Title: Bootstrapping in large panels with cross sectional dependence
Authors: Jan Ditzen - Heriot-Watt University (United Kingdom) [presenting]
Abstract: Bootstrapping standard errors and confidence intervals is standard in applied econometrics. However, the literature is missing applications for large panels with heterogeneous slopes. Large dynamic panels combine characteristics of time series and panel data. In addition, large panels can contain unobserved dependence across cross sectional units, often contained in the error term of a regression model. A bootstrap has to maintain the structure of the model across the time and the cross sectional dimension and of the error term. A wild bootstrap is proposed to maintain the error structure. Results of the bootstrap with residuals in- and excluding cross sectional dependence are compared. Bootstrap methods for a static and a dynamic model are explained and compared.