When simple economic rules need stronger guarantees

When can a simple rule be trusted?

Should a seller auction items separately? Should a decision-maker publish the clearest forecast? These choices matter because a rule can change others’ incentives, not just prices or information. This week’s six selected economics papers distinguish three guarantees: an outcome exists, we can compute it, and it achieves a desired objective. Computing an outcome with a specified guarantee establishes its existence; existence alone establishes neither efficient computation nor optimality.

Stability is not one guarantee

An equilibrium describes choices that nobody benefits from changing alone. Sam Ganzfried asks for more: worse local mistakes must become less likely as errors shrink, while earlier decisions account for one’s own possible future mistakes. His exact algorithm covers finite two-player zero-sum games—one player’s gain is the other’s loss—with perfect recall, meaning players remember their own information and actions, and rational input data. Its polynomial bit-complexity guarantee concerns the encoded explicit game tree, including numerical precision, not demonstrated practical speed. The method extracts an exact limit rather than merely choosing a tiny error probability (Ganzfried).

Julien Fixary, Rémi Castera and Rida Laraki instead study resource allocations across group interactions. Individual budgets and group capacities couple decisions: increasing one interaction can make another infeasible. They prove that their “coalitionwise stable” networks exist under continuity and componentwise quasiconcavity, restrictions on how payoffs vary with interactions. Two formulations of stability against small mistakes coincide under an additional componentwise monotonicity assumption. Unlike Ganzfried’s computation result, this establishes existence and relationships between stability concepts in a constrained model (Fixary and colleagues).

Auctions: optimality versus existence

Sho Mikami defines separate selling as running an optimal auction for each item independently. Buyers’ values add across items and are independent across buyers and items, with finitely many possible values. Each item’s distribution is identical across buyers but can differ across items. If at least two items have a positive possible value below their maximum, separate selling is exactly revenue-optimal for any buyer count of at least two if and only if “top selling” is optimal for one buyer: selling each item only at its highest possible value and price (Mikami).

The comparison covers dominant-strategy incentive-compatible mechanisms with ex-post individual rationality, and Bayesian incentive-compatible mechanisms with interim individual rationality. In plain terms: truthful reporting must be best regardless of others’ reports, with participation worthwhile after values are known; or truthfulness and participation are assessed in expectation given one’s own values. The distributional conditions highlight “top-value isolation”: high non-top values need a sufficient gap below the maximum or sufficiently little probability mass compared with the mass at the maximum—not rarity compared with a numerical price (Mikami).

Wei He, Yeneng Sun and Hanping Xu ask a different question about multiunit auctions. For any bidder count above one and strictly increasing, concave utility functions with at least one strictly risk-averse bidder, they construct independent, full-support distributions where monotone pure-strategy equilibrium fails. Such strategies prescribe deterministic bids that do not decrease as values rise. Under uniform pricing, their construction works for every supply of at least three units within an admissible strategy class; discriminatory pricing requires no such restriction. This does not rule out every equilibrium or describe every distribution (He and colleagues).

There is no contradiction with Mikami: revenue optimality and existence of a particular bidding pattern are different questions in different auction environments.

Information changes incentives

Vasudha Jain and Jaideep Roy study a public forecast about one uncertainty and expert advice about another, independent uncertainty. The forecast changes the parties’ expected conflict. In their uniform–quadratic model—uniformly distributed private conditions and squared-error losses—with a binary conflict state for the main characterizations, full revelation is optimal under optimal mediation. Commitment to an action rule can instead make no information preferable. Direct, nonbinding “cheap talk” can favour selective disclosure over both full revelation and silence. Crucially, the analysis selects the finest, most informative cheap-talk equilibrium at the study’s reports; these conclusions fail if uninformative “babbling” occurs there (Jain and Roy).

Philip R. Neary connects design with selection differently: symmetric binary Hopfield memory networks correspond exactly to two-action games with a quadratic potential, a shared score tracking individual incentives. Stable configurations correspond to pure-strategy equilibria; one-at-a-time neural updating corresponds to best responses. Designing intended stable outcomes still leaves the question of which emerges. This is a mathematical equivalence, not evidence that human decisions behave like neural memories (Neary).

What remains unresolved

These arXiv v1 preprints provide conditional theory, not measured market effects or consensus. Our evidence comprises abstracts and selected extracted passages, not complete papers; damaged equations limit verification. Abstract-only support cannot establish unreported methods or robustness, and missing excerpts do not mean missing analysis. Across this selection, the practical question remains which assumptions fit an institution well enough for its particular guarantee to guide decisions.