The Growth Aftermath of Natural Disasters

Similar documents
1 Basic concepts for quantitative policy analysis

International Trade and California Employment: Some Statistical Tests

Volume 30, Issue 4. Who likes circus animals?

Evaluating the statistical power of goodness-of-fit tests for health and medicine survey data

Do not turn over until you are told to do so by the Invigilator.

Impact of Internet Technology on Economic Growth in South Asia with Special Reference to Pakistan

WISE 2004 Extended Abstract

An Empirical Study about the Marketization Degree of Labor Market from the Perspective of Wage Determination Mechanism

The link between immigration and trade in Spain

emissions in the Indonesian manufacturing sector Rislima F. Sitompul and Anthony D. Owen

CONSUMER PRICE INDEX METHODOLOGY (Updated February 2018)

A Longer Tail?: Estimating The Shape of Amazon s Sales Distribution Curve in Erik Brynjolfsson, Yu (Jeffrey) Hu, Michael D.

Key Words: dairy; profitability; rbst; recombinant bovine Somatotropin.

The ranks of Indonesian and Japanese industrial sectors: A further study

6.4 PASSIVE TRACER DISPERSION OVER A REGULAR ARRAY OF CUBES USING CFD SIMULATIONS

Extended Abstract for WISE 2005: Workshop on Information Systems and Economics

Development and production of an Aggregated SPPI. Final Technical Implementation Report

Innovation in Portugal:

Volume 29, Issue 2. How do firms interpret a job loss? Evidence from the National Longitudinal Survey of Youth

A Two-Echelon Inventory Model for Single-Vender and Multi-Buyer System Through Common Replenishment Epochs

Introducing income distribution to the Linder hypothesis

Market Dynamics and Productivity in Japanese Retail Industry in the late 1990s

Firm Performance and Foreign Direct Investment: Evidence from Transition Economies. Abstract. Department of Economic, University of Texas at Arlington

Labour Demand Elasticities in Manufacturing Sector in Kenya

Optimal Issuing Policies for Substitutable Fresh Agricultural Products under Equal Ordering Policy

Saving Investment Correlation in South Asia- A Panel Approach

PRODUCTIVE PUBLIC EXPENDITURE AND IMPERFECT COMPETITION WITH ENDOGENOUS PRICE MARKUP: COMMENT

Hysteresis in Regional Unemployment Rates in Turkey

Sources of information

Supplier selection and evaluation using multicriteria decision analysis

Numerical Analysis about Urban Climate Change by Urbanization in Shanghai

Appendix 6.1 The least-cost theorem and pollution control

Analyses Based on Combining Similar Information from Multiple Surveys

Do Farm Programs Explain Mean and Variance of Technical Efficiency? Stochastic Frontier Analysis

LIFE CYCLE ENVIRONMENTAL IMPACTS ASSESSMENT FOR RESIDENTIAL BUILDINGS IN CHINA

The Effect of Outsourcing on the Change of Wage Share

International Trade and California s Economy: Summary of the Data

Consumption capability analysis for Micro-blog users based on data mining

Experiments with Protocols for Service Negotiation

Field Burning of Crop Residues

RULEBOOK on the manner of determining environmental flow of surface water

A NONPARAMETRIC APPROACH TO SHORT-RUN PRODUCTION ANALYSIS IN A DYNAMIC CONTEXT. Elvira Silva *

Stay Out of My Forum! Evaluating Firm Involvement in Online Ratings Communities Neveen Awad and Hila Etzion

To manage leave, meeting institutional requirements and treating individual staff members fairly and consistently.

Prediction algorithm for users Retweet Times

Spatial difference of regional carbon emissions in China

CHICKEN AND EGG? INTERPLAY BETWEEN MUSIC BLOG BUZZ AND ALBUM SALES

Household Budget and Calorie Consume of Livestock Products: Evidence from Indonesia SUMMARY

Direct payments, spatial competition and farm survival in Norway

MULTIPLE FACILITY LOCATION ANALYSIS PROBLEM WITH WEIGHTED EUCLIDEAN DISTANCE. Dileep R. Sule and Anuj A. Davalbhakta Louisiana Tech University

Driving Factors of SO 2 Emissions in 13 Cities, Jiangsu, China

Bulletin of Energy Economics.

A SIMULATION STUDY OF QUALITY INDEX IN MACHINE-COMPONF~T GROUPING

The Spatial Equilibrium Monopoly Models of the Steamcoal Market

Evaluating The Performance Of Refrigerant Flow Distributors

Economics Discussion Paper Series EDP-1308

Evaluation Method for Enterprises EPR Project Risks

Overeducation in Cyprus

The Labor Market Impacts of. Adult Education and Training in Canada

Problem Set 4 Outline of Answers

Impact of public research on industrial innovation

Influencing Factors and Evaluation Index of Farmers Financial Needs based on Analytic Hierarchy Process

Biomass Energy Use, Price Changes and Imperfect Labor Market in Rural China: An Agricultural Household Model-Based Analysis.

Development trajectory of energy consumption and carbon emissions in developing countries

Regression model for heat consumption monitoring and forecasting

Guidelines on Disclosure of CO 2 Emissions from Transportation & Distribution

Econometric Methods for Estimating ENERGY STAR Impacts in the Commercial Building Sector

Implementing Activity-Based Modeling Approach for Analyzing Rail Passengers Travel Behavior

The Substitutability of Labor of Selected Ethnic Groups in the US Labor Market

A Reevaluation of the Effect of Human Capital Accumulation on Economic Growth: Using Natural Disasters as an Instrument

Comparative Advantage, Information and the Allocation of. Workers to Tasks: Evidence from an Agricultural Labor Market. Andrew D. Foster.

Willingness to Pay for Beef Quality Attributes: Combining Mixed Logit and Latent Segmentation Approach

An Analysis of Auction Volume and Market Competition for the Coastal Forest Regions in British Columbia

The Effects of Incomplete Employee Wage Information: A Cross-Country Analysis. Solomon W. Polachek. and. Jun (Jeff) Xiang *

The Role of Price Floor in a Differentiated Product Retail Market

Profit Persistence in the Food Industry: Evidence from five European Countries

Gender Wage Differences in the Czech Public Sector: A Micro-level Case

The Impact of CO 2 Emission Cuts on Income

EVALUATING THE PERFORMANCE OF SUPPLY CHAIN SIMULATIONS WITH TRADEOFFS BETWEEN MULITPLE OBJECTIVES. Pattita Suwanruji S. T. Enns

Education and competence mismatches: job satisfaction consequences for workers

Relative income and the WTP for public goods

INTANGIBLE ASSETS AND HUMAN CAPITAL IN MANUFACTURING FIRMS

Experimental Validation of a Suspension Rig for Analyzing Road-induced Noise

PRICE VOLATILITY INFLUENCE ON AGRICULTURAL INCOME INSTABILITY. Dmytro Bilodid

The Long-Term Effects of Price Promotions on Category Incidence, Brand Choice and Purchase Quantity. Koen Pauwels. Dominique M.

Early warning models of financial distress. Case study of the Romanian firms listed on RASDAQ

Documento de Trabajo No. 01/00 Marzo Wage Differentials Between the Formal and the Informal Sector in Urban Bolivia.

ICT Intermediates, Growth and Productivity Spillovers Evidence from Comparison of Growth Effects in German and US Manufacturing Sectors

CENTRE FOR ECONOMIC PERFORMANCE DISCUSSION PAPER NO. 219 DECEMBER 1994 WAGES, EFFORT AND PRODUCTIVITY S. NICKELL AND D. NICOLITSAS

THE STUDY OF GLOBAL LAND SUITABILITY EVALUATION: A CASE OF POTENTIAL PRODUCTIVITY ESTIMATION FOR WHEAT

The Employment Effects of Low-Wage Subsidies

Discussion Papers No. 258, August 1999 Statistics Norway, Research Department

Managing Investigations Guidance Notes for Managers

DOES THE DOMESTIC INVESTMENT BENEFIT FROM THE INFLOWS AND OUTFLOWS OF FDI? PANEL DATA EVIDENCE FROM THE ASEAN-8 CONTRIES

Do Remittances Alter Labor Market Participation? A Study of Albania

EUROPEAN CONGRESS OF THE REGIONAL SCIENCE ASSOCIATION VOLOS- 2006

Welfare Gains under Tradable CO 2 Permits * Larry Karp and Xuemei Liu

Small Broadband Providers and Federal Assistance Programs: Solving the Digital Divide?

An Empirical Analysis of Search Engine Advertising:Sponsored Search in Electronic Markets 1

Economic Efficiency and Factors Explaining Differences. Between Minnesota Farm Households

Transcription:

WPS5002 Polcy Research Workng Paper 5002 The Growth Aftermath of Natural Dsasters Thomas Fomby Yuk Ikeda Norman Loayza The World Bank Development Research Group & Global Faclty for Dsaster Reducton and Recovery July 2009

Polcy Research Workng Paper 5002 Abstract Ths paper provdes a descrpton of the macroeconomc aftermath of natural dsasters. It traces the yearly response of gross domestc product growth both aggregated and dsaggregated nto ts agrcultural and non-agrcultural components to four types of natural dsasters droughts, floods, earthquakes, and storms. The paper uses a methodologcal approach based on poolng the experences of varous countres over tme. It conssts of vector auto-regressons n the presence of endogenous varables and exogenous shocks (VARX), appled to a panel of cross-country and tme-seres data. The analyss fnds heterogeneous effects on a varety of dmensons. Frst, the effects of natural dsasters are stronger, for better or worse, on developng than on rch countres. Second, whle the mpact of some natural dsasters can be benefcal when they are of moderate ntensty, severe dsasters never have postve effects. Thrd, not all natural dsasters are alke n terms of the growth response they nduce, and, perhaps surprsngly, some can ental benefts regardng economc growth. Thus, droughts have a negatve effect on both agrcultural and non-agrcultural growth. In contrast, floods tend to have a postve effect on economc growth n both major sectors. Earthquakes have a negatve effect on agrcultural growth but a postve one on non-agrcultural growth. Storms tend to have a negatve effect on gross domestc product growth but the effect s short-lved and small. Future research should concentrate on explorng the mechansms behnd these heterogeneous mpacts. Ths paper a product of the Development Research Group and the Global Faclty for Dsaster Reducton and Recovery s part of a larger effort n the department to study the man sources of vulnerablty and to dssemnate the emergng fndngs of the forthcomng jont World Bank-UN Assessment of the Economcs of Dsaster Rsk Reducton. Polcy Research Workng Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at ykeda1@ worldbank.org and nloayza@worldbank.org. The Polcy Research Workng Paper Seres dssemnates the fndngs of work n progress to encourage the exchange of deas about development ssues. An objectve of the seres s to get the fndngs out quckly, even f the presentatons are less than fully polshed. The papers carry the names of the authors and should be cted accordngly. The fndngs, nterpretatons, and conclusons expressed n ths paper are entrely those of the authors. They do not necessarly represent the vews of the Internatonal Bank for Reconstructon and Development/World Bank and ts afflated organzatons, or those of the Executve Drectors of the World Bank or the governments they represent. Produced by the Research Support Team

The Growth Aftermath of Natural Dsasters * Thomas Fomby Yuk Ikeda Norman Loayza Southern Methodst Unversty Georgetown Unversty The World Bank JEL Classfcaton: O11, O40, Q54 Key Words: Natural dsasters, economc growth, sectoral value added * We are grateful to Apurva Sangh, S. Ramachandran, Jamele Rgoln, Eduardo Olaberría, Claudo Raddatz and semnar partcpants at the World Bank for valuable comments, suggestons, and advce. Tomoko Wada provded excellent research assstance. Ths paper was commssoned by the Jont World Bank - UN Project on the Economcs of Dsaster Rsk Reducton. Partal fundng of ths work by the Global Faclty for Dsaster Reducton and Recovery s gratefully acknowledged. The fndngs, nterpretatons, and conclusons expressed n ths paper are entrely those of the authors, They do not necessarly represent the vews of the Internatonal Bank for Reconstructon and Development/World Bank and ts afflated organzatons, or those of the Executve Drectors of the World Bank or the governments they represent. 1

I. Introducton Ths paper provdes a descrpton of the macroeconomc aftermath of natural dsasters, specfcally tracng the economc growth response n the wake of these events. Its purpose s to contrbute to the analyss of the path of adjustment and recovery by tracng the yearly response of GDP growth --both aggregated and dsaggregated nto ts agrcultural and non-agrcultural components-- to four types of natural dsasters -- droughts, floods, earthquakes, and storms. As has been shown n recent papers (see, for nstance, Loayza, Olaberría, Rgoln, and Chrstaensen 2009), the analyss by sector of economc actvty and by type of natural dsaster s crucal to measure and nterpret ts complex effects on the economy. Apart from ths dsaggregated analyss, ths paper has four other features that set t apart. Frst, t traces the growth response n every year of and after the event. Ths focus on the annual frequency s necessary to characterze the detals of the adjustment path, rather than only explanng ts net permanent effect. For nstance, t s concevable that, say, an earthquake has no long-run consequences on economc growth whle havng a growth path of declne followed by recovery, whose characterzaton would be of nterest for the present analyss. Second, the paper uses a methodologcal approach based on poolng the experences of varous countres over tme to arrve at mean responses of growth to natural dsasters. Whle losng country specfcty, the methodology allows descrbng basc patterns n a sensble and robust manner. The econometrc methodology conssts of vector auto-regressons n the presence of endogenous varables and exogenous shocks, appled to panel, cross-country and tme-seres, data (for short, the methodology s descrbed as panel VARX). The full sample conssts of 87 countres representng all major regons of the world and 48 years coverng the perod 1960-2007. Thrd, the paper consders the dfference between advanced and developng countres. Some key papers n ths lterature have noted that although the mpact of 2

natural dsasters s not the same across countres, t s not erratcally heterogeneous ether (see Skdmore and Toya 2007, and Noy 2009, among others). Rather the mpact follows a more or less clear pattern, where poorer natons (n terms of economc, socal, or nsttutonal well-beng) tend to experence stronger effects from natural dsasters. In order to take ths mportant nsght nto consderaton, whle preservng the panel nature of the analyss, the paper conducts the econometrc study not only on the full sample of countres but also on two separate groups: poor or developng countres (62) and rch or advanced countres (25). Fourth, the paper expands the analyss by consderng the potentally dfferent effect of severe vs. moderate natural dsasters. Dsasters of moderate magntude are less dffcult to handle than severe ones. Thus, n the presence of moderate natural dsasters, governments and prvate organzatons can deploy, redstrbute, and relocate ther physcal and human resources to compensate for the losses and reactvate the economy. Under some condtons, moderate dsasters may even brng about an ncrease n economc growth by rasng land productvty (n the case of floods) or nducng captal transformaton (n the case of earthquakes). However, f the dsaster s of such magntude that t overwhelms publc and prvate responses, ts effect s lkely to be more detrmental. At the end of ths ntroducton, the paper offers a comprehensve revew of the new and nterestng lterature dealng wth the macroeconomc mpact of natural dsasters. Nevertheless, at ths pont, we hghlght three papers that are most closely connected wth ths study. The frst s the paper by Loayza, Olaberría, Rgoln, and Chrstaensen (2009). In a sense, that paper and the present study can be regarded as companon papers. Produced almost concurrently, the two studes take advantage of dsaggregaton by type of dsaster, sector of economc actvty, and level of economc development n order to enrch the analyss and elucdate the nterpretaton of results. The focus of Loayza et al. (2009), however, s not on the path of adjustment and recovery but on the net effects n the medum to long terms, for whose analyss t uses perod 3

averages rather than annual data. Therefore, nstead of employng a panel-varx approach to trace yearly responses, Loayza et al. uses GMM-System estmator (desgned for panels wth large cross-secton and short tme-seres dmensons) to obtan average net effects. The second s the paper by Hochraner and Mechler (2009). It assesses the macroeconomc consequences of natural dsasters by comparng the gap between a counterfactual GDP and observed GDP. The counter factual s constructed usng the projecton of past GDP under the assumpton of a no-dsaster scenaro. The paper fnds that natural dsasters on average lead to negatve effects on GDP. Although Hochraner and Mechler s paper dffers from ours regardng the methodologcal approach, t s smlar on the mportance of separatng natural dsasters accordng to type and estmatng ther effects ndependently. Thus, t fnds that typcal (or medan) storms, earthquakes, and droughts have a negatve mpact on GDP, whle floods show a postve mpact one. As shown below, these results are consstent wth most of our fndngs. The thrd paper s by Raddatz (2009). In ths case, the methodologcal approach seems to be smlar to ours regardng the use of an autoregressve model appled to panel data to assess the macroeconomc consequences of natural dsasters. There are, however, some mportant dfferences. Raddatz concentrates on the effects of dsasters on aggregate GDP growth, whle we also analyze the effects on agrcultural and nonagrcultural sectors, fndng dfferng effects on each of these sectors of the economy. Although Raddatz also recognzes the mportance of dsaggregatng by type of dsaster, he emphaszes a way of groupng them that, whle popular n the lterature, may mask contrastng effects. Such s the case of clmatc natural dsasters, whch group together floods and droughts. We separate them and fnd that they have radcally dfferent mpacts on economc growth. Another dfference between Raddatz analyss and ours s that we dfferentate between relatvely moderate dsasters and extremely severe dsasters to capture possble non-monotonc effects. On the other hand, Raddatz contrbuton extends n dmensons that we do not explore. He fnds that nether the nflow of foregn 4

ad nor the ntal level of ndebtedness of the country sgnfcantly affects the growth mpact of natural dsasters. On the other hand, he fnds that the level of economc development does nfluence the mpact of natural dsasters. It s ths dmenson of the heterogenety across countres that we emphasze n ths paper. Before proceedng wth the lterature revew, we now provde the outlne of the paper. Secton II presents the descrpton of the data, ncludng detals on the sample regardng countres, perods, and frequency of observatons; and on the varables used n the analyss concernng defntons, sources, and summary statstcs, wth specal attenton to the measures of moderate and severe natural dsasters. Secton III ntroduces the econometrc methodology, ncludng an exposton of the VARX method, and two mportant specfcaton tests dealng wth exogenety assumptons and lag structures. Secton IV presents the basc results, dscussng and contrastng the effects of droughts, floods, earthquakes, and storms, focusng mostly on the sample of developng countres. Secton V offers some concludng remarks. Lterature revew Economc research n ths feld s stll n an early phase of development. In general, the results on the macroeconomc mpact of natural dsasters seem to be ambguous. A close examnaton n recent studes further demonstrates that these effects may depend on economc, socal, and nsttutonal condtons, as well as on the type of natural dsaster and sector of the economy. Rasmussen (2004) assessed the mpacts of natural dsaster ncdences usng a cross-country sample for the perod 1970 through 2002. The data were obtaned from the EM-DAT database of the Centre for Research on the Epdemology of Dsasters (CRED), whch s the major source of data on natural dsasters used n most studes. Accordng to CRED, a natural dsaster s defned as a stuaton or event whch overwhelms local capacty, necesstatng a request for external assstance. The database conssts of dsaster events whch fulfll at least one of the followng crtera: ten or more people reported 5

klled; 100 or more people reported affected; declaraton of a state of emergency; or call for nternatonal assstance. These dsasters nclude hydro-meteorologcal dsasters such as floods, wave surges, storms, droughts, landsldes and avalanches; geophyscal dsasters such as earthquakes, tsunams and volcanc eruptons; and bologcal dsasters such as epdemcs and nsect nfestatons. To provde a comprehensve pcture, he compared the frequences and mpacts of dsasters across countres by employng four measures, ncludng the number of events dvded by land area, the number of events dvded by populaton, the number of affected persons dvded by total populaton, and damage dvded by GDP. He found that developng countres, partcularly small sland states n the Eastern Carbbean Currency Unon (ECCU), face hgher relatve costs than advanced countres when measured n terms of the number of person affected and the value of the damage. The author also assessed the short-term mpacts of 12 major dsasters occurred n the ECCU and observed ts negatve effects on economc output as well as external and fscal balances. The analyss showed that natural dsasters led to a medan reducton of 2.2% n the same-year real GDP growth. Moreover, a medan ncrease n the current account defct amounted to 10.8% of GDP n the dsaster year. The medan publc debt was also observed to ncrease by a cumulatve 6.5% over three years followng dsaster events. Closely related to ths approach, Heger, Julca, and Paddson (2008) nvestgated the macroeconomc mpact of natural dsasters wth the specfc focus on the Carbbean regon. Ther analyss was based on the annual dataset that ncluded sxteen Carbbean states over the 1970-2006 perod, drawn from the EM-DAT database. The authors frst selected proxes for natural dsasters through a smple OLS estmaton. They dentfed the frequency of dsasters, the estmated costs of dsasters, and the number of total affected as the major explanatory varables for dfferent macroeconomc outcomes. Wth those varables n the correspondng OLS regresson analyss, the results llustrated that natural dsasters negatvely mpact growth, fscal balance, and external balance. These results concde wth those of Rasmussen s presented above. Another sgnfcant fndng 6

was that when a country reles on export or mport specalzaton, larger damages occur n response to dsasters. The authors conclude that dversfcaton of the economy can help mtgate the effects of natural dsasters. Usng a panel vector auto-regresson model, Raddatz (2007) examned the dynamc mpacts of external shocks, ncludng natural dsasters, on the volatlty of output. Focusng on low-ncome countres, he uses a sample of 40 countres over the perod from 1965 to 1997. For the dsaster measurement, the author employs the annual data on the number of dsastrous events, compled from the EM-DAT database. The analyss ndcated that the effects of external shocks n general on per capta GDP are modest and contrbute to only a small porton of ts volatlty, leadng the author to conclude that output volatlty s largely determned by nternal causes rather than external shocks. However, shocks derved from some natural dsasters dd appear to have and mportant effect. In partcular, t was observed that clmatc dsasters lead to a decrease of 2% n real per capta GDP one year after the dsaster, whle humantaran dsasters reduce t by 4%. On the other hand, geologcal dsasters were found to be nsgnfcant n terms of contrbuton to the varance of output. A recent study by Noy (2009) nvestgated the short-run macroeconomc response to natural dsasters usng a panel dataset over the perod 1970-2003. Taken from the EM- DAT database, three measures of dsaster damages were employed: the number of people klled; the number of people affected; and the amount of drect damage. In lght of potental factors that can nfluence the dsaster mpacts, the author took nto account dfferences n populaton sze, sze of economy, and tmng of ncdences. The regresson of annual GDP growth rate on the dsaster measure and other control varables revealed that the mpact of natural dsaster s statstcally sgnfcant when t s measured as the amount of property damage ncurred. As other studes suggest, t was also found that the macroeconomc costs were much hgher n developng countres than n developed countres. Noy further analyzed the determnants of these negatve macroeconomc effects followng dsasters. He concluded that hgher level of lteracy, better nsttutonal 7

qualty, hgher per capta ncome, hgher government spendng, and more open economes along wth better fnancal condtons are lkely to contrbute to countres macroeconomc performance after natural dsasters. On a smlar lne, several studes have documented that economc development plays an mportant role n mtgatng a countres vulnerablty to catastrophc ncdences. Skdmore and Toya (2007) nvestgated the effects of the level of development on dsaster mpacts, usng a dataset of natural dsasters ncurred n 151 countres over the perod from 1960 to 2003. The analyss ncluded two patterns of dataset obtaned from the EM-DAT. One used the number of klled to assess the dsaster mpacts, whle the other consdered economc damages. The OLS regresson analyss demonstrated that human and economc damages from natural dsasters are generally reduced along wth economc development. In partcular, the results showed that deaths and damages were lower n countres wth hgher level of educatonal attanment, greater degree of openness, more developed fnancal sector, and smaller governments. The authors suggest that polcymakers could consder further efforts n developng economc and socal nfrastructures, whch can contrbute to decreasng natural hazards. Takng a dfferent approach n explorng the mpacts of captal and labor losses on short-term growth, Casell and Malhotra (2004) tested the emprcal valdty of the predctons of the Solow growth theory. The theory suggests that a declne n the captallabor rato resultng from a natural dsaster would lead to an ncrease n the country s growth rate, whle an ncrease n the captal-labor rato would curtal t. In ther emprcal analyss, the total number of people klled, njured, and affected by dsasters were used to calculate the percentage loss n the labor force, whle the mmedate damage as a percentage of GDP was used as a proxy for the loss n captal stock. The data were compled from the EM-DAT database for a sample of 172 countres for the perod between 1975 and 1996. Usng the real per capta GDP growth rate n the dsaster year to estmate the Solow model, ther emprcal analyss found that sudden losses of captal and labor dd not brng about a change n the economc growth as expected by the Solow 8

growth model. The results, however, reman questonable gven the proxes used to measure captal and labor destructons and the tmng of the growth response. Jaramllo (2007) presented a comprehensve analyss of the lnk between natural dsasters and economc growth both n the short-run and long-run usng a panel dataset of 113 countres over the perod 1960-1996. However, dsasters that develop through extended perods, such as droughts and famnes, as well as nsect nfestatons and epdemcs are excluded from the analyss. The type of dsasters examned by Jaramllo nclude earthquakes, floods, wld fres, wnd storms, waves and surges, extreme temperatures, volcano epsodes, and sldes. Takng country and year fxed effects nto account and controllng for trade openness and foregn ad, the author examned the short-run effects of dsasters on economc growth, followed by an analyss of the longrun effects. For the short-run, Jaramllo assessed the mpacts on GDP growth n the dsaster year and the followng year, whereas for the long-run, he tested for the cumulatve dsaster effects over the perod 1960-1996 on the GDP per captal level n 1996. The regresson results ndcated that short- and long-term effects of dsasters are determned by countres ncome level, populaton, and the type of dsaster. On the whole, t was found that the effects of dsasters on GDP growth rate vared from 0.9% decrease to 0.6% ncrease dependng on the dsaster type. Focusng specfcally on the long-term macroeconomc mpacts of natural dsasters, the frst comprehensve emprcal research was done by Skdmore and Toya (2002). In ther cross-country analyss, the authors use average per capta GDP growth over the perod 1960-1990 and the total number of sgnfcant dsaster events observed n respectve countres durng the same perod. The dsasters studed cover clmatc and geologc dsasters. The results revealed that clmatc dsasters have postve effects on the long-run economc growth as they nduce hgher captal accumulaton and total factor productvty than before. It s argued that total factor productvty s the predomnant factor n promotng growth after dsasters. By contrast, geologc dsasters were observed 9

to affect growth negatvely as t deterorates physcal captal and decreases human captal due to the ntal loss of lfe. Followng Skdmore and Toya s fndngs, Cuaresma, Hlouskova, and Oberstener (2008) examned the long-run effects of natural dsasters by analyzng the drect relatonshp between foregn technology absorpton and dsaster ncdences. Earler studes argued that dsasters can provde countres wth opportuntes to renew technologes, thereby promotng long-run growth. The authors assess ths argument by usng gravty model to analyze foregn knowledge spllovers between the G-5 countres and a sample of 49 developng countres. Accordng to the regresson results, natural catastrophc rsk negatvely affected knowledge transfers from the ndustralzed to developng countres. The authors further found that countres wth hgher levels of development are more lkely to be better off than countres wth lower levels of development through captal upgradng followng natural dsasters. Hallegatte and Ghl (2007) added busness cycle framework to the study of dsaster mpacts. They analyzed the effects of exogenous shocks, ncludng natural dsasters and stochastc productvty stocks, on economc behavor. Employng a Non- Equlbrum Dynamc model wth endogenous busness cycles, they found that total GDP losses resultng from natural dsasters are hgher when occurrng durng expansons than durng recessons. The reason s that because pre-exstng dsequlbra are wdened by exogenous shocks n the former phase, whereas the shocks are mtgated by the exstence of unused resources n the latter case. The paper drew the concluson that the phase of the busness cycle durng whch a dsaster occurs affects the degree of the macroeconomc response. As dscussed above, whle some studes found common patters n the determnants of a country s vulnerablty to catastrophc events, researchers have not come to a consensus on the mpacts of natural dsasters on economc growth. Ths paper attempts to help dsentangle ths ambguty by usng a better-grounded econometrc 10

methodology and a conceptually drven dsaggregaton by type of natural dsaster and sector of economc actvty. II. Data A. Perods, frequency, samples (groups of countres) To perform our estmatons, we use pooled cross-country and annual tme-seres data coverng 87 countres over the perod 1960-2007. The panel s unbalanced, wth some countres havng more observatons than others. We refer to the data as all countres. Then we splt the data nto two groups: rch countres and developng countres. We classfy 25 Arab and OECD countres nto the frst group and the other 62 countes nto the second group. Table II.1 gves the lst of countres of these groups. B. Varables, defntons, sources The man varables used n the paper are dvded n three groups. Frst, to study the mpact of natural dsasters on the economy, we defne three types of growth varables. The frst s the growth rate of real per capta Gross Domestc Product (GDP). The others are the growth rates of real per capta value added n the two major sectors of the economy, the agrcultural sector and the non-agrcultural sector. All of them are measured as the log dfference of per capta output (n 2000 US dollars), where per capta output s obtaned by dvdng the value added of each sector by the total populaton. Second, as a varable whch represents the role of external condtons that may affect the growth performance across countres, we use shocks to the Terms of Trade (TOT). Terms of trade shocks are measured by the growth rate of the terms of trade (export prces relatve to mport prces). The dea s to capture shfts n the demand for a country s exports. Data for all the above varables were obtaned from the World Bank (WDI, 2008). 11

The last set of varables represents the role of natural dsasters on the growth performance across countres. Data for natural dsasters were obtaned from the Emergency Dsasters Database (EM-DAT) mantaned by the Center for Research on the Epdemology of Dsasters (CRED). EM-DAT provdes the number of casualtes (people confrmed dead, reported mssng, and presumed dead), the number of people njured, and the number of people affected. People affected are those requrng mmedate assstance durng a perod of emergency. Also, people reported njured or homeless are aggregated wth those affected to produce the total number of people affected (we refer to ths number as total affected ). Throughout the paper, we assume that natural dsaster varables are (block) exogenous wth respect to the growth varables and shocks to the terms of trade. 1 C. Moderate and severe natural dsasters As mentoned above, we dvde natural dsasters nto four categores: droughts, floods, earthquakes, and storms. The measure of ntensty of natural dsasters, gven by: ND t k,, s drought f k = 1, flood, f k = 2, k t NDt, = (1) earthquake f k = 3, storm t, t, t, f k = 4, where ntensty k t,, j k k klledt,, j + 0.3* total affectedt,, j =, (2) populaton t, k k = 1, f ntenstyt,, j > 0.0001, ND t,, j (3) = 0, otherwse, 1 For the exogenety of natural dsaster varables, see secton III.b.. 12

J t, j= 1 k k ND = ND, (4) t,, j and J descrbes the total number of type-k events (k = 1, 2, 3, and 4 correspond to drought, flood, earthquake, and storm, respectvely) that took place n country durng year t. The followng steps descrbe how to create the ntensty measure. Frst, for each k event of type-k dsaster, we create a varable ntensty t,, j measurng the magntude of the event relatve to the sze of the economy, that s, the sum of the number of casualtes k k ( klled t,, j ) and 30% of the total number of people affected ( total affected t, j ), dvded by the populaton (equaton (2)) 2. Then we construct a dummy varable ND k t,, j whch takes the value of 1 f k ntensty t,, j s greater than 0.01% (equaton (3)). Fnally, for each type of dsaster, the respectve dummy varables ND k t j, j = 1,..., are summed up to obtan,, J the ndcator value ND, to assess the total magntude of type-k dsasters n country k t durng year t (equaton (4)). Many practtoners pont out that the mpact of moderate dsasters and extremely severe dsasters on the economc performance dffer, not only n ther magntude, but also n ther dynamc characterstcs. To capture the partcular effects of severe dsasters, we construct a second measure of ntensty, sevnd,, as follows: k t sev. drought sev. flood, f k = 2, k t sevndt, = (5) sev. earthquake f k = 3, sev. storm t, t, t, f k = 1, f k = 4, where 2 Ths ntensty measure s smlar to the one establshed by the Internatonal Monetary Fund (IMF, 2003), and used by Becker and Mauro (2006). 13

ntensty k t,, j k k klledt,, j + 0.3* total affectedt,, j =, (6) populaton t, k k = 1, f ntenstyt,, j > 0.01, sevnd t,, j (7) = 0, otherwse, J t, j= 1 k k sevnd = sevnd. (8) t,, j Here, for the dummy varable for the ntensty of ndvdual severe dsaster, sevnd, k t, j, we set the threshold at 1% of the populaton, whle we appled the threshold of 0.01% for general or moderate dsasters. In secton IV, we show the results of two types of estmaton, n whch () only moderate dsaster varables are ncluded (the basc model), and () both moderate and severe dsaster varables are ncluded. D. Summary statstcs Regardng the growth varables ntroduced n the early part of ths secton, a few observatons deserve some comments. Frst, we should pont out that the growth performance of the dfferent sectors vares wdely n each country. As shown n Table II.2, durng the perod 1960-2007, the non-agrcultural sector has had much hgher average growth rate (1.7% n developng countres, 2.1% n rch countres) than the agrcultural sector (0.31% n developng countres, 0.93% n rch countres). Also, Table II.3 shows that the correlaton between the growth rates of non-agrcultural sector wth the agrcultural sector s qute low (0.1095 n developng countres and 0.0173 n rch countres). The consderable dspartes among the growth performances provde some grounds to suspect that natural dsasters could have had dverse effects on the dfferent sectors of the economy. 14

III. Methodology A. Econometrc method namely, The econometrc model we adopt here s a fxed-effects Panel VARX model, y t, = α + Φ1y t 1, + Φ2y t 2, + Φ3y t 3, + Θ0xt + Θ1xt 1, + Θ2xt 2, + εt,, (9) where the country ndex s = 1, 2,, M and the tme ndex for each country s t = 1, 2,, T. The fxed effect for each country s represented by α. Hereafter, the total number of observatons for all countres n the panel s denoted by T = M T = 1. The endogenous varables vector s denoted by the (2 1) vector y t whle the (4 1) exogenous varables vector x t represents the occurrences at tme t of the dsasters, respectvely, drought, flood, earthquake, and storm. In equaton (9) we assume the homogenous error structure E ( ε ε t, t, ) = Ω for all t and where ε t, s the (2 1) vector of errors of the system. Furthermore, we assume ndependence of the errors wthn equatons, E ε ε ) = 0, j, and across equatons, E ε ε ) = 0, for any t and s where j. ( t, t, j ( t, s, j Model (9) s appled to three dfferent groups of countres: All of the countres, Developng countres, and Developed Countres. We choose to estmate Model (9) by OLS to the demeaned seres resultng n the so-called wthn-fxed-effects estmator. As ponted out by Nckell (1981), gven that Model (9) s dynamc, f T s small and fxed, such an estmator s nconsstent as the number of countres, M, goes to nfnty. However, n our case we consder the number of countres fxed and snce n each groupng of the countres consdered here the number of avalable observatons, T, s at 15

least 778. 3 In ths case, the bas of the wthn-fxed-effects estmator should be neglgble. Hereafter, we refer to the wthn-fxed-effects estmator smply as the OLS estmator, wth the coeffcent estmates beng denoted by Φ, = 1, 2, 3, and Θ, = 0, 1, 2. or Model (9) can be wrtten more compactly as 2 3 2 ( I Φ1L Φ2L Φ3L ) y t, = α + ( Θ0 + Θ1L + Θ2L ) xt, + εt,, Φ L) y = α + Θ( L) ε, (9 ) ( t, t, where L denotes the usual lag operator. To nsure that (9 ) produces a steady state, we 2 3 requre that all of the roots of the determnant equaton ( I Θ L Θ L Θ L ) 0 le ˆ ˆ 1 2 3 = outsde of the unt crcle. Invertng (9 ) produces the multpler form of Model (9): y 1 1 = Φ L) Θ( L) x + Φ( L) ε. (10) t, ( t, t, The mean responses from the occurrences of natural dsasters are therefore captured by the lag polynomal 1 Ψ( L) = Φ( L) Θ( L). (11) It follows that the coeffcents of the lag polynomal Ψ (L) can be obtaned by matchng the coeffcents n the expresson Ψ ( L) Φ( L) = Θ( L). (12) Ths gves rse to the solutons 3 The number of observatons avalable n the sample of rch countres, wth non-agrcultural growth rate as an endogenous varable. 16

Ψ 0 = Θ 0 (13) Ψ = + (14) 1 Θ1 Ψ 0Φ1 Ψ = + (15) 2 Θ 2 + Ψ1Φ 2 Ψ 0Φ 2 Ψ = + (16) 3 Ψ 2Φ1 + Ψ1Φ 2 Ψ 0Φ3 Ψ s = Ψ s 1Φ1 + Ψ s 2Φ 2 + Ψ s 3Φ3 for s 4. (17) Now let [ Φ Φ Φ Θ Θ ] Π = (18) 1 2 3 0 1 Θ 2 denote the coeffcent matrx of (9). The coeffcent matrx Φ s 2 2 for = 1, 2, 3 and Θ s 2 4 for = 0, 1, 2. Therefore, the coeffcent matrx Π s (2 (6 + 12)) = (2 18). Let π = vec(π). Then π s a (2(18) 1) vector wth the frst 18 elements beng the autoregressve and current and lagged natural dsaster coeffcents from the frst equaton and the second 18 elements beng the correspondng coeffcents from the second equaton. Let ψ = vec Ψ ) denote the (2(4) 1) vector of the s-perod delay mean s ( s responses due to natural dsasters. The frst 4 elements represent the s-perod delay mean responses of the frst endogenous varable to the natural dsasters whle the second 4 elements represent the s-perod delay mean responses of the second endogenous varable to the natural dsasters. Moreover, let πˆ denote the vector of the OLS estmates of equaton (9). Then t can be shown under farly general condtons that 1 T ( πˆ π) N( 0,( Ω Q )) (19) 17

where Ω = E ε ε ) s the varance-covarance matrx of the error terms of (9) and ( t, t, Q = plm( X' X / T ) where X s a (T 18) desgn matrx of the form ' X 1 ' X 2 X = ' X T (20) ' ' ' ' ' ' ' where X ( y y y x x x ). t = t 1 t 2 t 3 t t 1 t 2 In mplementng the result of equaton (19), we need consstent estmates of Ω and Q. These estmates are obtaned as follows: T Ω ˆ 1 = εˆ tεˆ t T t= 1 (21) and vector Q ˆ = X' X / T. (22) Let Ψˆ s ( πˆ ) denote the estmated s-perod delay mean responses to the exogenous x t where the dependence of these estmates on the coeffcent estmates πˆ s made explct. One way to obtan standard errors for these estmates s to use Monte Carlo methods. Frst, randomly draw a (36 1) vector from the dstrbuton 1 ( ˆ ˆ 1 (1) N ( πˆ, Ω Q )). Denote ths vector by π. Calculate ˆ (1 Ψ ( ) s π ). Repeat ths process T for, say, a total of 10,000 tmes. Then to get, for example, the 90% confdence nterval for the frst element of Ψ s, say Ψ s1, we need the 5 th percentle, Ψ s1, and the 95 th percentle, Ψ s1, from the smulated values of Ψ s1 resultng n the 90% confdence 18

nterval for Ψ s1, namely, ( Ψ s1, Ψ s1 elements of B. Dagnostc tests Ψ s are smlarly constructed.. Indvdual and panel unt root tests ). The confdence ntervals for the remanng Before we can proceed to buld a VARMAX panel model for analyzng the effects of natural dsasters on varous endogenous varables, we need to determne the statonary forms of the endogenous varables we are gong to be usng n our analyss. In ths study we chose as the endogenous varables of nterest (1) the log of real GDP per capta, (2) the log of real agrcultural value added per capta, (3) the log of real nonagrcultural value added per capta, and (4) the log of terms of trade. We chose to use the log transformaton of the varables because of the varance stablzng characterstcs of the transformaton and the fact that, f a unt root s contaned n the logged varables, then dfferencng them yelds a very straght-forward nterpretaton of the dfferenced data, namely percentage change. We proceeded to pursue unt root testng n these varables n two ways: seresby-seres unt root tests and panel unt root testng wth ndvdual country effects as n the Levn, Ln, and Chu (2002) and Im-Pesaran-Shn (2003) panel unt root testng frameworks. These unt root tests are, of course, dependent on the specfcaton of the determnstc parts of the unt root test equatons. That s, does the data contan a trend or not? Is the data wthout trend but has a non-zero mean as compared to a zero mean? To obtan consstent statstcal hypothess test results one must properly specfy the determnstc parts of the data under the alternatve hypothess of statonarty. In ths ven we tested the sgnfcance of the trend n the above four seres by testng the sgnfcance of the ntercept n the followng AR(2) equaton of the varable n queston, country-bycountry: z t = α + φ1 zt 1 + φ2zt 1 + ε t. (23) 19

In equaton (23) z t represents a partcular country s varable n queston and represents the frst dfferencng operator. We specfed a second-order autoregresson to ensure that the resduals of the equaton would be whte nose thus mplyng that OLS t- statstcs nvolvng the ntercept α would be approprate for testng for the presence or absence of trend. In the case that the null hypothess H : α 0 was supported, we 0 = concluded that the data does not have a trend n t. On the other hand, f the alternatve hypothess of H : α 0 was supported, we concluded that the data has trend n t. Wth 1 respect to the log of real GDP per capta and log of real non-agrcultural value added per capta, the preponderance of tests ndcate trend s present (52 of 87 null hypotheses rejected for the former and 47 of 87 null hypotheses rejected for the latter). In contrast, for the log of real agrcultural value added per capta and the log of terms of trade, the preponderance of tests ndcated that trend s absent (15 of 87 null hypotheses rejected for the former and 1 of 87 null hypotheses rejected for the latter). Thus, for the producton run of unt root tests, we choose to treat all of the log of real GDP per capta and log of real non-agrcultural value added per capta seres as havng trends n them whle the log of real agrcultural value added per capta and log of terms of trade seres had no trend n them but non-zero means. 4 As a result of these trend tests we chose to use an ntercept and determnstc trend n testng for unt roots country-by-country n the log of real GDP per capta and log of real non-agrcultural value added per capta seres n the augmented Dckey-Fuller and Phllps-Perron unt root test equatons whle for the log of real agrcultural value added per capta and the log of terms of trade, we chose to use only an ntercept n the augmented Dckey-Fuller and Phllps-Perron unt root test equatons. Of course, when testng for the suffcency of the frst dfference n producng statonarty n a seres, we checked the frst dfference of the seres for unt roots usng the approprate determnstc terms mpled by dfferencng. In partcular, when testng for the statonarty of the frst 4 Detaled test results are avalable from the authors upon request. 20

dfference of the log of real GDP per capta and the frst dfference of the log of real nonagrcultural value added per capta we ncluded only an ntercept n the test equaton. In contrast, when testng for the statonarty of the frst dfference of the log of real agrcultural value added per capta and the log of terms of trade we set the ntercept to zero n the test equaton. In contrast to the country-by-country unt root tests, the panel unt root tests of specfc tme seres assume as the null hypothess that a unt root exsts for all of the countres, wth country dstncton comng only from havng separate determnstc terms for each country (.e. dfferent ntercept effects or dfferent ntercept effects as well as dfferent trend effects for each country). The dfference between the Levn, Ln, and Chu (2002) and Im-Pesaran-Shn (2003) panel unt root tests resdes n the form of the alternatve hypotheses assumed by the tests. In the Levn, Ln, and Chu test the alternatve hypothess takes the form of a common statonary frst-order autoregressve coeffcent across all of the countres whereas the Im-Pesaran-Shn test assumes all of the frst-order autoregressve coeffcents are statonary but that they can possbly take on dfferent statonary values. Both tests are, of course, all-or-none tests n the sense that test results mply that ether (1) all of the countres gven seres have unt roots n them or (2) all of the countres seres are statonary of the same degree (as n the Levn, Ln, and Chu test) or dfferent degrees (as n the Im-Pesaran-Shn) test. The beneft of the panel unt root tests are that, n the case of short tme seres n the panel, the power of the unt root tests are ncreased when one or more of the panel seres are non-statonary as compared wth country-by-country unt root tests. The results of the above unt root tests appled to the four seres are summarzed n Table III.1. 5 The left half of the table pertans to unt root tests of the non-trendng seres (log of real agrcultural value added per capta and log of terms of trade) whle the rght half of the table pertans to the unt root tests of the trendng seres (log of real GDP 5 All of the results reported n Table III.1 were produced by EVews 6.0. 21

per capta and log of real non-agrcultural value added per capta). In addton, the top half of the table (Secton A) reports the unt root tests of the levels whle the bottom half of the table (Secton B) reports the unt root tests of the frst dfferenced data. Furthermore, n each secton the results of four unt root tests are reported, the frst two tests beng country-by-country unt root tests whle the latter two tests are the panel unt root tests. 6 The results reported n Table III.1 are summarzed as follows: Log of real agrcultural value added per capta. The preponderance of the ndvdual unt root tests ndcates the presence of a unt root. The panel unt root tests lkewse ndcate the presence of unt roots. After frst dfferencng the seres seems to be statonary. Log of Terms of Trade. The results for ths seres are smlar to those of the prevous non-trendng seres except for the sgnfcance of the Levn-Ln-Chu panel test where the p-value s less than 5% n the levels of the data. In contrast the Im-Pesaran-Shn panel test (wth a flexble alternatve hypothess) ndcates a unt root at the 10% level. Evdently, the log of terms of trade s near statonary. Despte ths splt decson on the exstence of a unt root we decded to treat ths seres as havng a unt root and to model ts dfferences as beng statonary. Log of Real GDP per capta. The preponderance of the ndvdual unt root tests ndcates the presence of a unt root. The panel unt root tests lkewse ndcate the presence of unt roots. After frst dfferencng the seres seems to be statonary. 6 Note n the case of the frst dfference of the non-trendng data, the Im-Peseran-Shn test s not reported as EVews does not accommodate the zero mean case. 22

Log of real non-agrcultural value added per capta. The same conclusons hold that hold for the log of real GDP per capta. Unt roots are present and the frst dfferenced seres appears to be statonary. In summary, the test results of Table III.1 ndcate that, when buldng meanngful VARMAX panel models to examne the mpacts of varous natural dsasters on developng countres GDP and agrcultural, non-agrcultural value added, and terms of trade, the growth rate forms of these endogenous varables should be used.. Block exogenety tests The VARX model presented n the prevous subsecton s dependent on the assumpton of exogenety of the natural dsaster varables. Whle all varables n the model are assumed to be endogenous n a smple VAR model, a VARX model allows some of the varables to be exogenous. In ths secton, we present the hypothess testng method about the exogenety of the dsaster varables and ts results. Here, we are nterested n the exogenety of the dsaster varables as a group, wth respect to shocks to the terms of trade and one of the growth varables (GDP growth, agrcultural growth, or non-agrcultural growth). Wthout assumng the exogenety of the dsaster varables, we can rewrte Model (9) as a smple VAR of order p as follows: x y t, t, = α + 1 = α 2 + p h= 1 p h= 1 A C h h x x t h, t h, + + p h= 1 p h= 1 B D h h y y t h, t h, + u + v t,, t,, (24) where 23

x y t, t, droughtt, floodt, =, earthquake t, stormt, TOTt, = GDP / Agr. / Non agr. growtht,, (25) and α 1 and 2 α are the fxed effects for country. In equaton (24) we assume the homogenous error structures: E u u ) = Ω, E( u v ) = Ω, E( v u ) =, ( t, t, 11 t, t, 12 t, t, Ω21 and E ( v t, v t, ) = Ω22 for all t and, where t, u and v t, are the errors of the system. The group of varables represented by x s sad to be block-exogenous wth respect to the varables n y f B h = 0 for h = 1,, p. To check the exogenety of the dsaster varables, we can perform a lkelhood rato test wth the null hypothess, H : B =, h = 1,, p. Ths test can be done wth 0 h 0 runnng OLS regressons of each of the dsaster varables on p lags of all of them and p lags of all of the elements of y. Let denote û t, the (4 1) vector of sample resduals from these regressons and ˆΩ 11 ther varance-covarance matrx. Next, run OLS regressons of each of the dsaster varables only on p lags of them, wthout lagged ˆ t varables of y. Let denote u (0) the (4 1) vector of sample resduals from the second set of regressons and Ω ˆ 11 (0) ther varance-covarance matrx. If T {log Ωˆ (0) log ˆ }, (26) * 11 Ω11 where T s the number of observatons, s greater than the crtcal value for a χ 2 (4 2 p) varable, then the null hypothess s rejected and the concluson s that some of the dsaster varables are helpful n forecastng y,.e., the dsaster varables are not blockexogenous wth respect to the varables n y. 24

Table III.1 dsplays the results of the block exogenety test. As t shows, the null hypothess s not rejected n any of three samples, wth any of growth varables, and wth p = 1, 2, 3, at 5% of statstcal sgnfcance. At 10% of sgnfcance, the null hypothess s rejected only n 2 cases out of 27 cases, when we use the sample of rch countres and nclude the agrcultural growth n y, wth p = 1 and 3. These results strongly suggest the use of VARX model, over the use of VAR model n whch all varables are treated as endogenous.. Lag structure Before estmatng the panel VARX model, we need one crucal pece of nformaton. That s the number of lags to nclude for each varable n the model. To dentfy the lag structure, some statstcal crtera can be used. A well-known crteron s Akake s nformaton crteron (AIC) (Akake (1973)), gven by l K AIC = 2, (27) T and an alternatve s Schwarz s Bayesan nformaton crteron (SBC) (Schwarz (1978)), whch s gven by 2l + log( T ) K SBC =, (28) T where T s the number of observatons, K s the number of parameters n the model 7, 1 ˆ ˆ ε ε l = T 1+ log(2π ) + log det, 2 T (29) 7 In our basc model, K = 2(2p + 4(q + 1)). 25

and εˆ s the (T 2) matrx of the error terms of Model (9). Models wth a lower AIC or SBC are preferred. Both crtera add a penalty that ncreases wth the number of regressors or lags. Table III.2 shows the AIC and SBC statstcs for the models wth three dfferent endogenous varables (GDP growth, agrcultural growth, and non-agrcultural growth) and three dfferent groups of countres (all countres, developng countres, and rch countres). p and q represent the number of lags for the endogenous varables and the exogenous varables, respectvely. In most cases, the results suggest ether the models wth p = q = 1, or the models wth p = q = 2. Clearly, SBC tends to favor more parsmonous models than AIC, because the penalty for ncreasng the number of lags s larger for SBC. Based on the nformaton crtera values, we selected the lag length 2 as our basc lag structure. From a statstcal pont of vew, there s lttle to choose between the lag length 1 and 2, snce we have the mxed results from the nformaton crtera. The latter one, however, provdes much rcher dynamcs of the mean responses of the endogenous varables to exogenous shocks. As the goal of ths paper s to study the dynamc effects of natural dsasters, ths s reason enough to select the lag length 2. We apply ths lag structure to all of our models homogenously to smplfy the nterpretaton. IV. Results We now report and dscuss the man results on the growth consequences of natural dsasters. We organze the presentaton by type of dsaster droughts, floods, earthquakes, and storms. For each of them, we consder the effects on GDP per capta growth and ts major components, agrcultural and non-agrcultural per capta valueadded growth. We frst estmate these effects usng the sample of all countres (Table IV.1). Then, to gan further nsght on the development angle of the ssue, we dvde the 26

sample nto developng countres (Table IV.2) and advanced countres (Table IV.3). Focusng on the sample of developng countres (for whch the effects are stronger), we then consder the dfferng mpact of moderate and severe natural dsasters (Table IV.4). The estmaton of the VARX model renders a wealth of results, from whch we choose those that are most pertnent to the man objectve of the paper. Snce we are nterested n tracng out the dynamc path of adjustment n the aftermath of the dsaster, the most relevant results are the mean response of growth to a gven natural dsaster for each year after the event. Snce the effects are small and non-sgnfcant a few years after the event, we only report the mean responses for years 0, 1, 2, and 3 of the event (where year 0 s when the dsaster occurred). We ndcate whether these responses are statstcally greater or smaller than zero, accordng to the Monte Carlo smulatons explaned n the methodologcal secton of the paper. Furthermore, we report the cumulatve effect of the event, whch corresponds to the sum of mean responses for the 4 years after the event. We organze and present these results n several tables, as ndcated above. In addton, we present a graphcal representaton of the mean responses for each natural dsaster for the sample of developng countres, together wth ther correspondng confdence bands ndcatng 10% tals of the dstrbuton of effects (Fgures IV.1-4). The confdence bands are obtaned through the Monte Carlo smulatons mentoned above. The majorty of the dscusson refers to the results obtaned wth the sample of developng countres. For comparson purposes, we also dscuss the results from the sample of all countres (of whch developng countres represent nearly 80%) and the sample of advanced countres. Fnally, we offer some robustness analyss regardng the lag structure of the VARX model (Appendx Tables A.1 and A.2, and Fgures A.1-A.4). In partcular, we use a more restrctve lag structure, p = q = 1, whch, as mentoned n the prevous secton also receved support from the nformaton crtera tests. The results are broadly smlar to those usng the preferred longer lag structure. The man dfference s that when only 27

one lag s allowed, the mean responses correspondng to later years are smaller and less sgnfcant. A. Droughts Droughts have an overall negatve effect on GDP growth. As expected, the effect s stronger for agrcultural growth, but t s also negatve for non-agrcultural actvtes. For agrcultural growth, the negatve effect of droughts s larger on the year of the event. There s a sgnfcant recovery on the followng year, but the cumulatve effect remans sgnfcantly negatve. For non-agrcultural growth, the negatve mpact s felt on the year of the drought and also a couple of years afterwards, ndcatng the presence of delayed effects. In the sample of developng countres, the cumulatve negatve response to droughts s 1.7 percentage ponts (pp) for GDP growth and 1.6 pp for agrcultural growth. The pattern of results just descrbed apples to the samples of all countres and of developng countres. For advanced countres, there s also a negatve response on the year of the drought but t only apples to agrcultural growth. Furthermore, n the subsequent years agrcultural growth recovers so substantally that the cumulatve effect of droughts for advanced countres s essentally zero. Turnng to the analyss of severe vs. moderate cases, the strongest negatve effects (n sze and statstcal sgnfcance) come from severe droughts. The year of the event, severe droughts have twce the negatve mpact on GDP growth than moderate droughts. Furthermore, severe droughts nduce larger volatlty of growth, whch means that they produce a larger drop the year of the event and a stronger recovery n the followng year. In the case of GDP growth, ths recovery s suffcently strong so that the cumulatve effect of severe droughts s comparable to that of moderate droughts (1.5-2.0 pp). However, n the case of agrcultural growth, the recovery s nsuffcent and, then, the negatve cumulatve mpact of severe droughts (2.0 pp) s twce as large as that of 28