Orhan Torul bio photo

Orhan Torul

Associate Professor of Economics at Boğaziçi University

  X LinkedIn   Google Scholar   ORCID E-Mail

Thirty-Five Minutes of Silence

On May 15, 2024, the U.S. Bureau of Labor Statistics loaded a subset of the April Consumer Price Index files onto its public website roughly thirty minutes before the scheduled 8:30 release. For half an hour, inflation figures that routinely move bond, currency and stock markets around the world sat on a government server, reachable by anyone who thought to look for them, with no announcement that they were there.

Incidents of this kind are unfortunate for the statistical agency and useful for researchers. Economists who study how quickly markets absorb news usually face an awkward obstacle: we rarely know the exact moment at which the news became available, so we have to infer it. Here we do know it, because the agency afterwards confirmed the time. In a recent paper, my co-author Emrehan Aktuğ and I use that timestamp to ask a simple question. When important economic news quietly becomes available, how long does it take a betting market to notice?

The question is not merely academic. Prediction markets, where people buy and sell contracts that pay out if an event happens, are increasingly quoted as real-time forecasts on financial news and in policy discussion. Their usefulness in that role depends on how quickly they absorb information. In the professional bond and currency markets, a scheduled economic announcement is reflected in prices within seconds. For the retail betting platforms that anyone can join, there has been very little comparable evidence, particularly minute by minute.

The Setting and the Data

The leaked report showed inflation running slightly cooler than expected, 0.3% for the month against a forecast of 0.4%. The gap is small, but its meaning was clear to anyone who follows monetary policy, since softer inflation makes interest rate cuts by the Federal Reserve more likely. That gives us a clean test. If several different markets price the odds of rate cuts, we can watch the clock and see which of them reacts first.

We follow three sets of prices, minute by minute:

  • Polymarket’s “How many Fed rate cuts in 2024?” contracts: our main subject. Polymarket is a betting platform where anyone can trade contracts on future events. This particular market offered a separate contract for each possible answer, from zero cuts up to six or more. Each contract pays one dollar if that answer turns out to be right, and nothing if it does not. The price therefore reads directly as a probability, so a contract trading at 30 cents means the market puts the chance at roughly 30%. We combine the contracts into a single figure, the expected number of cuts.

  • CME Federal Funds futures: our professional benchmark. This is the contract banks and funds have used for decades to bet on, and hedge against, Federal Reserve decisions. It prices the same underlying question as the Polymarket contracts, but it trades on an established exchange among institutional participants.

  • Bitcoin and Ethereum: background only. Both are sensitive to interest rate news, but only indirectly, so we do not treat them as a benchmark. They are here to show what a market watched around the clock by automated trading systems was doing at the same moment.

1. The Professional Market Reacted Within Seconds

The futures market moved almost at once. By the time the official 8:30 release arrived, Fed Funds futures had already made close to 60% of the entire price move they would make between 8:00 and 9:00. Someone, or more likely some piece of software, found the files on the server, read them, and traded on them before the government had announced anything.

The crypto prices are consistent with this. Bitcoin rose 0.31% and Ethereum 0.12% during the half hour in question. Only the Bitcoin move is larger than these assets typically drift before a CPI release, and with just four comparison mornings we cannot make much of it statistically. We report it as a reminder that continuously monitored markets were awake, not as evidence that they were pricing the leak.

2. Polymarket Did Not Move for Thirty-Five Minutes

Over the same half hour, Polymarket’s prices did essentially nothing. The first real price change comes at 8:35. We do not choose that moment by hand; a statistical procedure designed to find the point at which a series changes behavior locates it on its own. That is thirty-five minutes after the files went up, and five minutes after the official release, by which time the news was on every financial screen in the world. Once the adjustment began it did come through, and by 9:00 the contracts had moved roughly six times as far as they had during the entire half hour of the leak.

The order book makes the point more vividly than any statistic. On Polymarket, as on any exchange, there is a best price at which someone is currently willing to buy and a best price at which someone is willing to sell. Between 8:00 and 8:29, across every contract in the series, those two prices changed a total of three times. On four ordinary CPI mornings we use for comparison, the same half hour saw 11, 16, 20 and 50 changes. That is not a market adjusting slowly; it is a market whose quotes were largely being left untouched.

3. The Thirty-Five Minutes Are Really Two Different Stories

The obvious conclusion is that the betting market simply failed. We think that is too quick, and separating the thirty-five minutes into two parts is the most important thing the paper does.

Consider first the thirty minutes from 8:00 to 8:30, when the information was available but unannounced. Here the silence is close to what economic theory would predict, for a reason that comes down to the cost of trading. Buying a contract on Polymarket meant paying slightly more than the quoted middle price, and selling meant accepting slightly less. That difference, the bid-ask spread, was one to two cents on these contracts. Since each contract pays a dollar or nothing, one cent of price is one percentage point of probability. A trader therefore had to believe the true odds had shifted by at least half a percentage point, and more likely a full point, before trading was worth doing at all.

How far had the odds actually shifted? By about half a percentage point at most, because the inflation surprise itself was so small. So the expected gain was, at best, exactly what the trade would cost, and more probably less. A trader who had found the leaked file and done this arithmetic would rationally have done nothing, and we make no claim of failure for this window.

The last five minutes, from 8:30 to 8:35, are a different situation entirely. By 8:30 the report was public, the professional futures market had already made most of its move, and no informational advantage remained to be protected. The news was by then far larger than the one to two cent cost of trading, so the explanation that works for the first window no longer applies. This five-minute delay, occurring after the whole world already knew, is what we call informational inertia, and we use the term only for it.

What the Episode Does Not Show

An episode like this invites overstatement in both directions, so three qualifications deserve emphasis:

  • It is not about time zones: Polymarket blocked U.S. users at the time, which raises the possibility that 8:00 in the morning in New York simply found its traders asleep. The clock says otherwise. That hour is 1:00 in the afternoon in London, 2:00 in the afternoon across continental Europe and 8:00 in the evening in Singapore, all ordinary trading hours for the platform’s largely non-American user base. Being awake, however, is not the same as watching. What the platform lacked was the automated machinery that alerts professional traders the instant a data file appears.

  • It is not a verdict on prediction markets as a class: what the episode exposes is the cost of one particular setup in one particular month. Polymarket in May 2024 had no market makers paid to keep its quotes fresh, no automated news feeds, and none of the fast direct data connections professional venues rely on. The gap we measure concerns who was watching, not the technology used to match trades. Bitcoin trades on a conventional centralized exchange and Polymarket does not, but that difference is not what separated them on May 15. Automated systems were watching Bitcoin and the futures market alike. Nothing comparable was watching Polymarket.

  • It is not a general estimate: this is one incident. We looked for others and found none, since no comparable verified early release has been documented on Kalshi, PredictIt or any other platform. With a single event and four comparison mornings, our statistical tests are better understood as setting bounds than as proving a hypothesis, and we present them that way. One question does deserve a straight answer, namely whether the silence we found is simply too faint for our tests to detect. It is not. The smallest movement our design could reliably have caught is more than twice the movement we actually observed.

Why the Episode Is Worth Recording

Prediction markets are good at pooling the judgments of many people over days and weeks, and that is a real strength. At the scale of minutes, though, a market is only as alert as the people and programs watching it. In May 2024, on this platform, almost nobody was, and an accidental early release is what made that visible.

Polymarket has changed a great deal since then. Trading volume, market depth and automated participation have all grown substantially, and the same incident today might well be priced within seconds. That is precisely why the episode is worth writing down. It gives us a before picture, a measurement of how quickly a major betting platform reacted to real news before the money and the machines arrived, against which later measurements can be compared.

Whether closing the remaining gap requires paid market makers, better technical access for automated traders, cheaper movement of funds between platforms, or simply more traders, is an open question about market design. Future episodes, on Polymarket or on competitors such as Kalshi and PredictIt, will help answer it.

References