Technical Summary: Robust Procurement: Bayesian Design under Worst-Case Approval Constraints

1. Problem Statement

The paper addresses the problem of optimal procurement design when a Bayesian designer (an expert) must secure approval from a non-Bayesian authority (e.g., a regulator, supervisor, or senior management). The designer possesses a conjectured model of the environment, consisting of a value function $V^\star$ and a cost distribution $F^\star$, and seeks to maximize expected welfare under this model. However, the authority is skeptical of this specific model and requires that the chosen mechanism perform satisfactorily across a set of admissible models $\mathcal{A} = \mathcal{V} \times \mathcal{F}$, representing uncertainty about the true value of the good and the supplier's cost technology.

The core tension lies in the designer's need to balance efficiency (maximizing surplus under the conjectured model) with robustness (ensuring the mechanism yields the highest possible worst-case payoff guarantee to the authority). The designer's objective is to select a mechanism that maximizes the authority's worst-case payoff over $\mathcal{A}$, and among those mechanisms, choose the one that maximizes expected welfare under $(V^\star, F^\star)$. These are termed robustly optimal mechanisms.

2. Methodology and Framework

The Two-Step Optimization

The authors formalize the designer's problem as a lexicographic optimization:

1. Step 1 (Guarantee Maximization): Identify the "shortlist" of mechanisms ($\mathcal{M}_{SL}$) that maximize the welfare guarantee $G(M) = \inf_{(V,F) \in \mathcal{A}} W(M; V, F)$. The authority only approves mechanisms in this set.
2. Step 2 (Selection): The designer selects the mechanism from $\mathcal{M}_{SL}$ that maximizes expected welfare under the conjectured model $(V^\star, F^\star)$.

Environment

• Agents: A monopolistic seller with private marginal cost $\theta \in [\underline{\theta}, \bar{\theta}]$ and a buyer (designer/authority).
• Uncertainty: The set of admissible value functions $\mathcal{V}$ includes a "smallest" inverse demand function $\underline{P}$ (lowest marginal value). The set of admissible cost distributions $\mathcal{F}$ includes all CDFs supported on $[\underline{\theta}, \bar{\theta}]$, with the worst-case often being a Dirac distribution at the highest cost $\bar{\theta}$.
• Mechanisms: Initially analyzed as direct quantity mechanisms $(q, t)$, later extended to price regulations in markets with downstream demand.

Key Analytical Tools

• Worst-Case Characterization: The authors demonstrate that the worst-case welfare for any incentive-compatible (IC) and individually rational (IR) mechanism is determined by the lowest possible demand function $\underline{V}$ and the highest possible cost $\bar{\theta}$.
• Majorization Constraints: The robustness requirement translates into integral constraints on the quantity schedule $q(\theta)$. Specifically, the cumulative quantity procured from types above $\theta$ must not exceed the cumulative efficient quantity under the lowest demand, adjusted for deadweight loss.
• Ironing Techniques: When the robustly optimal mechanism deviates from the standard Baron-Myerson solution, the authors employ ironing techniques (similar to Toikka, 2011) to handle non-monotonicities in the modified virtual cost.

3. Key Results

A. Robustly Optimal Quantity Mechanisms (No Downstream Market)

When uncertainty concerns only the cost distribution (or demand is fixed), the robustly optimal mechanism is a Baron-Myerson-with-quantity-floor mechanism:

• Structure: The quantity schedule is $q^\star(\theta) = \max\{q^{BM}(\theta), q_\ell\}$, where $q^{BM}$ is the standard Bayesian optimal quantity and $q_\ell$ is the efficient quantity at the highest cost and lowest demand ($D(\bar{\theta})$).
• Implications:
  • High-Cost Sellers: Procurement is increased relative to the Bayesian optimum (up to the efficient level $q_\ell$) to protect against the possibility that high-cost firms are more prevalent than conjectured.
  • Low-Cost Sellers: Procurement remains identical to the Bayesian optimum because the welfare gain from contracting with these types exceeds the worst-case guarantee even if the cost distribution is misspecified.
  • Intermediate-Cost Sellers: If demand uncertainty is also present and the standard quantity floor mechanism fails the robustness constraints, procurement from intermediate-cost types is reduced further below the Bayesian level. This "double distortion" guards against demand overestimation.

B. Robustly Optimal Price Regulations (Downstream Markets)

When a downstream market exists, the designer can regulate prices. The robustly optimal price regulation is a Baron-Myerson-with-price-cap:

• Structure: The price is set as $p(\theta) = \min\{z^\star(\theta), \bar{\theta}\}$, where $z^\star(\theta)$ is the virtual cost under the conjectured model.
• Transfers: Transfers are contingent on realized demand to ensure ex-post individual rationality.
• Comparison: The price cap protects the buyer against unexpectedly high costs (by capping the price at $\bar{\theta}$), but it may induce over-procurement from high-cost sellers under the conjectured model if the cap is not binding in the Bayesian optimum.

C. Quantity vs. Price Regulation

The paper provides a novel comparison of regulation instruments under robustness constraints. While both instruments offer the same worst-case welfare guarantee, their performance under the designer's conjectured model differs:

• Quantity Regulation Dominates: When optimal markups (virtual costs) are large. In this case, the designer wishes to procure little from most sellers. Fixing quantities limits the buyer's exposure to demand misspecification.
• Price Regulation Dominates: When demand uncertainty is substantial (specifically, when the gap between conjectured and lowest demand is large relative to markups). Price regulation allows quantities to adjust to realized demand, avoiding the need for the designer to impose severe downward distortions on intermediate-cost types to satisfy the authority's robustness constraints.

4. Contributions and Significance

Theoretical Contributions

1. Lexicographic Robustness: The paper formalizes a specific approach to robustness where the designer uses a conjectured model to select among worst-case optimal mechanisms. This differs from approaches that treat all models symmetrically or weight them by distance from the conjecture.
2. Reshaping the Efficiency-Rent Tradeoff: The authors show that robustness fundamentally alters the standard tradeoff. It necessitates increasing procurement from high-cost types (to prevent worst-case welfare collapse) while potentially decreasing procurement from intermediate types (to guard against demand overestimation).
3. Regulation Comparison: The paper identifies the first conditions under which quantity regulation outperforms price regulation in a robust setting. This contrasts with the standard Bayesian result (where price regulation is often preferred for flexibility) and the classic Weitzman (1974) result (which relies on the relative slopes of marginal cost and benefit curves). Here, the dominance depends on the magnitude of markups relative to demand uncertainty.

Significance

The paper claims significance in providing a framework for contract design in environments where approval is conditional on robustness. This is highly relevant for:

• Public Procurement: Where government agencies must justify contracts to conservative regulators who doubt the accuracy of private cost/value estimates.
• Monopoly Regulation: Where regulators must ensure consumer protection against model misspecification.
• Auction Design: Where experts design mechanisms for entities that lack confidence in the underlying valuation models.

The authors emphasize that their approach captures the tension between efficiency and rent extraction even under robustness concerns, a feature often lost in other robust design literature. They also note that their results extend to settings with more permissive robustness requirements (e.g., requiring a fraction $\gamma$ of the maximal guarantee) and alternative forms of cost uncertainty.

Limitations and Scope

The analysis assumes the designer and authority share the same objective function (maximizing buyer welfare net of transfers) but differ in their confidence in the model. The paper does not address situations where the authority has a different objective function. Furthermore, the comparison between price and quantity regulation relies on the assumption that demand is learned ex-post, giving price regulation its "best shot" at outperforming quantity regulation; even under this favorable scenario, quantity regulation can dominate.