Optimal Procurement Design: A Reduced-Form Approach

1. Problem Statement

The paper addresses the design of optimal procurement mechanisms in environments where quality is unverifiable and unobservable to the buyer at the time of contract signing.

• Context: Unlike standard markets with standardized goods, procurement often involves custom goods (e.g., medical devices, software, infrastructure) where the true quality $q$ is only revealed long after the transaction.
• The Conflict: Buyers face a trade-off between price competition and quality concerns. In standard second-price auctions, aggressive bidding by low-quality sellers can lead to the "winner's curse" or the selection of low-quality goods (the "lemons" problem).
• Objective: The author seeks to maximize a weighted average of the buyer's expected payoff and social surplus, subject to the constraint that the buyer's expected payoff must be non-negative (Individual Rationality).
• Model Setup:
  • One buyer and $n$ symmetric sellers.
  • Seller $s$ has private quality $q_s \in [0, 1]$, distributed i.i.d. with density $f$ and CDF $F$.
  • Seller's cost is $q_s$; Buyer's valuation is $v(q)$.
  • The buyer cannot verify $q$ ex-post, so contracts depend on reported types.

2. Methodology: The Reduced-Form Approach

The paper employs a reduced-form approach, transforming the mechanism design problem from choosing ex-post allocation rules to choosing interim allocation probabilities.

• Transformation:
  • Instead of solving for the full mechanism $\{p_s(q), t_s(q)\}$, the author solves for the interim allocation probability $P(q)$, which is the probability a seller with type $q$ wins given others report truthfully.
  • By the Revelation Principle, the problem is reduced to finding a decreasing function $P: [0, 1] \to [0, 1]$.
• Feasibility (Border's Condition):
  • The set of implementable interim allocations is characterized by Border's (1991) condition.
  • Using majorization theory (Kleiner, Moldovanu, and Strack, 2021), the condition is translated: A quantile interim allocation $\tilde{P}(s)$ (where $s = F(q)$) is implementable if and only if it is weakly majorized by the allocation rule of always trading with the lowest quality seller, denoted $\tilde{P}^*(s) = (1-s)^{n-1}$.
  • Intuitively, this means the mechanism cannot discriminate between qualities more than the "lowest-quality-wins" rule.
• Optimization Problem:
  • The objective is to maximize a linear functional of the form $\int h_\gamma(F^{-1}(s)) \tilde{P}(s) ds$, where $h_\gamma$ is the weighted virtual surplus.
  • Constraints: $\tilde{P}$ must be decreasing, weakly majorized by $\tilde{P}^*$, and satisfy the buyer's non-negative payoff constraint.
  • The solution involves ironing (Myerson, 1981): If the virtual surplus is not monotonic, the optimal allocation "flattens" (pools) over intervals where the surplus function is increasing, replacing the optimal pointwise choice with a constant average.

3. Key Results

A. The Optimal Mechanism: Bid-Rstricted Auction (BRA)

The paper proves that the optimal mechanism is a Bid-Restricted Auction (BRA).

• Structure: Sellers are restricted to submitting bids only within a specific collection of disjoint intervals $[b_i, \bar{b}_i]$. Bids in the "gaps" between intervals are forbidden.
• Equilibrium: Like a second-price auction, the BRA induces a weakly dominant strategy equilibrium:
  • If a seller's quality $q$ falls within a bid interval $[b_i, \bar{b}_i)$, she bids her true quality $q$.
  • If $q$ falls in a gap $(\bar{b}_{i-1}, b_i)$, she bids the lower bound of the next interval, $b_i$.
  • If $q$ is above the highest allowed bid, she does not participate.
• Pooling: Sellers with qualities in a "gap" (pooling region) all submit the same bid ($b_i$). This creates a random allocation (lottery) among them if they win, effectively pooling their qualities.
• Payment Reduction Rule: A novel feature is introduced to prevent "underbidding across gaps."
  • If the winner is the only bidder in a specific interval and the second-lowest bid is from a higher interval, the winner receives a reduced payment: $p = \frac{b_j + k \cdot b_{j-1}}{k+1}$ (where $b_j$ is the second-lowest bid, $b_{j-1}$ is the lower bound of the winner's interval, and $k$ is the number of other bidders at $b_j$).
  • This rule eliminates the incentive for a seller in a gap to bid just below the gap to win with certainty, ensuring the pooling strategy remains a dominant strategy.

B. Special Cases and Generalizations

• Buyer-Optimal ($\gamma=1$):
  • If the buyer's virtual surplus is decreasing (mild quality concerns), the BRA reduces to a standard second-price auction with a reserve price.
  • If the virtual surplus is increasing (severe quality concerns), the optimal mechanism is random selection (degenerate bid interval).
  • If the virtual surplus is single-peaked, the BRA corresponds to the Lowball Lottery Auction (LoLA) (a second-price auction with a floor price and reserve price).
  • If the virtual surplus is single-dipped, the BRA involves two bid intervals separated by a gap, requiring the payment reduction rule.
• General Objective (Weighted Average + IR Constraint):
  • When maximizing a weighted average of buyer payoff and social surplus, the BRA remains optimal.
  • Augmented BRA (aBRA): If the buyer's non-negative payoff constraint binds (i.e., the unconstrained optimal mechanism yields negative buyer payoff), the mechanism is modified to include a random reserve price.
    • An extra bid $B$ is introduced.
    • Sellers bidding $B$ are qualified to win only with probability $\zeta \in (0, 1)$.
    • This allows the mechanism to partially admit high-quality sellers (who generate social surplus but hurt the buyer's payoff) without fully excluding them, satisfying the constraint while maximizing the weighted objective.

4. Key Contributions

1. Unified Framework: The paper generalizes previous results (Manelli & Vincent, 1995; Lopomo et al., 2023) by identifying optimal mechanisms for a broad class of procurement problems, not just specific virtual surplus shapes.
2. Reduced-Form Solution: It successfully applies majorization theory and linear optimization techniques to solve procurement design problems with unverifiable quality, providing a rigorous mathematical foundation for "pooling" regions.
3. Novel Mechanism Design:
   • Introduces the Bid-Restricted Auction (BRA) as the general solution.
   • Proposes the Payment Reduction Rule to sustain equilibrium in the presence of bid gaps.
   • Develops the Augmented BRA (aBRA) with random qualification to handle individual rationality constraints.
4. Insight on Quality Concerns: The paper demonstrates that "global" quality concerns might require abandoning competition entirely, but "local" quality concerns can be efficiently addressed by excluding specific bid intervals rather than abandoning the auction format.

5. Significance

• Theoretical: It bridges the gap between standard auction theory and procurement problems with unverifiable quality, showing that simple, dominant-strategy mechanisms can be optimal even in complex environments with asymmetric information.
• Practical: The findings offer actionable guidance for public and private procurement.
  • It suggests that restricting the range of allowable bids (e.g., disallowing bids in a specific "danger zone" of low quality) can improve outcomes compared to standard open auctions.
  • It provides a theoretical justification for mechanisms like the Lowball Lottery Auction and explains when and why random allocation or payment reductions are necessary.
  • The aBRA offers a solution for government agencies that must ensure "value for money" (non-negative payoff) while still pursuing social surplus, a common dilemma in public procurement.

In summary, Zhang demonstrates that by restricting bids to specific intervals and adjusting payment rules, a buyer can effectively mitigate quality concerns while maintaining the simplicity and incentive compatibility of second-price auctions.