Image: Andasol Solar Power Station, Spain. Credit: kallerna, (CC BY-SA 4.0) via Wikimedia Commons.



As we get ever closer to the climate change tipping point, action is needed now to limit the damage we have done to our planet. Not only is green energy part of the solution, it can also provide a major financial boost to the global economy, as Rupert Way explains.


Profile: Dr Rupert Way


Dr Rupert Way is an Associate at the Institute for New Economic Thinking at the University of Oxford.

He has a background in mathematics and his current research focuses on innovation and technological change, energy system modelling, and understanding strategies for accelerating the sustainable energy transition.

Late last year, I, alongside co-authors Doyne Farmer, Matthew Ives and Penny Mealy, published a paper in the journal Joule, called ‘Empirically grounded technology forecasts and the energy transition’[1]. The paper made quite an impact, because contrary to the usual story we hear about climate change – that decarbonisation is difficult and expensive – we estimate that a rapid transition to a clean energy system is likely to save at least $12 trillion (~£9.7tr) globally, compared to continuing our current levels of fossil fuel use.

In the paper, we present cost forecasts for four ‘key green technologies’: solar energy, wind energy, batteries, and electrolysers. The lesser known of these, electrolysers, are devices that combine electricity and water to produce hydrogen, which can either be used as a fuel directly, or to make other synthetic fuels; if the electricity is clean then the product is known as green fuel. Our forecasts have two distinguishing features. First, they are probabilistic, so rather than forecasting a single cost point, they give an entire range of possible future costs. Second, they are based on the experience curve model of technological change. This is a widely observed pattern, in which costs decline at roughly a constant rate with each doubling of cumulative production: the more we produce a given technology, the better we become at producing it, and the cheaper it becomes. This is also known as the learning curve, or learning-by-doing.

The paper explores the implications of these forecasts for the cost of the energy system as a whole, and for the transition to a sustainable energy system over the coming decades. In short, our clean technology cost forecasts are lower than those used by most major energy models, and as a consequence, our assessment of how much the green energy transition is likely to cost is far lower too.

We forecast that if we continue to roll out clean technologies on their current growth trends for the next decade or so, then decarbonising the global economy by around 2050 is achievable and will generate trillions of dollars of net economic savings, compared to continuing with a fossil fuel-based system. With smart policy choices, going green fast is the cheapest.


Falling costs


This will come as a surprise to many, so some essential context is required. First, the costs of key green technologies have been falling steadily for many decades due to innovation, technological progress, and policy-driven deployment.

Solar photovoltaics (PV) were first used in 1958 to power satellites in orbit, and since then their costs have fallen by over 99.9%. Similarly, wind power and lithium-ion batteries took off commercially around 1980 and 1991, and have since fallen in cost by around 90% and 98%, respectively. In contrast, despite short-term volatility (of which we are all now acutely aware), fossil fuel prices have barely changed in over a century. Adjusted for inflation, oil, coal, and gas now cost around the same as they did a hundred years ago. Different technologies improve at very different rates, and although nobody yet fully understands why, we can still make predictions based on the data we observe.

In the decade from 2010-2020 alone, the costs of PV, wind and electric vehicle (EV) batteries plummeted by 87%, 54% and 88%, making solar and wind the cheapest forms of electricity generation to build in most places on the planet. Electric cars are now cheaper to run than petrol cars, and will soon be cheaper to buy too. Costs may rise and fall temporarily due to supply chain disruptions and other global events, but the long run trend is driven by innovation, so they are likely to continue falling.


Image: Aerial picture of an agrivoltaics solar plant with walls of vertical bifacial modules near Aasen, Donaueschingen Germany. Credit: Tobi Kellner, (CC BY-SA 4.0) via Wikimedia Commons.

Graph: Long-run cost trends for various energy technologies. Adapted from [1]. See [1] for definitions and further details.

Second, the major energy models used to inform the Intergovernmental Panel on Climate Change (IPCC) have unfortunately failed to adequately capture these dramatic cost improvements so far. For the last two decades, they have systematically overestimated future costs of key green technologies, an astonishing fact freely acknowledged throughout the energy modelling community. The models in question include the half-dozen large-scale Integrated Assessment Models, which simultaneously model the global economy, energy system, and greenhouse gas emissions; and the International Energy Agency’s (IEA’s) World Energy Model, which is used to produce its highly influential annual World Energy Outlook (WEO) report. Throughout the 2010s, none of these considered the possibility that key green technology costs might fall as fast as they did in reality; they simply did not model such a scenario. Of the thousands of scenarios produced, none projected that the global average cost of solar electricity in 2020 would be as low as the 50 $/MWh value that actually occurred. Similarly, the electrification of transport was not considered a viable option in most energy models up until very recently, yet EVs are now poised to take over from internal combustion engine vehicles due to superior cost, performance, and environmental characteristics. By overestimating their costs, models made deploying green technologies at scale appear artificially expensive.

Image: Credit: D-Kuru, Public domain, via Wikimedia Commons.

Third, and finally, the cost forecasting method we use has been backtested and validated on data for over fifty different technologies (by co-author Doyne Farmer and colleagues in prior studies). The validation procedure involved collecting as much historical technology data as possible, and using it to make thousands of forecasts from various dates in the past, and then comparing these forecast values with what actually happened. This allowed the accuracy of the forecasting method to be assessed empirically, based on real data, and confirmed that our probabilistic forecasts are consistent with how technology costs have developed historically. In other words, if we had used this method in the past, then the realised cost values that actually occurred would have fallen within the probability distributions as expected – the forecasts would have been reliable. Of course, it’s possible that our future forecasts turn out to be inaccurate, but there’s no reason to think this would happen – indeed, this is the whole point of the elaborate backtesting procedure. To the best of our knowledge, these are the most scientifically justified forecasts we can make at this time.


Potential net gain


With these facts in mind, hopefully the conclusion of our paper is now less surprising: due to the high likelihood of further clean technology cost declines, the energy transition should not be thought of as a net burden on society, but rather a potential net gain. By rapidly switching to cheap, clean energy technologies in most sectors of the economy, we can drive costs down further and most likely achieve savings big enough to outweigh the high upfront investments required to decarbonise a minority of stubborn but critical ‘hard-to-abate’ sectors (aviation, marine transport, steel, cement, fertiliser etc.). The faster we make the transition, the larger the savings. To understand the result in a bit more detail, some background on energy system modelling is required.

Image: Credit: AleSpa, CC BY-SA 3.0, via Wikimedia Commons.


The idea of modelling anything at all decades into the future is daunting, let alone a complex socio-economic system such as the global energy system. Far-sighted models, in astronomy, geology, or climate science, for example, are tightly bound to extremely accurate and rigorously tested physical laws and phenomena. But the future of technology and the economy is far more uncertain. Nobody could have reliably predicted the invention of steam engines, vaccines, computers, or the internet, nor how they would change the world. Technological change depends on invention, innovation, and policy choices, all of which involve randomness and path-dependency. An unexpected innovation here or a clumsy policy there may determine our path for years to come, and are often impossible to predict. Furthermore, near-sighted models, in fields such as engineering, finance, or biology, are often designed with direct experimental validation in mind. But there are no experiments we can do to discover what societies will be like fifty years from now.The one thing we can be fairly sure of is that people will generally do whatever technologies allow them to do to improve their lot, so capturing long-run technological change well, and exploring the landscape of technological possibilities in as scientific a manner as possible is essential.

Image: Credit: Spike, CC BY-SA 4.0, via Wikimedia Commons.

Faced with the incredible uncertainty of multi-decadal time scales, it’s hard to know where to start. A few moments of reflection though, and a simple but far-reaching observation is obvious: all things being equal, if a technology is cheaper than its competitors then we should expect everyone to use it. So arguably the most important question in modelling the future of the energy system is: how much are different energy technologies likely to cost, and with what probabilities?


Empirically grounded answers


Our paper is designed to provide some empirically grounded answers to these questions. The journey towards this point began over a decade earlier. In 2010, the IEA’s flagship WEO report was projecting that solar in 2020 would cost 260 $/MWh (it ended up costing only 50 $/MWh). Around this time, complexity scientist Doyne Farmer, who would later become my boss and co-author, started working on the problem of technology forecasting, based on his broad knowledge of technologies, forecasting techniques, and the academic literature on technological change. The research question was: how predictable is technological change? Or, to put it another way, what is the best forecast we can make, and how wrong is it likely to be? (Incidentally, Farmer and co-author Makhijani predicted in 2010 that by 2020 solar and wind would both be cheaper than coal and nuclear power. This turned out to be correct, and no other forecasts from that time even came close.)

Through a series of papers published between 2011 and 2019, Farmer, with many co-authors, developed a foundation for scientifically making forecasts. This involved collecting historical data, developing different methods of forecasting technology costs, and testing their comparative performance. The work built on the large existing literature on technological change, in particular the learning curve concept introduced by Theodore Wright in 1936 and extended by Kenneth Arrow in 1962, and Moore’s law, introduced by Gordon Moore in 1965.

Throughout the 2010s, as this work progressed, the costs of key green technologies kept falling, and year after year all the major energy models kept revising down their future cost projections, and increasing their deployment projections (because lower costs mean higher deployment). How could this keep happening? The reason appears to be a group of assumed limits to technological progress buried deep within the models, the values of which turned out to be quite inappropriate (we go into further detail in our paper). In socio-technical system models, it is often our a priori assumptions about what is feasible that determines the results, and that seems to have been what happened here. The consequences of this repeated error are significant. Had modellers and global climate policymakers taken seriously the possibility that there was a viable route to solar, wind, EVs and heat pumps becoming the cheapest options by the early 2020s, this may well have catalysed faster progress in emissions reductions. Investors would have piled into the sector and supporting policies would have been set, just as we are seeing now, ten years and many gigatonnes of CO2 emissions later.




The upshot of Farmer and team’s work is that we are now able to make technology cost forecasts that are consistent with how technologies have improved in the past, and quantify the level of uncertainty history tells us we should have about future costs. The emphasis on probabilities, and the degree to which we expect our forecasts to be wrong, is essential, yet has so far not featured prominently in most energy models. Our forecasts are based on both long-run technology improvement trends (the experience curve), and historical cost volatility (the extent to which costs have bounced up and down due to unpredictable real-world events), and these are unique to each technology.

With this new forecasting method in hand, the next step was to embed it within an energy system model. This required designing a new energy system model around the limitations inherent to the forecasting method due to its dependence on data (we can only model technologies for which enough historical data exists). The resulting model has relatively low resolution, making it good for modelling large, sweeping changes, while not spending too much time on fine details of the system. It focuses on exploring the space of technology scenarios by extending or curbing existing technology growth trends, in different combinations, and then uses the probabilistic experience curve method to estimate total costs. With this simple framework we are able to model both a continuation of the existing system, and a rapid transition to a clean energy system based on high electrification and green fuels for hard-to-decarbonise sectors; and anything in between. The model structure meets our initial goals of being transparent, comprehensible, closely tied to data, easily updatable with the latest data, and free from unnecessary modelling constraints that may cause a systematic underestimation of technological progress.


Image: Credit: Yottanesia, CC BY-SA 4.0, via Wikimedia Commons.

Distinct energy system scenarios


In the published paper we use our model to analyse three very distinct energy system scenarios: a Fast Transition, a Slow Transition, and a No Transition. We forecast the cost of individual technologies in each scenario, and sum these to give total energy system costs. We then calculate the present discounted value of each scenario (discounting is a way of accounting for the fact that we generally don’t value the future as highly as we value the present, often for good reason). The cost reductions expected in key green technologies due to rapid, early deployment, make the Fast Transition scenario the cheapest, by around $5-12tr in present discounted terms, depending on the discount rate. This means that, at the level of the global economy, it is worth pursuing this scenario immediately, as it has higher value now than the other scenarios. We should be moving quickly right now to capture the benefits of a rapid energy transition. Other models have underestimated long-run technological improvement in a few key technologies, but once we take this into account, the benefits of moving fast to tackle climate change are overwhelming. If one counts the avoided economic damages that would be caused by continuing to use fossil fuels for decades to come – floods, drought, storms, heatwaves etc. – then the benefits of a fast transition are in the many tens to hundreds of trillions of dollars in savings.

In the eight months since our paper was published, many other reports have recommended faster action on climate change on the basis of clean technologies’ now low costs. The IEA’s Renewables forecast to 2027, published in December 2022, showed that the world appears to be on a pathway similar to our modelled Fast Transition scenario. Governments and investors are coming round to the huge opportunities of energy transition, and the easy ways to save energy and money with solar, wind, batteries, heat pumps, EVs, and many other smart solutions. Because the economic benefits are now within sight, the policy environment has improved considerably, with policymakers all over the world scrambling to find ways to reduce energy costs and emissions simultaneously. The Inflation Reduction Act in the US, and many other policies worldwide, have kickstarted investment in clean energy technologies, and spending is flooding into the sector. Increased investment will lead to faster technology improvements and lower costs.


A long way to go


The reality is that much of the energy transition is simply unforecastable. The best we can do is use the most scientifically supported forecasting methods we can, lean on the data as much as possible, and try to faithfully capture the most important trends that are likely to shape the system in the long-run. There is a very long way to go, but thankfully the data is moving much faster than most of the models predicted. With a major effort to keep ramping up clean technologies on their current growth trends for another decade, it looks as though we really can make a huge dent in global emissions, and finally get on a path to a cleaner, greener, cheaper, net-zero world by 2050.

Image: Rampion Wind Farm, UK. Credit: Wpwh81, (CC BY-SA 4.0), via Wikimedia Commons.

References [1] Way, R., Ives, M.C., Mealy, P., and Farmer, J.D. (2022). Empirically grounded technology forecasts and the energy transition. Joule 6, 2057–2082.

Dr Rupert Way


Institute for New Economic Thinking at the Oxford Martin School

University of Oxford