st: Problems with -xtabond2- (Arellano-Bond estimator)

[Sergey <fractal2004@list.ru>]

Dear Statalisters,

I have a problem when estimating a dynamic panel data model using -xtabond2-

Number of periods (years) - 5, number of observations - 82, number of variables - 15 (except year effects)

The command is the following:

xtabond2 y L.y x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12 x13 x14 x15 year3 year4 year5, gmmstyle(L.y x1, laglimits (1 3)) ivstyle(x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12 x13 x14 x15 year3 year4 year5, equation(level)) robust small

As a result I have AR(2) in first differences (p=0.029), 
Sargan test of overid. restrictions: chi2(19) = 63.17 Prob > chi2 = 0.000
Hansen test of overid. restrictions: chi2(19) = 36.36 Prob > chi2 = 0.010

The panel is low and I couldn't use more lags. 
Are there any ways to improve the specification of model? 
Are these tests valid in this situation? 
Should I increase the number of instruments? 

As an alternative I estimate the model with the lagged dependent variable only and without fixed effects by -xtpcse-:

xtpcse y L.y x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12 x13 x14 x15 year3 year4 year5, correlation(ar1)

But in this case I am confused by very small standard errors. Is there any test to choose between a model with lagged dependent variable and a dynamic panel data model?

Thank you very much.

---------
Sergey

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/