Arellano-Bond GMM Estimator
Arellano-Bond Generalized Method of Moments Estimator · Also known as: AB-GMM, Difference GMM, first-difference GMM, Arellano-Bond estimator
The Arellano-Bond GMM estimator is the standard approach for dynamic panel data models in which the lagged dependent variable appears as a regressor. By first-differencing to remove fixed effects and using deeper lags as instruments, it yields consistent estimates even when the error is serially correlated and regressors are endogenous.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
+13 more
When to use it
Use the Arellano-Bond estimator when (1) you have panel data with a moderately large number of individuals (N) and a short time dimension (T), (2) the lagged dependent variable belongs in the model, and (3) some regressors are likely endogenous. It is the appropriate choice when T is small (T < 10) because System GMM can suffer from weak instruments in that case. Do not use it when T is large relative to N, as the number of instruments explodes and over-identification tests lose power — fixed-effects IV or System GMM may be preferable. Also avoid it when there is no genuine dynamic relationship or when instruments are weak (test with first-stage F-statistics).
Strengths & limitations
- Eliminates individual fixed effects without assuming they are uncorrelated with the regressors.
- Handles endogenous regressors by exploiting internal instruments (lagged levels), requiring no external data.
- Asymptotically efficient under the chosen moment conditions; two-step GMM achieves the semiparametric efficiency bound.
- Formally tests overidentifying restrictions (Sargan-Hansen J) and serial correlation (AR(1), AR(2)) to check model validity.
- Widely implemented in Stata (xtabond2), R (plm, pgmm), and Python (linearmodels), making replication straightforward.
- Instrument proliferation: when T grows, the instrument count can exceed the number of individuals, making the Sargan test unreliable and coefficient estimates imprecise. Limit lags with the collapse option.
- Weak instruments in short panels: lagged levels may be poor predictors of first differences when the autoregressive coefficient is near unity, biasing estimates toward OLS. System GMM with level equations can help.
- Two-step standard errors are severely downward-biased in small samples without Windmeijer's finite-sample correction.
- Assumes no second-order serial correlation in levels; if the original error is AR(1), the AR(2) test will reject and the estimator is invalid.
Frequently asked
What is the difference between Arellano-Bond and System GMM?
Arellano-Bond (Difference GMM) uses only first-differenced equations instrumented by lagged levels. System GMM (Blundell-Bond 1998) adds the levels equations instrumented by lagged differences, which improves efficiency when the autoregressive parameter is near one but requires an additional stationarity assumption on the initial conditions.
How do I know if I have too many instruments?
A practical rule is that the instrument count should be less than the number of individuals (N). You can also check: if the Sargan p-value is suspiciously high (e.g. > 0.99), the test has been weakened by instrument proliferation. Use the collapse option in xtabond2 or restrict lag depth.
What does a significant AR(2) test mean?
A significant AR(2) statistic in the first-differenced residuals indicates second-order serial correlation, which implies the lagged levels used as instruments are correlated with the error. This invalidates the estimator. You should use instruments lagged three or more periods and re-test.
Can I use Arellano-Bond with unbalanced panels?
Yes. The estimator accommodates unbalanced panels: each individual contributes the available time periods to the moment conditions, so gaps in the panel reduce the instrument count for that individual but do not break the approach.
Should I use one-step or two-step GMM?
Two-step GMM is asymptotically more efficient, but its standard errors are severely downward-biased in finite samples. Always apply Windmeijer's (2005) finite-sample correction (the default in xtabond2). With that correction, two-step GMM with robust standard errors is generally preferred over one-step.
Sources
- Arellano, M., & Bond, S. (1991). Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Review of Economic Studies, 58(2), 277-297. DOI: 10.2307/2297968 ↗
- Roodman, D. (2009). How to do xtabond2: An introduction to difference and system GMM in Stata. Stata Journal, 9(1), 86-136. DOI: 10.1177/1536867X0900900106 ↗
How to cite this page
ScholarGate. (2026, June 3). Arellano-Bond Generalized Method of Moments Estimator. ScholarGate. https://scholargate.app/en/econometrics/arellano-bond-gmm-estimator
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Difference GMMEconometrics↔ compare
- Dynamic Panel Data ModelEconometrics↔ compare
- Fixed Effects ModelEconometrics↔ compare
- Instrumental Variables in Health ResearchHealth Economics↔ compare
- Panel System GMMEconometrics↔ compare