Developing countries are arguably the most affected by the increasing regularity and costs of natural disasters. In addition to the often impressive casualty ...
Issue 1 - 2012
Coping with loss: the impact of natural disasters on developing countries' trade flows Jorge Andrade da Silva and Lucian Cernat* * The views expressed in this document are the authors' and do not necessarily reflect those of the European Commission.
Abstract Developing countries are arguably the most affected by the increasing regularity and costs of natural disasters. In addition to the often impressive casualty figures, there can be systemic trade and development implications as well. Results from a simple model suggest that exports of affected small developing countries decline by 22% (whereas exports of larger developing countries are not significantly affected), and that such effects tend to last for about 3 years.
Introduction Increasing regularity and costs of natural disasters
The latest Annual Disaster Statistical Review (CRED 2011) reports a total of 385 natural disasters in 2010, with associated casualties surpassing 297 thousand people and costs estimated at around 95 billion Euros. The underlying increase in the regularity and costs of disasters (Munich Re 2006, UNEP 2005) justifies the growing interest in their trade impact. Indeed, direct potential effects include casualties, but also physical damages to infrastructure and disruptions to supply chains, international trade and economic activity.
Focus on small developing countries
In the current exercise we employ a simple gravity model to try to infer the impact of natural disasters on countries' exports. Particularly, we take into account two salient aspects in the literature: the focus on developing countries, as the level of development is arguably important in determining the losses from natural disasters (Peduzzi et al. 2002)i; and country size, as it is often indicated that small countries are most heavily impacted in economic terms, as well as being less capable to overcome the adverse effects without outside assistance (Lee and Logez 2005).
Editor: Lucian Cernat For further information:
http://ec.europa.eu/tra de/analysis/chiefeconomist/ ISSN 2034-9815
The Data In constructing our natural disaster dataset we take into account earthquakes, floods and volcanic eruptions from the EM-DAT database, the only publicly available global database on the subject (Peduzzi et al. 2009). We use the number of casualties for the identification of natural disasters. Furthermore,
Data employed concern earthquakes, floods and volcanic eruptions
and to avoid known measurement problems, we define the disaster variable as a combination of both absolute and relative measures . Concretely, we consider that a natural disaster has taken place if, in any given year and for any given country, there were either: floods or earthquakes causing 500 or more casualties, or volcanoes causing at least one casualty; or floods or earthquakes causing a minimum of 1 casualty per 100 000 inhabitants . ii
iii
Data for GDP and Exports are taken from the IMF, and distance and other (historical and cultural) control variables from CEPII. The latter include dummies for contiguous countries, countries with a common official language, and countries with colonial links. The distance variable refers to geodesic distances as calculated by CEPII on the basis of latitudes and longitudes of the most populous agglomerations. In defining a small country dummy variable, we use a 20 million population as threshold. According to the UN (2004), countries with populations under 20 million account for only 11.6 per cent of the world population, even if accounting for 77.2 per cent of countries. To be clear, the 20 million threshold is not meant as a policy variable or rule. We intend to apprehend from it what happens to the trade impact of disasters when we look at countries that are smaller in (population) size, but still big enough in order for them to figure in our dataset.
Estimation Study employs a simple gravity equation
In a first model we try to identify whether there is evidence of impact from disasters on the exports of affected developing countries taken as a whole. In the second model, we highlight the case of small developing countries. We employ a simple gravity regression of the form: Ln ( EXPij ,t ) = α + β .Ln (GDP i ,t ) + γ .Ln (GDP j ,t ) + δ .DISASTER
i ,t
+
∑ ϕ .Control
ij
+
+ ∑φ.Quartert + ∑ κ .Exporteri + ∑ χ .importerj + ε
Where EXP is country i’s (USD) Exports to country j, GDP is country i’s (USD) GDP. Set i of countries is composed of GSP countries for which we find quarterly GDP and trade data, and j are the OECD countries (with EU27 as an aggregate) . Index t refers to the 92 quarters in the sample, 1988Q1 through 2010Q4. We include time specific fixed effects (Quarter ), as well as exporter and importer fixed effects (respectively Exporter and Importer ) . We use STATA’s panel command xtreg. ij
i
iv
t
v
i
j
Results We see that all main variables included in model 1 (Table 1, columns 1 and 2) show the expected sign and are significant. Namely, the GDP of countries involved in trade contribute to higher export levels, while distance affects them negatively. The significant control variables also give a positive contribution to exports, as expected. Our disaster variable suggests that the average GSP exporter will have its exports decreased by 9% as a result of the disaster . vi
TABLE 1. Estimation results Random-effects GLS regression Dependent variable: Ln(EXPij) Model 1 (1)
% impact (2)
Model 2 (3)
% impact (4)
Main variables: Ln(GDPi) Ln(GDPj) Distance Disaster Disaster*small
0.41 0.83 0.00 -0.09
*** *** *** ***
0.41 0.83 -0.02 -9.01
0.42 0.83 0.00 -0.04 -0.25
*** *** ***
0.42 0.83 -0.02
***
-22.14
Control variables:
Exports of affected small developing countries decline on average by 22%
Exports of larger developing countries are not significantly affected
Effects tend to last for about 3 years
Small country Contiguous country Common language Colony Same country Nr obs Nr groups
2.12 *** 0.90 *** 1.67 *** -1.08 39360 531
-2.29 2.12 0.90 1.67 -1.08
*** *** *** ***
39360 531
Note 1: the % impact of disaster dummy variables is computed according to Kennedy (1981) Note 2: *** stands for significance at the 1% level
In model 2 (columns 3 and 4) we investigate whether the impact of natural disasters on exports of small countries is significantly different from that of the average GSP country. The interesting aspect of this preferred model is that, whereas the remaining variables continue to show as previously estimated, the introduction of the Disaster*small interaction dummy in the analysis suggests that the previously observed impact on the average developing country was in fact driven by the small (developing) countries in the sample . This result is in line with the findings of Gassebner et al. (2006), who conclude that the smaller and less democratic a country is, the more are its trade flows reduced in case it is affected by a disaster. The result also suggests that exports of larger developing countries are not significantly affected by the disasters in the sample. We perform robustness checks on the disasters' time span and on the aggregation of the disaster variables included in model 2. The disasters' time span, i.e. the period in which we try to capture the impact of the disaster, is here allowed to vary: we depart from the previously used 1 year (4 quarters) duration, rerunning the same model specifications for 2 and 3 years (8 and 12 quarters). The results remain unaffected. This partial evidence for a particular impact duration is supported by additional regressions (not reported but available) where we simultaneously specify disaster dummies by year (one for the first year after the disaster, one for the second, and so on), and which indicate that the negative impact for small developing countries is on average felt for around 3 years. The second robustness check is performed by running regressions for the different types of disasters separately. Concretely, we substitute, in different regressions, the combined disaster variable with, in turn, the flood, earthquake and volcano variables. Given the smaller number of disasters in any each regression, the disaster results do not always show as significant in this case. When they do, nevertheless, they support the same conclusions. Results also vii
showed however, less robustness when certain variations of the Poisson estimator were employed, in alternative to Ordinary Least Squares.
Conclusions Our results suggest that natural disasters impact negatively on the exports of small developing countries, which decline on average by 22%. A second result is that the negative effects of natural disasters on the export performance of small developing countries are estimated to last for about 3 years.
The Commission's Trade, growth and development communication
Given that small developing countries seem to be at higher risk of having their exports negatively affected by natural disasters, such countries could be the focus of measures dedicated at either reducing their export vulnerability to disasters in the first place, or at minimizing the negative export impact of disasters when they occur. In the recent European Commission's Communication on Trade, growth and development (EC 2012), a number of ways are proposed to improve the effectiveness of the EU in regards to the preparedness for natural disasters and the use of trade measures to help mitigate the effects of natural disasters.
TABLE A1. GSP exporter countries in the sample Country
ARGENTINA ARMENIA AZERBAIJAN BELARUS BOLIVIA BRAZIL BRUNEI DARUSSALAM China COLOMBIA COSTA RICA ECUADOR EGYPT GEORGIA GHANA GUATEMALA INDONESIA INDIA Iran JAMAICA JORDAN KAsAKHSTAN KENYA KYRGhiSTAN CAMBODIA MACAO (Special Administrative Region of China) MOROCCO MEXICO Mozambique MAURITIUS MALAYSIA PERU PHILIPPINES RUSSIA EL SALVADOR West. Samoa THAILAND TUNISIA TANZANIA UKRAINE Venezuela SOUTH AFRICA
With Small Disaster(s) Country
X X
X X X X X X
X X X X
X X X X X X X X
X X X
X X X X X X X X X X
X X X
X
Note: Highlighted cells indicate small countries with disasters; the total of disasters in the sample amounts to 97; over 50% are floods, the remaining are earthquakes and vulcanic eruptions
X X X X
References CRED (2011), "Annual Disaster Statistical Review 2010", Centre for Research on the Epidemiology of Disasters, Université catholique de Louvain EC (2012), "Communication from the European Commission to the European Parliament, the Council and the European Economic and Social Committee; Trade, growth and development", http://trade.ec.europa.eu/doclib/docs/2012/january/tradoc_148992.pdf Gassebner, Keck and The (2006), "The Impact of disasters on international trade", WTO Staff Working Paper ERSD-2006-04 March 2006 Lee and Logez (2005), “Trade Interests of the Tsunami Affected Countries”, OECD Trade Policy Working Papers, No. 23, OECD Publishing Munich Re (2006) "Annual Review: Natural Catastrophes in 2005", Knowledge series / Topics Geo, Munich Re Group Peduzzi, Hao, Herold (2002), "Global risk and vulnerability index, Phase II: development, analysis and results", UNDP/BCPR Peduzzi, Hao, Herold, Mouton (2009), "Assessing global exposure and vulnerability towards natural hazards: the Disaster Risk Index", Natural hazards and earth system sciences, 9, 1149-1159 Shepherd (2008), "Notes on the Theoretical Gravity Model of International Trade", Niehaus Center, Princeton University & GEM, Sciences Po UN (2004), "World Population Prospects, The 2004 Revision, Volume III, Analytical Report", Department of Economic and Social Affairs, Population Division UNEP (2005), "Environmental Management and Disaster Reduction", Concept Paper for the World Conference on Disaster Reduction, Japan 2005 In Peduzzi et al. (2002), similar exposure to hazards results in different casualties depending on the level of development, with the most developed countries representing 15% of human exposure to hazards, but accounting for only 1.8% of all victims. As forwarded by Peduzzi et al. (2009), if on the one hand taking absolute casualty figures picks up more disasters on the most populous countries (say, China, India), on the other hand focusing on relative figures (e.g. the number of casualties per 100 000 inhabitants) will identify more disasters in small countries. Munich Re (2006) considers disasters totalling more than 500 casualties (and with damage in excess of USD 500 million) as "devastating catastrophes". Due to limitations in data availability, we focus on the casualty numbers alone. Additionally, we try to capture the distinct character of natural disasters caused by volcanic eruptions, regarding the rapidity of onset and predictability aspects, by introducing an (arguably discretionary) lower threshold for the identification of volcano related disasters. In our sample, this OECD / EU27 group accounts for over half of GSP countries' total exports. Table A1 in the annex provides an overview of the developing countries in the sample. In doing this, we follow Shepherd (2008). Together, these fixed effects amount to a total of 92+41+13 control dummies (not reported), and should help us control for changes in trade policy over time. To compute the estimated percentage change in the dependent variable, we refer to Kennedy (1981), where the author shows that the consistent estimator is pˆ j = [exp( cˆ j ) / exp( 0 .5Vˆ ( cˆ j ))] − 1 , where C^j is the OLS estimator of Cj and V^ (C^j) its estimated variance. We control for the small country status with an extra dummy, which is significant and has the expected sign. i
ii
iii
iv
v
vi
vii
