Title: Bayesian Predictive Density Estimation for a Chi-squared Model Using Information from a Normal Observation with Unknown Mean and Variance

Abstract: In this paper, we consider the problem of estimating the density function of a Chi-squared variable with $n_2 > 0$ degrees of freedom, $W$, on the basis of observations of another Chi-squared variable with $n_1 > 0$ degrees of freedom, $V$, and a normal variable, $\boldsymbol{X}$, under the Kullback-Leibler loss. We assume that these variables have a common unknown scale parameter and that the mean of $\boldsymbol{X}$ is also unknown. We compare the Bayesian predictive densities with respect to the invariant prior and a hierarchical shrinkage prior. The hierarchical Bayesian predictive density depends on $\boldsymbol{X}$ while the invariant Bayesian predictive density does not. A sufficient condition for the hierarchical Bayesian predictive density to dominate the invariant Bayesian predictive density is obtained for the case of $n_2 = 2$. Furthermore, we show that many of the calculations can be generalized for the case where $n_2$ is an even integer possibly greater than $2$.