Learning in repeated multiple unit combinatorial auctions: An experimental study

The motivation of this paper is to understand trader behaviour and learning in a complex setting where finding a best strategy might not be intuitive. The assertion made is that feedback information can help in updating strategies through repeated bidding processes. The paper explores this assertion through the results of a series of repeated multiple unit combinatorial auction laboratory experiments where item and package traders interact under three information treatments: 1) basic information feedback on market prices and status of their own bids; 2) basic information feedback and all winning bids; and 3) market prices and the status of all bids. We compare bidding behavior with a local optimal package selection model. We then estimate an experience weighted attraction learning (EWA) model of bidding behavior. We observe that package traders follow price feedback information more closely than item traders, especially in the basic treatment information. With additional information package traders substantially deviate from best response bidding strategy resulting in a loss of efficiency. Finally, item traders tend to remember their past experiences more than package traders in low information environments. In high information environments the trend is reversed. The implications of this study could be significant for market design. The standard assumption that more information in combinatorial market design is better for traders may not hold in all cases.