Sensitivity Analysis for Assessing Climate Change Impacts on the Economy in an Integrated Assessment Model

Climate scientists, economists, social scientists, and policymakers are adopting the tools and techniques of risk analysis in their efforts to identify future impacts of climate change. A basic model that links in a fundamental manner the physical climate system and the economy is the integrated assessment model (IAM). These models have been derived using varying tenants of economic systems and practices under the impacts of climate change. As such, IAMs typically include a rather comprehensive set of endogenous variables that define climate sensitivity, damage functions, and economic factors, which include growth, consumption, and returns. In a basic sense, the endogenous variables are written in various forms as functions of exogenous variables that typically identify the physical climate system, the economy, and the interaction between the two systems.

Over the past several decades, IAMs have evolved to provide the frameworks of many investigations and policies to identify the impacts of climate change on the economy. Because IAMs are designed to model a variety of physical and socio-economic under varying degrees of intricacy, they must represent complex interactions such that it is often difficult to identify the direct relationships between endogenous and exogenous variables. Beyond identifying relationships among the variables, it is equally or more difficult to identify uncertainties in key variables and how those uncertainties permeate through the model formulation and analyses.

While IAMs have been used extensively to link physical climate change and economic conditions, they have been strongly criticized as they do not account for important factors such as acute physical risks, tipping points, societal attitudes, and severe aggregation across political and physical boundaries. Another major criticism is that IAMs do not capture deep uncertainty. That is the propagation of uncertainty in exogenous input through the model to the analyses of the model output. To address this latter concern, sensitivity analysis (SA) has been proposed as a technique by which uncertainties in key model exogenous parameters may propagate and define uncertainties in model endogenous variables.

In its most basic form, SA is often defined as a “one-factor-at-a-time” (OFAT or OAT) methodology. These methods systematically implement changes (i.e., one standard deviation increase or decrease) in exogenous parameters and then examine corresponding changes in the modeled responses. In practice, this is straightforward but many studies have concluded that this type of SA is not adequate to disentangle the path of uncertainty throughout the model processing.

An alternative that has just become more widely applied is that of a global sensitivity analysis (GSA). In GSA, exogenous variables are varied simultaneously in a repeated fashion, and a partitioning of the variance is conducted to account for the role of uncertainty in each parameter and combinations of parameters.

In this presentation, an SA is applied to a simplified IAM by first applying an OFAT method to examine the spectrum of uncertainty in a key model output, which is the social cost of carbon. The OFAT results are compared to a GSA and the ranking of variables defined to drive the most uncertainty in the GSA is compared to the ranking from an OFAT. The key result hinges on whether the sensitivity rankings follow what would be considered a physically and economically consistent flow or influence of the exogenous to endogenous variables and whether that is consistent between the OFAT and GSA methods. While many of the criticisms of IAMs remain valid, the application of the two SAs is used to address the issue of uncertainty in an IAM.