A note on the long-run determinants of economic growth

Abstract

Variable selection is the issue of major concern in practical regressions. This note provides a simple and efficient method to examine the robustness of predictor variables in cross-country economic growth models. Our results confirm the general findings of Sala-i-Martin et al. (2004), indicating the importance of a number of same predictor variables. In addition, we also identify that some other variables are associated with economic growth.

Keywords:

• economic growth
• variable selection
• model uncertainty
• boost

JEL Classification:

• C19
• D89
• E10

Funding

This research is partially supported by the National Natural Science Foundation of China [grant number 71171075], [grant number 71221001], and [grant number 71031004].

Notes

¹ In empirical economy, a precursor to apply boosting methods is Kostov (2010). Kostov (2010) applies a component-wise gradient boosting algorithm to deal with the issue of spatial weight matrix uncertainty. The basic idea is to choose the most ‘appropriate’ spatial weight matrix amongst a predetermined set of weight matrixes.

² The function f(.) can be expressed in different forms, e.g. linear components, nonparametric smooth components and spatial effects and interaction surfaces (Kneib et al., 2009).

³ Kneib et al. (2009) point out that ‘the variable selection procedure from boosting algorithm does not automatically choose the most significance variables’.

⁴ In this article we consider only 86 countries since there have some missing value for Iceland and Fiji. We are grateful to the authors for sharing the data.

⁵ All the computations are carried out using the mboost package in R Project (see Hofner et al., 2012).