Photovoltaic learning rate estimation: Issues and implications

This paper surveys the results of estimating learning rate (LR) equations for the photovoltaic (PV) industry at the world level, and reports new results, placing emphasis on estimation issues, and other shortcomings surveyed recently. The results are reported in detail, one relevant finding being that the learning rate parameter might reach values substantially higher than those usually reported (18-20%). This result, however, does not necessarily translate to other energies. The relevance of selecting the estimation sample, dynamic specification, and omitted variables in simple standard specifications for the estimated learning rate is highlighted. A solution for the LR in dynamic non stationary models is presented. The modeling of silicon prices is also discussed, and the concept of the total learning rate (TLR) is introduced. Probability confidence intervals for the main estimated learning rate parameters are analyzed, and the time decomposition of PV module prices is discussed, highlighting the role of fossil energy prices. It is found that the total LR might reach values above 27% with a 95% probability.