Hi,
I would very much appreciate your help and comments on my questions on xtabond2
I am trying to estimate a dynamic panel data model using an unbalanced data set
with more than 50,000 observations, over 6,000 firms in 37 countries over the
1991 ? 2002 period. Each firm has a minimum of four consecutive years of data.
I am estimating the model using the One Step Robust GMM System estimator using
the command xtabond2 in STATA 8.0.
I use year, industry and country dummies to control for time, industry and
country specific effects.
I am using STATA Command as follows:
xi:xtabond2 Investmentratio l.Investmentratio Isquared Salesratio CFratio i.year
i.Industry i.country, robust gmm(Investmentratio Salesratio CFratio, lag(2 3))
where
Investmentratio= ratio of Investment to capital= dependent variable
l.Investmentratio = lagged dependent variable = (lagged Investment ratio)
Isquared = square of lagged dependent variable
CFratio = ratio of Cash Flow to capital
Salesratio= ratio of sales to capital
I am treating the lagged dependent variable, the cash flow ratio and the sales
ratio as endogenous (by including all of them in the gmm style option).
The problem I am facing is that the validity of the instruments is rejected by
the Hansen J test of over- identifying restrictions. The number of instruments
is 88.
Chi2(31)=196.04 Prob>chi2=0.000
None of the coefficients is statistically significant.
I tried running the above equation for each country at a time. Surprisingly, I
got P-values = 0 for the Hansen J test of over- identifying restrictions for
five developed countries: UK, Italy, Germany, France and Australia. I tried
running the equation for the whole data set, while excluding these five
countries but still the validity of instruments got rejected with P-values=0.
I would very much appreciate your help on the following questions:
Question (1): Do I have to instrument for each endogenous variable (Investment
ratio, Cash Flow ratio and Sales ratio) by including each one of them in the gmm
list of instruments as I already have?, or can I use only one of them as
instrument in the gmm style option ?.
For instance, can I only use the cash Flow ratio as instrument with lags (2 3)
as follows:
Command: (here I am only accounting for time specific effects)
xi:xtabond2 Investmentratio l.Investmentratio Isquared Salesratio CFratio i.year
, robust gmm(CFratio, lag(2 3))
I tried the above specification, and the validity of the instruments get
accepted (P value is 0.8).
Hansen test of overid. restrictions: chi2(14) = 8.82 Prob > chi2 = 0.842
Question (2): Is a P-value of 0.8 usual to get in a test for over- identifying
restrictions ?.
Question (3): Can I use the instruments dated (t-2) only or the instruments
dated (t-3) only ?.
If yes, what would be the specified lags in that case ?.
Question (4): How to control for firm specific effects ?. Are they automatically
accounted for once I use firm/ country as panel identifier in STATA ?.
I would very much appreciate your help and time,
Thank you in advance,
Best regards,
Dahlia Anwar El- Hawary
Consultant
Financial Sector Operations and Policy Department
World Bank
Tel: 202 473 5238
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
