A review of uncertainties in technology experience curves

The use of log-linear experience curves (or learning curves) relating reductions in the unit cost of technologies to their cumulative production or installed capacity has become a common method of representing endogenous technical change in energy-economic models used for policy analysis. Yet, there are significant uncertainties in such formulations whose impact on key model results have been insufficiently examined or considered. This paper reviews the major types of uncertainty in log-linear experience curves and their effect on projected rates of cost reduction. Uncertainties are found not only in the learning rate parameter of a log-linear model, but also in the functional form that determines the shape of an experience curve. Evidence for alternative forms such as an S-shaped curve is reviewed along with case studies that demonstrate the uncertainties associated with cost increases during early commercialization of a technology—a phenomena that is widely recognized but rarely quantified or incorporated in learning models. Additional factors discussed include the effects of learning discontinuities, institutional forgetting, and the influence of social, economic and political factors. We then review other models of causality, which aim to improve modelers’ ability to explain and predict the influence of other underlying processes that contribute to technology cost reductions in addition to learning. Ignoring other types of underlying mechanisms can create a false sense of precision and overestimate the true contribution of learning. Currently, however, uncertainties in such multi-factor models remain large due to the difficulties of estimating key parameters (such as private-sector R&D investments) and extending models of a specific technology to a broader suite of technologies and cost projections. Pending the development and validation of more robust models of technological change, we suggest ways to significantly improve the characterization and reporting of current learning model uncertainties and their impacts on the results of energy-economic models to help reduce the potential for drawing inappropriate or erroneous policy conclusions.