Graduate Robustness of time series analysis

[monsmatglad]

I have a time series model constructed by using ordinary least square (linear).
I am supposed to provide some general comments on how one would improve the robustness of the analysis of a time series model (in general).
Are there any general advice apart from expanding data, making it more frequent and increasing the number of lags?

Mons

 

[BWV]

None of those makes a T.S. regression more robust other than more data. Robust means that the model can work if the assumptions about normally distributed variables, homoskedasticity, no autocorrelation etc are relaxed. Examining outliers and the impact they have is a good starting place. Testing the model out of sample using GMM to correct for autocorrelation are other methods.

 

[FactChecker]

I
Are there any general advice apart from expanding data, making it more frequent and increasing the number of lags?

Increasing the number of lags will have the opposite effect if you are talking about including terms with less statistical significance.

It's not clear to me what "making it more frequent" means. If that means decreasing the time step of the time series, then that may help. Especially if the current time step is too large to capture important high frequencies. Do not make the mistake of assuming that the Nyquist sampling frequency is adequate. It is the minimal sample frequency that will give perfect accuracy if you have an infinite time-length sample. Any finite time-length sample gives less than perfect accuracy.