PurposeThe research questions to be answered by this meta-analysis are as follows: What is the average effect in the literature of an increase in debt on a country’s economic growth? Is the direction of this link positive, negative or zero? Is there, and to what extent, a certain degree of heterogeneity in the results of the studies analyzed? If heterogeneity exists, what influences it? Is there publication bias in this area of research? If so, in which direction?Design/methodology/approachThe methodology employed in the development of a meta-analysis of the literature regarding the debt–growth relationship is based on the seminal paper by Stanley and Jarrell (2005). In this research, we endeavor to adhere as closely as possible to the reporting guidelines established by the Meta-Analysis of Economics Research Network (MAER-Net) (Stanley et al. (2013)), which have been recently updated by Havránek et al. (2020). Therefore, we will first define the effect size and describe the coding phase of the studies. Subsequently, we will present the forest plot and analyze publication bias. The theoretical model adopted will be introduced, and the results concerning the analysis with fixed effects, random effects and the moderator analysis will be shown. Finally, several meta-regressions will be estimated. Additional material can be found in online Appendix.FindingsFirst, with regard to publication bias, the analysis indicated a positive asymmetry of the funnel plot, which led to an over-representation of studies with positive effect sizes. The estimated average effect size, as determined by this analysis, is situated between −0.5 and −0.9. Additionally, the substantial prevalence of p-values below 0.05, as evidenced by the three-parameter selection model and the p-curve analysis, indicates the presence of publication bias. Statistically significant results at the 95% level are more likely to be published than results with p-values exceeding the 0.05 threshold. The mean effect size is −0.2 in the multi-level analysis, while it is slightly larger in absolute terms in the analysis of the entire sample and zero in the reduced sample (where the average PCC for each study is considered). Heterogeneity is a prominent feature of the data, with differences observed both within and between studies. The within-study variability is more pronounced than the between-study variability. Heterogeneity persists when moderators are analyzed. Among these, the moderators that lower the level of heterogeneity and thus explain the different estimates across studies are region, income and development level, the variables used as proxies and the methodology used.Originality/valueFirst, the paper sets out to quantify the debt–growth nexus, using all the relevant literature. In the most recent crises, policymakers have taken expansive fiscal policy measures to stimulate the economy. Many academics concur with this approach. Given the limited spending capacity of some economies, new debt instruments have been adopted, for example, in Europe, to finance NRRPs. It is therefore of great interest to ascertain the extent to which new debt issuance is correlated with higher economic growth and to examine how this correlation varies over time and across different geographical locations. The meta-analysis by Heimberger (2023) primarily focuses on the nonlinearities of this relationship, which we do not rule out characterizing. In contrast, here we propose the first comprehensive meta-analysis on the linear relationship between public debt and growth. Furthermore, this study introduces the use of the partial correlation coefficient (PCC) as an effect size. This is the inaugural study of this coefficient in the context of the debt–growth relationship, and it offers several advantages. Indeed, the variable employed is unitless, thereby facilitating the comparison of studies conducted with disparate methodologies, samples and numbers of regressors. Nevertheless, our estimates can serve as a reference point for those who wish to propose supplementary meta-analyses employing a distinct effect size. Moreover, the analysis is robust in all its points, as several methods are proposed both to identify publication bias, to analyze moderators and to assess the average effect size and heterogeneity. Indeed, one novelty is the use of estimation techniques such as deep moderator analysis and the BMA method for metaregression. Finally, for the first time, we include both continuous and categorical moderators (which are transformed into dummy variables only for the Bayesian model averaging analysis). In addition, we analyze additional moderators not considered in previous meta-analyses, such as region, external debt, income level, proxies for dependent and independent variables and focus on nonlinearities.
Read full abstract
Jan 15, 2025
Z Matisakova 
Just Published
Open Access