Multi-Dimensional Screening with Endogenous Information Disclosure

Yang Cai, Yingkai Li, Jinzhao Wu

TL;DR

In a world where a seller can control not just the price of their products but also what the buyer knows about them, simple pricing strategies become surprisingly powerful. This paper proves that by jointly designing prices and information disclosure, a monopolist can capture at least half of the optimal revenue using a single uniform price, and can even extract the full surplus under common statistical conditions like negative correlation or symmetry.

What Problem Does This Paper Address?

Selling multiple products to a single buyer is a classic headache in economic theory, known as multi-dimensional screening. In the standard model, the buyer knows exactly how much they value each item, and the seller tries to guess those values to set prices.

The results in this standard setting are often depressing for fans of simplicity: Hart and Nisan (2019) famously showed that if values are correlated, simple "deterministic" pricing (just sticking a price tag on each item) can be infinitely worse than complex "lottery" mechanisms (e.g., "pay $10 for a 30% chance of getting Item A and a 70% chance of Item B").

This paper flips the script by making information endogenous. It asks: What happens if the buyer doesn't initially know their own values and relies on the seller to provide information (like reviews, previews, or certifications)? By controlling the "information structure," can the seller make simple pricing mechanisms competitive again?

Methods

The authors model a joint-design problem where the seller commits to two things simultaneously:

  1. A Pricing Mechanism: A menu of prices for the products.
  2. An Information Policy: A rule for what signal the buyer receives about their valuations (e.g., "You will like Product A more than Product B").

The core technical innovation is the introduction of (Coarse) Horizontal Disclosure. Instead of telling the buyer exactly how much they value an item (vertical information), the seller tells them which item or set of items they would prefer (horizontal information).

By keeping the information "coarse"—for example, telling a buyer their favorite item is in a specific subset without revealing the exact dollar value—the seller can guide the buyer toward an efficient purchase while preventing the buyer from gaining too much "information rent" (the profit a buyer keeps because the seller is in the dark).

Datasets

As a theoretical paper in mathematical economics and computer science, this work does not use empirical datasets. Instead, it employs rigorous mathematical proofs and stylized numerical examples to demonstrate revenue gaps and optimality bounds.

Key Insights and Ideas

  • Information as a "Taming" Force: The "unbounded gap" between simple pricing and optimal revenue in classic theory exists because the buyer knows too much. When the seller designs the information, they can "tame" the distribution of the buyer's posterior beliefs, making simple pricing highly effective.
  • The Power of Uniform Pricing: Surprisingly, charging the exact same price for every item (uniform pricing) is a robust strategy. It eliminates "cannibalization"—where a buyer switches to a cheaper, less-ideal product—because the seller is indifferent about which item is sold.
  • Horizontal vs. Vertical Info: The seller's goal is to reveal enough horizontal information to ensure the buyer picks the "right" product (maximizing welfare) while withholding vertical information to ensure the buyer pays close to their maximum willingness to pay (maximizing revenue).
  • Full Surplus Extraction: In many "natural" cases—such as when product values are negatively affiliated (if you like one, you're less likely to like the other) or exchangeable (symmetric)—the seller can actually capture 100% of the possible value using just simple prices and horizontal disclosure.

Key Results

  • The 50% Guarantee: For any distribution of values, there exists a uniform pricing mechanism that captures at least 50% of the optimal revenue (Theorem 1). This bound is shown to be tight.
  • Beating Uniformity: By allowing different prices for different items, the seller can guaranteed at least 50.17% of the optimal revenue (Theorem 2). While the improvement seems small, it proves that price differentiation strictly improves the worst-case guarantee.
  • Two-Product Case: For a two-product setting, simple pricing is guaranteed to capture at least 2/3 of the optimal revenue (Proposition 1).
  • Optimality Conditions: The paper identifies that for negatively affiliated or exchangeable distributions, deterministic pricing is not just "good"—it is revenue-optimal (Propositions 2 and 3).

Limitations and Future Work

  • The "Gap" remains: While the authors prove a 50.17% guarantee for multiple prices, they acknowledge this is likely not the tightest possible bound. They conjecture that the true worst-case guarantee for pricing mechanisms might be closer to 2/3 (66.7%).
  • Unit-Demand Constraint: The paper focuses on "unit-demand" buyers (those who want at most one product). Extending these results to "additive" buyers (who might want multiple items) or buyers with more complex combinatorial preferences remains an open challenge.
  • Computational Complexity: While the paper proves these mechanisms exist, it does not focus on the algorithmic efficiency of finding the optimal coarse horizontal disclosure for a given distribution.

Summary updated

Abstract

We study multi-product monopoly pricing where the seller jointly designs the selling mechanism and the information structure for the buyer to learn his values. Unlike the case with exogenous information, we show that when the seller controls information, even uniform pricing guarantees at least half of the optimal revenue. Moreover, for negatively affiliated or exchangeable value distributions, deterministic pricing is revenue-optimal. Our results highlight the power of information design in making pricing mechanisms approximately optimal in multi-dimensional settings.
BibTeX
@misc{cai2025multi,
  author = {Yang Cai and Yingkai Li and Jinzhao Wu},
  title = {{Multi-Dimensional Screening with Endogenous Information Disclosure}},
  year = {2025},
  eprint = {2502.17809},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2502.17809},
}
Figure