Marshall Lecture 2023: Behavioral Macroeconomics via Sparse Dynamic Programming.

This article develops a tractable framework to model boundedly rational (BR) dynamic programming agents who simplify their understanding of the world by selectively attending to key state variables and ignoring others, formalized through a "sparse max" operator. Applying this framework to canonical macroeconomic and finance models—including the life-cycle consumption-savings problem, the neoclassical growth model, and Merton's portfolio choice problem—it shows that BR agents exhibit realistic behaviors such as delayed retirement saving, partial inattention to interest rates and taxes, and muted responses to policy changes, leading to larger and more persistent macroeconomic fluctuations. The approach preserves much of the structure of rational dynamic programming while incorporating behavioral features like inattention and simplification, allowing for source-dependent attention and partial myopia without requiring full Bayesian updating. This methodology offers a unified, microfounded way to analyze bounded rationality across a broad class of dynamic economic problems, with implications for fiscal policy effectiveness, the failure of Ricardian equivalence, and the Lucas critique.