The question of whether crowd-aggregated forecasts beat structured expert judgment has accumulated enough data to move past ideology. Two 2026 papers, one from the Federal Reserve and one from academic epidemiologists, bracket the answer: prediction markets outperform professional forecasters in domains where participants are financially incentivized and informationally competitive, and they underperform specialists when neither condition holds.
Where Markets Win: Finance and Policy Rates
A February 2026 working paper authored by economists at the Federal Reserve Board, Northwestern University, and Johns Hopkins University analyzed trade-level data from the largest federally regulated prediction market platform, overseen by the CFTC, to evaluate its accuracy against the Federal Reserve Bank of New York's Survey of Market Expectations, Bloomberg consensus estimates, and fed funds futures.1
The paper found that the platform's forecasts for the federal funds rate and CPI provide "statistically significant improvements over fed funds futures and professional forecasters, all while providing continuously updated full distributions rather than infrequent point estimates," per Diercks, Katz, and Wright (Federal Reserve Finance and Economics Discussion Series, February 2026).1 The practical implication of that last clause matters as much as the accuracy finding itself: for federal funds rate forecasts 150 days ahead, the prediction market's mean absolute error was similar to professional forecasters, but unlike the periodic survey, the market offered a continuously updating full distribution.1
That frequency advantage compounds over time. A panel of economists submitting quarterly point estimates cannot reprice in real time when a jobs print or a Fed governor's remarks shifts the probability distribution. The market-implied probabilities respond sharply and sensibly to major developments, with intraday repricing around FOMC communications that daily survey data misses entirely.
The calibration record in electoral markets extends this pattern. The Iowa Electronic Markets outperformed traditional opinion polls 74% of the time across several U.S. presidential elections, according to a 2026 working paper from the University of Iowa. Across a larger sample, Clinton and Huang (2025) analyzed over 2,500 prediction markets with approximately $2.4 billion in trading volume across the 2024 election cycle and found accuracy, measured as the percentage of markets correctly predicting outcomes better than chance, varied significantly by platform: 93% for the most tightly regulated venue, 78% for the CFTC-registered exchange, and 67% for the highest-volume decentralized platform.
That spread is the key finding, not the headline averages. The lowest-volume, most tightly capped platform produced the most accurate forecasts; the highest-volume, most widely cited platform produced the least accurate. This does not prove that position caps cause accuracy improvements, but it is consistent with the hypothesis that per-trader limits preserve forecasting quality by preventing concentrated positions from dominating visible prices. Volume is necessary but not sufficient. Market structure shapes calibration in ways that aggregate volume statistics conceal.
Where Specialists Win: Expert Ensembles Beat Thin Crowds
A May 2026 preprint by Dudley and Magdaleno tested whether prediction market performance in elections and finance generalizes to infectious disease forecasting, evaluating market-implied forecasts during 2025 and 2026 for both weekly influenza hospitalizations and monthly measles cases.2
Across both settings, prediction markets failed to outperform standard benchmarks. For influenza, markets were competitive with low-performing individual models in the CDC's expert forecasting ensemble but were dominated by the ensemble itself; even combining market forecasts with the ensemble, the best combination assigned zero weight to the markets. For measles, markets were outperformed by simple statistical baselines.
Participants in these markets are self-selected and typically lack epidemiological expertise, but markets can respond in real time to emerging news and unstructured signals in ways that standard forecasting pipelines cannot. The problem is that real-time responsiveness is only an advantage when participants know what signals to respond to. For disease dynamics driven by pathogen biology and surveillance reporting schedules, the crowd's edge, spotting fast-moving news, cannot overcome the specialist's edge, understanding the underlying process.
The researchers diagnosed two sources of market inefficiency: placement of probability mass on impossible outcomes (for example, decreasing values in cumulative hospitalization forecasts) and low trading volume. The first failure is a domain-knowledge problem: participants who do not understand that cumulative counts cannot decrease will misprice systematically. The second failure is a liquidity problem: thin markets attract fewer informed traders and are more susceptible to idiosyncratic noise.
What the Pattern Actually Shows
The literature is not resolving a binary contest. It is mapping a competency boundary. Prediction markets aggregate information effectively when three conditions align: participants have genuine informational advantages (not just opinions), liquidity is deep enough to attract skilled traders and discipline noise, and the contract structure maps cleanly onto the underlying event. Accuracy is highest in liquid, high-volume markets, with contracts above $500,000 in volume averaging materially better calibration; below that threshold, calibration degrades sharply.
Expert panels, meanwhile, hold the advantage wherever deep domain knowledge is required to interpret the signal. When experts provide probabilistic rather than point forecasts, both the number of experts and diversity are helpful for the panel's overall statistical accuracy, which in turn correlates positively with lower forecast error. The epidemiological ensemble that outperformed crowd markets is itself a probabilistic aggregation, just one confined to specialists.
For allocators evaluating the information value of market-based forecasts, the relevant question is not "do prediction markets beat experts" but which forecast source is epistemically better positioned for a specific domain and whether the market is liquid enough to express that position with discipline. The Fed's working paper and the Dudley-Magdaleno preprint, read together, answer both halves of that question more precisely than the headline narratives around either one.