VECM model and Granger causality

[Francois Durant]

Hi,

I would like to test (short-run) Granger causality in a VECM model with only one lag. I have found a document that does it for 3 lags. I thought I would be able to edit the code easily but somehow I am stuck:
https://socialsciences.mcmaster.ca/m...out%20full.pdf

Here is the data I have:

* Example generated by -dataex-. For more info, type help dataex
clear
input float(monthyear lconsumption lincome)
675    363     560
676   4982    8538
677   5008   15146
678  79305  265707
679  77967  354004
680  47894  211406
681  39764  155116
682  31204  149535
683  36509  161429
684  40262  178205
685  36904  145206
686  49472  196574
687  41860  172077
688 134783  593005
689 382292 2095681
690 512925 2605281
691 498831 2252573
692 493515 2075523
693 510854 1864678
694 629511 1949752
end
format %tm monthyear

Here is the code I have. The last line does not work:

tsset monthyear
vecrank lconsumption lincome, lags(1)
vec lconsumption lincome, lags(1)
test ([D_lconsumption]: LD.lincome )

Here is the error I get:
". test ([D_lconsumption]: LD.lincome )
LD.lincome not found
"

In the link provided, the code is very similar with 3 lags. I just do not know the syntax to get the short-term Granger causality test to work with one lag. Please help. Thanks.

[Francois Durant]

Hi,

I was wondering if anyone had a chance to look into it. Please let me know.

[George Ford]

Look at the e(b) matrix to see what you got as variable names.

[Francois Durant]

I have a [1,4] matrix. Here it is:


Would you be able to help me use it? Thanks!

[Francois Durant]

Based on the matrix content, I guess I do not understand how the variables are named. Therefore, I am having a hard time to input the proper name on "test". please help. Thanks!

[Francois Durant]

Is there someone who can help me extract those vectors and run the test? I am really stuck there, your help is much appreciated! Thank you in advance!!

[George Ford]

You need 2 lags to use the approach you've outlined above, which appears to be how people do it.

https://www.youtube.com/watch?v=UgbMUS9eWPw

[Francois Durant]

What if I only have one lag? I cannot change my model to 2 lags.

[George Ford]

When you have 1 lag, only the error correction term is in the model. There is no lag of the variables. When you have 2 lags, it includes the lags of the variables permitting the test. So you really only have 1 lag of the variables.

Run it, you'll see it in the results.