Parts | Days | Selection | Search | Downloads | Help

AKSOE: Arbeitskreis Physik sozio-ökonomischer Systeme

AKSOE 8: Economic Models and Evolutionary Game Theory II

AKSOE 8.3: Talk

Tuesday, March 27, 2007, 15:00–15:30, H8

Evolutionary learning in auctions — •Konrad Richter — Gardegasse 3/9 1070 Vienna; konrad_richter@mckinsey.com

Current auction theory relies crucially on the assumption that all bidders are perfectly rational and therefore bid homogeneously according to their Nash Equilibrium bidding strategies. This paper investigates computationally via an Agent Based Model whether evolutionary learning - in particular Best Response Learning - in repeated auctions could justify this assumption of NE bidding.

The simulations show that evolutionary Best Response learning does only lead for auction formats with dominant strategies to the NE. In general, however, the NE is not reached. Instead, repeated auctions show non-trivial dynamics such as clustered volatility and autoregressive behavior. This dynamics leads to greater risks for sellers and buyers and to a suboptimal allocation of goods.

In conclusion, the paper argues that auction theory could benefit from focusing more on the dynamics of repeated auctions than only on the properties of Nash Equilibria and highlights some potential future research fields