Truth and Robustness in Cross‐country Growth Regressions*

Abstract

We re-examine studies of cross-country growth regressions by Levine and , 1997b). In a realistic Monte Carlo experiment, their variants of Edward Leamer's extreme-bounds analysis are compared with a cross-sectional version of the general-to-specific search methodology associated with the LSE approach to econometrics. Levine and Renelt's method has low size and low power, while Sala-i-Martin's method has high size and high power. The general-to-specific methodology is shown to have a near nominal size and high power. Sala-i-Martin's method and the general-to-specific method are then applied to the actual data from Sala-i-Martin's original study.

Labour Market Performance in OECD Countries: A Comprehensive Empirical Modelling Approach of Institutional Interdependencies

“…This method uses inclusion probabilities of individual variables to approach the true model. 6 Alternatively, Krolzig and Hendry (2001) and Hoover and Perez (2004) suggest a general-to-specific approach based on statistical tests belonging to the frequentist strand. Hendry and Krolzig (2004) provide a programme PcGets for model selection purposes which relies on a general-to-specific procedure and searches along multiple paths.…”

What is Really Good for Long-Term Growth? Lessons from a Binary ClassificationTree (BCT) Approach

“…As can be seen, two of the studies that each claims to be the "gold standard" (Hoover andPerez 2004 andSDM 2004) only have two common significant variables-fraction of population that is Confucian and the number of years an economy is open-out of the universe of possible variables. These two common variables do not give much comfort as to what is really important for growth.…”

“…They claim that their models: (i) are statistically well-specified, i.e., with white noise errors; (ii) are valid restrictions of the general model; and (iii) encompass every other parsimonious regression that is a valid restriction of the general model; and (iv) are very cost effective ("We ran one equation", Hendry and Krolzig 2004). In contrast to the extreme bounds approach, Hoover and Perez (2004) and Hendry and Krolzig (2004) find that there are in fact only five variables that can robustly explain growth, in ascending order: (i) number of coups and revolutions; (ii) fraction of the population that is protestant; (iii) fraction of the population that is Confucian; (iv) equipment investment as a fraction of GDP; and (v) the number of years a country has been an open economy.…”

“…The proponents of this approach, such as Hoover and Perez (2004) and Hendry and Krolzig (2004), claim that their "method has proved superior to extreme-bounds methods in isolating the truth-when the truth is to be found" (Hoover and Perez 2004). They claim that their models: (i) are statistically well-specified, i.e., with white noise errors; (ii) are valid restrictions of the general model; and (iii) encompass every other parsimonious regression that is a valid restriction of the general model; and (iv) are very cost effective ("We ran one equation", Hendry and Krolzig 2004).…”

“…Interestingly, Hoover and Perez (2004) give short shrift to most of the favorite growth theories, such as "institutions rule," "it's geography, stupid', financial factors, health or human development. As underscored by SDM, "the multiplicity of possible regressors is one of the major difficulties faced by researchers trying to make sense of the empirical evidence on economic growth."…”