Abstract
A standard result from auction theory is that bidding truthfully in a second price auction is a weakly dominant strategy. The result, however, does not apply in the presence of Cost Per Action (CPA) constraints. Such constraints exist, for instance, in digital advertising, as some buyer may try to maximize the total number of clicks while keeping the empirical Cost Per Click (CPC) below a threshold. More generally the CPA constraint implies that the buyer has a maximal average cost per unit of value in mind. We discuss how such constraints change some traditional results from auction theory. Following the usual textbook narrative on auction theory, we focus specifically on the symmetric setting. We formalize the notion of CPA constrained auctions and derive a Nash equilibrium for second price auctions. We then extend this result to combinations of first and second price auctions. Further, we expose a revenue equivalence property and show that the seller's revenue-maximizing reserve price is zero. In practice, CPA-constrained buyers may target an empirical CPA on a given time horizon, as the auction is repeated many times. Thus, his bidding behavior depends on past realization. We show that the resulting buyer dynamic optimization problem can be formalized with stochastic control tools and solved numerically with available solvers.
Introduction
Standard auction theory, exemplified by the Vickrey auction, assumes bidders are utility maximizers seeking to maximize the surplus between their private value and the final price. However, the digital advertising landscape operates under different constraints. Advertisers often manage performance targets, such as a maximum average Cost Per Click (CPC) or Cost Per Order (CPO).
When a buyer is constrained by an average cost across a high volume of auctions rather than a maximum price for a single event, traditional bidding strategies become suboptimal. This research addresses how these CPA constraints fundamentally alter bidding behavior, Nash equilibria, and optimal auction design for the seller.
Methodology
The study employs a combination of game theory and stochastic control to model the CPA-constrained environment:
- Static Modeling: The bidder's optimization problem is formalized as maximizing expected value subject to an ex ante (average) cost constraint. Lagrangian multipliers are used to derive the optimal "best reply" strategy.
- Equilibrium Analysis: A symmetric Nash equilibrium is derived for second-price auctions and generalized to include first-price and "soft-floor" auction formats.
- Stochastic Control: Given that auctions are repeated over time, the Hamilton-Jacobi-Bellman (HJB) equation is applied to model how a buyer should dynamically adjust their bidding multiplier based on past performance to satisfy their CPA target by the end of a campaign.
- Numerical Simulation: The dynamic problem is solved using the
BocopHJBsolver, utilizing synthetic distributions (e.g., uniform, exponential, and power-law) to validate theoretical findings.
Theoretical Insights
The Bid Multiplier Strategy
In a second-price auction with CPA constraints, bidding one's true value is no longer optimal. Instead, the optimal strategy involves a linear bid multiplier: $b(v) = \alpha v$. This provides a theoretical foundation for the "bid multipliers" widely used in ad-tech. The multiplier $\alpha$ is determined by the target CPA ($T$) and a competition factor ($\gamma$).
The Zero Reserve Price Paradox
In standard auctions, sellers utilize reserve prices to extract more revenue. However, in a CPA-constrained environment, the optimal reserve price for the seller is zero. Since the constraint applies to the average cost, any reserve price merely reduces total volume and social welfare without increasing the seller's expected revenue.
Dynamic Adaptation and Results
As the number of bidders increases, the competition factor $\gamma$ tends toward 1, causing second-price auctions to behave similarly to first-price auctions regarding final payments. Furthermore, the paper demonstrates that revenue equivalence holds: whether using first-price, second-price, or mixed formats, the seller's expected revenue remains constant under CPA constraints.
In repeated auction settings, a stochastic controller proves significantly more robust than a static multiplier. When cost spikes occur, the stochastic controller dynamically reduces the bid multiplier toward the end of the time horizon to ensure the average CPA constraint is met.
Limitations and Future Work
The current model assumes symmetric bidders with identical CPA targets and value distributions. Preliminary analysis of asymmetric settings suggests more complex dynamics, where aggressive buyers might attempt to force competitors to violate their constraints. Additionally, the model assumes independent private values, whereas real-world advertiser values for specific users are often correlated. Future research could explore these asymmetric dynamics and the impact of CPA-constrained bidding on long-term market stability.
Summary updated