Event Date
"Macroeconomic Effects of Production Networks: A Deep Learning Approach" (joint with Jesus Fernandez-Villaverde, Christian Matthes and Felipe Schwartzman)
Abstract: We develop a scalable deep learning framework to globally solve high-dimensional dynamic general equilibrium models with production networks. Exploiting the symmetry of the economic environment, our algorithm overcomes the curse of dimensionality, and we apply it to a calibrated multi-sector model of the U.S. economy. Global non-linearities dampen ordinary aggregate fluctuations relative to perturbation methods, yet they slow the recovery from large shocks. The mechanism is a bottleneck in investment hubs, the few sectors that produce most investment goods: their nonlinear resource constraints block the reallocation of resources and reproduce the slow recovery of output and investment after the Great Recession. Perturbation solutions miss this, overstating volatility and predicting a counterfactual rebound.