Unsafe Bets

After receiving enormous backlash for giving Donald Trump 28.6 percent odds of winning the U.S. presidential election in 2016, Nate Silver responded to his naysayers loud and clear that he was not only not wrong, but right for giving Trump those odds. The problem was not with the numbers themselves but with the manner in which they had been read. Silver argued that, insofar as other models gave Trump worse odds (at around 17 percent), his odds should’ve indicated that something was amiss. “From the standpoint . . . of skilled gamblers and like-minded folks,” such as himself, Silver wrote in his recent book On the Edge, his 2016 election forecast “was a damned good forecast.” Why? “For a simple reason: if you’d bet on it, you would have made a lot of money.”

As much as Silver insisted upon being right, or at least the most right, he now concedes that the utility of his forecast to the average member of the public and not to the skilled gambler is, at best, unclear. Silver’s writing regarding the 2024 U.S. election—which he has consistently compared to a coin-flip—has echoed this. “Go vote,” he wrote in a recent Substack post, “if you care about what happens in this election, don’t sweat the polls.” 

Silver departed the polling and sports website he founded in 2008, FiveThirtyEight, in 2023, leaving to start his aforementioned Substack, Silver Bulletin. A year later, during the summer of 2024, he joined a company called Polymarket as a paid advisor. There, Silver found a community that could benefit from his work: “skilled gamblers and like-minded folk” very much included. 

Polymarket is a “de-centralized betting platform”—a betting website where bets are placed in cryptocurrency—where users can wager on pretty much anything. The sorts of wagers, or “prediction markets,” it hosts are submitted by users via the messaging platform Discord. Running through a few of them will quickly give you an idea of the kind of place it is. “Caroline Ellison prison time?,” “Premier League Winner,” and “Will Israel invade Lebanon in September?” have all featured on Polymarket’s homepage in recent months. Startlingly large piles of money are riding on these bets too: more than $13 million was bet on whether Israel would invade Lebanon in September. As of this writing, over $3 billion has been bet on who will win the presidential election. 

This October, the twenty-six-year-old founder of Polymarket, Shayne Coplan, sat down at Mainnet—“the largest annual crypto event in New York”—to be interviewed by Silver. (Neither Silver nor Coplan responded to requests for comment.) There, Coplan recounted the origin story of Polymarket. At the onset of the pandemic, he told Silver, it was clear that we needed prediction markets to tell us when the vaccine would come, when New York was going to open back up, and when the mask shortage would be over. These markets, Coplan said, would allow people like him to figure out the odds of these sorts of things actually happening, and free them “to not have to listen to the talking heads on Twitter.” The idea was that Polymarket’s prediction markets would require users to put their money where their mouth is. And, as Coplan made clear to Silver, Polymarket’s “product-oriented approach”—which Coplan credits with separating Polymarket from the pack of lesser-known competitors—makes putting your money where your mouth is very easy. For example, the shares you purchase on Polymarket to place a bet are priced at the odds of the thing—e.g., a Kamala Harris victory—happening. Right now, one Harris-will-win share is priced at a bit under forty-one cents; so, according to Polymarket, Harris has around 41 percent odds of winning the presidency.

Polymarket is backed by major venture capital and cryptocurrency investors, including Peter Thiel’s Founders Fund; Vitalik Buterin, the inventor of Ethereum; and Balaji Srinivasan, the tech baron known for advocating for the creation of autonomous “network states.” All told, Polymarket has raised from those three and others over $100 million as of May. But as many dollars as $100 million is, what’s more intriguing about Polymarket—and perhaps why tech’s most ideologically driven barons, Thiel and Srinivasan, are so close to the project—is that the platform is trying to be something more than just a betting website. In fact, on X, Bukerin flatly denied that Polymarket was a gambling platform, writing that, instead, Polymarket should be seen as “social epistemic tool” useful for the public to get “a view of how important certain events are and what kind of things are likely to happen.”

Prediction markets have existed in some vague form for centuries. In the early 1500s, for example, it was a popular pastime in Italian city-states to wager on papal elections, tennis, and even the outcomes of pregnancies. But Coplan’s self-described inspiration for Polymarket, which he outlined to Silver at Mainnet, came later. He cited Friedrich Hayek’s seminal essay “The Use of Knowledge in Society,” published in 1945, which conceives of markets as vehicles of information first and foremost. Then there is the work of George Mason economics professor Robin Hanson, who’s sometimes called the godfather of prediction markets. In the early aughts, Hanson was involved in a bid to start something called the Policy Analysis Market, through the Defense Advanced Research Project Analysis Market (DARPA). It billed itself as “a market in the future of the Middle East” where individuals could trade futures on geopolitical developments in the region (and, according to their old website, perhaps North Korea).

In 2003, Oregon Senator Ron Wyden called the DARPA project “morally offensive” and “grotesque,” claiming that it allowed for “trading on terrorism.” The project was shut down shortly after. But history repeats itself; fast forward two decades, and Wyden’s concerns are as relevant as ever (though as the Democratic Party’s preeminent crypto ally, he’s stayed mum this time around). Open polymarket.com, and click on any market to do with the Middle East, like, for example, “Israel x Hezbollah Ceasefire in 2024.” At the top of the page there’s a disclaimer, added after concerns were raised around the tactlessness of a “Hezbollah” section: 

The promise of prediction markets is to harness the wisdom of the crowd to create accurate, unbiased forecasts for the most important events to society. That ability is particularly invaluable in gut-wrenching times like today. After discussing with those directly affected by the attacks, who had dozens of questions, we realized that prediction markets could give them the answers they needed in ways TV news and Twitter could not.

Which attacks are being referred to is unclear. But click around some more. One market just closed: “Will [Yahwa] Sinwar leave Gaza by October 31?” The answer, according to Polymarket, is “no” because—and this was added as additional clarification after the bet was over—“As per the rules, Sinwar needs to leave Gaza himself for this market to resolve to ‘Yes.’” The denizens of Polymarket protested, arguing that his body was taken to Israel, so he did leave. A ruling was adjudicated by Polymarket’s arbiter, UMA, a so-called “de-centralized oracle service.” The answer remained “no.”

Outside of the morally repugnant Sinwar markets—there are others, like “Yahya Sinwar still Hamas leader before September?”—Polymarket also hosts markets which quantify death on a grander scale. These include the aforementioned market, “Will Israel invade Lebanon in September?,” and its accompanying “Will Israel invade Lebanon before November?” In addition to disregarding the substance of the question being posed and treating war like a casino, these two markets also reveal a structural issue bubbling under the surface of the platform. 

“Will Israel invade Lebanon in September?” was initially resolved to “yes” on October 1, but the decision was disputed and resolved to “no” shortly after; “Will Israel invade Lebanon before November?” was resolved to “yes” on October 5 and remained that way after two disputes. The former sounds suspect when you read the fine print: according to the market’s conditions, Israel had to invade Lebanon by 11:59 p.m. EST on September 30 for the bet to resolve to “yes.” A simple Wikipedia search of “2024 Israeli invasion of Lebanon” turns up an October 1 date for the invasion, but before 2:00 a.m. Eastern European time, and far before 11:19 p.m. EST on September 30. It’s strange then, that the “Will Israel invade Lebanon in September?” ruling was changed after going to UMA, the de-centralized oracle.

Only individuals with purchasable UMA tokens get to vote on whether to overturn decisions. (The price of one token, which accounts for one vote, is currently hovering around $2.) This creates a potential conundrum: Could someone with a lot of money on the line in a controversial prediction market buy more UMA tokens and swing the vote? Could, say, one very wealthy UMA token holder almost single-handedly out-vote every other UMA token holder? During the dispute of “Will Israel invade Lebanon in September,” ninety-four UMA individual token holders voted “yes,” and eighty voted “no,” but one of the eighty was a former Goldman Sachs managing director named Kevin Chan, who held significantly more tokens than anyone else—a little over three million. This prompted its own question of market manipulation: What if Chan had a lot of money placed on “no”—in an anonymized account—in the prediction market, helping create a result that he would then lock in using his UMA voting power? To note, Polymarket has overruled UMA before, as recently as June, but the platform does so rarely, and their intercession prompts a new set of concerns. What’s to stop Polymarket from siding with their wealthier users, the ones who individually provide the most liquidity to the markets—who make Polymarket work?

These are the sorts of concerns held by many chronically online X users obsessed with crypto. Like @Ape6867, who urged members of the community to file a complaint with the Commodity Futures Trading Commission over UMA—a common refrain if you spend enough time perusing the comment sections of disputed Polymarket predictions markets. But who cares what @Ape6867thinks? What any of these anonymous online characters think? It’s all rather stupid-sounding—a de-centralized oracle? These are reasonable questions. 

The first more obvious point of why this matters is that there are millions, and in the case of the election, billions, of real dollars on the line. There are also more than two hundred thousand monthly active traders on Polymarket, but only about 10 percent of users have made a profit. Then there is Polymarket’s new, increasingly mainstream role in the world. Bloomberg Terminal, the financial services platform favored by Wall Street, now incorporates Polymarket election odds. Most major publications also track Polymarket election odds. And in an informal conversation last spring, a BlackRock employee quite matter of factly told me that the company was making significant internal decisions based partly on Polymarket odds. The market “Ethereum ETF approved by May 31?,” about the odds of the SEC approving an Ethereum ETF submitted for approval by BlackRock, was how BlackRock was internally tracking their application, the employee claimed.

If you think about it hard enough, it begins to make sense why a company, or anyone, looking for specific information would go to Polymarket for answers: given that the platform has no way of preventing insider trading, you might assume that its odds are informed by desirable insider information. That’s actually sort of the point: prediction markets exist, in theory, to produce the most accurate numbers, which are only possible with insider information. Robin Hanson has argued as much. But when someone bets over $28 million on Trump winning, like the linked accounts Fredi9999, Theo4, PrincessCaro, and Michie did this October, does that actually indicate that they know something the general public doesn’t? Does it indicate they have a lot of money and don’t care about throwing it around? Or is it an attempt at market manipulation?

Speculation about the identity of Fredi9999 and the associated accounts reverberated throughout both the Polymarket community and mainstream news outlets. Late last month, the company revealed to the New York Times that the accounts were held by one person, a French national “with extensive trading experience and a financial services background.” Polymarket told the Times that it found no manipulation had occurred, and that the individual agreed “not to open further accounts without notice.” Yet it’s unclear how exactly Polymarket will know whether the bettor holds up their end of the deal: to even identify the user, the company reportedly had to recruit the help of corporate investigations firm Nardello and Company.

In a New Yorker interview with Kyle Chayka, Coplan pushed back against fear mongering about “whales,” or big-spending accounts, influencing the odds of their markets. “Everyone’s making a fuss, but it’s only sixty per cent,” he said on the subject of Trump’s odds on Polymarket at the time. “It’s still nearly neck and neck,” he continued, mirroring his colleague Silver’s claims that the 2024 election outcome is essentially a coin-flip. While this may be accurate—although 60/40 is pushing it—Coplan’s response partially serves to deflect future backlash about the accuracy of the prediction, to remind people that if something has, for example, 40 percent odds of happening, that means it still happens 40 percent of the time. But it also obscures the fact that the odds themselves and their attached returns, not just the events being predicted, are what reel people in. 

For now, Polymarket, which operates off-shore although most of its employees are based in New York City, is technically off-limits to American bettors, a problem easily fixed with a free VPN. Last month, a federal appeals court ruled that the Polymarket competitor Kalshi—a regulated prediction market—can offer U.S. election betting to American customers. Whether or not this specific election-betting ruling will entice Polymarket into opening into the United States is unclear, especially given that its presidential election market is not only far and away its largest but the largest among all of its competitors. The law isn’t the only factor deciding if Polymarket is here to stay or not, however. There’s also the question of whether the platform will ever make any money, as it currently does not (much like most of its users). Ironically, the sort of rationalism behind Polymarket—the idea that an accurate prediction requires an unregulated, efficient marketplace that punishes bad, and seemingly most, predictions—has rarely been applied, even in the press, to Polymarket itself. 

The question of what Thiel, Srinivasan, Bukerin, and even Coplan stand to materially gain from Polymarket in the long term remains unanswered. The most likely and most boring answer is lots of money: Coplan has hinted at fees being part of Polymarket’s future. While this would introduce some ethical concerns—like that Polymarket would itself be profiting off of its Middle East markets—it would also hardly be a surprise. However Polymarket eventually decides to make money, its exponential growth in traffic and its market capture would indicate that the company will probably figure it out. At least, barring a huge post-election drop-off in traffic, an FTX-style collapse (which no one should rule out), or legal trouble. And Polymarket’s potential legal issues are not limited to regulators and Commodity Future Trading Commission oversight; in its current state, it has no “know-your-customer” (KYC) procedures—that is, no identity verification processes—which means that anyone can anonymously create a “wallet” to place bets on the platform. This shroud of secrecy not only works to protect insider traders but also serves to make the inner workings of Polymarket opaque. 

Yet only to a degree: while no one knows who is behind the accounts buying and selling shares, the activity of the anonymous accounts is quite visible. A recent investigation by the crypto research firm Chaos Labs found that around one-third of trading volume across all of Polymarket’s prediction market appeared to be “wash trading,” an illegal form of market manipulation where shares are simultaneously, and sometimes repeatedly, bought and sold to give the impression of market activity. The investigation also found that, while Polymarket was then reporting $2.7 billion in trading volume on its presidential election market, the real volume was likely closer to $1.75 billion. Chaos Labs found that Polymarket had been counting shares purchased, like a Kamala-will-win share, as dollars spent, rather than the actual value of shares—practices hardly suitable for a gambling platform angling for mainstream legitimacy, let alone one with aspirations of being a news source and corporate oracle.