Download PDF

Temperature and US Economic Activity: Evidence from Disaggregated Data
Abstract
We use sub-national data to examine the relationship between temperature and growth
within the US and the European Union. Different from previous studies based on the country-level
data, we find that the optimal temperature is much lower. Because most of production takes place
in regions with temperatures above the optimal temperature, even modest temperature increases
(e.g., 1 °C warming) have statistically and economically significant (negative) impact on the GDP
growth of the US and the European Union. Our results suggest more proactive climate policy.
1
Introduction
A consensus in the climate-economy literature is that the impact of climate change on
developed economies is little. See Poterba (1993), Stern (2006), Nordhaus (2008), and Tol (2009)
for early discussion. For instance, Schelling (1992) state: “I conclude that in the United States, and
probably Japan, Western Europe, and other developed countries, the impact [of climate change]
on economic output will be negligible and unlikely to be noticed” (p.6). More recently, .Dell, Jones
and Olken (2012) in a world sample of 125 countries find that temperature increases reduce not
only income levels but also growth rates, but only in developing countries. Burke, Hsiang and
Miguel (2015) find non-linear effects of temperature on economic growth in a sample of 166
countries, with growth peaking at an annual average temperature of 13 °C and declining strongly
at higher temperatures. Because developed countries are typically cold, Burke, Hsiang and Miguel
(2015) reinforce the consensus that climate change has trivial impact on developed economies. For
instance, they conclude: “Europe could benefit from increased average temperatures” (p. 3). The
common explanation for such results in the literature is that developed economies are better able
to adapt to climate change. (e.g., Mendelsohn, Nordhaus, and Shaw, 1994; Kahn, 2005; Olmstead
and Rhode, 2011; Hornbeck, 2012; Barreca et al., 2013).1
Although developed economies are better able to mitigate the impact of high temperatures,
mitigation cannot be complete, partially due to costs of adaptation and defensive investment.
Research has shown that temperature increases adversely affect production and labor productivity
in the US, the most developed economy. For instance, Fisher, Hanemann, Roberts, and Schlenker
(2012) find a negative impact of temperature increases on US agriculture. Cachon, Gallino, and
Olivares (2012) find that high temperatures reduce productivity and automobile production at the
1
See Dell, Jones and Olken (2014) for an excellent review.
2
plant level. Graff Zivin and Neidell (2014) show that temperature increases at the higher end of
the distribution reduce labor productivity in industries with high exposure to climate such as
agriculture, forestry, fishing, and hunting; mining; construction; transportation and utilities; and
manufacturing. There is also experimental evidence that temperature affects labor productivity
(e.g., Seppänen, Fisk, and Lei. 2006). Motivated by such evidence, recently, Deryugina and Hsiang
(2014) challenge the consensus in the literature by examining the effect of daily temperature on
annual personal income in United States counties, and find that increases in daily temperature
above 15 °C significantly reduce personal income at the county level.
We intend to extend Deryugina and Hsiang (2014) along the following three dimensions.
First, while Deryugina and Hsiang (2014) focus on the impact of daily temperatures, we study that
of long-run temperatures (i.e., 5-year average temperatures). From a benefit perspective, high
temperatures over the long run have more negative impact on production, resulting in larger
marginal benefits from mitigation. From a cost perspective, technological progress over the long
run may also help reduce marginal costs of mitigation. Taken together, economic agents might be
more willing and able to mitigate the effects of long-run temperatures instead of daily temperatures,
suggesting that focusing on long-run (not daily) temperatures might shed more light on if
mitigation makes developed countries invulnerable from climate change. Second, while Deryugina
and Hsiang (2014) focus on one country (i.e., the US), we study the US as well as the European
Union (EU), the largest economy in the world. Examining both the US and the EU helps
understand if it is the US or developed economies in general that are immune to climate change.
Third, different from Deryugina and Hsiang (2014) who focus on the impact of temperatures on
the level of (personal) income (level effects), we investigate growth effects of temperatures. As
Dell, Jones and Olken (2012) point out, level effects are transitory, but growth effects are
3
permanent. “For example, a temperature shock may reduce agricultural yields, but once
temperature returns to its average value, agricultural yields bounce back. By contrast, the growth
effect appears during the weather shock and is not reversed. A failure to innovate in one period
leaves the country permanently further behind” (Dell, Jones and Olken, 2012, p.73). Thus, growth
effects may be particularly important to understand the long-run effects of climate change.
To examine the relationship between temperatures and growth within an economy (e.g.,
the US or the EU), we utilize a unique dataset, the geophysically-scaled economic data set (GEcon)
developed by Nordhaus (2006), which provides consistent economic and geography data across
the global at the sub-national level. Our results can be easily summarized. With sub-national data,
we find a significant nonlinear relationship between temperature and growth within the US (EU),
with the optimal temperature much lower than previously thought. Because most of production in
the US and the EU takes place in regions with temperatures above the optimal temperature, even
modest temperature increases (e.g., 1 °C warming) have statistically and economically significant
(negative) impact on the GDP growth. Our results therefore suggest more proactive climate policy.
The remainder of the paper is organized as follows. Section 2 introduces the data. Section
3 describes the empirical methodology. Section 4 presents the results. Section 5 concludes the
paper with a brief summary.
2. Data
Previous studies (e.g., Dell, Jones and Olken, 2012; Burke, Hsiang and Miguel, 2015)
typically use country-level data to estimate the impact of temperatures on economic growth. There
are two potential issues with using country-level data. First, as Nordhaus (2006) points out, “for
many countries, averages of most geographic variables (such as temperature or distance from
4
seacoast) cover such a huge area that they are virtually meaningless” (p. 3511). Second, if there is
a nonlinear relationship between temperatures and growth within a country, the impact at the
average national temperature will be a biased estimate of the average impact across the country,
because of Jensen’s inequality. We therefore follow Nordhaus (2006) and Deryugina and Hsiang
(2014) and utilize sub-national data.
More specifically, we use the geophysically-scaled economic data set (GEcon) developed
by Nordhaus (2006), because the GEcon data covers not only the US but also the EU. The GEcon
data estimate gross output at a 1-degree longitude by 1-degree latitude resolution at a global scale.
The GEcon cell is approximately 100 km by 100 km, which is generally smaller than the size of
the major subnational political entities for most large countries (e.g., states in the US) and
approximately the same size as the second level political entities in most countries (e.g., counties
in the US). The conceptual basis of gross cell product (GCP) is the same as that of gross domestic
product (GDP) as developed in the national income accounts, except that the geographic unit of
the latitude-longitude grid cell is used instead of the political boundaries. A full description of the
data and methods can be found at the project web site (http://gecon.yale.edu).
We use GEcon 4.0, which is the most recently updated version. GEcon 4.0 only provides
the average temperature and precipitation from 1980 to 2008. Professor Nordhaus generously
provides annual temperature and precipitation data from 1980 to 2008. GCP and population are
available at five-year intervals, namely 1990, 1995, 2000 and 2005. We compute the 5-year growth
rate in GCP and population as the log first difference. Thus, we lose 1990 observations. For ease
of exposition, we refer to our 5-year periods ending 1990, 1995, 2000, and 2005 as periods 0, 1, 2,
and 3. To match the 5-year GCP growth, we compute 5-year average temperatures and
5
precipitation. To ensure that our results are not contaminated by low quality data, we use only the
best quality data in GEcon (i.e., quality = 1).
Table 1 reports the summary statistics. For the US sample, there are 1,203 unique cells
with required economic and geographic data. The 50th percentiles (p50) of 5-year GCP growth,
temperature, precipitation, and population growth are 14%, 8 °C, 532 millimeters (mm), and 6%,
respectively. For the EU sample, there are 860 unique cells. The 50th percentiles (p50) of 5-year
GCP growth, temperature, precipitation, and population growth are 9%, 9 °C, 657 millimeters
(mm), and 0%, respectively. Therefore, the US and the EU are different economies, making them
complementary in addressing our research question if developed economies in general are not
significantly affected by climate change. To mitigate the effects of outliers, we winsorize our data
at the 1% and 99% levels.
3. Empirical methodology
In this section, we develop an empirical framework for the analysis of the relationship
between temperature and growth, and discuss how we aggregate the impact across cells.
3.1 Empirical framework
Following Dell, Jones and Olken (2012), we employ a Cobb-Douglas type production
function to model the long-run relationship between temperatures and growth within a cell.
Yit  eit Ait Kit Lit
(1)
where cells are indexed by i and (5-year) periods are indexed by t, Y is total output (i.e., GCP), A
represents productivity, K measures capital, L stands for labor, and  is a disturbance term. Let
Kit  Rit Lit , where R is the capital-labor ratio. Then we have,
6
Yit  e it Ait Rit Lit  
(2)
Taking log first difference in the production function, we obtain the following dynamic growth
equation
yit  ait  rit  (   )lit   it   it 1
(3)
where lower-case letters represent the log first differences of the corresponding variables.
We generalize Dell, Jones and Olken (2012), and assume that productivity growth, ait,
follows
ait  bi  ct  dTit
(4.1)
where T is temperature, b is the cell fixed effects that capture unobserved time-invariant
differences between cells (e.g., agricultural versus high technology), and c is the time fixed effects
that account for common shocks in productivity (e.g., innovations in information technology). We
assume a similar structure for the change in the capital-labor ratio, rit
rit  ei  ft  gTit
(4.2)
where e is the cell fixed effects that capture unobserved constant differences between cells, and f
is the time fixed effects that account for common trends in the capital-labor ratio. Combining Eqs.
(3), (4.1), and (4.2), we have
yit  (bi  ei )  (ct  ft )  (d  g )Tit  (   )lit  ( it   it 1 )
(5.1)
We rewrite Eq. (5.1) as
yit  i  t  Tit  lit  it
(5.2)
where i  bi  ei ,  t  ct  f t ,   d  g ,      , and it   it   it 1 . Our fixed-effects
panel model utilizes the within variation to identify the causal effects of temperature on growth. If
mitigation is not complete, cells that experience higher temperature may suffer lower growth rates.
7
Because our identification strategy relies on the exogenous variation in temperature within a cell
over time, it cannot be confounded by cross-sectional differences across cells.
Recent research (e.g., Deryugina and Hsiang, 2014; Graff Zivin and Neidell, 2014)
suggests that the impact of temperature on economic activities may be nonlinear, and that
precipitation may be relevant. We thus augment our base model Eq. (5.2) with the linear splines
of temperature and precipitation
yit  i  t   mTitm    m Pitm  lit  it
m
(6)
m
where Titm ’s are the linear spline of temperature, and Pitm ’s are that of precipitation. We follow
Deryugina and Hsiang (2014) and use 3 °C-wide temperature bins. Thus, for the US sample, m is
set to 12, and the knots are -9, -6, -3, 0, 3, 6, 9, 12, 15, 18, and 21. That is, the first temperature bin
is T < -9 °C, the second one is -9 ≤ T < -6, and so on. For the EU sample, m is set to 7, and the
knots are 0, 3, 6, 9, 12, and 15. The mean impact of temperature on GDP growth is then
f (Tit )   mTitm .
m
3.2 Impact of climate change
With the parameter estimates based on our linear spline regression model, we conduct a
thought experiment to gauge the impact of temperature increases on growth. Essentially, we
compare the GDP growth under two scenarios. One is the “no warming” scenario in which
temperatures are assumed to stay at their 1995 levels (“counterfactual”), and the other is the
“warming” scenario in which temperatures increase.
The GDP of an economy is the sum of underlying cells’ GCP within that economy,
Yt   Yit . Thus, the cumulative GDP growth from period 1 to period t in the economy can be
i
8
written as gt   wi git , where wi 
i
Yi 0
(the production weight of cell i in period 0 within the
Y0
economy). Because the GDP growth in period t is approximated as the log first difference (recall
Eq. 3), the cumulative growth can be approximated as git   yit . Then, the cumulative GDP
t
growth of the economy is
gt   wi  yit   wi yit
i
t
t
(7)
i
Substituting Eq. (6), we have
gt   wi ( f (Tit )  X it )
t
(8.1)
i
m m
where X it  i  t    Pit  lit  it . Similarly, the GDP growth under the counterfactual is
m
gt   wi ( f (Ti1 )  X it )
t
(8.2)
i
Thus, the impact of climate change on the cumulative GDP growth is
gt  gt   wi ( f (Tit )  f (Ti1 ))
t
(9)
i
The annualized impact then is
gt  gt
. Within our sample period from 1995 to 2005, we use the
t
historical temperatures to estimate the impact of climate change on growth. Beyond our sample
period, we use temperature projections.
4 Empirical results
In this section, we present the linear spline panel regression results, and estimate the impact
of climate change on the US and the EU.
9
4.1 Nonlinear relationship between temperature and GDP growth within an economy
Burke, Hsiang and Miguel (2015) document a global nonlinear relationship between
temperature and growth with the country-level data, and find that growth peaks at an annual
average temperature of 13 °C. To show that our GEcon sample is compatible, we collapse our celllevel data to the country level, and run a linear spline panel regression model with the country and
time fixed effects. We plot the mean impact of temperature on growth, f (Tit )   mTitm , in Figure
m
1. It is interesting to see that although our panel length is much shorter (we have three periods,
while Burke, Hsiang and Miguel (2015) have 50 periods) and we focus on the long run relationship
(not the annual relationship), there is still evidence that growth peaks at an annual average
temperature of about 15 °C in the GEcon sample, if we use the country-level data. However, as
we have pointed out, country-level temperatures might not be informative to identify the
relationship between temperature and growth. We therefore explore the temperature-growth
relationship with the cell-level data.
We estimate three versions of Eq. (6) for the US sample, and report the panel regression
results in Table 2. Model (1) only includes the linear spline of temperature as well as the cell and
time fixed effects. Consistent with the country-level results, there is also a significant nonlinear
concave relationship between temperature and GDP growth at the cell level. However, different
from the country-level results, the optimal temperature is 6 °C, well below that based on the
country-level data (e.g., 13 °C in Burke, Hsiang and Miguel (2015)). Adding the linear spline of
precipitation in Model (2) does not change the results qualitatively. Further enhancing the model
with the population growth in Model (3) produces similar results. To visualize the mean impact of
m m
temperature on GDP growth, we plot f (Tit )   Tit for Model (1) and Model (3) in Panels A
m
10
and B of Figure 2. As we can see, GDP growth increases with temperature, but only until 6 °C.
This is the central finding of the paper. What explains the difference in the optimal temperatures
between country- and cell-level regressions? First, as Nordhaus (2006) points out, for large
economies such as the US, national average temperatures are “virtually meaningless”. Second, if
there is a nonlinear relationship between temperatures and growth within an economy, the impact
at the national average temperature will be a biased estimate of the average impact across the
country, because of Jensen’s inequality.
Are the US results sample specific? To answer this question, we repeat the same analysis
for the EU. First, we estimate three versions of Eq. (6) for the EU sample, and report the panel
regression results in Table 3. Regardless of the version of the model we estimate, there is a
significant nonlinear concave relationship between temperature and GDP growth at the cell level
within the EU. However, different from the country-level results in Burke, Hsiang and Miguel
(2015) but consistent with the cell-level results in the US sample, the optimal temperature is about
6 °C, well below that based on the country-level data in Burke, Hsiang and Miguel (2015). To
m m
visualize the mean impact of temperature on GDP growth, we plot f (Tit )   Tit for Model (1)
m
and Model (3) in Panels A and B of Figure 3. Again, GDP growth increases as temperature
increases, but only until 6 °C. Thus, the nonlinear relationship between temperature and GDP
growth with the optimal temperature at about 6 °C is not unique to the US sample, but applies to
many developed economies.
4.2 Impact of climate change on the US and the EU
11
Because the optimal temperature is much lower than previously found, our results suggest
that even developed economies that have low national average temperatures may still be adversely
affected by climate change. We quantify the impact for the US and the EU in this section.
4.2.1 Heterogeneous temperature increases across cells
We first examine temperature increases across temperature bins. The idea is to understand
if warming is homogenous or heterogeneous across temperature bins. If warming is homogenous
across temperature bins within an economy, we could apply national temperature projections, for
instance, from the Intergovernmental Panel on Climate Change (IPCC, 2014), to evaluate the
impact of climate change on GDP growth. Otherwise, we would have to take into account the
heterogeneity across temperature bins.
For the US sample, we classify the cells into 12 temperature bins based on their
temperatures in 1995, compute temperature changes at the cell level, and plot the distributions
(box plots) of temperature changes within each temperature bin in Panel C of Figure 2. The
horizontal line indicates the increase in the national (area-weighted) average temperature. As we
can see, warming is not homogenous in the US, as cells with temperatures below (above) the
optimal temperature of 6 °C experience more (less) warming.
For the EU sample, we conduct a similar analysis, except that we classify the cells into
seven temperature bins based on their temperatures in 1995. Again, warming is not homogenous:
cold and warm cells seem to experience more warming than the cells around the optimal
temperature of 6 °C. Thus, we might mis-estimate the impact of climate change on GDP growth if
we use national temperature projections to characterize the warming at the cell level.
12
Because structural temperature projections at the cell level are not available, we use a
reduced-form approach by assuming “change as usual” in the Schmalensee et al. (1998) sense.
That is, we assume that the temperature change in a cell will continue at its historical mean rate.
Because historical temperature changes manifest historical trends in natural, economic, and social
conditions (e.g., climate policy), as long as these conditions change at roughly the historical pace,
our reduced-form forecasts are unbiased. We estimate the mean temperature increase in a cell with
the mean temperature increase in the temperature bin that the cell falls in 1995. Doing so enables
us to use bootstrap to generate sampling distributions of mean temperature increases for each cell.
Specifically, the sampling distributions of mean temperature increases are based on 5,000 random
bootstrap replications with each replication consisting of 30% of cells within each temperature bin.
4.2.2 Heterogeneous GDP distribution
Eq. (9) suggests that the impact of climate change on GDP growth also depends on the
distribution of GDP. For instance, if most of production occurs in cells with temperatures below
the optimal temperature, climate change helps increase GDP growth, and vice versa. We therefore
examine the distribution of GDP. For the US sample, we classify the cells into 12 temperature bins
based on their temperatures in 1995, compute the GDP weights of the cells, and plot the histogram
of the combined GDP weight across temperature bins in Panel D of Figure 2. Interestingly,
production occurs mainly in cells with temperatures above the optimal temperature, suggesting
that temperature increases due to climate change could have significant negative impact on US
production.
For the EU sample, we conduct a similar analysis, except that we classify the cells into
seven temperature bins based on their temperatures in 1995. Again, GDP is mainly produced in
13
the cells with temperatures above the optimal temperature, implying that climate change could
have substantially negative impact on the EU.
4.2.3 Impact of climate change
Within our sample period from 1995 to 2005, we use the historical temperatures to estimate
the impact of climate change on growth. Beyond our sample period, we use temperature
projections based on the sampling distributions of mean temperature increases. For instance, the
~
projected temperature of cell i in 2010 is Ti 2005  T m , where Ti 2005 is its temperature in 2005, and
~
T m is the mean 5-year temperature increase in temperature bin m that cell i falls in based on a
bootstrap replication. Again, we use 5,000 bootstrap replications to construct the 95% confidence
interval of the impact estimate.
For the US sample, we first plot the projected national (area-weighted) average
temperatures in Panel D of Figure 2. Based on our reduced-form approach, the US national average
temperature is projected to increase by 1 °C, from about 9 °C in 2005 to 10°C in 2050. This is
within the IPCC interval estimates for the increase in the global mean temperature by the end of
the 21st century (2081–2100) relative to the 1986–2005 period, which is from 0.3°C to 4.8°C
(IPCC, 2014). If the optimal temperature were 13 °C, the projected climate change would not have
much negative impact on the US GDP growth. However, as we have discussed, the optimal
temperature of 13 °C based on the country-level data may not be informative, if there is a nonlinear
relationship between temperature and GDP growth within an economy. Therefore, we estimate the
impact of climate change on the US GDP growth based on the parameter estimates in Model (3)
of Table 2, and report the results in Panel F of Figure 2. Strikingly, based on the cell-level
temperature-growth relationship, climate change has significantly negative effects on the US GDP
14
growth. By 2050, with just about 1 °C warming, the annual GDP growth in the US would be 0.82%
lower relative to the counterfactual case in which temperatures stay at their 1995 levels. This
impact is not only statistically but also economically significant, given that the GDP growth in the
US has been about 3.1% since 1960 based on the data from Bureau of Economic Analysis (BEA).
The significant impact is due to that most of production takes place in cells with temperatures
above the optimal temperature of 6 °C.
Burke, Hsiang and Miguel (2015) use the country-level data and find that “Europe could
benefit from increased average temperatures” (p. 3). This again holds if the optimal temperature
were 13 °C. However, our cell-level estimates suggest a much lower optimal temperature for the
EU, and potentially large negative impact of climate change on GDP growth (because again most
of its production occurs in cells with temperatures above the estimated optimal temperature). To
quantify the impact, we repeat the similar exercise. First, based on our reduced-form approach, the
EU (area-weighted) average temperature is projected to increase by about 1 °C, from about 9 °C
in 2005 to about 10°C in 2050, which is again within the IPCC interval estimates for the increase
in the global mean temperature by the end of the 21st century (2081–2100). Second, based on the
parameter estimates in Model (3) of Table 3, we estimate and report the impact of climate change
on the EU GDP growth in Panel F of Figure 3. Consistent with the US results, based on the celllevel temperature-growth relationship, climate change has significantly negative effects on the EU
GDP growth. By 2050, with just about 1 °C warming, the annual GDP growth in the EU would be
1.43% lower relative to the counterfactual case in which temperatures stay at their 1995 levels.
This impact is not only statistically but also economically significant. Our US and EU results thus
suggest that the country-level data may have underestimated the impact of climate change on GDP
growth in developed economies.
15
5. Conclusions
A consensus in the climate-economy literature is that the impact of climate change on
developed economies is little. In a recent study, Burke, Hsiang and Miguel (2015) reinforce this
consensus, and conclude that “Europe could benefit from increased average temperatures” (p. 3).
We argue that if there is a nonlinear relationship between temperature and GDP growth within an
economy, the country-level results in previous studies (e.g., Burke, Hsiang and Miguel, 2015) may
underestimate the negative impact of climate change on GDP growth. We thus use the sub-national
data to examine the relationship between temperature and GDP growth in the US and the EU. With
sub-national data, we find that the optimal temperature is much lower than previously thought.
Furthermore, both warming and production are heterogeneous across sub-national units within a
country, with most of production taking place in regions with temperatures above the optimal
temperature. As a result, even modest temperature increases (e.g., 1 °C warming) have statistically
and economically significant (negative) impact on the GDP growth of the US and the EU.
Therefore, our results suggest more proactive climate policy.
16
References
Barreca, Alan, Karen Clay, Olivier Deschenes, Michael Greenstone, and Joseph S Shapiro. 2013.
Adapting to climate change: the remarkable decline in the US temperature-mortality
relationship over the 20th century. Tech. rep., National Bureau of Economic Research.
Burke, Marshall, Solomon M. Hsiang, and Edward Miguel. 2015. Global non-linear effect of
temperature on economic production. Nature, forthcoming.
Cachon, Gerard P., Santiago Gallino, and Marcelo Olivares. 2012. Severe Weather and
Automobile Assembly Productivity, Columbia Business School Research Paper No. 12/37.
Dell, M., B. F. Jones and B. A. Olken. 2009, Temperature and Income: Reconciling New CrossSectional and Panel Estimates, American Economic Review 99, 198-204.
Dell, M., B. F. Jones and B. A. Olken. 2012, Temperature Shocks and Economic Growth: Evidence
from the Last Half Century. American Economic Journal: Macroeconomics 4 (3): 66–95.
Deryugina, Tatyana, and Solomon M. Hsiang. 2014, Does the Environment Still Matter? Daily
Temperature and Income in the United States. NBER Working Paper 20750.
Fisher, Anthony C. W. Michael Hanemann, Michael J. Roberts, and Wolfram Schlenker. 2012.
The Economic Impacts of Climate Change: Evidence from Agricultural Output and
Random Fluctuations in Weather: Comment. American Economic Review 102 (7): 3749–
60.
Graff Zivin, J and M Neidell, 2014. Temperature and the Allocation of Time: Implications for
Climate Change, Journal of Labor Economics, 32: 1-26.
Hornbeck, Richard. 2012. \The Enduring Impact of the American Dust Bowl: Short-and LongRun Adjustments to Environmental Catastrophe." The American Economic Review 102
(4):1477-1507.
17
Intergovernmental Panel on Climate Change, 2014, The IPCC Fifth Assessment Report,
http://www.ipcc.ch/.
Kahn, Matthew E. 2005. \The Death Toll from Natural Disasters: the role of Income, Geography,
and Institutions." The Review of Economics and Statistics 87:271-284.
Mendelsohn, R. O., W. D Nordhaus, and D. Shaw, 1994, The Impact of Global Warming on
Agriculture: A Ricardian Analysis, American Economic Review 84, 753-71.
Nordhaus, W.D., 2006, Geography and Macroeconomics: New Data and New Findings. Inaugural
Article, Proceedings of the National Academy of Sciences 103, 3510-3517.
Nordhaus, William. 2008. A Question of Balance: Weighing the Options on Global Warming
Policies. Yale University Press.
Olmstead, Alan L. and Paul W. Rhode. 2011. \Adapting North American wheat production to
climatic challenges, 1839{2009." Proceedings of the National Academy of Sciences 108
(2):480-485.
Poterba, James M. 1993. Global Warming Policy: A Public Finance Perspective. Journal of
Economic Perspectives 7 (4):pp. 47{63.
Schelling, Thomas C. 1992. Some Economics of Global Warming. American Economic Review
82 (1): pp. 1-14.
Schmalensee, R., Stoker, T. M., Judson, R. A., 1998. World carbon dioxide emissions: 1950-2050.
Review of Economics and Statistics 80, 15–27.
Seppänen, Olli, William J. Fisk, and Q. H. Lei. 2006. Effect of Temperature on Task Performance
in Office Environment. Berkeley, Calif.: Lawrence Berkeley National Laboratory.
Stern, Nicholas. 2006. Stern Review: The Economics of Climate Change. Cambridge University
Press.
18
Tol, Richard SJ. 2009. The economic effects of climate change." Journal of Economic Perspectives
23 (2):29-51.
19
Figure 1 Global nonlinear relationship between temperature and growth with the country-level data
6
E(GDP growth)
4
2
0
0
5
10
15
20
Temperature (Celsius)
25
30
We collapse the GEcon cell-level data to the country level, and run the following linear spline panel regression model
with the country and time fixed effects.
yit  i   t   mTitm  it
m
where countries are indexed by i and (5-year) periods are indexed by t, yit is the GDP growth, i is the country fixed
m
effects, t is the time-fixed effects, and Tit ’s are the linear spline of temperature, We plot the mean impact of
f (Tit )   mTitm
m
temperature on growth,
, in Figure 1.
20
Figure 2 Temperature and growth at the cell level in the US
Panel A: Mean impact of temperature
based on Model (1)
Panel B: Mean impact of temperature
based on Model (3)
0
0
E(GDP growth)
E(GDP growth)
-1
-1
-2
-2
-3
-3
-4
-4
-15
-12
-9
-6
-3
0
3
6
9
12
Temperature (Celsius)
15
18
21
24
-15
95% confidence intervals
-12
-9
-6
-3
0
3
6
9
12
Temperature (Celsius)
15
18
21
24
95% confidence intervals
Panel C: Distributions of temperature increases
Panel D: Distribution of GDP
2
30
1.5
25
1
20
.5
15
0
10
-.5
te
m
te
m
p(
. ,te
9)
m
p(
-9
,
te
6)
m
p(
-6
,-3
te
)
m
p(
-3
,0
)
te
m
p(
0,
3
)
te
m
p(
3,
6
)
te
m
p(
6,
9)
te
m
p(
9,
12
te
m
)
p(
12
,1
te
5
m
)
p(
15
,1
te
8)
m
p(
18
,2
1)
te
m
p(
21
,.)
0
Panel E: Temperature projections
te
m
-1
p(
. ,9)
p(
-9
,
te
6)
m
p(
-6
,-3
te
)
m
p(
-3
,0
te
)
m
p(
0,
3)
te
m
p(
3,
6)
te
m
p(
6,
te
9)
m
p(
9,
12
te
m
)
p(
12
,1
te
5)
m
p(
15
,1
te
8)
m
p(
18
,2
1)
te
m
p(
21
,.)
5
Panel F: Impact of climate change on GDP growth
0
11
-.2
Annual GDP growth
Annual GDP growth
10.5
10
-.4
-.6
9.5
-.8
9
-1
2000 2005 2010 2015 2020 2025 2030 2035 2040 2045 2050
Year
2000 2005 2010 2015 2020 2025 2030 2035 2040 2045 2050
Year
We plot the mean impact of temperature on GDP growth for Model (1) and Model (3) in Panels A and B. Model (1)
is:
yit  i   t   mTitm  it
m
Model (3) is:
yit  i  t   mTitm    m Pitm  lit  it
m
m
21
where cells are indexed by i and (5-year) periods are indexed by t, yit is the GDP growth, i is the cell fixed effects, t
m
m
is the time-fixed effects, lit is the population growth, Tit ’s are the linear spline of temperature, and Pit ’s are that of
precipitation. We follow Deryugina and Hsiang (2014) and use 3 °C-wide temperature bins. For the US sample, m is
set to 12, and the knots are -9, -6, -3, 0, 3, 6, 9, 12, 15, 18, and 21. That is, the first temperature bin is T < -9 °C, the
second one is -9 ≤ T < -6, and so on.
In Panel C, we classify the cells into 12 temperature bins based on their temperatures in 1995, compute temperature
changes at the cell level, and plot the distributions (box plots) of temperature changes within each temperature bin.
The horizontal line indicates the increase in the US (area-weighted) average temperature.
In Panel D, we classify the cells into 12 temperature bins based on their temperatures in 1995, compute the GDP
weights of the cells, and plot the histogram of the combined GDP weight across temperature bins.
In Panel E, we plot the national average temperature projections based on the sampling distributions of mean
temperature increases. The shaded area is the 95% confidence interval of the impact.
In Panel F, we report the impact of climate change on the US GDP growth based on the parameter estimates in Model
(3) and our temperature projections. The shaded area is the 95% confidence interval of the impact.
22
Figure 3 Temperature and growth at the cell level in the EU
Panel A: Mean impact of temperature
based on Model (1)
Panel B: Mean impact of temperature
based on Model (3)
1
1
.5
E(GDP growth)
E(GDP growth)
.5
0
-.5
0
-.5
-1
-1
-1.5
-1.5
-3
0
3
6
9
12
Temperature (Celsius)
15
18
-3
21
0
3
6
9
12
Temperature (Celsius)
15
18
21
95% confidence intervals
95% confidence intervals
Panel C: Distributions of temperature increases
Panel D: Distribution of GDP
1
50
40
.5
30
20
0
10
5,
.)
p(
1
te
m
2,
15
)
p(
1
te
m
p(
9
,1
2)
,9
)
te
m
p(
6
,6
)
te
m
te
m
p(
3
,3
)
Panel F: Impact of climate change on GDP growth
0
10.5
10
Annual GDP growth
Annual GDP growth
p(
0
te
m
te
m
Panel E: Temperature projections
te
m
p(
.,
0)
5,
.)
p(
1
2,
15
)
te
m
p(
1
p(
9
,1
2)
,9
)
0
te
m
te
m
p(
6
,6
)
p(
3
te
m
te
m
te
m
p(
0
p(
.,
0)
,3
)
-.5
9.5
-.5
-1
9
-1.5
2000 2005 2010 2015 2020 2025 2030 2035 2040 2045 2050
Year
2000 2005 2010 2015 2020 2025 2030 2035 2040 2045 2050
Year
We plot the mean impact of temperature on GDP growth for Model (1) and Model (3) in Panels A and B. Model (1)
is:
yit  i   t   mTitm  it
m
Model (3) is:
yit  i  t   mTitm    m Pitm  lit  it
m
m
23
where cells are indexed by i and (5-year) periods are indexed by t, yit is the GDP growth, i is the cell fixed effects, t
m
m
is the time-fixed effects, lit is the population growth, Tit ’s are the linear spline of temperature, and Pit ’s are that of
precipitation. We follow Deryugina and Hsiang (2014) and use 3 °C-wide temperature bins. For the EU sample, m is
set to 7, and the knots are 0, 3, 6, 9, 12, and 15.
In Panel C, we classify the cells into seven temperature bins based on their temperatures in 1995, compute temperature
changes at the cell level, and plot the distributions (box plots) of temperature changes within each temperature bin.
The horizontal line indicates the increase in the EU (area-weighted) average temperature.
In Panel D, we classify the cells into seven temperature bins based on their temperatures in 1995, compute the GDP
weights of the cells, and plot the histogram of the combined GDP weight across temperature bins.
In Panel E, we plot the average temperature projections based on the sampling distributions of mean temperature
increases. The shaded area is the 95% confidence interval of the impact.
In Panel F, we report the impact of climate change on the EU GDP growth based on the parameter estimates in Model
(3) and our temperature projections. The shaded area is the 95% confidence interval of the impact.
24
Table 1 Summary statistics
Panel A: US
N
St. Dev.
p1
p5
p10
p25
p50
p75
p90
p95
p99
5-yr GDP
growth
3807
0.18
-0.32
-0.14
-0.08
0.06
0.14
0.21
0.31
0.37
0.46
5-year average
temperature
4836
9
-13
-9
-6
2
8
13
18
20
23
5-year average
precipitation
4836
523
106
169
229
319
532
1028
1344
1530
2650
5-yr Population
growth
3813
0.11
-0.16
-0.04
-0.01
0.03
0.06
0.08
0.11
0.15
0.23
Panel B: EU
5-yr GDP
5-year average
5-year average
5-yr Population
growth
temperature
precipitation
growth
N
2841
3440
3440
2841
St. Dev.
0.16
5
260
0.05
p1
-0.54
-2
318
-0.11
p5
-0.17
0
412
-0.07
p10
-0.07
2
464
-0.06
p25
0.02
6
559
-0.03
p50
0.09
9
657
0.00
p75
0.15
11
811
0.02
p90
0.23
14
1079
0.05
p95
0.29
16
1248
0.06
p99
0.44
18
1664
0.10
Table 1 reports the summary statistics for the US and the European Union (EU). N: number of observations; St.
Dev.: standard deviation; p1 – p99: 1st percentile – 99th percentile.
25
Table 2 Nonlinear relationship between temperature and GDP growth in the US
Model (1)
Model (2)
Coef.
t
Coef.
t
Temperature (., -9)
0.2317
8.52
0.2322
8.49
Temperature (-9, -6)
0.1517
7.65
0.1474
7.01
Temperature (-6, -3)
0.1591
13.09
0.1629
12.05
Temperature (-3, 0)
0.1719
11.15
0.1715
10.31
Temperature (0, 3)
0.1227
6.92
0.1222
6.85
Temperature (3, 6)
0.0303
1.04
0.0300
1.07
Temperature (6, 9)
-0.0445
-2.11
-0.0350
-1.75
Temperature (9, 12)
-0.1239
-4.47
-0.1156
-4.12
Temperature (12, 15)
-0.1524
-4.40
-0.1404
-3.93
Temperature (15, 18)
-0.0776
-2.50
-0.0876
-2.69
Temperature (18, 21)
0.0293
0.84
0.0109
0.30
Temperature (21, .)
-0.0890
-1.53
-0.1031
-1.52
Precipitation (., 200)
0.0012
1.73
Precipitation (200, 300)
-0.0002
-0.98
Precipitation (300, 400)
-0.0009
-4.40
Precipitation (400, 500)
0.0003
1.32
Precipitation (500, 600)
-0.0001
-0.45
Precipitation (600, 700)
-0.0003
-1.36
Precipitation (700, 800)
0.0004
1.40
Precipitation (800, 900)
0.0003
1.48
Precipitation (900, 1000)
0.0000
-0.12
Precipitation (1000, 1100)
0.0001
0.22
Precipitation (1100, 1200)
0.0002
1.03
Precipitation (1200, .)
-0.0002
-3.93
Population growth
R2
0.54
0.56
N
3609
3609
Cell Fixed Effects
Yes
Yes
Time Fixed Effects
Yes
Yes
We estimate three versions of the following equation for the US sample.
Model (3)
Coef.
0.2304
0.1545
0.1378
0.1179
0.1315
0.0065
-0.0530
-0.1375
-0.0978
-0.0844
-0.0076
-0.0942
0.0003
-0.0004
-0.0008
0.0001
0.0001
-0.0003
0.0003
0.0002
0.0001
-0.0001
0.0002
-0.0001
1.8508
t
9.60
7.91
12.96
10.41
9.68
0.29
-3.24
-6.28
-3.58
-3.09
-0.33
-1.87
0.68
-1.87
-4.64
0.50
0.43
-1.43
1.11
0.92
0.63
-0.35
1.15
-2.98
24.08
0.72
3609
Yes
Yes
yit  i  t   mTitm    m Pitm  lit  it
m
m
where cells are indexed by i and (5-year) periods are indexed by t, yit is the GDP growth, i is the cell fixed effects, t
m
m
is the time-fixed effects, and lit is the population growth. Tit ’s are the linear spline of temperature, and Pit ’s are that
of precipitation. We follow Deryugina and Hsiang (2014) and use 3 °C-wide temperature bins. For the US sample, m
is set to 12, and the knots are -9, -6, -3, 0, 3, 6, 9, 12, 15, 18, and 21. That is, the first temperature bin, Temperature
(.,-9), is T < -9 °C, the second one, Temperature (-9,-6), is -9 ≤ T < -6, and so on. All t-statistics are based on clustered
standard errors, which are robust to correlation within cells.
26
Table 2 Nonlinear relationship between temperature and GDP growth in the EU
Model (1)
Model (2)
Coef.
t
Coef.
t
Temperature (., 0)
-0.0315
-1.44
-0.0190
-0.84
Temperature (0, 3)
0.0247
1.18
0.0438
1.63
Temperature (3, 6)
0.1424
1.87
0.1433
1.93
Temperature (6, 9)
-0.0444
-1.11
-0.0603
-1.48
Temperature (9, 12)
-0.1130
-2.66
-0.1324
-3.30
Temperature (12, 15)
-0.0671
-1.37
-0.0593
-1.19
Temperature (15, .)
-0.1493
-5.03
-0.1776
-5.88
Precipitation (., 400)
0.0006
0.77
Precipitation (400, 600)
0.0003
2.41
Precipitation (600, 800)
0.0002
1.13
Precipitation (800, 1000)
0.0002
1.18
Precipitation (1000, 1200)
0.0002
1.06
Precipitation (1200, 1400)
0.0001
0.26
Precipitation (1400, .)
0.0006
1.80
Population growth
R2
0.46
0.47
N
2580
2580
Cell Fixed Effects
Yes
Yes
Time Fixed Effects
Yes
Yes
We estimate three versions of the following equation for the US sample.
Model (3)
Coef.
0.0006
0.0666
0.1525
-0.0526
-0.1337
-0.0742
-0.1877
0.0006
0.0003
0.0002
0.0002
0.0003
0.0001
0.0006
0.8108
t
0.03
2.17
2.12
-1.31
-3.27
-1.38
-5.54
0.77
2.47
1.41
1.22
1.27
0.63
1.90
2.51
0.47
2580
Yes
Yes
yit  i  t   mTitm    m Pitm  lit  it
m
m
where cells are indexed by i and (5-year) periods are indexed by t, yit is the GDP growth, i is the cell fixed effects, t
m
m
is the time-fixed effects, and lit is the population growth. Tit ’s are the linear spline of temperature, and Pit ’s are that
of precipitation. We follow Deryugina and Hsiang (2014) and use 3 °C-wide temperature bins. For the EU sample, m
is set to 7, and the knots are 0, 3, 6, 9, 12, and 15. All t-statistics are based on clustered standard errors, which are
robust to correlation within cells.
27