Multivariate GARCH causality testing

[willb]

Dear Tom,

I am interested in conducting variance causality tests from a multivariate GARCH (BEKK) model. Hafner and Herwartz (2004) propose a Wald statistic in their working paper, "Testing for Causality in Variance Using Multivariate GARCH Models." How can this be implemented in RATS? Is there a procedure that exists?

Thanks.

[TomDoan]

Sure. Just do a GARCH, then a TEST(ZEROS) instruction on the proper set of coefficients. In the GARCHMV.RPF example, the following would do tests for causality between Japan and the block of European currencies:

Code: Select all

garch(p=1,q=1,mv=bek,pmethod=simplex,piters=10) / xjpn xfra xsui
test(zeros,title="Test for variance causality of JPN to FRA/SUI")
# 11 12 20 21
test(zeros,title="Test for variance causality of FRA/SUI to JPN")
# 13 16 22 25

The first tests the A and B's at positions (1,2) and (1,3). (Because the A's and B's are multiplied transpose first in RATS, (i,j) is the effect of i on variance j), and the second tests the A's and B's at positions (2,1) and (3,1).