TL;DR
This paper argues that Pay-Per-Action (PPA) advertising systems, while seemingly a solution to the drawbacks of Pay-Per-Click (PPC), introduce new vulnerabilities. Specifically, the ability for advertisers to define multiple actions and manipulate action reporting creates incentives for "skewed bidding," where advertisers strategically overbid on actions the platform might underestimate, leading to market inefficiencies and reduced benefits of PPA.
What Problem Does This Paper Address?
Online advertising platforms primarily use Pay-Per-Click (PPC), where advertisers pay for each click on their ad. However, PPC has limitations: not all clicks are equally valuable (e.g., a click on a hotel ad for a specific city might not lead to a booking), estimating conversion rates can be difficult for niche searches, and click fraud is a persistent issue where publishers might artificially inflate clicks for revenue. Pay-Per-Action (PPA) advertising was proposed as a solution, where advertisers pay only when a specific, valuable action (like a purchase or sign-up) is completed on their website. This paper investigates potential problems with PPA systems that arise from their design, particularly concerning advertiser bidding strategies and market efficiency.
Methods
The authors employ a stylized theoretical model of an advertising auction to analyze the incentives of advertisers in a PPA system. They consider a second-price auction for sponsored links, where the winning bidder pays based on the second-highest aggregate bid. The core of their analysis involves modeling advertiser utility and deriving equilibrium bidding strategies under various assumptions about action probabilities, advertiser knowledge, and the platform's estimation capabilities. They introduce the concept of "skewed bidding," where advertisers strategically adjust their bids based on discrepancies between their own estimates of action probabilities and the platform's estimates. The paper also explores scenarios with multiple actions, maximum per-action bids, and risk aversion, as well as the impact of endogenous action probabilities (where firms can manipulate them).
Datasets
Not discussed in the paper.
Key Insights and Ideas
- PPA's "Holy Grail" has hidden flaws: While PPA aims to align payments with valuable outcomes, the paper reveals that its design can lead to strategic manipulation by advertisers.
- Skewed Bidding is the core issue: When advertisers have better information about action probabilities than the platform, they have an incentive to "over-bid" on actions the platform underestimates. This allows them to win auctions with lower effective payments relative to their true valuation.
- Multiple Actions amplify manipulation: Allowing advertisers to define and bid on multiple actions significantly increases the complexity and potential for strategic behavior, as they can exploit differences in platform estimates across these actions.
- Endogenous Action Probabilities are a major threat: The ability for firms to manipulate action reporting (e.g., by temporarily disabling links or faking reports) is a critical vulnerability that can undermine the entire PPA system, even with sophisticated fraud-detection algorithms.
- Market design matters: The paper suggests that restrictions on bidding (e.g., maximum per-action bids) or limiting PPA to a single action can mitigate some of these issues, but they may not fully restore the intended benefits of PPA.
Key Results
- The paper demonstrates that advertisers have an incentive to engage in "skewed bidding" when their estimates of action probabilities differ from the ad platform's estimates. This can lead to advertisers overbidding on actions that the platform has underestimated.
- This skewed bidding can result in market inefficiencies, favoring firms that are better at gaming the system over those with higher intrinsic valuations.
- The ability of firms to manipulate action probabilities (e.g., by destroying links or misreporting actions) can lead to significant revenue loss for the platform and undermine the core principle of PPA.
- While risk aversion can mitigate some of these bidding distortions, it can also introduce new inefficiencies and increase the cost of participation.
- The authors suggest that PPA systems are likely to be more successful when restricted to a single action or when specific restrictions on bid and action rate variations are implemented.
Limitations and Future Work
The authors acknowledge that their model is stylized and simplifies real-world complexities. For instance, they assume risk-neutral firms for much of the analysis, though they briefly explore risk aversion. The paper also notes that a full analysis of equilibrium behavior with endogenous action probabilities is beyond its scope. Future work could involve developing more realistic models that incorporate dynamic bidding, learning, and more sophisticated platform estimation algorithms. The authors also suggest that practical implementations of PPA might need to restrict the number of actions or impose temporal constraints on bidding and reporting to maintain efficiency and prevent manipulation.
Summary updated