Johansen Cointegration in Eviews

[Kalec]

Hi

When I do the Johansen Cointegration test in Eviews, I see that the default setting was 1 to 4 lag interval.
Can anyone please tell me how to decide the number of lag interval to be used in EViews for this Johansen cointegration test?

Thanks and kind regards

Kalec

[theologos]

Hi,

Initially run the VAR model of your interest without caring for the optimal lag structure. From the VAR window now select: view/ lag structure / Lag length criteria.
A new window appears where the Lag specification is desired. Specify the maximum lag length (depending on the frequency of your data) and press ok. After re-running the VAR model with the optimal lag length you may proceed to the Johansen cointegration approach.

Regards

[Kalec]

Thanks theologos

Below is the outcome from Eviews, and I noticed that the optimal lag length suggested by different criteria is different. I wonder which of the criteria should I rely more on? and how could I support myself relying on such criteria please?

ScreenHunter_01 Aug. 17 14.44.jpg (30.54 KiB) Viewed 55458 times

ScreenHunter_02 Aug. 17 14.45.jpg (113.31 KiB) Viewed 55451 times

Thanks

Kalec

[theologos]

Hi,

As you have realized there is no unique way to select the optimal lag length, therefore is quite easy to support your selection. Whenever multiple lag lengths appear two strategies can be followed:
1) Consider all the available criteria and let democracy to do the job
2) Choose one criterion (I usually choose the LR statistic)

Regards

[Kalec]

Thanks for your help theologos

[cointthesis]

I am also trying to determine the maximum lag length I should indicate for my lag length criteria test, and I was wondering if you could advise me.

I am examining daily (5 days a week) data from 2005-2010. What maximum lag length would be appropriate for such data, or how could I determine the most suitable maximum lag length?

[wedira]

Hi,

Initially run the VAR model of your interest without caring for the optimal lag structure. From the VAR window now select: view/ lag structure / Lag length criteria.
A new window appears where the Lag specification is desired. Specify the maximum lag length (depending on the frequency of your data) and press ok. After re-running the VAR model with the optimal lag length you may proceed to the Johansen cointegration approach.

Regards

Hi,

does it mean that I need to run VAR in levels, which aren't stationary (they are I(1)) and then:
1. I find the optimal lag length, then
2. I apply this lag lengt to Johansen cointegration test, then
3. I run VEC estimation with the number of lags and number of cointegrating vectors found in point 1. and 2.

Am I right?

Thanks in advance for your comments, regards

[Arun.stat]

Hi Wedira, you are done completely correct. As you said correctly to determine the optimal lag length, you need not make data stationary. This is because here you are not estimating the model coefficients. Non-stationarity matters in deriving the asymptotic properties (i.e. distribution, consistency etc.)of the estimators of coefficients, especially when underlying process is I(1)

HTH

Thanks,

[wedira]

Thanks :)

[dealsfe]

how can i read Johansen cointegration test? i am a beginner. what should i read?

[hgaronfolo]

Hi,

I have used the method of I) estimating a VAR, II)View > lag structure > lag length criteria.

The lag selection indicates that I should use 8 lags. However, I am estimating my model on annual data, so 8 lags seems quite high, no?

The variables in my model are Market Capitalization and GNI.

Furthermore, if I increase the included lags in the lag selection to e.g. 12 lags, the lag selection will indicate that 12 lags is the best option. The optimal number of lags to include seems to go towards infinity, why is this? Can somebody explain this to me?

Thank you very much.

[fboehlandt]

Hello everyone,
I'm uncertain as to the validity of the approach suggested here. It is the purpose of a VEC to model the long-term cointegration relationship as well as susceptibility to short-term shocks at the same time. Thus, I'm not sure you can estimate the lag structure independent of the cointegration specification. You could calculate the IC for various lag lengths and deterministic trend assumptions and record the results in a matrix. However, I'm not sure if the IC are comparable. The only solution I have found to so far was to justify lag lengths on economic grounds. Someone mentioned the approach outlined in this thread is according to the EViews manual. Can anyone specify where? I was not able to find it. I would also appreciate any feedback on other solutions. Thanks

[Stephan122013]

HI, I am running the Johansen cointegration tests on eviews. Can anybody show me the code to freeze the output table so I save some important elements in a matrix. I have 29 series for both items irq and srq.

I tried the following and it didn't work:

for !i=1 to 29
freeze(table) coint(a,4) irq_!i srq_!i
next


Thanks a lot!

[vas001]

Hi
I have uploaded a word file which may help you guys. Please let me know whether it helps you.

Vasanth Naik

[wailycalcal]

can anyone help me?
is there a long run relationship or not in time series data that i regress?(significant level=0.01)
one of independent variable is significant at 2nd difference while others are significant at 1st difference
should i use VECM, check the short run relationship using Granger causality test?