Why prediction markets are still in the exploratory stage

Original Title: Why Prediction Markets Are Still in Beta
Original Author: Nick Ruzicka
Original Translation: SpecialistXBT, BlockBeats

Prediction markets are having their moment in the spotlight. Polymarket’s coverage of the presidential election has made headlines, Kalshi’s regulatory victory has opened up new territory, and suddenly, everyone wants to talk about this “world truth machine.” But behind the hype lies a more interesting question: If prediction markets are so good at forecasting the future, why haven’t they gone mainstream?

The answer isn’t sexy. The problem lies in infrastructure—in the US, it’s regulation. (For example, Kalshi obtained approval from the US Commodity Futures Trading Commission (CFTC), while Polymarket operates offshore), but infrastructure issues remain widespread. Even in regions where prediction markets are legal, the same fundamental challenges persist.

The dominant platforms in 2024 are solving these problems by throwing money at them. According to Delphi Digital researcher Neel Daftary, Polymarket has spent about $10 million incentivizing market makers, at one point paying over $50,000 per day to maintain liquidity on its order books. Today, those incentives have collapsed to just $0.025 per $100 traded. Kalshi has spent over $9 million on similar projects. None of these are sustainable solutions—they’re just band-aids on structural wounds.

Interestingly, the challenges holding back prediction markets aren’t mysterious. They’re well-defined, interrelated, and—for the right entrepreneurs—relatively easy to solve. After talking to teams in the space and analyzing the current landscape, we found five issues that come up repeatedly. Think of them as a framework, a shared vocabulary to help us understand why prediction markets, despite their theoretical promise, are still in beta.

These are not just problems—they are a roadmap.

Problem One: The Liquidity Paradox

The most fundamental issue is liquidity. Or more precisely, the chicken-and-egg problem that turns most prediction markets into ghost towns.

The mechanism is simple. New markets launch with low liquidity. Traders face poor execution—high slippage and price impact make trading unprofitable. They leave. Low volume scares off professional liquidity providers, who need steady fees to offset risk. Without liquidity providers, liquidity remains scarce. The cycle repeats.

The data backs this up. On Polymarket and Kalshi, most markets have less than $10,000 in trading volume. Even the larger markets lack enough depth to attract institutional investors for meaningful participation. Any large position causes double-digit price swings.

The root cause is structural. In typical crypto liquidity pools (like ETH/USDC), you deposit two assets and earn fees as traders swap—value is preserved on both sides, even if the price moves against you. Prediction markets are different: you hold contracts that become worthless if they lose. There’s no rebalancing, no residual value—there are only two outcomes: half the assets go to zero.

Worse, you get “picked off.” As settlement approaches and outcomes become clearer, informed traders know more than you do. They buy the winning side from you at favorable prices while you’re still pricing based on outdated probabilities. This “toxic order flow” continually bleeds market makers.

In 2024, Polymarket switched from an automated market maker (AMM) model to a central limit order book for this reason: order books let market makers pull quotes instantly when they sense they’re about to get picked off. But this doesn’t solve the root problem—it just gives market makers some defensive tools to slow their losses.

These platforms sidestep the issue by directly paying market makers. But subsidies can’t scale. For flagship markets—presidential elections, major sports events, hot crypto topics—this works. Polymarket’s election markets are liquid. Kalshi’s NFL markets compete with traditional sportsbooks. The real challenge is everything else: the vast number of markets where prediction markets could matter, but where volume can’t justify millions in subsidies.

The current economic model is unsustainable. Market makers don’t profit from spreads, but from platform payouts. Even protected liquidity providers (who can only lose 4-5% per market) need ecosystem grants to break even. The question is: How do you make liquidity provision profitable without burning cash?

Kalshi’s successful model is starting to emerge. In April 2024, they brought in Wall Street giant Susquehanna International Group as their first institutional market maker. The result: 30x more liquidity, contract depth up to 100,000, and spreads under 3 cents. But this requires resources retail market makers can’t provide: proprietary trading platforms, custom infrastructure, and institutional capital. The breakthrough isn’t higher rebates—it’s getting the first institution to treat prediction markets as a legitimate asset class. Once one joins, others follow: lower risk, benchmark pricing, and volume grows naturally.

But here’s the catch: institutional market makers need specific conditions. For Kalshi, that means CFTC approval and clear regulatory rules. For crypto-native and decentralized platforms—those without regulatory moats or massive scale—this path doesn’t work. These platforms face a different challenge: how to bootstrap liquidity without regulatory legitimacy or guaranteed volume? For everyone except Kalshi and Polymarket, the infrastructure problem remains unsolved.

What Entrepreneurs Are Trying

Quality-weighted maker rebates reward liquidity that improves trading—shorter execution times, larger quotes, tighter spreads. This is pragmatic, but doesn’t solve the root issue: these rebates still need to be funded. Protocol tokens offer an alternative—subsidizing LPs with token issuance rather than VC money, similar to Uniswap and Compound’s launch playbooks. Whether prediction market tokens can accrue enough value to sustain long-term issuance remains unclear.

Tiered cross-market incentives provide diversified liquidity across multiple markets, spreading risk and making participation more durable.

Just-in-time (JIT) liquidity provides funds only when needed. Bots monitor pools for large trades, inject concentrated liquidity, collect fees, and withdraw immediately. This is capital efficient but requires complex infrastructure and doesn’t solve the core problem: someone still eats the risk. JIT strategies on Uniswap V3 have enabled over $750 billion in volume, but activity is dominated by well-capitalized players, with razor-thin returns.

Continuous combinatorial markets challenge the binary structure itself. Traders aren’t limited to discrete “yes/no” options but can express views across a continuous range. This aggregates liquidity that would otherwise be split across related markets (Will Bitcoin hit $60,000? $65,000? $70,000?). Projects like functionSPACE are building this infrastructure, though it hasn’t been tested at scale yet.

The most radical experiments ditch order books entirely. Melee Markets applies Bonding curves to prediction markets—each outcome has its own curve, early participants get better prices, and strong conviction is rewarded. No professional market makers needed. XO Market requires creators to inject liquidity using LS-LMSR AMMs; as funds flow in, market depth increases. Creators earn fees, aligning incentives with market quality.

Both solve the cold start problem without professional market makers. Melee’s downside is inflexibility (positions are locked until settlement). XO Market allows continuous trading but requires creators to front capital.

Problem Two: Market Discovery and User Experience

Even if you solve liquidity, there’s a more practical problem: most people can’t find markets they care about, and even if they do, the experience is clunky.

This isn’t just a “UX problem”—it’s structural. Market discovery issues directly worsen liquidity problems. Polymarket has thousands of markets online at any time, but volume is concentrated in a handful of areas: elections, major sports, and hot crypto topics. Other markets are deserted. Even if a niche market has some depth, if users can’t naturally find it, volume stays low and market makers eventually leave. Vicious cycle: no discovery, no volume, no sustainable liquidity.

Market liquidity is extremely concentrated. In the 2024 election cycle, Polymarket’s top markets captured the vast majority of trading activity. After the election, the platform still had $650–800 million in monthly volume, but it was spread across sports, crypto, and viral markets. The thousands of other markets—local issues, niche communities, oddities—were almost entirely ignored.

User experience barriers make this worse. Polymarket and Kalshi’s interfaces are designed for people who already understand prediction markets. Regular users face a steep learning curve: unfamiliar terminology, converting odds to probabilities, what it means to “buy a YES,” and so on. For crypto-native users, this is tolerable. For everyone else, these frictions kill conversion.

Better algorithms help, but the core issue is distribution: matching thousands of markets to the right users at the right moment, without causing choice paralysis.

What Entrepreneurs Are Trying

The most promising approach is to serve users directly on platforms they already use, rather than making them learn a new one. Flipr lets users trade Polymarket or Kalshi markets by tagging a bot directly in their Twitter feed. For example, when a market is mentioned in a tweet, users can tag @Flipr and trade without leaving the app. It embeds prediction markets into the conversational layer of the internet, turning social feeds into trading interfaces. Flipr also offers up to 10x leverage and is developing copy trading and AI analytics—basically, it’s aiming to be a full-featured trading terminal that just happens to live inside Twitter.

The deeper insight: for startups, distribution matters more than infrastructure. Instead of spending millions to bootstrap liquidity like Polymarket, aggregate existing liquidity and compete on distribution. Platforms like TradeFox, Stand, and Verso Trading are building unified interfaces that aggregate odds from multiple platforms, route orders to the best venue, and integrate real-time news feeds. If you’re a serious trader, why bother switching between platforms when you can use a single, more efficient interface?

The most experimental approach treats market discovery as a social problem, not an algorithmic one. Fireplace, affiliated with Polymarket, emphasizes investing with friends—recreating the energy of betting together, not alone. AllianceDAO’s Poll.fun goes further: it builds P2P markets among small friend groups, letting users create markets on any topic, bet directly with peers, and have outcomes decided by creator or group votes. This model is highly localized, highly social, and by focusing on community rather than scale, sidesteps the long-tail problem entirely.

These are not just UX improvements—they’re distribution strategies. The winning platform may not have the best liquidity or the most markets, but will best answer the question: “How do you get prediction markets in front of the right users at the right time?”

Problem Three: Expressing User Views

This statistic should worry anyone bullish on prediction markets: 85% of Polymarket traders have negative account balances.

To some extent, this is inevitable—prediction is hard. But part of the reason is a hard flaw in the platforms. Because traders can’t effectively express their views, the platforms force them into suboptimal positions. Have a nuanced theory? Too bad. You can only make binary bets: buy, don’t buy, or choose position size. No leverage to amplify conviction, no way to combine multiple views into one position, no conditional outcomes. When traders can’t express conviction efficiently, they either overcommit capital or underbet. Either way, the platform captures less flow.

This problem splits into two distinct needs: traders who want leverage to amplify single bets, and those who want to combine multiple views into one bet.

Leverage: Continuous Settlement Solutions

Traditional leverage strategies don’t work in binary prediction markets. Even if your prediction is correct, market swings can wipe you out before settlement. For example, a leveraged “Trump wins” position could be liquidated during a bad polling week, even if Trump wins in November.

But there’s a better way: continuous-settlement perpetual contracts based on real-time data streams. Seda is building true perpetuals based on Polymarket and Kalshi data, letting positions settle continuously rather than waiting for discrete events. In September 2025, Seda enabled perpetuals (initially at 1x leverage) for real-time odds on the Canelo vs. Crawford fight on testnet, proving the model’s viability for sports betting.

Short-term binary options are another increasingly popular trading style. In September 2025, Limitless surpassed $10 million in trading volume, offering binary options on crypto price movements. These markets provide implicit leverage through their payout structure while avoiding liquidation risk during the contract’s life. Unlike fixed-payout options, binary options settle at fixed times, but their immediacy (hours or days, not weeks) gives retail traders the fast feedback they crave.

The infrastructure is maturing rapidly. In September 2025, Polymarket launched 15-minute crypto price markets with Chainlink. Perp.city and Narrative are experimenting with continuous information flow trading based on poll averages and social sentiment—true perpetuals that never resolve to a binary outcome.

Hyperliquid’s HIP-4 “event perpetuals” are a breakthrough—they trade on changing probabilities, not just final outcomes. For example, if Trump’s win probability rises from 50% to 65% after a debate, you can profit without waiting for election day. This solves the biggest problem with leveraged trading in prediction markets: even if you’re ultimately right, you can get liquidated by volatility. Platforms like Limitless and Seda are gaining traction with similar models, showing that the market wants continuous trading, not binary bets.

Combinatorial Bets: An Unsolved Problem

Combinatorial bets are different. They express complex, multi-faceted hypotheses like: “Trump wins, Bitcoin breaks $100,000, and the Fed cuts rates twice.” Sportsbooks can do this easily because they act as centralized institutions managing distributed risk. Contradictory positions offset each other, so they only need to collateralize the maximum net loss, not every possible payout.

Prediction markets can’t do this. They act as custodial agents—every trade must be fully collateralized once made. Costs quickly balloon: even small combinatorial bets require market makers to lock up orders of magnitude more capital than a sportsbook would for the same risk.

The theoretical solution is net margin systems that only collateralize the maximum net loss. But this requires complex risk engines, real-time correlation modeling across unrelated events, and possibly a centralized counterparty. Researcher Neel Daftary suggests starting with professional market makers underwriting limited market combinations, then scaling up. Kalshi is doing this—initially offering only same-event combinatorial bets, since it’s easier to model correlations and manage risk within a single event. This is insightful, but also acknowledges that true combinatorial markets—“choose your own adventure” style—may be impossible without centralized management.

Most prediction market entrepreneurs see limits to these novel market types: leverage caps on short-term markets, pre-approved event combinations, or simplified “leveraged trading” that the platform can hedge. The user expression problem may be partly solved (e.g., continuous settlement), but other parts (e.g., arbitrary combinatorial markets) remain out of reach for decentralized platforms.

Problem Four: Permissionless Market Creation

Solving the expression problem is one thing, but a deeper structural issue is: Who has the right to create markets?

Everyone agrees prediction markets need diversity—region-specific events, niche community topics, weird one-off events that traditional platforms would never touch... But permissionless market creation has always been a challenge.

The core issue is that hot topics have short lifespans. The most explosive trading opportunities often arise from breaking news and cultural events. For example, a market like “Will the committee revoke Will Smith’s Oscar for slapping Chris Rock?” could see huge volume in the hours after the event. But by the time a centralized platform reviews and lists it, interest has faded.

But fully permissionless creation faces three problems: semantic fragmentation (ten versions of the same question split liquidity into useless pools), cold start liquidity (zero initial liquidity makes the chicken-and-egg problem extreme), and quality control (platforms flooded with low-quality markets, or worse—legally risky assassination bets).

Both Polymarket and Kalshi have chosen a curated approach. Their teams review all markets to ensure quality and clear resolution criteria. This builds trust but sacrifices speed—the platform itself becomes the bottleneck.

What Entrepreneurs Are Trying

Melee uses a pump.fun-style strategy to address the cold start phase. Market creators get 100 shares, early buyers get decreasing amounts (3, 2, 1, etc.). If the market gains traction, early participants get outsized returns—potentially 1000x or more. It’s a “market of markets,” where traders bet early on which markets will grow. The core idea: only the highest-quality markets—built by top creators or truly meeting market demand—will attract enough volume. Ultimately, quality markets will naturally rise to the top.

XO Market requires content creators to provide liquidity using LS-LMSR AMMs. Creators earn fees by paying to seed markets, aligning incentives with market quality. Opinion market platforms like Fact Machine and Opinions.fun let influencers monetize cultural capital by creating viral markets around subjective topics.

Theoretically, the ideal is a hybrid, community-driven model: users stake reputation and liquidity to create markets, then community admins review them. This allows permissionless, rapid creation while maintaining quality. But no mainstream platform has pulled this off yet. The fundamental tension remains: permissionlessness brings diversity, admins ensure quality. Breaking this tradeoff will unlock the localized, niche markets the ecosystem needs.

Problem Five: Oracles and Settlement

Even if you solve liquidity, discovery, expression, and creation, there’s still the most fundamental challenge: Who decides what happened?

Centralized platforms have teams make decisions—efficient, but a single point of failure. Decentralized platforms need oracle systems to handle arbitrary questions without ongoing human intervention. But how to resolve these outcomes remains the hardest part.

As researcher Neel Daftary outlined for Delphi Digital, emerging solutions are a multi-layer stack that routes questions to the appropriate mechanism:

Automated data feeds for objective outcomes. In September 2025, Polymarket integrated Chainlink for instant settlement of crypto price markets. Fast and deterministic.

AI Agents for complex questions. Chainlink tested AI oracles on 1,660 Polymarket markets, achieving 89% accuracy (99.7% for sports). Supra’s Threshold AI oracle uses a multi-agent committee to verify facts and detect manipulation, ultimately providing a signed result.

Optimistic oracles like UMA suit ambiguous questions, proposing outcomes that disputing parties can challenge by staking funds. It’s game-theoretic but works well for clear questions.

For high-stakes disputes, use reputation-based juries, where voting power is tied to on-chain track record, not just capital.

The infrastructure is maturing rapidly, but market settlement remains the trickiest issue. If the settlement mechanism fails, trust is destroyed; if it works, it can scale to millions of markets.

Why These Problems Matter

Liquidity, market discovery, trader expression, market creation, and settlement are all interrelated. Solve liquidity, and you make markets attractive, improving discovery. Better discovery brings more users, making permissionless market creation viable. More markets mean greater demand for robust oracles. It’s a system, and right now, the system has bottlenecks.

But there’s opportunity: existing projects are stuck in their current models. The success of Polymarket and Kalshi is built on certain assumptions about how prediction markets should work. They’re optimizing within established constraints. The next generation of builders can ignore those constraints entirely.

Melee can try different Bonding Curves because they’re not trying to be Polymarket. Flipr can embed leverage into social feeds because they don’t need US regulatory approval. Seda can generate perpetuals from continuous data streams because they’re not limited by binary resolution.

This is the real advantage for prediction market entrepreneurs. Not copying existing models, but attacking the root problems directly. These five issues are table stakes. Platforms that solve them won’t just win market share—they’ll unlock the full potential of prediction markets as coordination mechanisms.

2024 proved prediction markets can be adopted at scale. 2026 will prove they can work anywhere.

“Original link”