st: No valid instruments for system GMM

[natalie.stata@web.de]

Dear statalist,

I would like to estimate an agricultural production function using panel data of approximately 20,000 individuals over 20 years. Applying system GMM - due to unobserved heterogeneity and simultaneity (inputs and output are determined together e.g. due to profit-maximizing behaviour of the firm), I estimate the following:

     xtabond2 y l.y x1 x2 x3 x4 x5, robust gmm(y x4 x5) iv(x1 x2 x3) small noconst

where y is output (crop yields), x1-x3 (e.g. weather) are strictly exogenous input variables and x4-x5 are endogenous variables (e.g. fertilisers). Furthermore, I added a lagged dependent variable on the right-hand side as it is highly significant. This specification passes the autocorrelation tests, but not the Sargan test of valid instruments. If I, however, specify using lag (18 19) (the maximum nr. of lags possible for my dataset) it passes the Sargan test at the 5% level, but do not obtain any significant estimates, except for the lagged dependent variable:

     xtabond2 y l.y x1 x2 x3 x4 x5, lag(18 19) robust gmm(y x4 x5) iv(x1 x2 x3) small noconst

Is there any systematic way of finding valid instruments for the system GMM? Or is there another estimator that I can use to handle simultaneity where I don't need instruments from outside the dataset (I know there is one called "opreg" from Olley and Pakes, but I need to specifiy exit of a firm which is not included in my data).

I would greatly appreciate any comments on this problem!

Best regards,
Natalie
___________________________________________________________
Schon gehört? WEB.DE hat einen genialen Phishing-Filter in die
Toolbar eingebaut! http://produkte.web.de/go/toolbar

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