Economic theory: information, incentives and the limits of simple predictions

A route through the selected papers

Read these six papers in three pairs: information design, AI-assisted signaling, then risky choice and auction equilibrium. The connections concern how assumptions shape theoretical conclusions, not a consensus about policy. The sequence moves from designing communication to questioning the foundations of predictions.

Information design: architecture versus incentives

In Designing Public Information for Strategic Advice, a public study concerns an uncertainty independent of the expert’s private knowledge, yet changes the conflict between expert and decision maker. Within the uniform–quadratic model, with a binary conflict state for the main characterizations, optimal mediation favors full revelation; commitment to an action rule may favor no information; unmediated cheap talk can favor selective disclosure over both. The cheap-talk analysis selects the finest equilibrium at the reports used, not just any informative equilibrium; its results fail if uninformative babbling is played there. These are conditional prescriptions, not a general argument against precise forecasts.

Persuasion by Parts changes the information architecture instead. Separate sources observe independent components of an additive payoff and signal only their own component. The receiver acts with probability g(y), where y is the posterior mean of the total payoff and g is nondecreasing. For this response class, a known g permits an action-probability guarantee of a constant times the square of the joint-persuasion optimum; the exponent is tight. When g is unknown, one experiment constructed from the prior guarantees a constant times that optimum to the power 2 + ε, for every positive ε, with the constant depending only on ε.

The distinction matters: these are guarantees for inducing action, not welfare rankings. Moreover, the separate sources have no incentive constraints because one designer commits to all their rules. Fragmented observation and strategic disagreement are different obstacles to communication (Persuasion by Parts; Strategic Advice).

Cheaper signals, costlier distinctions

The AI papers connect through costly signaling. In The Verification Bottleneck, researchers privately know their quality, an editor optimally allocates scrutiny, and reviewers supply time according to its expected scientific contribution. Cheaper production makes imitation easier. Beyond a technological threshold, reviewer capacity and verified knowledge weakly decline; with observable quality, production advances weakly increase both. This is a model comparison, not a measured effect of AI on journals.

The Market for AI-Assisted Signaling adds providers choosing technologies and prices while individual adoption is unobserved. Easier imitation can increase the output stronger senders need to demonstrate ability. Welfare can fall even with full revelation: accurate assessment does not imply economical signaling. Unlike the peer-review model’s verified-knowledge metric, this comparison explicitly weighs resource costs.

Verification offers a useful contrast, not a contradiction. Under Assumptions 1–6, at fixed production technology, better verification technology or a finer exogenous, nonmanipulable partition of researcher types weakly raises attainable verified value at every fixed reviewer capacity in The Verification Bottleneck. It also weakly raises greatest-equilibrium reviewer capacity and verified knowledge—not necessarily those quantities in every equilibrium, nor economic surplus generally. In the AI-signaling market, a higher external minimum verification capacity can instead induce a provider to switch from separating abilities to pooling them, worsening assessment. The interventions, strategic responses and outcome measures differ.

Preferences and equilibrium: distinct limits

Bang-Bang Correlation Sensitivity supplies an axiomatic comparison. Under continuity, completeness and betweenness, indifference between identically distributed alternatives under every pairing at one full-support marginal implies correlation insensitivity throughout the relevant outcome domain. The characterization separates sensitivity to joint outcomes from departures from classical independence; its testable restrictions are not experimental confirmation.

The multiunit-auction paper instead constructs counterexamples to monotone pure-strategy equilibrium with at least one strictly risk-averse bidder. For uniform pricing with at least three units, the result excludes admissible monotone equilibria in constructed distributions; discriminatory pricing needs a constructed finite supply but no admissibility restriction. It does not exclude every equilibrium or establish failure for every distribution. This is an equilibrium-existence boundary, not the preference paper’s representation exercise.

Open questions and evidence limits

The advice paper leaves correlated demand and local conditions without a formal general treatment, including strong correlation. Whether its disclosure logic travels to such environments remains unresolved (Strategic Advice). Across these pairs, applying results requires choosing which institutional constraints, response classes and equilibrium selections fit the setting.

All six supplied records contain abstracts and selected paper-text excerpts, not abstract-only evidence or complete papers. Abstract-only evidence cannot support unreported methods or qualifications; missing excerpts do not establish missing results. PDF extraction visibly disrupts equations and reading order, limiting proof-level assessment. These are arXiv preprint versions; the supplied evidence does not establish peer-review status. This guide describes selected coverage, not field-wide trends or validated policy prescriptions.