The report cites four datasets: federal funds rate, CPI, election accuracy, error rate. All four support its thesis. Not one contradicts it.
That is the first thing I check in any research document. It is also the first thing Tiger Research's closing installment on prediction markets gets wrong. A study that surfaces only confirming evidence has not tested a hypothesis. It has assembled a case.
State the evidence base plainly. Kalshi, a CFTC-registered designated contract market. The Iowa Electronic Markets, an academic project running since 1988. Neither issues a token. Neither settles on-chain. In the technical and tokenomics sections, the author writes "N/A" — repeatedly, and correctly.
That candor deserves credit. It also creates a problem the report never resolves. Its strongest evidence comes from markets that structurally cannot be replicated by the token-incentivized platforms its Web3 readers will actually use.
The argument itself is old. Real money at stake, prices update on new information, the resulting curve is a probability distribution. IEM did this with paper and electronic matching thirty-eight years ago. What is new is not the mechanism. What is new is the packaging: probability as a subscription data product, sold to institutions that will never place a trade.
That is the real claim to audit. Not whether prediction markets are accurate. Whether they have become infrastructure. The distinction matters because infrastructure earns rent, and rent attracts imitators who mistake the packaging for the product.
The commercial signal is unambiguous. A Federal Reserve Board working paper describes Kalshi as a high-frequency, continuously updating benchmark. ICE and Tradeweb now distribute the platform's data to institutional clients. Google Finance, CNBC, CNN and the Wall Street Journal display the quotes. The venue has become a data supplier, and the interface has been standardized to FIX-grade expectations.
That last point is the hardest engineering milestone in the entire report, and it occupies one sentence.
The settlement layer is missing, and its absence is the tell.
Prediction markets have exactly one technical bottleneck that matters: resolution. Who decides whether an event occurred, and what happens when two parties disagree. This is where every on-chain prediction market has bled. Optimistic oracle disputes, ambiguous resolution criteria, governance votes that retroactively redefine contract terms. The report covers none of it.
I spent two weeks in 2020 reconstructing Compound's borrow-rate logic in Python after noticing a rounding discrepancy in the assembly. The bug was small. The lesson was not: the exploitable surface is never where the whitepaper points. In prediction markets, that surface is adjudication. A centralized DCM handles it with internal rules plus regulatory exposure. A chain handles it with an oracle and a dispute window. The two are not comparable, and presenting one as validation of the other is a category error.
The token question invalidates the extrapolation.
Here is the part Web3 readers should read twice. Every cited data source — Kalshi, IEM — operates without liquidity mining, points, or airdrops. The report's conclusion, that prediction market prices outperform traditional indicators, therefore holds only for markets where participants are informed and capital is at risk.
Introduce a token subsidy and the causal chain inverts. Incentives recruit uninformed volume. Uninformed volume widens spreads and drags quotes away from genuine belief. The probability curve distorts. And the product — which is the curve, not the liquidity — is destroyed.
DeFi treats TVL as the objective. Prediction markets treat liquidity as an input and price accuracy as the output. Buying the input by diluting the output is not growth. It is a bug wearing a growth chart.
| Signal | What it actually measures | Weight | |---|---|---| | TVL / volume | Subsidy persistence | Low | | Spread and depth during zero-incentive weeks | Genuine informed participation | High | | Dispute history on resolved contracts | Settlement integrity | High | | Institutional data redistribution | Acceptance of the price as source of truth | Decisive |
Authority borrowing is doing quiet work.
A Federal Reserve Board working paper is a researcher's output. It is not a policy statement, not an endorsement, and not peer-reviewed. The report's phrasing — the Board "describes" Kalshi as a benchmark — reads as institutional sanction. It is not. Downgrade it to "noticed by policy researchers" and the section loses most of its weight.
In the absence of data, opinion is just noise. In the presence of misattributed authority, it is worse than noise — it is noise with a citation attached.
The regulatory argument is strategy dressed as advice.
The report recommends platforms engage regulators and position themselves as monitoring tools. Read that carefully. It is not a compliance suggestion. It is a reframing operation: convert gambling into public-interest information infrastructure, and use that framing to argue for federal preemption of state gaming law.
It may work. It also tells you where the actual bottleneck sits. When a research series saves regulatory positioning for its final installment, the constraint is not technology and not liquidity. It is legal.
And that moat does not transfer. A CFTC-registered DCM carries full BSA/AML obligations and a federal charter. A permissionless on-chain venue carries four simultaneous exposures — event contracts, potential securities treatment, money transmission, state gaming statutes. The compliance premium is not portable. Chain-native platforms inherit the narrative and none of the cover.
What the bulls got right.
Ignore the accuracy framing and the distribution argument stands on its own. ICE does not do brand favors. Exchange groups attach to data feeds when they expect recurring subscription revenue, and they vet latency, uptime and data dictionaries before they do it. That vetting is a harder engineering bar than any accuracy statistic in the report.
I watched this from the inside in 2025, designing custody risk protocols for an Australian bank. The hard problem was never the ledger. It was reconciling SQL-side audit trails with chain-side finality without breaking either. Infrastructure is boring, and boring is the point. If prediction market data has cleared institutional distribution, it has passed a test most crypto protocols never attempt.
The second thing the bulls got right is citation lock-in. A number quoted by a central bank working paper and a major financial daily becomes a reference point. Reference points resist migration. Feeds that are merely used get replaced. Feeds that are cited do not.
That is the moat. Not the matching engine. Not the user interface. The citation.
One structural opportunity the report ignores entirely: the probability oracle. Traditional oracles transmit settled facts — prices, rates, timestamps. A prediction market transmits a distribution over events that have not happened. That is a different data primitive, and it composes. Option pricing inputs, insurance triggers, DAO risk dashboards. The niche exists. Nobody has validated it.
Watch the institutional feed contracts, not the accuracy charts. Watch dispute resolution disclosures, not volume. And when the next prediction market launches with points and a liquidity program, ask the only question that matters: what does the spread look like in a week with no incentives?
If the answer is unavailable, the product is unavailable. Silence in the ledger is loud.