Methods to make probabilistic predictions of the cost and deployment of specific technologies based on historical data are being developed by the Complexity Economics Programme at the Institute for New Economic Thinking (INET), Oxford Martin School.
Instead of relying on optimization, which does not resemble real-world policy implementation, the methods rely on historical data. Information on specific technologies’ future cost and deployment can support MoFs in planning investments and identifying appropriate technologies.
- Integrated assessment models (IAMs) and the International Energy Agency’s energy models are the most widely used tools for technology forecasting. IAMs are optimizing models that aim for paths that maximize economic growth contingent on climate change. However, implementing such paths requires a benevolent global decisionmaker, which is not how policies have been implemented historically. Moreover, IAMs are limited to testing policies that can be implemented as a tax (or equivalent).
- IAMs often invoke incorrect assumptions, including regarding technologies’ minimum costs and maximum deployment rates.
- The team at INET studies future technology costs and deployment based on past costs and deployment. Technologies with costs that have decreased in the past will very likely see costs continue to decrease, and future deployment can be forecast based on past deployment and by invoking S-curves, i.e., exponential growth followed by leveling out as maturity is reached.
- Probabilistic technology cost and deployment predictions from historical models can help identify which technologies to support financially and which are likely to prevail globally.
- A limitation of this approach is that historical technology models assume typical behavior and do not consider potential changes under different policies. Additionally, the models indicate which technologies could be good bets, but not how support is best provided.
Macrocosm Inc. has developed Excel-based tools that make the models accessible via a user-friendly interface. Alternatively, the models can be implemented inside other models; the Python code is publicly available on GitHub.
INET is also designing agent-based models for the energy system that represent individual companies and estimate their future profits and losses under different policy scenarios. This will enable the testing of policy combinations (rather than just tax[-equivalent] policies) in different countries from 2025; more resources are required before implementation.
Keywords
data-drivenempirical methodsmodelstechnologytime series
