Dynamic Panel Data Model
Also known as: dynamic panel model, panel data model with lagged dependent variable, DPD model, Arellano-Bond model
The dynamic panel data model extends standard panel regression by including a lagged value of the outcome variable as a regressor, capturing persistence and adjustment dynamics. Because the lagged dependent variable is correlated with the unit-specific fixed effect, ordinary OLS or within estimators are biased; GMM-based methods using internal instruments are the standard remedy.
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.
+18 more
When to use it
Use the dynamic panel data model when the outcome is persistent over time and theory suggests that past values directly influence current outcomes — typical in macroeconomics (growth, investment, trade), finance (firm leverage, dividends), and health economics (health stock). The panel must have multiple time periods (T >= 3 for Arellano-Bond; T >= 4 preferred) and a reasonable number of cross-sectional units (N >> T is the classic setting). Do not use it when T is very large relative to N (time-series asymptotics dominate instead), when the panel is severely unbalanced with many gaps, or when the lagged dependent variable is theoretically unmotivated — in those cases standard fixed-effects or VECM frameworks may be more appropriate.
Strengths & limitations
- Explicitly models persistence and partial adjustment in the outcome, capturing dynamics that static models miss.
- GMM estimation with internal instruments eliminates individual fixed effects without requiring external instruments.
- Consistent even when regressors are endogenous or predetermined, not strictly exogenous.
- The Arellano-Bond m2 and Hansen J-tests provide built-in diagnostics for instrument validity.
- Handles unbalanced panels and heteroscedastic or autocorrelated errors with the two-step GMM variance estimator.
- Requires T >= 3; performance degrades rapidly with very short panels (T = 2 gives no instruments for Difference GMM).
- Instrument proliferation — using all available lags — can overfit the model and weaken the Hansen J-test; best practice caps instrument count below N.
- Asymptotic properties rely on large N; with small N the two-step standard errors are downward biased and Windmeijer (2005) finite-sample correction is essential.
- The validity of internal instruments rests on the assumption of no second-order serial correlation in levels errors — an untestable assumption that must be scrutinised via the m2 test.
- Cannot recover time-invariant effects or long-run elasticities directly; System GMM or ARDL-based approaches may be needed for those goals.
Frequently asked
Why not just use OLS with a lagged dependent variable in a panel?
OLS ignores the unit fixed effect, which is correlated with the lagged dependent variable. This creates upward 'dynamic panel bias' (Nickell bias) that grows worse as T is small. The within (FE) estimator corrects for the fixed effect but introduces a downward bias of order 1/T, so it is also inconsistent for short panels.
When should I use System GMM instead of Difference GMM?
System GMM (Blundell-Bond 1998) is preferred when the autoregressive coefficient alpha is large (close to 1) or regressors are highly persistent, because Difference GMM instruments become weak in those cases. System GMM adds moment conditions in levels, substantially improving precision, but requires the additional assumption that initial conditions are uncorrelated with the fixed effects.
How many instruments is too many?
A common heuristic is to keep the instrument count strictly below N (the number of cross-sectional units). With many periods, use collapsed instruments or restrict the lag depth. Instrument proliferation makes the Hansen J-test uninformative (p-value near 1) and can cause finite-sample bias.
What if the m2 test is significant?
A significant m2 statistic indicates second-order serial correlation in the differenced residuals, which violates the identifying assumptions of Difference GMM. In this case, extend the lag depth of instruments to t-3 or higher, or reconsider the model specification, as the standard instrument set is invalid.
Can I include time fixed effects alongside unit fixed effects?
Yes. Including time dummies (period indicators) is standard practice to control for common shocks affecting all units simultaneously, such as global recessions. They also help satisfy the strict exogeneity of time dummies that supports the moment conditions.
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 ↗
- Hsiao, C. (2003). Analysis of Panel Data (2nd ed.). Cambridge University Press. ISBN: 978-0521522717
How to cite this page
ScholarGate. (2026, June 3). Dynamic Panel Data Model. ScholarGate. https://scholargate.app/en/econometrics/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
- Difference GMMEconometrics↔ compare
- Fixed Effects ModelEconometrics↔ compare
- Panel Data AnalysisEconometrics↔ compare
- Panel Fixed Effects ModelEconometrics↔ compare