FEIVESON, ALAN
>
> Hello - I am thinking of using -xtregar, fe- to compare 4
> time series
> (assumed to be AR1 - ha!) to see if they can be considered
> to have arisen
> from the same process. For starters, I wish to use the
> simplest kind of
> comparison model - to check for series-specific offsets by fitting a
> fixed-effects model. Never having done this before, I have
> some questions.
>
> 1. Each series was observed at different unequally spaced
> times. So under
> conditions can I get away with interpolating them to get
> values at common
> equally spaced points?
>
> 2. These series all exhibit some sort of trend, but it is
> not linear or
> easily described by an simple function of time. What if I
> use the average
> series to represent trend, as would be the case if this
> were some sort of
> ANOVA? For example if ybar is the variable representing the
> average series,
> I would attempt something like
>
> . xtregar y ybar,i(series_no) fe
>
> What does using the average do to the standard errors? Is
> there a better
> way?
One guess is that interpolation will produce something
smoother than most empirical series, and so mess up
the autocorrelations. No doubt you could roughen them
up in some way.
One limited handle on this is that at least two
interpolation methods have been coded in Stata,
the linear interpolation method of -ipolate-
and the cubic interpolation method of -cipolate-
(SSC). (The latter is not a spline method; as
far as I am aware none of the spline routines
written in Stata do cubic spline _interpolation_,
a gap which itself needs interpolation.) Thus
you could interpolate two ways and see whether
and what results are sensitive to how you do it.
Nick
[email protected]
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
