A Reassessment of the Finance & Growth Nexus using Monte Carlo Evidence

The small sample performance of system Generalized Method of Moments (GMM) and an alternative Transformed Maximum Likelihood (TML) estimator in Monte Carlo simulations is investigated, implementing  a data generating process that tracks important features inherent in dynamic panels used in finance & growth regressions. It is allowed for weak exogeneity of the additional control variable and violations of initial mean stationarity. System GMM estimators tend to be biased towards Ordinary Least Squares (OLS) estimates when either the variance of the unobserved heterogeneity is large compared to the variance of the idiosyncratic error term or when initial mean stationarity is violated. Collapsing the instruments matrix tends to improve on mean bias and root mean squared error (rmse), reducing the overall instrument count. TML coefficients are generally downward biased but robust to initial mean stationarity violations and more reliable in estimating the lagged dependent variable. Results from an application to the finance & growth literature suggest that a positive and monotonic relationship might well be an artifact of an upward bias of system GMM towards OLS estimates. An inverse U-shaped relationship is also not statistically robust.