The Discount Rate Is the Whole Argument

The common discourse regarding future environmental harm rests on a comparison between costs incurred now and damages avoided later. That comparison requires converting future money into present money, and the conversion requires a number: the discount rate. It is the single most consequential parameter in climate economics, it is not an empirical quantity in the way that a temperature sensitivity or an emissions factor is, and it is almost never discussed outside the technical literature. That combination is worth examining directly.

The arithmetic is unforgiving over long horizons. A hundred billion dollars of damage a century from now is worth roughly twenty-five billion today at a one and a half percent discount rate, and roughly one billion at four and a half percent. The Stern Review, using a rate near the low end, concluded that aggressive immediate abatement was overwhelmingly justified. William Nordhaus, using a rate near the high end and broadly the same climate science, concluded that a more gradual ramp was optimal. The two most influential analyses in the field reached opposing policy conclusions not because they disagreed about physics but because they disagreed about a parameter, and much of the ensuing public debate proceeded as though it were about the physics.

What the parameter encodes is worth unpacking. In the standard Ramsey formulation, the discount rate combines three things: a pure rate of time preference, the expected growth rate of consumption, and the elasticity of marginal utility, which governs how much less an additional dollar matters to a richer person. Only the growth term is straightforwardly empirical. Pure time preference — how much less a future person's welfare counts simply because they are in the future — is an ethical position, and Stern's near-zero value and Nordhaus's positive one are moral disagreements dressed in notation. The growth term carries its own difficulty: it assumes future generations will be richer, which justifies asking them to bear costs, but the whole premise of climate damage is that it makes them poorer than they would otherwise be. Importing a historical growth rate into an analysis of the thing that might interrupt it is a form of circularity that deserves more attention than it gets.

Then there is the problem that discounting handles worst. Martin Weitzman's argument, sometimes called the dismal theorem, is that when the probability distribution of climate outcomes has a fat tail — a small but non-negligible chance of catastrophic warming — the expected value of damage can be unbounded, and the cost-benefit apparatus stops producing usable answers. Under those conditions the appropriate frame is not investment but insurance. Households do not calculate the expected value of their house burning down and compare it to the premium; they pay to eliminate a tail risk they cannot survive. Applied to climate, this shifts the question from "what abatement level maximizes expected net present value" to "what are we willing to pay to rule out the outcomes we could not recover from" — a question that discounting simply does not answer.

The practical response in policy has been partial and quiet. Official social cost of carbon estimates have swung by an order of magnitude across administrations and methodologies, driven substantially by the discount rate rather than by new damage evidence. The UK Treasury's Green Book applies declining discount rates over long horizons, acknowledging that a constant rate cannot be right across a century. These are sensible adjustments, but they leave the central problem intact: an ethical judgment about our obligations to people not yet born is being made inside a technical parameter, where it is invisible to almost everyone affected by it. The more honest practice would be to present cost-benefit results across a range of rates, state plainly what each rate implies about intergenerational obligation, and let the ethical argument happen where it belongs — in the open, rather than in a footnote about model calibration.