Working Paper Series: Special Edition of 2016 to 2018 Interns

Bond (1998). This technique is more effective given that the data was used to construct panels having a small “t” and large “i”. Applying the panel GMM controls for country-specific effects and controls for any simultaneity issues that may arise from regressors being endogenous with the response variable. AB-BB (95, 98) 37 estimators iron out these problems by using “ n” 38 lags of the first difference of regressand as instruments. The first difference transformation removes cross sectional effects and this allows equation (3) to be transformed as follows: � − , − � = + � , − − , − � + � − , − � + � − , − � + ( − , − ) (7) The assumption here is that the error term is not serially correlated and the regressors in their lagged form are weakly endogenous. This strategy is called Difference GMM and consist of the following moment functions: [ _ , − ∗ ( − , −1 )] = 0 ≥ 2, t [1,4], > (8) [ , − ∗ ( − , −1 )] = 0 ≥ 2, t [1,4], > (9) [ , − ∗ ( − , −1 )] = 0 ≥ 2, t [1,4], > (10) One drawback of DGMM is that when differencing equation (1), the process removes long-run cross-country information at current levels, and if the column vector , consists of annual persistence, the lagged levels of the variables within X will be weak instruments in the GMM estimation. To rectify this, AB-BB (95, 98) proposed the System GMM estimator. Blundell and Bond (1998) demonstrated that the SGMM panel estimator is able to enhance efficiency by reducing biases and erroneousness characterized by the difference GMM, especially as it relates to problems of weak instruments. Whilst maintaining the estimator’s consistency, the system GMM method modifies equations 8-10 to yield the following moment conditions:

37 Shortening for Arellano and Bover (1995) and Blundell and Bond (1998) 38 n≥1. This paper will use one (1) lag

132