On Sun, 8 Jan 2012, Sung Jin Lim wrote:
> Thank you for answer
>
> I have checked the again my estimation procedure...
> and I noticed that there is the some mistake in AIDS case...
> that means that the estimated coefficients of AIDS model regressed by GRETL
> has the same coefficient...
> It is almost same in the 4th digit.
>
>
> But the Rotterdam Case...I am not sure..
> maybe there is a mistake on the estimation...method...I lost my confidence
> on it
> anyway..following is the code I wrote
> and I attached the data
[...]
I tried your script using gretl and R (R 2.14.1 with systemfit
1.1-10). The results were essentially the same. Here is my script:
<hansl>
open book8.xls
genr DivisiaQ = sw1*dq1 + sw2*dq2 + sw3*dq3 + sw4*dq4
genr wdq1 = sw1*dq1
genr wdq2 = sw2*dq2
genr wdq3 = sw1*dq3
genr wdq4 = sw2*dq4
genr ddp1=dp1-dp4
genr ddp2=dp2-dp4
genr ddp3=dp3-dp4
system name="Rotterdam"
equation wdq1 DivisiaQ ddp1 ddp2 ddp3
equation wdq2 DivisiaQ ddp1 ddp2 ddp3
equation wdq3 DivisiaQ ddp1 ddp2 ddp3
end system
# Unrestricted estimation
estimate "Rotterdam" method=sur
restrict "Rotterdam"
b[1,3] - b[2,2] = 0
b[1,4] - b[3,2] = 0
b[2,4] - b[3,3] = 0
end restrict
# Restricted estimation, iterated
estimate "Rotterdam" method=sur --iterate
foreign language=R --send-data
library(systemfit)
e1 <- wdq1 ~ 0 + DivisiaQ + ddp1 + ddp2 + ddp3
e2 <- wdq2 ~ 0 + DivisiaQ + ddp1 + ddp2 + ddp3
e3 <- wdq3 ~ 0 + DivisiaQ + ddp1 + ddp2 + ddp3
system <- list(e1, e2, e3)
# Unrestricted estimation
fit1 <- systemfit(system, method="SUR")
print(summary(fit1))
R <- matrix(0,3,12)
q <- matrix(0,3,1)
R[1,3] <- 1
R[1,6] <- -1
R[2,4] = 1
R[2,10] = -1
R[3,8] = 1
R[3,11] = -1
# Restricted estimation, iterated
fit2 <- systemfit(system, method="SUR", restrict.matrix=R,
restrict.rhs=q, maxiter=80)
print(summary(fit2))
lrtest(fit2, fit1)
end foreign
</hansl>
The one thing that seems like a bug in gretl is that you're able to
estimate the "Rotterdam" system using 3SLS without generating an
error, since you have not specified any instruments. I'll look into
that. (Gretl seems to be falling back to consider all the regressors
as exogenous, but that seems too lax.)
Allin Cottrell
