Why do you want to estimate it manually, if you already estimated the system in Eviews? Anyway, if you do not want to use Eviews further, then you can save the estimated error correction terms (cointegrating equations or cointegrating group) in the proc menu. Beware, these EC terms will already be based on some restrictions. Alternative restrictions would lead to different results.

Please join this Group .

https://www.facebook.com/groups/macroeconomicmodelling/

This is Group Macroeconomic Modelling, you can post any question you have or help fellow economist with their queries

This is my take Sidra, hope they will help you:

1) Don't use EViews for VECM. When I tried to replicate some important pepers I couldn't do it in there (versions previous 8). Use for example JMulTi or Structural VAR which are free btw, if you tried those, you will find very different coefficients!!. Moreover the use of the determistic terms in some especification is hard to reconcile in theory, see for example the texbook of Juselius, The Cointegrated VAR Model, chapters 5 and 6. At least check if the variables are I(1) or I(0), as Pandelis suggest.

2) But if you want to use Eviews and calculate the long run relation manually anyway (i supose that by manually you´re saying that you take the coefficients from Eviews and in a Excel sheet you just evaluete it) at least chech for the significance in the alphas (loading matrix) in the two coint system, see week exogenous tests literature... You probably want to check for reductions on the VECM, check Applied Time Series Econometrics, Lutkepohl.

3) You can reduce a system of equations to one equation, at least in theory. It is refered as I(0) reduction, David Hendry has good examples. One is in the PcGive 13 manual vol II...

4) Be careful how you use the estimations and where you are going to present those if you are not trully confident of your analysis.

Sorry for the mispelling, i was typing fast and hurry... good luck!

Hi, I am having the same problem. I am estimating the long run relationship among five variables in CATs in RATs. I've got two cointegration vectors which are un-identified. So I have to impose restrictions and I am not sure how to do that. The software helps in that and may suggest solutions but I would like to understand the basis of that. Can anyone help?

Hi Hana,

I´m not a full expert but I have faced the problem that you have and this is what I did:

1) CATs is an old software, and great one --actually the guys that start the VECM literature have their codes in RATS/CATS-- but the problem is that it has few web economies, few people knows it now, I would compare it with an old language like latin. But anyway, if you want to estimate there it is fine, the downside is that the way you impose the restrictions is a little bit complicate and you have to know about matrixes. The restrictions look like this: beta = H*Phi. You need to fill the H matrix(es) to get the restrictions you want. The manual that shows you how.

2) If you want to understand what you are doing I highly recomend the textbook of K. Juselius, The Cointegrated VAR Model, you can have access to the data on the book and replicate the results --don't remember if there is also the code--. Try to get the paper of Hendry and Juselius (2001) "Explaning Cointegretion Analysis: Part II" (Part I, is also a great one!). And finally the book of Lutkepohl, Applied Time Series Econometrics: there is an example there of how to impose restrictions, one with three variables and two coint rel, you will understand more how to do it.

I hope you get it right.

best regards,

got back to the theory!! there are reasons why > 1 coin. Johansen & Juselius papers jointly test the PPP & UIP within a VAR system and they reported > 1 coin according to theory. My recent paper of Global economic review (2012), 41:3, 259-277 also did similar analysis.

I've read papers reported 2 or 3 coin. But instead of reporting 2 or 3 ECTs, they did a joint hypothesis on ECTs and reported a significant F-test.

rgds,

chan

Hi,

Thanks indeed for your reply. It is helpful actually.

The problem is that my supervisor does not agree to use Eviews at all. I have to work on CATs and hopefully, it will be fine.

I have another question about erogeneity. If the alpha is not significant, it is indication for weak exogenous variable if I am right. So, does this mean when I set alpha equal to zero so the long run relationship will just disappear from the corresponding equation only. For example lets say we have [y,x,z], then if z a weak exogenous variable then it will appear in the CI, its alpha equal to zero, and in VECM, we are going to have an equation for delta(x) but without EC term. Am I right? please if any can explain this I would really appreciate it.

Regards

Hanan

Hi Hanan,

I agree with your supervisor in not using Eviews, maybe for the unit root tests but not for the VECMs.

Yes, if a variable z is weakly exogenous that means that the z equation don't have feed back from the level variables, and you can estimate efficiently the model with two equations [y,x]. But notice that when you have 3 variables and two coint equations you need to test two alphas (one per coint equation) by restricting the two simultaneously. If you have more than one variable that is WE then you have to test them jointly as well. The test is a Wald-type test with a Xi-squared distribution with "r" degrees of freedom (r, is the number of coint. rel.)

I'm almost certain that you can do this on CATs, don't know exactly what is the command. See also Juselius (2006) pp. 197-198.

Best,

Rodrigo

Hello all,

I need help and I hope some one will respond.

I have a problem reading the long term cointegration equation I estimated using Johansen Cointegration test. The test confirms that there are 3 cointegrating equations (2 lags).

1. Which one is my long term equation? Is it the one which reads "Normalized cointegrating coefficients" or "Adjustment coefficients"?

2. In the former, two of my dependent variables are normalized (to 1.000). Does this mean that their coefficient is zero in the long run and do not have any role?

3. If I have 3 cointegrating equations, I have to look at the output which says "3 Cointegrating Equation(s)". Am I right?

Thank you!

Dear Pandelis Mitsis,

Thank you very much for your response. I appreciate your help!

I have 5 independent variables, all I(1). The model passes normality, heterskedasticity and serial correlation tests. The cointegrating parameters are all significant, 4 of them with expected signs and 1 of them not but I will try to explain why. Given these,

1. Just to confirm, my long run equation is the FIRST equation [which says 1 Cointegrating Equation(s):]?? because there are 5 Cointegrating Equation(s) reported in my model.

2. Concerning the long run equation from Johansen test: can I derive numerical inference from them? Like an increase in x by 1% leads to a 3% rise in Y? Or just the direction of relationship? Where can I find such an inference?

3. I read from another post that the adjustment coefficients just under the FIRST cointegrating equation are my error correction terms. They are all negative but some of them are insignificant. Is this a problem?

4. I haven't done IRFs/variance decomposition. Is it a must to do it?

Thank you once again and I'm looking forward to reading from you!

Thank you for your clear explanation and time!!!!!!!!!!

Please what o a situation where I have 7 independent variables and 3 cointegrating equations?

could b three  different coin path among the 7 variables. In other worlds, three variables may bear the adjustments towards long run equilibriums. Kindly do the theoretical and diagnostic checks before final conclusion.  

Thanks, I appreciate

However, when there are multiple cointegrating vectors ARDL Approach to cointegration cannot be applied. Hence, Johansen and Juselius(1990) approach becomes the alternative .

Source: Autoregressive Distributed Lag (ARDL) cointegration technique: application and interpretation Emeka Nkoro1 and Aham Kelvin Uko2

https://faculty.washington.edu/ezivot/econ584/notes/cointegration.pdf

A cointegrating vector contains coefficients that produce a linear combination of I (1) variables that is I (0). It is common to normalize the first coefficient to 1. A system of 3 integrated variables that are also cointegrated can have either 1 or 2 cointegrating vectors. Multicollinearity occurs when two or more predictors in the model are correlated and provide redundant information about the response. A common way to detect multicollinearity is by using the variance inflation factor (VIF), which measures how much the variance of a regression coefficient is inflated due to multicollinearity. The value for VIF starts at 1 and has no upper limit. A VIF value that exceeds 5 or 10 indicates a problematic amount of collinearity. Common Method Bias (CMB) is a systematic error that occurs when the same method is used to measure multiple variables in a study. It can lead to spurious relationships between variables and can affect the validity of the results. There are several methods to test and control for CMB, including Harman’s one-factor test, marker variable technique, and partialling out the first unrotated factor as a “general factor” attributable to social desirability or common method bias.