Estimation of Granger Causality through Error Correction

[upani]

Dear All,

As we know granger causality has two channels in ECM framework. The first one is through the first differences of the variables ( standard granger causality test with stationary variable or taking the first difference). The second one is through the error correction term(s). I am not able to understand how the the granger causality is tested through error correction term(s) (see Yang, Bessler and Leatham, 2001).

I will be grateful to you if you can suggest some ways to do it.

With regards,
Upananda

[TomDoan]

You've asked what's effectively the same question three times now. They're testing weak exogeneity. The word "Granger" doesn't even appear in the paper (except in the references). Weak exogeneity is one part of Granger non-causality in a VECM, but the paper doesn't look at the joint test with the lagged differences. If you want a test for Granger non-causality in the context of a VECM (again, that's not what they did), just run an error correction regression and do a joint test on the error correction term and the lagged differences of the "other" variable. That has standard asymptotics. Note, BTW, that the only two interesting hypotheses are alpha=0 and alpha=0 plus lagged differences=0. Lagged differences=0 without also alpha=0 is a test that has no real meaning. See Short- and long-run causality.

[upani]

Dear Sir,

As i am new to the rats and programming. Can you just tell me how to do the the joint test. I have tried write by looking at some of the post. Whether the code will be like this ?

@JOHMLE(LAGS=5,det=rc,cv=cvector)
# LFP LSP

equation(coeffs=cvector) eq
# LFP LSP constant

system(model=vecm)
variables LFP LSP
lags 1 to 5
ect eq
end(system)

estimate
linreg DLFP
# DLFP{1 to 4} DLSP{1 to 4} z{1}
@RegSTRTest(threshold=z,d=4)

With regards,
Upananda

[TomDoan]

No. Just

linreg DLFP
# DLFP{1 to 4} DLSP{1 to 4} z{1}
exclude(title="Test for GC in VECM")
# dlsp{1 to 4} z{1}

at the end. @REGSTRTEST is a test for threshold effects and has nothing to do with Granger causality.