We consider one-object private value auctions in which at the time of the auction bidders do not know their ex post values for the object but have beliefs about these. Bidders observe from past similar auctions ex post values and bids but not the beliefs held at the time of the auction. A share of bidders -referred to as data-driven- form their theory about other bidders’ behaviour simply based on such available data while the remaining share is rational. We first study simple auctions such as second-price or first-price auctions and note than when private information is correlated among bidders, data-driven bidders rely on a mis-specified model leading them to choose their bidding strategy similarly as in auction models with interdependent values. After noting that inefficiencies typically arise both in first-price and second-price auctions, we show that inefficiencies are inevitable in all auction-like mechanisms for generic distributions of private information (exhibiting correlation). Whether efficiency can be achieved in more general mechanisms is briefly discussed.

Please sign up for meetings here: docs.google.com/spreadsheets/d/1UnW-JOf7vD54h9QKWYWXQS1Wc-H4oP7nbywDdeJvr4o/edit#gid=0