Bayesian Analysis of Systems of Seemingly Unrelated Regression Equations Under a Recursive Extended Natural Conjugate Prior Density

Authors: 
Jean-Francois Richard, Mark F.J. Steel
Duke University, Catholic University of Louvain

Nov 30 1987

A "Recursive Extended Natural Conjugate" prior density is proposed for systems of seemingly unrelated regression equations. Compared to other classes of Extended Natural Conjugate prior densities it offers the advantage that analytical expressions are available for the prior convariance matrix of the regression coefficients. The posterior analyzes combines together analytical and numerical techniques. Numerical integration is based on Monte-Carlo importance sampling. An application to a system of four wage equations for EEC countries serves to evaluate different techniques discussed in the paper.

Manuscript: 

PDF icon 1987-06.pdf