Learning

• Value-Based Reinforcement Learning Matches Otherwise Challenging-to-Explain Consumption Patterns Out of Covid Stimulus Payments (JMP)

Inequality & Heterogeneity

• (Early Work) The Effect of Local Economic Shocks in a Geospatial Model with Moving (w/ Man Chon Iao, Doruk Gokalp )

Statistics/ML

• The AI Macroeconomy: A New Set of Benchmarks for Multiagent RL Models (Workshop Presentation 2022 ICML)

​

• “Bayesian Exploration Networks” (Mattie Fellows*, Brandon Kaplowitz*, Christian Schroeder De Witt, Shimon Whiteson), Under Review, ICML 2024: https://arxiv.org/abs/2308.13049 (* Equal Contribution)

​

• EarlyWork (planned for Neurips 2024): “Provable Convergence to Nash Equilibria with Public Beliefs in Large Imperfect Information Extensive-Form Games” (Coauthors: Gabriele Farina, Sobhan Mohammadpour, Sam Sokota). A Followup to Both ReBeL (Brown et al. 2020) and “Abstracting Imperfect Information Away from Two-Player Zero-Sum Games” (Sokota et al 2023), we establish proofs and rates of convergence for the Public Belief Game, and use it to develop an improved deep RL algorithm that we test on large instances of “Liar’s Dice”, a game in the poker-family, infeasible to solve via traditional counterfactual regret minimization. We efficiently achieve near-0 regret, a new SOTA result. Economically, this opens up the possibility of studying large models of imperfect infor- mation with learning, with strategic incentives, such as those that would occur during stock-market crashes, Federal Reserve guidance or other informational shocks.