Dear all, I have a dynamic panel model and I'm testing for crime persistence, T=13 and n=134. My variables are: dep var: log(homicide); explanatory vars: lag of log(homicide), log(Gini), log(GDPpc), unemployment rate, male 15-24yrs, pry educ, sec educ, rule of law, corruption and death penalty. Using xtabond2 in Stata13, lag of log(homicide) is endogenous and while other explanatory are weakly exogenous. I did the regression and my results are: N=1463, n=134, instruments=66, lags = 3, AR(1) = 0.009, AR(2)=0.068, Hansen=(0.101). I feel that I can accept this result by simply saying that 'I cannot reject AR(2) at 5% level of significance' will I be correct? or is there a way of correcting for AR(2)? I've read both Roodman's papers on this topic and I will greatly appreciate expert contributions. Thank you.
The test tells you that using the second lag as an instrument won't be valid as it is correlated with the disturbance (I guess that you reported the p-value for the AR-2 process). So start instrumenting with the third (and higher order) lags, and thereafter test if you can reject the AR-3 in the disturbance...
thanks Norbert, for the response. I used 3 lags and the problem is that Stata only reports p-values for AR(1) and AR(2). Given that AR(2) is 0.068, I just want to know if I can accept this result since the null of 'no serial autocorrelation' is rejected at 5% significance level. Will I be correct by doing so?
Statistically speaking you are correct: you cannor reject the null-hypothesis at 5%. Nevertheless the model seems to suffer from misspecification. You should consider adding another lag of the dependent variable, it will probably take care of the 2nd order autocorrelation.. The Hansen test p-value is low, so your instruments are not valid. Also, the number of instruments is very high, do you limit the maximum number of lags used for the instrumentation?
thank you so much Peter, but how do I add another lag of the dependent variable? do you mean I should use up to 4 lags, because I have used 3 lags already. I also thought my Hansen statistic is in good shape since we've been told that one should be wary of a high p-value....that's what Roodman said. Also, is 66 instruments too much? I used the 'collapse' option to minimise the number of instruments and all the p-values just went 'awry'. I just need a way to unravel this GMM....the methodology looks so simple but gruesomely technical.....this is an emergency....please help me.
Please I run xtabond2 and the AR(2) is dot. Any problem with it? how do I correct it?
xtabond2 has AR(3) option for reporting the diagnostics, you only add AR(3) on the syntax. And also you should try to lags further than t-2 due to it has autocorrelation in the second order of differenced residuals. Apart from AR tests, you should try to use Sargan-Hansen J Test for overidentification problem. I hope it will help.
Dear Ngozi, how did you resolve finally this issue related to the improvement of Ar(2). I'm facing the same issue. Thanks for your help.
- Badges
- Science topic
- Similar topics
- Instrumentation Engineering
- Instruments