I'm using panel data with N around 51 entiities.I have data from 1985 onwards and hence taken the average of all variables data which makes T=6. I'm using xtabond2 command and running twostep system GMM. I'm not able to reject null for both AR(1) and Ar(2). From papers on arellano bond, it is that we expect to reject AR(1) but not AR(2) which is not true in my case. Can anybody explain what can be the reason behind that . Please cite any papers that can shed more light on this issue.
Thanks
An analysis of the residuals may help. There might still exist low level components, which ought to be eliminated before fitting the model. These compnents may not show on high order AR but present in lower orders. ... my suggestion is to establish whether any of the components are present, then run a diagnostic check on the model ...
The AR1 and AR2 statistics measure first- and second-degree serial
correlations. Given the use of first-difference transformations, we
expect some degree of first-order serial correlation, although this
correlation does not invalidate our results. However, the presence
of second-order serial correlation does signal omitted variables (see, also Andres and Vallelado, 2008)
- Badges
- Science topic
- Similar topics
- Instrumentation Engineering
- Instruments