The risk of being misled by climate economy models


This post deconstructs the following statement:

If reducing greenhouse emissions had economic benefits then we would do it anyway without new policy.

The statement above is used by economists to argue against the introduction of policies to reduce greenhouse gas emissions on the basis that the costs would outweigh the benefits of reducing climate change. It is part of a wider narrative that regulatory policy can only lead to economic costs. However, the statement is perhaps one of the most perverse conclusions from neoclassical economics. It depends on a raft of assumptions that run contrary to real-world experience. Further, as discussed below, if just one assumption is taken out, the conclusion changes.

Sadly, economists and (in particular) economic modellers, have played a key role in turning this fallacy into accepted reality. They have done this by using simple optimisation-based approaches that make strong assumptions about human behaviour. Often the modellers do not critically question or even fully understand these assumptions. 

It should be noted that there is nothing wrong with optimisation per se. It is useful to know how to minimise costs in a system. However, once assumptions of optimisation extend into human behaviour, it is easy to draw false conclusions because the models portray an idealised version of a complex reality.  As I mentioned in Optimisation and simulation models – how, and when, to use them, “when an optimisation tool is used to estimate the effects of a policy. All the assumptions that were previously used to solve an optimisation problem suddenly become predictions of real-world behaviour”. Unfortunately, most climate-economy models operate in this way. 

William Nordhaus, using the Dynamic Integrated Climate Economy model (DICE), was a front runner in climate-economy modelling and also an early adopter of the optimisation-based approach. His work earned him the Nobel Memorial Prize in Economic Sciences in 2018. Since then, a new generation of ‘Integrated Assessment Models’ have been developed; with more detail about different technologies, but still based on the same neoclassical assumptions and with the same potential to produce misleading results. 

The figure below provides an example of some results. It is taken from the Intergovernmental Panel on Climate Change’s (IPCC’s) Fifth Assessment Report (AR5) and provides estimates of the impacts of reducing emissions to meet a 2°C climate target. Note that the axis is expressed in costs, so virtually all the models show negative impacts.

Screenshot 2020-02-19 at 20.44.50

Source: Clarke et al (2014)

However, the way the models get these results can be derived easily from the underlying assumptions. The baseline case is obtained through the optimisation process. Adding policy to meet the temperature target turns it into a constrained optimisation. As the statement at the top of this post suggests, with new policy, things can only get worse. The results are therefore little more than a restatement of the assumptions. Here we break down those assumptions.

First, the models assume that individuals are ‘rational’ and behave in an optimal manner. This ignores findings by behavioral economists (and others) that indicate the contrary. Secondly, the models assume that markets function perfectly, ignoring the role of institutions in shaping the functioning of markets.

Second, the models also assume ‘perfect’ knowledge. As Keynes noted, people can only be sure to select the best option if they know all the possible options. Perfect knowledge is therefore required to make it work. In some models, the assumption of perfect knowledge is extended to knowledge of the future (i.e. perfect foresight). But in the context of the modern economy, it is not realistic for agents to know all the available products out there (let alone what might be there in the future).

Finally, we also need to think carefully about technology. The low-carbon transition is primarily one of technology development and adoption, but optimisation-based models have little to say about either. Most notably, the paths of gradual diffusion that we get from innovation theory are not compatible with a system where everyone instantly switches to the cheapest (i.e. optimal) technology.

There are further assumptions, including the use of representative agents and an unrealistic treatment of finance (see “The role of money and the financial sector in energy-economy models used for assessing climate and energy policy”. None of these assumptions are realistic – and, as shown below, they are not necessary either. Unfortunately, for optimisation modellers though, changing the assumptions would mean a fundamental rethink of how their models work.

Alternative approaches

In the meantime, there are some alternative modelling approaches available. For example, Cambridge Econometrics’ E3ME model is based on post-Keynesian foundations. It allows for conditions of uncertainty and does not assume fully rational behaviour. The model’s results show that rates of technology uptake and debt dynamics are important in determining policy impacts. Policy that impacts on technology will affect both the composition and rate of economic growth, while additional investment will create short-term stimulus effects and higher debt levels that may reduce long-run growth.

Contrary to the results published in the IPCC report, E3ME shows that some climate policies can lead to positive impacts on GDP and employment levels (see “Unlocking the inclusive growth story of the 21st century” and “Renewable energy benefits: measuring the economics”). Other policies lead to costs; the important point, however, is that the direction of results is not pre-determined by unrealistic assumptions.

E3ME is not the only model to replace assumptions about perfect knowledge and optimal behaviour with a more empirical approach (see e.g. DEFINE (Dynamic Ecosystem-FINance-Economy) model and Energie und Klima Modelle). But these models are very much in a minority against the ‘consensus’ that is presented elsewhere, meaning that it can be hard for results to gain traction in the wider scientific community. More models are needed to fill this wide gap in scientific knowledge.

The way forward

Using their optimisation models, economists have been largely responsible for creating the illusion that addressing climate change is necessarily an economic cost. As more climate policy is introduced, they will again be called upon for policy assessment and evaluation. They have a chance to make up for previous mistakes but, with the next IPCC assessment report (AR6) due soon, it is not clear that this chance will be taken.

It is therefore up to all of us to question the interpretation of these models’ results. And the next time someone says that there must be costs, point to the example of solar power. Solar is now the cheapest source of electricity in many parts of the world, benefitting rich and poor regions alike (regardless of their current climate policies). However, solar costs are only so low because of previous government interventions in Germany and other European countries. In summary, these interventions have led to tangible economic benefits.

In the real world there are many more examples. In models often not. But we must be clear that the statement at the top of this post refers to the results from the models and not the real world.

Hector Pollitt is a Director and the Head of Modelling at Cambridge Econometrics@HectorPollitt.

6 thoughts on “The risk of being misled by climate economy models

  1. this is not my field, but as someone with an unremarkable mainstream economics education, I find this post pretty hard to understand. I have seen arguments that the Nordhaus approach does not do a good job of capturing the costs of fighting climate change because, for example, it neglects the possibility of substituting green techs for brown without loss of productivity, and i don’t want to defend those models because I don’t know enough them, but I don’t think the problems originate from things like optimisation, assuming rationality or ignoring uncertainty.

    As you no doubt know, one of the most fundamental ideas in context of climate change is negative externalities. So when you write:

    “If reducing greenhouse emissions had economic benefits then we would do it anyway without new policy”

    Do you mean “economic benefits” as in benefits to society as a whole, or benefits in the sense of private financial returns? If reducing greenhouse emissions is the most profitable thing to do from a private returns perspective, then yes maybe people would “do it anyway”. Otherwise no, firms will tend not do things that benefit society but which lose them money, that’s the whole point of externalities. So no (sensible) mainstream economist is going to argue against the introduction of policies to reduce greenhouse gas emissions because the costs must outweigh the benefits otherwise we’d be doing it anyway. That’s sheer nonsense and i am quite sure Nordaus would make no arguments like that. .

    That means that regulations etc. are not necessarily costly from society’s point of view – it depends if they are counteracting market failures such as externalities. So if your model is telling you regulation etc. can only be bad, it’s not because of optimising, rationality etc. it’s because you don’t have externalities etc. in the model. There are lots of lots of mainstream models with optimization and all the assumptions you blame, that recommend interventions. Here’s one:

    The example you give of German subsidies getting the solar market moving is pretty much a text-book case of why government intervention might be needed because private actors won’t do it by themselves – see e.g. box 1-1 here:

    Click to access 130066-REPLACEMENT-PUBLIC-WBG-Strategic-Use-of-Climate-Finance-Sept2018.pdf

    if you want to argue that investments to go green will have benefits because innovation is demand-led and the economy is operating below capacity so could use some stimulus, well fine your model needs to incorporate that idea but you can do that with rational optimising agents with perfect knowledge.

    So whilst mainstream climate models may or may not be rubbish, it looks to me like you have misdiagnosed the cause.


  2. Hi – many thanks for the thoughtful comments. Let me try to clarify a bit.

    I am talking in macro terms, so do mean economic benefits to society and not individual firms. I fully agree that one of the reasons we have not seen stronger climate policy is the lobbying of the firms that will lose out.

    This comes to the point on externalities and the logical mainstream policy response of a carbon tax. Similarly on innovation, where there are positive externalities, the logical response is subsidies. If I have interpreted correctly, you are saying that if you account for these externalities in the optimisation function then you could still get better outcomes in a low-carbon case – and it is reporting GDP rather than a more aggregate welfare function that is the issue. So the issue is not the models per se, but how they are applied/reported.

    I am arguing that GDP in its narrow definition can increase in a scenario with lower emissions, even ignoring any feedbacks from climate (see next paragraph). This seems to be the main point we disagree on.

    I view the main issue as fundamental uncertainty, which I think is even more relevant to this topic than most others – e.g. we don’t know what the climate damages will be and therefore how to set a carbon tax rate. On a more basic level, we see many examples of firms that could improve energy efficiency to their own advantage but don’t, e.g. either because they are not aware of the options or simply are focused on other concerns. In neoclassical terms you could call this another market failure (knowledge gaps), which active policy could correct. Similarly, we see that it can take some time for new/better technologies to be adopted because firms don’t want to take risks on new equipment.

    Regarding spare capacity, I can only see two ways you would get it under the conditions you set – either you put a value on not using something (as in theories of ‘voluntary’ unemployment) or relax the assumption of perfectly flexible markets with prices adjusting to equilibrium – otherwise all resources will be used. Again you could regard markets not adjusting as a market failure and put in policy to correct. In contrast, Keynes saw it more as a natural outcome even in perfect markets, again related to uncertainty and firms wanting to hold cash in case of emergency.

    To summarise, I think we are coming at the same issue from two different directions. You are suggesting starting from the standard theory and then making allowances for where the assumptions don’t hold, i.e. the market failures. I am suggesting that the market failures are too numerous and we start from a position where we don’t use these assumptions – in part because it would be difficult to identify all the places that we would need to step in (likely the case in the results in the chart).

    The European Commission reflects this in its modelling of energy policy – two models are used, one is a modified CGE model that is based on rational behaviour but constrainted to reflect many real-world constraints, the other is the Cambridge Econometrics post-Keyesian model referenced in the post. Happy to send some links comparing the results from the models if you are interested.

    Hope that helps, and thanks again.


  3. Thanks Hector

    Well if you mean benefits to society, anyone making this argument “If reducing greenhouse emissions had economic benefits then we would do it anyway without new policy” is obviously nonsense because the whole point of externalities, public goods is that beneficial things won’t happen without intervention.

    I don’t really understand this ” if you account for these externalities in the optimisation function then you could still get better outcomes in a low-carbon case”

    in a model with externalities, optimising by self-interested private agents won’t give you the best outcomes, but optimising by a social planner with tools such as taxes and regulations at their disposal.can.If the “low carbon case” is the best outcome that will not be obtained by laissez-faire, then taxes and regulations etc. could achieve it.

    “I am arguing that GDP in its narrow definition can increase in a scenario with lower emissions, even ignoring any feedbacks from climate (see next paragraph). This seems to be the main point we disagree on.”

    I did not write anything about whether GDP can increase with lower emissions. Whether it can or not will depend on technology and productivity and efficiency of resource allocation and utilization.

    I agree that if the truth is that there are lower emissions technologies that firms could profitably adopt but which they don’t know about, your model will be over optimistic about them being adopted without some policy intervention. I didn’t think the problem with Nordaus at al. was being over optimistic about firms adopting green tech, whereas in reality they won’t because they don’t know about it.

    yes you need to do something that deviates from perfect market assumptions to end up with a model that has persistent capacity underutilization. If the models you don’t like don’t incorporate market failures that are both present in reality and are relevant in this context, then there’s your problem – optimisation, rationality etc. are not the culprits.

    I don’t think I am really starting from any position, I am just pointing out that mainstream economics does not support the proposition that if something is economically beneficial, we’d be doing it already, or that regulations etc. are necessarily costly


  4. OK, thanks – agreed on most points. I would argue that these features are closely interconnected, however. For example, it seems inconsistent to suggest that agents have perfect knowledge on everything apart from new technologies. Having said that, I’m fully aware that this is what some people are doing…

    Turning back to the modelling, there is a practical and even more intimate link that is imposed by what the optimisation algorithms can solve (to be clear, these are optimising the system, not for individuals) – which generally means things have to be (log) linear. As far as I’m aware, this also means you need to assume implicitly a social planner to manage the system. It can handle internalising externalities through a basic pricing mechanisms but is limited on possibilities for more realistic treatments of technology and market structures.

    This is fine if the question you are asking is how to design a system at least cost, given the assumptions and constraints. But less appropriate for scenario analysis, where the assumptions become proxy for real-world behaviour. Incidentally, I’m speculating a bit now, but to model the structure you are suggesting (individual optimisation with system-level constraints) I think you would need to resort to an agent-based framework. I came across a non-climate example recently, will try to dig out.


  5. Indeed – in these models, the estimates of the damages from climate change are big unkowns as well, but are pretty critical in determining results.


Leave a Reply to Patrick Carter Cancel reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s