Cluster Detection in Laboratory Auction Data: A Model-Based Approach

Abstract

The benchmark risk-averse equilibrium model does not explain some of the outcomes obtained in experiments with first-price auctions. Nonetheless, the presence of non-linear bidding and the wide dispersion of bids have received little attention in the literature. I focus on these issues and revisit previous laboratory evidence with the help of model-based clustering techniques. The rejection of equilibrium models is found to be mostly due to the significance of non-linear bidding rules and the unexplained heterogeneity. With the use of a mixture model, the observations are classified into four groups or clusters. Significant differences between individuals and clusters are found, but so is a persistent within individual variation, which leads us to conclude that subjects do not commit to one particular bidding strategy and alternate across several processes.

Key words: Clustering, Experimental Auctions, Non-linear bidding.
JEL: C91, D44, D03.