×

zbMATH — the first resource for mathematics

On assessing prior distributions and Bayesian regression analysis with \(g\)-prior distributions. (English) Zbl 0655.62071
Bayesian inference and decision techniques, Essays Hon. Bruno de Finetti, Stud. Bayesian Econ. Stat. 6, 233-243 (1986).
[For the entire collection see Zbl 0608.00012.]
Considered is the normal linear multiple regression model \(y=X\beta +u\), where \(u\) is \(N(0,\sigma 2I_ n)\)-distributed. Several approaches to use prior information on the parameters \(\beta\) and \(\sigma\) for estimating \(\beta\) are discussed. As a new approach to the problem the author proposes to use Muth’s rational expectations hypothesis and an imaginary sample based on the model above with u replaced by an \(N(0,g^{-1}\sigma 2I_ n)\)-distributed error, \(0<g<\infty\), to get a reference informative prior distribution. It turns out that this is in the inverted-gamma-normal form. For a particular version the posterior distributions are then calculated and sampling properties of the posterior mean are derived.
The approach is shown to be equivalent to a least squares approach where \(\beta\) is restricted to lie in an ellipsoid with prescribed center and magnitude and determined by the matrix \(X'X\).

MSC:
62J05 Linear regression; mixed models
62F15 Bayesian inference
62C10 Bayesian problems; characterization of Bayes procedures