i have done an article of VAR and VCM , i have got the same results

thanks sir

1i have read what is written professor but i think to do the co integration butter than VAR model in this case.also you should check the stationirty ,it would be in 1(1) level .

you should check the stationarity first, if you have I(1) data, it is better to use VECM. otherwise, you can use VAR. ( you may also think about ARDL )

I presume that you are dealing with the logs of wages and food prices. If you have no cointegration with your three variables the implication is that there is no long-run relationship between them. (they are not cointegrated). I would have a look at the possibility of adding other variables to the analysis. It would have been better if you had started with a review of the economics of your project and started with a more general model or some encompassing model of competing theories.

If there is no cointegration in levels the basic approach would be to use a var in first differences, In your case the first differences of the logs of the two variables correspond to wage and food inflation. Again your main concern would be missing variables.

Another possibility is that there is some break in your series due to a change in the method of calculation or some policy change (e.g. a social contract operating over some sub period). Have you checked that there is no seasonality problem?

Question 2. If first differences are stationary there is no reason to assume that there is cointegration.

Question 3: No.

Question 4: I would regard the normality tests as specification tests. Perhaps the non-normality is caused by some outlier caused by a break in the series caused by an event such as mentioned above. If this is the case modelling the event may not only remove the non-normality but also give you cointegration.

Johansen is a lot more difficult than just applying the tools of an econometric package to a data set. You do need a deep knowledge of the economics of the processes and of the data. If you intend to work with this methodology you should read Juselius (2006) The Cointegrated VAR Model, Oxford.

Dear John C Frain,

Many Thanks for your valuable comment. I really appreciate and thankful for giving your valuable time to help me. Last couple of days I was thinking about your comments.

Structural Break & seasonality: Actually, before checking the stationarity, I checked the structural break but I did not check for seasonality. However, now again suddenly some questions arises my mind by seeing your comments regarding structural break and seasonality.

Structural Break:

I checked the structural break (SB) in STATA firstly by fitting a regression model as ' reg W1(wage type 1) W2 P ' and then I estimate the SB for unknown break 'estat sbsingle'. Here my questions are :

1. Shall I check the structural break (e.g., single break) of the series by fitting normal regression of the variables in their levels? (Both the variables are non-stationary at their levels),

2. And secondly, which direction I will fit this regression model like, food price = wage, or wage = food price, ( When economic theory based on the weight of the fitted coefficient, so I have no idea in advance before finding the suitable VAR or VECM )

3. Is there anyway to check the the SB for single series? and how feasible it will be? (I found several test in Eviews like Perron and Bai, Global etc. but confused)

Seasonality:

All the series are monthly. But I am worried about the use of lag length in all of these time series test. However, I used 'varsoc' commad for each series to get the appropriate length. My questions are:

1. If the normal first difference make the series stationary then still, we need to test seasonal unit root?

2. what is seasonal difference mean for monthly data, is it D12 (twelve difference) like first difference or difference after 12 lags (or meaning the same)?

3. If we use de-trended or de-seasonal (-filtered) data in fitting VAR then are not we lose main characteristics of the data before fitting the model? (Though you answer it before but for my correctness, you are saying that variables should be in their levels before fitting VECM, I mean all the commands like varsoc, vecrank etc. followed by original variables in levels)

Sorry for such long questions. I also sorry If i asked very basic questions.

Dear Alqaralleh and Alalaya, Thanks also for the comments.

I think that it is preferable to look at your variables themselves rather than subject them to rely on statistical tests. Are there breaks in the method of calculation of the series. Were there policy changes that had an effect on the variables (e.g. national wage agreements, price controls etc.). This kind of problem is discussed in Juselius (2006). Incidentally the appendix to that book sets out the steps that should be completed in a cointegrated VAR analysis. Some of this book is not for the mathematically challenged but the appendix and the discussions of dummy and intervention variables is straight forward. In a VECM environment stability is usually examined with rolling estimates of the system.

If your seasonality is fairly constant you could add centered seasonal dummies to your Johansen setup.

If there is no cointegration are you missing some variables? If you are sure that there is no cointegration and all your variables are I(1) then you can do a VAR in first differences and look at impulse response functions and Granger causality.

dear,

please. If you are sure that there is no co integration at 0(1) or after ADF test, you should move to solve the problem by adding new variables,then you can check structural break ,and to use the VAR model.