The framing around institutional prediction markets assumes a gap that needs bridging: the markets exist, institutions are circling, and what stands between them is a stretch of missing infrastructure – clearing, margin, prime brokerage and surveillance. Build the scaffolding, the assumption runs, and the institutions follow. A panel at A-Team Group’s recent TradingTech Summit New York, convened to set out a blueprint for exactly that, located the gap elsewhere. Institutions are already in prediction markets – but as consumers of the data, not participants in the trading. That distinction arguably matters more than the blueprint does.
The institutional product today is the signal. Prediction market pricing on a geopolitical outcome, a macro release, a policy decision is, as panellists described it, very high signal-to-noise information – and it can be organised, given identifiers and formatted in a way that is familiar to institutional participants and pushed through pipes those firms already connect to. A growing number of institutions are doing this now, using the data to inform risk management, to help instruct a position or hedge an exposure in more traditional markets. None of that requires a firm to trade a prediction market contract. It requires only that the market produces a price worth reading.What has not arrived is participation. The reason is unglamorous and was stated plainly on the panel: the markets are too small. It would be very hard, one line of argument held, for an asset manager to actually express a thesis in a prediction market at current notional scale. The activity that does exist is still weighted heavily toward sports and cultural events – fine for retail, but not the macro and geopolitical contracts that would carry institutional size. So the consuming relationship runs well ahead of the trading one, and everything the blueprint concerns itself with – the clearing arrangements, the margin models, the surveillance frameworks – is scaffolding for a participation phase that follows the data, rather than arriving alongside it.
The signal leads, the venue lags
That ordering inverts the usual assumption about how a market institutionalises. The conventional path has the venue mature first – the infrastructure gets built, the liquidity deepens, and the data becomes valuable as a by-product of a market that institutions are already trading. Prediction markets are running the sequence backwards. The data has become institutionally valuable while the venue is still, by the panel’s own account, nascent.
This is partly a question of what the signal is for. The value, as one panellist put it, is informational rather than implementational – getting a read on a geography, a policy, some other driver of risk or return, at a signal-to-noise ratio that is hard to source elsewhere. That read is useful to an institution regardless of whether it places a prediction market trade, and it is useful precisely because the people setting the prices have money at stake. The same feature that makes the data worth reading – real capital expressing real conviction – is what a forecasting model, however capable, cannot fully replicate, and the panel was sceptical that better models reduce the value of the markets rather than augment the work of reading them.There is a limit to how far the informational phase runs on its own, and it is the same limit that keeps institutions out of the trading. A market read for signal does not need to be deep; a market traded at institutional size does. Until that depth exists, the data relationship is the whole of the institutional relationship.
Fragmentation before consolidation
If there is a single structural reason the participation phase has not arrived, it is the shape of the market that would host it. The current direction of travel is verticalisation – exchanges operating their own clearing houses, each vertical a self-contained stack. The panel was consistent that this fragments liquidity, and consistent too about why that matters: fragmented liquidity limits the venues at which institutions are willing to participate, because an institution will not spread size across a dozen thinly capitalised books with a dozen different sets of resolution criteria.
The template the discussion kept reaching for was the options market in its multi-exchange era – fifteen, sixteen, seventeen venues, but one centralised clearing source underneath them. The contrast is the point. Prediction markets are fragmenting at the clearing layer, not just the execution layer, and that is the harder fragmentation to live with. The expected resolution is consolidation toward a smaller number of larger venues, the way the exchange-traded derivatives world settled into an equilibrium dominated by a handful of established exchanges. One reading held that this is simply where the capital efficiency lies: a consolidated structure offers margin offset and netting that a fragmented one cannot, and the institutions that need those efficiencies will wait for them.
The practical consequence is what this does to entry timing. If the structure has to settle before institutions arrive in size, then the order of events is fixed: consolidation first, participation second. Panellists were unwilling to put a date on it, and one declined the invitation to forecast the timing at all, observing that you could even run a prediction market on the question! But the sequence was not really in dispute. A market this divided does not attract institutional flow until it stops dividing, and the lowered barriers to launching new venues – cloud infrastructure has removed the warehouse of servers that a new exchange once required – mean the fragmenting pressure is, for now, still pushing the other way. The barrier to entry has fallen for venues at precisely the moment the market is likely to need fewer of them.
Surveillance and the cross-market problem
The surveillance discussion identified another challenge. The exposure that matters is not someone trading an event contract on misappropriated non-public information, serious as that is. It is cross-product, cross-market abuse: a position taken on an event contract for a macro outcome, paired with a simultaneous trade in the equities market that the same outcome moves. The two legs sit in two different systems. Very few vendors capture prediction markets in the same place as conventional markets, and most legacy surveillance stacks treat them as entirely disparate – which means the abuse that spans both is, structurally, the abuse no one is positioned to see.
The detection method has to change as well as the coverage. Standard mark-to-market anomaly detection, one panellist argued, is not the right approach when contracts move within milliseconds; the speed defeats a model built to flag price moves after the fact. The direction indicated was information-flow monitoring – a near real-time capture and understanding of how information moves, rather than a retrospective read of where prices ended up. Underneath that is a regulatory question the panel returned to more than once: existing market-manipulation rules were written for established markets, and the work now is to apply that definition to event-based instruments rather than to invent a new framework for them. The procedures are well established; what they need is adaptation, not reinvention.
Where the panel did not agree was on who holds the first line of defence. One position held that the exchange is the first line – the venue closest to the contract, with the clearest view of how it resolves, carrying the initial surveillance obligation. Another argued the opposite: that the first line belongs with the broker, on the futures and FX model, where know-your-customer and anti-money-laundering controls and an understanding of what the customer is actually doing sit closest to the flow, with the exchange as a second line and the regulator and the technology layered behind. Both positions are defensible and they point at different architectures. The question of where surveillance primarily lives is, at this stage, still open, and the answer will shape how the cost and the obligation are distributed across the market.
Building ahead of the rules
Running underneath all three threads was the view that the firms building controls before they are required to are the ones that win. The compliance frameworks, the data-integration controls, the risk assessments – the panel described organisations putting these in place ahead of any enforcement directing them to, on the expectation that the venues applying the most stringent standards now will be the ones left standing, the way the most compliant venues, rather than the most permissive, emerged as the survivors in crypto. The reference point that kept surfacing was that compliance, not engineering, has become the constraint worth solving for first.
That expectation is the one the whole structure rests on. The data is already in institutional hands and the relationship there is real. Everything else – the consolidated clearing that has not happened, the surveillance architecture still being argued over, the controls built ahead of the rules – is positioned for a participation phase whose timing no one on the panel would name. Prediction markets, on the panel’s own reckoning, may be close to unrecognisable inside a year. Whether that reshaping is the one the blueprint anticipates, or whether the institutional relationship stays where it already works – in the data, not the trade – is the question the next several quarters will answer.
Subscribe to our newsletter
