arXiv subjects

Henrique de Oliveira

Publications and source records attributed to Henrique de Oliveira.

3 recordsLinked to original sources

arXiv2026

Robust Aggregation of Correlated Information

Decisions are typically made by drawing on multiple sources of information, yet the correlation among these sources is often hard to assess. We study robustly optimal strategies\,---\,those that maximize the worst-case payoff guarantee across all correlation structures. With two states and two actions, the robustly optimal strategy is strikingly simple: use only the best information source and ignore the rest. With more actions, a robustly optimal strategy decomposes the problem into local comparisons and assigns each comparison to the source that is best at that comparison. With more states, the same specialization logic survives through a concavification characterization: the decision problem is decomposed into payoff-relevant components, and each component is assigned to its best source. These results offer a new rationale for why ignoring information can be optimal and highlight the benefit of using specialized information sources when correlations are ambiguous.

econ.TH↗

arXiv2026

Posterior-Separable Costs and Menu Preferences

We consider an agent with a rationally inattentive preference over menus of acts, as in de Oliveira et al (2017). We show that two axioms, Independence of Irrelevant Alternatives and Ignorance Equivalence, are necessary and sufficient for this agent to have a posterior-separable cost satisfying a mild smoothness condition, called joint-directional differentiability. Viewing the decision-maker's problem as a Bayesian persuasion problem, we also show that these axioms are necessary and sufficient for solvability by a unique hyperplane. When the cost function remains invariant for different priors, we show that these axioms imply uniformly posterior separable costs that are differentiable.

econ.TH↗

arXiv2025

Rationalizing dynamic choices

An analyst observes an agent take a sequence of actions. The analyst does not have access to the agent's information and ponders whether the observed actions could be justified through a rational Bayesian model with a known utility function. We show that the observed actions cannot be justified if and only if there is a single deviation argument that leaves the agent better off, regardless of the information. The result is then extended to allow for distributions over possible action sequences. Four applications are presented: monotonicity of rationalization with risk aversion, a potential rejection of the Bayesian model with observable data, feasible outcomes in dynamic information design, and partial identification of preferences without assumptions on information.

econ.TH↗