Designing online auctions with past performance information

We investigate the value of past performance information in the context of keyword advertising auctions, where advertisers differ both in valuation-per-click and in the numbers of clicks they can generate (their performance). We focus on weighted unit-price-contract (UPC) auctions, in which bidders bid unit prices and pay accordingly if they win, and their bids are weighted by factors based on their own past performance. We characterize the efficient and the revenue-maximizing weighting factors and apply our framework to study Yahoo!'s and Google's auction designs, each of which can be viewed as a special case of weighted UPC auctions.