Instrumental Variable Estimation of Dynamic Linear Panel Data Models with Defactored Regressors under Cross-sectional Dependence

This paper develops an instrumental variable (IV) estimator for consistent estimation of dynamic panel data models with error cross-sectional dependence when both N and T, the cross-section and time series dimensions respectively, are large. Our approach asymptotically projects out the common factors from regressors using principal components analysis and then uses the defactored regressors as instruments to estimate the model in a standard way. Therefore, the proposed estimator is computationally very attractive. Furthermore, our procedure requires estimating only the common factors included in the regressors, leaving those that influence the dependent variable solely into the errors. Hence aside from computational simplicity the resulting approach allows parsimonious estimation of the model. The finite-sample performance of the IV estimator and the associated t-test is investigated using simulated data. The results show that the bias of the estimator is very small and the size of the t-test is correct even when (T,N) is as small as (10,50). The performance of an overidentifying restrictions test is also explored and the evidence suggests that it has good power when the key assumption is violated.