Are the models underpinning global efforts to tackle climate change fit for purpose? Alex Teytelboym and Penny Mealy of the Institute for New Economic Thinking at the Oxford Martin School look at why a different approach should be tried.

The economics of climate change - “the greatest market failure that the world has ever seen” – is one of the most difficult intellectual challenges facing economists today. As Nicholas Stern observed, climate economics must “be global, deal with long time horizons, have the economics of risk and uncertainty at center stage, and examine the possibility of major, non-marginal change”. No wonder then climate economics is so far behind climate science: never before has the academic community been called upon to make such long-term predictions about so complex a system, with such patchy data, using insights and models spanning the vast breadth of physical and social sciences.

Unfortunately, current analytical techniques for assisting climate change decision-making are not particularly helpful. The state-of-the-art tool for policy assessments used by governments and the Intergovernmental Panel on Climate Change (IPCC) is called an Integrated Assessment Model (IAM). These models aim to capture interactions between economic, energy and climate systems and forecast the impacts of climate change on the world economy. However, despite a quarter of a century of concerted efforts by first rate economists and scientists to scrutinize, revise and improve these models, the current suite of IAMs have come under severe criticism, with MIT economist Robert Pindyck even describing them as being “close to useless”.

What’s wrong with IAMs?

At their core, IAMs use a traditional micro-founded macroeconomic model in which agents (firms and consumers) make decisions about how much to consume, save, invest and innovate. Within the model, agents’ choices and consequent economic activities create emissions, which accumulate in the atmosphere. The model’s physical science component calculates how these emissions increase the global temperatures and alter precipitation patterns. If agents choose to invest in new low-carbon technologies, they generally become cheaper and more widely adopted by current and future generations. These models are often used by policy makers to analyse how government policies, such as carbon taxation, efficiency standards or research subsides are likely to affect agents’ decisions and reduce emissions to a sustainable level.

The IAMs receiving the most criticism are those that include a damage function. The damage function aims to estimate how much global output will be lost due to climate change. In theory, a model that encompasses climate damage risks is particularly advantageous, as it provides a fuller account of the estimated costs and benefits of any climate policy as opposed to only the economic costs of reaching a particular climate target. However, in practice, damage functions are extremely poorly estimated, making these models, in Stern’s words, “grossly misleading”. As data on economic damages from climate change is non-existent at worst and patchy at best, decision making that fails to account for vast uncertainties about damages could result in poor policy. Although the IPCC stays clear of these kinds of IAMs, they continue to influence government policy decisions around the world.

Within the academic community, some economists are aiming to better deal with climate damage variance and uncertainty by constructing sophisticated dynamic stochastic general equilibrium models of the sort used by central banks to develop macroeconomic forecasts. These elegant models have been adapted in clever and nuanced ways to suit the climate economics context. But depending on how society values its future and how uncertain damages are, these models suggest that optimal carbon prices (the price that equates the economic benefits of emitting an extra tonne with its costs) could be anywhere between $25 and $4200 per tonne!

Unfortunately, damage functions are not the only thing wrong with present day IAMs. In a recent paperpublished in Environmental and Resource Economics with Cameron Hepburn and Doyne Farmer, we argue that many characteristics of modern day equilibrium-based macroeconomic models make them unsuitable for appropriately capturing a myriad of factors important in climate change decision-making.

One factor is inequality – many highly aggregated macroeconomic models struggle to capture distributional consequences of climate change. Although climate change is likely to hit the poorest of this world hardest, much analysis is performed at a country or regional and ignores effects that exacerbate income and consumption inequality.

Another factor is technological change. Despite being one of the critical factors influencing the costs, benefits and pace of transitioning to a low carbon economy, macroeconomic modeling of technological innovation, learning and adoption is still quite primitive. Present techniques based on worldwide learning curves impose unrealistic assumptions about perfect diffusion and technological spillovers. Further, even the most sophisticated models of endogenous technological change (in which agents can change innovation rates by choosing to invest more in research and development activities) fail to appropriately account for path-dependency and lock-in effects in infrastructure networks, or the influence of innovation on market structure evolution (such as a shift towards a decentralized rather than a centralized energy sector).

A possible way forward

In our paper, we suggest that an alternative type of modeling, known as agent-based modeling might be helpful. Agent based models (ABM) are computer simulations of heterogeneous agents that interact according to specified rules. Even fairly simple rules can mimic the evolution of complex socio-economic systems. Agent-based models have been successfully applied in a range of contexts to model a variety of different phenomena such as conflict dynamics, pandemics, urban planning, electricity markets, emissions trading markets, housing markets, financial markets, technology adoption, labour networks, business cycles, the evolution of market structures, and even large-scale macro-economies.

Could we use ABMs for modeling climate economics?

Since ABMs are quite flexible computer simulations, they can more easily capture non-linear dynamics and feedbacks involved in complex innovation processes. They are also well equipped to simulate agents with different characteristics (such as incomes, preferences and behavioral tendencies). This allows them to better understand distributional and inequality issues, and behavioural aspects associated with technology adoption. Finally, ABMs can tolerate the presence of large shocks and uncertainty, potentially giving us a more realistic and prudent appreciation of the adverse consequences of global warming.

Climate change is arguably the most important issue facing decision makers today. Yet we are probably hitting a dead end with existing IAMs. Pindyck now argues that we should do away with models and instead base climate policy on expert opinions. While this approach is transparent and intellectually honest, expert opinions might in turn be based on equally flawed mental models. We instead believe there is still a chance to develop useful, scientifically rigorous and well-calibrated climate decision making tools – and agent-based models could hold the key. Of course, many economists fear paradigm shifts and generally prefer sticking to the models they know. But for the economics of climate change, there might be just enough time to try a different approach.