There are 2 non-stationary time series of different integrated orders. Also the co-integration test shows that series is not stationary. Now how can I build a VAR model and how I can interpret the output?
How did you conclude on this ? What are these orders ? What R program did you use ? In a paper I co-authored we have allowed for a "partially nonstationary" model in error-correction form. It was for VARMA models, not only VAR's. It was also not for R but in Fortran. The final submitted version is on ReserachGate but contact me for the published version.
Thanking you for kind inputs!
I used adf.test in R to check for stationary. As going ahead with non-stationary will result in spurious regression issue; even AR or VAR is only applicable for stationary series.
Moreover the issues are:
1) Series are not stationary and have different order of integration (in I(d) )
2) We can't apply Johenson test for checking of cointegration that needs the series of same integrated order that too order 1
Can I difference the series and do the operations with obtained stationary series? Will there be any effect on Granger causality, VAR model and Impulse response?
How many series ? What values for d ?
No. If you difference the series, you will get a non-invertible VARMA model.
Have a look at MÉLARD, Guy, ROY, Roch, SAIDI, Abdessamad, "Exact Maximum Likelihood Estimation of Structured or unit root Multivariate Time Series Models", Computational Statistics and Data Analysis 50, No. 11, 2006, 2958-2986. DOI information: 10.1016/j.csda.2005.06.009. I am not sure it will solve the problem but perhaps by differencing only the series with the highest d. Probably you will need a VARMA anyway.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R