24 May 2024 Search

Scott Alexander writes, about bots on prediction markets,

Most of these bots are boring. They’re bots programmed to automatically buy some market once the price gets low enough, or to arbitrage basically-identical markets, or do some other technical finance maneuver.

From the point of view of a active prediction market, where a lot of well-informed traders are speculating about well-known events, then those bots are not especially interesting.

The place where the boring bots do make a big difference, though, is in incentivization markets.

An incentivization market is like a prediction market except that one trading strategy is to make an event being traded on either happen or not happen.

Some markets can be both. (A use case for Policy Analysis Market would have been for someone with advanced knowledge of a terrorist attack to trade on their knowledge, and create a price signal that could prevent the attack.)

One problem for incentivization markets to get over is the large number of thinly traded contracts. So a boring bot would be just what you need for things like:

  • trade across likely duplicate and dependent issues to create fewer and more lucrative opportunities for human experts (can be within or across projects)

  • bid up the price of FIXED based on encouraging CI results, enabling developers to get out of all or part of a position early

  • front-run issues on behalf of a developer based on their interests and available time

The market helps compensate for erratic LLM behavior—an unproductive bot will lose its stake and get shut down. A bot doesn’t have to be run for max earnings, either. An arbitrage bot, for example, could break even or get subsidized to lose a little to keep the market smooth.

Incentivization market needs more noisy traders, LLMs need a cheap way to evaluate whether they’re doing something sensible. Seems like a cookies and milk situation. More: How we get to the end of prediction market winter