Climate change is frequently described as an existential threat, posing significant risks to our lives, livelihoods and the global economy. This view, however, stands in stark contrast to empirical estimates of the impact of climate change on economic activity, which imply only modest economic effects. Do existing estimates not account for the full impact of climate change, or are the costs of climate change truly so small? This column argues that shifting attention to a new and comprehensive source of variation in the data – global temperature itself – implies drastically larger impacts of climate change than conventional estimates.
During the last Pleistocene ice age 20,000 years ago, thick ice sheets covered a third of Earth, including present-day Chicago and the entire UK. Yet, global mean temperature was only 4-5°C lower than it is today. While it is difficult to forecast exactly what will happen once accumulated greenhouse gas emissions will have added 3-4°C to pre-industrial global temperature by 2100, there is no doubt that it will be a very different world than the one we have been used to.
Thus, it is not surprising that climate change is one of the most prominent topics in today’s policy discussions. This importance is backed by a longstanding consensus among climate scientists, who have consistently warned us about the potentially catastrophic consequences of climate change for our livelihoods.
Consensus about the impact of climate change on the economy is less prevalent among economists. Classic studies of the impact of temperature on economic activity have estimated that a 1°C increase in temperature implies at most a 3% reduction in output in the medium run (Dell et al. 2012, Burke et al. 2015). In the face of continued background economic growth, the argument goes, a few degrees of warming by the end of the century seems hardly problematic.
That argument, however, relies on a critical leap of faith. Traditional estimates relate changes in a country’s output to changes in its local temperature. For instance, these estimates compare how France’s output changes relative to Germany’s when France happens to be a little warmer than usual in a year while Germany is not.
But local temperatures and global temperature are not the same things. For instance, global temperature includes ocean surface temperatures, while local temperatures do not. When the oceans warm, evaporation, air humidity content, and precipitation regimes can change in much more dramatic ways than when a given country happens to be a little warmer in a specific year. To understand the impact of climate change on the economy, we are ultimately interested in global temperature.
There are good econometric reasons to focus on local temperatures in economic research. It varies over time and across countries, which allows researchers to leverage panel fixed effects methods to evaluate the consequence of local temperature changes without worrying too much about confounding drivers of output. But the key question is then: Is the effect of local temperature comparable to the effect of global temperature?
We address this question in Bilal and Känzig (2024) and propose a direct estimate of the impact of global temperature changes on economic activity. We demonstrate that global temperature implies economic losses that are five to six times larger than local temperature. A 1°C increase in global temperature leads to a 12% decline in output at peak. By contrast and as in previous work, we find that a 1°C increase in local temperature leads to a 1.5% decline in output with the same data and the same empirical specification.
How do we reach these conclusions? Focusing on global temperature involves new challenges relative to previous work. Global temperature only varies over time, and we thus must exploit time-series variation. This feature dictates our main empirical specification. Standard de-trending techniques account for the gradual warming effect of slowly accumulating green gas emissions that leads to the jointly trending behaviour of temperature and output over our sample (1960-2019). We construct global and local temperature shocks – innovations of temperature relative to the trend – and use them as our explanatory variable in our local projections specification. Figure 1 shows the response of output to these temperature shocks.
Figure 1 The effect of a 1°C global and local temperature increase on output
Our baseline estimates show that global temperature has drastically different effects from local temperature. Naturally, the time-series nature of our identifying variation requires care in interpreting these conclusions. There are four particularly delicate features of our approach that we address one by one.
First, global temperature shocks may happen to coincide with the global economic and financial cycle. We control for rich measures of world and local economic performance and find virtually identical results. Second, reverse causality due to feedbacks from greenhouse gas emissions may affect our estimates. We use state-of-the-art climate models to explicitly adjust our estimates accordingly. We show that reverse causality works against us qualitatively, and is quantitatively negligible. Third, our estimated output response may be specific to a particular period of time. We show that, in fact, our estimates are remarkably stable across three subperiods (1900-2019, 1960-2019, and 1985-2019). Fourth, global temperature shocks may be driven by some countries more than others, and these countries may simultaneously be subject to unrelated economic shocks. We control for country fixed effects and region-specific time trends and obtain virtually identical results. Taken collectively, our robustness exercises ultimately support the view that our specification captures the causal effect of global and local temperature shocks on economic activity.
Why, then, does global temperature depress economic activity so much more than local temperature? We argue that global temperature captures damaging consequences of changes in the climate system more comprehensively than local temperature. We show empirically that global temperature shocks predict a large and persistent rise in extreme climatic events that cause economic damage: extreme temperature, extreme wind, and extreme precipitation (Deschênes and Greenstone 2011, Hsiang and Jina 2014, Bilal and Rossi-Hansberg 2023a, 2023b). By contrast, local temperature shocks predict a much weaker rise in extreme temperature and barely any rise in extreme wind speed and precipitation. This conclusion is consistent with the geoscience literature: extreme wind and precipitation are outcomes of the global climate that depend on ocean temperatures and atmospheric humidity across the globe, rather than outcomes of idiosyncratic local temperature realisations.
To understand the implications of our reduced-form results for welfare and decarbonisation policy, we develop a simple neoclassical growth model that coincides with the economic block of the dynamic integrated climate economy model of Nordhaus (1992). We use our reduced-form results to estimate structural damage functions that map temperature to economic fundamentals such as productivity. We contrast the impact of climate change when we target our new global temperature estimates with its impact when we target traditional local temperature estimates.
We assess the welfare effects of climate change with a counterfactual that gradually increases global mean temperature from 2024 and reaches 3°C above pre-industrial levels by 2100 – so 2°C above 2024 temperatures. Figure 2 displays our results with confidence bands. Under global temperature estimates, climate change implies precipitous declines in output, capital and consumption that reach 50% by 2100. These changes imply a 31% permanent consumption present welfare drop. Thus, climate change leads to economic losses comparable to the 1929 Great Depression, but experienced forever. By contrast, under local temperature estimates, the welfare impact is only 4%.
Figure 2 The economic impact of climate change
We gauge the implications of our results for decarbonisation policy by constructing a social cost of carbon – the dollar value of all present and future economic losses around the world associated with emitting one tonne of carbon and letting the climate warm accordingly. Under global temperature estimates, the social cost of carbon is $1,056 per tonne of carbon dioxide. This value is six times larger than the high end of existing estimates, prevailing carbon prices (Rennert et al. 2022, Känzig and Konradt 2023) and our value based on local temperature estimates ($151 per tonne).
The social cost of carbon based on global temperature reverses established policy trade-offs. Most decarbonisation interventions in the Inflation Reduction Act cost $80 per tonne of carbon dioxide abated (Bistline et al. 2023). A conventional social cost of carbon of $151 per tonne implies that these policies are cost-effective only if governments internalise benefits to the entire world, as captured by the social cost of carbon. However, a government that only internalizes domestic benefits values decarbonisation benefits with a domestic cost of carbon, which is much lower than the social cost of carbon. Under local temperature estimates, the domestic cost of carbon of the US is $30 per tonne, making unilateral emissions reduction prohibitively expensive. Under our new global temperature estimates however, the domestic cost of carbon of the US becomes $211 per tonne, and thus largely exceeds policy costs. In that case, unilateral decarbonisation policy becomes cost-effective for large economies such as the US.
Source : VOXeu
