Bending the learning curve

The aim of this paper is to improve the application of the learning curve, a popular tool used for forecasting future costs of renewable technologies in integrated assessment models (IAMs). First, we formally discuss under what assumptions the traditional (OLS) estimates of the learning curve can deliver meaningful predictions in IAMs. We argue that the most problematic of them are the absence of any effect of technology cost on its demand (reverse causality) and the ability of IAMs to predict all determinants of cumulative capacity. Next, we show that these assumptions can be relaxed by modifying the traditional econometric method used to estimate the learning curve. The new estimation approach presented in this paper is robust to the two problems identified but preserves the reduced form character of the learning curve. Finally, we provide new estimates of learning curves for wind turbines and PV technologies which are tailored for use in IAMs. Our results suggest that the learning rate should be revised upward for solar PV. Our estimate of learning rate for wind technology is almost the same as the traditional OLS estimates.