Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: ACPLOT vs AC

From   "Nick Cox" <[email protected]>
To   <[email protected]>
Subject   st: RE: ACPLOT vs AC
Date   Wed, 15 Oct 2003 14:23:48 +0100

Macneill, Stephanie J

> Hello. I have some time series data and would like to check
> whether there is
> any autocorrelation in my residuals. I generated a
> correlogram using AC and
> found that none of the autocorrelations were outside of the
> 95% confidence
> bands. When I repeated the analysis using ACPLOT I found
> that some of my
> points were outside of the SE bands. Which analysis should
> I be relying on?
> Why is there a difference?

-acplot- is a program I wrote in 1997, and it has been
lurking on SSC since then. In addition, it was
included in a publication in STB-46 by other authors,
without any reference to me that I can recall, not that
it matters. This continued a small tradition of large-scale larceny,
my -acplot- was (explicitly) based on an original -ac-
by Sean Becketti, itself published in STB-5 in 1992.

When Stata itself introduced an official -ac-, at some
later point, I imagine
that they changed the method of c.i. calculation,
as compared with either of these. That presumably
accounts for the discrepancies you noticed.

I'd add five very broad comments.

1. The theory of c.i.s for autocorrelation
is full of messy approximations. Different books
seem to emphasise different messes, arbitrarily
or as a matter of taste.

2. If your time series shows any hint of nonstationarity,
that theory non tenet aquam.

3. If the results of different formulae differ substantially,
it may be a hint that your sample is not large enough
to do much with it, in this respect.

4. Neither -ac- nor -acplot- knows if you feed it data
or residuals, and I doubt that the theory is exactly
the same, although this may come down principally or wholly
to a matter of degrees of freedom.

5. I would also plot residual vs last residual.

[email protected]

*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index