Re: [Gretl-users] I: Dumitrescu-Hurlin test and Granger causality test
Dear Sven,
yes, you are right :)
Thanks a lot
Giuseppe
On Thu, 2017-03-23 at 18:00 +0000, Giuseppe wrote:
Dear all,
> I am trying to compute the pairwise Dumitrescu-Hurlin test (a
Granger-causality test for panel data) in gretl.
> In order to do this, after having run a separate:
> var # y1 y2
> one for each cross-sectional unit in the panel, I need to retrieve
the value
of the F-test for the null hypothesis that no lags of variable
y2 (y1) are significant in the equation for variable y1 (y2).
> gretl computes these Granger causality tests and shows them by
default
> after the command "var", but it seems that this value
isn't stored
in any accessor...
> Is there an easy way to retrieve it?
Not that I'm aware of (and I may be wrong and/or there may be a
function
package for it, although I don't think so), but what about the
following
workaround:
For a Granger causality test from x on y you don't need to estimate a
VAR, you only need the equation for y. So I would estimate the single
equation for y with OLS, including the relevant lags, and then
testing
Granger causality with the "omit" command. Something like
this:
list causelags = lags(p, x) # p is the lag order
list targetlags = lags(p, y)
ols y const targetlags causelags
omit causelags --test-only --silent
yxteststat = $test
yxpval = $pvalue
Of course if you want to test in both directions, you have to do this
workaround again for x on the LHS (possibly with a loop to avoid code
duplication).
hth,
sven