Instrumental variable estimation of semiparametric dynamic panel data models: Monte Carlo results on several new and existing estimators

Is the relationship between inequality and growth affected by physical and human capital accumulation?

“…Two alternative procedures are used to estimate model (2.1): Berg et al [7], and Li and Stengos [21]. Without loss of generality, we will refer to the following, more compact representation of model (2.1) when discussing the estimation procedure:…”

“…3.3 can be consistently estimated, however, to do so while ignoring the one-way error structure of the model, which is essentially what Li and Stengos [21] do, yields an inefficient estimator. To overcome this, Berg et al [7] suggest a simple Feasible Generalized Least Squares (FGLS) procedure whereby a within transformation is performed (i.e. each country's observations are expressed in deviations from their country-specific means), thus eliminating the nation-specific terms (υ i ).…”

“…3.3. 7 Once the coefficient estimates are obtained, the next step is to partial-out the linear terms from Eq. 3.1:…”

“…a standard normal pdf), and h is the window width. 9 3.3 Robust estimation procedure While the estimation procedures discussed above ( [7], and [21]) are fairly common in the semiparametric literature, both face inherent difficulties in estimating Eq. 3.1.…”

“…3.3 virtually eliminates the male and female education covariates. 10 To overcome this difficulty, an iterative estimation technique, based loosely on the Berg, Li, and Ullah [7] and 6 The Li and Stengos [21] procedure is very similar to Berg et al [7], except that the one-way error component structure of the model is not exploited (i.e. no within transformation is performed prior to using instrumental variables to estimate the linearized, dynamic panel model), thus the Li and Stengos [21] procedure is consistent, but not efficient.…”