Rational inattention

Overview

Rational inattention (RI) is a theory of decision making under limited attention. It models the idea that acquiring, processing, and acting on information is costly, so agents optimally choose what to pay attention to and how precisely, rather than being fully informed about everything. Limited attention then shows up as coarse decisions, slow adjustment to news, or selective focus on a few variables. RI matters because it provides micro-foundations for widely observed phenomena—sticky prices, sluggish expectation updating, limited consideration sets in consumer choice, and managerial focus on a handful of KPIs—and helps design information, interfaces, and incentives that respect attention constraints.

Origins and Credit

Christopher Sims (2003) introduced RI using information theory: the cost of processing information is proportional to the reduction in uncertainty (measured by “mutual information”). Michael Woodford developed macroeconomic applications, linking RI to sluggish expectations. In discrete choice, Filip Matějka and Alisdair McKay (2015) showed RI delivers a logit-like choice structure with endogenous “consideration.” Glen Weyl, Andrew Caplin, Mark Dean, and Xavier Gabaix advanced related formulations (e.g., sparse attention) and empirical connections. Together, this literature bridges bounded rationality and standard optimization.

Core Idea and Mechanics

RI treats attention as a scarce resource that can be allocated across variables and problems. The setup has three steps:

1. Prior uncertainty: The decision maker starts with beliefs about an unknown state (e.g., demand, costs, customer type).
2. Choose an information structure: Before acting, she chooses what signals to observe and how precisely to observe them. More precise signals reduce uncertainty but are more “expensive” in attention.
3. Act on what is learned: After seeing her signal, she chooses an action (price, quantity, portfolio) to maximize expected payoff net of the information cost.

The canonical cost metric is mutual information, the expected reduction in uncertainty about the state from observing the signal. Intuitively, the more you tailor your decision to fine-grained states, the more information you must process. RI models either impose a “capacity constraint” (you cannot process more than a fixed amount of information) or a “penalty” (each bit of information costs attention). The agent then optimally coarsens decisions along low-stakes or low-variability dimensions and focuses attention where it pays most.

Implications flow from this optimization:

• Coarse and sticky actions: Prices, settings, or plans adjust only when signals cross endogenously chosen thresholds; small shocks are ignored. This micro-founds sticky prices and lumpy adjustments.
• Selective attention: More volatile or payoff-relevant dimensions receive more attention (e.g., a manager tracks a few KPIs closely and glosses others).
• Discrete choice with limited consideration: In RI models of product choice, the probability of choosing an option takes a logit form with choice probabilities shifting more for options with higher expected surplus—capturing attention-driven substitution patterns.

Key Assumptions and Conditions

• Information is costly: Reducing uncertainty consumes attention; costs are increasing and convex in “amount learned.”
• Endogenous information acquisition: Agents choose what to learn and how precisely, subject to a capacity or cost trade-off.
• Bayesian updating: Given chosen signals, agents update beliefs rationally and optimize actions (boundedness is in information, not in optimization).
• Quasilinear/standard payoffs (often): Expected utility or profit maximization given beliefs; tractability often uses Gaussian-quadratic or multinomial environments.
• Stable priors and environment: The value of attention depends on prior uncertainty and payoff curvature; comparative statics rely on these features.

Implications

• Managerial focus and dashboards: RI explains why simplifying KPIs, alerts, and thresholds can improve performance; attention should be allocated to volatile/high-stakes metrics.
• Pricing and operations: Firms optimally adopt review cycles and thresholds; more attention to categories with high margin or volatile demand and less to stable, low-stakes items.
• Marketing and consumer choice: Salience and simplification raise the probability of consideration and purchase. Complicated disclosures are ignored; clear, timely messages are more effective.
• Finance and treasury: Portfolios under-react to diffuse information but respond to salient signals; attention spikes around earnings or macro releases.
• Policy and communication: Central banks and regulators can shape outcomes via information design—clear, simple guidance induces more responsive expectations than complex, noisy signals.
• Organization design: Delegation, specialization, and process automation are tools to economize on scarce managerial attention.

Example in Practice

Retail pricing under attention constraints. A multi-category retailer manages prices for 20,000 SKUs. Demand in a few categories (seasonal apparel, fuel-sensitive goods) is volatile and margins are high; most staples are stable with low margins. The pricing team’s analytics capacity (and store execution bandwidth) is limited.

Applying RI principles, the retailer:

• Allocates forecasting and experimentation resources to the volatile, high-margin categories and sets dynamic thresholds that trigger price changes when demand or costs move sufficiently.
• Implements coarse pricing grids and longer review cycles for stable staples, absorbing small shocks without adjustment.
• Designs dashboards with alerts for only the top five KPIs per buyer (e.g., margin variance, stockouts), suppressing low-value noise.

Outcomes: fewer, larger price adjustments in staples; more frequent, precise changes in volatile categories; overall higher profit net of analytics and execution costs. The observed “stickiness” is not a mistake; it is optimal given attention limits.

Limitations and Common Misunderstandings

• Measuring attention costs is hard: Mutual-information costs are elegant but abstract; calibrations rely on functional assumptions (Gaussian signals, logit forms) that may not fit all settings.
• Confounding frictions: Inertia may stem from menu costs, contracts, or organizational frictions rather than inattention. Disentangling requires careful identification.
• High-stakes exceptions: As stakes rise, optimal attention rises; RI predicts sharp responsiveness when it matters. Apparent over-inattention in high-stakes contexts likely reflects other constraints.
• Welfare and paternalism: RI is descriptive and normative for the decision maker (they optimize given attention costs). For policy, the relevant question is whether attention costs are private (to be respected) or reflect fixable frictions (to be reduced).
• Dynamics and learning: Attention capacity can be reallocated over time; static RI predictions may miss habit formation, organizational learning, or automation that changes costs.
• Overuse of “salience” language: Not all salience effects are RI; some reflect framing preferences or misperception rather than optimal coarsening.

Related Concepts (names only)

Bounded Rationality; Limited Consideration in Discrete Choice; Sticky Prices and Information Frictions; Salience and Attention; Information Design (Bayesian Persuasion); Costly Information Acquisition (Rational Search).