CFE 2019: Start Registration
View Submission - CMStatistics
Title: Adaptive LASSO estimation of the spatial weights matrix Authors:  Miryam Merk - European University Viadrina (Germany) [presenting]
Abstract: Spatial econometric research is typically based on prior knowledge of the spatial weights matrix that characterizes cross-sectional interactions. For lattice data, spatial weights are commonly determined by contiguity, nearest neighbor assumptions, economic or social characteristics. These a priori specifications may lead to misspecifications of the model parameters, which are sensitive to the choice of the spatial weights. We therefore propose to select and estimate spatial dependence structures by using an adaptive Least Absolute Shrinkage and Selection Operator (LASSO). In the case of spatio-temporal models, the spatial dependencies of the process can be identified based on their observations over time. However, for purely spatial models the number of spatial links exceeds the sample size. The spatial weights are therefore estimated by cross-sectional resampling under the identifying assumption of sparsity. The estimation procedure employs two-stage least squares (2SLS) to account for endogeneity of the spatially dependent variable.