Market Making in Prediction Markets Sucks (5 minute read)
Prediction markets suffer from fragmented liquidity across platforms with no arbitrage equalization, leaving passive market makers to absorb losses while informed traders extract $228M in profits.
What: An analysis of why traditional market making fails in prediction markets, backed by data from 72M Kalshi and 150M Polymarket trades, showing how fragmentation, manipulation risk, and lack of pricing infrastructure create systematic losses for liquidity providers. Lotus proposes a cross-venue routing layer with RFQ and real-time volatility tracking as a solution.
Why it matters: The piece reveals structural problems that prevent prediction markets from scaling reliably: a January 2026 XRP contract exploit on Polymarket netted $231k by manipulating thin weekend liquidity, demonstrating how single-venue settlement and isolated order books enable price manipulation that would be impossible in mature financial markets with cross-exchange arbitrage.
Takeaway: If you're trading prediction markets, use aggregators to check prices across venues before placing orders—the article shows the same contract can vary 9 percentage points between platforms, costing real money on every trade.
Deep dive
- Analysis of 72M Kalshi and 150M Polymarket trades reveals the top 5% of skilled traders captured $228M over three years through persistent wealth transfer from uninformed participants, while passive liquidity providers consistently lost money
- Prediction markets exhibit severe fragmentation with the same event trading at vastly different prices across platforms (e.g., Bitcoin >$70k at 58¢, 62¢, and 67¢ on different venues) because arbitrage bots don't equalize prices like in crypto spot markets
- A January 2026 Polymarket exploit demonstrated the manipulation risk: a trader pushed an XRP contract to 70% on a 0.3% spot move during thin weekend liquidity, then executed a $1M spot buy to swing the contract to $1, netting $231k profit funded by the platform
- Single-venue pricing creates systemic vulnerability—a flash crash on one exchange could cause prediction markets using that exchange's oracle to settle wildly out of line with true consensus
- Favorite-longshot bias is endemic: bettors systematically overpay for unlikely outcomes (longshots win much less than implied odds) while favorites win slightly more often, creating an "optimism tax" that makers capture from takers
- Passive LPs in prediction markets function more like underwriters absorbing residual demand imbalances with binary outcome exposure, rather than earning traditional bid-ask spreads through inventory management
- Current mechanisms (LMSR, CFMMs, order books) guarantee expected losses for makers because they can't hedge binary outcome risk until settlement, leaving them exposed to informed traders and terminal risk
- Prediction markets lack a Black-Scholes equivalent—there's no standardized framework for quoting implied volatility or hedging event risk, forcing platforms to price ad hoc
- Information shocks produce heterogeneous responses: debate volumes mostly reversed, assassination attempts caused permanent repricing, while Biden's withdrawal saw frantic trading with little net price change, making it impossible for naive algorithms to distinguish noise from news
- Lotus's solution aggregates liquidity across venues like 1inch, adds a native RFQ layer for when order books can't support size, and tracks cross-market "implied belief volatility" to enable proper risk pricing
- The platform routes orders to capture arbitrage (58¢ vs 67¢ spreads), hedges inter-event risk by routing correlated markets (Trump wins vs Biden wins), and plans protocol-owned liquidity as a backstop during thin markets
- The infrastructure gap is fundamental: prediction markets have no multilateral clearing, no cross-exchange pricing standards, and no unified liquidity metric, leaving them fragmented where traditional derivatives solved similar problems decades ago
Decoder
- LMSR: Logarithmic Market Scoring Rule, an automated market maker mechanism that adjusts prices based on total shares outstanding
- CFMM: Constant Function Market Maker, a DeFi primitive (like Uniswap's x*y=k) that provides liquidity via mathematical formulas rather than order books
- RFQ: Request for Quote, a trading mechanism where liquidity providers privately quote prices for specific orders rather than posting public orders
- Favorite-longshot bias: Systematic pricing error where low-probability outcomes (longshots) are overpriced and high-probability outcomes (favorites) are underpriced relative to their true odds
- Adverse selection: When informed traders systematically trade against market makers who lack information, causing makers to lose on average
- Implied volatility: The market's expectation of future price fluctuations derived from option prices; prediction markets currently lack an equivalent framework
- Terminal risk: The exposure to final binary outcome (win/lose) that can't be hedged away before settlement, unlike continuous markets where positions can be unwound
Original article
Prediction markets have fragmented liquidity across venues with no fast arbitrage equalization, evidenced by an XRP contract exploit on Polymarket in January, where thin weekend liquidity allowed a trader to push the price to 70% on a 0.3% spot move, netting $231,000. Analysis of 72M Kalshi trades and 150M Polymarket trades shows the top 5% of skilled traders captured $228M over three years through persistent wealth transfer from uninformed takers, while passive LPs absorb binary-outcome inventory risk without earning typical bid-ask spreads. Without a Black-Scholes equivalent for prediction markets, current mechanisms, including LMSR, CFMMs, and order books, guarantee expected losses for makers. The proposed solution routes orders across venues, pools fragmented liquidity with a native RFQ layer, and tracks cross-market implied belief volatility in real time.