[Gretl-users] Re: Newey west standard errors

On 15/03/2026 05:50, Ramki S wrote: Hi, thanks for testing. First of all, it must be said that, when it comes to robust estimation, there are tons of slightly different ways to do things and a single different detail can affect the results visibly (especially in a small sample like the one you're considering). Take a look at Chapter 22 of the gretl User's Guide. Judging by the formatting of the output, it looks as if Gujarati and Porter are using Eviews; if that was true, I don't know what the precise formulae Eviews used, so... However, they seem to use a lag truncation parameter equal to 3, rather than the automatic value used by gretl, which is 2 in this case. You may want to use the "set hac_lag" command. But even in that case, result don't match exactly, although they qualitatively do. That said, I believe the results reported in the book are a little weird: the OLS coefficients reported on page 451 (the ones you posted) are different from the ones reported on page 336 (example 10.2), albeit only at the sixth decimal digit. That makes me thing that Gujarati and Porter may have used for the two tables two slightly different versions of the same dataset (decimal digits, perhaps?), and the table on page 451 may be impossible to replicate with publicly available data. ------------------------------------------------------- Riccardo (Jack) Lucchetti Dipartimento di Scienze Economiche e Sociali (DiSES) Università Politecnica delle Marche (formerly known as Università di Ancona) r.lucchetti(a)univpm.it http://www2.econ.univpm.it/servizi/hpp/lucchetti -------------------------------------------------------