Empirically grounded technology forecasts and the energy transition

The report examines how empirically validated probabilistic forecasting can improve estimates of future energy technology costs. It argues that traditional energy-economy models have systematically underestimated deployment rates for renewable technologies while overestimating their costs. Using historical data for solar PV, wind, batteries and electrolyser technologies, the authors construct probabilistic cost forecasts and apply them to model global energy system pathways. Across three scenarios, a fast transition to clean energy is shown to be economically advantageous, with expected savings of trillions of dollars relative to continued fossil fuel use.

Introduction

The report shows that the long-run trend for fossil fuels is static, with real prices for coal, oil and gas showing no persistent decline over 140 years. By contrast, solar PV, wind and batteries have recorded exponential cost declines of around 10% per year. Technology deployment trends also show rapid global uptake of these clean energy technologies, similar in scale to the rapid expansion of nuclear power in the 1970s but with consistent cost reductions.

Most integrated assessment models (IAMs) have underestimated renewable deployment and overestimated costs, leading to pessimistic energy transition scenarios. The report highlights that historical improvement rates justify the use of probabilistic forecasting grounded in past technological performance.

Results

The authors apply a stochastic version of Wright’s law to forecast costs for solar, wind, batteries and polymer electrolyte membrane electrolyser systems. Back-testing with over 50 technologies shows this method reliably captures observed volatility and long-term improvement trends.

Forecasts for fossil fuels use autoregressive (AR(1)) modelling consistent with their historical price behaviour. For solar, the 95% confidence interval for 2050 costs ranges from roughly USD 2–40/MWh under a fast transition pathway, compared with historical oil price uncertainty of USD 20–110 per barrel. Probabilistic forecasts remain sensitive to deployment rates, with faster deployment accelerating cost reductions.

IAM projections for solar and wind have persistently overstated future costs. For example, IAM projections from 2010–2020 assumed solar costs would fall by only 2.6% per year, whereas actual declines were around 15% per year. The report notes that widely used IAM floor-cost constraints have repeatedly been exceeded by real-world data.

From single technologies to a full system model

A simplified global energy system model is constructed, including fourteen key technologies and using exogenous scenario design. Three scenarios are analysed: Fast Transition (fossil fuel phase-out by ~2050), Slow Transition (phase-out by ~2070) and No Transition (continued fossil dominance).

The Fast Transition assumes rapid renewable deployment, large-scale electrification, sector coupling and substantial use of power-to-X fuels for hard-to-electrify sectors. The model also incorporates significant storage capacity, including enough energy storage to operate the system for one month without solar or wind generation.

How much will each scenario cost?

Using Monte Carlo simulation and a 2% discount rate, the expected annual system cost in 2050 is USD 5.9 trillion for the Fast Transition compared with USD 6.3 trillion for the No Transition. The expected net present saving of a fast energy transition is approximately USD 12 trillion at a 1.4% discount rate and around USD 5 trillion at a 5% rate. Across all discount rates, there is roughly an 80% probability that a fast transition is cheaper than continued fossil fuel reliance.

Discussion

The study concludes that a rapid clean energy transition is highly likely to reduce system costs even without accounting for climate damages or co-benefits. The authors recommend updating modelling assumptions in IAMs to reflect empirically grounded cost trajectories, highlighting that expectations of high transition costs are inconsistent with historical data.