Predicting Bidders' Willingness to Pay in Online Multiunit Ascending Auctions: Analytical and Empirical Insights

We develop a real–time estimation approach to predict bidders' maximum willingness to pay in a multiunit ascending uniform–price and discriminatory–price (Yankee) online auction. Our two–stage approach begins with a bidder classification step, which is followed by an analytical prediction model. The classification model identifies bidders as either adopting a myopic best–response (MBR) bidding strategy or a non–MBR strategy. We then use a generalized bid–inversion function to estimate the willingness to pay for MBR bidders. We empirically validate our two–stage approach using data from two popular online auction sites. Our joint classification–and–prediction approach outperforms two other naïve prediction strategies that draw random valuations between a bidder's current bid and the known market upper bound. Our prediction results indicate that, on average, our estimates are within 2% of bidders' revealed willingness to pay for Yankee and uniform–price multiunit auctions. We discuss how our results can facilitate mechanism–design changes such as dynamic–bid increments and dynamic buy–it–now prices.