In their latest NBER working paper on economic convergence, Kremer, Willis and You (2021) argue that there was a trend towards unconditional convergence (GDP per capita today depends on its GDP per capita in the past) since 1990 and absolute convergence since 2000 owing to broad faster catch-up growth and slower growth of the frontier economies. They also show that many of the correlates of growth such as human capital, policies, institutions, and culture also showed convergence with higher income country groups. As both correlates and growth have changed, their relationships (coefficients of growth regressions) have also changed. The shrinkage of growth regression coefficients has led to the narrowing of the gap between unconditional and conditional convergence. Absolute convergence has converged towards conditional convergence.
Major highlights from the working paper:
There was a steady trend towards unconditional convergence since the late 1980s and absolute convergence since 2000. Between 1985 and 1995, income per capita diverged by an average 0.5% annually, but between 2005 and 2015, it converged at a rate of 0.7%. The richest quartile of countries had the fastest growth in the 1980s, but the slowest growth since then as it became flat in the 1990s and then declined since 2000. However, the three other quartiles all experienced accelerating growth through 1990s and early 2000s.
They argue that the conditional convergence could be due to (i) faster spread of technologies due to globalization as well as greater capital and labor mobility (accelerates convergence), and (ii) convergence in growth correlates themselves such as human capital, policies, institutions and culture (these could narrow the gap between unconditional and conditional convergence).
The authors also explore whether growth correlates have changed over time by looking at four groups. First, enhanced Solow fundamentals, namely investment rate, population growth and human capital, which are fundamental determinants of steady state income. Second, short-run correlates, namely policies (political and financial institutions, fiscal policy) that can change in relatively short period of time. Third, long-run correlates, namely historical determinants of institutions and geography that change slowly, if at all. Fourth, culture (they consider 10 cultural variables such as views on inequality, political participation, the importance of family, traditions, work ethics, etc). They find that these correlates have not been highly persistent as many have undergone large changes and themselves converged substantially across countries, especially towards those of rich countries.
Establishing relationship between two trends -- towards convergence in income and the convergence of many of the correlates of growth itself-- is a bit challenging because causality can run both ways (for instance, converging income causes policies, institutions and culture to converge or it could be the other way round). By using omitted variable bias formula, they decompose the gap between absolute convergence (convergence across countries without conditioning on determinants of steady state income) and conditional convergence by looking at the product of the slopes of two relationships: correlate-income slopes and growth-correlate slopes.
They find that while cross-sectional relationships between income and the correlates have changed in levels, their slopes have mostly remained stable. However, growth-correlates regression coefficients have shrunk substantially and show little autocorrelation for correlates (institutional homogenization) except for Solow model fundamentals (investment rate, population growth, and human capital), which have remained stable. This flattening of growth-correlates relationship suggest that absolute convergence converged towards conditional convergence.
They argue that their results are consistent with neoclassical growth models, particularly after 1990. While conditional convergence held throughout the period (1985-2015), absolute convergence did not initially but as human capital, policies and, institutions improved in poorer countries, their explanatory power with respect to growth and convergence have declined. So, "the world has converged to absolute convergence because absolute convergence has converged to conditional convergence". Policies and institutions used to matter, but that they have converged, they matter less (and their effects are non-linear). Evidence support convergence in policies and institutions, but divergence in income over 1960-1990 (inconsistent with neoclassical growth model). However, convergence changed since 1990 and the outcome is consistent with neoclassical growth models (but not AK models such as endogenous growth models or poverty trap models that predict divergence).
*******
Meanwhile, Acemoglu and Molina (2021), in a follow up comment, argue that the results of Kremer, Willis and You (2021) working paper are driven by the lack of country fixed effects controlling for unobserved determinants of GDP per capita across countries. This "create a bias in convergence coefficients towards zero and this bias can be time-varying, even when the underlying country-level parameters are stable." They find no evidence of major changes in patterns of convergence and no flattening of the relationship between institutional variables and economic growth.
Pande and Enevoldsen (2021) argue that convergence towards development-favored policies that drove the trend outlined by Kremer, Willis and You (2021) also mean that this has happened with rising within-country inequality, resulting in more of the world's poor clustering in (lower) middle-income countries.
