Schwert suggested l = c((T/100^(1/4)) with either c=4 for small sample (T<100) or c=12 otherwise. As far as I know, this rule is widely accepted for the KPSS test too. Cheers, Greg
W dniu 2011-07-30 14:44, adkins.lee(a)gmail.com pisze: > I don't believe gretl doen not have a --test-downoption in kernel > estimation, which is what I believe KPSS involves. > > So-called 'optimal' bandwidth is usually a function of sample size. > Some use rules like 0.75T^(1/3) or 4(T/100)^(2/9). I believe that > Schwert suggested 12((T+1)/100)^(1/4) in picking lags for ADF tests. > With respect to KPSS, I don't know. Longer means less bias and less > precision. Shorter more bias, but greater precision. > > Cheers, > LA > On , "samyto ..." <samyto(a)hotmail.fr> wrote: > > > > > > > > > > > > Dear Gretl Users, > > > > My question concerns KPSS unit root test. After selecting my serie > and clicking on KPSS test; the default lag is set to 8. > > How do I find the best lag for my test ? Is there an automatic way > to select it ? Is the default lag the best one ? > > > > > > Thanks a lot for helping ! > > > > > > Samy > > > > > > > > > > > _______________________________________________ > Gretl-users mailing list > Gretl-users(a)lists.wfu.edu > http://lists.wfu.edu/mailman/listinfo/gretl-users -- Pozdrawiam, Grzegorz Konat
Schwert suggested l = c((T/100^(1/4)) with either c=4 for small
sample (T<100) or c=12 otherwise. As far as I know, this rule is
widely accepted for the KPSS test too. Cheers, Greg W dniu 2011-07-30 14:44, adkins....@gmail.com pisze: I don't believe gretl doen not have a --test-downoption in kernel estimation, which is what I believe KPSS involves. -- Pozdrawiam, Grzegorz Konat |