Modelling spatial weight matrices and lags in spatial panel models

Although each variable in a spatial econometric model can have its own spatial weight matrix, practitioners generally adopt one common pre-specified spatial weight matrix for all of them. This thesis breaks this practice for commonly used spatial econometric models with controls for fixed effects in space and time and different panel data settings, using both pre-specified and parameterized spatial weight matrices. The proposed quasi-maximum likelihood estimators of the parameters of these models are proven to be identified, consistent, and asymptotically normal. Three results in this thesis stand out. First, spatial autoregressive errors tend to go together with a sparse matrix and spatial moving average errors with a dense matrix. Second, indirect spillover effects, the focus of many empirical studies, can be severely biased when one common pre-specified spatial weight matrix is used. Third, identification problems that plagued the empirical literature trying to estimate general nesting spatial models are diminished if the spatial weight matrices are parameterized