Re: Correlation: Sets of Economic Data

From: John Gregory (jaygreg90_at_hotmail.com)
Date: 07/18/04 

Date: Sun, 18 Jul 2004 01:04:13 GMT

>>The data in the short term are generally linear In the longer term, the
log is often linear<<

So what am I to conclude from this? That such correlations are acceptable?
Unless you're trying to say the general trends can be reduced to a linear
representation in the long run and that's how it's done by econometricians?
If not... then I didn't need to know about econometricians at this point;
they didn't move me closer to an understanding.

>> The really big problem, which makes simple, naive correlations
> next to useless, is that the observations are time-series, with
> large auto-correlations. Google <specious correlation<<

And this tells me nothing. Since you consider the correlation described to
be "simple" and "naive", how about a little explanation; like a prof would
force you to give before the class. Your Google reference took me nowhere
but to an exchange you were having with someone else. I don't want to read
your mail.

Anyone else care to comment?

"Richard Ulrich" <Rich.Ulrich@comcast.net> wrote in message
news:8c4jf05bukjh02mpelljcd5krjmskibiu5@4ax.com...
> On Sat, 17 Jul 2004 18:54:01 GMT, "John Gregory"
> <jaygreg90@hotmail.com> wrote:
>
> > A member of our investment group (manufacturing engineer) decided to try
> > finding correlations between various economic indicators and the US
economy
> > in general. During the presentation, someone raised the question of just
how
> > useful it was to attempt to use the Excel software function to find
> > correlations when the tool is generally used on linear, not curvilinear
> > data. We were all left with a question mark; no one knew if this was a
valid
> > argument. If it IS, please comment and suggest another tool if possible.
> >
>
> The data in the short term are generally linear.
> In the longer term, the log is often linear.
>
> The really big problem, which makes simple, naive correlations
> next to useless, is that the observations are time-series, with
> large auto-correlations. Google <specious correlation> .
>
> As a group, econometricians are probably the most technical,
> most sophisticated users of regression models.
>
>
> --
> Rich Ulrich, wpilib@pitt.edu
> http://www.pitt.edu/~wpilib/index.html