Technological change in energy systems: Learning curves, logistic curves and input–output coefficients

Abstract Learning curves have recently been widely adopted in climate-economy models to incorporate endogenous change of energy technologies, replacing the conventional assumption of an autonomous energy efficiency improvement. However, there has been little consideration of the credibility of the learning curve. The current trend that many important energy and climate change policy analyses rely on the learning curve means that it is of great importance to critically examine the basis for learning curves. Here, we analyse the use of learning curves in energy technology, usually implemented as a simple power function. We find that the learning curve cannot separate the effects of price and technological change, cannot reflect continuous and qualitative change of both conventional and emerging energy technologies, cannot help to determine the time paths of technological investment, and misses the central role of R&D activity in driving technological change. We argue that a logistic curve of improving performance modified to include R&D activity as a driving variable can better describe the cost reductions in energy technologies. Furthermore, we demonstrate that the top-down Leontief technology can incorporate the bottom-up technologies that improve along either the learning curve or the logistic curve, through changing input–output coefficients. An application to UK wind power illustrates that the logistic curve fits the observed data better and implies greater potential for cost reduction than the learning curve does.