Regression models under competing covariance matrices: A Bayesian perspective