The effect of learning on climate policy under fat-tailed risk

Abstract

The effect of learning on climate policy is not straightforward when climate policy is concerned. It depends not only on the ways that climate feedbacks, preferences, and economic impacts are considered, but also on the ways that uncertainty and learning are introduced. Deep (or fat-tailed) uncertainty does matter for the optimal climate policy in that it requires more stringent efforts to reduce carbon emissions. However, learning may reveal thin-tailed uncertainty, weakening the case for emission abatement: learning reduces the stringency of the optimal abatement efforts relative to the no learning case even when we account for deep uncertainty. In order to investigate this hypothesis, we construct an endogenous (Bayesian) learning model with fat-tailed uncertainty on climate change and solve the model with stochastic dynamic programming. In our model a decision maker updates her belief on the total feedback factors through temperature observations each period and takes a course of action (carbon reductions) based on her belief. With various scenarios, we find that the uncertainty is partially resolved over time, although the rate of learning is relatively slow, and this materially affects the optimal decision: the decision maker with a possibility of learning lowers the effort to reduce carbon emissions relative to the no learning case. This is because the decision maker fully utilizes the information revealed to reduce uncertainty, and thus she can make a decision contingent on the updated information. In addition, with incorrect belief scenarios, we find JEL Classification: Q54

Managing Climate Change Under Uncertainty: Recursive Integrated Assessment at an Inflection Point

“…Because they use a model that is structurally different from DICE, it is difficult to be sure where these differences stem from. Our results are similar to the small consequences of uncertainty in Hwang et al (2014).…”

“…34 Some recent papers have adopted nonstandard approaches to solving the Bellman equation, such as using "logarithmic", state-separable basis functions (Hwang et al, 2013(Hwang et al, , 2014Hwang, 2016) and fixing policy for some number of periods before approximating the remaining continuation value as a linear function of the per-period payoff (Heutel et al, 2015(Heutel et al, , 2016. We here caution against using less theoretically grounded methods for three reasons.…”

“…We extend DICE to include uncertainty about the climate's sensitivity to emissions. Following recent economic (Hwang et al, 2014;Kelly and Tan, 2015) and scientific (Roe and Baker, 2007;Roe, 2009) literature, the uncertain parameters are the feedbacks that determine the climate's sensitivity to emissions. This uncertainty can generate fat tails in the distribution of climate sensitivity.…”

“…Fat-tailed uncertainty about future warming can reduce optimal emissions by a substantial amount, but the size of the effect is sensitive to our ability to learn about future warming (Hwang et al, 2014;Kelly and Tan, 2015). Uncertainty about warming also increases the cost of agreeing to inflexible limits on total warming rather than using new information to control policy optimally (Fitzpatrick and Kelly, 2016).…”

A Recursive Method for Solving a Climate–Economy Model: Value Function Iterations with Logarithmic Approximations

“…Only the Bellman equation, arguments of the value function, and the first order conditions should be changed according to models. For instance, Hwang et al, (2013Hwang et al, ( , 2014 solve uncertainty and learning models on climate change having up to 9 endogenous state variables with the method of this paper.…”