Consistent Estimation of Growth Regressions

Ordinary least squares is unlikely to estimate growth regressions consistently. A simple alternative method is formulated. Applying ordinary least squares and this method to data on 85 countries over the period 1964-1990 reveals that ordinary least squares is strongly biased. The consistent estimates are about twice as large in magnitude as the inconsistent estimates. This finding suggests that many inferences reached in the empirical growth literature are invalid and may be seriously misleading. In particular, the inference that economies converge slowly to parallel balanced-growth paths appears to result from biased estimates.