Robust Clearing Price Mechanisms for Reserve Price Optimization

TL;DR

This paper introduces robust reserve price mechanisms for repeated auctions that balance revenue maximization and bidder incentive compatibility, using noise-based techniques inspired by differential privacy, with theoretical guarantees and empirical validation.

Contribution

It proposes two novel, computationally efficient reserve price setting methods that are robust to bidder misreports and provide theoretical trade-offs between revenue and incentives.

Findings

The mechanisms effectively balance revenue and incentive compatibility.

Theoretical guarantees quantify trade-offs between revenue and bid-shading.

Empirical results validate the robustness and effectiveness of the proposed methods.

Abstract

Setting an effective reserve price for strategic bidders in repeated auctions is a central question in online advertising. In this paper, we investigate how to set an anonymous reserve price in repeated auctions based on historical bids in a way that balances revenue and incentives to misreport. We propose two simple and computationally efficient methods to set reserve prices based on the notion of a clearing price and make them robust to bidder misreports. The first approach adds random noise to the reserve price, drawing on techniques from differential privacy. The second method applies a smoothing technique by adding noise to the training bids used to compute the reserve price. We provide theoretical guarantees on the trade-offs between the revenue performance and bid-shading incentives of these two mechanisms. Finally, we empirically evaluate our mechanisms on synthetic data to validate our theoretical findings.