xtabond2 System GMM: repeated time values within panel and instrument specification

[Soma alanazi]

Dear Statalist members, I am working with an unbalanced firm-year panel dataset and I am trying to estimate a dynamic panel model using two-step System GMM with xtabond2. My panel identifier is gvkey and my time variable is fyear. When I run: xtset gvkey fyear I obtain: repeated time values within panel r(451) I understand that this indicates that there are multiple observations for some firm-year combinations. I checked the data using: duplicates report gvkey fyear and: duplicates tag gvkey fyear, gen(dup) tab dup There are duplicate firm-year observations. I do not want to use: duplicates drop gvkey fyear, force without first understanding why the observations are duplicated, because the observations may contain different values for other variables. My main question is therefore how I should correctly diagnose and resolve the repeated firm-year observations before estimating System GMM. After resolving this issue, I also need advice on the appropriate specification of xtabond2. I have a dynamic panel model with a lagged dependent variable and several explanatory and control variables. For simplicity, I will refer to them as: y = dependent variable x1 = main explanatory variable x2 = second explanatory variable controls = control variables I am considering a two-step System GMM specification with robust standard errors. I would like advice on: 1. How should I determine which regressors should be treated as endogenous, predetermined, or strictly exogenous? 2. How should I choose the lag ranges for the GMM-style instruments? 3. When should variables be specified using gmm() versus iv()? 4. Should I use collapse to limit instrument proliferation? 5. How should I determine whether the number of instruments is appropriate? 6. How should I interpret the Hansen test and AR(1) and AR(2) tests? 7. Is it appropriate to use two-step System GMM with robust standard errors for an unbalanced firm-level panel? I previously obtained a System GMM estimation with 53 instruments, but I am unsure whether the instrument specification is theoretically and econometrically justified. I would appreciate guidance on how to determine the appropriate specification rather than simply choosing a specification because the model successfully runs. Thank you.