Title: Efficient MCMC estimation for spatial econometrics models with convex combination of connectivity matrices
Authors: Nicolas Debarsy - CNRS (France) [presenting]
James LeSage - Texas State University (United States)
Abstract: An efficient Bayesian MCMC procedure is proposed to estimate spatial econometrics models in which the cross-section dependence scheme is modeled through a convex combination of connectivity structures, each one representing a motivation for interactions between observations. An MCMC approach has been previously proposed to estimate and perform inference on the parameters of the convex combination in these models. Even though such an approach works well when the number of matrices included in the convex combination is low (2 or 3), the computer time required to estimate these models increases a lot when more matrices are considered. This comes from the need to pre-compute the Jacobian of the transformation of these models for a grid of values for all parameters of the convex combination. We propose an efficient MCMC approach to estimate these models. It is constructed from block sampling based on reversible jump MCMC and relies on bounded inference MCMC based on the lower and upper bounded conditional distributions. We show that our approach is not affected by the number of considered matrices in the convex combination, nor by the number of draws performed.