The field of market design has produced exactly one Nobel Prize winner who regularly thinks about prediction markets. Alvin Roth, who shared the 2012 economics prize for his work on matching markets and market design, has spent decades studying how markets function — and sometimes fail to function — in the real world. When he weighs in on whether prediction markets actually deliver on their promises, the industry would be wise to pay attention.
His verdict? It’s complicated.
The Humility Gap
“I don’t know if they do better on elections,” Roth said in a recent interview, cutting through years of triumphalist claims from prediction market advocates. This kind of intellectual honesty is rare in an industry where platforms routinely tout their forecasting superiority over polls, pundits, and conventional wisdom.
Roth’s skepticism isn’t blanket dismissal. It’s the measured uncertainty of someone who understands that aggregating information through prices is theoretically elegant — but empirically messy. The gap between what prediction markets should do in theory and what they actually accomplish in practice is where most interesting economic questions live.
The prediction market industry has grown accustomed to celebrating itself. Polymarket just crossed the billion dollar line in trading volume. Kalshi’s valuation has surged to $40 billion. DraftKings has entered the arena. But a Nobel laureate expressing genuine uncertainty about core performance claims? That should give pause to anyone writing checks — or placing bets — based on the assumption that these markets have cracked the forecasting code.
What Market Design Actually Teaches Us
Roth earned his Nobel for work far removed from election betting. He helped redesign the National Resident Matching Program, which pairs medical school graduates with residency programs. He developed kidney exchange markets that have saved lives by enabling complex chains of transplants. These are markets that work because the rules were designed with specific outcomes in mind.
The lesson from that work is crucial: markets are designed objects. They don’t spring from nature fully formed. And the design choices — who can participate, what information flows freely, how prices form, what manipulation looks like — determine whether a market produces useful signals or noise dressed up as insight.
Prediction markets weren’t designed by economists thinking carefully about information aggregation. They emerged from a combination of academic experiments, offshore gambling operations, and crypto experiments in decentralization. The theoretical case for them is strong. The implementation is another matter entirely.
Consider the 2024 election markets, which drew unprecedented attention and volume. Did they outperform polling averages? The honest answer is: it depends on when you looked, how you measured, and what baseline you chose. Some platforms showed significant Trump leads before polls shifted. Others whipsawed in ways that reflected liquidity dynamics more than information flow.
The Liquidity Problem Nobody Wants to Discuss
One reason Roth’s uncertainty matters is that prediction markets suffer from a fundamental tension that market designers understand deeply: thin markets produce unreliable prices.
When a market has few participants, a single large trade can move prices dramatically. This isn’t new information being incorporated — it’s a wealthy bettor pushing numbers around. The distinction matters enormously for anyone trying to use prediction market prices as forecasts rather than entertainment.
The industry’s response has been volume growth. More money, more traders, more liquidity. And to some extent, this works. Major platforms have seen record volume in recent months. But volume concentrated in headline events doesn’t solve the thin market problem for everything else. Most prediction market contracts — the congressional races, the state ballot initiatives, the policy outcomes — trade with spreads wide enough to drive a truck through.
This is where academic prediction market research often diverges from commercial reality. The studies showing impressive forecasting performance typically involve carefully constructed markets with motivated participants. Commercial platforms face different incentives: maximize volume, maximize engagement, generate headlines. These goals don’t always align with producing accurate forecasts.
The Regulatory Dimension
Roth’s willingness to express uncertainty stands in contrast to the certainty expressed by both prediction market advocates and critics in Washington. The politics surrounding prediction market regulation has become increasingly polarized, with defenders claiming transformative information benefits and opponents warning of market manipulation and gambling harm.
Neither side has engaged seriously with the empirical question Roth raised: do these markets actually work? The regulatory debate has instead focused on categorical questions — are event contracts derivatives or gambling? — that determine jurisdiction without illuminating performance.
This is a missed opportunity. The CFTC’s event contract proposal could have required platforms to publish systematic accuracy data. Instead, the debate devolved into familiar turf wars between regulators. Democratic senators picked fights with the CFTC over enforcement priorities. State gaming commissions asserted authority. Nobody asked for receipts.
A market designer would approach this differently. What are we trying to accomplish? What design choices would get us there? How would we measure success? These questions remain largely unasked in the regulatory conversation.
The Uncomfortable Historical Parallel
Prediction markets existed before. The Iowa Electronic Markets have run since 1988. Various offshore operations have taken bets on elections for decades. Academic experiments date to the 1990s. Yet the field’s track record remains genuinely uncertain.
The Iowa markets produced some impressive calls and some notable misses. The 2016 election markets showed Clinton advantages that evaporated on election night. The 2020 markets struggled with the uncertainty of mail-in ballot counting. Each cycle, advocates point to moments of prescience. Critics point to moments of failure. Nobody has produced a rigorous, long-term comparison that settles the question.
This is actually strange. If prediction markets represented a genuine forecasting improvement, proving it should be straightforward: compile prices, compile outcomes, measure calibration. The fact that this definitive analysis doesn’t exist — or exists in forms that advocates don’t cite prominently — suggests the evidence is more ambiguous than the marketing implies.
What Roth’s Uncertainty Means for Investors
The prediction market industry has attracted serious capital based partly on claims about information aggregation. Venture money has flowed to Kalshi and its competitors. Traditional finance players have circled. The thesis: prediction markets are a new financial infrastructure, and early movers will capture enormous value.
That thesis may prove correct. But it rests on two distinct claims: that prediction markets work, and that working markets can be monetized at scale. Roth’s comments complicate the first claim without addressing the second.
If prediction markets turn out to be modestly better than polls sometimes, in some conditions, for some questions — well, that’s not nothing. But it’s not the revolution either. It’s an incremental improvement that may or may not justify the regulatory complexity, the operational costs, and the reputational risks these platforms navigate.
The honest investor asks: what is the actual evidence? And the honest answer, from someone who has spent a career designing markets: uncertainty.
The Path Forward
None of this means prediction markets are worthless. They clearly produce information. They clearly engage participants in ways that polls cannot. They clearly create incentives for research and analysis that might not otherwise occur. Polymarket’s latest markets generate real trading activity from people with genuine views about future events.
But the industry would benefit from the kind of intellectual humility Roth displayed. The claims should match the evidence. The marketing should acknowledge limitations. The regulatory arguments should address performance, not just categorization.
A Nobel laureate saying “I don’t know” is not an indictment. It’s an invitation to find out. The prediction market industry should accept that invitation rather than continuing to operate as if the answer were obvious.