adf test

["Leeds, Mark" <mleeds@mlp.com>]

i am performing the augmented dicke fuller test in Splus

using some code i found on the net which works fine.

 

when your doing the adf test, the lag length for

the regression is not obvious. so,

there's an algorithm recommended

in Dr. Zivot's book in

which you start out with a certain

lag length determined by a formula

( Schwert (1989 ). then you do the regression

and if the coefficient of the last lag is not siginiicant,

you decrease the lag length by 1 and do

it again. you keep doing this until you get

significance on the last lag or you run out of lags.

 

i want to automate this process which

is not that much of a problem but

I thought the lm object output could

be used to do this. I thought the

object woiuld contain the t values

of the coefficients ( so i could use

them in the algorithm ) but i looked

and I don't think it does. in that case,

i can calculate the t-statistics

using the other lm output

( assuming it's all there ) but

i just wanted to make sure that

this is the case and i didn't miss

something.

 

                            thanks.