QUESTION:
Can you tell me what the
advantages and/or disadvantages are of using a regression on residuals
procedure to control for
time in time series regression against the more usual method of including time
as an independent variable?
ANSWER:
What should be done is
to incorporate contemporaneous and lagged effects of possible exogenous series
and then to add whatever noise structure is
necessary. The noise structure could be as simple as an
adjustment for an unusual value, or a
seasonally "usual value", or a level shift or a time trend(s).
Additionally, there may be some stochastic or
probabilistic component that might be needed such as an
AR(1) process or more generally an ARIMA
process.
Don't despair AUTOBOX
speaks to these issues as it tests and automatically adjusts for these
potential
violations of the Gaussian assumptions.