This week, I would like to explore a subject that sits at the crossroads of my background as a scientific engineer, my experience in the business world, and my earlier life as a sports betting trader. I want to talk about prediction markets.<\/p>\n\n\n\n
The concept of prediction markets is not new. I remember studying the phenomenon when writing my book on sports betting in 2010<\/a>, drawing on an academic paper by Justin Wolfers<\/a> published in… 2004.<\/p>\n\n\n\n At the time, prediction markets were still in their infancy. Beyond a handful of near-obscure websites and markets created in academic research contexts, the largest prediction market was an online sports betting platform, Betfair, which offered wagers on a significant number of events but with sufficient and satisfying liquidity only for major competitions. That said, the reliability of this prediction market was well-established within the bookmaking world. As a young “trader” at Unibet at the time, responsible for setting real-time odds on certain sporting events, I had to constantly monitor that the odds offered to players, generated by our statistical model, did not diverge from Betfair’s “opinion.” When they did, I had to correct the machine by hand.<\/p>\n\n\n\n The beauty of prediction markets lies in their ability to efficiently aggregate the views of imperfectly informed participants. Prediction markets are, for example, more “complete” than opinion polls. Polls are notably imperfect: the respondent is passive (they are solicited regardless of their interest in the subject), expresses only their own opinion (when it hasn’t already been adjusted ex post by the polling firm analyzing the data), has no incentive to reveal their genuine preference (beyond simply not lying to the pollster), and cannot guarantee that their opinion will be acted upon (an electoral poll cannot guarantee that all respondents will actually vote). Prediction markets are the exact opposite: the bettor must take a position by placing themselves in the shoes of the group, aligning their words with their financial incentives, the very principle of a wager.<\/p>\n\n\n\n The accuracy of prediction markets was on display in the last two U.S. presidential elections. In 2020, prediction markets forecast that Biden would win 308 Electoral College votes<\/a>; the actual result was 306. In 2024, while polling firms predicted a neck-and-neck race between Trump and Harris, prediction markets foresaw a comfortable victory for the former<\/a>. That is precisely what came to pass.<\/p>\n\n\n\n Presidential elections represent ideal cases for prediction markets. To attract a sufficient volume of bets, the underlying events must concern a broad enough audience, even if each member of that audience has only incomplete or imperfect knowledge, and must have an outcome that cannot be influenced by a small number of privileged decision-makers. It is precisely this divergence of views that fuels disagreement and drives bettors’ willingness to place substantial wagers. This instinct has been formalized by the Kelly criterion<\/a>, which recommends increasing your stake the more your view diverges from the probability implied by the market. In practice, this means the most informed bettors, those with the sharpest private signal, have the strongest financial incentive to act on it, which is precisely the mechanism that drives price accuracy.<\/p>\n\n\n\n The increasingly rapid spread of information, combined with the legalization of sports betting markets in the United States and the cultural shift it has triggered, has fueled an unprecedented rise in prediction market platforms. The most prominent of them, Polymarket<\/a>, now records annual revenues exceeding $300 million<\/a> and could be valued at over $20 billion<\/a>, a sign of the strong future growth expected.<\/p>\n\n\n\n Unfortunately, prediction markets are not immune to the ailments that afflict other markets, financial or sports betting, and recent events prove it. The first pitfall is insider trading, whereby a small number of actors with so-called “privileged” information adjust their behavior in the market at the expense of uninformed participants. Last month, more than 150 members of the Polymarket platform had each bet $1,000 or more in favor of a U.S. strike on Iran<\/a>, just 24 hours before the attack took place. By revealing privileged information so transparently, prediction markets not only harm other participants, they also pose a threat to national security. This is not merely speculative: the timestamps of those wagers are public on-chain data, and several security researchers flagged the anomaly in real time. It raises a question that regulators have not yet answered: at what point does a prediction market become an inadvertent intelligence leak? Insider trading in connection with prediction markets currently benefits from a relative legal grey area, since there is no equivalent of the SEC in the United States or the AMF in France, the bodies responsible for investigating suspected insider trading in equity markets, operating in this space.<\/p>\n\n\n\n The second pitfall is corruption. Market depth is a necessary condition for prediction markets to function properly. Greater depth, however, comes hand in hand with growing financial stakes, making corruption operations more economically viable. The phenomenon is not new: in my book I discussed, for example, how certain athletes, particularly in lower divisions, could be tempted by criminals able to offer them more money to throw a match than they would earn by winning the tournament. The proliferation of betting offerings makes detecting such behavior increasingly complex. In sports betting, it is possible to wager not only on the result of a tennis match, but also on the outcome of the first point, the first game, and so on, details that can be subtly influenced.<\/p>\n\n\n\n Despite this, with a small leap of imagination, we can envision the usefulness of prediction markets withincompanies. In truth, this is not a new idea: in the late 1990s, HP used an internal prediction market<\/a> to improve the accuracy of its printer sales forecasts, achieving results more precise than official predictions. In 2005, Google launched “Prophit,”<\/a> an internal prediction market used as a way to surface employee sentiment in advance, for example on the likelihood of success of an internal project. Google attempted to open the concept to the general public before discontinuing the experiment in 2011. Today, many companies of sufficient size could implement a similar system, based on a virtual currency, through which employees would be asked to predict next semester’s sales figures, whether deadlines will be met for an ERP migration project (or, more realistically, estimate how late it will run), and so on.<\/p>\n\n\n\n Prediction markets are not a silver bullet. They are a remarkably efficient information aggrgation mechanism when the conditions are right: sufficient liquidity, diverse and independent participants, and outcomes that cannot be manipulated. When those conditions break down, markets fail as spectacularly as any other forecasting method. The real question is not whether prediction markets beat polls. It is whether we can design institutions, regulatory, technical, and cultural, that preserve their information-aggregating virtues while limiting the new risks their scale now creates. That is a problem worthy of both the engineer and the economist.<\/p>\n","protected":false},"excerpt":{"rendered":" This week, I would like to explore a subject that sits at the crossroads of my background as a scientific engineer, my experience in the business world, and my earlier life as a sports betting trader. I want to talk about prediction markets. The concept of prediction markets is not new. I remember studying the […]<\/p>\n","protected":false},"author":1,"featured_media":828,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-822","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-non-classe"],"_links":{"self":[{"href":"https:\/\/www.theFTexplained.com\/index.php\/wp-json\/wp\/v2\/posts\/822","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.theFTexplained.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.theFTexplained.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.theFTexplained.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.theFTexplained.com\/index.php\/wp-json\/wp\/v2\/comments?post=822"}],"version-history":[{"count":5,"href":"https:\/\/www.theFTexplained.com\/index.php\/wp-json\/wp\/v2\/posts\/822\/revisions"}],"predecessor-version":[{"id":827,"href":"https:\/\/www.theFTexplained.com\/index.php\/wp-json\/wp\/v2\/posts\/822\/revisions\/827"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.theFTexplained.com\/index.php\/wp-json\/wp\/v2\/media\/828"}],"wp:attachment":[{"href":"https:\/\/www.theFTexplained.com\/index.php\/wp-json\/wp\/v2\/media?parent=822"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.theFTexplained.com\/index.php\/wp-json\/wp\/v2\/categories?post=822"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.theFTexplained.com\/index.php\/wp-json\/wp\/v2\/tags?post=822"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}