What the Fed Found When It Tested Kalshi Against Its Own Decisions

Key takeaways

• A Federal Reserve working paper found Kalshi's macroeconomic markets produce well-calibrated, continuously updating density forecasts.
• The Kalshi median and mode posted a perfect record on the day before FOMC rate decisions, a statistically significant improvement on the New York Fed's own survey.
• Forecast errors on core CPI and unemployment came in almost identical to the Bloomberg consensus.
• This is the strongest pro-market evidence in the 2026 literature, and it comes from a hostile witness, since the central bank documented a betting venue pricing its own decisions about as well as the professional tools.
• The result describes the narrowest possible case: liquid, frequent, cleanly observable events with deep professional participation, which is a tiny share of category volume.
• Dr. Rajiv Sethi argues the mechanism's real edge is narrower still and points the other way, toward events with no historical precedent at all.

Why one working paper carries this much weight

Most claims about prediction market accuracy come from the platforms, from advocates, or from studies of election markets where the sample is a handful of events. The Federal Reserve paper is different. It tested a specific venue against the institution's own decision-making, on markets where ground truth arrives on a schedule, and it was published as a Finance and Economics Discussion Series working paper rather than as marketing. This piece covers what the paper found, why the finding is narrower than the headlines suggested, and where the evidence points next. Full treatment is in the forecasting chapter of the Hodl Up Prediction Markets Report 2026.

What the paper actually tested

The study examined Kalshi's macroeconomic markets, the contracts on FOMC rate decisions, core CPI and unemployment. Its central finding is that these markets yield well-calibrated density forecasts that update continuously, which is a stronger claim than simply getting the direction right. A density forecast assigns probability across the full range of outcomes rather than naming a single expected value, and calibration means those probabilities hold up against what actually happened.

Three results stand out.

The continuous updating is the part with practical value for a risk desk. A survey of professional forecasters is a snapshot taken at intervals. A liquid market reprices as information arrives, which means the probability is available at the moment a decision has to be made rather than at the moment the last survey closed. The paper's findings are the authors' own and not Board policy, and it remains a working paper, both of which belong in any citation of it. It is published at doi.org/10.17016/FEDS.2026.010.

The result is real and the band is narrow

Every condition that made the Fed result possible is missing from most of the category's volume.

Macro markets on FOMC decisions are liquid, they repeat on a published schedule, the outcome is perfectly observable, and professional participants with genuine models trade them. That combination is rare. On Kalshi, sports ran roughly 80% of cumulative volume, and the company told a federal court that sports contracts made up 80% to 90% of its listed contracts and revenue as of February 2026. The macro and political markets that carry the forecasting argument are the minority of the money. Our Kalshi vs Polymarket comparison breaks down the category mix on both venues.

Two further limits apply to reading the paper as a general endorsement. Calibration in a repeating, observable market says nothing about calibration in a one-off, contested, high-attention market, and separate 2026 research found the most socially visible markets produced the least accurate forecasts. And industry-wide Brier scores, including Polymarket's self-calculated 0.0843, are dominated by easy markets that resolve cleanly, which flatters the aggregate while saying little about the hard questions. Who actually sets those prices is its own problem, covered in The Wisdom of Crowds Is the Wrong Slogan for a Market 3% of Traders Price.

Sethi puts the edge somewhere almost opposite

The Fed paper locates value in liquid, repeating, cleanly observable events. Dr. Rajiv Sethi locates it in events with no precedent at all, where statistical models have nothing to calibrate against.

His examples are the June 2024 debate, when markets repriced within half an hour while polling-based models had to be suspended, and the pandemic, when more than 80 forecasting teams submitted projections to the CDC whose one-week-ahead forecasts disagreed by orders of magnitude. The reason those models struggled is structural rather than incompetence, because epidemiological models calibrated on historical data cannot absorb school closures, lockdowns or beach reopenings, and there was no comparable event within a century to calibrate against.

"It's a mechanism that gives you something when you're in uncharted waters."

The two positions are not in conflict so much as describing different products. The Fed found a well-behaved instrument in the one place it has enough professional flow to work. Sethi is arguing that the instrument's distinctive value shows up exactly where the alternative tools have nothing to offer, which is also where liquidity is thinnest and the evidence is hardest to gather. For a non-political test of the same idea, see the recent study on prediction markets and weather forecasts.

Both conclusions point the same direction for anyone building in the category. The forecasting value lives in a narrow band, it is genuine inside that band, and it is not what the volume is currently buying.

The full forecasting chapter, including the calibration literature and the concentration evidence, is in the Prediction Markets Report 2026. Our report summary covers the headline findings, and Sethi's full interview goes further on where the mechanism earns its keep.

None of this is investment advice.