Predicting the probability of winning sealed bid auctions: the effects of outliers on bidding models

Abstract

This paper is concerned with the effect of outliers on predictions of the probability of tendering the lowest bid in sealed bid auctions. Four of the leading models are tested relative to the equal probability model by an empirical analysis of three large samples of real construction contract bidding data via all-in (in-sample), one-out and one-on (out-of-sample) frames. Outliers are removed in a sequence of cut-off values proportional to the standard deviation of bids for each auction. A form of logscore is used to measure the ability to predict the probability of each bidder being the lowest. The results show that, although statistically significant in some conditions, all the models produce rather poor predictions in both one-out and one-on mode, with the effects of outliers being generally small