This grant is to create subsidized real money prediction markets on high impact topics such as geopolitical risk, nuclear proliferation, and AI development utilizing existing prediction market infrastructure (Polymarket). The goal of this program is to create highly accurate forecasts which respond in real time to events using the wisdom of the crowds and markets to aggregate and price information.
This project will be carried out in partnership with Polymarket.com. I will work with the Polymarket team to create a number of markets on high impact topics (Geopolitics, AI, etc.). Grant funds will be used to subsidize markets using an automated market maker (AMM) trading bot which will act as ‘dumb-money’ incentivizing user participation. All funds deployed will be sent from a non-American citizen in a jurisdiction which Polymarket operates legally in. Due to Polymarkets decentralized architecture and non-custodial wallets this process can be achieved while maintaining sole custody of funds. Markets will be resolved using Polymarket’s decentralized oracle system based on the UMA optimistic oracle.
Polymarket is the optimal partner for this project for a number of reasons. Polymarket is a decentralized platform and therefore there is minimal risk of grant or user funds being stolen or improperly accessed by the Polymarket team. Polymarket is also by far the largest prediction market with thousands of users and millions of dollars of volume traded. Furthermore, Polymarket does not collect any fees on transactions, deposits, or withdrawals. While they do not have the resources to execute this project themselves, they are highly interested and willing partners.
Total Grant Request: $8,000 - These funds would be deployed through AMMs in order to boost liquidity and act as the ‘dumb money’ in the market. Funds will be used until they are exhausted.
The idea of using prediction markets to create forecasts on high-impact topics is not novel, however past and current projects have been limited to small academic forecasting competitions such as Tetlock’s Good Judgment Project and play money markets such as Manifold. While these projects are not without merit, they have several disadvantages relative to real money markets.
Incentivize Research
Real-money markets create incentives for users to investigate questions thoroughly in order to maximize profits.
Accuracy
Real-money markets have been shown to be more accurate than play-money markets. This is simply a product of incentives. Prediction markets achieve accuracy by rewarding participants for adding information to the system. When financial rewards are absent, participants have little reason to divulge valuable private information.
Reward top Forecasters
This system of financial incentives not only generates greater accuracy in individual markets but inevitably cultivates a much stronger pool of forecasters than a play-money system would. On a play money platform bad forecasters can continue to publish under-researched and inaccurate forecasts indefinitely. Meanwhile, real money markets quickly weed out poor forecasters while strong forecasters and users with private information are rewarded for their knowledge. Metaculus, which is not a prediction market, has found technical solutions to this problem of bad forecasters distorting forecasts although it persists on play money prediction markets such as manifold.
Real money markets would allow top forecasters such as the Samotsvety team to profit from their forecasting talent. There is demonstrated interest by the Samotsvety team in participating in real money prediction markets, and they have currently profited close to $19,000 trading on Insight Prediction alone.
Grow the Forecasting Community
The possibility of profiting off of one’s forecasting ability will also further expand the forecasting community beyond its current status as a niche group of academics and effective altruists. Currently the forecasting community is incredibly homogenous; being almost exclusively composed of highly educated white men. The prediction market and gambling community is considerably larger and more diverse. By subsidizing participation, this project can introduce EA causes to new groups of people and help kick-start a self-sufficient ecosystem of forecasters interested in trading on geopolitics and technology.
Further test the accuracy of Real vs Play Money Predictions
There is a decent literature on this topic, however the data used to test this question is limited to sports and election trading/betting. Most notable is E. S. Rosenbloom & William Notz’s 2006 paper which found that real money markets were more accurate than play money markets when the odds of the event were not widely known. This is significant as it suggests that real money markets may be particularly useful for forecasting high impact events such as AI Development and geopolitics, which are harder to model than more frequent and studied events such as sports and elections. By creating liquid prediction markets on topics that are most relevant to EA cause areas, this experiment will provide valuable test data which can help inform how future investments in forecasting are allocated.
There are many factors that suppress prediction market participation. Prediction markets are zero-sum by nature and are in fact negative sum when one takes into account opportunity cost, transaction costs, and risk (this is particularly true for markets with longer time horizons). Furthermore, users could be subject to information asymmetries. No one wants to be the fish trading against sharks with private information. These factors in aggregate, along with the challenge of using crypto prediction markets such as polymarket have severely limited user participation.
Past attempts to create real-money prediction markets have failed due to regulatory impediments and perceived reputational risk. Platforms such as Intrade and Predictit have been shuttered by U.S. regulators. Meanwhile, academics such as Tetlock, Hannania, and Hanson have avoided using offshore prediction markets to conduct research. The recent advent of cryptocurrency however has allowed for the creation of a number of decentralized crypto exchanges which are safe and trusted by thousands of users.
I am the former Head of Markets at Insight Prediction where I developed some of the first real money markets on geopolitical topics. During my time as Head of Markets I ran markets that generated over a million dollars in volume including the market "Will Russia Invade Ukraine in 2022" which reached $400,000 in volume. I have a deep interest in Effective Altruism, forecasting, and political risk. My goal is to better understand the relationship between markets and geopolitics. You can reach me at ezra.brodey1@gmail.com. I love feedback!
Critique: How much does seeding an AMM subsidize trading? As in if I deposit $10k in an AMM, how are the rewards from this spread out amongst traders? My intuition is the first trader gets some huge fraction of the value, and then further traders are getting a very small fraction. (vs other schemes to subsidize markets, eg maker fee rebate, or volume rebates etc)
Answer: This is a very strong critique! AMMs are not the optimal way of subsidizing prediction markets. Personally, I think paying interest on shares owned would be the best, however there a lot of technical challenges which make this pretty difficult. In terms of who would profit from the AMM, I would set it up so that initial prices were as close to accurate as I could make them. The early users would probably profit more on average than later users but the real winners would be those who took advantage of large swings in probability later on. This is not optimal, and I would much rather have a more even distribution of returns to traders. Rather than thinking about the AMM as subsidizing specific users I think it’s more helpful to think about it in terms of making the market useable. Users hate leaving limit orders and this often leads to markets having low liquidity and large spreads which makes trading impossible even when two users would hypothetically trade against each other. The AMM also gives Polymarket an incentive to run the market. They reasonably don’t want to have a bunch of low liquidity markets up because this looks bad and so this would solve that problem. $8,000 isn't going to create the perfect prediction market but it can still go a long way!
Critique: Prediction markets are illegal. Why would you want to do something which might tarnish EA's already bruised reputation?
Answer: Prediction Markets are mostly illegal in the U.S. but they aren't illegal everywhere. Globally, they mostly operate in regulatory grey area (much like ACX mini-grants). Polymarket specifically has a sterling reputation of complying and working with American regulators and they have a top notch legal team! Furthermore these subsidies will be deployed by a trusted non-U.S. citizen with supervision of Polymarkets legal team.
I also think that EA's tarnished reputation has a lot more to do with immoral activities than illegal activities. Yes, the SBF fiasco was almost definitely illegal, but the recent racism incident with Bostrom and the Scottish Castle incident weren't illegal, just immoral/hypocritical.
Critique: Why would you bet on X-risk in a real money prediction market, if everyone's dead there is no point in having money.
Answer: This is true. I will not be running markets on existential risk scenarios but rather on questions which inform them. Eg. "Will Iran Test a Nuclear Weapon by 2024?" NOT "Will the U.S. Enter a Nuclear Exchange with Russia?".
Critique: People don't need monetary incentives to reveal private information. People release classified info on the War Thunder forum all the time.
Answer: I'm not arguing that people literally never reveal private information, but prediction markets give people an important monetary incentive towards revealing private info. An OpenAI employ with secret information about AGI progress is probably not going to sign up to Metaculus or Manifiold just to correct the forecasts but they might sign up for polymarket if they can make a quick $8,000.
Critique: Prediction markets are never going to work. Even at the height of crypto-mania these markets did not take off. It’s better to just to stick to non financial prediction platforms than try and use prediction markets.
Answer: Prediction markets have been growing rapidly over the past 2 years. Polymarket alone currently has millions of dollars in liquidity and volume and thousands of users. I'm not sure of exact numbers but I think it's safe to say that there are far more prediction market users than users of forecasting platforms.
Critique: $8,000 isn't enough to accomplish anything.
Answer: I agree that this project could be more impactful with a larger grant. Think of this as an experiment to test (just like ACX mini-grants) how to do subsidized prediction markets.