现代产业经济学系列讲座第301期

时    间：2026年7月3日15:00-16:30

地    点： 复旦管院思源楼326室

题    目：Sparse VARs Do Not Imply Sparse Local Projections & The Horizon Structure of Contagion

主讲人：Endong Wang 助理教授 德国曼海姆大学 经济系

主持人：俞学文 副教授

摘要: In this talk, Dr. Wang presents two complementary projects on horizon-specific Granger causality testing through high-dimensional local projections, with applications to connectedness in large financial and economic systems. The motivation is that spillovers often arise with delay and persist across horizons, so single-horizon measures cannot capture the full picture of systemic interdependence. The first project, Sparse VARs Do Not Imply Sparse Local Projections, shows that sparsity of the underlying VAR need not propagate to local projections, so that even a sparse short-run system can generate dense coefficients at longer horizons. Dr. Wang proposes a de-biased two-stage estimator that imposes sparsity only on the VAR and a robust inference procedure that dispenses with HAC corrections. The second project, The Horizon Structure of Contagion, builds a horizon-specific network from continuous Granger-causality measures computed from a "restricted" VAR, for instance, a Heterogenous autoregressive specification and a grouped network VAR, requiring sparsity only on the restricted law of motion. Throughout, Dr. Wang shows that post-double-selection and de-sparsified LASSO can fail for local projections, while the proposed method remains robust without horizon-by-horizon sparsity.

个人简介：Dr. Endong Wang is an econometrician whose research lies at the intersection of time-series and panel econometrics and empirical macroeconomics, with a focus on the identification and estimation of dynamic causal effects and mediation effects in macroeconomic causal channels and policy settings. He obtained his PhD from McGill University and is currently a Junior Professor of Econometrics in the Department of Economics at the University of Mannheim, Germany. His research develops local-projection methods for the identification of policy counterfactuals and the estimation of impulse responses, high-dimensional local projections, and panel shift-share designs.