xtabond2

xtabond2 is a Stata command that implements Arellano–Bond style generalized method of moments (GMM) estimators for dynamic panel data. Developed by David Roodman as an enhanced successor to xtabond, it provides system-GMM and difference-GMM estimators with features aimed at improving finite-sample performance and reducing instrument proliferation. It is widely used in econometrics for models that include a lagged dependent variable and endogenous regressors on panel data.

Key features and options include: support for system GMM and difference GMM estimators, one-step and two-step estimation with robust standard errors, and the Windmeijer finite-sample correction for the two-step estimator. The command uses gmmstyle and ivstyle specifications to define endogenous, predefined exogenous, and instrumental variables, and offers instrument reduction through options such as collapse and lag structures. It reports standard diagnostic tests, including the Hansen overidentification test and Arellano–Bond tests for first- and second-order autocorrelation in differenced residuals. The package also handles unbalanced panels and allows users to tailor the lag depth of instruments to mitigate instrument proliferation.

Usage typically involves specifying the dependent variable, its lagged terms, and the regressors, then choosing gmmstyle (for endogenous instruments) and ivstyle (for external instruments) with conventions for lags. Example syntax commonly cited is xtabond2 y L.y x1 x2, gmmstyle(L.y, lag(2 .)) ivstyle(x1 x2) twostep robust small. For guidance, users consult the help file and related literature, noting limitations related to instrument proliferation in small samples and the importance of appropriate lag selection.