Focusing on technological learning in learning rate assessments
How closely do learning rate estimates measure the thing they are supposed to measure – technological progress – and to what extent are they affected by technology-external factors? We investigated how exchange rates affect learning measurements and developed a method to solve the problem.
A few years ago, we performed an empirical analysis of the learning rate of concentrating solar power (CSP), subsequently published in Nature Energy. The learning rate describes how the cost of a technology decreases as the cumulative output increases, due to factors such as learning-by-doing and economies of scale: the more of something we have produced, the cheaper is the next unit, as we are better at using and producing it. For example, the International Renewable Energy Agency finds that large-scale solar photovoltaic (PV) power has had a learning rate of 35% for the last decade, meaning that its cost decreased by 35% each time the global PV fleet doubled.
During preparation of that article, we used Euro (EUR) as base currency – the currency to which all costs used in the analysis are converted – as most CSP stations had been built in Spain by European companies. However, as CSP had started growing outside Spain as well, it increasingly became a global technology. Anticipating reviewer criticism for the non-standard choice of the Euro as base currency for a global learning rate analysis, we changed the base currency to US Dollar (USD) just before submitting the manuscript. Surprisingly to us, this changed our results: for some technology configurations, the observed learning rates were only half as high in EUR as in USD. In one case, the observed CSP cost decreased over the investigated time span when described in USD but increased in EUR, as the USD appreciated strongly during this period. We included this currency-dependency finding in the supplementary material, and published the article with the USD-based analysis.
After publication, the issue remained at the back of my head, leading me to present the problem to my colleagues Bjarne Steffen and Tobias Schmidt – and although their initial reaction was a frown and a surprised “huh”, we decided to pursue the issue together. A week or two later, Bjarne mailed me a photo of his whiteboard with a set of equations for how to better handle the dependency on the base currency for technologies in global markets. Together with my group members Marc Melliger and Lana Ollier, we developed a deeper analysis of the problem and how to solve it, showing that the problem is large enough to affect the conclusions of energy models and technology forecasts – thus potentially biasing policy action, due to biased input data. Our solution is to include an adjustment factor that renders learning rates immune to exchange rate fluctuations.
Originally, we did not set out to identify or solve the exchange rate effect, but nevertheless we found it, and hope that our work will contribute to producing better estimates in the future, where technological development is increasingly likely to happen on a global scale. In this sense, this research process also shows the importance of staying curious and the role of unexpected findings to open new research avenues. It is not the “heureka”-moments that are most interesting in research, but those that begin with surprise – such as a “huh” and a frown – and go into unknown territory.