Project summary

Individuals, firms, and policymakers are making decisions about education, hiring, investment, retraining, and social insurance without credible forecasts of AI’s economic effects. Surveys and models are valuable, but they are episodic and rarely reward experts for acquiring information and regularly updating specific predictions.

We request $250,000 for a four-month feasibility and pilot-design effort to determine whether well-capitalized prediction markets can provide better signals of likely long-run developments in the economy caused by AI. This grant will be used to engage platforms and market makers, map legal and governance requirements, and create a working group of researchers to design prediction market contracts and to write a white paper summarizing the best approach to this project. Importantly, we will obtain estimates of the amount and nature of subsidies required for these markets to be informative and sustainable. The grant will not directly fund subsidies; instead, if this pilot work is promising, we will write up a larger grant proposal with the goal of implementing these markets. To understand the value of the pilot, it is important to understand the rough scope of the fuller grant proposal, which we estimate would cost on the order of $10 million.

What are this project's goals? How will you achieve them?

• Identify useful, measurable AI capability and economic-outcome questions.

• Design contract structures and horizons that attract informed trading.

• Estimate the amount of sponsorship required to ensure liquidity and price responsiveness.

• Produce a white paper that clearly explains what we’ve learned.

• Choose a preferred credible platform partner, as well as a market-making strategy, legal guidance, and governance pathway given the current regulatory uncertainty.

How will this funding be used?

1. Contract portfolio: Develop a variety of candidate contracts with primary and fallback data sources, resolution dates, revision and cancellation rules, a consideration of edge cases, and settlement examples. Families of markets will cover labor-market outcomes, AI capabilities, and adoption or economic passthrough.

2. Platform and partnership feasibility: Interview platforms, market-making institutions, researchers, data providers, and prospective sponsors to understand their perspectives. Compare relevant preexisting contracts on various platforms and understand risks to the project as a whole.

3. Assemble a team of collaborators to support the project. We will recruit experts on the economics of AI, prediction markets, mechanism design, and finance to assist with the design of the project. See Appendix for initial people and institutions we plan to reach out to, many of whom we’ve already spoken to in various contexts. 

4. Liquidity and subsidy design: Evaluate direct seeding, spread- and depth-of-order-book incentives, sponsorship pools, automated market makers, staged prizes, and other market design approaches.

5. Governance: Specify resolution sources, independent settlement review, how to address conflicts, and rules surrounding participant eligibility (for example, possibly excluding employees of frontier AI companies from participating in the markets).

Who is on your team? What's your track record on similar projects?

Andrey Fradkin (Associate Professor at Boston University), Ezra Karger (The Federal Reserve Bank of Chicago and the Forecasting Research Institute). We've also approached additional experts who are interested in the project.

We've both worked in large scale research projects. Andrey has co-created Webmunk (webmunk.org), which is a browser extension framework used by economics to conduct randomized control trials in digital platforms. Ezra works on constructed high-frequency indices that track economic indicators and has measured the economic effects of AI at the Forecasting Research Institute.

What are the most likely causes and outcomes if this project fails?

This project fails if:

• We can't raise enough money to subsidize prediction markets.

• If regulation makes these types of prediction markets illegal.

• If, even with incentives, there is not enough participation in the market.

• If market participants are not able to build models that substantively improve forecasts of the future economy. As a result, prediction markets provide little information over other forecasting methods.

How much money have you raised in the last 12 months, and from where?

0, we're just getting started.