You’ve heard it in the news. You’ve seen it in your feeds. The cost of solar is declining, rapidly. This is the first of two articles that will help readers understand the declining cost of solar. In this first post we cover learning curves and engineering assessments; these are the two major methods used explain the price decline.
Let’s get granular
Globally the cost of producing solar photovoltaics (PV) has declined since the 1980’s in response to and consequently encouraging the growth of the industry. The prevailing wisdom is that learning curve analysis is an appropriate explanatory theory of this well-documented price decline. However, factors such as policy, industry evolution and local market conditions help provide a complete explanation of the price decline.
Learning and experience curves
The learning curve theory is pervasive in the solar industry (and others) as an explanation of why the cost of producing something gets cheaper over time. The idea is simple: the learning curve effect can be summed up as the relationship between experience and efficiency. Experience curves are used to describe how unit costs decrease with cumulative production growth; because the more products you put out, the more experience you have at a task, the more efficiently you do it. The phenomenon happens exponentially. It’s intuitive and that probably explains its staying power.
Whilst applied in many fields, for solar PV this relationship is best illustrated by Swanson’s Law (a derivative of Moore’s Law) based on the observation and projection that in the solar industry there is a 20% reduction in PV module costs for every doubling of cumulative volume shipped. Learning curve analysis is supported by a host of academics and analysts who use it to simulate technological changes and how these changes will affect technology costs over time.
The solar PV industry has largely followed Swanson’s Law as can be seen in the chart below. With every doubling of cumulative PV module shipments, the average selling price decreased by 21,5%. This price decline is also referred to in learning curve analysis as a learning rate (LR), therefore globally solar PV has an LR of 21,5%.
Chart 1: Learning curve for module price as a function of cumulative PV module shipments, 1976 to 2015 (in MWp)
Source: ITRPV, 2016
The cost reductions in the solar PV industry related to learning curves are attributed to superior performing components (development of higher efficiency cells), lower costs of manufacturing (less wasted polycrystalline silicon), best practice integration, deployment methods and economies of scale. The remarkable rise of solar in the global market (and charts like the above) seemingly lend credibility to the learning curve hypothesis and the role of cumulative shipments as the key driver of cost reductions, however the story is more complicated.
The problem with learning curves
Single-factor learning curve analysis, similar to the one used above, has been criticised for its inability to account for the role of R&D, omitted variable bias and the tricky role of spillover effects. Some learning curve analyses have managed to account for the role of R&D in cost reduction and strong relationships have been found between learning-by-researching (R&D) and improvements in the production process.
Other researchers found that using learning curve analysis with cumulative output yields resulted in a wide range of LR estimates and correlation coefficients, indicating the presence of the omitted variable bias (a model that’s missing important variables). The implications of this bias are significant as it results in the attribution of cost reduction to the wrong variables as the analysis is capturing the influence of unevaluated variables such as input prices or scale effects. Pierre Taillant provides an excellent critique of the role of spillover effects in the decline of solar prices; by way of example, he argues the solar industry has surely benefited from increased computing power, manufacturing advances and silicon technologies but econometric methodologies to capture this effect are limited.
In sum, learning curves are not the perfect tool for explaining the cost reductions in solar PV as they likely exaggerate the cost reductions attributable to experience effects by failing to account for other important factors.
Therefore, learning from experience is only one factor in accounting for the cost reductions in solar PV. Rather than infer from industry numbers, other analysts seek to understand the decline in price using more traditional methods: they talk to industry insiders and assess the manufacturing components and the business.
Engineering assessments
Engineering assessments (bottom-up analysis) take a more granular approach to cost reduction potential by collecting technology-specific data and using qualitative analysis tools like industry surveys to provide a clearer picture of the landscape. This seems like a better approach. What does it have to say about learning curves?
A 2006 study of solar PV manufacturing found that cumulative capacity has a weak effect on the most important cost reducing factors (plant size, efficiency and inputs) and the experience variables “do not appear to have been major factors in enabling firms to reduce the cost of PV.” A scathing critique of the assumptions that underpin learning curve models.
To explain the price decline learning curve analyses point to improved manufacturing processes. Engineering assessments argue what we’re observing is a result of a scale and learning explains little. Clearly, there is an incoherence. It should be noted that both methods have been criticised for their inability to capture market forces and, ironically, some learning effects. But what they both agree on is the importance of demand. So in Part 2, we follow the demand.