The Argument About the Discount Rate

Nearly every disagreement among economists about how fast to decarbonise traces back to one number. Not to any dispute about physics, and not to disagreement about abatement costs. To the discount rate.

Discounting converts future costs and benefits into present values. It is unavoidable: comparing a cost today against a benefit in 2150 requires some exchange rate between them. The difficulty is that the choice of rate swamps everything else over century-long horizons. At around three per cent, a catastrophe arriving in 2150 has a present value close to a rounding error. At near zero, the same event dominates every other term in the analysis.

The Stern Review used a rate of pure time preference near zero. The reasoning was ethical: being born later is not a reason to count for less, a position with a pedigree running back through Ramsey, who called discounting future utility a practice arising from weakness of the imagination, and Sidgwick before him. On that view, a public appraisal has no standing to prefer ourselves to our descendants merely because we exist first.

William Nordhaus objected that appraisal should use rates revealed in actual markets. His argument is not that future people matter less as persons. It is that money spent on abatement could instead be invested at returns available in the economy, with the proceeds given to the future, so using a rate below those returns can leave future generations worse off than the alternative. Observed rates are the relevant benchmark because they describe the actual opportunity.

Both positions are coherent and neither contains a factual error. The disagreement is about what a public appraisal is for, which is a question in ethics that happens to have arrived inside a spreadsheet. The two approaches produced policy recommendations differing by decades in timing.

The choice is not confined to climate. Several governments have reduced the discount rates used for long-horizon appraisal, and the effect is to move a class of projects (flood defences, rail, reforestation) from failing a cost-benefit test to passing it, with no change in their engineering or their expected physical outcomes. It is among the least scrutinised levers in public spending, precisely because it looks technical.

Weitzman added an argument that is probably the strongest in the literature and certainly the least used. If the probability distribution over climate outcomes has a sufficiently heavy tail, a small chance of truly catastrophic warming, then the expected loss may not converge at all. Standard cost-benefit machinery works by comparing expected values, and an expectation that does not converge cannot be compared with anything. The implication is that the analysis should resemble insurance against ruin rather than optimisation of an average.

It has had little effect on the documents governments actually produce, and the reason is structural rather than dishonest. Insurance against an unbounded loss has no obvious price, and an appraisal process that must output a number cannot accept the answer that the number does not exist. So the tail is acknowledged in the text, excluded from the model, and the central estimate carries the decision.