Deep learning for solving dynamic economic models

• 成果类型: Article
• 署名作者: Maliar, Lilia; Maliar, Serguei; Winant, Pablo
• 署名单位: City University of New York (CUNY) System; Center for Economic & Policy Research (CEPR); Stanford University; Santa Clara University; heSam Universite; ESCP Business School; Institut Polytechnique de Paris; ENSAE Paris; Ecole Polytechnique
• 刊物名称: JOURNAL OF MONETARY ECONOMICS
• ISSN/ISSBN: 0304-3932
• DOI: 10.1016/j.jmoneco.2021.07.004
• 发表日期: 2021
• 页码: 76-101
• 关键词: Artificial intelligence; Machine Learning; Deep learning; Neural Network; Stochastic Gradient; Dynamic models; Model reduction; dynamic programming; Bellman equation; Euler equation; Value functio

摘要

We introduce a unified deep learning method that solves dynamic economic models by casting them into nonlinear regression equations. We derive such equations for three fun-damental objects of economic dynamics - lifetime reward functions, Bellman equations and Euler equations. We estimate the decision functions on simulated data using a stochas-tic gradient descent method. We introduce an all-in-one integration operator that facil-itates approximation of high-dimensional integrals. We use neural networks to perform model reduction and to handle multicollinearity. Our deep learning method is tractable in large-scale problems, e.g., Krusell and Smith (1998). We provide a TensorFlow code that accommodates a variety of applications. (c) 2021 Elsevier B.V. All rights reserved.