I want to apply the two-step system GMM to investigate the impact of ownership concentration on the CEO pay-performance relationship with 201 firms for 5 years of balanced panel data. I have applied the command given below.
xtabond2 TC L.TC ROEP ROET3 T3 LFSIZE LFAGE LEV RISK CEOD BSZE IND_P ID* YD*, gmmstyle(L.TC L.(ROEP ROET3 T3 LFSIZE LEV RISK CEOD BSZE IND_P), lag(0 1)) ivstyle(LFAGE ID* YD*) twostep robust small
DV is TC; IDVs and control variables are ROEP T3 LFSIZE LFAGE LEV RISK CEOD BSZE IND_P; ROET3 is the interaction variable; ID* are 5 industry dummy variables and YD* are 4 year dummy variables
(1) I want to confirm whether the command is correct or not?
(2) Please suggest some solution to meet all the assumptions along with retaining the significance of the coefficients.
My results were not up to the mark- the p-value of the Hansen and Sargan test was less than 0.1 and sometimes very high; AR(1) and AR (2) both are insignificant. I made some changes to the command like adding collapse to the equation to reduce the number of instruments, changed the classification of some variables from endogenous to exogenous but none worked.
A copy of few results is also attached for reference.
Hi, it seems to me that you have lots of regressors and you are assuming that all these regressors are either endogenous or predetermined. Indeed this is too much for the command to deal with, it would easily run into a weak intrument problem which causes the lack of signficance.
I'd recommend, first, to to start with a small model and then see how things change by adding variables, especially ID and YD. With such short time span it is likely that time dummies capture a lot of the variability.
Second, to see whether some variables can be considered exogenous or predetermined (using theory or by lagging regressors).
You should also check whether there is some multicollinearity with these dummies.
I hope this helps,
best,
P:E.
I agree with Piero.
Furthermore, Chhavi, did you rely on or refer to Roodman's papers on the use of GMM? They are still very important reading.
Best
S.
--
Webpage: http://docenti.unisa.it/001511/en/home; Repec Papers: http://econpapers.repec.org/RAS/pde148.htm; Academia.edu: https://salerno.academia.edu/DestefanisSergio ResearchGate: https://www.researchgate.net/profile/Sergio_Destefanis
@Sergio Destefanis Yes, I am referring to Roodman's paper but facing problem in understanding the technical part
@Piero Esposito Thank you for the suggestions. I will consider them and work on my model again.
Piero Esposito Is it compulsory to apply two-step system gmm with robust option. As given in the Roodman(2009) paper by construction two-step system gmm is more efficient then one-step system gmm. However, Weigh matrix is based on initial estimators, which makes standard errors downward bias. So robust option in two- step GMM implies Weindmeijers (2005) corrections. However, which this option all my coefficients becomes insignificant.
I have 67 countries and 16 years 12 independent variables. How should I proceed?
@Panika yes, using robust option in GMM is similar to calculating robust standard errors in case of Fixed/random effect regression. You can try changing classification of variables as endogeneous, predetermined or exogeneous
Panika Jain I agree with Chhavi in the seelction of endogenous variables. Here some suggestions: first, given your time span i'd check for stationarity of the series and use a fully stationary specification due to the fact that cointegration with 13 variables is very hard to test/estimate. Second, you can try to lag all variables so that you instrument only the lagged dependent variable. Third, you might try the difference GMM (or a standard IV ala Andersen and Hsiao) rather than system GMM. Last, notice that the Nickel bias falls with T, so with 16 time periods you might argue that the bias is not so important and use a more standard approach.
- Badges
- Science topic
- Similar topics
- Indexes