Aggregation versus heterogeneity in cross-country growth empirics

The cross-country growth literature commonly uses aggregate economy datasets such as the Penn World Table (PWT) to estimate homogeneous production function or convergence regression models. Against the background of a dual economy framework this paper investigates the potential bias arising when aggregate economy data instead of sectoral data is adopted in macro production function regressions. Using a unique World Bank dataset we estimate production functions in agriculture and manufacturing for a panel of 41 developing and developed countries (1963-1992). We employ novel empirical methods which can accommodate technology heterogeneity, variable nonstationarity and the breakdown of the standard crosssection independence assumption. We focus on technology heterogeneity across sectors and countries and the potential for biased estimates due to aggregation and empirical misspecification, relying on both theory and empirical evidence. Using data for a stylised aggregate economy made up of agricultural and manufacturing sectors we confirm substantial bias in the technology coefficients and thus any total factor productivity measures computed. Our empirical findings imply that sectoral structure is of crucial importance in the analysis of growth and development, thus strengthening the recent revival of research on structural change in development economics.