Bootstrap for seemingly unrelated regression

[Owen Gallupe]

Hi,

Since seemingly unrelated regression (SUR) is linear regression with correlated errors (at least that's my understanding), does it make sense to bootstrap it if there is residual non-normality and/or heterogeneity of variance in the linear regression models that comprise the SUR model? I assume this would relax the parametric assumptions, but I'm having a hard time finding confirmation of this approach.

E.g., say this model displays residual non-normality:

reg price length turn

Would it then make sense for the following SUR model to be bootstrapped?:

bootstrap, reps(1000) seed(1234): sureg (price length turn) (length turn)

Thanks.

Owen

[George Ford]

https://www.statalist.org/forums/forum/general-stata-discussion/general/1398586-sureg-estimation-with-robust-standard-errors

mysureg would do it for the variance, but there are other options.

don't worry about the non-normality of the residual. there are plenty of posts here on that issue.

[Owen Gallupe]

Thank you, George! Clearly my search approach needs refining since that was an obvious one. But thank you for pointing me in the right direction.

Owen