A model is said to be dynamic when one of the regressors is the lagged dependent variable. The usefulness of panel data for estimating dynamic models is self‐evident: it is impossible to estimate a dynamic relationship on cross‐sectional data while, in the case of time series data, such model cannot be precisely estimated without drawing on long enough a sample. By contrast, with panel data a dynamic model can be estimated over a set of individuals observed over a small number of time periods. The models presented in this chapter are well suited to “micro‐panels,” i.e., datasets where . For “macro‐panels,” characterized by a temporal dimension equivalent to, or bigger than, the cross‐sectional one, the appropriate models will be based on an adaptation of the methodology employed in unit roots tests and cointegration estimators to the specific issues of panel data.1
Among the many applied examples from the literature, one can mention:
- the estimation of per capita income convergence by regressing the growth rate as a function of the initial wealth level or, equivalently, regressing the level of per capita wealth as a function of the lagged wealth level;
- the analysis of the speed of adjustment of the labor force, obtained by regressing employment over different variables, including lagged employment;
- the dynamic analysis of consumption, based ...