Using the extended Ramsey rule, the socially efficient rate is the difference between a wealth effect and a precautionary effect of economic growth. This second effect is increasing in the degree of uncertainty affecting the future. In the literature, it is usually calibrated by estimating the historical volatility of the growth of GDP in a specific country. In this paper, I show that using cross-section data tends to magnify uncertainty, and to reduce the discount rate. Using a data set covering 190 countries over the period 1969-2010, I justify using a much smaller discount rate around 0.7% per year for time horizons exceeding 40 years.