Estimating the social cost of carbon

Welcome to Science Sessions, the podcast of the proceedings of the National Academy of Sciences, where we connect you with Academy members, researchers, and policymakers. Join us as we explore the stories behind the science. I'm Paul Gabrielson. Carbon dioxide emissions that cause climate change have a cost to society. Many economists have tried to quantify that cost in a dollar amount per ton called the Social Cost of Carbon. In a recent PNAS study, Francis Moore of the University of California Davis and colleagues reviewed studies on and modeling of the social cost of carbon. They also surveyed authors of the studies to understand what makes a high quality model. They concluded that previous studies likely underestimated what carbon emissions cost society. Francis, please introduce us to the social cost of carbon. What does this term mean and how is it calculated? The social cost of carbon is a quantification of all the damages from climate change to society, from a single ton of CO2 emissions. So if you think about what happens when we emit CO2, maybe from driving or from electricity production, that goes up into the atmosphere, or at least the portion of it stays up in the atmosphere for more or less and definitely, it has this long-term effect on the climate system. And in turn, that effect on warming and the hydrologic cycle have knock on effects on the economy, on society, on people's well-being, on the things we care about, around the world and going out far into the future. And so what the social cost of carbon conceptually is trying to do is trying to go through all of those impacts and add them up. In order to add them up, we have to convert them into common units and that units we use is dollars. So it's an economic valuation. Any one individual in the sense that they are exposed to climate change, they're going to be feeling a portion of that social cost of carbon. Many people in the US have experienced these smoky days, where it's either from local fires or it's from fires that are hundreds of miles distant, that smoke is blowing across the United States, that causes health damages that we can quantify. In theory, those should contribute to the social cost calculation. What's the background of this study? Economists have been attempting to quantify the social cost of carbon for a really long time. And so starting in the Obama administration, there was a process to develop a social cost of carbon for use in the benefit cost analysis of regulations that affect greenhouse gas emissions. The academic community has responded to that through really delving into some of these models and some of these questions and producing a rich and interesting body of work. What we were trying to do in this study is to pull a lot of those insights together. And so the paper really starts with a review of all the social costs of carbon estimates that have been published between 2020. And then we also collect a lot of information about what exactly the study was doing. So how did they set up their modeling? What did they assume about climate damages? What did that climate model look like? Various other kind of important aspects about how you think about risk, or how you think about the distribution, and equity implications of climate change. We collected a lot of variables from the studies about how they were doing the modeling, and then the social cost of carbon distribution that came out. Tell us about the methods of your study. Why did you survey authors of previous studies? You might think there are a whole number of reasons why you might get distortions or emissions or biases in the set of published social cost of carbon estimates. And so we went through and we decided we needed to complement that exercise with a survey of experts on the social cost of carbon in order to put these estimates into a kind of more holistic, fuller context. And so that's the second line of evidence where we go through and we survey all the authors that appear in the studies that we look at. And we essentially ask them, what do you think the social cost of carbon is in the literature? And then how do you understand that? Do you think that's overestimated? And then we asked them about some of these different modeling decisions that the studies made. And we asked them about their assessment essentially, the quality of some of those studies. We train a machine learning model, so random forest model. On the variation that we live in the literature. And so what that is trying to do is predict, given a certain type of modeling and a certain set of parameters, what would be the social cost of carbon that you would get out? The experts are telling us, well, we think this is a fairly good method, but we see that it's only in five percent of studies. Then papers that are doing that then would get a higher weight when we push them through the random forest model. So essentially what that's doing is rebalancing their estimates in the literature to more closely reflect the expert assessment. And then the results of that is what we call this synthetic social cost of carbon. And that's like final complete estimates from the study. What did you find? Why do previous studies underestimate the social cost of carbon? So starting with our first part of the study, which is this review of the literature, what we find is relatively high social cost of carbon, which is about $160 per ton of CO2. That mean estimate really come from the fact that we have a very long tail on the social cost of carbon distribution. And so there is very small probability of very, very high numbers. The second kind of main finding then is that experts in general really think the literature underestimates the true social cost of carbon. And they attribute that to a number of factors. Some of them are kind of related to what we call structural modeling assumptions. One of the most important that we identify this question of whether or not climate change damages affects economic growth rates or if it just affects the level of output. On average, our experts believe the literature is underestimating the social cost of carbon by about a factor of two. And they're attributing this to a number of factors. Then our final analysis where we pull these lines of evidence together to generate the synthetic social cost of carbon, that produces a mean estimate of 283. So that's quite substantially higher than the average we see in the literature. For instance, the US Environmental Protection Agency in 2023 released an update of its estimate of the social cost of carbon. And that central estimate of that was right around $200 per ton. Our mean estimate is quite substantially higher than that, even given that as part of the analysis, the EPA did a lot of important updates to the modeling. So what are the policy implications of a higher social cost of carbon than previously estimated? In this case, then we have the evidence that the social cost of carbon is higher. What that would suggest is taking action on climate change should be a correspondingly higher social priority. And that more expensive measures to reduce emissions would be justified on a benefit cost basis. We can think about everything from agricultural interventions to energy efficiency improvements to building out renewable electricity and electric vehicles. And from the social perspective where you have draw the line on, well, this is really too expensive. The social cost of carbon is a good guidepost for that. And so that's in theory how it connects into policy. More directly, it does have implications for how federal agencies do their benefit cost analysis. We kind of need to be able to put a number onto carbon emissions for regulations that are affecting carbon emissions. The EPA number is not too far from ours, but certainly we would be suggesting that it could be even higher. They're kind of older number from the set of models that we use as part of the Obama administration came out at around, say, like $50 per ton of CO2. And so both the EPA update and our numbers are very significantly higher than that. I would say it's a new consensus estimate from multiple lines of evidence that the $50 per ton range is really too low. What are the caveats or limitations of the study? What we are trying to do is to estimate what would model produce if they had different structures or they had different parameters, say, around the discount rate, which measures how much you value benefits in the future. To the extent that there are certain model combinations or interactions that are just missing from the literature, our statistical model is not going to be able to fill in the gaps. There are elements here that it would be quite helpful to have new original modeling studies that can fill in some important gaps in particularly some of these structural model elements. Yeah, we say we do the most we can with the literature that exists. Thanks for tuning into Science Sessions. You can subscribe to Science Sessions on iTunes, Spotify, or wherever you get your podcasts. If you liked this episode, please consider leaving a review and helping us spread the word.