TWO central contributions of this report are to characterize the uncertainty associated with the economic impacts of climate change and to estimate the extent to which these impacts will be borne unequally among Americans. In both cases, we note that average impacts gloss over an important aspect of the story: If the climate changes, there is a sizable chance that different types of impacts will be larger or smaller than the central estimate, and in many cases specific regions of the country experience impacts that differ substantially from the national average. While the primary purpose of this report is to provide empirically based, spatially explicit information about the risks businesses, investors, and households in different parts of the United States face from climate change, these insights are also important in how we price climate risk at a national level.
Both risk and inequality can increase the perceived costs of climate change above the expected cost of climate change; that is, the average impact we expect to see across possible futures and across regions of the United States. Uncertain outcomes and unequal impacts increase our perception (or valuation) of these costs because as individuals and as a society, we generally dislike uncertainty in our futures (e.g., individuals buy home insurance in part because the risk that a catastrophe could bankrupt a family is worrisome) and we dislike strong social inequalities (e.g., individuals donate money to charity in part to alleviate the hardships of poorer individuals). How much we dislike uncertainty and inequality affects how much these factors should influence our decision-making process, and they inform us of how much we should focus on future uncertainty or inequality in climate-change impacts relative to the average impact of climate change. In economics, the extent to which we are concerned about risk and inequality can be described by two factors:
Risk aversion: How averse are we to the uncertain possibility of bad future outcomes?
Inequality aversion: How much do we dislike having some individuals suffer greater losses than others, especially if proportionately greater losses fall on poorer individuals?
Both of these types of aversion reflect our underlying preferences and can thus be measured empirically, although it is possible that a decision maker may be more risk averse or inequality averse than one would estimate by observing individuals in a population. This might be true, for example, if increasing inequality has indirect effects on the economy or a population’s social well-being that are not understood or internalized by individuals within a population; it might also be true because the preferences of individuals acting collectively through democratic processes may differ from those of individuals acting individually in a market. It is worth noting that in many assessments of climate-change impacts, risk aversion and inequality aversion are assumed to be the same, although recent work suggests that the two need not, and very likely should not, be treated that way (Fehr & Schmidt 1999; Engelmann & Strobel 2004; Bellemare, Kröger, & Soest 2008; Crost & Traeger 2014).
Here we use our new results describing the probability distribution of impact across states within the United States to illustrate through example how decision makers could adjust their valuations of the damages from climate change to account for aversion to risk and inequality (see appendix E for mathematical details). In both cases, we summarize the additional costs imposed by risk and inequality as a premium, which is the additional cost that we would be willing to bear to avoid the inherent risk and additional inequality imposed by climate-change impacts. We assume the well-being of all Americans should be treated equally and consider how the value of mortality (using the VSL) and direct agricultural losses could be adjusted to account for the structure of their risk and their unequal impact, in large part because these two example sectors have nonlinear response functions that generate some of the largest variations in damage.
RISK AVERSION
Accounting for risk aversion stems from the observation that individuals and society at large dislike uncertainty in future costs. For example, suppose Anna has a salary of $40,000 this year. Further suppose Anna knows that if she stays at her current job, there is a 95 percent chance that she will get a 10 percent raise (a gain of $4,000) and a 1 percent chance that she will be fired (a loss of $40,000). The expected value of staying at her current job is therefore $41,800 (the sum of 95 percent times $44,000 and 5 percent times $0). Further suppose she has the opportunity to switch to a new job that also pays $40,000 but guarantees her employment next year (with no raise). If she is risk-neutral, then her current job is worth $1,800 more to her than the alternative; if she is risk-averse, she might nonetheless opt for the more certain alternative because she wished to avoid the possibility of being fired.
Following conventional practice, we measure risk aversion with a coefficient of relative risk aversion (RRA). An RRA of zero reflects risk neutrality; higher values reflect higher levels of risk aversion. Studies of the relative rates of return of safe investments (such as U.S. treasury bonds) and risky investments (such as stocks) suggest that the RRA reflected in U.S. financial investments is between 2.5 and 6, although it could be as low as 1 or as high as 12 (Ding et al. 2012). Another study of investments, aimed at separating risk aversion from preferences between current and future consumption, suggests an RRA of 9.5 (Vissing-Jørgensen and Attanasio 2003). Experimental results from a survey of individuals in the United States, the United Kingdom, Canada, Australia, and Mexico similarly suggest that the central third of individuals surveyed have values between 3 and 5, although one third of individuals surveyed have values less than 3 (half of these individuals are between 1.5 and 2.0), and one fifth of individuals surveyed have values greater than 7.5 (Atkinson et al. 2009).
We can use the RRA to turn the projected probability distributions of losses in each state in each period into certainty-equivalent losses per capita in that period; in other words, we find the losses that an individual would bear with certainty that have the same welfare impact as the distribution of losses characterized in this report (see appendix E). The risk premium is the difference between the certainty-equivalent loss and the expected actual loss: It is the hypothetical quantity one would be willing to pay just to avoid the uncertainty in climate effects (Kousky, Kopp, & Cooke 2011).
Because the production of commodity crops (maize, wheat, soy, and cotton) represents a small fraction (about 0.8 percent) of total economic output, it can have only a small effect on total income: In the 1-in-20 worst case for RCP 8.5 in 2080–2099 in the hardest-hit state, Iowa, agricultural losses constitute a 9 percent loss of overall output. The risk premium is therefore small relatively, ranging up to 7 percent of the value of the lost output for a high RRA of 10 (see the first row of table 15.1, where inequality aversion [IA] = 0). By contrast, because the mortality effects can be quite large—the 1-in-20 chance loss for RCP 8.5 in 2080–2099 is equivalent to 20 percent of output in Florida if measured using VSL—the risk premium can be significant. Even in the absence of inequality aversion, strong risk aversion can add as much as 16 percent to the value of the mortality losses (see the first row of table 15.2, where IA = 0).
TABLE 15.1 Combined inequality-risk premiums for agricultural impacts, 2080–2099
Note: RCP 8.5; premium as a percentage of expected losses for maize, wheat, cotton, and soy output
INEQUALITY AVERSION
Accounting for inequality aversion is important because most individuals dislike the notion that some individuals bear far more of a group’s cost than other members of a group. For example, in team efforts, most individuals usually find themselves unhappy if some members of the team do not “pull their weight,” thereby forcing others to do additional work to make up for this shortfall. In a more extreme example, if a foreign country were to invade a single U.S. state, Americans throughout the rest of the country would not simply stand by and let the population of that one state fend for itself; rather, the whole country would come to the aid of the invaded state. There are many similar cases, such as natural disasters, where the nation spends both effort and money to protect and support small groups of individuals because we do not believe those individuals should be left to suffer tremendous costs alone—instead, the country expends additional resources to share these burdens.
TABLE 15.2 Combined inequality-risk premiums for mortality impacts, 2080–2099
Note: RCP 8.5; premium as a percentage of expected losses, applying value of a statistical life
The degree of inequality aversion can be measured with a coefficient of inequality aversion (IA), analogous to the RRA. An IA of zero reflects inequality neutrality, implying there is no cost to inequality, while higher values reflect increasing levels of inequality aversion. Experimental results suggest that different individuals have a very broad range of IA values, with the central third of individuals having values between 2.0 and 7.5, one quarter having values between 0.5 and 1.5, and nearly a third having values greater than 7.5 (Atkinson et al. 2009).
In any given period, we can use the IA to turn the projected distribution of losses into an equivalent national, inequality-neutral loss (Gollier 2013). The inequality-neutral loss is a hypothetical economic loss that has the same welfare impact as the actual loss but, as a fraction of income, is shared evenly by all Americans. In other words, we find the level of loss that, if equalized across states, would yield the same welfare as the unequal cross-state distribution of output per capita (see appendix E). The inequality premium is the difference between the inequality-neutral loss and the expected actual loss; it is the hypothetical quantity one would be willing to pay just to avoid the additional inequality imposed by climate change impacts. Inequality-neutral losses will be larger if individuals who are initially poorer are harmed relatively more by climate change; they may be smaller if individuals who are initially richer are harmed relatively more.
Unlike the risk premium, it is possible for the inequality premium to be negative if climate change reduces initial wealth disparities, which can happen if climate change imposes sufficiently larger damage on wealthy populations than on poorer populations. In this case, the unequal distribution of climate impacts would lower the perceived cost of those impacts relative to their expected value. Thus, it is not obvious ex ante that accounting for inequality aversion will necessarily increase the perceived cost of climate change.
We note that we do not resolve differences in damage across counties within a state—accounting for such differences would likely increase the inequality premium—although cross-state impacts tend to be more unequal than cross-county impacts within each state. We also do not account for the distributional effects of climate change within a state by income or demographic group, which are likely more important than differences across counties.
For both agriculture and mortality, the inequality premium can be significant. Strong inequality aversion alone can increase the value of agriculture losses by up to 20 percent (see the first column of table 15.1, where RRA = 0), although the macroeconomic effects described in the preceding chapter dampen the inequality of direct agricultural impacts to some extent.
Strong inequality aversion can more than double the value of mortality losses, adding a 190 percent premium for an IA of 10 (see the first column of table 15.1, where RRA = 0). The large magnitude of the inequality premium for mortality arises because the mortality increase is highest in some of the nation’s poorest states and least in some of the richest. Among the poorest ten states, the per capita median mortality increase is 28 deaths per 100,000 people under RCP 8.5 in 2080–2099 (with additional deaths among the poorest states exceeding 30 per 100,000 in Florida and Mississippi); among the ten richest states, the average is a decrease in deaths of 3 per 100,000 people (with the reduction in deaths among the richest states exceeding 10 per 100,000 in Alaska, North Dakota, and Washington).
PUTTING IT TOGETHER
Thus far in this chapter, we separately analyzed the risk and inequality premiums for two types of impacts. However, we can combine both risk aversion and inequality aversion to compute an inequality-neutral, certainty-equivalent damage (see appendix E). This value is the hypothetical cost that, if shared equally among all individuals with certainty, would have the same welfare impact as the actual unequal distribution of state-specific risks. The combined inequality-risk premium is the difference between this hypothetical cost and expected damage; it is the cost of having unequal economic risks imposed by climate change. Calculating this premium helps us conceptualize how inequality in expected losses, inequality in the uncertainty of losses, and inequality in baseline income together increase the perceived value of climate-change damage.
For both agricultural losses and mortality, combining risk aversion with inequality aversion yields a higher inequality-risk premium. The magnitude of the increment from the combination partially reflects the magnitude of the individual effects. For agriculture, the increased premium from layering risk aversion (which is small in this impact category) on top of inequality aversion is small; for strong risk and inequality aversion, it amounts to a 25 percent increase in the value of losses compared to 20 percent from strong inequality and no risk aversion. For mortality, in contrast, strong risk and inequality aversion combined can add a premium of 250 percent compared to 190 percent for inequality aversion in the absence of risk aversion. If we focus on values most frequently observed in experiments (RRA and IA of roughly 4 [Atkinson et al. 2009]), which for RRA also coincides with the central Ding et al. (2012) estimate based on comparison of the prices of U.S. stocks and bonds, then the inequality-risk premium on agricultural loss and mortality are 13 and 90 percent of the expected loss. If instead we use an RRA of 10, close to the Vissing-Jørgensen & Attanasio (2003) estimate of 9.5, the mortality premium rises to 104 percent of the expected loss.
Overall, for the example impact categories we have assessed here, the inequality premium is substantially larger than the risk premium. While on its face this finding may be surprising—risk aversion is, after all, a key motivator for many policies and measures to manage climate-change risk—it is not when considered in the broader context of this analysis.
DECISION MAKING UNDER UNCERTAINTY
While our estimated probability distributions for the impacts we quantify represent a rigorous effort to assess probabilities in a framework that is both internally consistent and consistent with the best available science, they do not represent the only valid estimates. Among other factors, alternative climate-model-downscaling techniques, alternative probability weightings of climate models, alternative priors for the impact functions, and alternative assumptions about the changing structure of the economy would all change the estimates. There is no single correct approach.
Under such conditions of “deep” uncertainty, economists and decision scientists have developed a range of alternatives to the classical von Neumann-Morgenstern expected utility paradigm of estimating a single probability distribution that is interpreted as “true” and using this distribution to weight possible outcomes (as we did earlier when we estimated our risk premiums). One finding from this work is that many decision-makers are ambiguity averse: they view the non-uniqueness of the probability distribution as imposing a cost premium on top of the risk premium (Heal & Millner 2014). Another finding is that if catastrophic outcomes are possible and decision makers cannot rule them out, the remaining “fat tail” of the probability distribution of potential losses imposes an exceptionally large risk premium (Weitzman 2009).
Alternative approaches to the expected utility paradigm include a “maxmin” approach (i.e., choosing a course of action that among all possible worst-case outcomes is the least bad, an approach closely related to the “precautionary principle”) and an “α-maxmin” approach (i.e., making a decision based on a weighted mixture of the most likely outcome and the worst possible outcome). One might also choose to minimize regret—to find a strategy that across all possible futures minimizes the difference between the realized outcome and the best one could have done in the absence of uncertainty. These three approaches could all be applied using the impacts we estimate in this report to characterize the worst possible outcomes and the most likely outcomes. There are also additional alternative approaches that address the cost of ambiguity by estimating multiple probability distributions, each of which is assigned a probability of being correct (Kunreuther et al. 2012; Heal & Millner 2014).
Finally, we note that in this analysis, we have quantified only a subset of the potential costs of climate change, many of which cannot yet be rigorously assessed in an economic framework. Without the inclusion of the missing effects described in part 4, any evaluation of the worst possible outcomes would be incomplete. The true worst-case future is characterized not just by lost labor productivity and widespread heat-related deaths but also by (among other changes) international conflict and ecological collapse (Stern 2013). Part 4 surveys the gaps in our coverage.
