17.05.2013 12:38, Pindar:
Hi there,
I have a panel of bank data for 12 years and investigate on the
effect of interest rates on risk-taking.
Although every characteristic of the dataset says DPANEL I have
some problems getting correct estimations.
The main problem is one of multicollinearity: The time dummies and
two of my regressors lead to omitting two time dummies. Since not
all of them are significant I could kick them out before and have
the same time dummy specification through all models. The
multicollinearity problems starts in FE estimation and stays in
DPANEL.
Actually I wonder why this happens, and whether this is a signal for
misspecification.
Since
including the significant time dummies is crucial, I started to
replicate the --time switch.
This works for the GRTEL DPANEL method, but how to tell GRETL to
perform --dpdstyle without the --time switch?
Is this possible?
I found in Roodman (2006/2008) p. 42 the necessary "gynmastics" for
'Difference GMM' in stat code:
This example exactly imitates the regression for column (a1), Table
4 in Arellano and Bond (1991):
forvalues y = 1979/1984 { /* Make variables whose differences are
time dummies */
gen yr`y´c = year >= `y´
}
gen cons = year
xtabond2 n L(0/1).(L.n w) L(0/2).(k ys) yr198?c cons, gmm(L.n)
iv(L(0/1).w L(0/2).(k ys)
> yr198?c cons) noleveleq noconstant small robust
What is he doing here? This is what the --dpd flag does, right?
I detect the the loop, but don't know how to adapt the code in
HANSL.
(I have no experience with STATA code)
Thanks in advance
Leon