Spatial econometrics and the Lasso estimator : theory and applications
Abstract
This thesis links two topics of empirical economics: spatial econometrics and the Lasso estimator. Spatial econometrics is concerned with methods and models accounting for interaction effects between units. The Lasso estimator is a regularisation technique that allows for simultaneous variable selection and estimation in a high dimensional setting where the number of parameters may exceed the sample size. Three applied and theoretical articles are presented that demonstrate how spatial econometric research can benefit from high-dimensional methods and, specifically, the Lasso. The introduction in Chapter 1 presents a literature review of both fields and discusses the connections between the two topics. Chapter 2 examines the effect of economic growth on civil conflicts in Africa. The Lasso estimator is employed to generate instrumental variables, which account for non-linearity and spatial heterogeneity. The theoretical contribution in Chapter 3 proposes a two-step Lasso estimator that can consistently estimate the spatial weights matrix in a spatial autoregressive panel model. Chapter 4 is an application to the US housing market. A Lasso-based estimation method is considered that controls for spatial effects in a spatial error-correction model. Chapter 5 provides concluding remarks.