Climate change is a growing global threat to our environment and our economy. To determine appropriate policy measures to address global climate change, decisionmakers require transparent information both about the physical effects of climate change as well as the relationship between those effects and the economy.
To support this need, today we are releasing a working paper that adds to the body of investigative work on climate change and economic activity—The GDP-Temperature Relationship: Implications for Climate Change Damages. Our overarching findings are broadly consistent with the modeling results that have supported the federal government’s assessment of global climate damages through its social cost of carbon estimates. The top-line results of the economic effects of climate change from our study therefore will sound familiar to those who have followed this literature. However, there is also a novel twist in our approach that we hope will prove helpful toward encouraging more needed research in this area.
A large body of previous research has demonstrated the range of impacts that climate change can have on economic outcomes such as crop yields, human health, and energy consumption. Our study confirms previous estimates of economic damages under an unmitigated climate change scenario on the order of trillions of dollars in lost economic activity per year in the later part of this century. Previous research efforts have carried out aspects of our analysis on similar datasets. Whereas other studies have assessed the effects of climate change on the economy under specific statistical models, we bring a new approach by thoroughly evaluating hundreds of alternative models against objective performance criteria in order to determine which models perform the best—in a rigorous statistical sense.
The existing body of research on the relationship between climate and the economy can be broadly categorized into two approaches. The first is enumerative, or bottom-up, in that it seeks to understand the overall relationship between weather variables such as temperature and precipitation and the economy by summing up individual estimates of the effects on a given sector, such as agriculture, or the effects on human health. These relationships can be derived empirically through econometric analysis of relevant datasets or by modeling underlying physical processes. To assess the full effects of climate change on the broader economy based on the enumerative approach requires assessing the effects on each individual sector of the economy, accounting for potential interrelationships between the sectors, and summing the effects. The enumerative approach has informed the climate damage components embedded in “integrated assessment models” of the climate-economy relationship, including those used in the federal government’s estimation of the social cost of carbon.
An alternative to the enumerative approach is a top-down analysis that evaluates the effects of climate variables on an aggregate metric of economic activity such as gross domestic product (GDP), as we do in this study. Both of the approaches have relative merits and limitations. For example, the enumerative approach provides the ability to look in detail at individual sectors, but aggregating the damages from each sector and accounting for their potential interrelationships offers challenges, and achieving full coverage of all of the relevant sectors of economic activity is difficult at best.
One of the research challenges of the top-down approach, on one hand, is that there is not a clear theoretical framework to guide the statistical modeling. A myriad of statistical models is available to estimate the economic impacts of climate variations, while distinguishing the effects of climate variations from other drivers of overall economic activity. On the other hand, it is clear that the choice of statistical model can strongly influence the outcome of the analysis—even when applied to the same dataset—with some recent results estimating damages an order of magnitude greater than those embedded in official social cost of carbon estimates. In the absence of theory to guide the selection of a specific model, our work provides insight on the performance of a huge set of these models.
In our paper, we evaluate roughly four hundred different statistical models of the climate-economy relationship against a number of different cross-validation criteria, (i.e., how they perform at predicting out of sample, such as in forecasting), to determine their overall level of performance. A key finding from our study is that models with particular features provide distinctly better performance across these metrics. Once one focuses on these better-performing models, the very wide range of impacts evident across all the possibilities narrows considerably, with results that are consistent with the degree of impacts embedded in the most prominent integrated assessment models.
The results from these higher-performing models suggest that, under an unmitigated climate change scenario, the aggregate annual hit to GDP in 2100 would be centered around 1-2 percent per year, with about a 95 percent probability of being within a 3 percent loss to GDP. To put those percentages in dollar terms, the current global economy is roughly $80 trillion dollars in GDP annually; a 1 percent global loss would represent $800 billion and a 3 percent loss would represent $2.4 trillion in annual damages. At a 2–3 percent annual growth rate, the economy in 2100 would be 5–12 times today’s size, thereby further magnifying the level of future damages. The estimates from our study are roughly in line with those provided by the integrated assessment models used by the federal government to estimate global values of the social cost carbon.
Our work comes with an important set of caveats. The approach taken in this paper focuses on the effects of annual temperature fluctuations on GDP, but of course damages from climate change will also manifest themselves in ways that are not reflected in that relationship. For example, in prior estimates, roughly half of total damages from climate change are from non-market damages (such as damage to ecosystems, species, and other aspects of the environment), which our approach does not capture. Another limitation in applying our results—and others like them—to assess future damages from climate change is that the current trajectory of emissions is projected to take countries to temperature levels that are not represented in the historical record upon which statistical analysis is based. These limitations, as well as variation across models in expected GDP impacts, suggest that caution is warranted when these aggregate econometric data and methods are used to estimate the impacts of climate change on global GDP or are incorporated into integrated assessment models.
It is critical that there is continued development of science focused on estimation of the socioeconomic impacts of climate change.
