A0705
Title: Uncertain spillover effects and priors for spatial models
Authors: Nikolas Kuschnig - Vienna University of Economics and Business (Austria) [presenting]
Abstract: In an ever more connected world spillover effects are at the centre of a wide range of applied research. Spatial econometric models are commonly used to analyse such spillovers empirically. However, these models suffer from rigid specifications and strong assumptions regarding connectivity between units. We address these issues by adopting a fully Bayesian approach. Assumptions such as known connectivities can be loosened, with their forms being learned from the data instead. Weakly informative priors provide the foundation for a flexible framework, imposing regularisation where appropriate and limiting assumptions otherwise. The result is more credible and extensible empirical tools that natively account for uncertainty. We dismantle the spatial econometric framework, discuss prior information in the context, and sketch out a Bayesian approach. We introduce general purpose and specific probability models that allow flexible treatment of connectivity. A simulation exercise and an empirical application show the merits of this approach. Bayesian methods present a great opportunity for once again raising the bar in spatial econometrics.