Climate change is all about uncertainty: uncertainty governing the natural processes involved, uncertainty arisen due to its long-term nature, uncertainty on how agents react to those phenomena or uncertainty on how every event is modelled. Most Integrated Assessment Models (IAM) of climate change (including DICE) are deterministic and way too complex to permit a proper incorporation of uncertainty. Monte-Carlo methods are the most common approach to addressing uncertainty in the integrated assessment literature. However, Monte-Carlo methods, as implemented in this strand of literature, do not model decision making under uncertainty. They present a sensitivity analysis that averages over deterministic simulations.
The quantitative analysis of optimal climatic policies under uncertainty requires a recursive dynamic programming implementation of IAM.
Our formulation resembles closely that of Traeger (2014), who extends the small set of existing state of the art implementations of stochastic dynamic programming integrated assessment models. His main contribution is to reduce the number of states needed to represent the climate side of the DICE model without sacrificing its benchmark in capturing the interaction between emissions and temperature increase. Such reduction of the state space is crucial to permit additional state variables needed to capture uncertainty and avoid the curse of dimensionality. To model the adaptation behaviour, we rely primarily on the work by de Bruin et al (2009), which allows us to separately choose between every optimisation stage.
Our model reproduces well the basic properties of the original model. For instance, as it can be seen in the next figure, the model describes a hump-shaped path peaking around year 2200, abatement rate constantly grows until it reaches a ceiling after year 2200 and features a declining emission rate.
In this study we conduct a series of experiments covering a wide menu of uncertainties that may affect our model. We divide these uncertainties into four broad categories and provide an example of each group.
Optimal Mitigation-Adaptation mix under uncertainty
Next, we present an interactive picture describing the optimal (social planner) combination of mitigation and adaptation investment decisions under a menu of uncertainty scenarios: uncertain technology growth, Bayesian learning about climate sensitivity, and three tipping point alternatives. All magnitudes are compared to the equilibrium values under certainty.
In general, the introduction of uncertainty of any kind in the behaviour of variables is incorporated by the social planner through an increase in mitigation investment. The social planner is thus addressing potential severe environmental conditions in the future by tackling climate change sources, that is, total emissions of to the atmosphere instead of adapting climate change damages.
In this study we echo the IPCC’s call for greater integration of adaptation within Integrated Assessment modelling and extend the original DICE model by incorporating adaptation a la de Bruin, that is, specifying adaptation as a separate decision variable. Then, we perform a thorough analysis of the optimal balance between mitigation and adaptation under various stochastic scenarios.
All the results are qualitatively similar and point towards favouring mitigation with respect to adaptation as a method of insurance against potentially future adverse scenarios.
Future Avenues of Research: Many other additional features can be further incorporated into this model: uncertainty in the parameters governing the damage function, alternative damage function specifications, persistent effects of technology shocks,… Additionally, different types of adaptation could be jointly modelled. For example, Bosello et al (2010) construct a more involved framework in which different types of adaptation can be found. In particular, they distinguish between anticipatory adaptation (modelled as a stock variable), reactive adaptation (modelled as a flow variable) and accumulation of reactive adaptation knowledge.