behavior, but what they think voters in the US will do — and this has been shown to be at least as accurate as polling.
While some Ethereum-based markets have more liquidity than the IEM, they lack the sample size that would make them comparable to BetFair. Nate Silver’s aggregated polling site FiveThirtyEight gives Trump a 10% chance of winning, the collective intelligence of Ethereum-based markets indicate Trump potentially stands a 40% chance, higher even than the 36% chance in UK betting markets. Even though we will have to wait until Wednesday (or later) to know the how these prediction markets stack up this year, one thing is clear: Ethereum-based prediction markets can increasingly represent a simple and quick means to transform opinions into a single collective judgment, and may just improve the way we understand outcomes.