tag:blogger.com,1999:blog-2198942534740642384.post4039412527993702795..comments2023-10-24T03:16:41.009-07:00Comments on Econometrics Beat: Dave Giles' Blog: A Trick With Regression ResidualsDave Gileshttp://www.blogger.com/profile/05389606956062019445noreply@blogger.comBlogger3125tag:blogger.com,1999:blog-2198942534740642384.post-50475271775987139912017-10-16T01:35:40.807-07:002017-10-16T01:35:40.807-07:00Hi Prof. Thanks for the insights. However, suppose...Hi Prof. Thanks for the insights. However, suppose you need to do a multivariate test of heteroskedasticity for the whole SUR system rather than equation-by-equation. How do you do it in Eviews? Kindly refer to the following paper:<br />System Misspecification Testing and Structural Change in the Demand for Meats<br />Anya McGuirk, Paul Driscoll, Jeffrey Alwang, and Huilin Huang (1995).<br />There's need to take into account cross-equation correlation when conducting multivariate diagnostic tests.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-59803830936518559122012-02-23T16:03:07.776-08:002012-02-23T16:03:07.776-08:00Ben - thanks for the comment. Keep in mind that th...Ben - thanks for the comment. Keep in mind that these LM tests have only asymptotic validity, so there is no distinction between n and (n-k). Bottom line: no real problem.Dave Gileshttps://www.blogger.com/profile/05389606956062019445noreply@blogger.comtag:blogger.com,1999:blog-2198942534740642384.post-74934498975095381362012-02-23T15:27:16.774-08:002012-02-23T15:27:16.774-08:00Hey, Prof, won't tests implemented in this fas...Hey, Prof, won't tests implemented in this fashion be optimistic in terms of the degrees of freedom they associated with the estimated residuals? If you have n residuals from your original fitted model which estimated, say, k parameters, then we should be using n-k as our residual degrees of freedom but presumably this method assumes the residuals from the second model (which are equivalent to those from the first) have n-1 degrees of freedom (or possibly n if a no intercept / no regressor model is fit). So won't some tests be off?Benhttps://www.blogger.com/profile/03460160852554584365noreply@blogger.com