Comparing expert elicitation and model-based probabilistic technology cost forecasts for the energy transition

Area of Science:

• Energy economics
• Technology forecasting
• Climate policy analysis

Background:

• Accurate technology cost forecasting is crucial for effective energy and climate policy.
• Existing methods include expert elicitation and model-based approaches (e.g., Wright's and Moore's laws).
• Previous comparisons have not systematically evaluated these methods against observed data.

Purpose of the Study:

• To systematically compare the performance of expert elicitation and model-based technology cost forecasting methods.
• To assess forecast accuracy using historical data for energy technologies.
• To forecast 2030 costs for multiple energy technologies using both method types.

Main Methods:

• Generated probabilistic technology cost forecasts using expert elicitation and model-based approaches (deployment- and time-based).
• Validated forecasts by comparing them to observed costs in 2019 for six energy technologies.
• Produced 2030 cost forecasts for 10 energy technologies using both elicitation and model-based methods.

Main Results:

• Model-based methods demonstrated superior performance, with forecast ranges more frequently encompassing observed 2019 costs and medians closer to actual costs.
• All forecasting methods, including model-based and elicitation, underestimated technological progress, likely due to structural changes driven by policy and market forces.
• Elicitation methods produced narrower uncertainty ranges for 2030 forecasts compared to model-based methods, which varied based on technology modularity.

Conclusions:

• Model-based forecasting is generally more accurate than expert elicitation for energy technology costs.
• Current forecasting methods struggle to capture rapid technological progress influenced by market and policy shifts.
• Future research should prioritize developing and validating forecasting methods that better incorporate structural market changes and inter-technology correlations.