Author: Mikey, 1kx; Translation: 0xjs@Golden Finance
Augur is the first on-chain prediction market and one of the earliest applications launched on Ethereum. Its vision is to allow anyone to bet any size on anything. Augur's vision failed to materialize years ago due to numerous issues. Lack of users, poor settlement UX, and high gas fees led to the closure of the product. We have come a long way since then: block space is cheaper, order book designs are more efficient. Recent innovations have solidified the permissionless and open source nature of cryptocurrencies, allowing for an unconstrained global liquidity layer where anyone can become a market participant through liquidity provision, market creation, or betting.
Polymarket is the emerging market leader with around $900 million in volume to date, while SX Bet has amassed $475 million to date. Still, there is plenty of room for growth, especially when compared to the sheer size of sports betting, a subcategory of traditional prediction markets. In the United States alone, sports betting processed more than $119 billion in transactions in 2023. This volume figure stands out even more when considering the volume of physical and online sports betting in all other countries, as well as other types of prediction markets such as politics and entertainment.
This article aims to analyze how prediction markets work, the current bottlenecks that need to be addressed for further mass adoption, and some of the ways we think they can be addressed.
How do prediction markets work?
There are many ways to design prediction markets, most of which can be divided into two categories: order book models and centralized AMM (automated market maker) models. Our argument is that order book models are the superior design choice because they allow for better price discovery, maximum composability, and ultimately scaled trading volumes.
For the order book model, each market has only two possible predefined outcomes: yes and no. Users trade these outcomes in the form of shares. At market settlement, the correct shares are worth $1.00 and the wrong shares are worth $0.00. Before the market settles, the price of these shares can range from $0.00 to $1.00.
In order for stocks to be traded, there must be liquidity providers (LPs); in other words, they must provide bid and ask prices (quotes). These LPs are also called market makers. Market makers provide liquidity in exchange for a small profit on the spread.
Example of a specific market: If something has an equal probability of happening, such as a coin toss landing heads, then both the Yes and No stocks should theoretically trade at $0.50. However, like any financial market, there tends to be a spread, which allows for slippage. If I want to buy the Yes stock, my strike price may end up being closer to $0.55. This is because my counterparty (liquidity provider) is intentionally overestimating the true odds to gain potential profit. The counterparty may also be selling the No stock for $0.55. The $0.05 spread on each side is the reward the liquidity provider gets for providing the quote. The spread is driven by implied volatility (expectation of price movement). Prediction markets essentially guarantee realized volatility (actual price movement), simply because the stock must eventually reach a predetermined value of $1 or $0 by some predetermined date.
To illustrate the market maker scenario for the coin toss prompt above:
The market maker sells 1 share of "Yes" stock at $0.55 (which is equivalent to buying 1 share of "No" stock at $0.45)
The market maker sells 1 share of "No" stock at $0.55 (which is equivalent to buying 1 share of "Yes" stock at $0.45)
The market maker now has 1 share of No stock and 1 share of Yes stock, paying a total of $0.90
Regardless of whether the coin lands heads or not, the dealer redeems $1.00, making a $0.10 difference
The other main way to solve prediction markets is through centralized AMMs, which are used by both Azuro and Overtime. For the purposes of this article, we won’t go into these models too much, but the analogy in DeFi is GMX v2. Funds are pooled together and act as the sole counterparty to the platform’s traders, and the pool relies on external oracles to price the odds provided to users.
What bottlenecks currently exist in the prediction market?
Prediction market platforms have been around and known for a long time, and if there was true product-market fit, escape velocity would have already occurred. The current bottleneck can be attributed to the simple conclusion that there is a lack of interest on both the supply side (liquidity providers) and the demand side (bettors).
On the supply side, problems include:
Lack of liquidity due to volatility: Polymarket’s most popular markets tend to be those where the concept is novel and relevant historical data is scarce, making it difficult to predict the outcome and accurately price it. For example, it is difficult to predict whether OpenAI CEO Sam Altman will return to his post after rumors of possible mishandling of AGI circulated, as there are no past events that closely correlate to this situation. Market makers will offer wider spreads and less liquidity in uncertain markets to compensate for implied volatility (i.e. the wild price action in the Sam Altman CEO market, where consensus flipped 3 times in less than 4 days). This makes it less attractive to whales who want to place large bets.
Lack of liquidity due to too few experts: Despite hundreds of market makers earning rewards on Polymarket every day, many long-tail markets lack liquidity simply because there are not enough participants with expertise who want to make markets. Example markets include “Will celebrity x be arrested or charged for y?” or “When will celebrity x tweet?” This will change over time as more prediction market types are introduced, data becomes richer, and market makers become more professional.
Information asymmetry: Since order makers provide buy and sell prices that any taker can trade at any time, takers have the advantage of making positive expected value (EV) bets when they have favorable information. In the DeFi market, such takers can be called toxic flow. Arbitrageurs on Uniswap are a typical example of toxic flow takers because they use their information advantage to continuously extract profits from liquidity providers.
In one Polymarket market, “Will Tesla announce a Bitcoin purchase before March 1, 2021?”, a user bought $60,000 worth of Yes shares at odds of about 33%. This market was the only one the user had ever participated in, and it can be assumed that the user had favorable information. Leaving aside the legal issues, the market maker providing this quote could not have known that the taker/bettor had this favorable information at the time, and even if the market maker initially set the odds at 95%, the taker might still have bet because the true odds were 99.9%. This results in the market maker being bound to lose money. In prediction markets, it is difficult to predict when and how large a toxic flow will occur, making it even more difficult to provide tight spreads and deep liquidity. Market makers need to price in the risk of a toxic flow occurring at any time.
On the demand side, the main issues are:
Lack of leverage tools: Without leverage tools, prediction markets have a relatively inferior value proposition for retail investors compared to other crypto speculation tools. Retail investors want to earn "generational wealth", which is more likely to be achieved on memecoins than betting on capped prediction markets. Since their inception, early bets on $BODEN and $TRUMP have provided far greater upside than predicting stocks for Biden or Trump to win the presidency.
Lack of exciting short-term markets: Retail bettors are not interested in bets that will be settled in a few months, a conclusion that can be confirmed in the sports betting world, where a large amount of retail trading today occurs on live betting (ultra-short term) and daily events (short term). At least not yet, there are not enough short-term markets to attract a mainstream audience.
Is there any way to solve these problems? How can we increase the volume?
On the supply side, the first two problems (illiquidity due to volatility and illiquidity due to fewer experts) will naturally decrease over time. As the size of individual prediction markets grows, the number of professional market makers and those with higher risk tolerance and capital will also increase.
However, rather than waiting for these issues to abate over time, the lack of liquidity can be addressed head-on through a liquidity coordination mechanism originally invented in the DeFi derivatives space. The idea is to allow passive stablecoin depositors to earn yield through a vault, which deploys market-making strategies in different markets. This vault will act as the primary counterparty to traders. GMX was the first protocol to do this through a pooled liquidity provision strategy that relies on oracles for pricing, while Hyperliquid was the second notable protocol to deploy a native vault strategy, but with the distinction that liquidity is provided on CLOBs. Both vaults have been profitable over time because they are able to act as counterparties to the majority of non-toxic flows (retail users who tend to lose money over time).
Hyperliquid Vault PNL has been growing continuously
Native vaults allow protocols to easily bootstrap liquidity on their own without relying on other vaults. They also make long-tail markets more attractive; one of the reasons Hyperliquid has been so successful is that newly listed perpetual assets include a lot of liquidity from day 0.
The challenge of building a treasury product for prediction markets is preventing poison flows. GMX prevents this by charging high fees on its trades. Hyperliquid implements a market maker strategy with wide spreads, a 2 block delay on taker orders to give market makers time to adjust their quotes, and prioritizes market maker order cancellations within a block. Both protocols create an environment where poison flows cannot enter because they can find better price execution elsewhere.
In prediction markets, toxic flows can be prevented by providing deep liquidity at wider spreads, selectively providing liquidity to markets that are less susceptible to information advantages, or hiring sharp strategists who can gain information advantages.
In effect, the native vault can deploy $250,000 of additional liquidity, buying at 53 cents and selling at 56 cents. A wider spread helps increase potential vault profits because users will bet while accepting worse odds. This is in contrast to quotes at 54 and 55 cents, where counterparties may be arbitrageurs or savvy individuals looking for a good price. This market is also relatively less susceptible to information asymmetry problems (there is less inside information and insights, which are often released to the public relatively quickly), so the expectation of toxic flows is lower. Vaults can also use information oracles to provide insights into future line moves, such as pulling odds data from other betting exchanges or collecting information from top political analysts on social media.
The result is greater liquidity for bettors, who are now able to place larger bets with less slippage.
There are several ways to solve or at least reduce the information asymmetry problem. The first few are related to order book design:
Progressive Limit Order Book: One way to combat poison flow is to increase the price based on the speed and size of the orders. If buyers are certain that something will happen, then a reasonable strategy is to buy as many shares as possible at a price below $1.00. Additionally, if the market eventually gets favorable information, it is also wise to buy quickly.
Contro is implementing this GLOB idea and launching it on Initia in the form of an interwoven Rollup.
If the Tesla $BTC market were to occur in a GLOB model, the taker would have to pay a much higher fee than 33% per “yes” share because of the “slippage” that occurs given the combined velocity (a slice) and size (huge) of the orders. Regardless of the slippage, he would still make a profit because he knows his “yes” share will ultimately be worth $1, but this at least covers the market maker’s losses.
One could argue that if the taker executes a DCA strategy over a long period of time, they could still suffer very little slippage and pay closer to 33% of the “yes” stock price per share, but in this case it at least gives the market maker some time to withdraw its bid from the order book. Market makers may withdraw bids for several reasons:
It suspected a poison flow because there were a large number of taker orders coming in
It was convinced there was a toxic stream because it checked the recipient’s profile and found that he had never placed a bet before.
It wants to rebalance its inventory and no longer wants to be too one-sided because it is selling how many "yes" stocks and accumulating how many "no" stocks - maybe the market maker initially has orders worth $50,000 on the sell side, accounting for 33%, and orders worth $50,000 on the buy side, accounting for 27% - its initial goal is not to have a directional bias, but to remain neutral so that it can earn profits through symmetrical liquidity provision
Winners’ Take: In many markets, some of the profits are redistributed to those with favorable information. The first example is peer-to-peer web2 sports betting, specifically Betfair, where a fixed percentage of the user’s net winnings is redistributed to the company. Betfair’s take really depends on the market itself; in Polymarket, it might make sense to charge a higher net winning take for newer or long-tail markets.
This concept of redistribution also exists in DeFi in the form of order flow auctions. Bots running in the background extract value from information asymmetry (arbitrage) and are forced to return the proceeds to the people participating in the transaction, who may be liquidity providers or users who make transactions. To date, there have been many PMFs for order flow auctions, and CowSwap* is pioneering this category with MEVBlocker.
Static or dynamic taker fees: Polymarket currently does not charge taker fees. If implemented, proceeds could be used to provide liquidity rewards in markets with high volatility or susceptible to toxic flows. Alternatively, higher taker fees could be set in long-tail markets.
On the demand side, the best way to address the lack of upside is to create a mechanism that allows for upside. In sports betting, multiple bets are becoming increasingly popular with retail bettors because they offer the opportunity to "win big." A multiple bet is a bet that combines multiple individual bets into a single bet. For the multiple bet to win, all of the individual bets must win.
One user won over $500,000 with an initial bet of just $26
There are three main ways to increase the number of users of crypto-native prediction markets:
Multiple bets
Sustainability
Tokenized Leverage
Multiple bets: Technically, this is not feasible on Polymarket’s books, as bets require upfront capital and the counterparty is different for each market. In practice, the new protocol can fetch odds from Polymarket at any given point in time, price the odds for any multiple bet, and act as the sole counterparty for the multiple bets.
For example, a user wants to bet $10 on:
If placed individually, these bets have limited returns, but when they are combined into multiple bets, the implied return soars to about 1:650,000, which means that if every bet is correct, the bettor can win $6.5 million. It is not difficult to imagine how multiple bets can earn PMF among cryptocurrency users:
The cost of participation is low, you only need to invest a small amount of money to win a lot
Sharing a multiple bet slip can quickly go viral on crypto Twitter, especially when someone wins a jackpot, creating a feedback loop with the product itself
There are challenges in supporting multiple bets, namely counterparty risk (what happens when multiple bettors win large multiple bets at the same time) and odds accuracy (you don’t want to offer bets that underestimate the true odds). Casinos have solved the challenge of offering multiple bets in the sports world, and it has become by far the most profitable part of sports betting. Even if some bettors get lucky and win big, the profit margin is about 5-8 times higher than offering single market bets. Another added benefit of multiple bets is that there are relatively fewer toxic streams in multiple bets compared to single game markets. The analogy here is: why would a professional poker player who lives and dies by expected value put money into the lottery?
SX Bet is a web3 sports betting application chain that has launched the world's first peer-to-peer multi-bet system, which has reached $1 million in multi-bets in the past month. When a bettor "requests a multi-bet", SX creates a private virtual order book for the multi-bet. Programmatic market makers listening through the API will have 1 second to provide liquidity for the bet. It will be interesting to watch the liquidity and traction of non-sports multi-bets increase.
Perpetual Prediction Markets: This concept was briefly explored in 2020 when the former leading exchange FTX offered perpetual contracts for the US election results. You could go long the price of $TRUMP and redeem for $1 per share if he won the US election. As the odds of Trump actually winning changed, FTX had to change margin requirements. Creating a perpetual mechanism for a market as volatile as a prediction market presents a lot of challenges with collateral requirements, as the price could be $0.90 one second and $0.10 the next. As a result, there may not be enough collateral to cover the losses of someone who did the wrong thing. Some of the order book designs explored above can help compensate for the fact that prices can change so quickly. Another interesting thing about the FTX $TRUMP markets is that we can reasonably assume that Alameda is the primary market maker for these markets, and without natively deployed liquidity, the order books would be too thin to generate much volume. This highlights the value of native liquidity vault mechanisms for prediction market protocols.
Both LEVR Bet and SX Bet are currently working on the perpetual sports betting market. One benefit of sports leverage is that “yes or no” stocks will see smaller price swings, at least in most situations. For example, a player making a basketball shot might increase the team’s chance of winning from 50% to 52%, since the average team might make 50 shots per game. A 2% increase on any one shot is manageable from a liquidation and collateral requirements perspective. Offering perpetual contracts at the end of a game is another matter, since someone might make the “game-winning shot” and the odds could change from 1% to 99% in half a millisecond. One potential solution is to only allow leveraged betting up to a certain point, since any event after that would change the odds too much. The viability of perpetual sports betting also depends on the sport itself; a hockey goal changes the expected outcome of a game far more than a basketball shot.
Tokenized Leverage: Lending markets that allow users to borrow against their prediction market positions (especially long positions) could increase volume for professional traders. This could also lead to more liquidity, as market makers can borrow against positions in one market to trade in another. Tokenized leverage may not be an interesting product for retail bettors unless there are abstracted loop products, such as those that Eigenlayer is attracting. The overall market may be too immature for such an abstraction layer to exist, but these types of loop products will eventually appear.
Beyond the pure supply and demand side of things, there are other nuanced ways to increase adoption:
From a UX perspective: Switching settlement currency from USDC to a yield-bearing stablecoin will increase participation, especially in long-tail markets. This has been discussed a few times on Twitter; holding a market position that expires at the end of the year has a large opportunity cost (e.g. betting on Kanye West to win the presidency earns 0.24% APR, while earning 8% APR on AAVE).
Additionally, gamification aimed at improving retention can really help attract more users in the long run. Something as simple as a “daily betting streak” or “daily contest” works well in the sports betting industry.
Several industry-level tailwinds will also increase adoption in the near future: the combination of growing virtual and on-chain environments will unleash entirely new speculative demand, as the number of short-term events will eventually be infinite (think AI/computer-simulated sports) and the level of data will be abundant (making it easier for market makers to price outcome odds). Other interesting crypto-native categories include AI gaming, on-chain gaming, and general on-chain data.
Accessible data will lead to increased levels of betting activity from non-humans, more specifically autonomous agents. Omen on Gnosis Chain* is pioneering the idea of AI agent bettors. Since prediction markets are a game where the outcome is already determined, autonomous agents can become increasingly adept at calculating expected value, potentially with far greater accuracy than humans. This reflects the idea that it may be more difficult for AAs to predict which memecoins will become popular because there is a more “emotional” element to what makes them successful, and humans are currently better at feeling emotion than AAs.
All in all, prediction markets are a fascinating user product and design space. Over time, the vision of allowing anyone to bet on anything at any scale will become a reality.
Thanks to Peter Pan, Shayne Coplan, Sanat Kapur, Andrew Young, taetaehoho, Diana Biggs, Abigail Carlson, Daniel Sekopta, Ryan Clark, Josh Solesbury, Watcher, Jamie Wallace, and Rares Florea for their feedback and comments!
*Denotes a 1kx portfolio.
