On the convergence and robustness of reserve pricing in keyword auctions

Abstract

Reserve price becomes a critical issue in mechanism design of keyword auctions mostly because of the potential revenue increase brought up by it. In this paper, we focus on a sub-problem in reserve pricing, that is, how to estimate the bids distribution from the truncated samples and further calculate the optimal reserve price in an iterative setting. To the best of our knowledge, this is the first paper to discuss this problem. We propose to use maximum likelihood estimate (MLE) to solve the problem, and we prove that it is an unbiased method for distribution estimation. Moreover, we further simulate the iterative optimal reserve price calculating and updating process based on the estimated distribution. The experimental results are interpreted in terms of the robustness of MLE to truncated sample size and initial reserve price (truncated value), and the convergence of subsequent optimal reserve price in the iterative updating process is also discussed. We conclude that MLE is reliable enough to be applied in real-world optimal reserve pricing in keyword auctions.