Bayesian econometrics and how to get rid of those wrong signs
1988
Griffiths, W.E. (New England Univ., Armidale (Australia). Dept. of Econometrics)
Despite advantages of the Bayesian approach, applied econometric work has generally been dominated by the sampling theory approach. A simple regression example with one coefficient is used to describe the Bayesian approach using 3 different priors: a natural conjugate informative prior, a noninformative prior, and a prior with inequality restrictions on the sign and possibly magnitude of the coefficient. The differences between the sampling theory and Bayesian approaches are highlighted. Some practical problems with the first 2 priors are suggested as possible reasons for the non-adoption of the Bayesian approach. The inequality-restricted prior provides a practical and meaningful alternative which is likely to increase the appeal of the Bayesian approach. The implications are outlined of extending the simple one coefficient model to one where the error variance is unknown and then one where there is an unspecified number of coefficients. An example is provided of how to compute Bayesian inequality restricted estimates using the econometric computer program SHAZAM.
显示更多 [+] 显示较少 [-]