Dynamic Panel Data Model
Also known as: dynamic panel model, lagged dependent variable panel model, Arellano-Bond type dynamic panel, GMM dynamic panel
The dynamic panel data model extends standard panel regression by including one or more lagged values of the outcome variable as regressors. Because past outcomes directly predict current outcomes, the model captures persistence and adjustment dynamics — but it also introduces a correlation between the lagged dependent variable and the individual fixed effect, rendering OLS and standard fixed-effects estimators inconsistent. GMM-based approaches developed by Arellano-Bond and Blundell-Bond resolve this problem.
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.
When to use it
Use a dynamic panel data model when theory or prior evidence suggests that past outcomes affect current outcomes (habit persistence, partial adjustment, path dependence) and your data form a panel with N units observed over T periods, typically with large N and small T. It is especially appropriate when individual heterogeneity is suspected (warranting fixed effects) but standard within-group estimators would be inconsistent because of the lagged dependent variable. Avoid it when T is long relative to N (use time-series methods instead), when the number of instruments is close to or exceeds N (instrument proliferation), or when the panel is severely unbalanced with many missing periods.
Strengths & limitations
- Handles the Nickell bias that afflicts fixed-effects OLS when a lagged dependent variable is included.
- Controls for unobserved individual heterogeneity without requiring instruments for the fixed effect itself.
- Exploits internal instruments (past lags), so external instruments are not required in most applications.
- System GMM substantially improves efficiency over difference GMM when the outcome is highly persistent (alpha close to 1).
- Accommodates endogenous regressors beyond the lagged outcome by treating them similarly with additional moment conditions.
- Requires large N relative to T; with small N the GMM estimator can behave poorly.
- Instrument proliferation — using too many lag depths — inflates the instrument count, weakens the Hansen test, and can produce spuriously good fit.
- The AR(2) and Hansen tests lose power in small samples; a non-rejection does not guarantee instrument validity.
- Model specification (lag length, which regressors to treat as endogenous) requires careful theoretical justification.
Frequently asked
Why can't I just use fixed-effects OLS with a lagged dependent variable?
Fixed-effects (within) OLS is inconsistent when T is small because the demeaned lagged outcome is mechanically correlated with the demeaned error — the 'Nickell bias'. This bias is O(1/T), so it is negligible when T is large but severe in typical short panels.
When should I prefer system GMM over difference GMM?
System GMM is preferable when the autoregressive coefficient alpha is close to 1 or the regressors are highly persistent, because lagged levels become weak instruments in the first-differenced equation. System GMM adds a levels equation with lagged differences as instruments, restoring relevance. When persistence is low, difference GMM is simpler and equally efficient.
How many instruments is too many?
A common rule of thumb is to keep the instrument count below the number of cross-sectional units N. Collapsing the instrument matrix (using one instrument per variable per lag depth instead of one per time period) or restricting the lag window are standard remedies for instrument proliferation.
What does it mean if the AR(2) test rejects?
The Arellano-Bond AR(2) test checks for second-order autocorrelation in the first-differenced residuals. Rejection implies that epsilon_it in levels is serially correlated, which invalidates y_{i,t-2} as an instrument. You must use deeper lags (starting from t-3) or reconsider the error structure.
Can the dynamic panel model handle time-fixed effects?
Yes. Time dummies should generally be included to control for common macroeconomic shocks affecting all units in a given period. Without them, the GMM instruments may absorb period-specific variation that should be attributed to aggregate trends.
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 ↗
- Blundell, R., & Bond, S. (1998). Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics, 87(1), 115–143. DOI: 10.1016/S0304-4076(98)00009-8 ↗
How to cite this page
ScholarGate. (2026, June 3). Dynamic Panel Data Model. ScholarGate. https://scholargate.app/en/econometrics/panel-dynamic-panel-data-model
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.
- Arellano-Bond GMM estimatorEconometrics↔ compare
- Dynamic Panel Data ModelEconometrics↔ compare
- Panel Fixed Effects ModelEconometrics↔ compare
- Panel Random Effects ModelEconometrics↔ compare
- Panel System GMMEconometrics↔ compare