Why does the Automated Market Maker (AMM) mechanism fail in the prediction market space?

Written by: Melee

Translated by: AididiaoJP, Foresight News

In July 2017, Hayden Adams was fired by his employer Siemens, where he worked as a mechanical engineer. His college roommate Karl Floersch was working at the Ethereum Foundation at the time and often discussed smart contract topics with him. Adams had previously paid little attention to these matters. Now unemployed and seeking something to do, he decided to listen and learn.

The Birth of Automated Market Makers (AMMs)

Floersch recommended a blog post by Vitalik Buterin, which discussed running on-chain exchanges using mathematical formulas instead of order books. The principle was not to match buyers and sellers directly, but to allow traders to exchange with an asset pool, with prices automatically set based on the ratio of tokens in the pool. At that time, no usable version had been released. Adams took it as a learning project, received a $65,000 grant from the Ethereum Foundation, and launched Uniswap in November 2018.

Its formula was almost childishly simple: x * y = k.

Two tokens are placed in a pool, and their product remains constant. When someone buys one token, they must deposit the other, causing the ratio within the pool to change and the price to adjust accordingly. No order book, no matching engine, no professional market makers. Anyone can deposit tokens into the pool and earn fees from each trade.

This made automated market makers the cornerstone of decentralized finance. Uniswap, Curve, Balancer, and dozens of other protocols handle billions of dollars in trading volume. On-chain order books are slow and expensive, and traditional market makers have no interest in participating in tokens with only a few hundred holders. AMMs enable anyone to create liquidity markets for any asset at any time. Before AMMs, launching a new asset required permission and infrastructure. After their emergence, all you need is a liquidity pool.

The benefits are obvious. Consequently, prediction markets naturally sought to adopt them as well.

AMMs and Prediction Markets

Prediction markets and token markets face the same cold start problem. You need liquidity providers willing to trade first, and traders willing to provide liquidity first. Little known is that Robin Hanson, years earlier, proposed an automated market making scheme for prediction markets based on his 2002 logarithmic market scoring rule.

He believed he had solved the cold start problem in theory. But in practice, the scheme faced the same fundamental issue that every attempt to automate liquidity in prediction markets would encounter later: the formula cannot distinguish between tokens that fluctuate forever and rights that expire.

Prediction market outcomes are binary. They settle at either one or zero. In token exchange pools, both assets can fluctuate indefinitely, and the AMM formula is effective because both tokens are designed not to reach zero.

Early Polymarket used an AMM based on the logarithmic market scoring rule. Augur also experimented with similar schemes. If automated liquidity pools are effective for token swaps, they should also work for betting on elections.

But that is not the case.

Why AMMs Fail in Prediction Markets

When a prediction market event settles, one side is worth one dollar, the other zero. For liquidity providers, the mathematical outcome is almost brutal. As the market approaches settlement, the pool automatically rebalances toward the losing side.

Impermanent Loss

The “impermanent loss” traders refer to in DeFi becomes a permanent “permanent loss” here. Every market will settle, and every liquidity pool will eventually hold a set of shares worth zero.

In typical DeFi pools, trading fees can offset impermanent loss over time.

But in prediction markets, the loss is structural and inevitable. The only question is how much liquidity providers lose. Protocols have tried to persuade users to deposit assets into such pools through liquidity mining, rewards, and various incentive schemes. All these are just different ways of subsidizing users to lose assets at a slower rate.

Price Discovery

Another issue is price discovery. AMMs price assets based on the ratio within the pool and a fixed formula. For tokens, the “correct price” is inherently a moving target, and the approximate value driven by the formula is sufficient. But prediction market prices should represent probabilities. The slippage introduced by the constant product curve distorts signals, especially in markets with low liquidity, where a single trade can shift implied probabilities by several basis points.

Is a Central Limit Order Book (CLOB) Better Than an AMM?

Polymarket was quick to realize this. By late 2022, the platform migrated from an AMM based on the logarithmic market scoring rule to a central limit order book. AMMs are designed for continuous token exchanges across price ranges. Prediction markets, however, require precise probability pricing on binary outcomes with known final results. These are fundamentally different problems.

The revolutionary feature of AMMs for tokens—permissionless market creation, instant liquidity provision, and independence from professional market makers—is precisely what prediction markets need. The problem is that the specific mechanism of the constant function formula, designed for token swaps, becomes difficult to sustain once faced with binary outcomes and inevitable settlement.

The challenge for prediction markets is how to develop infrastructure that can reflect the actual settlement process of such markets and reproduce the desired effects.