Online Program Home
  My Program

Abstract Details

Activity Number: 250 - SPEED: Sports and Business
Type: Contributed
Date/Time: Monday, July 31, 2017 : 2:00 PM to 2:45 PM
Sponsor: Business and Economic Statistics Section
Abstract #325284
Title: Deep Learning Econometrics
Author(s): Guanhao Feng* and Nicholas Polson and Jianeng Xu
Companies: and University of Chicago and University of Chicago
Keywords: Neural Network ; Long Short-Term Memory ; Difference-in-Difference (DID) ; Regression Discontinuity Design (RDD) ; Kernel Density Estimation (KDE) ; Instrumental variables (IV)

We develop applications of deep learning estimation in various econometrics methods. These applications involve complex data generating process with either nonlinear equations, interactive effects, sparse or latent factor structures. Deep learning estimators, such as neural network or long short-term memory models, can be developed wherever there is a "black box" structural estimation that considers a GMM objective function or likelihood-based probabilistic model. For specific econometric modeling, we show how deep learning estimators apply to nonlinear difference-in-difference (DID), regression discontinuity design (RDD), nonparametric kernel density estimation (KDE), and instrumental variables (IV). We also provide a Bootstrap inference method for the deep learning estimator. We demonstrate the advantages and disadvantages of the deep learning estimator over the traditional econometric estimators in an extensive simulation study.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2017 program

Copyright © American Statistical Association