Accounting for the risk of extreme outcomes in an integrated assessment of climate change

The potential for climate catastrophes, represented by 'fat-tailed' distributions on consequences, has attracted much attention recently. To date, however, most integrated assessment models have either been largely deterministic or deterministic with ex-post sensitivity analysis. The conclusions of such analyses are likely to differ from those employing models that accurately characterize society's joint preferences concerning time and risk, especially when distributions are fat-tailed. Using a dynamic stochastic general equilibrium model adapted from Nordhaus's DICE model, we show that failing to accurately account for risk can lead to substantial underestimation of the net benefits of greenhouse gas abatement. A robust finding of our analysis is that a lenient 'policy ramp' emissions reduction strategy is preferable over a more aggressive strategy--such as that advocated by the Stern Review--only if the model does not account for uncertainty about the climate system, the carbon cycle and economic damages, and specifies a consumption discount rate that is counterfactually higher than the historical global weighted average cost of capital of 4.0%. In the debate over uncertainty and time discounting, our results imply that what matters most in climate change assessment is the inclusion and particular specification of uncertainty rather than the precise choice of discount rate.