Solow residuals without capital stocks

Abstract

We use synthetic data generated by a prototypical stochastic growth model to assess the accuracy of the Solow residual (Solow, 1957) as a measure of total factor productivity (TFP) growth when the capital stock in use is measured with error. We propose two alternative measurements based on current investment expenditures: one eliminates the capital stock by direct substitution, while the other employs generalized differences of detrended data and the Malmquist index. In short samples, these measures can exhibit consistently lower root mean squared errors than the Solow–Törnqvist counterpart. Capital measurement problems are particularly severe for economies still far from their steady state. This drawback of the Solow residual is thus most acute in applications in which its accuracy is most highly valued. As an application, we compute and compare TFP growth measures for developing countries in the Heston–Summers dataset.