One way to determine whether you have a dynamic model is to estimate a dynamic model by adding lagged dependent variables and testing whether the lagged dependent variables are significant. If the lagged dependent variables are significant you should use a dynamic model and a dynamic panel estimator. If the lagged dependent variables are not significant you can use a standard panel (fixed effects) estimator. GMM estimators are preferred to the (LSDV) fixed effects estimator when you have panel data and a dynamic model (it includes a lagged dependent variable as a regressor). The GMM dynamic panel estimators are appropriate for large N and small T. For a broader discussion and how to apply GMM methods in Stata see: Roodman, D. (2009). How to do xtabond2: An introduction to "Difference" and "System" GMM in Stata. The Stata Journal, 9(1), 86-136. However, if T is large GMM estimators can become unreliable because the number of instruments becomes large and the instrumented variables can be overfitted and so may not remove the endogenous components of the lagged dependent variable(s) as intended. When N is not large and T is moderate you may wish to use a bias corrected LSDV estimator to deal with dynamic panel bias, although these assume that all variables other than the lagged dependent variable are strictly exogenous. To apply a bias-corrected LSDV estimator to a potentially unbalanced dynamic panel in Stata see: Bruno, G. (2005). Estimation and inference in dynamic unbalanced panel-data models with a small number of individuals. The Stata Journal, 5(4), 473-500. An example paper that uses Bruno's estimator is given in: Goda, T., Stewart, C. and Torres García, A. (2020) ‘Absolute Income Inequality and Rising House Prices’, forthcoming in Socio-Economic Review. An example application of GMM dynamic panel estimators is: Matousek, R., and Nguyen, T. N., and Stewart, C. (2017) ‘Note on a non-structural model using the disequilibrium approach: Evidence from Vietnamese banks’, Research in International Business and Finance, 41, pp. 125 – 135.

Thank you so much Sir

A panel data model is dynamic when the regressors include the endogenous variable lagged by at least one period. The presence of this autoregressive component on the right-hand side of the equality leads to an endogeneity problem, because this lagged variable will be correlated with the individual effects present in the error term. To overcome this endogeneity problem, one must resort to instrumental variable (IV) estimation method or generalized moments method (GMM), taking care to select the instruments carefully (see weak instruments issues above).

Thank you everyone

Dear Chris, How are You? Maybe You couldn’t remember me, I was just a listener during your lectures at London Met UNI back in 2011 (Roman Matousek was my supervisor). Since then, it’s been a while. I appreciate Your effort, however, when our panels are not stationary, and the resids suffer from heteroskedasticity, the system GMM (B-B approach) is the right choice even though the lagged value is not significant, is it not? In that specific case xtreg with fixed effects is not the solution, or am I wrong? Furthermore, Kripfganz (2019) has suggested really nice STATA developments at the STATA conference in London..xtdpdgmm command is just brilliant to my mind..letting you test even under-identification, not just the overidentification according to Sargan-Hansen techniques. Although, there are still some critical issues mentioned by Windmeijer (2018) that I’m dealing with.

Hi Tomas, I remember teaching at London Met back in 2011 although I've now moved to Kingston and Roman is at Queen Mary I believe. I hope all is good with you. My understanding is that while the Blundell-Bond System GMM estimator is appropriate for "random walk-like" variables it still assumes the data are stationary. My interpretation is that GMM should be used for stationary dynamic panel data and system GMM is more appropriate when some variables are near unit root processes. With small T it may be difficult to assess this (although some variables may be regarded as intrinsically nonstationary a priori, such as GDP and prices). I've not read Kripfganz (2019) and there may developments there to take into account. However, if the data are nonstationary there is a case for using panel cointegration techniques, which are available. Although care needs to be taken to account for cross-sectional dependence and in interpreting the alternative hypothesis. Many of these techniques specify the alternative hypothesis as at least one cross-sectional unit in the panel being cointegrated (stationary).