Deriving experience curves: A structured and critical approach applied to PV sector

Experience curves are widely used for cost estimates in energy-economy models and are proposed as a forecasting tool for projecting the future environmental impact of emerging technologies. However, further application is limited by data availability and methodological challenges related to modelling the dynamic relationship between cost, different kinds of learning, and scale effects. This paper systematically compares existing experience curves using empirical data from the PV sector. We compare the cost forecast of the assessed experience curves, derive the learning rates over different periods, and draw parallels to the environmental experience curve. Our results show that the single-factor experience curve (SEFC) is the most stable model, showing consistent performance across different technological eras, train-test splits and validation methods. Two-factor and multi-factor experience curves exhibit higher sensitivity, with their performance metrics varying significantly based on the data subsets used. Diagnostic tests are important to examine the robustness of the results. For the environmental experience curve, data quality and model explanatory power are lower, yet there is potential for its applicability in projecting environmental impact and energy use. Policymakers and modellers should consider the specific technological era when using learning rates for decision-making. Our findings indicate that learning-by-doing provides a steady learning rate across all experience curves. In the early stages of technological maturity, cost reductions in the PV industry are driven by learning-by-innovation, which is later dominated by economies of scale.