Integrated Assessment of Climate Change in an Unequal World of Uncertain Economic Growth

A central policy question of climate economics is how much warming we should optimally accept, balancing the costs of reducing CO2 emissions with the damages from climate change. In economic theory, a global carbon price equal to the social cost of carbon (SCC) would implement this first best response to climate change. The SCC is calculated with integrated assessment models (IAM), which combine economic dynamics and climate dynamics. In this project, we contribute to our understanding how the optimal SCC is influenced by uncertainty: should we reduce our emissions more (or less)? And, even more importantly, what drives this uncertainty premium (or discount)? Traditionally integrated assessment models are deterministic. To account for parametric uncertainty ex post, models are run multiple times varying the uncertain parameters' values. The results are then averaged. This first order approximation to risk does not capture the behavioral response to uncertainty: Uncertainties mean we take actions to contain or reduce these directly; we in other act differently in their presence than if were they absent. We use a structural model of the relevant dynamics and interactions, describing the specific uncertainties at hand and how they enter the dynamic systems. Two main approaches have been taken to the risk premium: numerical modelling (calculating a number) and analytic modelling (deriving a formula that provides intuition). In this project, we merge these two approaches. The project focuses on climate uncertainty: how emissions translate into temperature increases in the long run - the equilibrium climate sensitivity. Its value is highly uncertain, due to the complexity of the climate systems and numerous uncertain feedbacks. Our central contribution to the literature is an analytic formula for the climate uncertainty premium for a rather generic IAM. It shows why we should increase our emission reduction efforts, and by how much and clarifies the distinct roles of the following elements: (1) the choice of social preferences, (2) the characteristics of the damages from climate change, and (3) the future policy response to uncertainty. Since the formula is approximate and relies on some simplifying assumptions, we test its quantitative predictions against a fully stochastic, integrated assessment model - a recursive implementation based on the DICE model. Our formula does indeed match the numeric uncertainty results closely for a wide range of preference-, growth- and damage specifications. We show that the quantitatively most important contribution to the climate uncertainty premium comes from the convexity of the damage function: how much faster damages get worse when temperatures increase. While it seems intuitive on its own, it contradicts and expands the previous literature which identified so-called prudence as the main driver of precautionary behavior in the face of uncertainty. A prudent decision maker is less averse to risk the more well off they are: by increasing their wealth, they can lessen the impact risk aversion has on their well-being in the face of uninsurable risk. But, due to the structure of the climate change challenge (the non-linear interactions between the climate and the economy) and the way the climatic uncertainty enters the problem (it influences the social cost of carbon in three distinct ways), this idea does not carry over to the context of climatic uncertainty and optimal emission reduction decisions. Quantitatively we find that for our baseline specification of the DICE model the increase in the social cost of carbon is modest: climatic uncertainty, despite implying potentially a more than 4.5 degree Celsius warmer planet from doubling of atmospheric CO2 (with 15% probability), only carries a risk premium of around 25%. Specifications with more convex, cubic damages increase this number to up to 75%. In ongoing work, we explore closely related aspects of climatic uncertainty for optimal climate policy. Firstly, we investigate the impacts of anticipating learning on optimal policy. How fast and by how much the uncertainty is reduced in the future is a priori an important aspect of the policy challenge. One robust numeric finding in this work is that the anticipation of faster learning does not lead to more lenient climate policy: there is no 'let us wait and see'-effect. Currently we work on explaining this result analytically. Secondly, we take a closer look at uncertainty preferences. We explore as an alternative the implications of ambiguity aversion: a dislike not only of uncertainty itself but also the inability to precisely quantify it. Applying smooth ambiguity aversion preferences to the numeric stochastic DICE implementation, we find that modelling climate sensitivity as deeply uncertain and the decision maker as averse to ambiguity does not change optimal policy notably. We explore ways to explain this a priori unintuitive result analytically.

This postdoctoral project let to a postdoctoral researcher achieving full time employment in academia working as a researcher and lecturer.

My proposed research strives for a better representation of the fundamental climate policy trade-off between mitigating and accepting damages. I intend to design an integrated assessment model that answers the following two questions: How to optimally aba te emissions when uncertain damages are distributed unevenly among a heterogeneous population? How to optimally mitigate climate change when the required technologies need to be developed first and the outcome of that process is uncertain? To achieve thi s, I wish to advance the literature along three dimensions: 1) I intend to analyze scientific and economic uncertainty in greater detail. 2) I plan to introduce endogenous technological innovation in a new way. In DICE family models for example abatement costs and emission intensity are exogenous. Likewise, most models feature exogenous growth, neglecting the question how we create and steer growth to be green. 3) Few existing IAMs treat distributional concerns. Within the standard economic framework the highly uneven distribution of wealth today easily turns climate policy into a tool for economic redistribution. I will develop a recursive dynamic programming version of the DICE model to address these 3 issues. The recursive structure permits me to trea t uncertainty in great detail. In contrast, most other models simply average deterministic runs, in each of which decisions are made under certainty. With recursive modeling, decisions are made under uncertainty. The recursive structure also allows for a better representation of risk preferences (Epstein and Zin's disentanglement of risk aversion and consumption smoothing). A similar approach will permit me to treat distributional concerns more elaborately than in the standard model.