# Simulate Arellano-Bond

I have fitted a dynamic panel data model with Arellano-Bond estimator in gretl, here is the output:

Model 5: 2-step dynamic panel, using 2332 observations
Included 106 cross-sectional units
H-matrix as per Ox/DPD
Dependent variable: trvr

coefficient   std. error     z       p-value
---------------------------------------------------------
Dtrvr(-1)    0.895381     0.0248490    36.03    2.55e-284 ***
const        0.0230952    0.00226823   10.18    2.39e-024 ***
x1          -0.0263556    0.00836633   -3.150   0.0016    ***
x2           0.127888     0.0171532     7.456   8.94e-014 ***

Sum squared resid    605.9396   S.E. of regression   0.510180

Number of instruments = 256
Test for AR(1) errors: z = -4.29161 [0.0000]
Test for AR(2) errors: z = 1.62503 [0.1042]
Sargan over-identification test: Chi-square(252) = 105.139 [1.0000]
Wald (joint) test: Chi-square(3) = 2333.35 [0.0000]


I have 2 questions about the results:

1. How do I assess the fit?
2. How can I simulate from the model?

Regards

-
@teucer, what do you mean by simulate? You want to get the forecasts? –  mpiktas Feb 3 '11 at 10:28
@mpiktas I want to plug in new values for x1, x2 and get the probability distribution for trvr (which should be the sum of two normals, right?). So it is a simulation and not a forecast... –  teucer Feb 3 '11 at 12:53
@teucer, in general Arrelano-Bond estimator does not assume normality. Furthermore it is used to estimate conditional expectation $E(y|x)$, so it is not a good idea to use it for simulating the probability distribution of $y$. –  mpiktas Feb 3 '11 at 13:10
@teucer, why do you insist on simulating? If you want to get the feel how model works for different future values of x1 and x2, simply plug them into model and calculate trvr. What is the purpose of your simulation? –  mpiktas Feb 3 '11 at 14:09
@teucer, look at the coefficients, lag coefficient is 0.9, this is very large. This means that almost everything is explained by the lagged value. If the values of x1 and x2 are small they contribute little, since the corresponding coefficients for the them and the constant are small. –  mpiktas Feb 3 '11 at 16:59
show 5 more comments