Understanding xtabond2 syntax

[gagandeep sharma]

Hello everyone,

I'm trying to understand the syntax of xtabond2 for implementing system gmm on my dynamic panel dataset. The variable of interest (Div2) as well all the control variables are lagged (t-1) in the main equation. I have run the following two versions of xtabond 2 and have produced very different results, especially Hansen test statistic.

1. xtabond2 lendgrwth_w l.lendgrwth_w l.Div2_w l.totalcapratio_w l.liquidity_w l.size_w l.fundstruct_w l.profitability_w l.gdpgrowth_w l.npl_w l.allow_w l.repochange, gmm( l.lendgrwth_w l.Div2_w l.totalcapratio_w l.liquidity_w l.size_w l.fundstruct_w l.profitability_w l.gdpgrowth_w l.npl_w l.allow_w l.repochange, lag (5 6) collapse eq(level)) iv (i.year, eq(level)) twostep robust ar(3)

2. xtabond2 lendgrwth_w l.lendgrwth_w l.Div2_w l.totalcapratio_w l.liquidity_w l.size_w l.fundstruct_w l.profitability_w l.gdpgrowth_w l.npl_w l.allow_w l.repochange, gmm( lendgrwth_w Div2_w totalcapratio_w liquidity_w size_w fundstruct_w profitability_w gdpgrowth_w npl_w allow_w repochange, lag (5 6) collapse eq(level)) iv (i.year, eq(level)) twostep robust ar(3)

The obvious difference is that in the first specification I use gmm(l.dependent var l.independent var l.control var) whereas in the second specification I use gmm (dependent var independent var control var). Which of the specification is logically correct, and why?

It would also be great help if someone could explain what does eq(level) and eq (diff) imply for the model specified, and which option should I use?

[Sebastian Kripfganz]

Why are you using lags 5 and 6 as instruments? And even more questionable: Why are you using only instruments for the level equation?

More on GMM estimation of dynamic panel data models:
XTDPDGMM: new Stata command for GMM estimation of linear dynamic panel models

[Pedro Antonio]

I am using the command

xtabond S R G, pre(t83 t84 t85 t86 t87) lags(3) endogenous(R, lag(3,.)) endogenous(G, lag(3,.))

but I want to exclude estimations of R and G in levels. How can I get it?

Thanks

[Sebastian Kripfganz]

The command has a nolevel option.

[Pedro Antonio]

I can't see that option when typing help xtabond:

options Description
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Model
noconstant suppress constant term
diffvars(varlist) already-differenced exogenous variables
inst(varlist) additional instrument variables
lags(#) use # lags of dependent variable as covariates; default is lags(1)
maxldep(#) maximum lags of dependent variable for use as instruments
maxlags(#) maximum lags of predetermined and endogenous variables for use as instruments
twostep compute the two-step estimator instead of the one-step estimator

Predetermined
pre(varlist[...]) predetermined variables; can be specified more than once

Endogenous
endogenous(varlist[...]) endogenous variables; can be specified more than once

SE/Robust
vce(vcetype) vcetype may be gmm or robust

Reporting
level(#) set confidence level; default is level(95)
artests(#) use # as maximum order for AR tests; default is artests(2)
display_options control spacing and line width

coeflegend display legend instead of statistics


It is a problem of my version (Stata 14)?

Pedro

[Sebastian Kripfganz]

Sorry, my mistake.

The command xtabond estimates the model in first differences. I thus do not really understand your initial question about levels.

[Pedro Antonio]

I'm affraid I do not understand properly the output. I estimate

xtabond S R G, lags(3) endogenous(R, lag(2,.)) endogenous(G, lag(2,.))
note: R dropped because of collinearity
note: G dropped because of collinearity

Arellano-Bond dynamic panel-data estimation Number of obs = 1,325
Group variable: id Number of groups = 265
Time variable: year
Obs per group:
min = 5
avg = 5
max = 5

Number of instruments = 58 Wald chi2(9) = 3372.26
Prob > chi2 = 0.0000
One-step results
------------------------------------------------------------------------------
S | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
S |
L1. | -.5151975 .0711475 -7.24 0.000 -.6546441 -.3757509
L2. | .0664364 .0601437 1.10 0.269 -.0514431 .1843158
L3. | -.0870036 .0165012 -5.27 0.000 -.1193454 -.0546618
|
R |
--. | .8064682 .0193631 41.65 0.000 .7685173 .8444191
L1. | .4851564 .0673851 7.20 0.000 .353084 .6172289
L2. | -.1964957 .0665926 -2.95 0.003 -.3270147 -.0659767
|
G |
--. | .7129053 .1072076 6.65 0.000 .5027823 .9230283
L1. | .6473446 .2349174 2.76 0.006 .186915 1.107774
L2. | -.885298 .2014909 -4.39 0.000 -1.280213 -.4903831
|
_cons | .0112254 .0023659 4.74 0.000 .0065884 .0158625
------------------------------------------------------------------------------

R and G have been dropped becuase of collinearity, but what do mean the coefficients in red?

[Sebastian Kripfganz]

I am not entirely sure what the meaning is of the notes about R and G being dropped. Clearly, they are not dropped from the regression itself because coefficients for them are estimated.