VAR models on raw or filtered/smoothed data?

[Paulius Pranskevicius]

Dear all,
I just started to learn time series analysis and I'm reading Becketti's book at the moment. One general question that I didn't catch an answer to was, do I apply VAR models (as well as other models) on raw data or smoothen/filtered data? To be more precise, do I want to, for instance, de-trend, fix autocorrelation, and/or make my series stationarity prior to applying VAR models?

In general, I find it very hard to extract any general steps of how time series analysis proceeds, that is, what are the first steps, second, third? Are they any general "ways" how one analysis time series, I mean, maybe it makes way more sense to first test stationarity, autocorrelation than checking trend, cycle, etc?

Hope my question is not too confusingly written.

Thank you for your answers!

[Justin Niakamal]

do I apply VAR models (as well as other models) on raw data or smoothen/filtered data?

If you have monthly or quarterly data, I would use seasonally adjusted. If you have annual, it makes less sense to do any sort of adjustment.

do I want to, for instance, de-trend, fix autocorrelation, and/or make my series stationarity prior to applying VAR models?

In general, you should work with stationary series in a VAR framework. If you have cointegrated variables, then you can use a VECM. If not, use first differences on the non-stationary variables. For other issues check help var postestimation .

A good starting point for you questions about how to proceed is this Statablog and Bruce Hansen's notes (both below)

https://blog.stata.com/2016/08/09/ve...ions-in-stata/

https://www.ssc.wisc.edu/~bhansen/390/390Lecture25.pdf

Hope this helps.