Specification Bias in Seemingly Unrelated Regressions