Auctions Between Regret-Minimizing Agents
This addresses strategic behavior in automated auctions, with implications for auction design and AI ethics, though it appears incremental as it builds on known auction theory and regret minimization.
The paper analyzes repeated auctions where software agents use regret-minimizing algorithms, finding that in second-price auctions, players have incentives to misreport valuations, while in first-price auctions, truthful reporting is a dominant strategy.
We analyze a scenario in which software agents implemented as regret-minimizing algorithms engage in a repeated auction on behalf of their users. We study first-price and second-price auctions, as well as their generalized versions (e.g., as those used for ad auctions). Using both theoretical analysis and simulations, we show that, surprisingly, in second-price auctions the players have incentives to misreport their true valuations to their own learning agents, while in the first-price auction it is a dominant strategy for all players to truthfully report their valuations to their agents.