In article <9qrjkc$buq$[EMAIL PROTECTED]>,
David B <[EMAIL PROTECTED]> wrote:
>When the exogeneous variable is an integrated process, standard procedures
>are not valid (spurious regression).
>
>However, if you consider the exogeneous variable as determined (non
>stochastic), regression procedures work again.
>
>My interpretation is that regression procedures are always valid since
>nothing can prevent you from considering the variables as fixed. But that it
>is shown that in the very special case where the exogeneous variable is
>*truly* generated by an integrated process (if such thing can exist) it
>won't work.
Before we can attempt to resolve this paradox for you, you'll have to
state more precisely what your definition is of "an integrated
process", and in what sense you think using a variable generated by
such a process as an explanatory variable in a regression is "not valid".
Quoting from whatever source claimed that would be best.
There is certainly nothing wrong with using standard regression when
an explanatory variable is randomly generated, from whatever sort of
stochastic process you please, as long as the regression residuals are
independent. Nor is it wrong to do regression when the residuals are
correlated, as long as you compute standard errors taking this into
account. In both cases, I assume that your goal is prediction, not
validation of causal hypotheses. There may be something you have in
mind that is wrong, but you'll have to explain more for us to tell
what it might be.
Radford Neal
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
