Bayesian inference for the correlation coefficient in two seemingly unrelated regressions

We study the problems of hypothesis testing and point estimation for the correlation coefficient between the disturbances in the system of two seemingly unrelated regression equations. An objective Bayesian solution to each problem is proposed based on combined use of the invariant loss function and the objective prior distribution for the unknown model parameters. It is shown that this new solution possesses an invariance property under monotonic reparameterization of the quantity of interest. The performance of the proposed solution is examined through a simulation study. Furthermore, the solution is illustrated by an application to the real annual data for analyzing the investment model.