Why Most Prediction Market AMMs Are Structurally Doomed

Liquidity is a liability. Prediction market AMMs like Polymarket's LMSR treat liquidity as a static asset, but it is a dynamic liability that must be hedged. Each liquidity position is a short volatility position on the market's outcome, exposing LPs to asymmetric, unbounded loss.

The AMM misprices tail risk. Constant function market makers like Uniswap v2 are designed for correlated assets, not binary outcomes. This creates a systematic mispricing of extreme probabilities, allowing informed traders to extract value from LPs until the pool is drained.

Compare to order books. Centralized platforms like Kalshi or traditional sportsbooks use a bookmaker model, where the house actively manages risk and capital. The passive, automated AMM model cedes control to the most informed counterparty, guaranteeing its capital is the exit liquidity.

Evidence: Historical data from Augur v1 and Gnosis shows that over 90% of liquidity providers in prediction markets end with a net loss. The mechanism is a mathematical certainty, not a market inefficiency.

CFMMs price liquidity, not information. A Uniswap v2 pool treats a bet on 'Trump wins' identically to a swap for ETH. The AMM's pricing curve reacts only to trade size, ignoring the fundamental value shift from new poll data.

This creates a free option for informed traders. A well-informed bettor exploits the stale price, extracting value from the LP's inventory. This is the adverse selection problem that bankrupts traditional bookmakers.

Prediction markets are zero-sum, DeFi pools are not. In a Uniswap pool, LP profits from volatility. In a prediction market, one side's gain is the other's loss; LPs are the predictable losers.

Evidence: Platforms like Polymarket avoid CFMMs, using order books or peer-to-peer models. The failure of early AMM-based markets like Augur v1's scalar markets demonstrated unsustainable LP losses.

The core failure is mispricing. Automated Market Makers like those used by Polymarket or PlotX price binary outcomes via a bonding curve. This model assumes a normal distribution of information, but real-world events are fat-tailed. The AMM consistently undervalues low-probability outcomes, creating a permanent delta between the market price and the true statistical price.

This mispricing is an arbitrage subsidy. Sophisticated actors like Wintermute or Jump Crypto execute a delta-neutral strategy. They buy the undervalued 'yes' shares on the AMM while hedging the risk via a traditional sportsbook or OTC desk. This 'risk-free' arb extracts value from the protocol's liquidity pool with every trade, acting as a continuous fee on LPs.

The system becomes insolvent at resolution. At event settlement, the AMM must pay out all winning shares at $1. The treasury's value is the sum of all shares, but the mispriced asset distribution means the pool lacks sufficient stablecoins to cover liabilities. The insolvency is mathematically guaranteed for sufficiently mispriced, high-conviction events, rendering LPs' 'yield' a mirage of their own capital.

Uniswap v3 is not a panacea. Its concentrated liquidity model works for stable, mean-reverting assets like ETH/USDC, not for binary outcomes. Prediction markets require liquidity across the entire 0-1 probability range, creating a massive capital efficiency problem that v3-style concentration cannot solve.

The liquidity is fundamentally ephemeral. Unlike a DEX pool where LPs profit from volatility, prediction market LPs face guaranteed adverse selection. Informed traders extract value from LPs, making passive liquidity provision a structurally losing game, as seen in early platforms like Augur and Gnosis. This is the AMM's adverse selection tax.

Compare to order book models. Centralized platforms like Polymarket use a maker-taker model where professional market makers provide tight spreads because they can hedge risk and manage inventory. On-chain AMMs cannot replicate this without sacrificing decentralization or introducing trusted intermediaries.

Evidence: The total value locked in prediction market AMMs is negligible compared to DEXs. Platforms using hybrid or non-AMM models, like Polymarket (order book) or Kalshi (CFTC-regulated), consistently demonstrate higher liquidity and tighter spreads for equivalent markets, proving the model mismatch.

Static liquidity pools are information-blind. They cannot distinguish between noise traders and informed actors, creating a structural arbitrage opportunity. This leads to predictable losses for LPs, mirroring the adverse selection problem in traditional finance.

Information asymmetry destroys LP returns. In markets like Polymarket or Augur, an informed bettor's edge comes directly from uninformed LPs. The AMM's constant product formula guarantees this loss, making passive liquidity provision a negative-sum game without massive subsidies.

The solution is dynamic, information-sensitive mechanisms. Protocols must move beyond the Uniswap V2/V3 model. Mechanisms must price liquidity based on signal, similar to how Robinhood or Citadel internalize order flow, or how KeeperDAO coordinates MEV.

Evidence: Analysis of Polymarket liquidity pools shows LP returns are consistently negative after fees during high-volatility events. This proves the model is fundamentally extractive, not sustainable.