Screening with Advertisements

Technical Summary: Screening with Advertisements

Problem Statement
This paper investigates a seller's revenue-maximizing mechanism in a setting where a desirable good is sold alongside an undesirable "bad" (specifically, advertisements). The seller generates revenue from two sources: direct payments from the buyer and third-party payments (from advertisers) whenever the bad is allocated. The buyer possesses two-dimensional private information: their valuation for the good ($x_1$) and their disutility from the bad ($x_2 \leq 0$). The magnitude $|x_2|$ represents the buyer's willingness to pay to avoid the bad.

The core challenge is a multidimensional screening problem where the allocation of the "bad" generates external compensation. This fundamentally alters the seller's incentives compared to standard multi-good models, as one dimension represents a "bad" rather than a "good." The paper seeks to characterize the optimal mechanism among three commonly used pricing structures:

1. Good-Only Posted Price: Selling the good without ads.
2. Single-Bundle Posted Price: Selling a bundle of the good and ads.
3. Ad-Tiered Posted Price: Offering a menu with an ad-supported tier and an ad-free premium tier.

Methodology
The analysis employs the duality framework established by Daskalakis, Deckelbaum, and Tzamos (2017) (DDT), originally developed for multi-good settings. The seller's revenue maximization problem is reformulated using a transformed measure $\mu$, derived from the buyer's type density and the third-party payment structure.

Key methodological steps include:

• Transformation: The objective function (expected revenue) is expressed as an integral of the buyer's indirect utility $u$ against a transformed measure $\mu$. The introduction of third-party payments modifies $\mu$ by adding terms dependent on the payment $k$.
• Duality: The paper relies on DDT's strong duality result, which states that a finite-menu mechanism is optimal if and only if a specific integral inequality holds for all convex, monotone test functions compatible with the menu's allocation structure.
• Simplification: Recognizing that verifying inequalities against all convex functions is computationally intractable, the authors introduce a new class of tractable, interpretable two-dimensional orthant conditions. These conditions serve as sufficient tests for optimality by restricting the verification to specific geometric regions (orthants) of the type space.
• Assumptions: The analysis assumes a constant third-party payment $k$ (independent of buyer type) and imposes a mild regularity condition on the density function $f$ (the McAfee-McMillan or "MM" condition), adapted to include the third-party revenue term.

Key Results
The paper provides a near-characterization of optimality for the three canonical mechanisms, establishing necessary conditions and sufficient conditions (via orthant tests) for each.

1. Good-Only Posted Price:

   • Necessity: Optimal only if the third-party payment $k$ is low (specifically $k \leq |x_2|$). The transformed measure $\mu$ must assign zero mass to the regions of buyers who purchase and those who do not, and the marginal distribution of $\mu$ along the good dimension must satisfy specific integral inequalities.
   • Sufficiency: Requires an additional orthant condition ensuring non-negative mass on specific lower-right orthants within the purchasing region.

2. Single-Bundle Posted Price:

   • Necessity: Optimal only if $k$ is high (specifically $k \geq |x_2|$). The sign pattern of $\mu$ along the boundaries of the type space must align with the bundle allocation.
   • Sufficiency: Requires an orthant condition ensuring non-negative mass on upper-right orthants within the purchasing region.

3. Ad-Tiered Posted Price:

   • Necessity: Can be optimal for a broader range of $k$. It requires zero mass on the three distinct regions (non-purchasers, ad-tolerant purchasers, and ad-averse purchasers) and specific integral conditions on the marginal measures.
   • Sufficiency: Requires orthant conditions for both the ad-tolerant and ad-averse purchasing regions.

Comparative Statics and Economic Interpretation
The paper derives a clean comparative static based on the magnitude of the third-party payment $k$:

• Low $k$: The "Good-Only" regime is optimal. Ad exposure is unprofitable; the seller excludes the bad entirely to extract surplus from ad-averse users.
• Intermediate $k$: The "Ad-Tiered" regime is optimal. Third-party revenue becomes significant enough to make ad exposure profitable for tolerant users, while ad-averse users are still served via a premium ad-free tier. This regime bridges the other two.
• High $k$: The "Single-Bundle" regime is optimal. Advertising revenue dominates; the seller bundles ads for all types, as the opportunity cost of foregone ad revenue outweighs the benefit of offering an ad-free option.

When the type distribution is uniform in the bad dimension, these regimes partition the parameter space strictly: $k \leq |x_2|$ implies Good-Only; $|x_2| < k < |x_2|$ implies Ad-Tiered; and $k \geq |x_2|$ implies Single-Bundle.

Significance and Contributions
The paper claims several theoretical and methodological contributions:

• Qualitative Departure from Standard Models: Unlike standard multi-good models where the sign pattern of the transformed measure $\mu$ is fixed by the type distribution, this model shows that the sign pattern of $\mu$ shifts systematically with the third-party payment $k$. This creates a transparent geometric evolution of the optimal mechanism.
• Tractable Sufficiency Checks: The paper addresses the computational intractability of the general DDT conditions (which require testing against all convex functions). By introducing orthant conditions, the authors provide a practical, easily verifiable set of sufficient conditions that preserve geometric and economic intuition.
• Handling "Bads": The analysis extends multidimensional screening theory to environments containing a "bad" dimension with external compensation, demonstrating how this specific feature reshapes incentive constraints and optimal allocation geometry.
• Unified View of Hybrid Monetization: The results offer a unified theoretical explanation for the prevalence of hybrid monetization strategies (Good-Only, Ad-Tiered, Single-Bundle) in digital markets, linking the choice of mechanism directly to the level of third-party monetization.

The authors note that while the sufficiency conditions are strong, they are not necessary in fully two-dimensional regions due to the complexity of the cone of convex functions. However, for distributions with closed-form densities, these tests provide a transparent and computationally simple alternative to numerical optimization.