The Pitfalls of Climate Models
By Andrew Garber
Economists overwhelmingly support pricing greenhouse gas emissions. The reasoning is simple: for the invisible hand to guide society to the socially optimal level of greenhouse gas emissions, consumers must take into account the social cost of carbon, i.e., the cost that greenhouse gas emissions impose on others. But how do we determine this cost? Integrated Assessment Models (IAMs) attempt to answer this question. They usually translate changes in the concentration of greenhouse gases in the atmosphere to temperature increases and then to economic damage. From there, one can calculate the social cost of carbon. Unfortunately, IAMs have failed to arrive at a conclusive social cost of carbon; indeed, they disagree on this cost by an order of magnitude. The wide range of social costs of carbon points to a deep problem: IAMs are unreliable because they rely on uncertain empirical and normative premises, only some of which are resolvable by more empirical work.
First, it is unclear how much economic damage greenhouse gas emissions are likely to cause. This uncertainty in part comes from our ignorance over how sensitive the average global temperature is to increases in the concentration of greenhouse gases in the atmosphere. Climate scientists know that greenhouse gases increase the average global temperature, but the mechanisms by which they do so are complex: climate feedback loops and tipping points guarantee that there is no simple relationship between emissions and temperature increases. For example, since water absorbs more sunlight (and thus heat) than ice, as the oceans warm and ice melts, the oceans warm even faster. Given this positive feedback loop, once the ice sheets shrink enough, they might reach a point of no return at which they would inevitably disappear. Nevertheless, climate scientists have greatly narrowed the range of likely temperature increases from greenhouse gas emissions, and continue to do so.
The harder problem is determining the economic damage from warming; in other words, what is the “loss function” that accurately maps temperature increases to monetary costs? The loss function is usually defined as the ratio of GDP in a world with climate change to the GDP in a world without climate change, both as functions of time. The models agree that the loss function is decreasing (i.e., warming hurts growth) and approaches zero asymptotically. Unfortunately, that is where the agreement ends. Due to a lack of empirical evidence, there is little information that can determine which function is most accurate, so economists have arbitrarily chosen functions to be the loss function. Even if these functions were to fit the data from temperature rises so far, the difference in functions would lead to different predictions of economic damage when extrapolated. As a result, different damage functions lead to different recommended policies to tackle climate change. To make matters worse, we do not know which functions fit the data so far. Economists lack a comprehensive picture of how much damage global warming has caused so far, since the damage ranges across many industries and is thus difficult to measure.
Second, most IAMs fail to account for the risk of catastrophic outcomes, instead focusing on scientists’ best guess of what will happen. Catastrophic outcomes, in which there is a massive drop in output, are unlikely but not impossible. Temperature increases by greater than ten degrees celsius are possible and would likely cause an environmental catastrophe almost unrivalled in Earth’s history. Even warming well within ten degrees celsius could cause catastrophes, such as conflict over mass migration and devastating damage to coastal communities through sea level rises and extreme storms. The improbability of catastrophes does not justify ignoring them. Because the magnitude of climate catastrophes would be staggering, even small probabilities of their occurrence warrant serious concern. Uncertainty about whether a catastrophe will occur should only increase the importance of preventing its occurrence, given risk aversion.
Third, IAMs rely on a crucial normative parameter: the rate at which we ought to discount future generations’ welfare. The worst effects of global warming are decades down the line, so future generations are likely to bear most of the burden of global warming. The choice of how we should weigh not-yet-existent humans’ welfare against our own is thus crucial. Unlike the above problems, the problem of determining the proper discount rate cannot be solved by careful empirical analysis alone. The problem departs from the realm of climate science and economics to that of philosophy. One answer is that we should give future people’s welfare the same weight as the welfare of people living now, justifying a discount rate around zero. For example, we would fault our distant ancestors if they had recklessly taken actions that they knew would shorten our lives, even though a positive discount rate would mean that from their point of view, our welfare in the distant future would be trivial compared to theirs. But as is all but guaranteed in moral philosophy, it is easy to dispute this account. For example, just as we feel a stronger sense of obligation to our friends and families, perhaps we should also have stronger obligations to, and thus place more weight on, our current generation. Or perhaps we should reject utilitarianism and the aggregation of people’s wellbeing (although people holding this view probably have gripes with the very idea of deciding climate policy on a calculation of social cost).
Is there a way to avoid resorting to philosophical speculation to decide the discount rate? One way is to look at capital markets to deduce the discount rate. More specifically, one can infer the discount rate by checking the rate of returns on capital, the growth rate of consumption, and the elasticity of marginal utility. Although the latter parameter is uncertain, deriving the discount rate from observable parameters eliminates some of the uncertainty that moral philosophy introduces. Nevertheless, this certainty is an illusion. It is an old observation that one cannot derive what ought to be from what is, and that principle applies here: it does not make sense to say that the capital markets accurately reflect what is right. Doing so smuggles in a normative premise that brings us again to the realm of philosophy.
IAMs are far from perfect, but they still have their place. That our best guesses of the social cost of carbon are -- and might inevitably be -- highly uncertain is no reason to stop guessing. It is still important to price greenhouse gas emissions so that they more accurately reflect the social cost of carbon. What the arguments above show is that we might need more. IAMs and their calculations of social cost are important policy tools, but they might be insufficient to tackle the challenge of global warming. In particular, it might be appropriate for countries to mitigate the risk of climate catastrophe in the same way as they mitigate the risk of another catastrophe: nuclear war. Scientists and economists could outline the catastrophes with a non-negligible chance of occurring and governments should act to mitigate these risks, just as they did with nuclear proliferation. Under this framing, it might make sense for the government to assume a more prominent role in the private sphere to prevent catastrophe; mandates, not merely Pigouvian taxes, may be necessary. In this way, governments can reduce and adapt to the likely damages of global warming while preventing the worst outcomes from occurring. Given the many facets of the climate challenge, we will need every tool we have to overcome it.