Two Pricing Mechanisms in Sponsored Search Advertising

Wei Yang, Youyi Feng, Baichun XiaoGames 2013 · DOI

TL;DR

This paper analyzes two pricing mechanisms for sponsored search advertising: the Generalized Second-Price (GSP) auction and a novel "Auction with Posted Reserve Price" (APR). The authors find that while GSP generally yields higher revenue, APR offers practical advantages and can be optimized to approach GSP's revenue. They also demonstrate that the optimal reserve price in APR is influenced by market dynamics like the number of advertisers and positions, unlike in GSP.

What Problem Does This Paper Address?

Sponsored search advertising, a major revenue source for search engines, relies on auctions to sell ad placements. Search engines use reserve prices (minimum bids) to influence advertiser bidding and maximize revenue. However, the optimal reserve price strategy, especially in dynamic markets with varying numbers of advertisers and ad positions, is complex. This paper addresses the question of how different pricing mechanisms, specifically GSP and a new APR mechanism, affect revenue and how to determine optimal reserve prices in these contexts. The authors aim to compare these mechanisms and understand the factors influencing optimal reserve price setting.

Methods

The paper models sponsored search advertising as a simultaneous-move, one-shot auction. They analyze two pricing mechanisms:

  • Generalized Second-Price (GSP): The winner of the lowest position pays the maximum of the reserve price and the highest losing bid.
  • Auction with Posted Reserve Price (APR): The winner of the lowest position pays a fixed posted reserve price, regardless of other bids.

The authors focus on the "lower bound of the symmetric Nash equilibrium" (SNE) for advertiser bidding strategies, which is argued to be empirically relevant. They derive expected revenue functions for both mechanisms and use calculus to find the optimal reserve price that maximizes these functions. A key assumption is that advertisers' per-click values follow a distribution with an increasing generalized failure rate (IGFR), which ensures quasi-concave revenue functions and the existence of a unique optimal reserve price.

Datasets

Not discussed in the paper. The analysis is theoretical and relies on assumed probability distributions for advertiser values and arrival rates.

Key Insights and Ideas

  • GSP vs. APR Revenue: While GSP generally generates more revenue than APR, the difference diminishes as the reserve price increases. APR has practical advantages like mitigating collusion and providing clearer bidding signals.
  • Optimal Reserve Price in GSP: The optimal reserve price in GSP is determined solely by the distribution of advertisers' private values, similar to single-item auctions.
  • Optimal Reserve Price in APR: In contrast, the optimal reserve price in APR is influenced by the number of available ad positions, the number of advertisers, and their valuation distributions. This makes it more adaptable to dynamic market conditions.
  • APR's Higher Optimal Reserve Price: The optimal reserve price in APR is generally higher than in GSP. This is because APR allows bidders to retain more information rent, leading the search engine to set a higher price to compensate for potentially lower marginal revenue.
  • IGFR Assumption: The assumption of an increasing generalized failure rate (IGFR) for advertiser values is crucial. It ensures that the revenue functions are quasi-concave, guaranteeing a unique optimal reserve price and simplifying the analysis.

Key Results

  • The paper provides analytical solutions for the optimal reserve prices in both GSP and APR mechanisms under the IGFR assumption.
  • Theorem 1 states that for GSP, the optimal reserve price satisfies the standard condition $r^* - \frac{1-F(r^*)}{f(r^*)} = 0$, which is independent of the number of bidders or positions.
  • Theorem 2 shows that for APR, the optimal reserve price $r^*$ is the solution to $r^* - [1 + \theta(r^*)] \frac{1-F(r^*)}{f(r^*)} = 0$, where $\theta(r^*)$ depends on the bidder population, number of positions, and valuation distribution.
  • Corollary 2 concludes that the optimal reserve price in APR should be set higher than in GSP.
  • Corollary 3 demonstrates that in APR, the optimal reserve price is an increasing function of the arrival rate of advertisers.
  • Figure 1 visually supports these findings, showing GSP's higher revenue and the convergence of APR's revenue as the reserve price increases.

Limitations and Future Work

  • Exogenous CTR: The model assumes click-through rates (CTRs) for each slot are exogenous. In reality, CTRs can be affected by factors like slot occupancy, making the model more complex. Future work could incorporate endogenous CTRs.
  • Static Auction Model: The analysis is based on a static auction model. The authors suggest exploring dynamic models where search engines can update reserve prices over time based on incumbent and potential new entrants.
  • Bayes-Nash Equilibria: The paper focuses on the lower bound of SNE. Investigating Bayes-Nash equilibria for both GSP and APR could provide a more complete understanding and explore whether the Revenue Equivalence Theorem applies.
  • Endogenous Number of Bidders: While the number of bidders is endogenous in the model, the analysis of optimal reserve price is done for a given number of bidders. Further research could explore how the search engine should set reserve prices considering the endogenous nature of bidder participation.

Summary updated

BibTeX
@article{yang2013two,
  author = {Wei Yang and Youyi Feng and Baichun Xiao},
  title = {{Two Pricing Mechanisms in Sponsored Search Advertising}},
  journal = {Games},
  year = {2013},
  doi = {10.3390/g4010125},
}
Figure