Essays in spatial autoregressive panel data models

(eng) The purpose of this dissertation is to improve the applied researcher's toolbox to estimate spatial autoregressive panel data models. The first chapter of the dissertation derives impacts of a change in explanatory variables on the dependent variable in a spatio-temporal specification. An empirical application based on cigarettes sales illustrates the use of these impacts and their interpretation. The second chapter derives specification tests to assess the relevance of spatial autocorrelation in a fixed effects panel data model and provides an empirical application of the developed tests devoted to the Feldstein-Horioka (FH) paradox. It is shown that the traditional FH model is misspecified due to the omission of spatial autocorrelation in the specification. Accounting for spatial autocorrelation reduces the paradox. The third chapter extends the Mundlak approach to spatial Durbin panel data models to allow consistent estimation of spatial panel data models by random effects. The approach consists in correcting for the possible correlation existing between regressors and individual effects. The proposed method is applied on the determinants of house prices among Belgian municipalities. Finally, spatial autoregressive models traditionally assume the linearity of spatial autocorrelation, meaning that one only has to estimate one spatial autoregressive parameter. The fourth chapter of the dissertation explorates the possibility of nonlinear spatial autocorrelation and suggests an application on the US presidential election in 2000.