Time series


Fortran must die
Time for my biannual return to this :p

Here is a basic question I have been wondering at for a long time. When do you need to do time series and when do you not. For example, we provide a service, the result is in the future. Does that mean we should do (far more complex and difficult to analyze) time series regression? Or regular regression.

In reality most things have impact in the future not present. How can they not. Much of the analysis of time series deals with autocorrelation - but that is a pretty simple reality. You just use HAC standard errors and don't worry about it.

I think the real key to multivariate time series is that you do it when patterns change over time (another way of saying there is non-stationarity in the relationships). Or you do it when the impact covers multiple periods. So X has an impact now, and at some point or points in the future.