A0315
Title: Bayesian group lasso for spatial autoregressive model with convex combinations of different spatial weights
Authors: Zhengzheng Cai - Dongbei University of Finance and Economics (China)
Xiaoyi Han - Xiamen University (China)
Jianchao Zhuo - Xiamen University (China) [presenting]
Abstract: The spatial autoregressive (SAR) models with convex combinations of different spatial weights have been employed to capture spatial spillovers from different channels and identify the relative importance of each channel. The scenario is considered where different spillover channels exhibit group structure, with multiple spillover channels (spatial weights) being classified into several different groups. A new Bayesian group lasso prior is proposed to detect groups with and without spillover effects and to tackle the multicollinearity issue among multiple spatial weights within the same group. Simulation results suggest that the newly proposed prior performs well in parameter estimation. Lastly, the model is applied with the proposed prior to investigating the sovereign risk spillover effects among developed and developing countries and finding the co-existence of multiple transmission channels in the presence of contemporaneous risk spillover. Among possible spillover channels, the socioeconomic proximity index within informational channels plays the most crucial role, which confirms the empirical findings of a prior study. Additionally, marginal effect analysis suggests that the indirect effect of state spending is almost as large as the direct effect in view of future diffusion impacts.
