Abstract
This chapter shows how alternative specifications perform when they are used to estimate growth effects of variables with time series and panel datasets. Of the three methods compared, Rao and Singh's approach seems to work the best. This validates the earlier observations that models with stronger theoretical basis perform better. To gain further confidence in these findings, this chapter applies Sala-i-Martin's (1997) and Levine and Renelt's (1992) extreme bounds procedures to test the robustness of variables and the adopted specifications. This is an important but neglected issue in modeling economic growth and in many other empirical studies. The robustness test results imply that all the tested variables are important determinants of long-run growth, estimated with correct signs, plausible magnitudes, and with high confidence. Globalization and financial sector variables show small but statistically significant and robust relationships with long-run growth. Government spending and investment rate are growth enhancing while high rates of inflation are negatively associated with long-run growth. Analyses in this chapter imply that because of the consistent results obtained from alternative tests, Rao and Singh's (2007) extension seems to be robust and useful for empirical application. This result, therefore, provides greater confidence in the application and use of this new extension in empirical studies, especially for but not restricted to, country-specific studies.
TopIntroduction
This chapter compares the empirical performances of Rao and Singh, Senhadji and Barro’s conventional estimation methods. The latter has two variants – standard Barro approach mostly used for panel estimates and Barro-type ad-hoc regression used using alternative datasets. In ad-hoc regressions, often times, applied growth economists regress the growth rate of output (or its variant) against that of other determinants of growth. It must be stated that this is invalid, unless the growth rate can be based on long time period averages and the initial conditions are included. Therefore, application of this methodology is flawed. The Barro-regression unfortunately cannot be used with time series data because the included initial conditions conflict with the constant term in growth regressions. Recall that this method essentially involves regression some measure of the average growth rate of GDP against averages rates of growth of the hypothesized variable(s), the growth rates of conditioning variables and lagged levels of some initial conditions. Senhadji’s methodology applied to time series and panel datasets is promising and is expected to provide comparable estimates of the growth effects of variables. While there are some issues relating to the measurement of TFP, better TFP estimates will be a useful addition to this idea of Senhadji. Since Rao and Singh have shown advancements over Senhadji’s, the latter has further theoretical support and empirical advantages and at presents the state-of-the art approach. It is expected that unless better alternatives are made available, this approach will be highly useful.