Professional Documents
Culture Documents
Article 1 (Climate Change)
Article 1 (Climate Change)
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Fondazione Eni Enrico Mattei (FEEM) is collaborating with JSTOR to digitize, preserve and
extend access to this content.
1Fondazione
Eni Enrico Mattei (FEEM)
2Università
Ca’ Foscari Venezia
3Euro-Mediterranean Centre on Climate Change (CMCC)
4Università degli Studi di Milano
September 2015
ABSTRACT. This paper examines the determinants of climate related disasters and attempts to
estimate the presence of adaptive capacity in terms of per capita income and population density elasticities.
We find evidence of adaptive capacity in a “weak” form both in terms of income and population density
elasticities over our entire sample. That is, damages are in fact increasing with income and population, but less
than proportionally. There is also evidence of countries improving their adaptive capacity over the long run, but
of maladaptation occurring in the short run. Repeating the analysis splitting the sample by per-capita income
levels, we find that higher income countries show adaptive capacity in a “strong form”, i.e. damages decrease
with GDP, while lower income countries highlight exactly the opposite behavior. Finally, using Granger causality
tests for panel data, we find evidence of increase in GDP per capita Granger causing climate related damages
for lower income countries, but not in higher income countries.
* [email protected]
† [email protected]
The total estimated damage from climate related disasters between 2011 and 2013 was US$ 641 billion
while the average number of deaths and people affected between 2003 and 2012 were 106,000 and 216
million, respectively. How will these damages and fatalities change in the coming years? Under the theoretical
view point, expectations of both increasing and decreasing trends can be supported. Indeed, on the one hand
development associated to higher population density, physical capital, and ultimately GDP would itself
determine a higher exposure to climate stressors and thus expected damages. This trend can be exacerbated
by climate change that can increase the frequency and intensity of some form of climatological events (IPCC,
2012). On the other hand, development associated to more advanced technologies, knowledge and resource
availability would determine a higher adaptive capacity and thus lower expected damages.
A number of papers have studied the determinants of natural disasters originating a rapidly growing
body of research. The related, mostly econometric literature has focused on the linkages between damages
and different indicators of economic development. The general finding is that developed societies tend to be
less vulnerable; however, the relation between vulnerability and development is not always straightforward
with studies pointing to non-linear or even negative patterns. In a seminal paper Albala-Bertrand (1993) argues
that countries with weaker economies are more affected by climate related disasters. Horwich (2000) also
suggests that the level of wealth is critical for a country’s response to a natural disaster. Using quantile
regression, Keefer et al. (2011) test differential impacts of earthquakes, floods, and tropical cyclones across
various damage bins on disaster propensity. They find that mortality due earthquakes are lower in countries
with high payoffs to mortality prevention and mortality is higher in autocracies and in more corrupted
countries but without controlling for many other relevant socioeconomic variables. Mendelsohn et al. (2012)
found damage elasticity to income to be 0.42 and population density to be -0.20 concluding that both provide
an evidence of adaptation linked to development.
Kellenberg and Mobarak (2008) identify of a nonlinear relationship between economic development
and vulnerability to damages from natural disasters, with risk initially increasing with higher incomes as a result
of changing behaviors. Cavallo et al. (2010) using simple regression techniques conclude that countries with
higher population, land area, and GDP are more exposed and hence suffer higher damages from earthquakes.
Their findings are thus conflicting to the argument that richer countries have more adaptive capacity.
A stream of literature (Kahn, 2005; Skidmore and toya, 2007; Stromberg, 2007; Raschky, 2008; and
3
1
As coded in EM-DAT database.
4
The aim of the paper is to explain climate related economic damages by controlling for socioeconomic
indicators along with measures of risk, institutional capacity, and stability of democratic regimes. As a
conceptual starting point, we assume that countries solve an optimization problem with the goal to minimize
total costs from climate disasters. Without adaptation, for a particular climatological disaster, c, countries’
damages, Dc, can be written as2;
𝛼
𝐷 𝑐 = 𝛼1 ∙ 𝑌 ∙ 𝛼2 ∙ 𝑃𝑜𝑝 ∙ 𝐶𝐷𝑐 3 (1)
According to (1) damages will be higher as income, Y, and population density, Pop, increase. In (1) CDc
represents characteristics of the climatological disaster c. However, adaptation can protect against damages
from disasters. Assuming that adaptation (expressed as the capacity to reduce the damage) also depends
positively on income, population density, an adaptation function can be expressed as;
𝐴 = 𝛾1 ∙ 𝑌 𝛾1 ∙ 𝑃𝑜𝑝𝛾2 (2)
Finally, total damages from climatological disasters can be written as a multiplicative function of the
two functions expressed above;
𝛼
𝑇𝐷𝑐 = 𝐴 ∙ 𝐷𝑐 = 𝛼1 ∙ 𝛾1 ∙ 𝑌1−𝛾1 ∙ 𝛼2 ∙ 𝑃𝑜𝑝1−𝛾2 ∙ 𝐶𝐷𝑐 3 (3)
In (3) the exponents of (2) appear with the inverted sign as adaptation lowers damages. For the
purposes of this paper, we focus on income/wealth and population density as primary determinants of
adaptive capacity. We assume that γ1 = γ2 = 0 in (3) suggests no evidence of adaptation. In other words, if the
damage elasticities of income and population density are equal to 1 then it can be concluded that no
adaptation has taken place. However, if the coefficients of income and population density elasticity are lower
than 1 or negative then there is evidence of adaptation. We call “weak” adaptation the former case with
damages increasing, but less than proportionally with income and population density, and “strong” adaptation
the latter with damages decreasing with income and population density. On the other hand, if the elasticities
are greater than 1, meaning that increases in income and population density will lead to more than
proportional increases in damages from climate related disasters then we have evidence of maladaptation.
2
Adapted from Schumacher and Strobl (2011).
5
3
(EXPORT+IMPORT)/GDP
7
Table 1 reports the result of the simpler fixed effect models without lagged variables among the
explanatory. Statistically significant income elasticities turn out to be robust across the models, with values
ranging from 0.581 to 0.739. This suggests that as income of countries increase, damages from climatological
disasters do not increase proportionally. Thus in the overall dataset, there is evidence of what we call “weak”
adaptive capacity among countries with respect to income. It is likely that as a country’s income increases, it is
able to allocate greater funding towards adapting ex ante and ex post to climatological damages, e.g.
constructing dams to protect against flood damages and rehabilitation of population living in coastal areas.
Nonetheless, it is not possible to offset completely the higher exposure given by the more assets at risk.
In Table 1 in columns 2 and 3, coefficients for log of population density are also statistically significantly
positive, and lower than one. This further provides evidence of adaptation and suggests that an increase in
population density does not bring about a proportionate increase in climatological damages. This could be due
to greater precautions being taken in densely populated and highly urbanized areas and governments
undertaking more programs and projects to protect against climatological disasters.
Column 3 also shows, in accordance with intuition, that the larger the agricultural share in GDP, the
higher the damages from climatological disasters. This is a consequence of a higher share of the economy
particularly sensitive to natural disasters such as floods and droughts. The polity variable is negative and
statistically significant, suggesting that stability and democratic qualities of regimes plays an important role in
helping to reduce damages from climatological disasters. The same holds for the openness index which
suggests that well developed trade relationships can help a country to smooth negative consequences of
climate disasters. This can happen through an easier substitution of foreign demand and supply for the
domestic ones should the latter be negatively affected by climate impacts and also, more directly, through an
easier access to foreign aids. However, the coefficients for the share of government expenditure of GDP and
risk are not statistically significant.
In this section we explore the impact of lagged covariates on climatological damages and attempt to
highlight any differential impact between the short and the long run. We also include temperature and
precipitation as indicators of climatic exposure variables. The first striking difference with the static analysis
reported in Table 1 is that the coefficient for contemporaneous income elasticity in Table 2 is greater than one
and statistically significant, suggesting evidence of maladaptation in the short-run. However, the elasticity is
lower than one in the long-run. This seems to point out that adaptive capacity of countries does improve with
respect to income, but that this is a process requiring time. This finding is strengthened by the coefficient for
population density elasticity; it is just slightly lower than one in the short-run, but much lower than one in the
long-run, providing further evidence that adaptive capacity is a dynamic process. In the specific case of
population this could indicate more protective adaptation and development programs being undertaken over
Results in Table 2 also show that government expenditure is positive and significant only in the long
run. This outcome, similar to that of Toya and Skidmore (2007) seems to point out that countries with higher
share of public expenditure over GDP tend to dynamically experience higher climate related losses. This can be
explained as a signal of lower efficiency of the public vs the private sector. In our sample this behavior can be
driven by the presence of developing countries more easily characterized by ineffective relief efforts and
corruption in government programs. Similarly, but with opposite effect, regime characteristics: democratic
10
It is critical to examine whether the results from the above specifications are robust across different
income levels. In this regard, we use country and year fixed effects along with standard errors clustered by
country in a further set of regressions. Results are reported in Table (3). The major message conveyed by the
analysis suggests, somewhat similarly to Kellenberg and Mobarak (2008), a bell-shaped relation between
damage and per capita income, or, differently said, evidence of insufficient adaptive capacity, if not
maladaptation, (at) in low income (levels) countries and efficient “strong” adaptive capacity (at) in high income
(levels) countries. The GDP per capita elasticities for the low income countries are indeed positive and larger
than one (ranging from 1.701 to 1.960). This could be interpreted as the dominance, at low per-capita income
levels, of the “assets-at-risk” or “exposure” effects over adaptive capacity. Accordingly, increasing GDP raises
vulnerability. At the higher income levels, the relation reverses: per capita GDP coefficient is larger than one,
but negative. Thus more GDP reduces damages highlighting that with development more resources are
available for investment in resilient infrastructure and disaster preparedness, the factors most likely influencing
adaptive capacity, eventually overcoming the “assets-at-risk” effect.
11
The analysis by income bins also confirms that larger shares of government expenditure on GDP have a
positive relationship with damages. Interestingly, this is not a characteristic of low income countries (even
though the related coefficient is considerably higher for particularly low income countries) but a generalized
outcome. This result provides some evidence that larger public intervention in the economy may introduce
inefficiency also in the action against disaster related damages.
The risk variable is positive and statistically significant for all the income bins; suggesting that the
higher the ratio of those killed to those affected, the greater the impact on climate damages for countries. This
is particularly troublesome for relatively poorer countries which have larger population exposed to natural
disasters. It is worth recalling that in Table 2 the coefficient for this variable was not statistically significant. One
particular reason for this could be the fact that we have fewer observations in the regressions segmented by
income class which could be influencing the standard errors. The coefficients for the agricultural share of GDP
are also no longer statistically significant.
Increased openness seems to reduce damages from climate disasters across all income levels, further
reiterating that increased integration with international trade can be an important damage smoothing
component. Estimated results also show that regimes with more democratic qualities have a negative impact
12
The analysis conducted in the previous sections highlights indirectly that development, measured by
per capita GDP is a driver of adaptive capacity measured by lower damages at least in developed regions. In
this section we dig deeper in this relationship. As already noted, it can be argued that the causal relationship
between GDP per capita and climate related damages is bidirectional; i.e. damages could determine
subsequent income or income could determine subsequent damages. It is also possible to have no
interdependency between these two variables.
We test explicitly this link verifying whether per capita GDP Granger causes (lower) damages, if it does
then there is further evidence of adaptation. In this case we are not assuming the exogeneity or endogeneity of
the underlying variables a priori. GDP per capita Granger causes climate related damages if the lagged GDP per
capita helps forecast climate related damages. The standard Granger causality test for time series data is not
applicable in the case of panel data and needs to be constructed ad hoc. We follow the procedure of Hurlin and
Venet, (2001), Hurlin, (2005, 2007) allowing more efficiency, control for individual heterogeneity, a reduction in
identification problems, modelling temporal effects without aggregation bias present in time series data, and
the usual increased robustness due to the use of panel data (Greene, 2008 and Baltagi, 2005).
The Hurlin heterogeneous non-causality hypothesis (HENC) test, verifies the existence of a causal
relation in the Granger sense from the variable x to y in some or at least one country in the panel even though
the relation is no homogenous across the entire panel.
The linear panel data model can be written as:
(𝑘) (𝑘)
𝑦𝑖𝑡 = 𝛼𝑖 + ∑𝐾
𝑘=1 𝛾𝑖 𝑦𝑖𝑡−𝑘 + ∑𝐾
𝑘=1 𝛽𝑖 𝑥𝑖𝑡−𝑘 + 𝜖𝑖𝑡 (5)
(𝑘) (𝑘)
Where 𝛾𝑖 and 𝛽𝑖 are various coefficients of 𝑦𝑖𝑡−𝑘 and 𝑥𝑖𝑡−𝑘 for country i.
We test the causal patterns for four income groups and the coefficients of each group are tested
against the null hypothesis of zero causality i.e. β=0. The FHENC test statistic in practice amounts to build an F-
test where the sum of squared residuals (SSR) from this restricted regression (SSR2ij) is then compared to those
from the unrestricted model (SSR1);
(𝑆𝑆𝑅2,𝑗 −𝑆𝑆𝑅1 )/(𝑛𝑛𝑐 𝑝)
𝐹𝐻𝐸𝑁𝐶 = 𝑆𝑆𝑅 (6)
1 /[𝑁𝑇−𝑁(1+𝑝)−𝑛𝑐 𝑝]
In (6) nnc is the subset of countries for which β is restricted to zero, while nc is the subset of countries
13
Hence, we differenced the data for this purpose. Then, in order to conduct the panel Granger non-
causality test by Hurlin (2004 and 2005), the requirement is Ti > 5 + 2K (Ti being the time span for country i and
K represents the autoregressive lag orders). The panels in our dataset are often relatively short, which limits
the degrees of freedom. As a result the analysis is based on K = 1. Table (5) shows the results from the panel
Granger non-causality test between the first-differences of log of GDP per capita and log of climate damages.
14
We find no evidence of changes in GDP per capita Granger causing higher climate damages for the
entire panel. However, for countries with GDP per capita below $5,000 increasing GDP per capita seems to
Granger-cause higher climate related damages. In other words, for countries at the earlier stage of
development, GDP per capita is a driver (in a Granger sense) of damages (and not vice versa). However, the link
goes in the opposite direction that one would expect, as growth brings about more damages. This is further
evidence of maladaptation or of increased disaster vulnerability at earlier stages of economic development
which is brought about by more assets (and population) at risk not yet compensated by sufficient investment in
vulnerability reduction. Echoing one of the theoretical justifications of the Environmental Kuznets Curve-like
behavior (see e.g. Ruttan, 1971; Antle and Heidebrink, 1995), it could be argued that protection against
weather related disasters is in a way a superior good which enters societal preferences only after more basic
needs are fulfilled. For the panel related with income above $15,000, the test statistics are negative, but not
statistically significant, suggesting that there is no evidence of increasing GDP per capita Granger causing a
reduction in climate related damages. While our regression results suggest that higher income countries are
better at improving their adaptive capacity over time, this may not be enough to significantly reduce their
monetary damages due to climatological disasters. None of the test statistics in column 2 are statistically
significant, suggesting that higher damages do not Granger cause the GDP per capita.
4. Conclusions
In this paper we tried to detect the existence of ongoing adaptation or of the working of adaptive
capacity against climate related damages examining a panel combining climate disaster, socioeconomic, and
climatic data between 1980 and 2012 for 104 countries. When the analysis is conducted over the entire data
15
16
Albala-Bertrand, J. (1993). Political Economy of Large Natural Disasters. Oxford University Press Inc., New
York.
Angrist, Joshua D., and Jorn-Steffen Pischke. (2009). Mostly Harmless Econometrics: An Empiricists
Companion. Princeton: Princeton University Press.
Antle, J.M. and G. Heidebrink. (1995). Environment and Development: Theory and International Evidence.
Economic Development and Cultural Change. 43(3), 603-625.
Baltagi, B.H. (2005), Econometric Analysis of Panel Data, New York: Wiley.
Besley, T., and Burgess, R. (2002). The political economy of government responsiveness: Theory and evidence
from India. Quarterly Journal of Economics, 117, 4: 1415–1451.
Cavallo, E., A. Powell and O. Becerra (2010), Estimating the Direct Economic Damage of the Earthquake in Haiti.
The Economic Journal 546(120): F298–F312,
Eisensee, T., and Stromberg, D. (2007). News floods, news droughts, and U.S. disaster relief. Quarterly Journal
of Economics, 122, 2.
Greene, W. H. (2008), Econometric Analysis, 6. ed, Pearson Prentice Hall: Upper Saddle River, New York.
Healy, Andrew, and Neil Malhotra. (2009). “Myopic Voters and Natural Disaster Policy." American Political
Science Review. 103(3): 387-406.
Horwich, G. (2000), Economic Lessons of the Kobe Earthquake, Economic Development and Cultural Change,
Vo. 48, pp. 521-42.
Hurlin, C. and Venet, B. (2001) “Granger causality tests in panel data models with fixed coefficients”, Eurisco
working paper no. 2001-09, University of Paris-Dauphine.
Hurlin, C. (2004a), “Testing Granger Causality in Heterogeneous Panel Data Models with Fixed Coefficients”,
Miméo, University Orléans.
Hurlin, C. (2005). Granger causality tests in panel data models with fixed coefficients. Revue Economique, 56, 1-
11.
Hurlin, C. (2007). "Testing Granger Non-Causality in Heterogeneous Panel Data Models with Fixed Coefficients,"
Working Papers 1547, Orleans Economic Laboratorys, University of Orleans.
IPCC (2012), “Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation. A
Special Report of Working Groups I and II of the Intergovernmental Panel on Climate Change [Field,
C.B., V. Barros, T.F. Stocker, D. Qin, D.J. Dokken, K.L. Ebi, M.D. Mastrandrea, K.J. Mach, G.-K. Plattner,
S.K. Allen, M. Tignor, and P.M. Midgley (eds.)]. Cambridge University Press, Cambridge, UK, and New
York, NY, USA, 582 pp.
Kahn, M. E. (2005). The death toll from natural disasters: The role of income, geography, and institutions. The
Review of Economics and Statistics 87, 2, 27-284.
Keefer, Philip and Neumayer, Eric and Plumper, Thomas (2011) Earthquake propensity and the politics of
17
19