This looks like an exam question. Please see your instructor.

Assuming all the variables are I(1) and your objective is to find a long-run relationship, VECM is generally used. Please note that "all variables are stationary at I(1)" is not apposite.

I have two variables non stationary at level and non stationary at first difference, so i transform the data to first difference, and they became stationary at level. Can I perform VECM model with those two variables. Thank yo

If all your variables are l(1) you can either differentiate them and then use any model suitable for your research question (panel, ARMA, VAR,.....). Or you can take advantage of this integration and use the cointegration and VECM model that gives you the long and short run relationships.

There are two questions in this debate.

My advice to Nabaraj Gautam and Hammache Souria is that they should get a good textbook and try to understand the concepts of stationarity, non-stationary, integrated, and cointegrated series. There is a lot more to understanding these matters than could be answered on a forum such as this. I usually recommend Enders (2014) Applied econometric time series, Wiley to economists who wish to be familiar with time series. It is a nice compromise between applied time series and theory. There are several graduate-level texts that provide additional material that may be needed.

The following points are non-rigorous and are intended as a guide

@john frain students have been known to copy incorrectly. Don't blame the examiner just yet. If the stationarity condition was correctly written would you see any possibilities. I did my last time series model work many years ago but a multiple choice question on research methods seems strange don't you think? David Booth

Unless of course it is a methods course exam. Best wishes to all.

David Eugene Booth If the stationarity condition was correctly written, the logic of my answer would be similar to 9 in my previous answer. You need to know the context to answer the question. Sometimes Johansen would be best, other times a combination of Johansen and ARDL (bounds test) and on other occasions, I might use ARDL. If there was a previous part to the question it might help. Take Care, John

@ John Frain, I certainly agree with your comment. I hope the instructor would as well. Best wishes to all, David Booth

Anis UR Rehman Your video does not even mention I(1) variables. Applying the methods in the video to I(1) variables will lead to spurious results in the absence of cointegration. There is also a mention of the stepwise regression. Stepwise regression is not a valid method of finding an appropriate regression as it usually leads to the wrong results.

conduct the EG cointegration test to det, then follow up from thereermine long run effect

Nabaraj Gautam If all variables are non-stationary at level but stationary at the first difference I(1), then Johansen cointegration and VECM model are used to determines the long-run and short-run relationship respectively.

Hammache Souria If two variables are non-stationary at level but stationary at the first difference, then Johansen cointegration and VECM model are used to determines the long-run and short-run relationship respectively.

If all the variables are stationary in the same order then it is theoretically suggested to use the Johanson Co-integration test instead of ARDL bound test for cointegration. But, it is not wise to use VECM if there is no co-integrating equation in the model. You can use ECM when you intend to find out the long-run relationship.

If all variables are stationary at the first difference I(1), then the vector error correction model (VECM) is a good choice for modelling the relationship between the variables. The VECM is a dynamic model that allows for both short- and long-run relationships between the variables.