I originally wrote this article about Automatic Market Makers (AMM) 3 years ago as an attempt to impress my future boss at Kalshi. The PM space has changed so, so much since June 2022, but the I think the majority of the article still rings true.
What has changed since then?
Polymarket did in fact move to an order book, using an on/off-chain model. So the part about Poly’s issues with execution and front-running is outdated.
Liquidity in prediction markets overall is WAY up, thanks to efforts like Polymarket’s liquidity incentives and Kalshi inking a deal with a huge market maker in SIG.
Order flow is also way up as prediction markets hit the mainstream in 2024 and brokers like Robinhood and WeBull started integrating Kalshi’s markets
There’s now a Cambrian explosion of prediction markets as many race to grab a slice of this growing pie
Many are on-chain, and many are pursuing an regulated off-chain strategy as well
GaetanD, whom I quote in this article as a retired trader, has come out of retirement and is printing money on Kalshi’s Rotten Tomatoes markets
I had written that Kalshi’s trading hours were a boon to market makers, but now Kalshi’s trading hours are now 24/7 so that’s not true anymore. “Circadian rhythms in shambles.”
I’m reposting this now for a few reasons:
Selfishly, to say “I told you so!”, especially the part about the UMA protocol
As a cautionary tale to builders looking to implement AMMs in their prediction market exchange
That said, I don’t think AMMs are worthless. They could prove useful as an early-stage bootstrapping strategy that then transitions into a mature orderbook/market maker. More on that in a later post.
Speaking of which, you should subscribe! It’s literally free! No paid tier or paywalls. I’m publishing 2-3x a week.
Also, Lucas at Adjacent News has a fantastic deep dive into the history, mathematical foundation, and technical nitty-gritty of how decentralized AMMs work. A must-read if you’re a market structure nerd!
The original article follows, unedited:
Prediction markets are a growing asset class with humble beginnings. But prediction markets (PM) are still relatively niche and struggle with low amounts of liquidity. Historically, in central limit order book (CLOB) markets, this problem was solved by the advent of designated market makers (DMM). What started out as a couple dozen guys in top hats on Wall St. has turned into a massive and lucrative industry dominated by high-frequency trading algorithms.
One recent innovation is the introduction of blockchain-based Automatic Market Makers, or AMMs, as a way to decentralize the concept and bootstrap liquidity. “They have applied a deterministic pricing rule in the context of digital asset exchange, redefined the process of liquidity provisioning for market making, and democratized access to global pools of capital,” wrote Dmitry Berenzon. To decentralize the markets, these AMMs rely on the public to participate as liquidity providers (LP) and stake their currency as the liquidity that traders trade against in exchange for fees collected.
While I do think decentralized exchanges that use AMMs are probably an important (and at least fascinating) innovation, as an experienced PM trader I don’t believe on-chain AMMs are the future of the space. Here, I cover a few of the reasons why this approach isn’t good for traders and results in a lower quality and inefficient market.
On-chain trading is not performant
On-chain trading is slow and cumbersome. “Web2” trading is plenty fast — NYSE may settle 5 billion trades on a typical trading day, easily clearing 150,000 transactions per second. But, most traders on PolyMarket, the pre-eminent on-chain PM, are all too familiar with the feeling of submitting an order and waiting longer than 20 seconds to see it go through (or not, more on that later). Polymarket uses Polygon, a protocol that, in the absolute best conditions, might be capable of a tenth of NYSE’s throughput. In PMs, traders traffick in information, and the ability to react to new information quickly is crucial.
Traders don’t appreciate having all of their trades public
PM trading is inherently competitive and trading on-chain exposes every trader’s moves for anyone to see. The most obvious downside to this is tailing — who wouldn’t want to see what trades lanceuppercut is making as they are making them, and then follow behind? If a well-known trader whose wallet was doxxed were trying to accumulate a larger position, it would be impossible without alternate accounts or misdirection.
The non-obvious downside is actually just the opposite of tailing. On-chain trading opens up the risk of being front-run on all of your trades. The best PM traders have been repeatedly targeted by this strategy. “It’s frustrating when you’re on Polymarket to submit a $1k order to buy, but because you used the Polymarket UI, somebody else is able to pay a little bit more [in gas fees] to get their transaction processed before yours so they make the exact same trade, yours fails and you see the price go up 5 cents,” says GaetanD, a well-known and now-retired PM trader. “Any benefits of the blockchain are outweighed by that cost right there, alone.”
One potential benefit of a public ledger of trades is rooted in the philosophical ideal of information aggregation. If the point of a prediction market is to aggregate the known information and opinions of monetarily-incentivized parties, then one could argue that having every trade exposed would be the most radical version of that transparency. If one followed PM prices as a way to inform their expectations, one could also benefit from getting to see the explicit opinions of specific traders.
“I don’t find that angle compelling,” said jipkin, another pre-eminent PM trader. “There’s not much value in knowing specifically who is making what trade; ultimately smart people can and often will be on opposite sides of any given trade. On top of that, sometimes sharp traders will disguise their action by using an alternate address to make trades.” Ultimately, exposing all of the trades will incentivize deception, create a lower quality of life for traders, and get in the way of efficient price discovery.
Traders like to make markets, too
Typically, the term “market maker” refers to non-directional parties — traders or institutions that don’t intend to take any particular opinion on the asset’s value. These market makers intend to profit by buying and selling the spread and pocketing the difference. But, experienced prediction market traders know that prices can be incredibly volatile and that the art of being a profitable PM trader involves more than just forecasting the result — it also involves predicting all of the swings in between the present and the resolution. If one can predict how the market will react to new information and in what order, making limit orders with the intent of profiting off of this volatility is not possible in an AMM market.
Central limit order books, or CLOBs, are the most obvious and simple way of offering this functionality. For example, in the recent Democratic primary for Texas 28th congressional district, a sophisticated trader could have predicted that the price for Henry Cuellar will tank before recovering strongly, due to quirks of how that district reports results.
If a trader wanted to exploit this in an AMM market, the trader would have to be present and wait for the swings to occur to buy the shares she wants. In theory, the trader could program a bot that watches the price and executes trades, but that would be essentially just creating their own personal CLOB. In a CLOB market, it’s as simple as setting the limit orders at the prices they like and going to bed, which is exactly what I did that night.
“Disintermediation” is not a good thing for prediction markets
One of the philosophical underpinnings of crypto and blockchain itself is the idea of disintermediation, or getting rid of the middleman between parties. In this case, the middleman would be the exchange (e.g. PredictIt, Kalshi) and the parties are traders. And, amongst all of the hype of Web3, there is a religious fervor bubbling up that aims to decentralize or tokenize all sorts of assets from music royalties to home titles, often with the default assumption that a decentralized version of anything is inherently better than a centralized version.
In my view, disintermediation is not necessarily a good thing for PMs. Any experienced PM trader is well aware (and maybe even fond) of the infamous “rules cuck”, where ambiguity in the rules of deciding the outcome of a market results in a market flipping, often from 95–5 to 5–95. This kind of thing happens often enough that traders stand to benefit from the decision-making of a trusted and perhaps regulated intermediary to adjudicate things fairly. In order to decentralize the process, Polymarket uses UMA, an “optimistic oracle”, to resolve market outcomes. But, in the case of a rules fight, the way things are resolved is by a vote amongst UMA token holders, where one token equals one vote. On top of that, the majority of UMA tokens are concentrated in just a few wallets. It really goes without saying how potentially problematic that is — I would much rather have my assets governed by an exchange that is at the mercy of the CFTC than the whims of anonymous crypto bros.
AMM Liquidity Providers face exotic financial risk
Market making in any PM is inherently biased towards loss-making do to information asymmetry and adverse selection. The volatility of PMs makes it difficult to be delta-neutral, and AMM LPs are effectively the counterparty to every trade. So if an LP adds liquidity for a 50 YES / 50 NO market that then goes to 70 YES / 30 NO, when that LP pulls the liquidity they will be provided shares that make their position net NO. If one wanted to collect fees profitably and be neutral to the outcome of the market, they would have to stick with markets that see a lot of volume but little volatility. But that’s extremely rare in PMs — generally, new volume occurs when new information comes out, and when new information comes out the price will move one way or the other.
For LPs in an AMM system, this manifests itself as the risk of being the final bagholder. For example, LP’ing a market with potential for sudden-death resolution like “Will a picture of Kim Kardashian and Pete Davidson kissing surface by the end of the month?” is risky, because as soon as that picture is released traders will buy “yes” to a price approaching 100, and the LP will be stuck with all of the worthless “no” shares and lose everything.
It’s true that the same risk exists for anyone, directional or non-directional, who makes a limit order in a sudden death market. But in that case, central limit order books (CLOB) offer much more flexibility to those traders to react. The only strategy that LPs in an AMM market have is deciding how much liquidity to provide and when.
AMMs can’t handle linked margin
Linked margin, often referred to as “MECNET”, “collateral return”, or “negative risk”, is the concept of automatically returning funds to traders who have a set of positions that guarantees some payout. Kalshi has an excellent article explaining how this works in detail.
Linked margin is important because it allows traders to provide the same amount of liquidity (in the form of limit orders) across several NO contracts with less capital than otherwise. It also incentivizes efficient pricing because illogical pricing (e.g. 40/40/40 market) will be arbitraged away easily and automatically in a linked margin CLOB system.
For this to work, there needs to be a centralized party to be the custodian of the capital to be able to return it. But that wouldn’t work in a set of liquidity pool-based markets because each one would have to have its own pool and the capital belongs to the LPs and the pools are not connected. This results in markets that require more collateral than necessary, locked up for longer than necessary, to achieve a comparable amount of liquidity in a CLOB.
Is there a better way to be a PM market maker?
Market making in PMs is very difficult, and generally loss-making due to adverse selection and information asymmetry. So, market makers will need to employ every possible technique and advantage they can scrape together, much in the same way the biggest institutional market makers do with traditional equities and derivatives.
For example, a program that watches Twitter for relevant information and reprices assets accordingly, or a trading strategy that reacts to customer flow in a Bayesian manner to minimize downside risk. A CLOB allows for a massive design space of sophisticated market-making strategies, which at least gives market makers a shot at being profitable. Kalshi, a CLOB-based PM platform, even has trading hours to give market makers a chance to pull orders before the market opens in case market-changing information came out overnight. LPs in an AMM market can only decide when and how much money to put in and have no such flexibility.
Ultimately, it is hard to justify the existence of on-chain, AMM-based PMs aside for one thing: the ability to bet on things that regulated entities would not touch. From Britney Spears’ conservatorship to the ship stuck in the Suez Canal, you can make a market for anything as if you exist outside of the established legal and financial systems. But other exchanges have already paid the price for doing PMs in an unregulated manner — it’s simply not sustainable. Standard CLOB markets are tried and true and better for traders in nearly every way.