Need help in xtdpd command for two-step GMM

[Chua Mei Shan]

Hi everyone,

I need help from you all to verify the command below:

. xtdpd bvstd1 l.bvstd1 size tang prof ndts mtb y08-y17   oilgas basicmaterials industrial consumergoods healthcare consumerservices
>  telecommunication utilities financials technology, twostep dgmmiv(bvstd1, lagrange (2 .)) lgmmiv(bvstd1, lag(2)) div(L2.size L2.t
> ang L2.prof L2.ndts L2.mtb y08-y17, nodifference) liv(LD.size LD.tang LD.prof LD.ndts LD.mtb y08-y17) artests (2)

FYI, I am doing the two-step GMM for dynamic capital structure model.
The independent and dependent variables are based on owned lagged (t-2 and deeper) and first differencing (t-1 and deeper).

Pls help.

Thanks& Regards,
Chua

[Sebastian Kripfganz]

A few comments:

• With the twostep estimator, you should always use the vce(robust) option to compute the Windmeijer-corrected standard errors.
• Time dummies should only be specified as instruments for the level model, not both the level and first-differenced model; see also https://www.statalist.org/forums/for...m-time-dummies.
• Are your variables oilgas basicmaterials industrial consumergoods healthcare consumerservices telecommunication utilities financials technology constant over time? If so, then there coefficients are (asymptotically) not identified under the usual assumption that all first differences used as instruments for the level model are uncorrelated with all time-invariant variables. The reported coefficient estimates in that case are just spurious because in finite samples this correlation is usually not exactly equal to zero. In any case, the instruments would be either very weak or otherwise invalid. You need to find level instruments but then place strong assumptions about their uncorrelatedness with the unobserved time-invariant effects; see Kripfganz and Schwarz (2015). Estimation of Linear Dynamic Panel Data Models with Time-Invariant Regressors. ECB Working Paper 1838.

[Chua Mei Shan]

Hi Mr Sebastian Kripfganz,

Thank you for your comments and working paper. The variables for "oilgas basicmaterials industrial consumergoods healthcare consumerservices telecommunication utilities financials technology" are constant overtime since they are the industry dummies.

Regarding the vce(robust) option, it is a must to include into two-step estimator. Since according to Windmeijer, 2005, the two-step estimator only biased in small sample. Consider the sample size of 400 firms that i have, would it be large enough?

However, to show u my effort, i've made changes on my command:

xtdpd bvstd1 l.bvstd1 size tang prof ndts mtb y08-y17   oilgas basicmaterials industrial consumergoods healthcare consumerservices
>  telecommunication utilities financials technology, twostep dgmmiv(bvstd1, lagrange (2 .)) lgmmiv(bvstd1, lag(2)) div(L2.size L2.t
> ang L2.prof L2.ndts L2.mtb, nodifference) liv(LD.size LD.tang LD.prof LD.ndts LD.mtb y08-y17 oilgas-technology) artests (2) vce(ro
> bust)

Am i doing it in correct order?

Regards,
Chua

[Sebastian Kripfganz]

Personally, I would always use the Windmeijer correction. It does not do any harm even in larger samples but helps a lot in small samples. 400 firms might be large from a macroeconomic or macrofinancial perspective but it is still considerably smaller than many microeconomic sample sizes.

From just looking at your code, the specification now seems reasonable.

[Chua Mei Shan]

Noted. Thanks a lot for your help.

[Chua Mei Shan]

Hi Mr Sebastian Kripfganz,

How about if i am doing for xtabond command? Am i correct if

xtdpd bvtd1 L.bvtd1 size tang prof ndts mtb y08-y17  oilgas basicmaterials industrial consumergoods healthcare consumerservices telecommunication utilities financials technology malaysia singapore indonesia thailand, dgmmiv(bvtd1, lagrange (2 .)) div(size tang prof ndts mtb y08-y17 oilgas basicmaterials industrial consumergoods healthcare consumerservices telecommunication utilities financials technology malaysia singapore indonesia thailand)  div(L2.size L2.tang L2.prof L2.ndts L2.mtb y08-y17 oilgas basicmaterials industrial consumergoods healthcare consumerservices telecommunication utilities financials technology malaysia singapore indonesia thailand, nodifference) twostep vce(robust)

Pls advice.

Thanks & Regards,
Chua