The State of Prediction Markets

The prediction market sector in 60 seconds

What was dismissed as a passing trend around the 2024 election cycle has ballooned into a class of derivatives that does billions of dollars of volume every week.

Polymarket still has the edge in volume, depth, and breadth - but Kalshi is closing the gap in the overall standings. Overall, the sector is thriving. In the beginning of 2024, you could sneeze and slip the price of every market by 10 cents, but by the fourth quarter of 2025, Polymarket and Kalshi had thousands of markets that could absorb a $10K trade within 1¢ of price impact.

This diversification can be visualized through the Herfindahl-Hirschman Index (HHI), which can tell us when an ecosystem becomes overreliant on too few market leaders.

We see that, for all of the worry, the end of the 2024 election season represented only a momentary high point in the concentration of prediction market activity.

Prediction market sector deep dive

The important thing to re-learn is that these are not all one product. Prediction markets are simultaneously:

• A trading venue, with execution, spreads, and risk transfer,
• A media/data product, that people cite and lean on for headlines,
• A hedging platform, where traders can de-risk unique tail risks,
• And a decision-support tool or a way to derive signal via market incentives

This category has been so hard to analyze because people argue “gambling vs forecasting,” but that’s the wrong axis. The right question to ask is which user intent you’re serving, and how that maps onto distribution, liquidity, and settlement.

User Archetypes

Speaking of users, prediction markets have to be specific about who their target market is and, more importantly, who it is not. People come to use these platforms for many reasons, but here are some of the usual participants:

Information traders: These users attempt to locate market mispricing and capitalize. They need tight spreads, reliable settlement, and prefer low fees. You could throw “information-advantaged” (both legitimate and not) traders in here too, which we can explore later.

Narrative commentators: These folks express and contest narratives in real time and generally focus on crypto & politics. They value a wide breadth of available markets to be hyper-specific about their views and don’t necessarily need super deep liquidity.

Gamblers: Value a comfortable balance of market depth and breadth, with an overweighting towards sports markets and a propensity for combo bets/parlays and social attention loops.

Hedgers: Generally prefer to trade markets with clear financial/economic exposure (macro releases, policy outcomes, rate decisions) and value deep liquidity and very clear resolution criteria.

Bots/agents: Kind of a wildcard to think through, but agents are OK with placing tiny trade amounts and racking up miniscule gains, but they need clean tool access and prefer low fees.

Once the platform has located its target audience, it can begin optimizing for what they truly value in a prediction market.

Rails vs Wrappers and how to map products to users

From the platform perspective, we find it helpful to shape a mental model around two layers of the stack:

Rails = where liquidity, execution, and settlement live.

Wrappers = where distribution, UX loops, and packaging live.

The team that owns the distribution does not necessarily have to own the base-layer liquidity. There can be stack competition between: who owns the rails, who owns the wrapper, and who can think up clever combinations between the two that serve a target audience.

In practice, we see this manifest in several distinct “product archetypes.” Here’s how it breaks down.

Narrative markets such as Polymarket (could also add in Opinion, Limitless, XO Markets, etc.) benefit from the ability to have high specificity in their markets. They are able to take the topics that everyone is chatting about around the watercooler (or, let’s face it, on X) and turn them into real-deal betting markets. Their edge is their ability to capture attention and headlines, the speed-to-list new markets, and their de facto position as offering “odds” as a form of media.

The obvious risk of a fast-and-loose market generation philosophy is that resolution disputes are inevitable. When money is on the line and the rules are even a little vague, there will be disagreements. And even if they are trending down against adoption (now just 4 disputes per 10,000 markets), the edge cases make the news and frustrate users.

Cadence Markets leverage the dopamine spikes caused by consistent feedback loops involved in high-cadence markets, with sports betting being the number one vertical. These businesses can be extremely profitable, and they’re able to charge higher fees as sportsbetters will generally accept a higher percentage fee compared to, say, an investor or arbitrageur. Venues like Kalshi and Robinhood have begun implementing parlays/combo bets, which ratchet up the profitability of the platforms.

Robinhood and Coinbase are actually very interesting case studies of the Rail vs. Wrapper distinction, with both exchanges using Kalshi to solve their cold-start problem during the initial spin up phase. Robinhood has since added their own markets (capturing the Rail business) and Coinbase plans on integrating “contracts from additional prediction market platforms in the coming months” (establishing themselves as the Wrapper of all Wrappers).

Spinning up these combo bets can be difficult from a practical and a regulatory point of view. Practically, establishing a RFQ (Request for Quote, where these multi-leg custom bets are offered to many market makers to see who can quote the best price) is more technically difficult than a simple orderbook and tends to fracture liquidity into many small and bespoke buckets instead of keeping dollars in a single book. On top of that, sports betting is already top-of-mind in the eye of regulators, especially at the state level. And by slapping an RFQ system in the mix, this might start to look (to some) more like a dealer-packaged derivative than a simple CLOB matching. (Late breaking: CFTC rules around these may be easing!)

Hedging Markets are what we perceive to be around the corner. Institutional giants Cboe and Nasdaq are joining in on the party, looking to roll out prediction market options in the near future. We see the focus of this subset to surround real financial exposure (macro releases, policy outcomes, and rate decisions to name a few) so sophisticated investors can specify exact risks they would like to hedge out. Note that CME has been early, but they’re off partying with DraftKings and FanDuel, with a larger focus on retail/sportsbetting at the moment.

Hedging markets will require dead-simple resolution criteria and standardized market creation. These guys aren’t gonna mess around with Zelensky’s suit, they’ll have markets like “Will Crude Oil futures be above or below $90.00 at 4PM ET on March 6th, 2026?”. These are for risk transfer, not for entertainment. Not very fun, but useful to the poor sap trading airline stocks.

Polymarket vs Kalshi

Speaking of the two titans, let’s address the elephant in the room. But I want to be clear up front, this isn’t about “who wins”; both can be right at the same time because they’re optimizing different parts of the stack and different product archetypes.

It all comes down to: where does price discovery live, and what kind of user-product bundles does each rail naturally produce?

The pair’s differences started at the very beginning, with Polymarket setting up an onchain architecture on Polygon and pursuing a later-stage compliance path, while Kalshi built entirely offchain and went down the regulation path before selectively exploring onchain components.

This resembles the perps “engine vs settlement” split: where is matching performed, where is settlement finalized, and how do you anchor your trust? With perp DEXs, this winning architecture seems to be either - a) an entirely off-chain orderbook + onchain settlement or verification OR b) a high-performant (and mostly centralized) onchain orderbook. It remains to be seen whether Kalshi or Polymarket will arrive at a hybrid conclusion such as this, but I’d argue that their final form is yet to be revealed.

Polymarket is optimizing for long-tail execution scaling with permissionless market generation to cater to whatever their users want to bet on. This hyperspecificity uniquely positions them as the “place to trade narratives” and basically becomes the likelihood extension of the topic du jour, turning attention into a public probability surface.

This positions them as a “probability” brand, and a plausible path is that they continue to parlay this positioning into data partnerships with news organizations and trading venues.

Their vibe-check-as-a-service architecture means they no longer solely have to rely on trading fees to meet their bottom line. In other words, they become a “truth API” and organizations pay to get a direct feed.

At the beginning of 2025, Kalshi made the decision to push heavily into sports. Part of this is likely just product physics: after the 2024 election cycle, the category’s “always-on” attention moved away from politics, and sports is the highest-frequency substrate available - there’s always another match, another slate, another moment to trade.

This pivot is also the cleanest expression of what we call Cadence Markets: a subset of prediction markets optimized for high feedback loops and habitual usage. In Cadence Markets, the product goal isn’t “the world’s most philosophically pure probability estimate” - it’s getting users to return, wager, and repeat. That changes everything about what you build and what you can monetize.

Combos (parlay-like packaging) are the retention superweapon here. They let users express richer views, feel like “builders” rather than simple bettors, and keep them engaged across a calendar of events instead of a handful of headline markets.

But combos also force a structural reality: you need someone willing to quote the bundle. In practice, that pushes you toward RFQ-style mechanics and tighter risk management, because you’re no longer just matching two-sided flow in a single contract - you’re packaging legs into a single exposure that must be priced and warehoused, at least temporarily.

The distribution strategy fits the archetype. Retail sports flow is far more UX-sensitive than fee-sensitive - most users only care about frictionless onboarding, fast feedback, and the ability to close positions easily. That’s precisely why Kalshi’s partnerships with brokers/wrappers (like Coinbase and Robinhood) matter: a large distribution surface concentrates flow, concentrated flow tightens execution, tighter execution improves retention, and better retention attracts more liquidity provision. That flywheel is hard to replicate if you’re not plugged into where users already trade.

The takeaway is that we’re witnessing specialization - we’re watching rails and wrappers recombine into different product archetypes. Kalshi’s cadence-first path is one extreme; Polymarket’s narrative-first path is the other. The next section goes deeper: we have the current frontrunning rails, but where does the rest of the ecosystem stand?

Where the money is actually being made

With both leaders’ private valuations pushing into the 11 figures, they have to show a path to profit at some point, so it’s no surprise that Polymarket recently joined Kalshi in charging fees on most markets. But the comparison is not quite apples-to-apples, with each side using their fee structure to signal which users they truly optimize for.

A few things jump out. Polymarket undercuts Kalshi in every single fee-bearing category at every price point up to 90¢. Even Crypto, Polymarket’s most expensive category at a 1.80% peak effective rate, is roughly half of Kalshi’s 3.50% at 50/50 odds. Kalshi only becomes competitive at extreme near-certainties (90¢+), where its effective rate drops below 0.7% - precisely the price range where institutional hedgers operate. That’s not an accident.

Meanwhile, Kalshi’s halved S&P/Nasdaq rate (0.035 vs 0.07) is a targeted concession to win index-derivative flow from TradFi. But even at half rate, it still costs more than Polymarket’s Economics category at every price except 99¢. The message: Kalshi is betting that compliance credibility and settlement enforceability justify the premium, while Polymarket is betting that lower fees plus permissionless market creation will win on volume.

And then there’s Polymarket’s decision to make geopolitical markets completely free - it signals that the real monetization isn’t in trading fees at all, but in the data surface those markets create.

Remember to think in terms of the stack - execution fees are not the only place to charge rent. Often, the largest rents sit at:

Rails - Where execution and settlement live

Wrappers - Where distribution and monetization live

While rails capture the trading fees, the wrappers have freedom to experiment with financialization. They could charge:

• Attention rent by owning the default location where probabilities are consumed,
• Flow rent by controlling the funnel/routing choices to different rails,
• Cross-sell rent by pushing add-ons to users while leveraging the high-engagement prediction markets,
• Or embed their own custom monetization through strategies like spreads on funding rails, subscriptions, premium tools, or even just plain old ads.

As wrappers (like brokers and media) distribute across many rails, they will demand standardized contract specifications with comparable markets, so rails that lead this charge could be rewarded with quick integrations.

An entire ecosystem of applications, powered by this novel primitive, and it goes far beyond the minimizing “Polymarket vs. Kalshi” debate - we’re talking about dozens of killer businesses at the periphery that are about to be enabled. But most of these businesses will only work where liquidity is real and contract specs are legible.

And who will be paying for these services?

Again, it breaks down by the type of market we’re talking about and what layer of the stack is in focus:

• Cadence Markets: rail fees are funded by retail entertainment flow; wrapper economics are funded by retention loops and cross-sell.
• Narrative Markets: rail fees are funded by attention-driven trading; wrapper economics can be funded by probability-as-data distribution (publisher/institutional embeds) plus power-user tooling.
• Hedging Markets: rail fees are funded by hedgers and informed traders; wrapper economics are funded by workflow embedding (brokers/terminals) and trust/permissioning.

Of course, this is a simplification. After all, Polymarket has sports markets and Kalshi has politics and crypto, but we see these as examples of cross-selling to users, not the platforms’ main courses. You go to Kalshi to bet on the Super Bowl and realize they have midterm markets too, how convenient. It’s the same as 7-11 enticing you with BigGulps and Funyuns when you go in to get your daily scratchers, it’s just good business.

The stack explains where value can accrue. User behavior determines whether it does. If Cadence Markets are powered by high-frequency retail flow, Narrative Markets by attention + positioning, and Hedging Markets by informed hedgers, then the next question is: do the observed flows actually match those stories? Let’s look at who trades, how they trade, and what that implies about where true price discovery lives.

User behavior and onchain patterns

For prediction markets especially, who trades and how they trade matters more than in most asset classes because each individual market is really a one-off micro-asset. Trading price of the markets only becomes signal when the right cohort is in the trenches.

We’re gonna track four distinct behaviors across the sector and use it to answer some key questions:

• Participation breadth & retention: is this a one-off spike or are habits forming?
• Concentration: can we assume wisdom-of-the-crowds or is everything whale dominated?
• Trading posture: are users mostly opening simple positions or actively managing risk?
• Liquidity assumptions: are market makers helping users easily enter/exit?

The answers to these questions will clue us into the health of the sector as well as how well the leaders are optimizing for their target audience.

Participation breadth and retention (what adoption looks like)

Top-line numbers are looking great, with Polymarket breaking 300K unique returning users per week and consistently adding 30-50K new users to go along with it.

This broadening participation and improving retention tells us that prediction markets as a sector have graduated from an event-driven curiosity to a mainstay. People aren’t just checking into the markets when an election is in full swing, they are using it to check the pulse of the world (and to participate in the commotion).

Concentration and whale dominance (is “wisdom of the crowds” possible?)

For a prediction market to, well, predict, it requires a critical mass. This is real science by the way, studies have shown (Standford/UCSD study, PDF warning) that “aggregate opinions of groups are often more accurate than those of the constituent individuals.” Which is why Polymarket absolutely smoked pollsters in the 2024 election cycle.

Polymarket in particular has faced heavy scrutiny for alleged insider trading taking place on its platform and has recently partnered with the likes of Palantir and TWG to shore up its defences. Regardless, we’d like to see that it isn’t simply a few big whales making wild swings on most markets, but a wide cross-section of users from all over the world.

On Polymarket in particular, we see “whale dominance” declining, with the top 100 traders by volume going from >95% of platform volume in 2024 to 61% today. We see this as a maturity signal, reducing manipulation/narrative risk and making Polymarket’s probability-as-media roadmap much more credible.

What trading posture tells us about who uses these markets

Casual users and traders display different patterns of behavior - and we can see it in the data.

For Kalshi, we see a clear preference for traders to take the “YES” side over “NO,” especially in 2025, by a 2.49x margin.

When we remember that the majority of Kalshi volume has been in sports markets, this begins to make sense. On most of these markets, buying “Toronto YES” (as below) is economically similar to buying “Phoenix NO,” so the YES/NO split is better read as: how users choose to express the view (with the affirmative framing and UI defaults), not a directional asymmetry in economics.

Put another way, people prefer to click the positive proposition, betting on their team to win rather than on their opponent to lose.

This suggests that we’re witnessing a lot of retail activity, since you’d expect a more balanced weighting from a price-sensitive buyer (assuming better prices are distributed evenly between the YES and NO sides for a large enough pool of markets).

As far as price takers on the Polymarket side, we see that SELLs dominate the BUYs by a wide margin. This supports a “more trader-like posture” story, suggesting that a meaningful share of takers are hedging/liquidating, not only opening fresh directional bets.

This all tracks with our assumption that Cadence Markets (like Kalshi) cater to retail flow (who you could classify more on the “gambler” side) and Narrative Markets (like Polymarket) are more suited for trader activity (with participation from our Information traders, Narrative commentators, and some Hedgers).

Near-resolution dynamics and the market-maker question

Unlike traditional options markets, market makers face off in a game of chicken when it comes to holding their positions close to expiration. The binary payoffs (0 vs. 1) make inventory risk a serious concern for market makers, with small information shocks or late news meaning a complete wipe out of inventory PnL. The knee-jerk reaction to this would be to: quote smaller, widen spreads, or reduce presence altogether as the expiration draws nearer.

We observe this phenomenon in the data, with spreads and price impact actually increasing as resolution draws nearer, indicating that liquidity thins out when the end draws near.

That is our interpretation of the data, anyway. It could be argued that other forces are at play, such as a tendency for traders to size heavier into markets that are close to expiry or for traders to accept higher slippage to jam their bids in at the late stages. This would go along with the idea that platform-aligned market makers have obligations to support markets to completion. We leave it up to the reader to come to their own conclusions.

It’s important to understand the user posture because that tees up which wrappers will win (brokers vs terminals vs media), and near-resolution liquidity dynamics shape which product archetypes can scale (Cadence vs Narrative vs Hedging).

So the debate isn’t “gambling vs forecasting.” It’s whether these products can reliably convert attention into tradeable markets without breaking at the moments that matter most - near resolution, during shocks, and in thin long-tail contracts. Every major innovation we’re seeing is aimed at one of those failure points: packaging (combos), liquidity provision, distribution surfaces, and new contract formats that broaden what can be traded. Let’s look at what’s actually new.

Key innovations (what’s actually new)

This is not going to be a one-stop-shop for every new app in the prediction market space. Instead, we’re going to use this section to look for leverage points - which changes are actually capable of moving liquidity, trust, retention, and/or distribution.

The clean way to read this wave of innovation is as constraint-hunting: teams are trying to unlock liquidity formation (often via distribution and market-making mechanics), harden trust under stress (settlement integrity and reputational resilience), or expand what users can express with one trade.

A few innovations (especially probability-as-data) are less about changing execution and more about packaging the market’s output for distribution, but they still tie back to the same constraints by shaping where attention and legitimacy accrue. Once you map these moves onto rails and wrappers, the category starts looking like a stack that stratifies by cohort and by distribution surface. We’ll close by stating a base case and a bull case, and where we could see it all breaking down.

Our short-term outlook on prediction markets

No “state of” report would be complete without a healthy dose of peering into our crystal ball. Here’s the no-nonsense base, bull, and bear checklist items we’re watching for.

We’re past the point of wondering if prediction markets will meaningfully penetrate the global zeitgeist. Whether you consume the majority of your content through Twitter, YouTube, or mainstream media, this infrastructure has found its way into the conversation.

The question now shifts from “does this have legs?” into “who will capture meaningful market share?”

Our argument has been that the key to sussing out the victors will come down to:

• Who can win the rails for each market archetype and
• Who can wrap their product to best hunt the constraints of: liquidity formation, trust, or expressiveness

The end-state will not be a single winner, more likely it will look like a stratified stack across Narrative Markets, Cadence Markets, and Hedging Markets, each bundled differently across rails and wrappers.

In closing

The data suggests the category has crossed the threshold into infrastructure - but it’s still asymmetrical: execution quality and market breadth don’t scale evenly across rails.

That’s why this is a stack war: rails compete on settlement credibility and depth, while wrappers compete on distribution, retention, and packaging. Narrative Markets, Cadence Markets, and Hedging Markets are simply different equilibrium bundles of those layers, optimized for different user intents. The durable advantage goes to whoever can standardize enough to scale distribution without losing the permissionless edge that keeps markets culturally relevant.

Where does each player stand today? Here’s how we see it: