In microeconometrics, it is sometimes advised to always use heteroscedasticity-robust standard error (instead of testing for heteroscedasticity and using generalized or weighted least squares or something else). This is the case because heteroscedasticity or autocorrelation / serial correlation affects the interval estimates only but not the point estimates.
If I am applying OLS in a time series context,
- is there a similar advise to simply address autocorrelation (even if disputed)?
- does it depend on whether the time series is short or long (small vs large T)?
I asked a similar question before, but this one is much more specific to the time series context, and I am happy even with suggestions not every agrees with.
