IPCC, IAMs, and the Politics of 'Optimal' Carbon Paths

The Models Behind the Policy

When world leaders gather at climate summits — COP meetings, G7 sessions, UN General Assembly side events — the targets they debate, the pledges they make, and the timelines they set are shaped by a class of economic models that most of the public has never heard of: Integrated Assessment Models, or IAMs. These models combine simplified representations of the global economy with simplified representations of the climate system to project how different emissions pathways will affect both economic output and global temperatures over decades and centuries. They are the quantitative backbone of the IPCC’s Working Group III reports (which focus on mitigation), and they inform the cost-benefit analyses that governments use to evaluate climate policy.

IAMs are, in a very real sense, the most consequential economic models in the world. The question of how much to spend on climate mitigation, how fast to reduce emissions, and how to distribute the costs across countries and generations depends critically on what these models say. And what they say depends, in turn, on a set of assumptions — about discounting, damage functions, technological change, and risk — that are far more contested, far more value-laden, and far more political than their technical appearance suggests.

Understanding IAMs is not optional for anyone who wants to engage seriously with climate policy. And understanding their limitations is essential for anyone who wants to engage honestly.

What IAMs Do

At their core, IAMs link three components:

An economic module that projects GDP growth, energy use, emissions, and the costs of emissions reduction over time. Most IAMs use some variant of neoclassical growth theory: output is produced by capital, labor, and energy; technological progress increases efficiency over time; emissions reduction is achieved through a combination of energy efficiency improvements, fuel switching, and carbon capture, all of which have costs.

A climate module that translates cumulative emissions into atmospheric concentrations of greenhouse gases, and concentrations into temperature changes, sea level rise, and other physical impacts. The climate modules are simplified versions of the general circulation models used by climate scientists — they capture the broad dynamics (more CO2 means higher temperatures) without the full spatial and temporal resolution.

A damage function that translates temperature changes into economic losses. This is the most critical and most contested component. The damage function specifies how much GDP is lost for each degree of warming — and because IAMs typically run for centuries, small differences in the damage function produce enormous differences in projected costs and, therefore, in policy recommendations.

The modeler sets a policy scenario (a carbon tax, a cap-and-trade system, a technology mandate), runs the model forward in time, and reads off the results: the emissions pathway, the temperature trajectory, the economic costs and benefits. By comparing different scenarios, the model identifies the “optimal” policy — the one that maximizes the present discounted value of economic output net of climate damages and mitigation costs.

The Nordhaus DICE Model

The most influential single IAM is William Nordhaus’s DICE model (Dynamic Integrated Climate-Economy), first published in 1992 and updated regularly since. Nordhaus, a Yale economist who received the Nobel Prize in 2018, built DICE as a tool for cost-benefit analysis of climate policy. The model is transparent, publicly available, and has been the reference point for climate economics for three decades.

DICE’s core structure is straightforward. It models the world as a single economy that produces output using capital and labor, generates CO2 emissions as a byproduct, and can reduce emissions at a cost (the “abatement cost function”). Warming reduces output through a damage function that is quadratic in temperature: damages increase as the square of the temperature change. The model optimizes over time, choosing the path of carbon taxation that maximizes the discounted sum of per capita consumption.

The results of DICE have been remarkably consistent across versions: the optimal policy is a moderate, gradually increasing carbon tax that limits warming to somewhere around 3 to 3.5 degrees Celsius by 2100 — well above the 1.5 or 2 degree targets endorsed by the Paris Agreement. Nordhaus’s model suggests that aggressive near-term mitigation is too expensive relative to the damages it avoids, and that a more gradual approach — starting with a low carbon tax and ratcheting it up over time as the economy grows and damages mount — is economically rational.

This conclusion has been enormously influential and enormously controversial. It has been used to argue against ambitious near-term emissions reduction and in favor of incremental, market-based approaches. And it depends critically on three assumptions that are each deeply contested.

The Discount Rate War

The most famous debate in climate economics concerns the discount rate: the rate at which future costs and benefits are converted into present values. A higher discount rate makes future damages less important relative to present costs, favoring gradual action. A lower discount rate makes future damages loom larger, favoring aggressive action now.

Nordhaus uses a market-based discount rate — roughly 4-5 percent per year, calibrated to observed returns on capital. His reasoning is descriptive: the discount rate should reflect how people actually trade off present and future consumption, as revealed by market interest rates. At this rate, damages occurring a century from now are heavily discounted. A trillion dollars of climate damage in 2126 is worth only about $20 billion today.

In 2006, the British economist Nicholas Stern, commissioned by the UK Treasury, published the Stern Review on the Economics of Climate Change, which used a much lower discount rate — about 1.4 percent. Stern’s reasoning was prescriptive (or ethical): the discount rate should reflect the moral weight we give to future generations’ well-being, not the market’s revealed preference for present consumption. There is no ethical reason, Stern argued, to count a future person’s well-being as less important than a present person’s (beyond a small probability that humanity goes extinct). At Stern’s lower discount rate, climate damages loom much larger, and the case for aggressive near-term mitigation becomes overwhelming. The Stern Review called for immediate, large-scale investment in emissions reduction — spending 1-2 percent of global GDP per year to avoid future damages of 5-20 percent of GDP.

The Nordhaus-Stern debate is not a technical disagreement that can be resolved by better data. It is a philosophical disagreement about intergenerational ethics. Should we discount the welfare of future generations because we can earn a return on capital today (Nordhaus)? Or should we treat all generations’ welfare equally, discounting only for the pure uncertainty that the future world will exist (Stern)? Economists have lined up on both sides, and the debate has generated a vast literature. But the core question is not an economic one — it is a moral one, and the pretense that it can be settled by the model rather than by ethical judgment is itself a form of politics.

The practical stakes are immense. At Nordhaus’s discount rate, the “optimal” warming is 3+ degrees Celsius. At Stern’s discount rate, the optimal warming is close to the Paris target of 2 degrees or below. The difference between these two outcomes — one degree of additional warming — represents an entirely different future for human civilization, and the models give you either answer depending on which discount rate you plug in. This is not a model providing objective guidance; it is a model amplifying the modeler’s priors.

The Damage Function Debate

If the discount rate determines how much we care about the future, the damage function determines how bad we think the future will be. And the damage functions used in most IAMs are, to put it diplomatically, poorly grounded.

The standard DICE damage function is a smooth, quadratic curve: damages equal a coefficient times the square of the temperature increase. At 2 degrees of warming, DICE projects damages of roughly 2 percent of global GDP. At 4 degrees, about 8 percent. These numbers sound large in absolute terms but are small relative to projected GDP growth: even under significant warming, the model shows the world getting richer over time, just somewhat less rich than it would be without warming.

There are several problems with this.

Empirical basis. The damage function is not derived from a detailed bottom-up assessment of climate impacts. It is a reduced-form estimate, calibrated to a handful of studies and expert surveys, that purports to summarize the total economic impact of warming in a single number. The studies underlying it are, by necessity, extrapolations from observed relationships between temperature and economic output — relationships that may break down at warming levels outside the range of historical experience. We have no empirical observations of what happens to the global economy at 4 degrees of warming, because it has never happened in the history of human civilization.

Missing damages. Most IAMs omit or undervalue a long list of climate impacts that are difficult to quantify but potentially enormous: biodiversity loss, ecosystem collapse, water scarcity, forced migration, conflict, the psychological toll of living in a degraded environment, and the irreversible loss of cultural and natural heritage. These “non-market damages” do not appear in GDP — the metric that IAMs optimize — and are therefore invisible in the cost-benefit calculus.

Distributional blindness. The standard damage function expresses damages as a percentage of global GDP, ignoring how those damages are distributed. A 4 percent reduction in global GDP that falls entirely on the world’s poorest countries is very different from a 4 percent reduction spread proportionally. Climate damages are in fact expected to be highly regressive — falling most heavily on tropical, low-income countries that have contributed least to emissions. IAMs that aggregate globally systematically understate the human impact of climate change.

Tipping points and tail risks. The smooth, quadratic damage function implies that damages increase gradually and predictably with temperature. But climate science identifies a number of potential tipping points — thresholds beyond which changes become self-reinforcing and irreversible: the collapse of the West Antarctic ice sheet, the dieback of the Amazon rainforest, the disruption of Atlantic thermohaline circulation, the release of methane from Arctic permafrost. These tipping points are not captured by smooth damage functions, and their consequences could be catastrophic and abrupt.

Weitzman’s Dismal Theorem

The most powerful intellectual challenge to the standard IAM framework came from Martin Weitzman, a Harvard economist who died in 2019. Weitzman argued, in a series of papers beginning in 2009, that the key question in climate economics was not the expected cost of warming but the risk of catastrophic warming — and that standard cost-benefit analysis was fundamentally incapable of handling this kind of risk.

Weitzman’s “Dismal Theorem” showed that when the probability distribution of climate damages has a “fat tail” — meaning that extreme outcomes are more likely than a normal distribution would suggest — the expected value of damages can be infinite, or at least arbitrarily large, regardless of the discount rate. The intuition is that even a small probability of civilizational collapse outweighs any finite cost of prevention. If there is a 5 percent chance of warming exceeding 6 degrees Celsius, and a nonzero probability that 6 degrees of warming is incompatible with organized human civilization, then the expected cost of inaction is enormous — far larger than what smooth damage functions suggest.

Weitzman’s argument was not that catastrophe is certain or even likely. It was that the uncertainty itself — our profound ignorance about the tail risks of climate change — should dominate the policy calculus. Standard cost-benefit analysis, which focuses on the most likely outcome and treats uncertainty as a nuisance to be averaged away, is the wrong framework for a problem where the worst-case outcomes are existential.

The Dismal Theorem was controversial among economists — Nordhaus published a direct rebuttal — but it resonated with climate scientists, who had long warned that the tails of the temperature distribution were more dangerous than the mean. The practical implication was that precautionary action — spending more now to reduce the risk of catastrophic outcomes — was justified even if the expected damages (calculated with a smooth damage function) appeared manageable.

What IAMs Leave Out

Beyond the specific debates about discounting and damages, there is a broader concern about what IAMs exclude from their framework entirely.

Inequality within and between countries. Most IAMs model the world as a single representative agent or a small number of regional blocs. They do not capture the distributional consequences of climate change — the fact that poor people, poor countries, women, indigenous communities, and future generations bear disproportionate risks. An “optimal” climate policy that is efficient in aggregate may be catastrophic for the most vulnerable.

Political feasibility. IAMs identify theoretically optimal policies — typically a globally uniform carbon tax rising over time — without addressing whether such policies are politically achievable. The gap between the economist’s optimal policy and what democratic politics can actually produce is enormous, and ignoring it does not make it go away.

Path dependence and lock-in. IAMs treat investment decisions as reversible and technology choices as flexible. In reality, energy infrastructure has a long lifespan (a coal plant built today will operate for 30-50 years), and technological and institutional choices create path dependencies that constrain future options. Early action on emissions reduction is more valuable than IAMs suggest because it avoids locking in carbon-intensive infrastructure.

Ethical and cultural values. The cost-benefit framework that IAMs embody treats all values as commensurable — expressible in monetary terms and tradeable at the margin. But some of the things at stake in climate change — the survival of indigenous cultures, the existence of coral reefs, the habitability of low-lying island nations — are arguably incommensurable, meaning they cannot be meaningfully compared to dollars of GDP without violence to the values involved.

Why “Cost-Benefit Analysis” of Climate Change Is More Political Than It Appears

The deepest problem with IAMs is not their technical limitations — every model is a simplification, and modelers are generally honest about what their models cannot do. The problem is how IAMs are used in policy discourse: as neutral, scientific arbiters of what climate action is “economically justified.”

When Nordhaus’s model says the optimal warming path is 3 degrees Celsius, this is often reported as a finding — as if the model had discovered an objective truth about the world. But the “finding” is an output of assumptions that the modeler chose: the discount rate, the damage function, the treatment of uncertainty, the aggregation across regions and generations. Different assumptions — each defensible, none uniquely correct — produce radically different “optimal” paths. The model does not resolve the underlying value judgments; it translates them into technical language that obscures their political character.

This matters because the framing of climate policy as a cost-benefit question — how much should we spend? what is the optimal amount of warming? — systematically favors the status quo. Cost-benefit analysis compares the costs of action (which are immediate, measurable, and fall on identifiable actors) with the benefits of action (which are diffuse, uncertain, and accrue to future generations). The framework is biased toward inaction not because it is designed to be, but because the costs of action are always more visible and concrete than the costs of inaction.

An alternative framing — what is the maximum warming that is consistent with human dignity and planetary stability? what policies are needed to stay below that threshold? — starts from a different set of values and produces different conclusions. This is the framing implicit in the Paris Agreement, which set 1.5-2 degrees as a political target before anyone ran an IAM to check whether it was “optimal.” The target reflects a judgment about what is acceptable, not a calculation of what is efficient. And the policy question becomes not “what is the optimal amount of warming?” but “how do we achieve the target at the lowest cost?” — a very different question, with very different answers.

Neither framing is purely objective. Both embed values and political choices. The contribution of IAMs is to make the trade-offs explicit and quantitative, which is genuinely useful. The danger is pretending that the trade-offs are resolved by the model rather than by the people who built it and the societies that must live with the consequences. The politics of “optimal” carbon paths is, in the end, not about getting the model right. It is about deciding what kind of future we want — and for whom — and then building the models to serve that decision rather than to substitute for it.