Unconditional Transformed Likelihood Estimation of Time-Space Dynamic Panel Data Models

Location
University House - Room 204
Dates
Tuesday 9th February 2016 (13:00-14:00)
Download the date to your calendar (.ics file)

Speaker: Sebastian Kripfganz (University of Exeter)

Speaker biography

Prior to joining the University of Exeter Business School in 2015 as a Lecturer in Economics, Sebastian Kripfganz completed his doctoral studies at Goethe University Frankfurt where he also held a position as a research and teaching assistant.

His international experience includes a research visit at Michigan State University on invitation by Professor Jeffrey Wooldridge and a short-term consultancy in the Poverty Reduction and Economic Management Network of The World Bank in Washington, DC.

He can also look back to a 19-months professional experience in the Economics Department of Bayerische Landesbank after he received his Economics diploma from the University of Mannheim. His research interests centre on dynamic panel data and spatial econometrics.

Abstract

Panel data sets allow to account for unobserved unit-specific heterogeneity, as well as time-series and cross-sectional dependence. I derive the unconditional transformed likelihood function and its derivatives for a fixed-effects panel data model with time lags, spatial lags, and spatial time lags that encompasses the pure time dynamic and pure space dynamic models as special cases. In addition, the model can accommodate spatial dependence in the error term. Consistent estimation in short panels requires making proper allowance for the influence of the initial observations. I demonstrate that the model-consistent representation of the initial-period distribution involves higher-order spatial lag polynomials. Their order is linked to the minimal polynomial of the spatial weights matrix and, in general, tends to infinity with increasing sample size. An appropriate truncation of these lag polynomials becomes necessary unless the spatial weights matrix has a regular structure. The finite sample evidence from Monte Carlo simulations shows that the proposed estimator performs well in comparison to a bias-corrected conditional likelihood estimator if parameter proliferation is kept under control. As an application, I use data from the Panel Study of Income Dynamics to estimate a time-space dynamic wage equation that I derive from a bargaining model. I find significant spillover effects among household members that give rise to a positive cohabitation premium. Furthermore, the theoretical bargaining model justifies a particular nonlinear restriction on the spatial time lag that simplifies the analytical derivations considerably, and is also empirically supported.