Are prediction markets arriving in institutional workflows faster than the industry has worked out what it is buying? The valuations are real. The distribution deals are real. The research interest is real. What is not settled is a set of questions sitting underneath the adoption story – and a closing fireside session at the recent A-Team Group/Eagle Alpha Alternative Data Conference in London was useful less for its answers than for how sharply it exposed those questions.
Is the legitimisation story evidence, or momentum?
The case for taking the category seriously is usually made through capital. A recent funding round valued Kalshi at $22bn. ICE has invested in Polymarket. Established marketplaces and feed providers have struck deals to carry event-contract data into institutional channels. Each is a real signal of intent.
But intent is not validation. A funding round prices a business, not the quality of a dataset. So the first question is whether the legitimisation story rests on evidence that the data forecasts well, or on the momentum of capital betting that it will. One claim deserves particular scrutiny: the much-quoted figure that event markets beat traditional macro forecasters by up to 40 per cent on error rates, often pinned to a recent Federal Reserve working paper. The paper itself makes a narrower claim – that prediction markets forecast about as well as established benchmarks, and beat some of them on specific measures – not the headline number. The gap is worth noting, because it asks what it says about a category marketing itself on a statistic its own cited evidence does not carry.Can regulation be a guide when it is always behind?
The second question looks like the obvious obstacle. The category comes wrapped in litigation: insider-trading cases, a US Senate ban on members of Congress trading these contracts, state lawsuits, criminal charges in at least one place. The natural assumption is that regulatory clarity will eventually settle how institutions can use the data.
But the session pointed to a more lasting problem. Regulators move slowly and markets move fast, and the gap is widening, not closing. If that is true, waiting for clarity is not a strategy. The question becomes whether institutions can build their own usage and compliance rules for a category the regulation has not caught up with – and whether “it was allowed” will ever be a sufficient test when the hard issues are ones the rules have not yet reached.
What are you actually getting a signal from?
This is where the surface questions give way to the one that really matters.
Prediction markets have a habit of moving ahead of major events – prices shifting before geopolitical developments, repositioning before announcements. The instinct is to treat that as a red flag. The more uncomfortable possibility, raised in the session and rarely said plainly elsewhere, is that it is not a flaw in the signal but the source of it. If much of the edge comes from people who know something the public does not, then the forecasting power and the compliance risk are not two separate things. They are the same thing.
That collapses the comfortable gap between “is it useful?” and “is it clean?” into one question: if the edge and the contamination cannot be separated, what is actually being bought – and can a regulated institution buy it at all? The art-market comparison made in the session is the right one. Value depends on provenance: who held the asset, how it changed hands. The open question is whether prediction market data can ever carry the provenance trail institutional use demands, or whether inference risk – the chance that a signal contains material non-public information however it is packaged – is simply built into the category. Nobody at the session claimed to have solved this. That is the point.
Even if provenance is sorted, can the signal be used?
Suppose the provenance problem could be managed. A separate set of questions would still stand, and they are practical rather than philosophical.
How does a team run vendor diligence in a market consolidating through M&A faster than the diligence can finish – where a preferred provider can be bought and its data pulled before the evaluation is even presented? How does a macro portfolio manager’s standard demand for 20 years of backtest history get met for contracts tied to single events, such as the outcome of one geopolitical crisis, where the history is a sample of one? How do you map an idiosyncratic event contract onto an equity index and get a repeatable, tradeable signal out of it rather than a one-off?
These questions decide whether the category is usable at all, and none has a settled answer. The barrier to adoption is not whether the signal exists. It is whether the apparatus around it – sourcing, validation, contracting, entitlement – can be made repeatable for a dataset that resists all four.
And what if the expensive data is not the edge?
A final question runs under the others and unsettles the premise. The category is sold as a premium dataset. But is alpha really a function of what data costs? The session offered the counter-case: some of the most valuable alternative data is free and public – regulatory insider-transaction filings, shipping-container volumes that once gave months of warning on semiconductor demand. If interpretation rather than acquisition is where the edge usually sits, the question is not only what prediction market data costs, but whether the same effort spent sourcing it would do more applied to data already in plain sight.
The question under the questions
Distribution, it turns out, is the easy part. Getting prediction market data onto a desk is close to solved. Knowing what it is, where it came from, and whether it can be trusted is the work that remains – and it is work the buyer, not the seller, will have to do.
Subscribe to our newsletter
