EcoSta 2019: Start Registration
View Submission - EcoSta2019
Title: Efficient Bayesian estimation for the space-time stationary condition with blocked sampling approach Authors:  Yoshihiro Ohtsuka - Tohoku Gakuin University (Japan) [presenting]
Abstract: An efficient posterior sampling algorithm is proposed for the spatial dynamic panel data model from the viewpoint of the Bayesian inference. The stationary condition of this model mutually depends on three parameters such as simultaneous and lagged spatial dependencies, and serial correlation. Thus, the interdependence between these parameters yield law convergence to their target marginal distribution. To accelerate sampling efficiency, Bayesian estimation algorithm for these parameters is developed by using a Tailored approximation and blocked Metropolis-Hastings (TaB-MH) algorithm. With the respect to an inefficient factor, the TaB-MH algorithm is superior to the random walk MH in the experimental studies and empirical analysis on regional data.