Conference Agenda

We study how Earth subsystems, such as the Amazon rainforest, interact with global climate change through their internal dynamics, heterogeneously amplifying aggregate risk. Our framework captures the long-term value of marginal changes in subsystem states, including feedbacks, via a reduced-form model reflecting realistic geophysical processes. In our quantitative application, explicitely modeling Amazon dynamics raises the global social cost of carbon (SCC) by approximately 6% and implies a marginal value of local carbon storage that is about 33% larger than conventional estimates. This implies that a marginal ton of carbon stored in the Amazon rainforest is 25% more expensive than another type of carbon emissions. These results highlight the need for global climate policy and local conservation to recognize subsystems as dynamically vulnerable systems rather than static stocks.

Anthropogenic climate change is subject to a multitude of highly uncertain feedback processes, making the long-run impact of current emissions also highly uncertain. At present, we cannot reliably quantify the likelihood of differing global warming scenarios.

Decision theory distinguishes between known, quantifiable risks and situations of ambiguity or deep uncertainty. A fully rational decision maker can respond differently to ambiguity and to risk, and real-world decision makers frequently do. We show how aversion to ambiguity affects optimal climate policy in an integrated assessment of climate change. We derive an analytic social cost of carbon formula for an ambiguity averse decision maker in a generic integrated assessment model. We also quantify the impact of recursive smooth ambiguity aversion for a stochastic dynamic programming implementation of DICE.

Previous and paralleling approaches suggest substantial ambiguity premia on the optimal carbon tax. Our results show that the ambiguity premium is very small and optimal policy deriving from the standard Bayesian model is robust to ambiguity concerns under moderately large ambiguity aversion if climate policy is endogenous and policy maker's have rational foresight.

We establish the existence of a Lindahl equilibrium in a dynamic, uncertain world with externalities and provide an algorithm to compute it. The resulting allocation is efficient, time-consistent, and implementable with emissions trading in a climate setting. We quantify the regional welfare effects of climate cooperation by building on a standard integrated assessment model, and incorporating uncertainty and negative emissions. Compared with competition in emissions, farsighted cooperation benefits everyone and yields the largest relative gains for low-income regions through three channels: lower damages under the efficient path, burden sharing of abatement consistent with efficiency, and equilibrium compensations that internalize the public-good externality. Unlike prior work, we show that a Lindahl equilibrium is not necessarily in the core, and that, quantitatively, free-riding remains profitable. Yet the net compensations required to support Lindahl's equilibrium are modest, billions of U.S.\ dollars, compared with benefits in the trillions–suggesting room for international climate negotiations.

This paper aims to explore the impact of ambiguity, ambiguity aversion, and model misspecification on mitigation dynamics when several mitigation options are considered. It develops a continuous-time endogenous-growth economic model allowing for ambiguity and model misspecification on (i) climate and investment dynamics and (ii) uncertainty around technological jumps for potentially disruptive decarbonisation technologies. The model further innovates by considering a relative degree of technology richness, by representing emission-free capital, carbon intensity reductions and negative-emission technologies. Given the high dimensionality of the model and the inherent difficulties encountered in optimal control in the presence of misspecification corrections, we solve the model using a recent deep learning method to solve complex high-dimensional partial differential equation, the Deep-Galerkin Method with Policy Iteration Algorithm (DGM-PIA) proposed by Al-Aradi et al. (2022). Our preliminary findings indicate that misspecification and ambiguity aversion can give rise to a wide variety of transition strategies, including lower reliance on uncertain technologies, like negative-emission mitigation options.