Michael D. Gerst

Agent-based modeling of climate policy

Model-based support of climate policy is scientifically challenging because climate change involves linked physical and social systems that operate on multiple levels: local, national, and international. As a result, models must employ some strongly simplifying assumptions. The most frequently used models typically assume hyper-rational and homogenous human behavior. These ensure tractability but, as a trade-off, abstract away the effects of less-than-rational decision-making and actor heterogeneity on domestic policy effectiveness and the influence of domestic constituents on international policy agreement. In this paper, we introduce a multi-level model framework, called ENGAGE, that relaxes some common modeling assumptions by adopting an agent-based approach. ENGAGE is styled after the Putnam two-level game, in which negotiators at the international level are constrained by the heterogeneous policy preferences and power of constituents at the domestic level. We proceed to provide a detailed description and demonstration of the prototype domestic-level module. Domestic actors include firms and households who function as agents within an evolutionary representation of economic growth, energy technology, and climate change. This allows an evaluation of policies that accounts for agent decision-making and social and technological change. Ultimately, we plan to use the ENGAGE model to simulate the two-way dynamic feedback between international agreements and domestic policy outcomes. Highlights? Introduces ENGAGE, a multi-level agent-based model of international climate policy. ? Prototype linked energy-economy model with endogenous technological change. ? Supports the design of robust climate change mitigation strategies.

Rare disasters and risk attitudes: international differences and implications for integrated assessment modeling

Evaluation of public policies with uncertain economic outcomes should consider societys preferences regarding risk. However, the preference models used in most integrated assessment models, including those commonly used to inform climate policy, do not adequately characterize the risk attitudes revealed by typical investment decisions. Here, we adopt an empirical approach to risk preference description using international historical data on investment returns and the occurrence of rare economic disasters. We improve on earlier analyses by employing a hierarchical Bayesian inference procedure that allows for nation-specific estimates of both disaster probabilities and preference parameters. This provides a stronger test of the underlying investment model than provided by previous calibrations and generates some compelling hypotheses for further study. Specifically, results suggest that society is substantially more averse to risk than typically assumed in integrated assessment models. In addition, there appear to be systematic differences in risk preferences among nations. These results are likely to have important implications for policy recommendations: higher aversion to risk increases the precautionary value of taking action to avoid low-probability, high-impact outcomes. However, geographically variable attitudes toward risk indicate that this precautionary value could vary widely across nations, thereby potentially complicating the negotiation of transboundary agreements focused on risk reduction.

The interplay between risk attitudes and low probability, high cost outcomes in climate policy analysis

Assessing the value of climate change mitigation requires an analysis framework that can account for societys attitude toward the risk of uncertain outcomes, especially those with low probability and high cost. For largely historical and computational reasons, this issue has not been adequately addressed by previous climate policy analyses. Using a novel stochastic version of a tractable global climate model, we demonstrate the importance of this shortcoming by showing how low probability, high cost outcomes interact strongly with risk attitudes to shape the results of quantitative analysis. Our results indicate that the relatively high levels of risk aversion implied by global investment behavior suggest that the large downside risk of economic catastrophe should weigh more heavily in policy consideration than the risk of over-mitigation. Further, this qualitative conclusion is robust to the particular specification of uncertainties concerning climate sensitivity and resultant economic damages. This conclusion is at odds with previous analyses that either assume low levels of risk aversion or employ numerical methods that underestimate disaster probabilities and therefore imply that the risk of over-mitigation should be of primary concern. This divergence suggests that more attention should be paid to the specification of risk attitudes and risk exposure in climate policy analysis.

Exploring Energy and Economic Futures Using Agent-Based Modeling and Scenario Discovery

By illuminating a range of possible futures, scenario analysis has proven valuable as a means for organizing and communicating the many uncertainties associated with predicting the development of the linked energy, economic, and climate systems. Thus far, scenarios have mostly been defined according to a sequential, piecewise process in which experts create plausible storylines that are then used as inputs to formal models. However, as the storylines are drafted separately from model construction, it is often difficult for models to engage completely with scenario themes. As a solution, methods of ‘scenario discovery’ have been proposed which apply statistical techniques to sets of model simulations to identify regions of the stochastic parameter space that result in distinctively different levels of policy performance. In our previous work, we described a novel multiattribute scenario discovery method and demonstrated application to ENGAGE, an agent-based model (ABM) of economic growth, energy technology, and carbon emissions. In the current contribution, we further demonstrate the utility of this approach by using ENGAGE to generate socioeconomic scenarios relevant to a given emissions target. We find that population growth, improvement in labor productivity, and efficiency of learning-by-doing regarding carbon-free energy technology are the key factors driving the success rate in achieving the selected target. This implies that these features should form essential elements of the storylines underlying socioeconomic scenarios if they are to provide a meaningful exploration of policy efficacy. Such results are consistent with those of more conceptual approaches. However, by being derived from the results of a quantitative model, our formulation is intrinsically consistent with practicable modeling assumptions and specifications.