[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: matsize & single exogeneity & autocorrel. in panel
I would be very grateful if someone could answer next questions.
First, I was trying to estimate panel data model in Intercooled Stata 7.0,
with N=930 and T=6, i.e. total 5580 observations and 12 variables, but
without any success. Is the only solution working in Stata/SE 8.0 where the
upper limit is 11000?
Second, since I suspect autocorrelation and single exogeneity problem in
the model (some regressors are correlated with individual effects), my
question is: what should I apply first - test for single exogeneity (Hausman
test: fixed effect (FE) v.s. random effect (RE)model) or test for
autocorrelation in RE model?
I thought to apply first Hausman specification test, because if there is
singly exogeneity problem in RE model, then AR(1) test is based on biased
and inconsistent estimates and residuals later. But, on the other hand, if
the remainder disturbances u_it follow an AR(1) process, and I test single
exogeneity first, Hausman test and Hausman-Taylor (HT) over-identification
test are in fact not valid. Any suggestions?
Third, I am not sure if I made mistake in defining HT set of instruments
A=[QX_1, PX_1, QX_2, Z_1], since following result of HT over-identification
test (FE AR(1) v.s HT AR(1)) is strange to me:
.xtreg lexAR ldistAR combordAR v97AR v98AR v01AR euAR ceftaAR croAR bjrAR
lpopexAR lpopimAR lgdpimAR lgdpexAR, fe
. xtivreg lexAR v97AR v98AR v01AR euAR ceftaAR croAR bjrAR (ldistAR lpopexAR
lpopimAR lgdpimAR lgdpexAR = combordAR QlpopiAR1 QlpopeAR1 QlgdpimAR1
Test: Ho: difference in coefficients not systematic
chi2( 12) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= -2.74 chi2<0 ==> model estimated on these
data fails to meet the asymptotic
assumptions of the Hausman test.
Thanks a lot in advance for all the suggestions.
* For searches and help try: