Introduction
In the landscape of digital advertising, the transition from manual bidding to automated strategies has fundamentally altered the dynamics of the auction house. Traditionally, auction theory has championed Marginal CPA (mCPA) bidding—where an advertiser bids their exact marginal value for a conversion—as the gold standard for profit maximization. However, in practice, many of the world's largest advertisers opt for Target CPA (tCPA), a strategy where they set a desired average cost and allow an algorithm to maximize conversion volume.
While classical theory often dismisses tCPA as sub-optimal or attributes its popularity to "cognitive ease," this paper provides a rigorous game-theoretic justification for its dominance. The authors argue that tCPA serves as a strategic defense mechanism in environments where the auctioneer cannot credibly commit to fixed auction rules.
The Problem: The Lack of Auctioneer Commitment
In a theoretical "with commitment" scenario, an auctioneer (such as Google or Meta) sets reserve prices and auction rules in advance and adheres to them regardless of the bids received. In reality, the digital advertising ecosystem often lacks this level of transparency. An auctioneer without commitment can observe submitted bids and then "greedily" adjust reserve prices to extract the maximum possible surplus from the bidder.
If a bidder uses mCPA and reveals their true marginal value, a greedy auctioneer can simply set the reserve price at that exact value, effectively reducing the bidder's profit to zero. This creates a "trap" for the advertiser, leading to a breakdown in auction efficiency.
Theoretical Framework
The authors analyze this dynamic using a multi-stage game-theoretic model, comparing mCPA and tCPA under two conditions: commitment and non-commitment. The study evaluates a Single-Bidder Model (facing an exogenous price landscape) and a General Model (incorporating multiple competitors). Central to their analysis is the concept of a credible equilibrium—a state where the auctioneer has no incentive to alter the rules ex-post, even when they possess the power to do so.
Analysis of Bidding Formats
The mCPA Trap
Under mCPA bidding without auctioneer commitment, the bidder anticipates that the auctioneer will set reserves to match their marginal bid. To protect their margins, the bidder is forced to "shade" their bids—submitting values significantly lower than their true valuation. This behavior causes the auction to function like a First-Price Auction, which is notoriously prone to inefficiencies. The result is fewer conversions for the bidder and lower overall market efficiency.
The tCPA Shield
In contrast, tCPA bidding changes the auctioneer's incentives. Under tCPA, the bidder's marginal bid is not static; it is dynamically adjusted by an auto-bidder to satisfy an average cost constraint. The authors demonstrate that in this format, the auctioneer’s revenue becomes proportional to the total volume of queries sold.
To maximize their own revenue, the auctioneer is incentivized to lower reserve prices to increase volume, rather than raising them to extract per-unit rent. This aligns the auctioneer’s profit motive with the bidder’s desire for conversion volume, creating an "endogenous credibility" where the auctioneer naturally gravitates toward a zero-reserve policy.
Key Results
- Bidder Preference: The paper proves (Theorems 1 & 3) that in any equilibrium where the auctioneer lacks commitment, a profit-maximizing bidder strictly prefers tCPA over mCPA. The tCPA format effectively prevents rent extraction.
- Auctioneer Revenue: Interestingly, the auctioneer also benefits. Under common market conditions (Theorem 2), the increased volume generated by tCPA bidding outweighs the loss from lower per-unit prices, leading to higher total revenue compared to mCPA.
- Market Efficiency: tCPA leads to welfare-optimal outcomes. Because bidders do not need to shade their bids to protect their profit, queries are more likely to be allocated to the advertisers who value them most.
- Mitigating Revenue Loss: The authors quantify the revenue loss suffered by auctioneers who cannot prove commitment, showing that the adoption of tCPA-like mechanisms can significantly mitigate these losses by restoring trust in the auction's fairness.
Conclusion
By shifting the focus from marginal values to average cost constraints, tCPA bidding creates a self-enforcing environment of transparency and efficiency. It acts as a "strategic shield" for advertisers, ensuring that even in the absence of formal commitment from the auctioneer, incentives remain aligned. This research suggests that the industry-wide shift toward auto-bidding is not just a matter of convenience, but a sophisticated response to the structural realities of modern digital auctions.
Summary updated