时  间：2020年10月28日12:00-13:30

线上平台与会议ID：ZOOM    ID: 898 3169 9235  (密码：202010)

主题：Testing for neglected nonlinearity in high-dimensional time series models: A deep neural network approach

会议主题：

We introduce a linearity test in the context of deep neural networks to detect nonlinearity in high-dimensional time series. This facilitates determination of linear versus a nonlinear structure in “big data” by testing for neglected nonlinearity in econometric models. We propose a Lagrange multiplier test that determines whether adding nonlinear components to the linear network enhances predictive performance. We illustrate by application to economic time series. Performance of the test is assessed by simulated time series from high-dimensional Lorenz systems, and to experimental time series from a high-dimensional model of Taylor–Couette flow. The proposed approach can play a valuable role in building big data econometric models and evaluating model adequacy. It also provides a way of assessing whether a nonlinear dimensionality reduction is preferable to linear one.

主讲人简介：

Esfandiar Maasoumi教授，英国伦敦政治经济学院博士，美国艾默里大学艺术与科学杰出经济学教授（Arts of Science Distinguished Professor of Economics）,《Econometric Reviews》主编，计量经济学期刊会士(Fellow of Journal of Econometrics)，美国统计协会会士（Fellow of American Statistical Association）。主要研究集中于计量经济学在收入不平等、流动性、社会福利以及金融资产定价方面等相关领域。Maasoumi教授的研究发表在《Journal of Political Economy》、《Review of Economic Studies》、《Econometrica》、《Biometrika》、《Journal of Econometrics》和《Journal of Applied Econometrics》等国际顶尖期刊。