How to do Xtabond2: An Introduction to Difference and System GMM in Stata

Abstract

, are increasingly popular. Both are general estimators designed for situations with "small T , large N " panels, meaning few time periods and many individuals; independent variables that are not strictly exogenous, meaning they are correlated with past and possibly current realizations of the error; fixed effects; and heteroskedasticity and autocorrelation within individuals. This pedagogic article first introduces linear generalized method of moments. Then it describes how limited time span and potential for fixed effects and endogenous regressors drive the design of the estimators of interest, offering Stata-based examples along the way. Next it describes how to apply these estimators with xtabond2. It also explains how to perform the Arellano-Bond test for autocorrelation in a panel after other Stata commands, using abar. The article concludes with some tips for proper use.

Logistics and economic growth: a panel data approach

“…As it does not consider unobserved time and regional effects, the OLS estimation of panel data tends to bias upwards and inconsistent, given the positive correlation between the lagged dependent variable and the error term (Ding et al 2008). On the other hand, although the Fixed-effects estimation is better than OLS due to its controlling for the omitted variable bias and the endogeneity problem, it does not eliminate dynamic panel bias since there is a negatively correlation between the lagged dependent variable and the error term, biasing its coefficient estimate downward (Roodman 2007). Interestingly, generally the OLS and Fixed-effects estimators can provide a bound for the true value of the coefficient of the lagged dependent variable (Sevestre and Trognon 1996), and therefore, good estimations should lie in the range between these values, or at least close to it (Ding et al 2008).…”

The factors determining firm start-ups in French regions and the heterogeneity of regional labor markets

“…Estimates are obtained from the xtabond2 procedure, seeRoodman (2009) and from the xtdpd procedure in STATA 13.…”

Crime and punishment: evidence from dynamic panel data model for North Carolina (2003–2012)

“…3 This potential danger has always been neglected in studies that have used DPD models (see, e.g., Han et al 2014). We follow Roodman (2009b) and use only two lags of lagged dependent variable (LDV) and endogenous variables in the matrix of instruments. 4 Furthermore, through a DPD model we can also deal with measurement error problems.…”