6533b7d2fe1ef96bd125f43a

RESEARCH PRODUCT

Convergence in the OECD: Transitional Dynamics or Narrowing Steady-State Differences?

subject

description

I. INTRODUCTION Research on growth and convergence has proceeded through several stages that can be described as a process of accommodating cross-country heterogeneity into the convergence equation. In the first stage, the world could be described as countries approaching to equal (absolute convergence) or to different (conditional convergence) steady states. In both cases--see Baumol (1986) Barro and Sala i Martin (1992), or Mankiw et al. (1992)--the assumption of parameter homogeneity of the underlying production function was assumed and not tested. Later, some researchers (Knight et al. [1993], Islam [1995], Durlauf and Johnson [1995], or Caselli et al. [1996], among others) began to challenge the view that the productivity shift parameter of the underlying production function is homogeneous across countries. From an econometric point of view, the transition from the first to the second stage made researchers resort to panel data methods that exploit the time dimension of data sets. A natural extension to accommodate heterogeneity was developed in Lee et al. (1997). In this third stage, the authors extended the use of panel data methods to allow for differences not only in the initial conditions (the constant term) but also in other coefficients of the production function and in the rate of technological progress itself. If there is widespread heterogeneity, there are substantial econometric difficulties in obtaining precise estimates of the speed of convergence. In particular, the estimated rate of convergence is biased downward in cross-section regressions (see Lee et al. [1997] or Caselli et al. [1996]), suggesting a slower convergence rate than is actually the case. (1) The pooled fixed-effect estimator would partially solve the problem of accommodating level effects across countries through heterogeneous intercepts, but in dynamic panels heterogeneity in speeds of convergence renders this estimator inconsistent, too. Pesaran and Smith 0995) show that the mean group estimator (i.e., estimating the convergence equation separately for each country using annual data and averaging the coefficients) is appropriate in this case. Nevertheless, as recognized by these authors, this procedure may also be subject to a small sample bias that can be important even for time dimensions as large as 30 years. In this article, we do not discuss further the advantages or shortcomings of these different econometric methods to estimate convergence equations. (2) Our aim is not presenting an alternative econometric method to estimate consistently the parameters in growth regressions; rather we focus on some implications of the different econometric techniques that so far have been extensively used in the convergence literature and that have been largely unexplored. To this end, we present three well-established estimation approaches as a benchmark of the way the literature has estimated convergence equations in some fundamental contributions, and we explore the implications of parameter heterogeneity within the standard Mankiw, Romer, and Weil theoretical framework. Cross-section and fixed-effects models lead us to conclude that the observed reduction in the dispersion of per capita incomes that has taken place among Organisation for Economic Co-operation and Development (OECD) countries since the 1960s is mostly a process of transitional dynamics. According to this interpretation, the dispersion of the long-run (steady-state) features of OECD economies has remained fairly stable during the past 40 years. If the dispersion of incomes has narrowed, it is because these countries started from very different initial conditions and have been approaching their own steady state. This is in accordance with the predictions of the standard exogenous growth model. Mean group estimates tell us a different story, namely, that the process of transitional dynamics has been negligible. OECD countries have never been too far from their steady states, and the reduction of dispersion is mostly explained in terms of convergence in steady states themselves, that is, in the long-run determinants of per capita income (savings rates, human capital accumulation, etc. …