The Universality and Predictability of Technology Diffusion

Tune in for the live stream on YouTube 

Abstract: Technology diffusion follows S-curves, in which deployment initially accelerates and then levels off.  We collect data for 47 technologies ranging from monasteries and missles to canals and mobile phones and show that the shape of their S-curves is remarkably universal.  On average, the Gompertz function explains more than half the variance in the level of technology diffusion at the point of maximum growth, suggesting that while each technology's story is different, the similarities are bigger than the differences.  We show that technology S-curve time series suffer from problems of nonstationarity, autocorrelation, heteroscedastic noise and severe estimation bias.  We develop a time series model that takes these problems into account, formulate a method for probabilistically forecasting future deployment and study how its forecasting accuracy varies as a function of forecasting horizon and stage of development.  Application to solar energy and wind indicates that the renewable energy transition will very likely happen quickly, displacing most fossil fuels within 20 years.