Hi all!As described in Hamilton 1994 p.610, I do regression on the seriesy_t = a + b*y_(t-1) + e_t, where e_t is a MA(k)-process. Say k=3 as an example.In Hamilton, it says that due to the MA-process, my t statistics should be adjusted with sqrt(S_T^2/lambda^2), where S_T^2 is the standard deviation of the residuals and lambda^2 is given by another formula involving laged residuals.The Hamilton book describes the case when testing for a=0 and b=1 jointly.I want to test a=0 and b=1 individually. In Splus there is an in-built Newey-West correction (which is the described in Hamilton). summary(ols.fit,correction="nw",bandwidth=3)I do this on my ols.fit, but I find that the Newey-West factor that is used to adjust my t statistics are different for the two variables, a and b. How can that be?I can´t see that neither S_T or lambda should be different for the two tests.Could anybody help me out here?muzzex

if that functions comes from the add on SplusFinmetrics ( or maybe even if it doesn't ), a good person to ask would be eric zivot or youshould join the Splus Newsgroup. Someone on that list might know but Eric would be the best to ask. i think his email is e.zivot@washington.edu( I use R now but I used to use Splus and SplusFinmetrics heavily ) but you can google him. also, he's a very busy person so it's probably best if you don't say that i recommended him. he gets flooded with questions because he wrote a lot of the econometrics tools that are in Splus and he's a real expert. mark

Using R, in the package NLME there is a function called GLS which I use frequently to fit models like this.