Sponsored Search Auctions with Markovian Users

Sponsored search involves running an auction among advertisers who bid in order to have their ad shown next to search results for specific keywords. Currently, the most popular auction for sponsored search is the “Generalized Second Price” (GSP) auction in which advertisers are assigned to slots in the decreasing order of their “score,” which is defined as the product of their bid and click-through rate. In the past few years, there has been significant research on the game-theoretic issues that arise in an advertiser’s interaction with the mechanism as well as possible redesigns of the mechanism, but this ranking order has remained standard. From a search engine’s perspective, the fundamental question is: what is the best assignment of advertisers to slots? Here “best” could mean “maximizing user satisfaction,” “most efficient,” “revenue-maximizing,” “simplest to interact with,” or a combination of these. To answer this question we need to understand the behavior of a search engine user when she sees the displayed ads, since that defines the commodity the advertisers are bidding on, and its value. Most prior work has assumed that the probability of a user clicking on an ad is independent of the other ads shown on the page. We propose a simple Markovian user model that does not make this assumption. We then present an algorithm to determine the most efficient assignment under this model, which turns out to be different than that of GSP. A truthful auction then follows from an application of the Vickrey-Clarke-Groves (VCG) mechanism. Further, we show that our assignment has many of the desirable properties of GSP that makes bidding intuitive. At the technical core of our result are a number of insights about the structure of the optimal assignment.

💡 Research Summary

The paper addresses a fundamental limitation of the widely used Generalized Second Price (GSP) auction for sponsored search: the assumption that each ad’s click‑through probability is independent of the other ads displayed on the page. In reality, users scan ads sequentially, and the presence of a high‑quality ad in an upper slot can dramatically reduce the chance that a lower‑slot ad will ever be seen or clicked. To capture this inter‑dependence, the authors introduce a Markovian user model. In this model a user starts at the top slot, and for each ad i she either clicks with probability α_i (which depends on the ad’s intrinsic quality and its position) or, if she does not click, moves to the next slot with probability q_i. The process terminates either when a click occurs or when the user decides to stop browsing. This sequential stochastic process is a simple Markov chain that directly links the probability of a click on any given ad to the product of the “survival” probabilities of all ads above it.

Under this model the expected social welfare (the sum of advertisers’ values for clicks) is no longer a simple sum of independent terms. Instead, the contribution of an ad placed in slot s is multiplied by the product of the q‑values of all ads in higher slots, reflecting the diminishing exposure of lower ads. Consequently, the classic GSP ranking rule—ordering advertisers by the product of bid and estimated click‑through rate (CTR)—does not maximize expected welfare.

The authors formulate the welfare‑maximization problem as a combinatorial optimization: assign a subset of n advertisers to k slots so that the sum of expected values is maximized. They show that the problem admits a dynamic programming (DP) solution with a clear structural property. Define a “efficiency‑to‑transition ratio” for each advertiser

\