Beyond the Oracle: Harnessing Market Intelligence

Background

When people think of intelligence nowadays, the first thing that comes to mind is probably LLM models, i.e. GPT, Claude, Llama etc.

But actually, the market itself may be the best form of general intelligence, as it is essentially the synthesis of all actions. AI itself is trained on huge amounts of information generated by the masses. That said, AI is passive, it needs goals and instructions (at least before we reach a true agentic world). To better harness and express market intelligence, we need something that captures the ever evolving minds of the crowd, something that is forward looking.

Enter prediction markets. A prediction market is a platform where participants can buy and sell contracts based on their belief about the potential outcomes of future events. These events can range from politics (e.g. election outcomes) and economics (e.g. interest rate changes), to entertainment and sports (e.g. competition outcomes).

(Source: ocularvc)

This concept is not new - early forms of prediction markets were believed to have existed more than 500 years ago, primarily for forecasting political outcomes.

In the early 2000s, prediction markets like Intrade and Betfair started to gain prominence, particularly during the US presidential elections. However, these were centralized platforms that were often limited by geographical restrictions, regulatory constraints, and the need for trusted intermediaries to manage funds and settle bets. This affected their ability to grow and scale.

• For example, Intrade was forced to shut down in 2013. The US Commodity Trade Futures Commission had sued Intrade to stop Americans from using the site, saying that it was illegally selling futures contracts, and this led to a sharp drop in the number of users.

In the late 2010s, with the rise of blockchain technology, prediction markets resurfaced stronger than before. This time, platforms leverage blockchain to create decentralized and global platforms that offer several advantages over past centralized variants:

However, prediction markets were never mainstream. Not until this year. Thanks to the 2024 US Presidential Elections, there is renewed interest and attention on this new form of market intelligence. In this piece, we will deep-dive into the mechanics of web3 prediction markets, specifically covering 1) the use cases and current landscape of prediction markets; 2) a case study on Polymarket; and 3) future trends.

1) Use Cases and Current Landscape

Besides offering users an opportunity to profit from their views/predictions of the future, there are other use cases of prediction markets, as outlined below:

DeFiLlama estimates that the current total value locked (TVL) in web3 prediction markets is around $140 million. This is down from the pre-election highs, where the TVL was at $545 million.

In Oct 2024, the combined monthly revenue of all prediction markets was estimated to be around $750,000, or about $9 million on an annualised basis.

The key players in the space are Polymarket, Azuro; and Drift (BET). Across these three players, the total betting volume has risen by over 550% in Q3 2024 to reach $3.1 billion, up from just $463.3 million in Q2 2024 (see image below).

(Source: CoinGecko and ocularvc)

In Nov 2024, there were about 290,000 monthly active traders on Polymarket alone, and over 300,000 new accounts were created on the platform. Based on the all-time leaderboard for Polymarket, the most active trader on Polymarket has generated $397 million in transaction volume, and the most profitable trader has earned over $22 million.

2a) Polymarket - A Case Study

So how do prediction markets work? This can be broken down into three sub-categories, namely its features; charging model; and dispute resolution process. As Polymarket is the dominant player currently, we will make reference to their operating model.

Features

• Markets - Typically tied to real-world events. Polymarket features a wide range of markets, from politics and crypto to pop culture and weather outcomes, while others may choose to focus on a particular niche, such as sports betting.
• Outcomes - It can be:
  • Binary, e.g. “yes” or “no” to whether an event will occur;
  • Multi-outcome, e.g. predicting which candidate will win a multi-candidate race or which team will win a tournament; or
  • Continuous, i.e. predicting outcomes across a range of values (e.g. stock prices or percentage of votes).
    • Polymarket features mostly binary outcomes. There are some multi-outcome markets, but for each option (e.g. candidate), it will be a binary trade (see example below). There is a probability/price indicated for each side of the trade, and after the event happens, a share of the correct outcome can be redeemed for $1, while a share of wrong outcomes will be worth $0.

(Source: Polymarket)

• Odds - There are two main ways that the prices/odds for the markets are decided:
  • One, through an order-book based system, similar to stock markets. Participants submit buy and sell orders; prices are determined by matching these orders.
  • Two, through an Automated Market Maker (AMM)-based system. In this system, every bid and ask is accepted. The price is determined and adjusted automatically based on an algorithm/mathematical formula that tracks the trading volumes.
    • Polymarket uses primarily an order-book based system.
    • When a market is created, there are initially zero shares and no pre-defined prices or odds. Those interested in buying “Yes” or “No” shares can place limit orders at the price they're willing to pay.
    • When offers for the “Yes” and “No” side equal $1.00, the order is "matched" and that $1.00 is converted into 1 “Yes” and 1 “No” share, each going to their respective buyers.
    • For example, if you place a limit order at $0.60 for “Yes”, that order is matched when someone places a “No” order at $0.40. This becomes the initial market price.
    • Subsequently, the prices displayed on Polymarket are the midpoint of the bid-ask spread in the order book — unless that spread is over $0.10, in which case the last traded price is used.
    • As seen with the market below, the probability/price of 37% is the midpoint between the 34¢ bid and 40¢ ask. If the bid-ask spread is wider than 10¢, the probability/price is shown as the last traded price.

(Source: Polymarket)

• Payment - Polymarket runs on the Polygon blockchain and users use USDC to place their orders. Polymarket also recently collaborated with MoonPay to allows users to use fiat to purchase USDC.
• Orders - Polymarket offers market, limit and AMM orders. However, at this juncture, there is no leverage option. After you place an order, Polymarket allows users to trade the shares they own before the event described by the market actually transpires.
  • With reference to the example above, let’s say that we purchased “Yes” shares for Ethereum hitting $3,000 when the odds were at 37%. If the odds increased after we placed our bet, we could decide to sell our shares for a higher price and lock in profits before the actual event happens/deadline hits. Of course, there is also the option of selling the shares at a loss if the odds decreased after we made our bet.

Fees

There are two main fees that decentralized prediction markets can charge:

• Trading fees, i.e. the platform charges a small fee each time a trade is executed.
• Deposit/withdrawal fees, i.e. a small fee is charged each time fiat/cryptocurrency enters or leaves the platform.

Polymarket currently does not charge any trading fees. However, they do take a 2% fee on net earnings from winning bets. Polymarket doesn’t take home any of this fee as revenue; instead, it is used to reward liquidity providers (as part of their Liquidity Rewards Program), and to pay for gas fees. Polymarket also does not charge any deposit/withdrawal fees.

• When asked about its pricing strategy, Polymarket’s Founder Shayne Coplan said in Jul 2024 that “We're focused on growing the marketplace right now and providing the best user experience. We'll focus on monetization later.”

Disputes

To resolve markets after the event is over, platforms typically rely on a mix of 1) oracles; and 2) community voting.

• In Polymarket’s case, they rely on both strategies through Universal Market Access (UMA)’s Optimistic Oracle and Data Verification Mechanism (DVM). A simplified chart to illustrate the market resolution process can be found below:

(Source: Polymarket and ocularvc)

• On the four possible outcomes that can take place after the debate period, these are the implications of the outcomes:
  • Proposer wins - in this case the proposer receives their $750 bond back plus half the disputer’s bond as a bounty. This is to prevent people from merely challenging proposals for fun.
  • Disputer wins – in this case, the reverse occurs, where the disputer receives their $750 bond back plus half the proposer’s bond as a bounty.
  • Too early to tell – in this case, proposer will get punished for resolving the markets too early. So the disputer receives their $750 bond back plus half the proposer’s bond as a bounty.
  • 50-50 – if any of the first 3 options are not appropriate, then the market price may resolve to a split, where it is 50% yes; 50% no. In this case, the proposer is punished for resolving the markets inaccurately. So the disputer receives their $750 bond back plus half the proposer’s bond as a bounty.

2b) Risks and Limitations

Thus far, there are varying views on the effectiveness of Polymarket, and prediction markets as a whole, as a source of market intelligence. We have outlined these different views below.

(Source: ocularvc)

Ocular’s view is that there needs to be three conditions in place for prediction markets to be used as a source of intelligence:

• One, prediction markets work best when the three factors of incentive, ability and timing are aligned.
  • On incentive: The individuals that you are polling will need to have a vested interest in an issue. This could be because it affects their daily lives; their other investments; or it’s a trending topic on social media that they would like to be a part of. Point is they need to feel engaged/involved in this issue for them to participate in the markets.
  • On ability: The public will need to have the required information to formulate their views. It can’t be too niche a topic or a topic that requires deep technical expertise, because in this case, the public may not be any wiser and it is hard for the results to be credible.
  • On timing: While the markets can be launched instantaneously, you need time to collect the public’s views and for the markets to react to new information. So, this will not work for time-sensitive decisions.
• Two, there needs to be sufficient liquidity. Ultimately, prediction markets works best when they are truly able to tap on the wisdom of the crowd. What this means is that the markets need to be at a certain scale, both in terms of the number of relevant individuals involved as well as the betting volume, for it to make sense and be useful.
• Three, it should not be used in isolation. Oftentimes, prediction markets will be driven by what is publicly available, e.g. what is mentioned in the news or on social media. There may be others who have private sources of data that may not be fully reflected in the trades, so it may be useful to seek them out to get a different perspective.

3) Future Trends

We see a trend of the extension of use cases of prediction markets:

Prediction markets can potentially be applied to decision markets. Instead of what “will” the outcome be, users vote on what “should” the outcome be. Prediction markets, while providing valuable insight, is more passive. People vote and wait for the outcome to be revealed, often having little influence over it. Decision markets on the other hand, are more active and more applicable for governance.

Case study: MetaDAO ($META)

MetaDAO is a project founded in January 2023 and invested by Colosseum and Paradigm. Its core product is Futarchy, where governance proposals are put forward for votes, and 2 conditional trading markets are concurrently spun up:

• When the market thinks that the proposal would increase the value of the token above a certain threshold, they can bid up the “Pass” token. Conversely, they bid up the “Fail” token.
• When voting ends, if the TWAP price of the Pass token is 3% higher than the Fail token, the proposal passes and will be implemented. Conversely, the proposal fails and markets will be reverted.
  • For example, the proposal can be on hiring a new CEO for the company. If the decision markets indicate that the value of the company’s stock will increase significantly if the CEO is hired, then the proposal will pass and the CEO will be hired.
  • The broad idea of the futarchy is illustrated below, and the specifics can be found here.

(Source: MetaDAO)

Futarchy was introduced in 2000 by economics Robin Hanson. He proposed it as a governance model combining prediction markets with traditional voting systems. In the Futarchy governance model, decisions are made based on predictions from decentralized markets. Instead of voting directly on policies, participants vote on measurable goals, such as economic growth. Prediction markets then forecast how proposed policies might affect those goals. The policy expected to achieve the best outcome, as determined by the market, is implemented. This approach leverages collective intelligence and financial incentives to guide decision-making.

Futarchy is better than Polymarket's binary betting model for scenarios where:

1. Decision-making is the goal, not just prediction.
2. Complex, long-term impacts must be considered.
3. Incentive structures need to align with societal or organizational objectives.

MetaDAO is currently working with six projects on decision-making, in addition to its own DAO governance. It has shown some early signs of success in decision-making, from preventing whales from purchasing $META at a deep discount, to directing firm resources away from new initiatives that the holders deem as distractions.

(Source: MetaDAO)

However, it is still early days for MetaDAO. There are several key constraints to MetaDAO’s model and it is to be seen whether this can work at scale:

• Oracle - Not every project has a token, and not every decision outcome can be measured precisely with metrics. It may also be difficult to tell impact on metrics by a relatively small decision.
• Liquidity - Concentration of users and wallets can lead to biased outcome. The UX may also be too technical for the broader population to get onboard with, and this may limit the pool of voters.
• Suitability - People in powerful positions may not want to transition decision making to markets. Participants also have to be an informed crowd. How do we ensure that users vote based on the long-term interests of the firm?

Other than MetaDAO, there are constant developments and innovation in prediction and decision markets:

• New markets - This year, there has been a lot of hype and chatter about prediction markets since it coincided with the Olympics and US Presidential Elections. The challenge for the sector will be to sustain interest in prediction markets, even after these cyclical events are over. Platforms can consider expanding beyond seasonal/one-off events, and explore different consumer base and categories, such as pop culture and social media, targeting female and youth participants.
• Advanced oracles - To establish new markets, new oracles may have to be built to scrape and fetch the relevant data to price new markets. Overlay is one such project that is looking to build oracles for unique markets, such as Counter-Strike skins and an AI index.
• Efficient arbitrage - Many of the platforms use an order-book based system (i.e buy and sell orders placed by users) to price their markets. As user behaviour and overall liquidity may vary across markets and platforms, there exists arbitrage opportunities both within and across platforms.
  • An example can be seen below with the odds for the US Presidential Election on Polymarket and Kalshi as of 28 Oct. A user could potentially purchase a contract for Kamala winning on Polymarket (at 33%) and a contract for Trump winning on Kalshi (at 62%), for a total cost of 95c. Given that the two events are mutually exclusive, the user will receive a payout of $1 regardless of the outcome, yielding a 5% arbitrage opportunity.

(Source: Polymarket)

(Source: Kalshi)

• To efficiently capitalise on these arbitrage opportunities, platforms may wish to:
  • Ensure that there are no duplicate or similarly worded listings on the platform;
  • Factor in the odds found on other platforms in their own pricing models; and/or
  • Build a trading bot to capitalise on such opportunities and minimise the spread across platforms.
• Capital efficiency - Today, to execute a trade on most platforms, users will need to have the capital on hand and once the trade is executed, the capital will be locked in on the platform. To promote greater capital efficiency, platforms can consider:
  • Introducing leveraged products so that users will be able to make outsized bets;
  • Allowing for the use of yield-bearing stablecoins/tokens to execute trades;
  • Tokenising the user’s position and allowing that token to be traded on other platforms; and/or
  • Designing a lending protocol that allow users to borrow against their positions.
• Anti-manipulation - To address price manipulation, platforms could consider setting a bet limit or restricting the number of accounts that an individual can open. Using cross examination can also be effective in small markets with thin liquidity: for example, asking “what you believe and what you think others believe”, and comparing the results.
• AI participation - To enable the resolution process, particularly for straightforward markets, to be done more efficiently, platforms can consider tapping on AI/large language models to obtain and verify the necessary information to resolve the markets. AI agents can also be trained to more efficiently research on truth and participate in future voting.

Conclusion

Ocular is keeping a close watch on the prediction market sector and its developments.

Today, prediction markets are used to generate income; hedge positions; engage communities; and gauge market sentiments. Moving forward, it can be used for collective decision-making/governance and many more interesting areas:

• Predicting the replicability of scientific papers;
• Aggregating private information (within corporations);
• Predicting the success of drug trials;
• Getting estimates on subjective matters (product pre-launch testing); and
• Reinventing news media – reporters and analysts to have skin in the game.

While there are existing inefficiencies/constraints with the sector that needs to be addressed (e.g. inconsistent liquidity; regulatory uncertainty; and oracle issues), we remain optimistic about the sector in the long-term, especially with the overlaps with AI.

We look forward to connecting with those building in the space and/or have views on how this sector will continue to evolve - if this resonates with you, do feel free to reach out to us at crypto@ocular.vc.