Dynamic panels with endogenous interaction effects when T is small

This paper compares the performance, measured in terms of bias, root mean squared error and computation time, of different estimators of the fixed effects dynamic panel data model extended to include endogenous interaction effects when T is small: (i) the bias corrected LSDV (BCLSDV) estimator based on Yu, De Jong and Lee (2008); (ii) the ML estimator based on Hsiao, Pesaran and Thamiscioglu (2002) and Bhargava and Sargan (1983) extended in this paper to include endogenous interaction effects; (iii) the GMM estimator based on Arrelano and Bond (1991) extended in this paper to include endogenous interaction effects; (iv) the ML estimator mixed with the BCLSDV parameter estimate of the endogenous interaction effects; and (v) the GMM estimator mixed with the BCLSDV parameter estimate of the endogenous interaction effects. It is found that the mixed ML/BCLSDV estimator outperforms the other estimators in terms of bias and root mean squared error when T = 5, and that the mixed GMM/BCLSDV estimator is a reasonable alternative for values of N greater than 500 due to differences in computation time.