The Long Wave of Conditional Convergence1

We calculate the time series of the speed of convergence for 21 high-income countries over the period: 1953-1996, using low-pass filtered time series of per-capita GDP which are thus isolated from the in-fluence of the short-run business cycle components. The observed patterns contradict the conventional ‘time-invariant speed of convergence ’ hypothesis. Furthermore, dynamic panel data analysis provides strong evidence of the existence of stationary long cycles in the per capita GDP time series. We develop and estimate a technology-diffusion-based endogenous growth model, which shows that the endogenous growth of the domestic knowledge stock can account for the long cycles observed in the data. Key words: trend reversion, speed of convergence, growth cycles