Handbook on the Economics of Disasters 1839103728, 9781839103728

Citation preview

HANDBOOK ON THE ECONOMICS OF DISASTERS

I dedicate this book to my wife, Kate, with all my love. Our journey together began with a honeymoon year in Japan when I was a Fulbright Scholar. This is also where my journey into the topic of disaster economics began. Since that time, we have raised six children together. Whatever I have learned about the economics of natural disasters over the past 25 years and shared in this book would not have been possible without you and your partnership. Thank you for everything.

Handbook on the Economics of Disasters

Edited by

Mark Skidmore Professor of Economics/Morris Chair in State and Local Government Finance and Policy, Department of Agricultural, Food, and Resource Economics/Department of Economics, Michigan State University, East Lansing, Michigan, USA

Cheltenham, UK • Northampton, MA, USA

© Editor and Contributors severally 2022 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical or photocopying, recording, or otherwise without the prior permission of the publisher. Published by Edward Elgar Publishing Limited The Lypiatts 15 Lansdown Road Cheltenham Glos GL50 2JA UK Edward Elgar Publishing, Inc. William Pratt House 9 Dewey Court Northampton Massachusetts 01060 USA

A catalogue record for this book is available from the British Library Library of Congress Control Number: 2022942625 This book is available electronically in the Economics subject collection http://dx.doi.org/10.4337/9781839103735

ISBN 978 1 83910 372 8 (cased) ISBN 978 1 83910 373 5 (eBook) Typeset by Westchester Publishing Services

EEP BoX

Contents List of contributors Acknowledgements

viii xvi

  1 Introduction to the Handbook on the Economics of Disasters Mark Skidmore   2 A taxonomy of natural disasters Mark Skidmore

1 13

PART I  THEORETICAL CONSIDERATIONS IN EVALUATING DISASTER IMPACTS   3 A few good models for economic analysis of disasters: can your model handle the truth? Yasuhide Okuyama

30

  4 Behavioral economic consequences of disasters Adam Rose

50

  5 The role of biases and heuristics in addressing natural disasters Howard Kunreuther and Wouter Botzen

72

  6 Risk preferences and natural disasters: a review of theoretical and empirical themes Laura Bakkensen and Marc N. Conte

86

PART II  EVALUATION OF DISASTER CONSEQUENCES SECTION I  Economic Impacts   7 Economic consequences of pre-COVID-19 epidemics: a literature review Ilan Noy and Tomáš Uher

117

  8 Natural disasters and economic growth: revisiting the evidence Jesús Crespo Cuaresma

134

  9 The impact of natural disasters on economic growth Eduardo A. Cavallo, Oscar Becerra and Laura Acevedo

150

10 Assessing the impact of natural disasters on industry gross domestic product in the United States Monica Escaleras, Anand Jha and Christopher J. Boudreaux

v

193

vi  Handbook on the economics of disasters 11 The fiscal consequences of natural disasters Tatyana Deryugina

208

SECTION II  Social Impacts 12 Natural disasters and self-reported well-being: a review of the literature Michael Berlemann and Marina Eurich

230

13 Preferences, behavior, and welfare outcomes against disasters: a review Yasuyuki Sawada

252

14 Effect of major disasters on geographic mobility intentions: the case of the Fukushima nuclear accident Eiji Yamamura, Chisako Yamane, Shoko Yamane and Yoshiro Tsutsui

275

PART III  RISK MANAGEMENT, RESILIENCY AND VULNERABILITY SECTION I  Insurance 15 The role of insurance in integrated disaster risk management with a focus on how insurance can support climate adaptation and disaster resilience Swenja Surminski 16 Supplying insurance for natural disasters: a retrospective study of property insurer strategies Patricia H. Born and J. Bradley Karl

294

319

SECTION II  Risk Assessment and Reduction 17 Expanded disaster risk assessment using agent-based modeling: a case study on floods in Sri Lanka Brian Walsh and Stéphane Hallegatte 18 Using weather modification to subdue severe weather Scott Knowles and Mark Skidmore

355 389

SECTION III  Resilience and Vulnerability 19 Advances in the empirical estimation of disaster resilience Noah Dormady, Adam Rose and C. Blain Morin

401

20 State capacity and vulnerability to natural disasters Richard S. J. Tol

434

Contents  vii SECTION IV  Recovery and Response 21 Small business recovery: lessons from Hurricane Katrina and the COVID-19 pandemic Maria I. Marshall, Bhagyashree Katare and Corinne B. Valdivia

459

22 Disaster challenges and entrepreneurial responses Virgil Henry Storr, Stefanie Haeffele and Alexander W. Craig

472

Index

487

Contributors Laura Acevedo has a master’s degree in economics from the Universidad de los Andes in Bogotá, Colombia. She has worked at the Inter-American Development Bank and the CEDE. Her work is focused on the macroeconomic effects of inequality and empirical analysis of the effects of natural disasters. Laura Bakkensen utilizes applied microeconomic and econometric techniques to study the economics of natural disasters, identifying current hazard risks and evidence of adaptation to hurricane damages and fatalities across the globe. Her research informs policy on insurance regulation, post-disaster aid, hurricane warnings, and public adaptation projects. Oscar Becerra is an assistant professor in the economics department of the Universidad de los Andes in Bogota, Colombia. His work is focused on the empirical analysis of the effects of natural disasters and the effects of public policies on labor market outcomes. Michael Berlemann studied economic sciences at Ruhr University Bochum. After graduating in 1994, he became a research assistant at Dresden University of Technology and finished his PhD in 1999 for his dissertation on “Politico-economic theories of inflation and business cycles.” In 2004, he became managing director of the Dresden branch of the ifo Institute of Economic Research. After completing his habilitation at Dresden University of Technology with a thesis on methods of inflation forecasting, he became a professor of economics, especially political economy and empirical economics in July 2007. In February 2022, he became scientific director of Hamburg Institute of International Economics (HWWI). His primary fields of research are applied econometrics, political economy, economics of climate change, macroeconomics, monetary economics, and migration. Patricia H. Born is Midyette Eminent Scholar and a professor in the Department of Risk Management/Insurance, Real Estate and Legal Studies at Florida State University. She received her MA and PhD from Duke University and her BA from the University of Michigan. Dr. Born teaches courses in risk management, insurance, and economics at all academic levels. Her research interests include insurance regulation, medical malpractice liability, health insurance, catastrophe modeling, and insurance market development. She has published numerous articles and is editor of Risk Management and Insurance Review. She has served as an expert witness and has provided consulting services on a wide range of cases relating to insurance and liability issues. Dr. Born is the current president of the Eastern European Risk and Insurance Association and has held leadership roles in many other academic organizations, including the American Risk and Insurance Association, the Asia-Pacific Risk and Insurance Association, the Western Risk and Insurance Association, and the Risk Theory Society. Wouter Botzen is full professor of economics of climate change and natural disasters at the Department of Environmental Economics, Institute for Environmental Studies (IVM), Vrije Universiteit Amsterdam (VU), and full professor of economics of global environmental change at the Utrecht University School of Economics, Utrecht University. His main research interests are climate change economics with a particular focus on risk, natural disaster insurance, viii

Contributors  ix behavioral economics of decision-making under risk, and natural disaster risk assessment and management. He has published more than 100 articles in international scientific journals on these themes. He twice won the best paper award from the Risk Analysis journal, won the best paper award for Journal of Flood Risk Management, and received the Lloyd’s Science of Risk Prize in 2014. Christopher J. Boudreaux is an associate professor of economics at Florida Atlantic University and a research fellow for the Phil Smith Center for Free Enterprise. His research interests include entrepreneurship, innovation, and the economic analysis of public policy. He received his PhD and MS from Florida State University and his BS from Nicholls State University. Eduardo A. Cavallo is a principal economist at the Research Department of the InterAmerican Development Bank (IDB) in Washington DC and adjunct professor at The Johns Hopkins University–Paul H. Nitze School of Advanced International Studies (SAIS). Prior to joining the IDB, Eduardo was a vice-president and senior Latin American economist for Goldman Sachs in New York. Eduardo had already worked at the IDB as a research economist between 2006 and 2010. Before that, he served as a research fellow at the Center for International Development (CID), a visiting scholar at the Federal Reserve Bank of Atlanta, and a member of the faculty at the Kennedy School of Government’s Summer Program. He holds a PhD in public policy, an MPP from Harvard University, and a BA in economics from Universidad de San Andres (UdeSA) in Buenos Aires, Argentina. Marc N. Conte is an applied microeconomist who studies topics in environmental economics, including climate change, ecosystem services, and natural disasters. He uses a variety of tools, including parametric and nonparametric econometrics, applied theory, laboratory experiments, and randomized field experiments, to explore his questions of interest. Alexander W. Craig is a visiting assistant professor of economics and business at Beloit College. Jesús Crespo Cuaresma is a professor of economics at the Vienna University of Economics and Business (WU), as well as director of economic analysis at the Wittgenstein Centre for Demography and Global Human Capital (WIC) and research scholar at the International Institute of Applied Systems Analysis (IIASA). His research interests are in the fields of applied econometrics, macroeconomics, economic growth, human capital, and economic policy. He has published extensively in renowned scientific journals such as Science, Proceedings of the National Academy of Sciences, Nature Communications, Nature Climate Change, Demography, European Economic Review, and Journal of Applied Econometrics, just to name a few. Tatyana Deryugina is an associate professor of finance at the University of Illinois. She is also a co-editor at the Journal of the Association of Environmental and Resource Economists (JAERE) and at Environmental and Energy Policy and the Economy. She serves on the board of editors of AEJ: Policy. She is affiliated with the National Bureau of Economic Research, the Institute for the Study of Labor (IZA), the E2e Project, and the CESifo Research Network. Professor Deryugina holds a PhD in economics from MIT, a BA in applied mathematics from UC Berkeley, and a BS in environmental economics and policy from UC Berkeley.

x  Handbook on the economics of disasters Noah Dormady is an associate professor at The Ohio State University’s John Glenn College of Public Affairs. His research focuses on the economics of energy and environmental markets and systems, and economic resilience. He is a research fellow at two US Department of Homeland Security Centers of Excellence: the Critical Infrastructure Resilience Institute (CIRI) at the University of Illinois, and the Center for Risk and the Economic Analysis of Threats and Emergencies (CREATE) at the University of Southern California. His work has been published in a broad array of government publications and academic peer-reviewed journals, including Energy Economics, The Energy Journal, Risk Analysis, Bulletin of the American Meteorological Society, International Journal of Production Economics, Natural Hazards Review, Journal of Public Policy, and Journal of Commodity Markets. He received his PhD from the University of Southern California. Monica Escaleras is a professor of economics at Florida Atlantic University and chair of the economics department. Her research areas include economics of natural disasters, political economy, development economics, and entrepreneurship. She received her PhD from Florida International University and her MS from University of Florida. Marina Eurich is a PhD student in economics at the Helmut Schmidt University and is working as a research assistant at the chair of Political Economy and Empirical Economics. Before that, she received a MSc in economics from the University of Heidelberg. Her research interest is in the consequences of climate change. Her recent articles have focused on the effects of extreme weather risks on individuals’ well-being by combining geo-referenced survey data with extreme weather data. Stefanie Haeffele is senior research fellow, senior program and operations director of academic and student programs, and senior fellow with the F. A. Hayek Program for Advanced Study in Philosophy, Politics, and Economics at the Mercatus Center at George Mason University. She is the coauthor of Community Revival in the Wake of Disaster (Palgrave Macmillan, 2015). Stéphane Hallegatte is the senior climate change advisor of the World Bank Climate Change Group. He joined the World Bank in 2012 after ten years of academic research. His research interests include the economics of natural disasters and risk management, climate change adaptation, urban policy and economics, climate change mitigation, and green growth. Mr. Hallegatte was a lead author of the 5th Assessment Report of the Intergovernmental Panel on Climate Change (IPCC). He is the author of dozens of articles published in international journals in multiple disciplines and of several books. He also led several World Bank reports, including Shock Waves: Managing the Impacts of Climate Change on Poverty in 2015, Unbreakable: Building the Resilience of the Poor in the Face of Natural Disasters in 2016, and Lifelines: The Resilient Infrastructure Opportunity in 2019. He also led the writing team of the Stern-Stiglitz High-Level Commission on Carbon Prices. He was the team leader for the first World Bank Group Climate Change Action Plan, a large internal coordination exercise to determine and explain how the Group will support countries in their implementation of the Paris Agreement. In 2018, he received the Burtoni Award for his work on the link between climate change adaptation and poverty reduction. Mr. Hallegatte holds an engineering degree from the Ecole Polytechnique (Paris) and a PhD in economics from the Ecole des Hautes Etudes en Sciences Sociales (Paris).

Contributors  xi Anand Jha is an associate professor of finance at Wayne State University. His research interests include corporate finance, social capital, and behavioral finance. He received his PhD from Kelley School of Business of Indiana University. J. Bradley Karl is an associate professor in the Department of Risk Management/Insurance, Real Estate and Legal Studies at Florida State University. He received his PhD from Florida State University. Dr. Karl teaches courses in risk management, insurance, and corporate finance. His research interests include insurance regulation, medical malpractice liability, distracted driving, and catastrophes, and Dr. Karl has published numerous academic articles. Dr. Karl is the current president of the Southern Risk and Insurance Association and is also a member of the American Risk and Insurance Association and the Risk Theory Society. Prior to joining Florida State University, Dr. Karl was department chair and IIANC-NCSLA W. Kurt Fickling Distinguished Associate Professor in Risk Management & Insurance in the Department of Finance and Insurance at East Carolina University. Bhagyashree Katare is an associate professor in the Department of Agricultural Economics, Purdue University. Her current research program focuses on empirical analysis of observational and experimental data to study the effects of policies, behavioral nudges, and economic interventions on individual health behavior and labor outcomes. Her research involves designing behavioral economic interventions to motivate healthy behavior and implementing nudges in a low-cost and non-invasive manner. This is important from a policy perspective, as employers, governmental policy makers, and other stakeholders are looking for low-cost and effective methods to motivate positive labor, health, and employment outcomes in individuals. Scott Knowles currently leads FP&A and data analysis at the Bill of Rights Institute. Before joining the Institute, he held FP&A and data analysis roles in the research, education, and sports apparel industries. He earned his MS in agricultural, food, and resource economics from Michigan State University and his BS in economics from Towson University. Howard Kunreuther is the James G. Dinan Professor Emeritus of Decision Sciences and Public Policy, and Co-Director Emeritus of the Wharton Risk Management and Decision Processes Center at the University of Pennsylvania. Howard has a long-standing interest in ways that society can better manage low-probability, high-consequence events related to technological and natural hazards. He is a fellow of the American Association for the Advancement of Science, a distinguished fellow of the Society for Risk Analysis, and recipient of the Shin Research Excellence Award from the Geneva Association and International Insurance Society in recognition of his outstanding work on the role of public-private partnerships in mitigating and managing risks. Recent books include The Ostrich Paradox: Why We Underprepare for Disasters (with R. Meyer, Wharton School Press) and Mastering Catastrophic Risk: How Companies Are Coping with Disruption (with M. Useem, Oxford University Press). Maria I. Marshall is a professor and Jim and Lois Ackerman Endowed Chair in Agricultural Economics, in the Department of Agricultural Economics. She is the director of the North Central Regional Center for Rural Development and director of the Purdue Institute for Family Business. Dr. Marshall has a nationally and internationally recognized extension, research, and teaching program focused on small and family business development. Her program thrust is

xii  Handbook on the economics of disasters to increase the viability and sustainability of small and family businesses as they develop and mature through their life cycles. Her research provides relevant information and publications to entrepreneurs, family business owners, and policy makers. C. Blain Morin is a PhD candidate in public policy and management at The Ohio State University’s John Glenn College of Public Affairs. He conducts research in economic resilience, energy policy, and environmental policy. He holds a MSc degree in biostatistics from Brown University and a BS degree in economics from The Ohio State University. Ilan Noy is the chair in the Economics of Disasters and Climate Change at Victoria University of Wellington, New Zealand. His research and teaching focus on the economic aspects of hazards, disasters, and climate change, as well as other related topics. He is the founding editor-in-chief of the Economics of Disasters and Climate Change journal. Yasuhide Okuyama is a professor at the University of Kitakyushu, Japan. He earned his doctoral degree in regional planning from the University of Illinois at Urbana-Champaign in 1999. He also holds master’s degrees from the University of Wisconsin-Madison (Urban and Regional Planning, 1994) and from the University of Tsukuba, Japan (Environmental Science, 1986). His primary research interests center on economic impact of disasters, regional science, input-output analysis, and urban and regional planning. He has published a number of articles in various academic journals and book chapters, and edited books, including Modeling Spatial and Economic Impacts of Disasters in 2004 with Professor Stephanie Chang of the University of British Columbia and Advances in Spatial and Economic Modeling of Disaster Impacts in 2019 with Professor Adam Rose of the University of Southern California. In addition, he has been contributing to research projects and consultation for organizations such as the World Bank, European Commission, Economic Research Institute for ASEAN and East Asia (ERIA), and Japan Bank for International Cooperation. In January 2022, he became the co-editor of Economic Systems Research, the international scholarly journal of the International InputOutput Association (IIOA). Adam Rose is a research professor in the University of Southern California Sol Price School of Public Policy, and Director Emeritus of USC’s Center for Risk and Economic Analysis of Threats and Emergencies (CREATE). He has spearheaded the development of CREATE’s comprehensive economic consequence analysis framework and has done pioneering theoretical and empirical research on resilience at the level of the individual business, industry, and regional/national economy. He has also completed dozens of case studies of disaster consequences, resilience, and recovery, including the September 11 terrorist attacks, several seaport disruption scenarios, and COVID-19. Professor Rose is the author of several books and more than 250 professional papers. He is the recipient of several honors and awards, including the Distinguished Research Award from the International Society for Integrated Risk Management (IDRiM), East-West Center Fellowship, American Planning Association Outstanding Program Planning Honor Award, Applied Technology Council Outstanding Achievement Award, Regional Economic Models Outstanding Economic Analysis Award, and DHS/CREATE Transition Product of the Year Award. He is also an elected fellow of the Regional Science Association International. Yasuyuki Sawada is currently a professor at the Faculty of Economics, University of Tokyo, Japan. From 2017 to 2021, he was chief economist of the Asian Development Bank (ADB).

Contributors  xiii His key research areas are development economics, economics of disasters, and field surveys and experiments. Professor Sawada obtained his PhD in economics from Stanford University. Mark Skidmore serves as professor and Morris chair in State and Local Government Finance at Michigan State University. He has joint appointments in the Department of Agricultural, Food and Resource Economics and the Department of Economics. He is a distinguished scholar at the Lincoln Institute of Land Policy and a fellow with the CESifo Institute. Professor Skidmore conducts research on public economics, regional economics, and the economics of natural disasters. Virgil Henry Storr is an associate professor of economics at George Mason University and the Don C. Lavoie senior fellow with the F. A. Hayek Program for Advanced Study in Philosophy, Politics, and Economics at the Mercatus Center at George Mason University. Swenja Surminski is the head of adaptation research at the Grantham Research Institute on Climate Change and the Environment, part of the London School of Economics and Political Science (LSE), overseeing research projects on climate risk management, sustainable finance, climate, and resilience strategies by applying a mix of interdisciplinary approaches. A political scientist and ecological economist by training, her work embraces environmental, social, and economic perspectives, delivering cutting-edge research with decision-making support across a wide range of stakeholder groups in industry and government. Swenja has published widely on these topics and is a contributing author to the IPCC and the EU Science for Disaster Risk Management Report, and lead author of the UK’s Climate Change Risk Assessment. Her work focuses on capacity building, translation, and knowledge transfer between science, policy, and industry, building on her ten-plus years of experience working in the insurance industry and as advisor to governments, private sector, and civil society, including as visiting academic at the Bank of England and chair of the Munich Climate Insurance Initiative. She appears regularly in print, TV, and online media. Richard S. J. Tol is a professor at the Department of Economics, University of Sussex; a professor of the economics of climate change, Institute for Environmental Studies and Department of Spatial Economics, Vrije Universiteit, Amsterdam, the Netherlands; and a research fellow of the Tinbergen Institute, CESifo, and the Payne Institute for Public Policy. He is an elected member of the European Academy. He holds an MSc in econometrics (1992) and a PhD in economics (1997) from the Vrije Universiteit Amsterdam. He specializes in the economics of energy, environment, and climate, and is interested in integrated assessment modeling. He is ranked among the top 100 economists in the world and among the top 100 most-cited climate scholars. He is the editor-in-chief for Energy Economics, a top field journal. He is the author of the only textbook on the economics of climate change. He is advisor and referee of national and international policy and research. He was an author of Working Groups I, II, and III of the Intergovernmental Panel on Climate Change, shared winner of the Nobel Peace Prize for 2007. Yoshiro Tsutsui is a professor of economics at Kyoto Bunkyo University. He obtained a PhD in economics from Osaka University in 1989. He previously taught at Konan University, Osaka University, and Nagoya City University. He has served as president of the Association of Behavioral Economics and Finance (2008–2009), president of the Japan Society of Monetary Economics (2008–2010), and as a visiting scholar at Yale University, UC San

xiv  Handbook on the economics of disasters Diego, and University of Amsterdam. He was also a research fellow at the Institute for Posts and Telecommunications Policy, the Research Institute of Economy, Trade and Industry, and the Policy Research Institute. His primary areas of teaching and research are behavioral economics, economics of happiness, and banking and finance. He was awarded the Nikkei Economic Book Prize in 1988, Zengin Foundation Award in 2001, and Japan Post President’s Award in 2007. Tomáš Uher has a master’s degree in business administration from the Czech University of Life Sciences in Prague, Czech Republic (awarded in 2016). He has been working as a researcher at the School of Economics and Finance at Victoria University of Wellington since 2020. Corinne B. Valdivia is a professor of agricultural and applied economics, and holds the D. Howard Doane Endowed Professorship in Agricultural Economics. A founding fellow of Cambio Center at the University of Missouri, she also codirects Cambio Center’s research, engagement, and education on Latinxs, and changing communities in the Midwest. The Center organizes the Cambio de Colores Conference. Dr. Valdivia’s disciplinary, interdisciplinary, and translational research focuses on how individuals, businesses, and communities adapt to changes driven by innovations, globalization, migration, and climate change. Her research on the agency of small businesses and households and their communities in the US, Latin America, and Africa seeks to inform on best practices and policy. Brian Walsh is a microeconomist with the World Bank’s Climate Change Group. He uses household surveys to predict resilience of the poor to shocks, for promotion of welfare, and decarbonization. Brian previously worked as a data scientist in disaster risk management and holds a PhD in particle physics. Eiji Yamamura is a professor of economics at Seinan-Gakuin University. He earned his BA and MA from Waseda University and a PhD from Tokyo Metropolitan University in 1995, 1999, and 2004, respectively. He has been the executive director of the Association of Behavioral Economics and Finance since 2021. His research topics are behavioral economics, sports economics, income distribution, and household behavior. He has published more than 100 papers in peer review journals, including Review of Economics and Statistics, Journal of Economic Behavior and Organization, Journal of Population Economics, Journal of Economic Geography, European Journal of Political Economy, Review of World Economics, Review of International Economics, Economics of Education Review, Public Choice, Economics of Governance, Southern Economic Journal, Review of Economics of the Household, Kyklos, Journal of Cultural Economics, Journal of Sports Economics, Social Science and Medicine, Sustainability, and Journal of Japanese and International Economies. Chisako Yamane is an associate professor of economics at Hiroshima University of Economics. She received her PhD in economics from Hiroshima University in 2007. She taught economics at Niigata Sangyo University for five years and was an associate professor at Okayama Shoka University for four years prior to assuming her current position in 2018. Shoko Yamane is CEO of Papalaka Research Institute, Ltd. Her research interests are in behavioral economics. She conducts research in the areas of peer effects and subjective happiness. She received her BA in psychology from Ritsumeikan University and her PhD in economics from Osaka University. She was an associate professor of faculty of economics

Contributors  xv in Kindai University from 2012 to 2019. In January 2020, she started her own business, Papalaka Research Institute, Ltd. She has also written several books, including Kyou kara tsukaeru Koudou Keizaigaku (Behavioral Economics that you can use from today) with Hirofumi Kurokawa, Shusaku Sasaki, and Youki Kohsaka, and Koudou Keizaigaku Nyumon (Introduction of Behavioral Economics) with Yoshiro Tsutsui, Shunichiro Sasaki, and Grzegorz Mardyla.

Acknowledgements I would like to thank all the contributors for their dedication, knowledge, and commitment to sharing the research in this handbook. I also thank my family for their patience with me as I worked on this project! I thank my longtime friend and coauthor in disaster economics research, Hideki Toya. We’ve worked together for 20 years. It has been my pleasure and honor. Finally, I would like to express my deepest appreciation for the professionalism and excellent support provided by the Elgar editorial team.

xvi

1. Introduction to the Handbook on the Economics of Disasters Mark Skidmore

1. INTRODUCTION This volume addresses developments in research on the economics of natural disasters: (1) theoretical considerations in modeling impacts and decision-making; (2) methods and applications for evaluating disaster impacts; and (3) approaches to evaluating and improving risk management, resilience, and adaptability. Theoretical considerations include behavioral consequences of disasters, biases and heuristics, risk preferences, and modeling the impacts of disasters. Empirical applications and discussion include assessments of disaster impacts on (1) economic growth, (2) government fiscal conditions, (3) well-being, (4) and migration. Topics in risk assessment and reduction include the following: (1) evaluation using agent-based modeling, (2) risk management and reduction, (3) weather modification to subdue severe weather, and (4) the use of financial markets for disaster mitigation. Also included are chapters on disaster vulnerability, resilience, recovery, and response. Immediately following this introduction, Chapter 2 provides an overview of disaster types and a history of what we know about catastrophic events that have occurred over time. This chapter provides a unifying backdrop for the remaining chapters. Chapter 2 is followed by a set of chapters offering discussions of important theoretical considerations in the evaluation of disaster impacts on economic activity. Part II of the volume offers chapters that focus on methods and applications used in disaster impact evaluation. Finally, Part III addresses disaster risk management, resilience, adaptation, and vulnerability.

2. EMERGENCE OF THE ECONOMICS OF NATURAL DISASTERS AS A FIELD OF STUDY With the exception of the substantial literature on the role of risk in decision-making,1 prior to 1990, economists had conducted relatively little research to evaluate the broader implications of disasters on various aspects of economic activity. Further, little work in economics had been conducted to evaluate the underlying socioeconomic determinants of disaster vulnerability and resilience. Albala-Bertrand (1993) is perhaps the seminal work on this aspect of the economics of natural disasters. He emphasized the role of social processes and political economy within countries to improve understanding of natural disaster impacts and vulnerability. The research of Zeckhauser (1996) is also important early work. Two factors seemed to have served as catalysts for increased interest among economists in the role disasters play in economics. The first is that some very large disaster events, including the Great Hanshin earthquake of 1995 and the Asian tsunami of 2004, captured the attention of economists. The Asian tsunami is estimated to have caused the most loss of life in modern 1

2  Handbook on the economics of disasters history, with 230,000 fatalities coupled with US$8.7 billion in damages (Fenn, 2014). These events, combined with increased computing power and data/statistical capacities, enabled economists to examine disaster consequences using more systematic and robust methods. In the wake of the Great Hanshin quake, Skidmore and Toya (2002) offered one of the first articles that examined the long-run relationship between economic growth and disaster exposure. Within a long-run macroeconomic framework, Skidmore and Toya identified positive relationships between a country’s propensity for climatic disasters and factor productivity, human capital, and economics growth. This somewhat surprising finding spurred additional research on the topic, including three chapters in the present volume. Since that time, researchers from a variety of subfields within economics have examined the multitude of ways disasters affect economic activity. See Skidmore and Lim (2020) for a collection of the most prominent research on the economics of natural disasters covering growth impacts; instability; monetary and financial flow impacts; regional impacts; and the determinants of vulnerability, resilience, recovery, and adaptation. Extreme and potentially catastrophic climatic, geologic, and galactic events are a part of the earth’s natural systems. Evidence on the earth’s surface from ancient catastrophic events are abundant. As discussed in Chapter 2 of this book, this evidence takes the form of huge craters left by meteorites, ancient coral reefs in cold-water pole regions, huge ancient lake beds in the Sahara Desert, massive Mammoth graveyards flash frozen in mud and ice, evidence of warm climate flora in the north and south poles, massive floods, ancient city ruins that are now under water, and so on. Compared to these ancient catastrophic events, natural disasters in modern human history are relatively small scale. Nevertheless, individuals, communities, nations, and the globe are exposed to significant life-changing . . . and economy-changing catastrophic events every year. Prior to the industrial revolution, the impact of disasters on capital stock among the poor was negligible. The modest shelters they built could easily be replaced. However, lacking knowledge of more sophisticated engineering principles, the wealthy could and did spend huge sums building structures that could withstand forces far in excess of likely disaster shocks (Alexander, 1993). Despite the major improvements in engineering and knowledge in modern times, the potential for loss of life and capital destruction is enormous. While the 2004 Indian Ocean tsunami was one of the most significant disaster events in modern history, the death toll of the 1918 Spanish Flu pandemic was far higher at an estimated 50 million (National Archive, 2022). Several societal factors play a role in determining vulnerability. Consider two earthquakes of similar magnitude: the Haiti quake in 2010 and the Japan quake in 1995. The Haiti quake resulted in 300,000 fatalities, but fatalities from the Japan quake were 6,434. The difference in the impacts is primarily due to the quality of infrastructure and buildings, as well as the level of economic development and government policies regarding preparedness and recovery (Skidmore, 2013). The potential risks to life and property from disasters require individuals, businesses, and governments of the world to consider how best to prepare for and respond to disaster events. For a variety of reasons, making such decisions is a challenge. First, there is often a high degree of uncertainty in assessing both the probability and magnitude of events, which makes decision-making more difficult, thus hampering the formation of insurance markets. Such uncertainty can lead to under-preparation and underinvestment in disaster protection. Sometimes societal protections can lead to perverse incentives that lead to reduced preparedness. In the United States, more than 70 federal programs offer disaster assistance

Introduction to the handbook on the economics of disasters   3 to households, businesses, not-for-profit organizations, and subnational governments. These federal programs include direct grants to individuals and communities, low-interest loans, public works projects to remove debris and rebuild, disaster unemployment benefits, mental health and legal services, environmental cleanup, and federal income tax deductions for uninsured losses. These programs help offset the burden of local disaster recovery. However, as described in Skidmore (2013), the funding mechanisms for these programs are not risk adjusted, and thus disaster relief can significantly reduce incentives to implement appropriate risk-reduction measures. The risks associated with building a vacation home on an exposed shoreline is not as high if one recognizes that federal government assistance is available should a hurricane strike. Similar perverse incentives exist for subnational governments too. Few would argue that governments should eliminate societal protections. Further, governments can and do implement policies to offset the misaligned incentives. For example, the Federal Emergency Management Agency (FEMA) requires that subnational governments develop hazard mitigation plans as a condition of accepting grant funding. In addition, risk perceptions and preferences can potentially change for those who experience catastrophic events, which can lead to over- or under-reinvestment in the aftermath of disasters. Economists have generally assumed that preferences are very stable. However, recent research demonstrates that significant shocks may alter a person’s risk preference profile in the short run and perhaps in the longer run too. In the context of disasters, people who have experienced significant shocks may have reduced risk tolerance, which may lead to suboptimal investment decisions. See Chapter 6 for a detailed discussion of these issues. Before turning to a discussion of key questions and a summary of the chapters in the handbook, consider the following: This is not a book about climate change. Many climate scientists have pointed to the likelihood that both the number and magnitude of climatic disasters could increase due to climate change. While this is a possibility, this book does not focus on climate change or the potential implementation of offsetting actions such as geoengineering. Rather, this book focuses on disasters, including disaster mitigation efforts, vulnerability, resilience, and their implications for economic activity. Second, we must acknowledge that the many facets of the interaction between the natural environment and human activity necessitate a multidisciplinary perspective. While economists have much to offer in improving our understanding of natural disasters and helping societies better prepare for and respond to disasters, much can be gained through partnerships with researchers from other disciplines. 2.1  Important Questions Economics of pandemics The COVID-19 crisis is at the forefront for many economists who conduct research in the disaster arena. Chapter 7 of this handbook provides an excellent summary of how past pandemics affected economic activity. Pandemics have resulted in more lives lost than any other type of disaster over the past 150 years. The magnitude of fatalities is primarily due to the Spanish Flu of 1918, which resulted in an estimated 50 million fatalities globally. The COVID-19 event is proving to be far less deadly but is still a substantial threat to the elderly and those with other underlying health conditions (Ioannidis, 2021). However, the way in which health authorities and governments have responded to the crisis has no precedent in history. Across the globe, governments closed schools, implemented work from home orders, and inhibited social and

4  Handbook on the economics of disasters economic activity. These measures were initially taken to slow the spread of the virus and thus prevent hospitals from being overwhelmed. In many places, the initial plan was to impose restrictions for a few weeks but were then extended again and again until a vaccine became available. Decisions regarding these lockdowns were based on polymerase chain reaction (PCR) testing, which had never been used for this purpose. To compensate for the slowdown in economic activity, governments in the industrialized world provided generous subsidies that were largely paid for with monetary liquidity provided by central banks. The COVID-19 gene therapeutic vaccine was developed in record time—a matter of months instead of the typical multiyear process. The short vaccine development period is attributed to the fact that COVID19 vaccines are based on a new and never-before-used gene therapeutic technology whereby a person is injected with material that reprograms his or her cells to produce the spike protein component of the coronavirus. The person’s immune system then responds to the foreign substance being produced by the body by eliminating the cells producing the spike protein. The goal is that in recognizing the spike protein, the body will build immunity to the coronavirus. Another remarkable development is the imposition of vaccine mandates in many countries across the globe with the ultimate goal of vaccinating as much of the global population as possible. Importantly, the mandates have led to the dismissal of millions of workers for their refusal of the vaccine treatments. All these developments are now and will continue to be studied by researchers in the coming years. What were the benefits and costs of the socioeconomic restrictions? Will it be possible to disentangle the impact of the COVID-19 illness from the restrictions on economic activity? What are the benefits and costs of the vaccines relative to other treatment options? What are the benefits and costs of the vaccine mandates? Answers to these questions are essential for determining policies in potential future pandemics. Individual versus public preparation for and responses to disasters In general terms, society must strike a balance between individual and public responsibility. This is no less true for disasters. To what extent is it the responsibility of private households and businesses to take appropriate precautionary measures against disasters, and what role should the public sector play? What guiding social and economic principles shed light on striking a balance? If governments intervene, what principles should be used to guide those interventions? Is it best for national or subnational authorities to take on those disaster-related responsibilities? Under what conditions is a global/centralized response appropriate, and when is a decentralized strategy more effective? Economic frameworks such as pricing/incentives, information access, freedom of choice, externalities/spillovers, public good theory, competition, economies of scale (in the provision of safety measures and in the scope of risk), and regulatory capture all inform the choices societies make. Few would argue that private preparation only is optimal, and few would make the case that all authority for disaster management should be concentrated in a single centralized government entity. Most agree that there is an appropriate balance between individual responsibility and public/collective action. There is also an appropriate mix between nationwide or even global policies and decentralized governmental decision-making. Societal preparation for extinction-level events How does society prepare for rare catastrophic/extinction-level events? As discussed in Chapter 2 of this book, there is abundant evidence that the earth has experienced extinctionlevel catastrophic events in the past. Should civilization prepare for a major asteroid strike, a

Introduction to the handbook on the economics of disasters   5 pole shift, or a massive solar burst? In this context, the goal cannot be to save lives or reduce economic damages but to preserve species, including Homo sapiens. These issues are not often discussed in the public sphere (though there are plenty of movies on the topic), but there is some evidence that governments have taken action. Underground seed banks, DNA storage, expansive underground tunnels and facilities, space exploration, and technologies to divert incoming asteroids are all examples of governmental efforts that directly or indirectly increase the resilience of the species to extinction-level events. Spatial spillovers While there are now numerous studies that consider the spatial aspects of disaster impacts, there is still much to be learned.2 There is clear potential for spillovers to occur in communities near disaster-affected areas. However, spillovers can affect activity in other parts of a country or even globally as the input and product supply chains are global in nature. For example, a disaster crisis in Taiwan, a center of semiconductor production, could affect economic activity on a global scale. While individual businesses are aware of potential supply chain effects, broader understanding is needed on how global supply chains and distribution systems can be affected by disasters. Understanding the long-run impacts of disaster exposure The early work of Skidmore and Toya (2002) offered some evidence of how disasters might affect total factor productivity and human capital investment decisions. That study was one of the first to suggest that disaster propensity could affect long-run decision-making. More work along these lines is needed to understand how past disasters could affect behavior today. Major disaster experiences are often embedded in folklore and can affect an entire culture and thus influence economic structure and decision-making. However, more than just the culture could be affected. In the field of psychology, scientists have documented how traumatic events can alter DNA (Morath et al., 2014; Vinkers et al., 2015). Trauma can affect decision-making and negatively affect well-being, even causing shortened life spans. What’s more, the altered DNA can be transferred to offspring. Youssef et al. (2018) conclude that there is “evidence of an enduring effect of trauma exposure to be passed to offspring transgenerationally via the epigenetic inheritance mechanism of DNA methylation alterations.” What are the economic implications for societies or people groups that have experienced repeated trauma from disasters or other causes? While there is research on how trauma affects health expenditures and individual economic outcomes (Bothe et al., 2020; Loughran & Heaton, 2013), there is little research extending that type of evaluation to economic activity at community or societal levels. Scales of analyses, methods, and data Research on the economics of disasters has used nearly all the tools of economics: econometric analysis (cross-section, time series, and panel data methods), economic modeling (static and dynamic), survey and experimental approaches, agent-based modeling, and cost-benefit tools. Researchers have examined economic choices and behavior in a variety of contexts and levels, including the individual (households and businesses), community and regional, and national/ cross-country levels. One of the challenges in conducting empirical research on disasters is the quality and nature of disaster data. Many studies use disaster data drawn from insurance companies. A commonly used source is the EM-DAT database, which is very useful in evaluating human casualties and monetary damages. However, Felbermayr and Gröschl (2014) rightfully

6  Handbook on the economics of disasters point out that in the context of evaluating the effects of disasters on economic growth, the use of insurance data may lead to biased estimations because selection into the database may be correlated with gross domestic product (GDP). To address this issue, Felbermayr and Gröschl developed the “GeoMet” database that comes from primary geophysical and meteorological databases.3 In summary, depending on the scope and purpose of the disaster research, selection of the appropriate data source can be very important. Chapter 2 provides a detailed discussion of natural disasters and extreme events that have occurred in the recorded history and over millions of years. While historical documents provide records of significant catastrophic events that societies have experienced, we know from the geologic record that the earth has experienced a number of extinction-level events dating back millions of years. Disasters such as major storms, hurricanes/typhoons, or perhaps a significant earthquake are somewhat expected in one’s lifetime. However, events such as major asteroid strikes, major solar storms, or physical pole shifts occur very rarely. Societies have the challenge of preparing for and responding to not only relatively frequent and less severe (localized) events but also low-probability/high-consequence global events. The tendency is high for people to ignore such possibilities and not prepare for them at all. Chapter 2 offers a summary of the frequencies, magnitudes, durations, sources, and consequences of disasters, including rare extinction-level catastrophes that have occurred in the past. The chapter also provides an important baseline information context useful for the other chapters in this book, which are summarized in the following.

3.  STRUCTURE OF THE VOLUME This volume contains chapters covering topics in three general categories: (1) theoretical considerations in evaluating disaster impacts; (2) empirical evaluation of disaster impacts; and (3) risk management, resiliency, and vulnerability. 3.1  Theoretical Considerations in Evaluating Disaster Impacts Chapter 3, “A few good models for economic analysis of disasters: can your model handle the truth?” is an indispensable resource for researchers seeking how to conduct economic analyses of disasters. Economists and researchers from many disciplines have used a variety of modeling approaches to evaluate the impacts of disasters on economic activity. However, choosing an appropriate modeling design is critical to the success of any project, and it is sometimes a challenge. Okuyama draws on his deep knowledge of economic modeling in the context of disasters. This chapter provides an extensive review of recent advancements in modeling, offering a useful discussion of the advantages and limitations of alternative approaches as well as insights for both micro- and macro-level modeling. In Chapter 4, “Behavioral economic consequences of disasters,” Adam Rose describes an analytical framework for estimating the behavioral effects and consequences of disasters. One of the challenges in evaluating disaster is disentangling the effects of disaster impacts from behavioral responses. As one illustration, a pandemic can lead to significant economic impacts, but the effect can be exacerbated by undue fear. People may reduce consumption behavior as a result of fear above and beyond the direct pandemic impact. While the actual pandemic risk

Introduction to the handbook on the economics of disasters   7 could dissipate, peoples’ perceived risk could lead to additional ongoing negative economic consequences. Rose offers an excellent discussion of these types of behavioral impacts, outlining approaches for measuring and accounting for them as part of the disaster impact analysis. Howard Kunreuther and Wouter Botzen offer an excellent overview on the role of biases and heuristics in addressing natural disasters in Chapter 5. The discussion centers on the idea that those at risk from disaster exhibit biases that result in under-preparation for disasters. The chapter offers an assessment of these biases as well as approaches to encourage people to take appropriate loss-reduction measures. Chapter 5 dovetails with Chapter 6, “Risk preferences and natural disasters: a review of theoretical and empirical themes” by Laura Bakkensen and Marc N. Conte who offer a review of the prominent models of risk preferences and empirical tools for eliciting preferences. They also discuss how risk preferences influence insurance purchases, location decisions, and the adoption of adaptive technologies. They close the chapter by examining how risk preferences may actually be altered by disaster experiences, making the case that changing risk preferences is an understudied issue in disaster mitigation and societal resilience. Chapter 7, “Economic consequences of pre-COVID-19 epidemics: a literature review,” offers a very useful review of the literature on the economic consequences of past pandemics. In 2020, the world was struck by the COVID-19 crisis. With the backdrop of the current crisis, the chapter summarizes the research on how highly infectious pathogens such as influenza or Ebola affect economies. This chapter, coupled with Chapter 4, is essential reading for those interested in conducting evaluation of the current COVID-19 crisis. They show that epidemics can and do have interconnected economic consequences through both supply- and demandside channels. They also discuss how the behavioral response of individuals, communities, societies, and governments is perhaps even more important than the impacts of disease prevalence and progression itself. 3.2  Evaluation of Disaster Impacts This part of the book begins with three applied chapters that evaluate the impacts of disasters on economic growth. There are now numerous articles that empirically evaluate the impacts of disasters on economic activity, and the conclusions of this research range from negative, to neutral, to positive growth impacts. Chapter 3 is very informative in this regard as is Chapter 2. The conclusions of the many studies examining the growth impacts of disasters depend on a variety of factors. Does the research focus on one or more highly developed countries with effective government programs and well-functioning insurance markets? Or is the focus on one or more developing countries that may lack such resources? Does the evaluation assess the short-run or long-run growth impacts? Does the examination focus on the effects of disaster events, or does it focus on how underlying exposure to disaster risk over the long run may affect culture and thus saving and investment patterns? How does one decide which extreme events should be included in the evaluation? Within a cross-country framework, if one chooses the number of fatalities as a key factor in the disaster selection process, then one is likely to weigh more heavily the experiences of developing countries because industrialized countries tend to experience far fewer fatalities. Chapter 2 also provides a discussion of the challenges associated with potential data sources and the biases that may be present in the selection process. This section of the volume also includes three chapters that offer evaluations

8  Handbook on the economics of disasters of the fiscal and social consequences of natural disasters. With this summary of some of the challenges associated with assessing the impacts of disasters on economic growth, the next chapters turn to a discussion of three disaster-growth studies within distinct contexts, resulting in three different outcomes. Economic growth impacts Chapter 8, “Natural disasters and economic growth: revisiting the evidence,” uses Bayesian model averaging techniques to thoroughly examine the association between natural disaster risk and GDP per capita in a long-run context. The evaluation suggests that there is no consistent relationship between disasters and growth over the 1970–2019 period for 123 countries. In contrast, in Chapter 9, “The impact of natural disasters on economic growth,” Acevedo et al. (2022) estimate the impacts of large disasters on economic growth over the 1970–2019 period. Here, Chapter 9 first focuses on larger events as measured by mortality, which means the emphasis is on developing countries. In this portion of the evaluation, the chapter finds negative effects. However, when physical intensity is used to determine inclusion of the disasters in the evaluation, the relationship between disasters and growth is negligible. This finding suggests that negative effects of natural disasters are more likely to be observed in developing countries. In contrast, Chapter 10, “Assessing the impact of natural disasters on industry gross domestic product in the United States,” finds positive effects in the context of an industrialized country. Here, the authors used county-level panel data over the 2001–2014 period to evaluate the impacts disasters have on industry growth. They offer the explanation that counties may experience positive growth effects following a disaster as federal government assistance and insurance funds flow in as part of the recovery/rebuilding process. Each of these studies is well done but offer seemingly inconsistent results, and yet the factors driving the differences across these studies increase our understanding: disasters can have differential effects depending on conditions, circumstances, study scope, and study design. Fiscal and social impacts In Chapter 11, “The fiscal consequences of natural disasters,” Deryungina offers a detailed discussion of the fiscal consequences of natural disasters on governments as well as how government policy plays an important role in recovery. This excellent chapter summarizes the literature evaluating the dynamics of how government budgets are affected by major disasters. At the country level, empirical research consistently shows that major disasters reduce tax revenue and increase spending, serving to create a temporary fiscal imbalance and fiscal stress. However, more research is needed to determine the returns to government mitigation efforts and the effect of post-disaster expenditures on economic activity. Another line of disaster research evaluates the degree to which disasters affect individual and societal well-being. In Chapter 12, “Natural disasters and self-reported well-being: a review of the literature,” offers an overview and review of the literature on the relationship between natural disasters and self-reported well-being. Generally, the evidence suggests that disasters have a negative effect on well-being, but the effect tends to be short-lived for most people. Interestingly, the authors also show how changes in self-reported well-being resulting from disasters can be used to quantify disaster losses. While this line of inquiry has expanded in recent years, additional research is needed, particularly within a panel data approach, to better identify the linkages between disaster and self-reported well-being.

Introduction to the handbook on the economics of disasters   9 Chapter 13, “Preferences, behavior, and welfare outcomes against disasters: a review,” offers an excellent discussion of the relationships between preferences, behavior, and welfare outcomes in relation to disasters. This chapter begins with a discussion of the disaster-poverty nexus, highlighting the vulnerability of the poor. The author then turns his attention to risk preferences and risk-coping behaviors in the context of physical, financial, human, and social capital assets. The chapter also provides a discussion of the role of disasters in forming individual and societal preferences, as well as the use of insurance mechanisms and other protective actions against disasters. The chapter concludes with a summary of the nexus between disaster exposure, changing preferences, changing individual and social responses, behavioral responses, and the overall welfare consequences of disaster exposure. Substantial literature is also available on disasters and migration, which is reviewed in Chapter 14, “Effect of major disasters on geographic mobility intention: the case of the Fukushima nuclear accident.” In Chapter 14, a panel of survey data from Japan is used to examine the intentions of people to move in the wake of the Great East Japan Earthquake, and Tsunami, and the resulting Fukushima nuclear accident. The analysis offers a valuable illustration of how survey data can be used to evaluate decision-making in the wake of disasters, demonstrating that differences in both real and perceived risks are important factors in mobility intentions. 3.3  Risk Management, Resiliency, and Vulnerability This part of the book contains chapters on topics related to risk management (primarily insurance), risk assessment and risk reduction, dealing with low-probability/high-consequence disasters, resilience, vulnerability, and recovery. Insurance One of the primary methods of protecting property against disaster risk is insurance, provided that there are well-developed financial markets through which insurance can be provided. In Chapter 15, “The role of insurance in integrated disaster risk management with a focus on how insurance can support climate adaptation and disaster resilience,” the author examines the role of insurance in integrated risk management. This chapter offers a discussion of how insurance can play a role in encouraging increased investment in disaster resilience, enhancing adaptation to changing climate, and helping to avoid the creation of new risks. The chapter discusses a range of concepts and experiences of how insurance mechanisms can be used to facilitate integrated solutions for disaster risk management. As a complement to Chapter 15, Chapter 16, “Supplying insurance for natural disasters: a retrospective study of property insurer strategies,” offers a comprehensive evaluation of the fundamental challenges of providing disaster insurance. The authors discuss the factors that determine insurers’ willingness to provide coverage in a time when the frequency and intensity of disasters appear to be increasing. In general, the chapter offers a useful primer on the challenges insurers face in providing disaster insurance. Risk assessment and reduction Chapter 17, “Expanded disaster risk assessment using agent-based modeling: a case study on floods in Sri Lanka,” offers a case study in the use of agent-based modeling in disaster risk assessment. The approach enables researchers to more effectively incorporate socioeconomic

10  Handbook on the economics of disasters resilience and welfare losses into the evaluation. The researchers provide an example of how to use this approach by evaluating flood risks in Sri Lanka, showing that average annual losses in well-being are US$119 million annually, far more than the estimated US$78 million in annual losses. The chapter provides a road map for using agent-based modeling in disaster risk assessment. While there is a large body of research evaluating the effectiveness of weather modification, relatively few economists have knowledge of weather modification activities, which are occurring all over the world. Chapter 18, “Using weather modification to subdue severe weather,” provides a summary and general evaluation of weather modification techniques in the context of subduing severe weather. A variety of weather modification technologies are being used to reduce the severity of damages from extreme weather, such as hail, drought, and increased rainfall and snowpack. The chapter offers a useful summary of weather modification activity globally and presents a brief discussion of approaches for evaluating the benefits and costs of these programs. Resilience and vulnerability The idea of resilience is discussed at length in the disaster literature. This work points to its importance, and yet resilience is a challenge to define, let alone measure. Chapter 19, “Advances in the empirical estimation of disaster resilience,” is an insightful discussion of resilience, noting the differences between resilience, reliability, and mitigation. The authors make distinctions between adaptive versus inherent resilience, and static versus dynamic resilience, and they present the key empirical methods for measuring resilience. This chapter provides very useful guidance for researchers interested in assessing resilience, especially at the microeconomic level. Vulnerability is another extensively studied topic in disaster research. Chapter 20, “State capacity and vulnerability to natural disasters,” presents an evaluation of the role of government in reducing the vulnerability of at-risk populations. This chapter reviews the literature on the role of government in reducing (or exacerbating) vulnerability. Generally, the body of research has shown that government quality and democratic institutions are associated with reduced vulnerability, whereas public corruption increases vulnerability. Following this review, the chapter offers further examination of vulnerability by considering the degree to which the structure of public finance influences harm from disasters. He finds that countries with a larger public sector are better able to prevent extreme events from doing harm and that countries which rely more heavily on income taxes tend to be less vulnerable. This chapter is an important resource for those seeking to better understand the role of government in reducing disaster vulnerability. Recovery and response Many researchers rely on interview and survey methods to assess conditions in the aftermath of disasters. Surveys can be conducted in a variety of venues and be targeted at households, businesses, and governments. Chapter 21, “Small business recovery: lessons from Hurricane katrina and the COVID-19 pandemic,” provides a review of the research on survey methods and applications in the context of disasters. The chapter also includes a survey-based evaluation of small business recovery in the aftermath of Hurricane Katrina, identifying factors such as US Small Business Administration disaster loans as important in the recovery process. While businesses recovery is critical in the aftermath of disasters, entrepreneurs also do much of the work of rebuilding following disasters, playing a significant role in recovery.

Introduction to the handbook on the economics of disasters   11 Sometimes businesses are criticized for price gouging in the wake of disasters. However, as discussed in Chapter 22, “Disaster challenges and entrepreneurial responses,” market process theory sets the stage for the idea that entrepreneurs respond to the potential for monetary profit and social reward by identifying major problems and solving them in the wake of disaster events. As one illustration, Horwich (2000) discusses the role of the Japanese mafia in providing critical supplies in the aftermath of the 1994 earthquake in Kobe, Japan. Though the role of entrepreneurship in the recovery is sometimes underappreciated, Chapter 22 demonstrates the importance of the entrepreneurial spirit in helping disaster victims manage uncertainty, repair damages, and negotiate the politics of disaster recovery. This chapter is an indispensable resource for those who want to increase their understanding of the role of business/entrepreneurs in the disaster recovery process.

NOTES 1. As examples see research on risk and portfolio choice (Hakansson, 1970; Merton, 1969; Sandmo, 1969), uncertainty related to income variance and savings decisions (Dréze & Modigliani, 1972; Dynan, 1993; Guiso et al., 1992; Kimball, 1990; Leland, 1968; Sandmo, 1970; Skinner, 1988; Zeldes, 1989), insurance and behavioral responses to risk and uncertainty (Kunreuther et al., 1995; Kunreuther, 1996), and economic responses to risks from natural disasters (Brookshire et al., 1985). 2. See, for example, LeSage et al. (2011), who evaluated business reopening in New Orleans following Hurricane Katrina, documenting the importance of spatial interdependence of reopening decisions. Others have documented the importance of spatial relationships within industry sectors. Examples include agriculture (Debolini, 2013), manufacturing (Lee et al., 2018), and hospitality (Xue et al., 2020). 3. Information on how to access GeoMet is available in Felbermayr and Gröschl (2014).

REFERENCES Albala-Bertrand, J. M. (1993). Political economy of large natural disasters: With special reference to developing countries. OUP Catalogue. Alexander, D. (1993). Natural Disasters. London: University College London Press. Axfors, C., & Ioannidis, J. P. (2021). Infection fatality rate of COVID-19 in community-dwelling populations with emphasis on the elderly: An overview. medRxiv. Retrieved from: https://www.medrxiv.org/content/10.1101/2021. 07.08.21260210v1. Accessed on October 25, 2021. Bothe, T., Jacob, J., Kröger, C., & Walker, J. (2020). How expensive are post-traumatic stress disorders? Estimating incremental health care and economic costs on anonymised claims data. The European Journal of Health Economics, 21, 917–930. Brookshire, D. S., Thayer, M. A., Tschirhart, J., & Schulze, W. D. (1985). A test of the expected utility model: Evidence from earthquake risks. Journal of Political Economy, 93(2), 369–389. Debolini, M., Marraccini, E., Rizzo, D., Galli, M., & Bonari, E. (2013). Mapping local spatial knowledge in the assessment of agricultural systems: A case study on the provision of agricultural services. Applied Geography, 42, 23–33. Dréze, J., & Modigliani, F. (1972). Consumption decisions under uncertainty. Journal of Economic Theory, 5, 308–335. Dynan, K. (1993). How prudent are consumers? Journal of Political Economy, 101, 1104–1113. Felbermayr, G., & Gröschl, J. (2014). Naturally negative: The growth effects of natural disasters. Journal of Development Economics, 111, 92–106. Fenn, C. (2014, December 25). The human and financial cost of the India Ocean Tsunami – Interactive. The Guardian. Retrieved from: https://www.theguardian.com/global-development/ng-interactive/2014/dec/25/human-financialcost-indian-ocean-tsunami-interactive. Accessed on October 25, 2021. Guiso, L., Jappelli, T., & Terlizzese, D. (1992). Earnings uncertainty and precautionary saving. Journal of Monetary Economics, 30, 307–337. Hakansson, N. (1970). An induced theory of the firm under risk: The pure mutual fund. Journal of Financial and Quantitative Analysis, 5(2), 155–178.

12  Handbook on the economics of disasters Horwich G. (2000). Economic lessons of the Kobe earthquake. Economic Development and Cultural Change, 48, 521–542. Kimball, M. (1990). Precautionary saving in the small and in the large. Econometrica, 58, 53–73. Kolassa, I. T. (2014). Effects of psychotherapy on DNA strand break accumulation originating from traumatic stress. Psychotherapy and Psychosomatics, 83(5), 289–297. Kunreuther, H. (1996). Mitigating disaster losses through insurance. Journal of Risk and Uncertainty, 12, 171–187. Kunreuther, H., Meszaros, J., Hogarth, R., & Spranca, M. (1995). Ambiguity and underwriter decision processes. Journal of Economic Behavior and Organization, 26, 337–353. Lee, K. H., Kang, S., Terry, W. C., & Schuett, M. A. (2018). A spatial relationship between the distribution patterns of hotels and amenities in the United States. Cogent Social Sciences, 4(1), 1444918. Leland, H. (1968). Saving and uncertainty: The precautionary demand for saving. Quarterly Journal of Economics, 82, 465–472. LeSage, J., Pace, R. K., Campanella, R., Lam, N., & Liu, X. (2011). Do what the neighbours do: Reopening businesses after Hurricane Katrina. Significance, 8(4), 160–163. Loughran, D. S., & Heaton, P. (2013). Post-traumatic stress disorder and the earnings of military reservists. Rand Health Quarterly, 3(3). Merton, R. C. (1969). Lifetime portfolio selection under uncertainty: The continuous-time case. The Review of Economics and Statistics, 51(3), 247–257. Morath, J., Moreno-Villanueva, M., Hamuni, G., Kolassa, S., Ruf-Leuschner, M., Schauer, M., & Kolassa, I. T. (2014). Effects of psychotherapy on DNA strand break accumulation originating from traumatic stress. Psychotherapy and Psychosomatics, 83(5), 289-297. National Archive. (2021). The flu pandemic of 1918. Retrieved from: https://www.archives.gov/news/topics/flupandemic-1918. Accessed on October 25, 2021. Sandmo, A. (1969). Capital risk, consumption, and portfolio choice. Econometrica, 37(4), 586–599. Sandmo, A. (1970). The effect of uncertainty on saving decisions. The Review of Economic Studies, 37(3), 353–360. Skidmore, M. (2013). Natural disaster preparedness and recovery: Issues and policy options. National Agricultural & Rural Development Policy Center. Retrieved from: https://aese.psu.edu/ . . . /policy-briefs/natural-disasterpreparedness-and-recovery. Accessed on October 25, 2021. Skidmore, M., & Lim, J. (2020). Natural disasters and their impact on cities. Oxford University Press. Retrieved from: https://www.oxfordbibliographies.com/view/document/obo-9780190922481/obo-9780190922481-0014.xml. Skidmore, M., & Toya, H. (2002). Do natural disasters promote long-run growth? Economic Inquiry, 40(4), 664–687. Skinner, J. (1988). Risky income, life-cycle consumption and precautionary savings. Journal of Monetary Economics, 22, 237–255. Vinkers, C. H., Kalafateli, A. L., Rutten, B. P., Kas, M. J., Kaminsky, Z., Turner, J. D., & Boks, M. P. (2015). Traumatic stress and human DNA methylation: A critical review. Epigenomics, 7(4), 593–608. Retrieved from: https://www. futuremedicine.com/doi/full/10.2217/epi.15.11?casa_token=eMFAIVoveG4AAAAA%3AiLKQ94T8HmUjp05aT Xc8giODDPF70LpFKvX4bpXaMuoTy51JSUkHEKh2jK7g22YWqqZ5JrVVOUya. Xue, B., Xiao, X., & Li, J. (2020). Identification method and empirical study of urban industrial spatial relationship based on POI big data: A case of Shenyang city, China. Geography and Sustainability, 1(2), 152–162. Youssef, N., Lockwood, L., Su, S., Hao, G., & Rutten, B. (2018). The effects of trauma, with or without PTSD, on the transgenerational DNA methylation alterations in human offsprings. Brain Sciences, 8(5), 83. Zeckhauser, R. (1996). The economics of catastrophes. Journal of Risk and Uncertainty, 12(2), 113–140. Zeldes, S. (1989). Optimal consumption with stochastic income: Deviations from certainty equivalence. Quarterly Journal of Economics, 104(2), 275–298.

2. A taxonomy of natural disasters Mark Skidmore

1. INTRODUCTION In this chapter, I provide an overview of the types of natural disasters humans have experienced over time to be used as a common definition and framework for “natural disaster” in all the chapters contained in this book. Having a common working definition is important because though there is a general understanding of what is meant by “natural disaster,” there is no precise and commonly accepted definition. What constitutes a natural disaster? In the context of disasters, the term “natural” typically refers to a naturally occurring extreme event such as a storm, cyclone, or geologic disturbance. However, whether any specific naturally occurring extreme event is considered a disaster also depends on the degree to which the event disrupts human activity. For example, a major meteorological disturbance such as a Category 5 hurricane that never makes landfall would probably not be considered a disaster. However, an event of the same magnitude that makes landfall can have devastating effects on people. For purposes of this book, for a natural disturbance to be considered a disaster, it must have some impact on people. In addition, whether an extreme event is characterized as a natural disaster also depends on how prepared communities/societies are for such events. In this regard, a storm or geologic disturbance could have a devastating impact in a developing country where preparation is inadequate and thus be considered a natural disaster, but an event of similar magnitude in an industrialized country might only result in relatively minor damages. Recognizing that any definition will be imperfect, developing a commonly accepted working definition of a “natural disaster” is necessary. Perhaps the most commonly accepted criteria for determining a natural disaster are provided by the Emergency Events Database (EM-DAT), available at www. em-dat.net. According to EM-DAT (Guha-Sapir et al., [n.d.]), an extreme event is included in this comprehensive disaster database if it results in great, widespread destruction and meets at least two of the following criteria: 1) Ten or more people reported killed 2) 100 or more people reported affected 3) A call for international assistance or a declaration of a state of emergency from the government This definition will be used in this chapter as a basis for summarizing the different types of disasters that people have experienced over time. The primary focus of this book, including this chapter, is on natural disasters, and not manmade/technological disasters such as war, anthropomorphic climate change, environmental contamination, and the like. All the different types of natural disasters are summarized in Figure 2.1. They are categorized as biological, geophysical and extraterrestrial, hydrological, and meteorological/ climatological. Biological events include epidemics, insect infestations, and animal stampedes. 13

14  Handbook on the economics of disasters NATURAL DISASTERS Biological • Epidemic ○ Viral Infectious Disease ○ Bacterial Infectious Disease ○ Parasitic Infectious Disease ○ Fungal Infectious Disease ○ Prion Infectious Disease • Insect Infestation • Animal Stampede

Geophysical/ Extraterrestrial • Earthquake • Tsunami • Volcano • Mass Movement (Dry) ○ Rockfall ○ Land Slide ○ Avalanche • Magnetic Pole Excursion/Reversal • Physical Pole Shift • Solar Storm • Solar Cycle • Asteroid/Meteor Strike

Hydrological

Meteorological

• Flood ○ General Flood ○ Flash Flood ○ Storm Surge/Coastal Flood • Mass Movement (Wet) ○ Rockfall ○ Landslide ○ Avalanche ○ Subsidence

• Storm ○ Tropical Cyclone ○ Extra-Tropical Cyclone ○ Local Storm Climatological • Extreme Temperature ○ Heat Wave ○ Cold Wave ○ Extreme Weather Conditions • Drought • Wildfire ○ Forest Fire ○ Land Fire

Figure 2.1  Natural disaster categories

Geophysical events include earthquakes, tsunamis, volcano eruptions, and dry mass movements. Note that geophysical events also include geomagnetic excursions/reversals and physical pole shifts, although these events are unlikely to occur during one’s lifetime. Infrequent extraterrestrial events include solar cycles, solar storms, and asteroid/meteor strikes. It is important to note that these disaster types sometimes interact. For example, geomagnetic excursions typically coincide with a weakening of the earth’s geomagnetic field, which in turn increases vulnerability to solar storms such as the Great Carrington event of 1859. A solar storm of similar magnitude today would likely result in significant damages to satellites, communications systems, and the electrical grid. Further, potential damages from solar storms are increasing because we are currently experiencing a magnetic excursion and an accompanying weakening of the earth’s geomagnetic field. Solar storms present a serious risk to technologically dependent societies. Solar cycles result in changes in energy received from the sun, which lead to periods of global cooling. The Maunder minimum, which led to a cooling period between 1645 and 1715 AD, had significant impacts on crop productivity in the northern and southern regions of the globe. Recent research by Zharkova (2020) demonstrates that we are again entering a solar minimum period, which began in 2020 and is expected to continue through 2053. Hydrological events include floods, storm surges, and mass movements (wet). Meteorological and climatological events include cyclones, local storms such as tornadoes, extreme temperature (extreme heat or cold), drought, and wildfire. Hydrological, meteorological, and climatological events occur frequently across the globe and are easier to forecast than other types of disaster events.

A taxonomy of natural disasters  15 The ability to forecast weather events as well as disaster characteristics such as frequency and magnitude play a critical role in how people prepare for and respond to such events. In the United States, about 45% of households are, to some degree, prepared for a disaster event (Ablah et al., 2009), and, according to Malmin (2020), one of the most important predictors of preparedness is having experienced a disaster in the recent past. In places that experience regular storms, we tend see a higher proportion of households with emergency supplies, whereas people who live in places with less frequent (but perhaps more severe) events are less prepared. Our perceptions of risk/exposure, whether accurate or not, play an important role in the degree to which we take precautionary measures. A question arises as to how to best prepare society for infrequent catastrophic events.

2.  DISASTER MAGNITUDES AND FREQUENCY This section summarizes the frequency and impacts of the different types of natural disasters with an emphasis on large-magnitude, infrequent natural disasters. Figure 2.2 presents information on the frequency of all types of natural disasters from 1900 to 2020 (taken from

Population (in Billions)

21 20

20

20

19

19

19

19

19

19

19

19

19

19

11

0

01

0

91

2

81

120

71

4

61

240

51

6

41

360

31

8

21

480

11

10

01

Number of Disasters

Natural Disaster Trends, 1901–2020 600

Year Number of Disasters

Population (in Billions)

Notes: (1) For 128 disaster events in EM-DAT, Guha-Sapir et al. (n.d.), which have inconsistent year information, the year included in “Disaster Number” is regarded as the start year, and the duration is assumed to be one year. (2) The earthquake event (1969-0155-MAR) was recorded to start in 1969 and end in 2019. Since the duration is unreasonably long, the end year is assumed to be 1969. Sources: (1) Centre for Research on the Epidemiology of Disasters (CRED). EM-DAT: The OFDA/CRED International Disaster Database. Brussels, Belgium: Catholic University of Leuven, Guha-Sapir et al. (n.d.). (2) Max Roser, Hannah Ritchie, and Esteban Ortiz-Ospina. (2013). World Population Growth. Our World in Data. Retrieved from: https://ourworldindata.org/world-population-growth.

Figure 2.2  Global natural disaster trends

16  Handbook on the economics of disasters EM-DAT, Guha-Sapir et al., [n.d.]). The dark line plots total disasters annually, whereas the green line plots global population per billion people. The number of recorded disaster events have increased dramatically over time. However, a significant portion of the increase is due to increased population. As population increases, more people are likely to be affected by extreme events. However, it is also true that a higher proportion of the global population lives in disaster-prone regions, further increasing the numbers affected. Generally, most of the increase occurred in the climate-related categories. The increase in climate-related disasters is due in no small measure to the fact that more people live in coastal areas that are prone to storms, cyclones, and hurricanes. While the number of geological events increased modestly in recent years, the number of disasters in this category has been relatively stable over time. In addition to the frequency, the magnitude of disaster events varies significantly. Some disaster types, such as tornadoes, are very intense but highly localized and of short duration. Other disasters, such as droughts or heat waves, may affect an entire region and last for weeks, months, or even years in the case of drought. Geologic events such as a landslide or even an earthquake can be limited in scope and of very short duration, whereas other deep quakes can lead to region-wide or even global impacts. The 2004 Asian earthquake/tsunami, which resulted in more than 230,000 fatalities across Asia, is one example of a broadscale disaster. How society manages disaster risk depends on the nature, frequency, predictability, and magnitude of events. Frequently occurring events such as weather-related phenomenon offer the opportunity for insurance markets to emerge as the probability of events and the expected impacts can be estimated (Skidmore, 2001). It is also easier for households, communities, and governments to prepare for these types of disasters. Less frequent events with impacts that are difficult to estimate create uncertainty, and thus insurance markets for these types of disaster tend not to emerge without governmental intervention, making it more difficult for households, communities, and governments to take precautionary measures (Skidmore, 2001). Of the disaster types, biological events have led to the most fatalities globally over the past 120 years, though they occur relatively infrequently. The EM-DAT records indicate that biological events such as pandemics resulted in 31 million fatalities since the turn of the twentieth century, but the database does not include full information about the Spanish Influenza pandemic of 1918 that resulted in more than 50 million fatalities. Taking into consideration the Spanish Influenza event, biological disasters have by far caused the most fatalities worldwide. The economic damages from disasters are substantial as well. While industrialized countries sustain the most direct economic damages, those damages are typically a smaller proportion of GDP than losses in developing countries (Toya & Skidmore, 2007). Further, industrialized countries with highly functioning capital markets and quality infrastructures recover much more quickly than developing countries (Horwich, 2000). Tables 2.1 and 2.2 summarize the number of events for each category of disaster (biological, geophysical, and climatic) over time and across geography, respectively. The period from 1901 to 1920 had the most disaster-induced fatalities. Note that the data from EM-DAT only count a small fraction of the 50+ million Spanish Influenza fatalities. Including the Spanish Influenza, the first two decades of the twentieth century were by far the deadliest of the past 120 years. Table 2.2 summarizes disaster events and fatalities globally. Asia experienced significantly more disasters and fatalities over the past 120 years, but the Asian continent also has the largest land area and population base. Africa has experienced the most biological events, whereas the Americas and Asia have had the most climatic disasters. While geophysical events occur all around the world, Asia has had by far the most geophysical events.

A taxonomy of natural disasters  17 Table 2.1  Natural disaster events and fatalities by disaster type Time Period Disaster Type

1901–1920 1921–1940 1941–1960 1961–1980 1981–2000 2001–2020

Biological Events

14

11

5

109

675

783

Biological Deaths

8,509,740

826,516

10,757

21,065

142,616

102,849

Geophysical Events

87

115

160

249

565

652

Geophysical Deaths

451,976

378,763

215,334

478,897

182,693

723,267

Climatic Events

61

126

341

1,261

3,925

6,680

Climatic Deaths

784,137

8,786,922

5,145,417

2,902,852

988,039

516,088

Notes: (1) Geophysical events include geophysical and extraterrestrial (one meteor strike) events. Climatic events include climatological, hydrological, and meteorological events. (2) If the data on fatality is missing, the fatality is assumed to be zero. (3) For 44 disaster events that span two periods, the fatalities for each year are assumed to be equal, and the fatalities for each period equal to the average fatalities times the duration of the events in that period, and the total fatalities for each disaster type are rounded to an integer. These disaster events are considered to occur in both time periods. (4) For 128 disaster events in EM-DAT with inconsistent year information, the year included in “Disaster Number” is regarded as the start year, and the duration is assumed to be one year. (5) The earthquake event (1969-0155-MAR) was recorded to start in 1969 and end in 2019. Since the duration is unreasonably long, the end year is assumed to be 1969. (6) Biological deaths recorded for the 2001–2020 period do not include COVID-19. Source:  Centre for Research on the Epidemiology of Disasters (CRED). EM-DAT: The OFDA/CRED International Disaster Database. Brussels, Belgium: Catholic University of Leuven (Guha-Sapir et al., (n.d.)).

Table 2.2  Natural disaster events and fatalities by continent Continent Disaster Type

Africa

Americas

Asia

Europe

Oceania

Biological Events

951

185

376

50

30

Biological Deaths

497,222

76,139

6,532,449

2,500,475

7,257

Geophysical Events

97

418

1,012

213

88

Geophysical Deaths

23,857

506,564

1,621,602

268,638

10,269

Climatic Events

1,841

3,263

4,975

1,690

586

Climatic Deaths

899,143

239,195

16,590,466

1,389,508

5,143

Notes: (1) Data on natural disaster events span from 1901 to 2020. (2) Geophysical events include geophysical and extraterrestrial events. Climatic events include climatological, hydrological, and meteorological events. (3) If the data on fatality is missing, the fatality is assumed to be zero. (4) The reason for the different total events between Table 2.1 and Table 2.2 is that the 44 disaster events that span two periods are considered to occur in both time periods in Table 2.1. (5) Biological deaths do not include COVID-19. Source:  Centre for Research on the Epidemiology of Disasters (CRED). EM-DAT: The OFDA/CRED International Disaster Database. Brussels, Belgium: Catholic University of Leuven.

18  Handbook on the economics of disasters

3.  LOW-PROBABILITY/HIGH-CONSEQUENCE DISASTERS Thus far, the discussion has focused on events that occur relatively frequently within the human life span. Most people are familiar with less frequent events such as earthquakes or volcanic activity. However, no one living today has experienced some types of infrequent major events that our ancestors endured. There is much evidence in the geologic, anthropologic, and archeologic records that reveals numerous catastrophic events in the past. A more detailed discussion of this evidence is given next. While it is unclear when such catastrophic events will occur again, such events will likely happen at some point in the future. From the perspective of preserving the human species as well as plant and animal species, it is prudent to consider how society should prepare for infrequent and extreme extinction-level events. 3.1  Meteorite/Asteroid Strike Geologists have long noted scars on the earth’s surface as evidence of catastrophic meteorite/ asteroid strikes. Perhaps the most recent major event occurred about 12 ka years BP (before present) when an asteroid struck Greenland, leaving a mark on the earth’s crust 16 miles across. Kjaer et al. (2018) offers an assessment of the 16-mile-wide depression beneath a kilometer of ice as evidence of the impact. Scientists believe this is among the 25 most significant known asteroid strikes. As shown in Table 2.3 (Wee, 2017), significant asteroid impacts are rare, but they can be devastating; some are considered extinction-level events. Asteroid/meteor strikes can trigger mega-tsunamis, volcanism, and can even disrupt the earth’s geomagnetic field (Wee, 2017). Table 2.3  Biggest asteroid impacts on earth Crater

Location

Acraman Crater

South Australia, Australia

Woodleigh Crater

Western Australia, Australia

Manicouagan Crater

Quebec, Canada

Morokweng Crater Kara Crater

Radius (in km) Estimated Impact Date 90

580 million years ago

40–120

360 million years ago

100

215 million years ago

North West, South Africa

70

145 million years ago

Nenetsia, Russia

65

70.3 million years ago

Chicxulub Crater

Yucatan, Mexico

170–300

Popigai Crater

Siberia, Russia

100

35.7 million years ago

Chesapeake Bay Crater

Virginia, United States

85

35 million years ago

65 million years ago

Source:  Wee (2017).

3.2  Geomagnetic Excursions and Solar Flares Geomagnetic excursions and solar flares are discussed together because vulnerability to solar flares depends on the strength of the geomagnetic field that protects the earth from harmful solar radiation/storms. Currently, evidence suggests that the geomagnetic poles are in the process of an excursion, which in turn is associated with the weakening of the earth’s

A taxonomy of natural disasters  19 geomagnetic field (Livermore et al., 2020). In fact, the geomagnetic field has been weakening since about 1700, and the rate of weakening has increased in recent decades; the overall field strength has decreased by about 8% since 1970 (Kumar, 2020). A distinction must be made between a full geomagnetic pole reversal, which occurs very infrequently on earth, and geomagnetic excursions, which have occurred about every 12 ka–14 ka years for the past 45 ka years. The discussion begins with a brief history of magnetic excursions that have occurred in the relatively recent past as well as the current excursion we are experiencing now. We will then turn our attention to solar flares and their potential impacts. The most recent excursion aside from the one currently underway is the Gothenburg Excursion, which occurred about 12–13 ka BP (Mörner, 1977). While the excursion occurred over about 300 years, there was significant movement over a 50-year period, which resulted in the geomagnetic north pole moving from the arctic region to the central pacific, and ultimately returning to the arctic. Prior to the Gothenburg Excursion, there was the Mono Lake Excursion, which occurred between 24 and 32 ka BP (Barbetti & McElhinny, 1976; Wang et al., 2019).1 During this event, the magnetic north pole wandered to the Eurasian continental area before returning to the arctic region. There is also the Laschamp Excursion that occurred about 40–42 ka BP (Jiabo, 2019). A very recent study by Cooper et al. (2021) demonstrated that the Laschamp event was a full magnetic reversal accompanied by a dramatic reduction in the geomagnetic field, which in turn caused significant changes in atmospheric ozone as well as climate and environmental shifts that led to the extinction of species (Channell & Vigliotti, 2019). Numerous studies now present geologic evidence from different parts of the world showing that periods of significant geomagnetic instability occurred in 12–14 ka BP, 24–32 ka BP, and ~40–42 ka BP. Over the past 42 ka years, three geomagnetic excursions (or periods of geomagnetic instability) have occurred about every 12–14 ka years. In 2019, the World Magnetic Model had to be recalibrated because the geomagnetic poles were moving faster than expected. A time-varying map from the National Oceanic and Atmospheric Administration (NOAA) offers a clear illustration of movement of the geomagnetic poles of the past several hundred years (https://www.ncei.noaa.gov/maps/historical_declination/). Over the past 30 years, the geomagnetic pole has moved rapidly (about 55 km per year) from northern Canada across the Arctic Sea toward Russia. During the same period, the geomagnetic field has weakened by about 5% per decade. While geomagnetic excursions affect animal migration patterns and force the adjustment of systems that rely on the earth’s geomagnetic field (in today’s modern world), a moving geomagnetic pole in and of itself does not always present a serious threat to life. However, corresponding to this movement is a change in the strength of the geomagnetic field, which can lead to greater exposure to harmful solar radiation. In most but not all geomagnetic excursion events, the strength of the geomagnetic field declines. A weakening of the field makes the earth and all living things more vulnerable to solar flares and radiation from space. We now turn our attention to this risk factor. Geomagnetic storms are created when a solar coronal mass ejection (CME) hits the earth’s magnetosphere. The rate at which the sun produces the CME varies and depends on the solar cycle. During solar maxima, the sun creates about three CMEs every day, whereas during solar minima only about one CME is created every five days. Whether a CME has any effect on the earth depends on whether the earth happens to be in the trajectory of the CME. Though the sun produces many CMEs, most do not intersect with the earth. However, when a CME hits the earth,

20  Handbook on the economics of disasters it creates a geomagnetic perturbation. The largest recorded geomagnetic storm was the Carrington Event of 1859, which damaged parts of the new US telegraph network. In 2012, A CME of similar magnitude just missed the earth in 2012 (https://science.nasa.gov/science-news/science-atnasa/2014/23jul_superstorm/). There is also evidence of CMEs hitting the earth farther back in history. For example, scientists believe a significant CME hit the earth in 774 AD (Plait, 2021). Scientists are concerned about another such event because it could potentially cause trillions of dollars in damages to communications systems and power grids, not to mention the potential health impacts on people due to greater radiation exposure (Stone, 2015). Communication and power outages could create major disruptions to human activity and could be deadly. A question arises as to how technologically dependent societies can reduce exposure to such CMEs. Solar minimums have also been shown to trigger seismic/volcanic activity (Ebisuzaki et al., 2011). 3.3  Solar Cycles

Solar activity (summary) curve

The preceding discussion noted that the frequency of CMEs depends on the solar cycle. However, the solar cycle also plays a significant role in the heating and cooling of the earth and, by extension, influences the climate. We now turn our attention to how the solar cycle affects the earth’s climate. Astrophysicist Valentina Zharkova is an expert in solar cycles and has written a number of scientific articles about the role of solar cycles in influencing terrestrial temperature. A recent study by Zharkova (2020) summarizes research on how the sun has entered the modern grand solar minimum (2020–2053), which she expects will lead to a reduction in terrestrial temperature. The sun experiences a 11-year cyclic variation in the number of sunspots, which influences the earth’s temperature. Overarching these shorter cycles are 350- to 400-year cycles as illustrated in Zharkova et al. (2015). The image demonstrates significant variation in solar activity modulating in 350 to 400 cycles. As shown in Figures 2.3 and 2.4, these cycles align closely with terrestrial temperature variation over time. For example, the Maunder minimum correlates with the cool period from 1645 to 1710, wherein the average temperature in northern Europe decreased by 1–1.5˚C. This change led to frozen rivers such as the Thames in England, long cold winters, and relatively cold summers. During this period, sea ice expanded south from the arctic, alpine glaciers stretched into valley farmland. Even the canals in the Netherlands froze. 800 Wolf minimum 600 400 200 0 –200 –400 –600 –800 1200 1400

Maunder minimum

1600

1800

Modern minimum 1

2000

2200

Modern minimum 2

2400

2600

2800

3000

3200

Calendar year

Source:  Image from Zharkova (2015). https://www.nature.com/articles/srep15689#rightslink. Used under Creative Commons License 4.0 (CC. BY 4.0).

Figure 2.3  Solar cycles

1366.8 1366.6 1366.4 1366.2 1366.0 1365.8 1365.6 1365.4 1365.2 1365.0 1364.8 1364.6 1364.4 1364.2 1364.0 1363.8 1363.6 1363.4 1363.2 1363.0

Solar irradiance reconstructed

IG8P PAGES/World Data Center for Paleoclimatology

1366.8 1366.6 1366.4 1366.2 1366.0 1365.8 1365.6 1365.4 1365.2 1365.0 1364.8 1364.6 1364.4 1364.2 1364.0 1363.8 1363.6 1363.4 1363.2 1363.0

1610 1620 1630 1640 1650 1660 1670 1680 1690 1700 1710 1720 1730 1740 1750 1760 1770 1780 1790 1800 1810 1820 1830 1840 1850 1860 1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010

Solar irradiance (W/m2)

1610 1620 1630 1640 1650 1660 1670 1680 1690 1700 1710 1720 1730 1740 1750 1760 1770 1780 1790 1800 1810 1820 1830 1840 1850 1860 1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010

A taxonomy of natural disasters  21

11

Temperature °C

10

Maunder

1880– 1915

Dalton

1945– 1977

9

8 1660 1680 1700 1720 1740 1760 1780 1800 1820 1840 1860 1880 1900 1920 1940 1960 1980 2000 2020 Year

Source:  Image from Zharkova (2020). https://www.tandfonline.com/doi/full/10.1080/23328940.2020.1796243. Reprinted with permission from Elsevier.

Figure 2.4  Solar irradiance reconstructed and terrestrial temperature Importantly, Zharkova’s models show that we have now entered another solar minimum period (2020–2053) in which she expects global temperatures to drop by at least 1˚C. Zharkova (2020, p. 6) offers a warning: The reduction of a terrestrial temperature during the next 30 years can have important implications for different parts of the planet on growing vegetation, agriculture, food supplies, and heating needs in both Northern and Southern hemispheres. This global cooling during the upcoming grand solar minimum 1 (2020–2053) can offset for three decades any signs of global warming and would require intergovernment efforts to tackle problems with heat and food supplies for the whole population of the earth.

22  Handbook on the economics of disasters Though the sun cycles occur over several hundreds of years, Zharkova and others have demonstrated that they can be forecasted. Should we heed their warnings and consider and prepare for a scenario where the terrestrial temperature drops rather than rises over the next 30 years?2 3.4  Geographic Pole Shift In the early part of the twentieth century, Alfred Wagner proposed the theory of continental drift. Wagner and other scientists proposed that at one time, the continents were a large single land mass that broke apart and drifted over very long periods of time. This theory is supported by how the shapes of the continents appear to fit together. In what is now known as plate tectonics, the outer layer of the earth’s crust (the lithosphere) contains a number of plates that move independently over a more viscous layer called the asthenosphere (Kious & Tilling, 1996). Today, the generally accepted view is that the movement of these plates occurs incrementally over long periods of time, and there is substantial evidence of this as a steady geologic process. However, we also have evidence of significant catastrophic geologic events that have occurred several times over the past 100 ka years. Perhaps the most widely known writing of the crustal displacement theory is that of Charles Hapgood (1958). Hapgood compiled research from many fields of study, including but not limited to anthropology, biology, climatology, geology, glaciology, and paleontology, which offered compelling evidence that until about 12 ka years ago, the north pole had been located in Hudson Bay, Canada. A north pole at that location offers an explanation for why a great ice sheet covered North America, extending all the way to Ohio. It also explains why during the same time frame, Alaska and Siberia had relatively mild climates. The evidence Hapgood presents comes from the scientific literature and is very compelling. Regardless, many have not accepted Hapgood’s pole shift hypothesis, but new evidence has emerged. Very recently, Carlotto (2020a, 2020b) has provided new evidence of a pole shift as well as evidence that a relatively advanced human civilization was present when the pole shifted to its current position. Further, he provides evidence that human civilization was in place in the previous three pole locations prior to Hudson Bay (Greenland, Norwegian Sea, and the Bering Sea) over the past 100,000 years. The two Carlotto articles are companion papers. The first article (Carlotto, 2020a) uses Google Earth to explore the alignments of 224 ancient sites. The ancient sites studied are often aligned with the north pole (19%), solstices (9%), lunar standstill (15%), magnetic north pole (5%), zenith passages (4%), and stars (1%). However, about 42% of the sites he explored were not aligned with any of the typical alignment hypotheses. Carlotto (2020b) proceeds to test Hapgood’s hypotheses of previous pole locations by exploring whether any of the “unaligned” ancient sites are aligned with the cardinal directions of the previous poles. In this article, Carlotto reports that all but 17 of the 95 unexplained sites are aligned with the previous pole locations. Several of these ancient sites and their alignments are provided in Figures 2.5, 2.6, 2.7, and 2.8 (reprinted with permission). Modern cities are sometimes built over earlier settlements, often preserving the alignments of the original sites (Aveni, 1980). As certain sites fell into ruin over time, they were rebuilt/ expanded. Thus, the sites that remain today may not be the originals, but rather relatively new structures that preserved the original alignment. There is now substantial scientific evidence that a major crustal displacement occurred about 12 ka years ago, which resulted in the geographic north pole moving from Hudson Bay to its current location. Graveyards of flash frozen mammoths and other creatures found in Siberia and

A taxonomy of natural disasters  23

a) Teotihuacan, Mexico

b) Tikal, Guatemala

c) Sri Martand Sun Temple, India

d) Haʻamonga ʻa Mau

Examples of sites aligned to the Hudson Bay pole. Photo credit: Apple Maps Source:  Image from Carlotto (2020a). https://doaj.org/article/3218dbb0fe93464ab6f4e3361a6d60df. Reprinted with permission from the Journal of Scientific Exploration.

Figure 2.5  Alignment with the Hudson Bay pole

Alaska suggest that the crustal shift that occurred was catastrophic. Hibben (1961) estimated that as many as 40 million animals died in North America at the close of the Ice Age. Hapgood hypothesized that the massive extinction was the result of crustal displacement, which resulted in a new pole location. The dates of the previous pole shifts are more difficult to decipher, but ice cores from Antarctica offer some useful information. Specifically, ice core samples from East Antarctica are more than a million years old. However, samples in West Antarctica are much younger, between a few centuries to 70 ka years (US ITASE, accessed March 2021). Limitations prevent the full presentation of this documentation in this chapter. While there is much more evidence of these previous pole shifts, it is difficult to pin down the dates of pole shifts that occurred prior to Hudson Bay. One thing we can conclude is that a geographic pole shift is not a one-time event; it is not even rare as there is overwhelming scientific evidence

24  Handbook on the economics of disasters

a) Tower of Babel, Babylon

b) Temple of Jupiter, Baalbek, Lebanon

c) The Parthenon, Athens

d) Tenochtitlan, Mexico City

Examples of sites aligned to the Greenland pole. Photo credit: Apple Maps Source:  Image from Carlotto (2020a). https://doaj.org/article/3218dbb0fe93464ab6f4e3361a6d60df. Reprinted with permission from the Journal of Scientific Exploration.

Figure 2.6  Alignment with the Greenland pole that such shifts have occurred many times, including the last four with human civilization in place to experience them. How did our ancestors cope with the movement of the entire crust of the earth? When will the next pole shift occur? Should we prepare for it, and, if so, how? As a final thought on the geographic pole shift, there is no consensus on what force would cause the crust of the earth to separate from the mantle, shift, and do so multiple times.

4. CONCLUSIONS During the average human life span, it would be common to experience a significant climatic event. One might even experience a significant earthquake or volcanic eruption, depending on where one lives. For such events, it is possible to prepare for or potentially purchase

A taxonomy of natural disasters  25

a) Chichen Itza, Mexico

b) El Tepozteco, Mexico

c) Caral-Supe, Peru

d) Brihadisvara Temple, India

Examples of sites aligned to the Norwegian Sea pole. Photo credit: Apple Maps Source:  Image from Carlotto (2020a). https://doaj.org/article/3218dbb0fe93464ab6f4e3361a6d60df. Reprinted with permission from the Journal of Scientific Exploration.

Figure 2.7  Alignment with the Norwegian Sea pole

insurance against. National and subnational government safety nets also provide support to people during crises. Is it possible to prepare for catastrophic events such as an asteroid strike, a physical or magnetic pole shift, or major geomagnetic storm? For potentially extinctionlevel events such as these, most people do not have the capacity to prepare. The magnitudes of such major catastrophic occurrences are so large that only society as a whole has the capacity to collectively engage in preparations. And yet, such governmental preparations are not discussed very often in the public sphere, and budgets for such activities are often nontransparent. However, we are aware of some deep underground facilities (Sepp, 2000), seed vaults (Simpson, 2015), a growing interest in off-world activity such as the newly created Space Force (US Space Force, 2021), and a very recent government contract for moon-based 4G networks (Browne, 2020).

26  Handbook on the economics of disasters

a) Nazca line, Peru

b) Temple of the Sun, Ollantaytambo, Peru

c) Knossos, Crete

d) Temple of the Winged Lions, Petra, Jordan

Examples of sites aligned to the Bering Sea pole. Photo credit: Apple Maps Source:  Image from Carlotto (2020a). https://doaj.org/article/3218dbb0fe93464ab6f4e3361a6d60df. Reprinted with permission from the Journal of Scientific Exploration.

Figure 2.8  Alignment with the Bering Sea pole

While everyday people can do some things to increase resilience and reduce vulnerability, large-scale catastrophic events require collective preparations. However, such preparations are largely not a part of public discourse except in movies and entertainment. The chapters in this book present a variety of research papers and approaches for examining the implications of disasters on different aspects of economic activity and decision-making. The purpose of this chapter is to provide an overview of all types of extreme events. The most difficult to evaluate are very extreme catastrophic events that rarely occur. Yet, to ensure the survival of the civilization, it is prudent to consider how societies prepare for extraordinary events that we know have occurred in the past and may occur again at some point in the future.

A taxonomy of natural disasters  27

NOTES 1. Barbetti and McElhinny (1976) estimate that the event occurred around 24 ka years BP. The more recent Wang et al. (2019) assessment is that the event occurred about 28 ka–32 ka years BP. 2. Many climate scientists have presented evidence of anthropomorphic-induced global warming/climate change. This book is focused primarily on natural disasters, not human-made or human-induced disasters.

REFERENCES Ablah, E., Konda, K., & Kelley, C. L. (2009). Factors predicting individual emergency preparedness: A multi-state analysis of 2006 BRFSS data. Biosecurity and bioterrorism: Biodefense strategy, practice, and science, 7(3), 317–330. https://doi.org/10.1089/bsp.2009.0022. Aveni, A. F. (1980). Skywatchers of Ancient Mexico. University of Texas Press. Barbetti, M. F., & McElhinny, M. W. (1976). The lake Mungo geomagnetic excursion. Philosophical Transactions of the Royal Society of London. Series A, Mathematical and Physical Sciences, 281(1305), 515–542. Browne, R. (2020). NASA is launching a 4G mobile network on the moon. CNBC. https://www.cnbc.com/2020/10/19/ nasa-is-launching-a-4g-mobile-network-on-the-moon.html. Accessed on May 2021. Carlotto, M. J. (2020a). An analysis of the alignment of archaeological sites. Journal of Scientific Exploration, 34(1) 13-35. Carlotto, M. J. (2020b). A new model to explain the alignment of certain ancient sites. Journal of Scientific Exploration, 34(2), 209-232. Channell, J. E. T., & Vigliotti, L. (2019). The role of geomagnetic field intensity in late quaternary evolution of humans and large mammals. Reviews of Geophysics, 57(3), 709–738. Cooper, A., Turney, C. S. M., Palmer, J., Hogg, A., McGlone, M., Wilmshurst, J., Lorrey, A. M., Heaton, T. J., Russell, J. M., McCracken, K., Anet, J. G., Rozanov, E., Friedel, M., Suter, I., Peter, T., Muscheler, R., Adolphi, F., Dosseto, A., Faith, J. T., . . . Zech, R. (2021). A global environmental crisis 42,000 years ago. Science, 371(6531), 811–818. https://doi.org/10.1126/science.abb8677. Ebisuzaki, T., Miyahara, H., Kataoka, R., Sato, T., & Ishimine, Y. (2011). Explosive volcanic eruptions triggered by cosmic rays: Volcano as a bubble chamber. Gondwana Research, 19(4), 1054–1061. Guha-Sapir, D., Below, R., & Hoyois, P. (n.d.). EM-DAT | The International Disasters Database. Université Catholique de Louvain – Brussels – Belgium. https://www.emdat.be/. Accessed on February 5, 2021. Hapgood, C. (1958). Path to the Pole. Adventures Unlimited Press. Hibben, F. C. (1961). The Lost Americans. Thomas Y. Crowell. Historical Declination Viewer. (2020, October 2). National Oceanic and Atmospheric Administration (NOAA). https://maps.ngdc.noaa.gov/viewers/historical_declination/. Horwich, G. (2000). Economics lessons of the Kobe earthquake. Economic Development and Cultural Change, 48(3), 521–542. Jiabo, L. (2019). Dynamics of the Geomagnetic Field During the Last Glacial. Doctoral Dissertation. Universität Potsdam. https://d-nb.info/1218404833/34. Accessed on April 13, 2021. Kious, W. J., & Tilling, R. I. (1996). This Dynamic Earth: The Story of Plate Tectonics. DIANE Publishing. Kjaer, K. H., Larsen, N. K., Binder, T., Bjørk, A. A., Eisen, O., Fahnestock, M. A., . . . & MacGregor, J. A. (2018). A large impact crater beneath Hiawatha Glacier in northwest Greenland. Science Advances, 4(11), eaar8173. Kumar, U. (2020, June 4). Earth’s Magnetic Field Is Weakening Down. Should We Be Worried? Live in the One. https://liveintheone.com/2020/06/04/earths-magnetic-field-is-weakening-down-should-we-be-worried/. Livermore, P. W., Finlay, C. C., & Bayliff, M. (2020). Recent north magnetic pole acceleration towards Siberia caused by flux lobe elongation. Nature Geoscience, 13(5), 387–391. Malmin, N. P. (2020). Historical disaster exposure and household preparedness across the United States. Disaster Medicine and Public Health Preparedness, 15(1), 58–64. Mörner, N. A. (1977). The Gothenburg magnetic excursion. Quaternary Research, 7(3), 413–427. Plait, P. (2021, January 6). In 774 AD, the Sun blasted Earth with the biggest storm in 10,000 years. SYFY WIRE. https://www.syfy.com/syfywire/in-774-ad-the-sun-blasted-earth-with-the-biggest-storm-in-10000-years. Sepp, E. M. (2000). Deeply Buried Facilities: Implications for Military Operations. Occasional Paper No. 14, Center for Strategic and Technology Air War College. Simpson, C. (2014). Food Security and the Doomsday Vault: Foresight or Fallacy. Interdisciplinary Environment Review, 10(3–4), 73–79. Skidmore, M. (2001). Risk, natural disasters, and household savings in a life cycle model. Japan and the World Economy, 13(1), 15–34.

28  Handbook on the economics of disasters Stone, M. (2015, August 20). What Would Happen if a Massive Solar Storm Hit the Earth? Gizmodo. https://gizmodo. com/what-would-happen-if-a-massive-solar-storm-hit-the-eart-1724650105. Toya, H., & Skidmore, M. (2007). Economic development and the impacts of natural disasters. Economics Letters, 94(1), 20–25. US International Trans Antarctic Scientific Expedition (ITASE). (n.d.). ITASE: HOME. http://www2.umaine.edu/ itase/. Accessed March 2021. US Space Force. https://www.spaceforce.mil/. Accessed May 2021. Wang, S., Chang, L., Xue, P., Liu, S., Shi, X., Khokiattiwong, S., . . . & Liu, J. (2019). Paleomagnetic secular variations during the past 40,000 years from the Bay of Bengal. Geochemistry, Geophysics, Geosystems, 20(6), 2559–2571. Wee, R. Y. (2017). Biggest asteroid impacts in earth’s history. World Atlas. https://www.worldatlas.com/articles/ biggest-asteroid-impacts-in-earth-s-history.html. Accessed on April 13, 2021. Zharkova, V. (2020). Modern grand solar minimum will lead to terrestrial cooling. Temperature, 7(3), 217–222. Zharkova, V. V., Shepherd, S. J., Popova, E., & Zharkov, S. I. (2015). Heartbeat of the sun from principal component analysis and prediction of solar activity on a millennium timescale. Scientific Reports, 5(1), 1–11.

PART I THEORETICAL CONSIDERATIONS IN EVALUATING DISASTER IMPACTS

3. A few good models for economic analysis of disasters: can your model handle the truth? Yasuhide Okuyama*

1. INTRODUCTION While natural and man-made hazards have increased in frequency and intensity around the world, understanding the economic cost of such events and the benefits of the countermeasures has become vital to not only the public sector but also the citizens who may potentially face hazards (Valles et al., 2021). Although researchers have studied various disasters using a wide range of economic modeling frameworks, these studies are only to reveal how complicated it is to construct models to handle the truth of disaster impacts in an economic context. Why is it so complicated to develop such models? One of the major reasons is that disasters, as the consequence of natural or man-made hazards, are unique events, with each disaster possessing a different set of attributes, such as the geographical extent of the affected areas, duration of hazard occurrence, and heterogeneity of damage over space and/or time. This has motivated disaster researchers to develop models emphasizing the most significant attributes, resulting in different types of modeling frameworks and, hence, different results from one study to another, even for the same event (Hallegatte & Przyluski, 2010). Despite the suggested standardization of the modeling for disaster impact analysis (e.g., Greenberg et al., 2007; Okuyama, 2007), several advancements have been made to such models in various directions through incorporating disaster-related features such as resilience, finer geographical resolution, shorter or longer time frames, or disaggregated economic agents. Each advancement of the modeling framework and structure has its advantages and limitations because a model is a partial representation of a real-world event. This chapter presents some of the recent advancements of modeling frameworks for disaster impact analysis, along with the discussions of their advantages and limitations; provides insights from empirical studies on disasters at the macro and micro level that can potentially contribute to further improvements of modeling; and deliberates the future opportunities in this line of research.

2.  ECONOMIC IMPACTS OF DISASTERS The definition and extent of economic impacts of a disaster are vague at best, and terms such as damage, loss, direct damage, and indirect loss are used interchangeably in various contexts (Okuyama, 2007). The guidelines in the Handbook for Disaster Assessment by the United Nations Economic Commission for Latin America and the Caribbean (UN ECLAC, 2014), which have been widely employed for recent disaster assessments in developing countries, define the economic influence of a disaster as twofold: effects of a disaster, including damages (to stock), losses (decline in flow production by business interruptions), and additional costs (costs required to adjust production due to the disaster); and impact of a disaster, including 30

A few good models for economic analysis of disasters  31 changes in household income, unemployment rate, gross domestic product (GDP), and fiscal balance.1 This ECLAC definition of losses can be further divided into first-order losses, that is, the decline in production level caused by damages, and higher-order effects,2 the ripple effects stemming from first-order losses through production linkages. While first-order losses can be measured based on the damage to production facilities employing capital-to-output ratios, for example, it is problematic to empirically measure higher-order effects as well as additional costs. This is because they spread within the disaster-hit economy and across other economies via interindustry linkages for higher-order effects and through various business decisions for additional costs (see Rose, 2004). As a matter of fact, higher-order effects and some proportion of additional costs result from a series of propagation stages, as follows (Oosterhaven, 2017): (1) decline of production due to destruction of production capacities (first-order losses), (2) negative forward effects from first-order losses (for nonreplaceable goods and services), (3) positive substitution effects for replaceable goods and services to nondamaged industry and potential additional costs (increased price and cost), (4) decline in intermediate and final demands in the disaster-hit region, (5) positive or negative changes in consumption patterns due to psychological impact (especially in the case of terrorist attacks and pandemics), and (6) positive backward effects from recovery and reconstruction activities. Therefore, some economic models are necessary to assess such impacts, including input-output (IO) models, computable general equilibrium (CGE) models, and econometric models.3 In brief, traditional IO models, also known as the Leontief models, are demand-driven models by which changes in final demand are transformed to changes in outputs through backward interindustry linkages. Because the IO model structure is rigid with fixed coefficients for production technology, value added, and imports, a simple IO model cannot deal with supply-side constraints and their resulting effects or substitution effects (1, 2, and 3 above) in a disaster. Supply-driven IO models, on the other hand, can accommodate supply-side shocks (but not substitution effects). However, while the supply-side IO model has inherent deficiencies and derives unrealistic outcomes for an economic expansion case,4 Rose and Wei (2013, p. 213) suggest that it is “less subject to criticism for a supply disruption” in a disaster situation “than for a supply acceleration” during an economic expansion. With all these limitations and deficiencies, IO models have been the most popular modeling framework for disaster impact analysis because of their simple structure, thrifty data requirement, and painless operability for calculation. CGE models have also been used for disaster impact analysis. Unlike IO models, CGE models are optimization models simulating general equilibrium across various markets. They can accommodate both demand and supply shocks, responses to price change via price elasticities, and technical and spatial substitution possibilities. Moreover, CGE models can endogenize resilience in the model as part of the economic system in a disaster both inherently by, for example, input substitution possibilities already built into the production process or, adaptively, for example, by improvising technological change after the disaster strikes (Rose & Liao, 2005). Because of such flexibilities to changes in various factors, the construction of a CGE model for a particular economy requires an enormous volume of data sets. In addition, some parameter values, especially those for elasticities of technical and/or spatial substitution, are often borrowed from other CGE models of different geographical levels, which “are likely not to be comparable” to the region in question (Greenberg et al., 2007, p. 90). Moreover, the operation of a CGE model demands expert knowledge of calibration and performing analysis. Despite these disadvantages, however, CGE models are considered ideal

32  Handbook on the economics of disasters modeling frameworks for disaster impact analysis that can cover all six propagation stages (Oosterhaven, 2017). These models have not been used without criticism. Notably, Albala-Bertrand (2013) contended that many quantitative analyses of disaster economic impacts using these models intrinsically involve the following three “interactive insufficiencies”: (a) low quality of input data, (b) limitations of the models employed, and (c) results interpretation issues. To some extent, these criticisms are common to all quantitative models for economic projections/ forecasts, which are based on a set of assumptions to reflect specific aspects of the reality. On the other hand, disasters pose unique features to economic models, such as sudden and intense changes in both demand and supply in either different or similar directions, resilience in production processes, and heterogeneous damage across industries and/or over space. Hence, several modifications to and extensions of the modeling frameworks have been proposed to not only cope with these disaster-related features but also overcome the limitations and disadvantages so that the models and their estimation results can alleviate the above interactive insufficiencies.

3.  RECENT ADVANCES IN DISASTER IMPACT MODELING This section examines the recent advances in disaster impact modeling by model type, that is, the IO model, CGE model, and agent-based model, in order to tackle their deficiencies and drawbacks, as well as incorporate disaster-related features. 3.1  Input Data Issues It is often said that the accuracy (or plausibility) of the estimation results using a quantitative model relies mostly on the sufficiency and quality of the input data. This is what AlbalaBertrand (2013) identified as one of the interactive insufficiencies by which many quantitative analyses of disasters employ imprecise input data, such as the data of damage and loss for a particular disaster. While the ECLAC methodology has established a common framework to measure and refine the data for damage and loss, it remains largely aggregated and sketchy even for applying to usual macroeconomic models. To cope with data limitations and make the input data more suitable to the model’s resolution and consistent across events, a series of methods for rapid assessment of damage and/or loss caused by a disaster has been proposed. The assessment of the damage and/or loss right after the event is difficult and problematic, not only because the degree of destruction on stocks (properties, such as buildings, facilities, and equipment) cannot be gauged immediately but also because their values, especially of damage, should be converted to monetary values to be used as input data in economic models. If a method exists that can assess the intensity of a natural hazard—such as the level of ground motion at a particular location in the case of earthquakes, wind speed in the case of severe storms, and hourly precipitation for flooding, for example—and translate it to the monetary value of damage and/or loss, it would make this process manageable and harmonized. Heatwole and Rose (2013) introduced a reduced-form regression model to estimate property damage in monetary value from the observed earthquake magnitude, the distance from the earthquake’s hypocenter, location, and so forth, for future earthquakes in the United States. Toyoda et al. (2020) proposed a similar approach for a real-time estimation of damage caused

A few good models for economic analysis of disasters  33 by an earthquake. They employ a simpler regression model than that of Heatwole and Rose’s, relying only on the categorized ground motion levels, which can estimate damage in a 250meter grid for the affected area. While these methods estimate damage from an earthquake, the production capacity loss rate, introduced by Kajitani and Tatano (2014), can estimate the rate of production capacity reductions due to the projected ground motion of an earthquake using the functional fragility curve and lifeline resilience factors. This approach results in production losses, by which the conversion process from damage data to loss data needed for IO analysis can be avoided. Whereas all these methods assume the availability of monitoring equipment for hazard intensity across various locations, which may be affordable in developed countries but not so in many developing countries, the satellite image of the nighttime light is proposed as an alternative for measuring the economic impacts of a disaster. Klomp (2016) studied the satellite images of the damaged areas after a disaster between 1990 and 2010, concluding that the amount of light visible from outer space is significantly reduced after a natural hazard strikes in the short term. However, changes of nighttime lights are surely caused either by the damage to facilities (buildings, houses, power lines, or equipment) or by business interruptions resulting from higher-order effects. In this sense, the impacts measured through the nighttime satellite images should be considered as the aggregated impacts of losses and higher-order effects, not as losses alone. Nevertheless, the economic impacts estimated using the nighttime satellite imagery can serve as a preliminary estimation in many developing countries where data collection on damage and loss requires a much longer time than in developed countries. 3.2  Input-Output Models Despite the inherent deficiencies, such as the inability to deal with supply shocks and/or substitution and having rigid coefficients, as discussed in the previous section, IO models and the variants have been a popular modeling framework for the economic analysis of disasters. Because of its simple structure, the IO model can be easily modified and/or integrated with other models, such as network models for transportation or lifelines. In order to alleviate some of the deficiencies, a series of extended IO models have been proposed. One of the most prevalent modifications is the adaptive regional IO (ARIO) model introduced by Hallegatte (2008), which adds a series of case-specific ad hoc assumptions, such as production capacity constraints, rationing scheme, and adaptation of demand and supply to the traditional Leontief IO model. These exogenously determined ad hoc assumptions of the ARIO model may be useful and acceptable with regard to the uniqueness of each disaster event (Albala-Bertrand, 2013). The ARIO model has been further extended to not only incorporate firm network within an economy to evaluate cost amplifications from the heterogeneity of losses and business interactions in a disaggregated manner (Henriet et al., 2012) but also to include production inventories to better represent production bottlenecks (Hallegatte, 2014). Variants of the ARIO model have been applied in several studies, namely, by Li et al. (2013) to a hypothetical flooding in London, and Guan et al. (2020) to the global supply-chain effects of COVID-19. In addition, the ARIO model with firm network has been extended to an agent-based model (ABM) for detailed analyses of supply-chain disruptions; a couple of them will be discussed later in the chapter. Another way to deal with supply constraints and spatial substitution effects within the IO framework is to combine it with other modeling frameworks. Koks and Thissen (2016)

34  Handbook on the economics of disasters proposed a recursive dynamic multiregional supply-use model,5 called the multiregional impact assessment (MRIA) model, which combines the IO framework with linear programming. The objective function is to minimize the value of total production over all regions given not only the conditions where supply should be equal to or larger than demand but also the constraints of maximum production capacity and of the consistency of regional imports and exports relationships. In this way, the MRIA model can use available production technologies based on the supply-use model, deal with both demand and supply side effects, and embrace multiregional substitution effects. Faturay et al. (2020) also employed a liner programming and IO model combination while establishing the model by maximizing the economy-wide total output subject to damaged production capacities and nonnegative final demand (net outputs). Because their model is based on a single-region IO table and they also assume that the pre-disaster economic structure is unchanged, spatial substitution effects are not considered. Using the two flooding scenarios in Italy, the estimation results from the MRIA model are systematically compared with those of the ARIO model and a multiregional CGE model6 for the same geographical area and the similar input data (Koks et al., 2015). The comparison among three models indicates relatively large differences on the national scale, most significantly between the ARIO model and the other two models. Because the MRIA and CGE models are flexible and optimized in terms of spatial substitutions, while the ARIO model can ration demand-supply gaps but with ad hoc fixed ratios, the ARIO model derived the results approximately three to six times higher than did the CGE model. Differences between the MRIA and CGE results appear relatively small, indicating that the spatial substitution effects for demand-supply gaps in the damaged regions contribute considerably to reducing the propagation of negative impacts across regions. Using a hypothetical open, two-region, and two-industry IO model, Oosterhaven and Bouwmeester (2016) demonstrated a nonlinear programming approach to relieve some of the IO model’s deficiencies using the principal of minimum information gain (Theil, 1967). Their model minimizes the information gain between the pre-disaster and post-disaster pattern of interregional and interindustry transactions through incorporations of fixed technical coefficients for short-run impact analysis, flexible trade coefficients for spatial substitution effects, and partial import and export substitutions to compensate for the lack of intermediate demands. With these features, they claim that their model “combines the simplicity of the IRIO (interregional IO) model with the plausibility of CGE approach” (p. 585). The proposed model was tested using a series of scenarios and obtained conceivable results, including the mixture of partially compensating demand, supply, and spatial substitution effects. This model was extended with a multiregional supply-use table for relaxing the assumption of fixed industry market share in the IO framework and applied to an empirical case of the 2013 flooding in Germany (Oosterhaven & Többen, 2017). Because of the flexibilities attained in these original and extended models, the results show “considerably lower disaster multipliers” (p. 407) than do those applying the traditional demand-driven IO models. A comparative analysis of several multiregional models based on the IO framework—including this nonlinear programming model, the aforementioned MRIA and two demand-driven IO models, the multiregional supply-use model, and the dynamic multiregional inoperability IO model—was conducted using the 2013 German flooding event (Koks et al., 2019). Because the nonlinear programming model and the MRIA model have the capability to incorporate supply shocks and the flexibility of spatial substitution, the estimation results were substantially different between two groups. The demand-driven IO models derived only negative impacts in all the German

A few good models for economic analysis of disasters  35 regions, whereas the programming models estimated not only smaller negative impacts but also some positive impacts at several nondamaged regions with the MRIA model because of its spatial substitution effects and constraint. Garcia-Hernandez and Brouwer (2021) also developed a model combining nonlinear programming and a multiregional IO model to assess the impacts of water supply disruptions under climate change in Canada, focusing more on supply shocks and less on substitution effects. Minimizing the weighted Euclidean distance between the target and baseline outputs, their model had a flexible optimization procedure that could identify the minimal gross output disruption or the least disruption to final demand. Their study highlights how the regional economy endogenously absorbs the water supply disruptions, thus not allowing substitution of local intermediate inputs with imports, while it is also capable of incorporating substitution effects across regions. Other approaches to incorporate supply shocks in the traditional IO framework include the mixed exogenous/endogenous IO model, which was applied to the 2011 Great East Japan Earthquake and Tsunami case (Arto et al., 2015). Unlike natural hazards that generate supply shocks but not demand declines simultaneously as the first-order impact, the recent COVID-19 pandemic presented economies with simultaneous supply and demand shocks. Supply-side shocks arose due to nationwide lockdowns for nonessential industries whose employees could not work remotely. Demand-side shocks became notable for travel and personal services industries in order to avoid exposure to the virus. Guan et al. (2020) employed the ARIO framework based on the World Input-Output Database (WIOD) to assess the global impacts of nationwide lockdowns by several countries. Similarly, but using a linear programming model with a set of rationing schemes, Pichler and Farmer (2021) evaluated the impacts of COVID-19 in Germany, Italy, and Spain based on the WIOD. These studies could capture the disaster impacts resulting from the simultaneous supply and demand shocks due to the COVID-19 pandemic. In contrast to the models that employ the standard demand-side IO relationship, a combination of a demand-driven Leontief model for backward (demand-side) effects and a supply-driven Gosh model for forward (supply-side) effects by rectifying the shortcomings of the supply-driven IO model was applied for a hypothetical port shutdown case (Rose & Wei, 2013). Their model measured all the impact propagation paths by covering the interdependence of economic sectors both upstream and downstream in the supply chain of the disrupted goods. After all, the simple structure of the IO framework makes it possible for these various modifications to alleviate the drawbacks of the Leontief model for disaster impact analysis. Meanwhile, various other disaster-related features, such as temporal variations of negative and positive shocks and the subsequent recovery and reconstruction activities and behavioral and policy measures, are integrated into the IO framework. Avelino and Hewings (2019) introduced the generalized dynamic IO (GDIO) model that synthesizes several IO models, including the ARIO with inventory model, the sequential interindustry model (Romanoff & Levine, 1981), and the Batey-Weeks demo-economic model (Batey & Weeks, 1989) to simulate not only inter-temporal adjustments and intra-temporal production and market clearing but also changes in the labor market in a disaster situation. Man-made disasters, such as terrorist attacks, impose a different set of damage and adverse effects to an economy, as evidenced in negative psychological impacts. Unlike natural hazards that are hardly intended to prevent their occurrence, terrorist attacks can be avoided through policies and countermeasures, such as tightened security. To simulate such interactions between defender and attacker, Hwang and Park (2019) constructed the game theoretic national interstate economic model (G-NIEMO)

36  Handbook on the economics of disasters that combines a competitive game situation between a defender and an attacker for micro-level behavioral strategies with a macro-level IO-based national interstate economic model for the US aviation system. G-NIEMO can evaluate defense strategies against terrorist attacks by deriving the probabilistic impacts in the national context. A wide variation of extended IO models proves its popularity and ease of use for disaster impact analysis because the deficiencies of the IO model can be alleviated. However, the extended models become rather complex and require additional data and knowledge. 3.3  CGE Models A CGE model can be defined as a “multi-market simulation model based on the simultaneous optimizing behavior of individual consumers and firms in response to price signals, subject to economic account balances and resource constraints” (Rose, 2004, p. 26). Therefore, it consists of an enormous number of equations representing each actor (such as producers, consumers, and governments) that responds to changes in markets within an economy. As such, constructing a CGE model necessitates the collection of a huge data set, and running and calibrating the model also require the CGE modeler’s expertise. This is one of the major reasons that CGE models have rarely been employed by emergency management practitioners, while the traditional Leontief IO model and its less-extended variants—many of which can be run with a spreadsheet software—have been used in practice. For example, the Hazard US (HAZUS) software developed by the Federal Emergency Management Agency (FEMA) in the United States included an IO model based on Cochrane’s rebalancing algorithm (Cochrane, 1997) for estimating economic impacts of disasters. Yet, the CGE models’ advantages over IO variants are conspicuous, namely, the integration of resilience in the production process in a disaster situation. One of the innovations that can make CGE models for disaster impact analysis accessible to and operational by the practitioners and decision makers is the economic consequence analysis tool (E-CAT) for aviation system disruptions due to man-made disasters introduced by Rose et al. (2017) and Chen et al. (2017). E-CAT is a reduced-form CGE analysis tool, not a CGE model per se, for rapid estimates of the economic impacts of a disaster. E-CAT was constructed as follows: while varying user input variables, such as magnitude of threats, duration of disruptions, economic structure, location, behavioral attributes, and resilience, a full CGE model comprising 58 sectors and 9 household groups, government institutions, and international trade is run with 100 random draw scenarios. Three different regional economic structures are further added to the scenarios to derive 400 unique results of GDP and employment. These 400 results are regressed over the user input variables using ordinary least squares (OLS) and quantile regression models to produce a series of reduced-form equations. The estimated liner equations are further used, along with a Monte Carlo simulation, to propagate uncertainties in the outcomes. Using this E-CAT, practitioners and policy makers can simulate the model by themselves, altering some of the user input variables to observe the economic consequences of such cases. Whereas E-CAT enables the CGE analysis of a disaster operational to nonexperts, this would make E-CAT analysis more of a black box method because the users cannot comprehend how the tool works inside. E-CAT has been extended to incorporate spatial and dynamic dimensions as the generalized, regional, and dynamic E-CAT (GRADECAT; Dixon et al., 2019), which also implies a broader coverage of input variables (capital damage and recovery expenditures) and outputs (GDP and GRP [gross regional product] for the short and long run as well as economic welfare).

A few good models for economic analysis of disasters  37 As discussed in the previous section, another issue with the CGE models pertains to the values of elasticities, which rely on external sources and may not reflect disaster situations. Two studies tackled this issue by adjusting elasticity values so that estimated impacts, such as the changes in GRP, match with the observed numbers. Kajitani and Tatano (2018) considered the 2011 Great East Japan Earthquake and Tsunami as the case study to estimate the values of the elasticity of spatial substitution for intermediate and final goods in a spatial (interregional) CGE (SCGE) model of Japan. The empirically observed data of the index of industrial production (IIP) is the monthly index of changes in the sectoral production level published by the Ministry of Economy, Trade and Industry of Japan. Because of the monthly data, the employed SCGE model imposed a short-run closure rule by which labor and capital movement is restricted and the elasticity of factor substitution is set at a lower magnitude. Trial-and-error comparisons between the observed data and estimated results revealed that the values of elasticity of spatial substitution for most manufacturing sectors appear to be one-third of the original (ordinary time) values for both intermediate and final goods, except for the automobile parts sector, which was extensively damaged by the event. Using the same event and the same IIP empirical data, Yamazaki et al. (2018) also estimated the values of the elasticity of not only substitution between labor and capital but also spatial substitution. Unlike Kajitani and Tatano’s static model with a short-run closure rule, Yamazaki et al.’s model uses a monthly recursive dynamic SCGE model over 11 months. Their estimation procedure for the elasticity values appears to be an iterative process through which the values of elasticity of not only factor substitution but also spatial substitution are adjusted sequentially, not simultaneously, to narrow the differences between the derived results and the IIP values. While the adjusted elasticity values reproduced the observed declines in production level effectively, the estimated values of the elasticity of spatial substitution vary by sector between the Kajitani and Tatano study and the Yamazaki et al. study because of the differences in damage data conversion, model structure (static vs. recursive dynamic), and adjusting the value of elasticity of factor substitution. Some other notable extensions and/or modifications of CGE models for disaster impact analysis include the measuring the economic resilience of infrastructure tool (MERIT) and its extension, the dynamic equilibrium seeking (DES) model. MERIT is a dynamic, multiregional, and multisectoral systems model incorporating core features of CGE models such as price and quantity adjustments, and constant elasticity of substitution (CES) functions. However, it is formulated based on the framework of system dynamics to seek out general equilibrium using finite difference equations instead of optimization in a standard CGE model (McDonald et al., 2018). The DES model extends MERIT by integrating a broader range of systems, such as socioeconomic attributes (demographics, land use, transportation) and environmental factors (resources, energy, climate change, ecosystem service) (McDonald & McDonald, 2020). While the results from CGE models are considered more plausible than from other economic models, Zhou and Chen (2020) examined the reliability of CGE models applied to disaster impact analyses. In their study, a meta-analysis of 253 CGE simulations in 57 empirical studies was performed using an OLS model that controlled various factors, such as input data types, disaster types, closure rules, existence of resilience function, spatial structures, and temporal structures. The findings include varying impacts by disaster type, the outcomes being sensitive to model specification and closure rules, and the effectiveness of resilience for reducing the impacts. While this study may be the first to (1) compare numerous disaster impact analyses

38  Handbook on the economics of disasters that use CGE models and (2) identify the tendencies of modeling strategies and outcomes, how their OLS results should be construed remains uncertain because the values of estimated impacts in the sample CGE studies are not random variables. 3.4  Agent-Based Models Unlike the preceding economic models that seek and/or assume an equilibrium, ABMs are simulation models that agents (individuals, groups, or institutions) follow for their respective decision-making rules. Each agent is assigned heterogeneous properties, and their simultaneous interactions result in a complex behavior. In terms of disaster impact analysis, several recent studies employed the ABM framework to investigate the impacts on supply chains in a disaster situation. Dulame et al. (2020) used an ABM to model the impacts of a sudden demand surge for essential products, such as bottled water, canned food, and toilet paper, during a disaster due to consumer panic based on groundless rumors and/or anxiety. While such a demand surge will surely increase the price of such products, assuming that the price stays the same as the pre-disaster level because of the prohibition policy of price gouging during a disaster and households (consumers) trying to maintain their stock level of bottled water, the study focuses on the strategies for counteracting such excess demand to mitigate supply disruptions in a supply chain network of bottled drinking water. Li et al. (2021) also modeled the supply chain network of the automobile parts industry using the ABM framework to examine how the network’s topological structure affected disruption propagation in both forward and backward directions. Their ABM is set up as an approximation of a probabilistic cellular automata (PCA) because a PCA model will become highly complex and infeasible with their case study. The agents (automotive-parts firms) follow the rules of transition probabilities between nondisrupted and disrupted states based on their respective forward and backward linkages. The investment cost against disruptions (such as increase in inventory level and improvements of supply-chain management, production flexibility, and risk mitigation plan) and their benefit (reduction of vulnerability in both directions) are analyzed. While the preceding two studies assessed the supply chain of specific products/industries, a few others, namely Inoue and Todo (2019) and Colon et al. (2021), explored the economywide impacts of a disaster with an ABM built on the ARIO firm-network model (Henriet et al., 2012). The ARIO firm-network model is a dynamic model representing a production network among firms in a regional economy by disaggregating the sector-level IO tables. The dynamic feature mimics Romanoff and Levine’s (1993) sequential interindustry model (SIM) and adds inventory for intermediate inputs to the firm’s production process.7 A couple of constraints for production capacity of each firm are imposed either by insufficient production capacity due to damage by a disaster or by limitations of intermediate supply, implying the insufficient inventory level. A randomized set of disaggregated IO tables are generated as “synthetic” (p. 154) firm-level IO tables from a regional sector-level IO table and the simple network characteristics (numbers of firms per sector, intermediate suppliers per firm, and clients per firm), and they are consistent with the original regional IO table. Even though the modeling framework in this study is based on a closed IO framework (no imports or exports) and does not allow substitution among suppliers in a disaster situation, the outcomes of random disturbances in the supply chain network reveal the advantages of the disaggregated approach over the aggregated

A few good models for economic analysis of disasters  39 approach, which may be “too optimistic as they represent the most favorable case in which risks and losses are optimally shared among all producers” (pp. 165–166), whereas in reality losses are distributed heterogeneously among firms within a sector. Inoue and Todo (2019) applied an extended version of Henriet et al.’s (2012) ARIO firm-network model to the 2011 Great East Japan Earthquake and Tsunami case with a comprehensive data set of supplier-purchaser linkages in Japan.8 They made the original ARIO firm-network framework an ABM by setting firms and consumers as agents using the following rules: firms have a Leontief (IO) production function, try to maintain their pre-disaster inventory level, and have a recovery function as well as a rationing policy that fulfills intermediate demand first in the case of limited production capacity. Their simulation results reproduced a better macroeconomic trajectory after the event in terms of the propagations of higher-order effects and recoveries from the event than did the aggregated IO models. Further extending Henriet et al.’s and Inoue and Todo’s models, Colon et al. (2021) constructed an ABM based on the ARIO firm-network model for the United Republic of Tanzania and integrated it with a road network model to explore the transport-supply chain nexus. Their model has 324 major transportation nodes and 42 sectors as well as households. Unlike with Inoue and Todo’s study, the supplier-purchaser linkages were estimated using a modified gravity model based on the size and location of firms by sector; moreover, households tended to purchase products only from the firms located on the same node. In this model, international imports and exports are allowed based on similar gravity relationships as above. The trade flows between suppliers and purchasers are distributed within the transportation model. When a node or a network link is disrupted, trade flows find an alternative pathway either by rerouting through the lowest-cost journey available, but adding extra cost if an alternative route exists, or by holding the supplies at the producer’s premises if an alternative route does not exist. The model simulates a oneweek disruption of each transportation asset and estimates the resulting higher-order effects. The results generate a set of criticality maps that illustrate the criticality of transportation infrastructure for the overall economy. Based on these studies, a disaggregated micro-level model clearly not only extracts the finer details of economic impacts among agents but also provides more accurate estimations than do aggregated macroeconomic models because a great deal of additional information is integrated and the heterogeneity of damage or loss is considered. 3.5  Theoretical Models The long-term effects of disasters have been analyzed using theoretical models, such as growth models, that have recently made advancements with the increasing empirical analyses of disasters, as discussed in the next section. Because the scope of this chapter pertains to quantitative modeling that is applicable to empirical cases, in this subsection, the findings and contributions of the results are highlighted based on a theoretical analysis of the disaster process to further improve quantitative modeling. Focusing on the impact of disasters on consumption, Hallegatte and Vogt-Schilb (2019) used a set of simple production functions to theoretically examine how asset damage due to a disaster needs to be addressed. Their analysis reveals that the use of an aggregate production function, assuming a single homogenous capital, leads to the systematic underestimation of output losses and suggests that they should be estimated by using the average, instead of the

40  Handbook on the economics of disasters marginal, productivity of capital. In addition, it also demonstrates that the decline in net present consumption is expected to be lower where the productivity of capital is higher when a disaster damages a similar fraction of capital. This implies that reinstalling the destroyed capital demands less forgone consumption in capital-scarce developing countries. The former finding of capital-to-output transformation relates to the input-data issue of quantitative models for the empirical analysis discussed above, while the latter result hints at the role of consumption in the quantitative models.

4.  EMPIRICAL ANALYSES OF DISASTER IMPACTS Because of more frequent occurrences of disasters and the increasing interests in the economic impacts of disasters, empirical studies have been undertaken of a particular disaster and statistical investigations of disaster impacts across countries and/or disaster types. While each event (hazard or disaster) is a unique phenomenon, these studies present information about and contribute to improving our understandings of disaster processes, which may lead to further progress in modeling strategies. 4.1  Micro-Level Studies As discussed in the previous section, ARIO-based ABMs have clear advantages over aggregated macro-economic models, such as IO and CGE models, in terms of providing detailed impact distribution and deriving more accurate estimations of higher-order effects. Simultaneously, their micro-level modeling is limited to include supply-chain linkages among firms but does not endogenize how firms behave in a disaster situation, especially whether or not firms may survive with various damage levels. Firm-level or plant-level survival analyses of a particular disaster have been conducted. For example, business survival after Hurricane Katrina of 2005 in the United States was analyzed using a linear probability model of survival based on the US Census Bureau’s longitudinal business data on about 10,000 business establishments, as well as location and damage data from FEMA (Basker & Miranda, 2018). The results exhibit the very low survival rates of the firms that suffered physical damage, especially small firms and less-productive establishments, while large firms and more productive firms of all sizes managed to rebuild their operations quickly. The study also reasons that the impact on the local economy from a large number of small firms exiting may have induced a second wave of business exits because of the decreased intra-regional and interindustry relationships, claiming that prompt and timely recovery and reconstruction are essential to mitigate the spread of higher-order effects. Similar studies were performed for historic events: the changes in firm-level capital accumulation were analyzed before and after the 1923 Great Kanto Earthquake, revealing that larger firms tend to increase their capital investment in more heavily damaged areas (Okazaki et al., 2019). The firm survival in Nagoya City, Japan, after the 1953 Ise Bay Typhoon was examined using a probit survival model and a log-linear performance model based on firmlevel statistics and localized damage index. The results revealed that although the flooding caused by the typhoon did not make firms exit from business, in general, firms’ growth of sales and employment decreased, while the considerable sectoral heterogeneity of impacts

A few good models for economic analysis of disasters  41 was observed, especially between the manufacturing sector and the wholesale and retail sector (Okubo & Strobl, 2021). Furthermore, plant-level survival analyses with finer spatial scale and organizational detail than firm-level analysis were conducted for the 1995 Kobe earthquake: Tanaka (2015) employed a difference-in-difference estimation method with the matching technique to control for pre-event plant characteristics to investigate the disaster impacts on the growth of capital, employment, and value added. The evaluation revealed that the effects on the latter two elements were considerably negative, and the effects on capital was positive only a few years after the event, implying that “plants in Kobe suffered from over-investment in physical capital and a failure to enhance productivity” (p. 39). While Tanaka’s study assumed all the plants in Kobe suffered the same damage, Cole et al. (2019) controlled the local spatial heterogeneity of damage using geo-coded plant locations and building-level surveys and employed a panel data of manufacturing plants during 1992 and 2007. Their decomposition analysis uncovered that, unlike Tanaka’s results, there was an increase in productivity after the event that appears driven by the exit of plants and the entry of new plants into undamaged areas. The survival analysis indicates that damaged plants are inclined to exit up until seven years after the event. These studies provide useful insights for micro-level modeling for firmor plant-level survival, which could be integrated with a micro-level modeling framework such as ABM, although such integration necessitates further additional data. Cole et al. (2019) remarks that the use of aggregated data and modeling, either of damage or agent, would lead to “considerable measurement errors due to the heterogeneous nature” (p. 403) of damage and interactions among agents, depending on the type of hazard.9 4.2  Macro-Level Studies This subsection discusses empirical studies of a particular disaster on its short-term economic impacts in an aggregated context. The empirical studies using cross-country data are beyond the scope of this chapter and are discussed elsewhere in this book. One of the difficulties for the empirical studies of a disaster is to isolate the disaster impacts, including first-order losses and higher-order effects, from the observed macroeconomic statistics, such as changes in gross product, income, and employment, because even a catastrophic disaster rarely influences the trends of the corresponding macroeconomy (Albala-Bertrand, 2007), except for small island nations and least developed countries. Hence, the analysis of disaster impacts needs to be dealt with at a regional (subnational) level, whereas the trends of macroeconomic indicators of a region are heavily influenced by the national economic trends and policies. Thus, the disaster impacts may be muddled and difficult to distinguish from other macroeconomic influences. The long-term economic impacts of the 1995 Kobe earthquake in Japan on the GRP per capita and the local government expenditure of the Hyogo Prefecture have been studied by duPont and Noy (2015) using econometric models with synthetic control methodology to isolate the earthquake’s impacts from the macroeconomic indicators. Their results reveal that the 1995 earthquake resulted in substantial adverse and long-term impacts with a per capita GRP reduction of 12%, which did not recover to the previous level during the study period (1975–2009), while Hyogo Prefecture’s expenditures increased considerably after 1995, recording “at least a 15% increase each year after the disaster (with peaks of almost 60%)” (p. 797). They concluded that its true cost may be more than twice as large as the estimates released after the event. Ohtake et al. (2012) investigated the same disaster’s impacts

42  Handbook on the economics of disasters on the labor market, examining the data on job replacements and job openings between 1993 and 2009. Using an ARIMA model and controlling cyclical fluctuations of economic trends, they found that, for full-time jobs, a mismatch between labor supply and demand existed after the earthquake and that the number of job placements recovered within four years but declined later. However, for part-time workers, the labor supply and demand numbers dropped steeply, while the number of job placements rebounded in about five years but later declined again. These two studies suggest that the economic impacts of the 1995 Kobe earthquake had a complex trend of ups and downs wherein massive reconstruction demand and fiscal stimulus were injected into the regional economy and made positive impacts in the short term; however, this was insufficient to offset the long-term trend of decline in the regional economy. These temporal dynamics in a relatively short term present one of the challenges that models for disaster impact analysis are expected to address by reflecting these empirical observations. In addition, the extent of spatial propagation of higher-order effects is another challenge for disaster economic modeling. The spatial spillover of disaster impacts is examined based on the 2008 flash flood in the state of Santa Catarina, Brazil, using a spatially extended version of the dynamic difference-in-differences models and municipal-level data (Lima & Barbosa, 2018). The results indicate that the municipalities which suffered damage showed a 7.6% decrease of GRP per capita but regained their pre-disaster level in three years, while the ones that did not suffer damage experienced a 0.5% to 1.4% decline of GRP per capita as spillover effects. The spillover effects appear small but not negligible compared to the disaster impacts found in the hazard-hit municipalities, which had a higher per capita GRP and population than the municipalities that were not hazard-hit before the event. A dynamic difference-in-differences model was also employed by Husby et al. (2014) to isolate the impacts of the 1953 major flood in the Netherlands on population and a subsequent disaster prevention program (the Deltaworks) to mitigate flooding risks at the flooded areas. The analysis exposed that the population growth level decreased by approximately 0.6% in the affected municipalities compared to the non-affected ones. However, this tendency did not continue over time because the disaster risk was mitigated by the Deltaworks, resulting in a 0.75% higher annual population growth rate compared to the non-affected municipalities in the long term. While the Deltaworks not only reduced the flood risks but also constructed additional amenities in the affected areas, such as freshwater supply and transportation infrastructure, this type of reconstruction policy may have changed the local economies, considerably leading to the population increase. As such, the authors suggest that investigations into the demographic and economic impacts of a disaster “should analyze the disaster in conjunction with mitigation efforts following in its wake” (p. 370). To what extent the cost and benefit of disaster impacts should be counted, ranging from the duration to the geographical area, has been a debatable issue, as discussed in the introduction of this chapter. A wider range of socioeconomic effects of more than 10,000 disasters, including migration, housing prices, median family income, and poverty rates, has been explored using a fixed-effect linear econometric model based on the data of all the counties in the United States from 1920 to 2010 (Boustan et al., 2020). The study revealed that the counties which experienced severe disasters faced greater out-migration, lower housing prices, and higher poverty rates. These findings are consistent with the theory that “given the durability of housing capital, lower demand due to persistent natural disasters leads to falling rents and acts as a poverty magnet” (p. 12). The change in net out-migration from a county due to severe natural hazards was estimated to reflect a 1.5% increase, while some extreme events, such as the 1923 Great Mississippi Flood and the 2005 Hurricane Katrina recorded a 12%

A few good models for economic analysis of disasters  43 increase of net out-migration. Because of the comprehensive data in this study, the derived estimates reflect the average of many severe disasters, instead of a specific one. This kind of cross-case study can yield more generalized (or averaged) results, and the socioeconomic aspect of disaster impacts can be incorporated in demo-economic models, such as the GDIO model discussed in the previous section. The ex-post empirical studies of a particular disaster provide the observed changes in some indices and values (such as gross production, income, and employment) retrospectively, despite the values being mostly aggregated ones. Yet, they are useful to compare with and validate the estimated impacts from the preceding models. Furthermore, the generalization of the results from the previously mentioned studies or other empirical studies on specific disasters for the improvement of models should be done with caution because some of the findings may be specific to unique features of locality, development level, institutional settings, and so forth, which may not apply to other places. In this context, a further accumulation of empirical studies is needed so that a meta-analysis can be performed to control unique features for possible extraction of the generalized tendencies.

5.  FUTURE OPPORTUNITIES In this chapter, economic models for disaster impact analysis and the potential contributions of empirical studies to the advances of models are discussed, focusing on the recent progress. Because models are a representation of the specific aspects of the event and economy or economies, they inherently have limitations and disadvantages because they lack some other features. The recent advances presented earlier aim to overcome some of the limitations and/ or disadvantages, while the sophistication and extensions of modeling frameworks have also been achieved. Yet, no model is a perfect fit for all cases, and, more importantly, for the assessment of empirical cases, each event is a unique phenomenon with its own distinctive characteristics that require careful treatment. Therefore, the challenges to advancing the modeling frameworks for disaster impact analysis still remain. One of the challenges to further improvement is the validation of the estimated results. Proposed models ought to be evaluated in terms of how accurate the derived estimations are in comparison with the true value of impacts. For this purpose, as discussed in the previous section, the results from empirical analyses of a particular event, which use various methods to extract the disaster impacts from the observed indicators, can be the benchmark value for the evaluation of the estimated impacts by models. Although these empirical studies deal more with aggregated indicators, rather than the disaggregated sectoral-level estimations from models discussed in this chapter, they are still useful for validating the overall consistency of the model’s results. Comparisons may be difficult due to the different timing between the impact estimation using models and the empirical retrospective studies of the same event. The former analyses are usually done within a few years after the event occurrence, while the latter studies are performed after several years, even after a decade or so, because they need a sufficient number of data points after the event for statistical analysis. The disaster researchers’ willingness to review their estimation results after the empirical studies uncovered the true values is challenged here. As discussed in Section 3, operability has been one of the disadvantages of the CGE models, although E-CAT and GRAD-ECAT models addressed this issue by developing a reduced-form

44  Handbook on the economics of disasters CGE analysis tool so that the practitioners and policy makers can simulate how the changes in policies influence the outcomes in a disaster situation. While IO models are considered simple and easy to use, the efforts to overcome limitations have made them more sophisticated with the ability to deal with more features, but they have also become more intricate and less user friendly, especially when being operated by nonexperts. Cochrane’s rebalancing algorithm built into the HAZUS’s Indirect Economic Loss Module, which offered a balanced trade-off between modeling capability and required technical knowledge (Banks et al., 2014), was popular and widely used for mitigation planning and cost-benefit analysis by practitioners. However, it has unfortunately been discontinued. Avelino and Dall’erba (2019) compared several recent extended IO models to search for an alternative to Cochrane’s algorithm and suggested that the dynamic inoperability input-output model (DIIM; Lian & Haimes, 2006) can be considered a candidate because of its feature that endogenizes the recovery path based on local demand conditions. Overcoming the limitations while retaining simple operability is the current challenge that IO models face. Further improvements of economic impact estimation from disasters, especially from infrastructure failures, focusing on IO models were discussed by Kelly (2015). His suggestions, such as “improving terminology and definitions” and “better modeling of physical and economic linkages” (p. 9), have been discussed in the literature for the former (e.g., Rose, 2004; Okuyama, 2007) and in Section 2 for the latter. His remaining two recommendations are briefly discussed here. First, the study claims that models for disaster impact analysis “assume the structure of the economy remains unchanged after a disaster” (p. 10). Although this is true for the simpler versions of IO models, it is no longer true for IO models with linear or nonlinear programming, which optimize the interregional economic structure after the event. It is also not true for CGE models, for which regional economic and spatial economic structures are set to be flexible and endogenized based on the changes in a series of factors and constraints. Meanwhile, a series of studies that empirically investigated structural changes which occurred after the 1995 Kobe earthquake were performed using a time series of IO tables. They revealed that apparent structural changes within the regional economy emerged in terms of regional IO coefficients and regional purchase coefficients, implying that interindustry relationships as well as import dependencies by industry were affected by the disaster (Okuyama, 2014, 2015; Okuyama & Yu, 2019). These empirical findings and the extent to which models with flexible features optimize economic structures can be compared to clarify Kelly’s concern. Further improvements to these models will be necessary when the optimally adjusted structure and the empirically observed structural changes appear to be different. Kelly’s other suggestion is that the models for disaster economic analysis are mostly deterministic in nature, “despite the probabilistic nature of natural hazards and the stochastic nature of economic consequences” (p. 10). This point, incorporating uncertainty in the analysis, appears the least addressed in the current models, except for E-CAT and GRAD-ECAT, which employ statistical methods and Monte Carlo simulations for variabilities of the estimation results to some extent. One of the ways to incorporate uncertainty is to perform a Monte Carlo simulation of existing models to derive a range of estimation results. Another way is to convert the input data to stochastic value when transforming from damage (stock) data to loss (flow) data so that the input data have upper and lower bounds with an error distribution. Inclusion of uncertainty can be achieved within a model, especially for CGE models (Rose & Liao, 2005), if the resilience10 functions can become stochastic rather than deterministic so that the

A few good models for economic analysis of disasters  45 counteractions to shocks become varied. Incorporating uncertainty analysis into the estimation of disaster impacts leads to estimation results with upper and lower bounds and a distribution, which not only becomes more informative and plausible to assess various mitigation and/or adaptation strategies but also makes the economic impact analysis more compatible with other fields of disaster research that have dealt with uncertainty. A few disaster features that have become increasingly important for disaster impact analysis over the years have not been included for discussion in this chapter. One of them is the analysis of distributional impacts in a disaster. As Albala-Bertrand (1993) pointed out, most casualties in disasters are “the poorest countries and their weakest socio-political groups” (p. 90). The distributional impacts within a country/region and among countries in cases such as the 2004 Indian Ocean earthquake and tsunami and the COVID-19 pandemic need to be investigated to elucidate the incidence of disaster policy, that is, who pays and who benefits from public policies on disaster mitigation and preparedness measures. Rose (2004) suggested the importance of disaster impact models in providing distributional estimates of impacts for decision makers and citizens to understand the nature of disaster impacts. Karim and Noy (2019) provide an excellent review of the literature in this area, and Hallegatte et al. (2020) illustrate the framework of distributional impacts by analyzing the recent empirical literature and suggesting policy implications for reduced inequality and poverty through disaster countermeasures and relief. Another disaster feature that is not discussed in this chapter relates to the behavioral changes in a disaster situation. Dacy and Kunreuther (1969) may have been the first to address this feature, pointing out that the sympathetic behavior of mutual aid in a disaster leads to the situation wherein “the supply and demand curves may shift in unexpected ways” (p. 64). Furthermore, Albala-Bertrand (1993) suggest that the “in-built societal mechanism” may offset the higherorder effects of a disaster on the economy and society. These features are considered to be positive changes emerging from the society and are different from “resilience” in production systems. In contrast, sympathetic sentiment toward the casualties and people affected by a catastrophic disaster may lead to self-restrained consumption in non-affected regions, becoming a negative impact as a result of demand decline. Okuyama et al. (1999) evaluated this type of demand decline at the 1995 Kobe earthquake in Japan with an interregional IO model of Japan. The behavioral effects of a terrorist attack scenario, such as an increase in supply cost of resources to the affected region and the decrease in demand for goods produced in the region due to a fear factor, were assessed using a CGE model based on a nationwide survey by Giesecke et al. (2012). Since the behavioral effects vary depending on the nature of the threat (hazard), for example, between natural hazards and man-made hazards, as well as the duration of the threats (short term or long term), this area of investigation demands not only case studies across various hazard types but also theoretical and analytical developments using knowledge of behavioral economics and psychology. Meanwhile, in this handbook, Rose (2021) discusses an analytical framework evaluating the behavioral effects of disasters and their consequences as a part of disaster impacts, especially on behavioral shifts resulting from the changes related to fear and risk perception. The challenges of the models discussed in this chapter present clear opportunities to further advance not only the modeling frameworks for disaster impact analysis but also our understanding of the disaster process. Such advances will surely unearth the truth of disaster impacts.

46  Handbook on the economics of disasters

NOTES * The author would like to express gratitude to Mark Skidmore, the editor of this handbook, and Adam Rose for their encouragement and comments that significantly improved the contents of this chapter. This research was partially supported by a JSPS Grant-in-Aid for Scientific Research JP20K01658. 1. The distinction between the definitions of effect and impact in the Handbook for Disaster Assessment (UN ECLAC, 2014) appears to be the difference between the consequences at the micro and macro level. Because the models discussed in this chapter deal with macro-level consequences, while some empirical studies reviewed in the later section deal with micro-level effects, the term “disaster impact analysis” is used to refer to studies employing these models in this chapter. 2. Rose (2004) suggested that the use of higher-order effects, instead of indirect effects (or impacts), commonly found in disaster-related literature, is more appropriate because of the conflict with the terminology used in IO models. Moreover, it also “is intended to be general enough to include both IO quantity interdependence and general equilibrium price interdependence effects” (p. 17). 3. Further overviews and discussions of these models can be found in Rose (2004), Okuyama (2007), Greenberg et al. (2007), Okuyama and Santos (2014), and Galbusera and Giannopoulos (2018). 4. More detailed discussions about the supply-driven IO models can be found in Oosterhaven (1996, 2012). 5. A supply-use model consists of a supply table (industry-to-commodity transactions) and a use table (commodityto-industry transactions), whereas an IO model is based on an IO table (industry-to-industry transactions). Note that this approach is often referred to as the “use-make” system of rectangular matrices because modern industries produce a range of commodities. A supply-use model provides more detailed information about how industries interact with each other through the supply and use of their products (commodities). A set of supply and use tables can be converted to an IO table, but not vice versa. For more detailed descriptions about the supply-use framework, please see Miller and Blair (2022). 6. This regional CGE model is called the Italian Economic Equilibrium System (IEES) in Koks et al. (2015) and was originally developed by Standardi et al. (2014). 7. Hallegatte’s (2014) ARIO-inventory model also includes an inventory function similar to the one found in Henriet et al. (2012); however, the ARIO-inventory model is an aggregated sector-level model, instead of a firm-level one, and has been applied to the analysis of 2005 Hurricane Katrina’s economic impacts, instead of the hypothetical cases in the Henriet et al.’s study. 8. Their model has 887,715 firms and 3,223,137 supplier-purchaser linkages. 9. Geophysical hazards, such as earthquakes, cause a heterogeneous distribution of damage, whereas meteorological hazards, such as flooding and severe storms, result in a relatively homogeneous damage over space. 10. Another challenge related to resilience in disaster impact modeling is to empirically measure the value of resilience for use in models. Dormady et al. (2021) in this handbook reviewed and summarized the recent literature on empirical estimation of resilience at the microeconomic level, while Hashiguchi et al. (2021) evaluated the resilience values at the macroeconomic level (country level) using the Organisation for Economic Co-operation and Development’s (OECD’s) annual Inter-Country Input-Output tables from 1995 to 2011.

REFERENCES Albala-Bertrand, J. M. (1993). The political economy of large natural disasters: With special reference to developing countries. Oxford, UK: Clarendon Press. Albala-Bertrand, J. M. (2007). Globalization and localization: An economic approach. In H. Rodriguez, E. L. Quarantelli, & R. R. Dynes (Eds.), Handbook of disaster research (pp. 147–167). New York, NY: Springer. Albala-Bertrand, J. M. (2013). Disasters and the Networked Economy. Oxson, United Kingdom: Routledge. Arto, I., Andreoni, V., & Cantuche, J. M. R. (2015). Global impacts of the automotive supply chain disruption following the Japanese earthquake of 2011. Economic Systems Research, 27(3), 306–323. Avelino, A. F. T., & Dall’erba, S. (2019). Comparing the economic impact of natural disasters generated by different input-output models: An application to the 2007 Chehalis River Flood (WA). Risk Analysis, 39(1), 85–104. Avelino, A. F. T., & Hewings, G. J. D. (2019). The challenge of estimating the impact of disasters: Many approaches, many limitations and a compromise. In Y. Okuyama & A. Rose (Eds.), Advances in spatial and economic modeling of disaster impacts (pp. 163–189). Cham, Switzerland: Springer. Banks, J. C., Camp, J. V., & Abkowitz, M. D. (2014). Adaptation planning for floods: A review of available tools. Natural Hazards, 70(2), 1327–1337. Basker, E., & Miranda, J. (2018). Taken by storm: Business financing and survival in the aftermath of Hurricane Katrina. Journal of Economic Geography, 18, 1285–1313.

A few good models for economic analysis of disasters  47 Batey, P., & Weeks, M. (1989). The effects of household disaggregation in extended input-output models. In R. Miller, K. Polenske, & A. Rose (Eds.), Frontiers of input-output analysis (pp. 119–133). New York, NY: Oxford University Press. Boustan, L. P., Kahn, M. E., Rhode, P. W., & Yanguas, M. L. (2020). The effect of natural disasters on economic activity in US counties: A century of data. Journal of Urban Economics, 118, 103257. Chen, Z., Rose, A. Z., Prager, F., & Chatterjee, S. (2017). Economic consequences of aviation system disruptions: A reduced-form computable general equilibrium analysis. Transportation Research Part A, 95, 207–226. Cochrane, H. (1997). Economic impacts of a midwestern earthquake. Quarterly Publication of NCEER, 11(1), 1–20. Cole, M. A., Elliott, R. J. R., Okubo, T., & Strobl, E. (2019). Natural disasters and spatial heterogeneity in damages: The birth, life and death of manufacturing plants. Journal of Economic Geography, 19, 373–408. Colon, C., Hallegatte, S., & Rozenberg, J. (2021). Criticality analysis of a country’s transport network via an agentbased supply chain model. Nature Sustainability, 4, 209–215. Dacy, D. C., & Kunreuther, H. (1969). The economics of natural disasters: Implications for federal policy. New York, NY: The Free Press. Dixon, P. D., Maureen, M. J., Rimmer, T., & Wittwer, G. (2019). Rapid assessments of the economic implications of terrorism events using a regional CGE model: Creating GRADECAT (generalized, regional and dynamic economic consequence analysis tool). In Y. Okuyama & A. Rose (Eds.), Advances in spatial and economic modeling of disaster impacts (pp. 121–161). Cham, Switzerland: Springer. Dormady, N., Rose, A., & Morin, B. (2022). Advances in the empirical estimation of disaster resilience. In M. Skidmore (Ed.), Handbook on the economics of disasters. Cheltenham, UK: Edward Elgar Publishing. Dulam, R., Furuta, K., & Kanno, T. (2020). Development of an agent-based model for the analysis of the effect of consumer panic buying on supply chain disruption due to a disaster. Journal of Advanced Simulation in Science and Engineering, 7(1), 102–116. duPont W., & Noy, I. (2015). What happened to Kobe? A reassessment of the impact of the 1995 earthquake. Economic Development and Cultural Change, 63(4), 777–812. Faturay, F., Sun, Y-Y., Dietzenbacher, E., Malik, A., Geschke, A., & Lenzen, M. (2020). Using virtual laboratories for disaster analysis—a case study of Taiwan. Economic Systems Research, 32(1), 58–83. Galbusera, L., & Giannopoulos, G. (2018). On input-output economic models in disaster impact assessment. International Journal of Disaster Risk Reduction 30, 186–198. Garcia-Hernandez, J. A., & Brouwer, R. (2021). A multiregional input-output optimization model to assess impacts of water supply disruptions under climate change on the Great Lakes economy. Economic Systems Research, 33(4), 509–535. Giesecke, J. A., Burns, W. J., Barrett, A., Bayrak, E., Rose, A., Slovic, P., & Suher, M. (2012). Assessment of the regional economic impacts of catastrophic events: CGE analysis of resource loss and behavioral effects of an RDD attack scenario. Risk Analysis, 32(4), 538–600. Greenberg, M. R., Lahr, M., & Mantell, N. (2007). Understanding the economic costs and benefits of catastrophes and their aftermath: A review and suggestions for the U.S. Federal Government. Risk Analysis, 27, 83–96. Guan D., Wang, D., Hallegatte, S., Davis, S. J., Huo, J., Li, S., Bai, Y., Lei, T., Xue, Q., Coffman, D., Cheng, D., Chen, P., Liang, X., Xu, B., Lu, X., Wang, S., Hubacek, K., & Gong, P. (2020). Global supply-chain effects of COVID-19 control measures. Nature Human Behaviour, 4, 1–11. Hallegatte, S. (2008). An adaptive regional input-output model and its application to the assessment of the economic cost of Katrina. Risk Analysis, 28(3), 779–799. Hallegatte, S. (2014). Modeling the role of inventories and heterogeneity in the assessment of the economic costs of natural disasters. Risk Analysis, 34(1), 151–167. Hallegatte, S., & Przyluski, V. (2010). The economics of natural disasters. CESifo Forum, 2, 14–24. Hallegatte, S., & Vogt-Schilb, A. (2019). Are losses from natural disasters more than just asset losses? In Y. Okuyama & A. Rose (Eds.), Advances in spatial and economic modeling of disaster impacts (pp. 15–42). Cham, Switzerland: Springer. Hallegatte, S., Vogt-Schilb, A., Rozenberg, J., Bangalore, M., & Beaudet, C. (2020). From poverty to disaster and back: A review of the literature. Economics of Disasters and Climate Change, 4, 223–247. Hashiguchi, Y., Yamano, N., & Webb, C. (2021). How thick is your armour? Measuring economic resilience to shocks in global production networks. Economic Systems Research. https://doi.org/10.1080/09535314.2021.1958764. Heatwole, N., & Rose, A. (2013). A reduced-form rapid economic consequence estimating model: Application to property damage from U.S. earthquakes. International Journal of Disaster Risk Science, 4, 20–32. Henriet, F., Hallegatte, S., & Tabourier, L. (2012). Firm-network characteristics and economic robustness to natural disasters. Journal of Economic Dynamics & Control, 36, 150–167. Husby, T. G., de Groot, H. L. F., Hofkes, M. W., & Dröes, M. I. (2014). Do floods have permanent effects? Evidence from the Netherlands. Journal of Regional Science, 54(3), 355–377. Hwang, H., & Park, J. Y. (2019). The game-theoretic national interstate economic model: Economically optimizing U.S. aviation security policies against terrorist attacks. In Y. Okuyama & A. Rose (Eds.), Advances in spatial and economic modeling of disaster impacts (399–421). Cham, Switzerland: Springer.

48  Handbook on the economics of disasters Inoue, H., & Todo, Y. (2019). Firm-level propagation of shocks through supply-chain networks. Nature Sustainability, 2, 841–847. Kajitani, Y., & Tatano, H. (2014). Estimation of production capacity loss rate after the Great East Japan Earthquake and Tsunami in 2011. Economic Systems Research, 26(1), 13-38. Kajitani, Y., & Tatano, H. (2018). Applicability of a spatial computable general equilibrium model to assess the shortterm economic impact of natural disasters. Economic Systems Research, 30(3), 289–312. Karim, A., & Noy, I. (2019). The empirical literature on poverty and natural disasters. In S. K. Sharma & E. Quah (Eds.), Economics of natural disasters (pp. 15–47), World Scientific Publishing Co. Pte. Ltd. https://doi. org/10.1142/9789814723237_0002. Kelly, S. (2015). Estimating economic loss from cascading infrastructure failure: A perspective on modelling interdependency. Infrastructure Complexity, 2(7), 1–13. https://doi.org/10.1186/s40551-015-0010-y. Klomp, J. (2016). Economic development and natural disasters: A satellite data analysis. Global Environmental Change, 36, 67–88. Koks, E. E., Carrera, L., Jonkeren, O., Aerts, J. C. J. H., Husby, T. G., Thissen, M., Standardi, G., & Mysiak, J. (2015). Regional disaster impact analysis: Comparing input-output and computable general equilibrium models. Natural Hazards and Earth System Sciences, 3, 7053–7088. Koks, E. E., Pant, R., Husby, T., Többen, J., & Oosterhaven, J. (2019). Multiregional disaster impact models: Recent advances and comparison of outcomes. In Y. Okuyama & A. Rose (Eds.), Advances in spatial and economic modeling of disaster impacts (pp. 191–218). Cham, Switzerland: Springer. Koks, E. E., & Thissen, M. (2016). A multiregional impact assessment model for disaster analysis. Economic Systems Research, 28(4), 429–449. Li, J., Crawford-Brown, D., Syddall, M., & Guan, D. (2013). Modeling imbalanced economic recovery following a natural disaster using input-output analysis. Risk Analysis, 33(10), 1908–1923. Li, Y., Chen, K., Collignon, S., & Ivanov, D. (2021). Ripple effect in the supply chain network: Forward and backward disruption propagation, network health and firm vulnerability. European Journal of Operations Research, 291(3), 1117–1131. Lian, C., & Haimes, Y. Y. (2006). Managing the risk of terrorism to interdependent infrastructure systems through the dynamic inoperability input-output model. Systems Engineering, 9(3), 241–258. Lima, R. C., & Barbosa, A. V. B. (2019). Natural disasters, economic growth and spatial spillovers: Evidence from a flash flood in Brazil. Papers in Regional Science, 98, 905–924. McDermott, T. K. J., Barry, F., & Tol, R. S. J. (2014). Disasters and development: Natural disasters, credit constraints, and economic growth. Oxford Economic Papers, 66, 750–773. McDonald, G. W., Smith, N. J., Kim, J. H., Brown, C., Buxton, R., & Seville, E. (2018). Economic systems modelling of infrastructure interdependencies for an Alpine Fault earthquake in New Zealand. Civil Engineering and Environmental Systems, 35, 57–80. McDonald, N. J., & McDonald, G. W. (2020). Towards a dynamic equilibrium-seeking model of a closed economy. Systems, 8(4), 42, 1–25. Miller, R. E., & Blair, P. D. (2022). Input-Output Analysis: Foundations and Extensions (Third Edition). Cambridge, United Kingdom: Cambridge University Press. Ohtake, F., Okuyama, N., Sasaki, M., & Yasui, K. (2012). Impacts of the Great Hanshin-Awaji Earthquake on the labor market in the disaster areas. Japan Labor Review, 9(4), 42–63. Okazaki, T., Okubo, T., & Strobl, E. (2019). Creative destruction of industries: Yokohama City in the Great Kanto Earthquake, 1923. The Journal of Economic History, 73(1), 1–31. Okubo, T., & Strobl, E. (2021). Natural disasters, firm survival, and growth: Evidence from the Ise Bay typhoon, Japan. Journal of Regional Science, 61(5), 944–970. Okuyama, Y. (2007). Economic modeling for disaster impact analysis: Past, present, and future. Economic Systems Research, 19(2), 115–124. Okuyama, Y. (2014). Disaster and economic structural change: Case study on the 1995 Kobe earthquake. Economic Systems Research, 26(1), 98–117. Okuyama, Y. (2015). How shaky was the regional economy after the 1995 Kobe earthquake? A multiplicative decomposition analysis of disaster impact. Annals of Regional Science, 55(2), 289–312. Okuyama, Y., Hewings, G. J. D., & Sonis, M. (1999). Economic impacts of an unscheduled, disruptive event: A Miyazawa multiplier analysis. In G. J. D. Hewings, M. Sonis, M. Madden, & Y. Kimura (Eds.), Understanding and interpreting economic structure (pp. 113–144). Heidelberg, Germany: Springer-Verlag. Okuyama, Y., & Santos, J. R. (2014). Disaster impact and input-output analysis. Economic Systems Research, 26(1), 1–12. Okuyama, Y., & Yu, K. D. (2019). Return of the inoperability. Economic Systems Research, 31(4), 467–480. Oosterhaven, J. (1996). Leontief versus Ghoshian price and quantity models. Southern Economic Journal, 62, 750–759. Oosterhaven, J. (2012). Adding supply-driven consumption makes the Ghosh model even more implausible. Economic Systems Research, 24, 101–111.

A few good models for economic analysis of disasters  49 Oosterhaven, J. (2017). On the limited usability of the inoperability IO model. Economic Systems Research, 29, 452–461. Oosterhaven, J., & Bouwmeester, M. C. (2016). A new approach to modeling the impact of disruptive events. Journal of Regional Science 56(4), 583–595. Oosterhaven, J., & Többen, J. (2017). Wider economic impacts of heavy flooding in Germany: A nonlinear programming approach. Spatial Economic Analysis, 12(4), 404–428. Pichler, A., & Farmer, J. D. (2021). Simultaneous supply and demand constraints in input-output networks: The case of COVID-19 in Germany, Italy, and Spain. Economic Systems Research. https://doi.org/10.1080/09535314.202 1.1926934. Romanoff, E., & Levine, S. H. (1981). Anticipatory and responsive sequential interindustry models. IEEE Transactions on Systems, Man, and Cybernetics, 11(3), 181–186. Romanoff, E., & Levine, S. H. (1993). Information, interindustry dynamics, and the service industries. Environment and Planning A, 25(3), 305–316. Rose, A. (2004). Economic principles, issues, and research priorities in hazard loss estimation. In Y. Okuyama & S. E. Chang (Eds.), Modeling spatial and economic impacts of disasters (pp. 13–36). New York, NY: Springer. Rose, A. (2022). Behavioral economic consequences of disasters. In M. Skidmore (Ed.), Handbook on the economics of disasters. Cheltenham, UK: Edward Elgar Publishing. Rose, A., & Liao, S. Y. (2005). Modeling regional economic resilience to disasters: A computable general equilibrium analysis of water service disruptions. Journal of Regional Science, 45, 75–112. Rose, A., Prager, F., Chen, Z., & Chatterjee, S. (2017). Economic consequence analysis of disasters: The E-CAT software tool. Singapore: Springer. Rose, A., & Wei. D. (2013). Estimating the economic consequences of a port shutdown: The special role of resilience. Economic Systems Research, 25(2), 212–232. Standardi, G., Bosello, F., & Eboli, F. (2014). A sub-national version of the GTAP model for Italy. Working paper of the Fondatione Eni Enrico Mattei, Milan, Italy. Tanaka, A. (2015). The impacts of natural disasters on plants’ growth: Evidence from the Great Hanshin-Awaji (Kobe) earthquake. Regional Science and Urban Economics, 50, 31–41. Theil, H. (1967). Economics and information theory. Amsterdam, the Netherlands: North-Holland. Toyoda, T., Cui, Q., Ikeda, M, Sato, S., Horie, S., Nakamura, H., & Fujiwara, H. (2020). A study on real-time estimation of direct economic losses caused by major earthquakes. Journal of Social Safety Science, 36, 1–10. UN ECLAC. (2014). Handbook for disaster assessment. Santiago, Chile: United Nations. Valles, A. C., Ferrer, M. M., Poljanšek, K., & Clark, I. (Eds.). (2021). Science for disaster risk management 2020: Acting today, protecting tomorrow. Luxembourg: Publications Office of the European Union. Yamazaki, M., Koike, A., & Sone, Y. (2018). A heuristic approach to the estimation of key parameters for a monthly, recursive, dynamic CGE model. Economics of Disasters and Climate Change, 2, 283–301. Zhou, L., & Chen, Z. (2020). Are CGE models reliable for disaster impact analyses? Economic Systems Research, 33(1), 20–46.

4. Behavioral economic consequences of disasters Adam Rose*

1. INTRODUCTION Many disasters generate dread and fear among the directly affected population and often also among those who believe they may be subsequently affected by the current or a similar future event. This “fear factor” is intentional in the case of terrorist attacks, accidental in the case of toxic spills, and intrinsic in the case of pandemics. The fear “causes” behavior that often exacerbates the direct losses. For example, 9/11 not only destroyed the World Trade Center (WTC) and caused business interruption (BI) among its former tenants, but it also caused a decline in airline travel and related tourism in the United States for nearly two years. The COVID-19 pandemic has resulted in a similar effect on air travel, as well as more general trepidation about venturing out of the house to engage in a broad range of other types of economic activity. Behavioral reactions are not limited to consumers. Workers may be apprehensive about returning to jobsites due to fear of lingering contamination (even after an all-clear sign has been issued) from a dirty bomb attack, toxic spill, or disease epidemic. Business owners may board up and shut down their stores in anticipation of riots. Governments may react prudently or overreact in forcing evacuations of entire regions in the path of a hurricane, or closing down all nonessential businesses in their jurisdiction during a pandemic. One challenge is to distinguish the part of this behavioral response that is a warranted reaction to risk from the portion representing an overreaction. A major question is whether these behavioral consequences should be included in benefitcost analysis (BCA) of projects, products, or policies intended to reduce disaster losses. We know that these effects take place, so one of the main issues is whether they can be isolated and measured in a way that avoids overlaps or double-counting with other effects in particular and whether they are consistent with BCA principles in general (see, e.g., Boardman et al., 2018; Farrow & Rose, 2018; von Winterfeldt et al., 2020). We need to be able to conceptualize and measure these effects so that they are neither overcounted or undercounted. A cornerstone of BCA is the measurement of impacts on resource utilization. In the case of behavioral effects, this often extends beyond the original affected market into other markets. An example again is the decline in airline travel following 9/11, where it is appropriate to moderate that negative impact by accounting for offsetting effects of people substituting other travel modes or other types of spending entirely for the reduced travel by air (Rose et al., 2009). Business shutdowns during the COVID pandemic result in production losses but are offset by increased consumer savings and subsequent purchases stemming from pent-up demand. Employees not coming to the work site during the pandemic can telecommute or return at a later date to work overtime or extra shifts to help businesses make up lost production (Walmsley et al., 2021b, 2021c). These are all examples of resilience, which reduces overall BI losses after the disaster has struck by using remaining resources as efficiently as possible (Rose, 2007, 2017; Cutter, 2016). The bottom line is to determine the net effect on resource utilization. This is also necessary for the impacts extending to other markets through indirect 50

Behavioral economic consequences of disasters  51 (multiplier, general equilibrium, supply chain) effects emanating from the direct behavioral impacts and resilience adjustments. The purpose of this chapter is to develop an economic analytical framework for estimating the behavioral effects of disasters and their consequences for disaster losses. The reduction of these losses represents some of the benefits of pre-disaster mitigation/interdiction and post-disaster resilience/recovery. We provide conceptualizations, definitions, classifications, examples, and empirical results of this category of economic consequences. We also examine methods used to measure behavioral reactions to fear for insight into improving their delineation. The analysis is intended to serve as the basis for the legitimate inclusion of behavioral consequences in BCA. Because we are interested in a comprehensive assessment of behavioral effects, we also cover resilience adjustments and extend our initial partial equilibrium analysis to the general equilibrium analysis level. We distinguish three major categories of behavioral responses affecting BI losses once the disaster strikes: • Mandatory avoidance behavior. This is typically enacted by government decree, as in business closures and general population stay-at-home orders to reduce the spread of COVID-19 during 2020, or evacuation orders in anticipation of a hurricane. • Voluntary avoidance behavior. This refers to individuals or groups refraining from engaging in economic activities, typically out of fear, as in the major downturn in airline travel following 9/11 or people avoiding crowded areas during the COVID pandemic. In this case, no viable inducements exist to influence people to alter their behavior. • Aversion behavior. This refers to cases where people are inclined to refrain from various activities but can be induced to alter their behavior by such mechanisms as price discounts to patrons of restaurants in areas recently affected by a chemical/biological/ radiological attack or providing salary premiums to workers to return to jobsites there.1 All of the aforementioned impacts are typically included in what has come to be known as economic consequence analysis (ECA). This is an established term representing applications of economic analysis to estimating the direct and indirect impacts of natural and man-made disasters (Dixon et al., 2020; Rose, 2009, 2015; Zhou & Chen, 2021). It represents a broader analytical framework than traditional economic impact analysis, which is typically applied to more straightforward activities such as opening a new mine or closing an automobile plant. The broadening incorporates features of behavioral linkages and resilience into the analysis.2 We will attempt to draw some boundaries on the inclusion of behavioral impacts in measuring the consequences of disasters, with special attention to the requirements of BCA. This will be done in the course of addressing questions such as the following: • To what extent can causation be demonstrated between the disaster and the behavioral impact? • Is there a limit on the inclusion of behavioral impacts with respect to time and space? • To what extent can behavioral effects spill across markets? • To what extent do we need to factor in offsetting effects? • To what extent can political responses be considered behavioral effects of disasters? • Is there a meaningful distinction between what has been termed avoidance versus aversion behavior?

52  Handbook on the economics of disasters • What other behavioral effects in addition to fear should be taken into account? • Does it matter whether behavioral effects are consistent with risk preferences based on science-based estimation of probabilities or subjective perceptions? • To what extent can standard economic analysis encompass behavioral responses? The following section provides some illustrative examples of behavioral consequences and the evolution of their measurement. Section 3 makes important distinctions in the realm of behavioral consequences and risk aversion. Section 4 explores the scope of these consequences and some potential lines of demarcation. Section 5 presents a categorization of behavioral consequences of disasters in relation to aspects of supply and demand at both the partial and general equilibrium levels of analysis. Section 6 summarizes the extent to which behavioral responses are amenable to standard welfare analysis and addresses the legitimacy of including behavioral consequences in BCA. Throughout the chapter, we identify aspects that are beyond its scope but that are prime candidates for future research. Finally, we acknowledge that the vast authority of the literature surveyed in the analysis presented is quite ethnocentric in focusing on the United States. In addition to its methodological contributions, our analysis reaches some important conclusions. Behavioral responses to disasters can often be substantial and hence merit inclusion in the examination of disaster consequences. There are some unique features of individual types of behavioral responses that need to be considered for accurate empirical estimation. Finally, the consequences can be moderated by resilience, but their influence on the bottom line is typically increased by their indirect or general equilibrium effects.3,4

2.  ILLUSTRATIVE EXAMPLES We begin with some examples of behavioral impacts of disasters, their potential limits, and how they have been modeled. The latter is intended to help demonstrate how these impacts represent standard aspects of economic causation and measurement. Behavioral effects are often related directly to a given product or market and can extend to the more general category of ripple or indirect effects, but a key question is where we should draw the line. A classic example of what most consider overstepping the bounds is the assessment of the economic consequences of 9/11. For example, Nobel Prize laureate Joe Stiglitz and others have suggested that we should include the trillions of dollars spent on the second Iraq War as a reaction to that disaster (New York Times, 2011). Is this a behavioral reaction to the disaster? If not, should it be considered within some other category of impacts? Another example is the case of a foot-and-mouth disease (FMD) epidemic (see, e.g., Oladosu et al., 2013). When an outbreak is first verified, it is standard practice to slaughter not only all the animals in the herd, but all the animals in the near and even distant vicinity. This is a form of mitigation against the spread of the disease, but a key question is how much of this is warranted as a precaution (margin of safety) and how much of it is an overreaction that represents a wasteful use of resources (see, e.g., Elbakidze & McCarl, 2006). Moreover, it is not clear that this distinction matters—rather, both types of reactions result in the loss of the market values of the animals that have died from the disease or been slaughtered to reduce its spread. Another aspect is the discovery of FMD in the United States causing other countries to ban imports of US beef. It is not always clear whether this is a justified precaution or just a political excuse

Behavioral economic consequences of disasters  53 to reduce imports in favor of domestic industry. Oladosu et al. (2013) adjusted final demand for US beef exports in simulating an FMD epidemic using a computable general equilibrium (CGE) model. The response is the standard demand shift and, moreover, is unlikely to be offset by substitution of spending in the United Kingdom on other imports from the United States to any significant extent. Is the inclusion of the decision by foreign governments to ban US beef imports a bridge too far in the estimation of losses from an FMD epidemic? At the same time, it did happen and had a significant effect on US beef production and exports. It thereby resulted in a negative impact on US GDP, though not necessarily according to some welfare measures, such as those using personal consumption as a proxy (see, e.g., Dixon & Rimmer, 2010; Rose et al., 2020), because consumers in the producing country were not affected directly. GDP impacts are important in their own right and are the major economic indicator used in ECA.5 Therefore, the aforementioned should be incorporated into the overall estimate of the economic consequences of the epidemic, whether we link them to behavioral considerations or some other cause (possibly a category of “policy responses”). One of the first examples of the inclusion of monetized estimates of behavioral impacts in disaster-related studies was by von Winterfeldt and O’Sullivan (2006) in the course of a BCA of an infrared jamming device to protect commercial airliners from shoulder-mounted missile launchers known as man-portable air defense system (MANPADS). The major component of the benefits was estimating the prevention of reduced airline travel demand if such an attack were successful. This was based on a crude measure of the reduction in airline travel following 9/11, including an estimate that the authors adapted from Gordon et al. (2007) to be as high as $400 billion. A similar inclusion was part of a study of the economic consequences of the WTC attacks by Rose et al. (2009), which de-trended the time-path of the nearly two-year decline in airline travel and related tourism for the pre-9/11 recession that year and estimated its impact at $85 billion in 2006 dollars ($120 billion in 2021 dollars). Moreover, there was further adjustment downward in the analysis by considering the substitution of other travel modes. This adjustment was accomplished by using the substitution possibilities in the consumption portions of a CGE model. Despite this offsetting effect, the fear-based effect was responsible for more than 80% of the net total of losses due to the WTC attacks.6 Rose et al. (2017a) performed a similar type of analysis with an even more detailed investigation of intermodal transportation substitution in an examination of the behavioral responses to attacks on an airliner and on an airport. The preceding are now classic examples of voluntary avoidance behavior motivated by fear (see, e.g., Gertz et al., 2019). This pertains to the reaction to COVID-19 of people avoiding retail shopping in stores, restaurant dining, public assembly gatherings, using public transportation, and even workplaces and schools. This reaction represents outright halts to economic activity (BI), where even pricing inducements have little or no effect, as in the height of the COVID pandemic (Byrd & John, 2021; Walmsley et al., 2021a) and others. Giesecke et al. (2012) identify another class of behavioral responses in investigating impacts of a dirty bomb attack in the Los Angeles financial district: (1) the need to provide a wage premium to employees to induce them to return to work in the once-contaminated (and likely perceived currently contaminated) area, (2) the need to provide investors with a higher rate of return in that area, and (3) the need to provide price discounts to customers of restaurants there. These responses have come to be known as “aversion effects” because people will respond to inducements (Rose et al., 2017b). Based on behavioral scenario–based experiments, the inducements were modeled by modifications in elasticities of demand in a CGE model,

54  Handbook on the economics of disasters which was then used to analyze the further, and more typical,7 general equilibrium effects. Essentially, the increased cost of doing business stifled economic activity both directly and indirectly. The results indicated that these three behavioral effect categories resulted in 15 times the GDP reduction stemming from ordinary brick-and-mortar plus quarantine impacts associated with the temporary closures for repair and decontamination of the area. Further distinctions between avoidance and aversion effects are explained in the context of a disease epidemic by Rose et al. (2017a) and Prager et al. (2017).

3.  BEHAVIORAL IMPACTS AND RISK Of course, aspects of behavior are ubiquitous in ordinary economic activity, as well as in disasters. Here we focus on a subset of behavior that manifests itself in an increase in the costs of disasters, typically with regard to their effect on the flow of goods and services (BI) and typically measured in terms of GDP and sales revenue, and sometimes in terms of economic welfare measures (Rose, 2004). Most of our focus will be on behavioral reactions stemming from fear brought on by the disaster, over and above ordinary fears or beyond pre-disaster risk aversion. We acknowledge that there are some counterpart behavioral responses that can reduce BI losses, such as acts of heroism and merely putting aside fears to resume normal activity, as was demonstrated, for example, in the aftermath of 9/11, through acts of patriotism to show that terrorists had not dampened the American spirit. Another example was emphasized in one of the first disaster impact studies in terms of mutual aid among communities (Dacy & Kunreuther, 1969).8 We begin with a taxonomy of general behavioral responses of the following three types: 1. Standard Behavioral Shifts Due to Changes in Preferences resulting from such factors as changes in tastes or product quality, which do not include elements of risk. 2. Standard Behavioral Shifts Due to Changes in Risk, but with tempered risk perceptions, of which there are two subtypes: a. Characteristics of products or economic activities becoming riskier due to such perceived existing factors as contamination or impending building collapse. b. Increase in probability of potential harm from disasters (e.g., fear of immediate terrorist attacks following an initial event, earthquake aftershocks, a new tornado warning, or longer-term repetitive flooding).9 3. Nonstandard Behavioral Shifts Due to Misperceptions of Risk, as when fear enters the picture as a result of extreme shock, intense media attention or rumor (social amplification of risk), or stigma effects. 3.1  Individual Behavior In the exposition to follow, we focus on the latter two types of behavior (standard behavioral shifts due to changes in risk and nonstandard behavioral shifts due to misperceptions of risk) because they are the ones most frequently associated with disasters. Some of the behavioral linkages are rational responses, while others are not. We attempt to delineate these two categories conceptually but acknowledge the difficulty of doing so empirically (e.g., distinguishing tempered [rational] risk perceptions from “exaggerated” ones). As illustrated by the case

Behavioral economic consequences of disasters  55 studies mentioned in this chapter, these behavioral responses typically incur additional costs, often quite large in magnitude. We offer a brief summary of some subtleties relating to aversion behavior following disasters. The term relates to the widely used concept of risk aversion. One of the first aspects to note is that this concept makes distinctions between three major categories of attitudes toward risk: “risk aversion” in a narrow sense, “risk neutrality,” and “risk seeking.” We can exclude the latter in the case of disasters, except for the few who engage in thrill-seeking as tornado chasers or who want to witness a hurricane firsthand.10 We emphasize that ordinary (pre-disaster) risk aversion is not aberrant behavior; it can in fact be a prudent reaction. The subcategory to which we refer pertains to extreme reactions. In such cases, the affected person might exaggerate the probability of a disaster, for example. However, we are focusing on disasters having actually occurred, so the person affected should be well grounded in terms of this consideration, and, therefore, the major behavioral response relates to reactive behavior affecting economic consequences. Gertz et al. (2019) refer to the aforementioned reactions as “inappropriate aversive behavior.” We focus on cases where fear11 enters the picture and exacerbates risk responses.12 In addition to individual overreactions to disaster risk, exacerbation is often caused by a phenomenon known as the “social amplification of risk,” where fear feeds on itself through word of mouth or media attention (see, e.g., Kasperson et al., 1988; Pidgeon et al., 2003). This has been measured effectively by various techniques (see, e.g., Rosoff et al., 2012; Zhao et al., 2019). Experiments based on behavioral scenarios using realistic examples of media coverage by Burns and Slovic (2010) have been integrated into ECA studies such as those of Giesecke et al. (2012) and Rose et al. (2017a).13 Attitudes toward risk are affected by the disaster itself in nuanced ways according to the type of impact. Shupp et al. (2017a) found that individuals who lost someone close to them in the aftermath of the tornado were in fact bolstered by the experience, which led to a decrease in risk aversion. Ambiguity aversion rose for individuals who suffered property damage, a factor that could affect longer-term reconstruction decisions. Shupp et al. (2017b) also found that disaster experience influences behavioral factors such as patience and trust. Beine et al. (2020) found increased risk aversion and impatience following an earthquake, which influenced migration decisions. They noted that these effects were in fact cumulative with regard to aftershocks. In terms of economic modeling and measurement in the context of disasters, the practice has been to identify a shift in a key behavioral parameter. Examples in relation to aversion are changes in product demand or in product and factor supply. These are typically reflected in changes in price elasticities of demand and supply for the case of aversion, and other types of parametric shifts in the case of avoidance behavior, which is not price responsive (see Section 6 for more details). We note that aversion behavior differs by type of disaster in terms of not only its duration but also its rate of onset. A good example noted by Gertz et al. (2019) is the comparable outcome between gradual sea-level rise and occasional riverine flooding. The former is a slow onset event that can go relatively unnoticed, while the latter can induce significant fear and the negative response, as in fleeing a potentially affected area prematurely.14 Note also that behavioral factors affect recovery from a disaster in terms of its time-path and duration, and thus can have a significant effect on BI as well. For example, behavioral factors affect the willingness of people to return to the location of a disaster rather than to resettle elsewhere, as in the example of the hesitancy of people to return to Fukushima (Nagamatsu et al., 2020). Behavioral factors also affect the pace and form of investment in recovery; for

56  Handbook on the economics of disasters example, funds for repair and reconstruction may be diverted somewhat to mitigating the future hazard if there is strong fear of its reoccurrence. These dynamic effects, however, are beyond the scope of this chapter, and the reader is referred to Xie et al. (2018) for a standard CGE analysis, and to Gertz et al. (2019), who develop an “anticipatory” CGE model, which has a relatively greater ability to incorporate them.15 3.2  Government Behavior Mandatory restrictions on economic activity during disasters typically have two behavioral elements: government decision-making and individual compliance. In this chapter, we focus on government decision-making and assume individual compliance is given (exogenous to the analysis). Of course, there are significant examples of noncompliance, but these are relatively minor for business closures, as in the case of COVID-19. However, stay-at-home orders are much more difficult to enforce, and noncompliance has probably been significant, though no definitive studies have been performed to date (Dash & Gladwin, 2007). In a related area of evacuations in the face of disasters, compliance problems arise as well. There are debates over whether force should be used and, if so, to what extent. Government involvement stems from the principle that one of government’s duties is to protect its citizenry. However, the counterargument is that government mandates in general, and some relating to disasters in particular, infringe upon civil liberties (Fairchild et al., 2006). The debate is strongly influenced by the features of the disaster. In the case of a pandemic, it is considered in the general public interest to close businesses due to contagion effects (actions by one individual entity having a significant effect on others), but in evacuation, it is more individual-based, though reference is often made to burdens on first responders who may have to risk their lives saving those who decide to shelter in place.16 Government decisions, of course, are not perfect, and they are subject to behavioral complications described elsewhere in this chapter because they are based on individual behavior. Government officials have their own perception issues and risk attitudes, as well as political considerations of staying in office, which affect the decision to mandate business or individual behavior. Depending on the circumstances, this can lead to delayed responses on closures (e.g., the mayor’s decision in the movie Jaws not to close the beach in order to maintain tourist revenues) and to overreactions (e.g., evacuating people prematurely under conditions of great uncertainty about the path of a hurricane or shutting down more businesses than are needed to minimize the economic impact no matter what the cost).17 Outright political considerations can often come into play as well, as in the example discussed earlier of countries using the threat of an FMD epidemic to ban imports from other countries, with the prime motivation being to help their domestic industry (Oladosu et al., 2013). More broadly, there is extensive literature on “government failure,” tracing back to research on bureaucracies by Niskanen (1971) and still being clarified and extended generally (Orbach, 2013) and in relation to disasters (Straub, 2020). Another critical factor is that managing disasters requires certain critical competencies and institutional capacities, which, when sorely lacking, as in the case of Hurricane Katrina, can greatly exacerbate disaster losses (Kapucu & van Wart, 2008; Waugh, 2006), especially when added to the behavioral considerations discussed earlier. As noted earlier, the intent of this chapter is not to provide a definitive way to conceptually or empirically sort out all of these factors. Moreover, for estimating bottom-line measure of economic consequences, it is not

Behavioral economic consequences of disasters  57 necessary to do so. Analytically, it helps to identify all causal factors, but often they reveal themselves after the fact, as in the case of the decline in airline travel and related tourism in the aftermath of 9/11 or the various types of losses following Hurricane Katrina. Alternatively, scenario or sensitivity analyses can be applied to cover the range of uncertainties about the relative influence of causal factors (Rose et al., 2017b). Overall, government behavior is not limited to its decrees but also applies to its information flows, which then encompass individual behavior in the case of both decrees and voluntary compliance. De Vericourt et al. (2021) employed an information design framework (Kamenica & Gentzkow, 2011) to analyze government strategic behavior in this regard in the context of COVID-19 pandemic confinement measures. They concluded that government strategy with regard to exaggerating or downplaying the severity of the situation was dependent on the relative preferences between economic activity and population health.18

4.  THE SCOPE OF BEHAVIORAL CONSEQUENCES The key issue that arises in the conceptualization estimation of behavioral consequences of disasters is their scope or extent. Previously, we mentioned examples relating to political reactions that range from the reasonable to the unreasonable, such as banning the importation of potentially contaminated meat products to spending trillions of dollars on foreign wars, respectively. Essentially the issue boils down to the following: • Which people are affected • Which areas are affected • What time period is affected19 The first relates to the age-old issue in BCA of who has standing, or whose welfare should count. Questions about the affected population overlap somewhat with the issue of the geographic area, but a simple example would be whether a policy affecting a national forest should only include its residents, actual visitors, or potential visitors (option demand). The more prevalent issue these days is, however, with regard to various population groups in a societal or socioeconomic context, for example, income or racial/ethnic groups. The increasing emphasis on equity and environmental justice are relevant in a couple of ways. The first is with regard to the value of a statistical life and the debate over whether the average value should be adjusted downward for the elderly (nonworking) population (see, e.g., Viscusi, 2020). With regard to equity/justice considerations, it has crystallized into issues of whether or not certain population groups should receive greater weights in the overall calculation. The US Department of Homeland Security (DHS) has typically encouraged analyses at the national rather than local or regional levels (Rose et al., 2017b). This issue has arisen more recently in relation to BCAs of interdiction technologies or policies relating to migrants crossing US borders (Farrow, 2020). Generally, except for cross-border issues, this is a straightforward decision and should be based on the population of greatest concern and in relation to the technology or policy under examination. As to the length of time that should be considered, one of the thorniest issues in the disaster literature is whether such events have long-run economic effects (see, e.g., Albala-Bertrand, 1993; Skidmore & Toya, 2002). We note that the focus of ECA is not typically on property

58  Handbook on the economics of disasters damage but on the flow of goods and services that emanate from the capital assets. On the other hand, BCA is usually applied to either property damage or BI, but typically not both.20 BI is typically conceived of as beginning when the disaster strikes and continuing until the relevant geographic area is recovered, or has achieved what is now termed the “new normal.” The latter is a way of truncating the losses from going on forever. Prime cases are Kobe, Japan, after the 1995 Great Hanshin Earthquake, where the port activity in that major city still has not recovered after 25 years, and the case of New Orleans after Hurricane Katrina, where most analysts suggest that a city two-thirds of its previous size is the more sustainable population level going forward. Probably the most controversial area of measuring behavioral effects relates to the political arena. Here, the matter is not initially one of necessity but one of choice (though we should keep in mind that this is also a distinction between mandatory avoidance and voluntary avoidance and voluntary aversion, which does not undercut the relevance of the latter two). One useful cutoff point here is whether the policy is related to the recovery of the original disaster or intended to mitigate or interdict future ones. That line of demarcation is typically applied, so that one would conclude that the second Iraq War was not related to the recovery of New York City or Washington DC, but rather to avoid the next terrorist attack or for other purposes entirely. A related issue is best exemplified by the concept of post-traumatic stress disorder (PTSD). This is a real and reasonably well-measured consequence of many major disasters, but it is typically not included in loss estimates, likely because it has not been traditional to do so or there is some skepticism, probably unwarranted by now, about the accuracy of such estimates. PTSD represents a type of behavioral response, but the economics of this issue are so complicated by psychology and health considerations that it is deemed beyond the scope of this chapter. All the issues just discussed are pertinent to both BCA and ECA, though the scope of the latter is likely to be broader than that of the former in general and with regard to behavioral responses. This is likely, to a great extent, due to the lack of formal rules of application in ECA, though BCA guidelines are not definitive on some of these issues either. Finally, we note the issue of making the distinction of whether various coping decisions, especially those associated with government policy, are warranted or are overreactions, or belong to the general behavioral category at all. Because many governmental post-disaster policies are related to fear of the population and/or the policy maker, we deem them relevant to the discussions in this chapter. In the strict sense, however, some are likely to maintain the position that only the overreactions should be considered in the behavioral category more narrowly defined.

5.  CATEGORIZATION OF BEHAVIORAL EFFECTS In this section, we offer a categorization of behavioral consequences of disasters in terms of basic economic elements at the partial and general equilibrium levels. We continue the distinctions made in previous sections, but present them in an organizing framework, as exhibited in Table 4.1. Note that the entries in the table represent the vast majority of cases for each of the types of avoidance and aversion responses; in some cases, there are notable exceptions that will be discussed as well.

Behavioral economic consequences of disasters  59 At the outset, we summarized a major distinction mentioned in passing above, but it bears specificity at the outset here: • Avoidance: Outright cessation of an activity, where any market inducements will not change the behavior (not price-sensitive) • Aversion: Inclination to reduce an activity, but where the reduction can be moderated by inducements (price-sensitive) The first impression of the entries in the table is that they are all very similar. This is primarily because the essences of the responses are related to the most basic elements of economics— supply and demand at both the partial and general equilibrium levels. This characterization is also helpful in linking the analysis to formal BCA measures, which focus on these basic elements as well, as will be done in Section 6. In the following summary, we also mention some subtle distinctions among the responses and their contexts.21 The simplest responses listed in Table 4.1 are the cases of mandatory avoidance, typically by government decree. In the case of COVID-19, this involves quarantines or stay-at-home orders of various sorts for the general population. It could also apply to a contaminated area hit by a chemical/biological/radiological/nuclear (CBRN) attack with respect to the physical area affected and/or population residing there. The counterpart under voluntary avoidance is people halting their shopping at establishments they deem unsafe. Both of these are represented by a truncation of demand at the partial equilibrium level, with the demand function shifting to a vertical line at zero demand. Similarly, there is an analogous depiction on the production side for the mandatory closures of businesses, represented by a truncation of supply, and at the extreme going to zero for affected enterprises.22 The counterpart in the voluntary avoidance partition would be employees staying away from the workplaces they consider unsafe, which would represent a full truncation of supply if complete, or a partial truncation if not, but in this case in terms of the supply of labor (factor market) rather than supply of goods and services (product market). Note also that mandatory closures dominate voluntary ones, in that it does not matter whether workers are inclined to stay away from their place of employment if those businesses are closed. Focusing further on the partial equilibrium level, public transit is basically the same in both cases, though the truncation is likely greater in the mandatory case because it does not allow for any leeway on the part of the customer. For voluntary avoidance, in this case and for other responses, the truncations may not be as great because there may be some threshold of risk below which people may engage in an activity, as in the willingness to risk some level of contamination or to place oneself in harm’s way in general. As to investment, the government may decree that certain geographic areas are off-limits for future rebuilding, as in a toxic waste spill (or buildup), CBRN release, or forbidding building in floodplains. This may be either temporary or permanent. Nevertheless, it essentially restricts the supply of investment opportunities. On the other hand, voluntary investment avoidance is a demand-side phenomenon and may be partial because some investors are scared off and others are not. As to mandatory avoidance of imports and exports, refer back to the example of banning the sale of meat products following the confirmation of an animal disease. These are characterized as demand truncations in the importing country banning the sale but a demand shift to the

60  Handbook on the economics of disasters Table 4.1  Categorization of behavioral consequences of disasters Behavior Type

Partial Equilibriuma

General Equilibrium

Adjustments (Resilience)

Literature

Avoidance (Mandatory) Quarantine

Demand truncation

Upstream supply chain

E-commerce; pent-up demand

Walmsley et al. (2021b)

Business

Supply truncation

Upstream and downstream

Telework; recapture

Walmsley et al. (2021b)

Public Transit

Supply truncation

Upstream supply chain

Other modes

Rose et al. (2017a)

Investment

Supply truncation

Upstream supply chain

Substitution; pent-up demand

Greenstone & Gallagher (2008)

Demand truncation

Upstream supply chain

Supply substitution

Oladosu et al. (2013)

Demand truncation

Upstream and downstream

Substitution; recapture

Whitehead (2003)

Import/Export

Evacuation

Avoidance (Voluntary) Shopping

Demand truncation

Upstream supply chain

E-commerce; pent-up demand

Walmsley et al. (2021c)

Workplace

Supply truncation

Upstream and downstream

Telework

Walmsley et al. (2021a)

Public Transit

Demand truncation

Upstream supply chain

Other modes

Rose et al. (2009, 2017a)

Investment

Demand truncation

Upstream and downstream

Relocation; dynamics

Dixon & Rimmer (2020)

Import/Export

Demand truncation

Upstream supply chain

Supply substitution

Oladosu et al. (2013)

Shopping

Demand shift

Upstream supply chain

E-commerce

Giesecke et al. (2012)

Workplace

Supply shift

Upstream and downstream

Telework; recapture

Giesecke et al. (2012)

Public Transit

Demand shift

Upstream supply chain

Other modes

Rose et al. (2009)

Investment

Demand shift

Upstream and downstream

Relocation; dynamics

Giesecke et al. (2012)

Import/Export

Demand shift

Upstream supply chain

Production recapture

Dixon & Rimmer (2020)

Aversion

a

See text for further explanation of entries.

Behavioral economic consequences of disasters  61 exporting country when this is not the ubiquitous response in foreign markets. Likewise, it is a supply truncation if it applies to all domestic beef sales but a supply shift otherwise. This is another case where one aspect of the supply-demand scissors trumps the other. Evacuation is actually very similar to quarantine in terms of placing a geographic area offlimits. The subtle difference is that it involves a geographic movement out of the area, which allows people to likely engage in more economic activity than in the quarantine case, though they do so elsewhere. A major distinction appears in the third partition of Table 4.1 with respect to demand and supply. As emphasized earlier, aversion behavior is not absolute, but it is price sensitive. Hence, supply and demand are not likely to be truncated as in the previous two categories. However, in the cases of supply and demand shifts associated with aversion, they also would not likely result in completely vertical supply and demand curves, but rather more standard, and possibly just parallel, shifts downward of either or both curves.23 Otherwise, each of the response types in the partial equilibrium column are basically the same between aversion and voluntary avoidance in terms of being related to demand or to supply. General equilibrium responses are best thought of in terms of supply-chain reactions in markets related indirectly to the partial equilibrium responses. The major distinction here is whether the supply-chain effect is upstream, downstream, or both. In the context of business operations, both upstream (relating to the demand for inputs) and downstream (relating to the supply of outputs) effects are operative, except for consumer goods, investment goods, and exported goods, which have no supply-side linkages because they do not stimulate, positively or negatively, any further production in the geographic area in which the impacts are being measured.24 In the table, we specify the relationship according to the majority of likely cases. For example, public transit is characterized as a consumer good, though we know that some aspects relate to business activity. Evacuation falls primarily on the affected population but, as an economic effect, is not limited to consumption and also refers to business closures. The adjustments column pertains primarily to partial equilibrium responses in terms of additional ramifications. Some of them could in fact be part of the general equilibrium responses as well, however. For example, telework is a way of contributing to economic production activity even when people shy away from the workplace. It represents a type of economic resilience—a strategy intended to use available resources as efficiently as possible to maintain function (Rose, 2004).25 Rose (2009b) has related resilience to the economic production function for such responses, primarily to the disruption of inputs, in terms of standard economic adjustments such as input substitution (e.g., goods and location), productivity changes (e.g., conservation and technological change), excess capacity, and inventories. These resilience tactics have been successfully estimated empirically for both hypothetical events (see, e.g., Kajitani & Tatano, 2009; Wei et al., 2020) and actual events (see, e.g., Dormady et al., 2022; Rose, 2009) in both partial and general equilibrium contexts. In addition to the telecommuting resilience example, others include consumers resorting to e-commerce or shifting their purchases over time in terms of pent-up demand (Walmsley et al., 2021a). In the case of public transportation, this involves shifts to other transportation modes (Rose et al., 2009, 2017a).26 In terms of investment, this refers to relocation and also has a dynamic element if the investment is delayed. In terms of evacuation, it involves the shift (substitution) of spending to other locations. Production recapture is a unique resilience tactic, and the only one that refers to the output, rather than input, side of the production function. It relates to the ability to make up for lost

62  Handbook on the economics of disasters production by working overtime or extra shifts once productive capacity is restored (Park et al., 2011). Various studies have indicated it has sizable potential—as high as 90%+—in manufacturing and mining industries and much lower, but still significant, potential in utilities, services, and agriculture (Dormady et al., 2022; Wei et al., 2020). Note also that production recapture is the supply-side counterpart of the demand-side pent-up demand. Overall, the adjustments noted typically reduce the negative impact of behavioral responses on economic consequences. There is every reason to venture, moreover, that their offsetting effects are as powerful for behavioral responses as they are for the same level of consequences caused by other types of disaster shocks.27

6. IMPLICATIONS FOR ECONOMIC WELFARE AND BENEFIT-COST ANALYSIS 6.1  Welfare Analysis Summary Much of the previous analysis relates to ECA because that is the area in which the author has worked most. However, the presentation does have strong bearing on BCA in terms of the potential inclusion of some subsets of behavioral responses that have been discussed. The intent is to help establish guidelines for their inclusion in both BCA and ECA, though those guidelines might differ between the two approaches. In this section, we first summarize a standard economic welfare analysis of the avoidance and aversion behavior. The workings of the three types of behavioral shifts noted in Section 4 as an underpinning for these phenomena could be analyzed using standard consumer theory (or production theory for business cases) and expected utility theory. However, this step can be skipped to proceed to the more straightforward result of these behavioral shifts in terms of changes in supply and demand because these are the more operational aspects of the analysis in terms of the actual measurement of behavioral linkages.28 This is the approach taken by the author (Rose, 2021b), who performs the micro welfare analysis according to the three categories of behavioral responses presented earlier. The demand and supply cases and shifts summarized in Table 4.1 serve as the basis for the analysis, which is able to delineate producer and consumer surplus losses and transfers for each of the cases. Some general results are derived. First, behavioral responses are likely to result in greater losses of surpluses in the case of mandatory closures than in the other two cases because of the more dramatic truncations in the former, ceteris paribus. Second, and in contrast to the other cases, Aversion behavior is more likely to have two effects: a downward shift in the demand curve, and then a movement along it but only up to a limit and then an ambiguous equilibrium. Note also that we have simplified the analysis for sake of exposition, as the same event could trigger voluntary avoidance behavior to some of those affected and aversion behavior to others (see, e.g., Giesecke et al., 2012). Moreover, the general equilibrium effects (indirect or supply chain) of behavioral responses can also be measured in the standard manner as performed in non-disaster contexts, though there may be limits as to the extent that they can be legitimately included in BCA (Farrow & Rose, 2018; US EPA, 2017). We know at the outset of this section that the legitimacy of this inclusion is often broader than intimated in standard BCA texts. For example, Boardman et al. (2018) advise against the inclusion of indirect effects because they do not account for the siphoning off of

Behavioral economic consequences of disasters  63 resources from existing activities. Only later in the exposition, and appearing as an afterthought, is it mentioned that this applies only to the case of full employment. We emphasize the asymmetry, however, of positive and negative economic impacts in the context of disasters. Most standard BCA texts focus on project evaluation or policy analysis, typically with expansionary ramifications, where the existence of full employment or the possibility that the project or policy will exceed the available labor (or capital) supply are serious concerns. Disasters are just the opposite, where the impacts of a negative effect of a decline in economic activity doesn’t usually run up against a labor supply ceiling, unless there are an extensive number of deaths/injuries or when mandatory avoidance is invoked. Thus, the standard caveat does not usually apply. 6.2  Behavioral Responses That Can Be Legitimately Included in ECA and BCA Practically all the examples of behavioral responses noted in this chapter have been included in various economic consequence analyses. Two areas of controversy in ECA stem from cases where impacts are related to political considerations or extend over what some would consider a too-lengthy time horizon. In both cases, there are likely to be concerns about the sufficiency of the causal links. With regard to whether behavioral economic consequences fall within the scope of BCA, we offer the following summary:29 1. Whose welfare should count? For standard applications of BCA, such as in the evaluation of projects and policies pertaining to reducing risk, all of those within the relevant geographic area who have standing and are affected directly and indirectly by behavioral economic consequences should, and typically are, included.30 2. How long of a time period should be considered? Most behavioral consequences are short-term reactions to a disaster. However, some truly catastrophic events can leave a lasting imprint on society, and hence it is reasonable to consider as long a time horizon as is necessary (though with appropriate time discounting). CBRN disasters have lasting effects, and, in behavioral analysis, it is not actual decontamination that counts but rather people’s perception of the situation and how they act on it. This overlaps with the fact that many such events are characterized by “stigma” effects that can last for decades (Schulze & Wansink, 2012). Other long-lasting implications can stem from such outcomes as PTSD, which can last for many years and have various types of additional costs associated with changes in behavior over and above the health effects.31 3. What geographic area should be considered? It is standard DHS policy, as well as that of many other federal government agencies, to perform analyses of projects and policies at the national level. This approach may be even more appropriate in terms of behavioral effects, which readily spill over the boundaries of the directly affected area of the disaster, given today’s extensive nationwide and worldwide media coverage. We offer the following conclusions regarding other key questions: 1. To what extent should, and can, behavioral effects that spill over to other markets be taken into account? These include concerns about displacement effects and double-counting of impacts (see, e.g., Farrow, 2020). Another set of indirect effects pertains to substitutions that arise from the behavioral shift. In many cases, this amounts to a “coping” response

64  Handbook on the economics of disasters that has come to be listed under the heading of resilience, for which numerous tactics have been identified and whose effects can be estimated using standard models. General equilibrium effects of behavioral shifts are fairly standard and should and can be estimated. 2. What is the boundary of what should be included as a behavioral effect? This issue pertains to evaluating the strength of causation between the original disaster and the behavioral effect in question. Unfortunately, there is no straightforward rule to apply regarding causal strength, nor is there likely to be one that receives universal consensus, especially when the additional activity is related to political considerations. However, it is reasonable to exclude behavior that is not a reaction to the original disaster events, but rather intended to prevent further disasters. In any case, when in doubt, analysts would be wise to perform sensitivity analyses to compare bottom-line losses with and without the follow-on activity. It is our contention that, aside from a few aforementioned reservations, which typically arise in only a minority of cases, the vast majority of nonstandard behavioral responses to disasters can be legitimately included in ECA and BCA. Empirical analyses and simulations have indicated that behavioral responses can increase BI losses by an order of magnitude or more over and above more limited estimates of the effects of property damage, loss of life, and infrastructure services. Hence, it is extremely important to be able to evaluate the ability of new products and policies that can reduce exacerbating behavioral responses.32 We offer a similar conclusion regarding performing BCA on behavioral impacts or countermeasures to reduce them, including resilience tactics. Again, these would simply be analyzed in relation to shifts or movements along supply and demand curves. We reiterate the difficulty in many cases of separating normal behavior and modified behavior in response to disasters. Such decompositions are useful, especially for policy-making, but they are not necessary if one is interested only in the bottom line of economic consequences of disasters or the benefits of reducing them.

7. CONCLUSION This chapter has intended to provide a formal foundation for the analysis and empirical measurement of this category of economic consequences. The major step toward their legitimate inclusion is that various behavioral responses can be translated into straightforward demand and supply reactions. This helps make them understandable and measurable. It is especially critical to the calculation of welfare measures in BCA, in contrast to ECA for which macroeconomic indicators, such as GDP and employment, still suffice and are not dependent on the explicit measurement of some demand and supply considerations. The reader is reminded of the importance of behavioral considerations in the overall estimation of economic consequences of natural disasters, technological accidents, and man-made disasters such as terrorism. Studies have shown that behavioral effects can be one or two orders of magnitude larger than the standard economic impacts typically calculated.33 At the same time, we have cited studies that indicate resilience strategies to prevent some BI caused by behavioral linkages can have a significant offsetting effect and hence should be included in the calculation, for which the net effect can still remain very large.

Behavioral economic consequences of disasters  65 Note that we have not always explicitly distinguished ordinary behavior and behavior unique to disasters in the course of this chapter. In some cases, this is not an important distinction, as many methodological approaches would simply use a shift in demand and supply, where the shifts might represent the totality of behavioral responses. Normal responses are likely to be built into most models, for example, in terms of ordinary substitutions among inputs by producers or goods/services by households or in absolute changes in economic activity in either one. If the behavioral needs of the modeling are focused on exaggerated behavior, then the behavioral shift can be incorporated as parametric changes reflecting limits on activity or shifts in supply and demand. This can serve as the basis for separating the consequences of unordinary behavioral responses from the ordinary ones. Of course, the difference between ordinary behavior and exaggerated behavior is more important in the case of policy analysis because the remedies to steer these two kinds of behavior in a positive direction are likely to differ significantly. In that context, it would be desirable to decompose these two sets of responses to gauge their relative influence and formulate policies according to priorities and potentials. There are likely many products and policies worthy of assessment that have the ability to dampen negative behavioral responses that exacerbate losses from disasters, especially when the countermeasures would appear to be so relatively inexpensive. Examples include the verified influence of risk messaging to alter behavior about fear of airline travel or information campaigns and financial inducements to increase vaccination “take-up” rates. In addition, more expensive options that assure thorough decontamination of toxic spill sites or the dispatch of security personnel to keep the peace are also likely to pass the benefit-cost test in most instances. We also acknowledge that we invoked a number of simplifying assumptions into our formal analysis. First, we have characterized behavioral responses effectively as “games against nature” or a nonadaptive adversary, in general, rather than explicitly analyzing strategic behavior with an adaptive adversary, including business-consumer interactions or either of these sets of entities in relation to government decisions. Second, we haven’t examined considerations of market power, nor have we analyzed the issue in the context of uncertainty. Third, we have not addressed any equity issues associated with behavioral responses, where there is an increasing call for the inclusion of this criterion in general into BCA. Loosening of these assumptions presents several areas of future research. However, the most important, and likely most challenging, will be the empirical measurement of behavioral responses, especially the distinction between ordinary and exaggerated ones.

NOTES * The author is senior research fellow, Center for Risk and Economic Analysis of Threats and Emergencies (CREATE), and research professor in the Price School of Public Policy, University of Southern California. I wish to acknowledge the helpful comments of Scott Farrow, Detlof von Winterfeldt, Richard John, Vicki Bier, Dan Wei, Chris Zobel, Gilberto Montibeller, and an anonymous reviewer. I also appreciate the helpful research assistance of Konstantinos Papaefthymiou. This research was supported by the Directorate of Science and Technology of the US Department of Homeland Security through the National Center for Risk and Economic Analysis of Threats and Emergencies (CREATE) under Task Order 70RSAT20FR0000046. However, any opinions, findings, conclusions, or recommendations in this document are those of the author and do not necessarily reflect views of the US Department of Homeland Security or the University of Southern California.

66  Handbook on the economics of disasters 1. Note that our focus is on behavioral reactions that typically exacerbate losses once the disaster strikes. There is another use of the term “avoidance” in the literature referring to protective or preventative behavior in advance of a disaster that typically refers to interdiction and mitigation, which reduce the frequency or magnitude of disasters (see, e.g., Boardman et al., 2018), but this is a far different area of inquiry and beyond the scope of this chapter. This approach is pertinent to reducing some behavioral responses, but has some significant limitations as will be discussed further later in this section. 2. Although ECA is typically applied to measuring the benefits of hazard reduction, it can also be applied to the cost side, including behavioral spillover effects of implementing countermeasures, such as patrons of sports stadiums and concert halls reducing their attendance due to frustrations about invasion of privacy, inconvenience, or delays (Rose et al., 2014, 2021). 3. Albala-Bertrand (1993) has noted that built-in societal mechanisms can offset some of these ripple effects, but this would just represent aspects of group or macroeconomic resilience (Rose, 2017). 4. We acknowledge the omission of several related considerations in our analysis. Our focus is on individual behavior, so we have excluded aspects of behavioral responses at the group level, such as mass panic (refer to the extensive literature in sociology and psychology on this topic). We have also abstracted from elements of time discounting and of individual values, except for individual risk attitudes. 5. We focus on “flow” measures such as GDP in this chapter. However, other types of disaster impacts should be acknowledged, such as effects on property values (see, e.g., Dormady et al., 2014, for the behavioral effects of an anthrax attack on real estate prices, and Smith et al., 2006, for the effects of a hurricane on real estate prices and migration). 6. Note that this proportion was greatly affected by the fact that approximately 72% of potential GDP losses were prevented by the rapid relocation of businesses and government agencies in the WTC shortly after the attack (Rose et al., 2009). This is an example of resilience, which also plays an important role in the estimation of disaster losses in ECA. For an exposition of the role of resilience in the ECA of disasters, refer to Rose (2009a, 2015) and Rose et al. (2017b). Note that the bottom-up analysis of the impacts of this terrorist attack focusing on resilience and behavioral linkages led to an estimate close to that determined by an econometric approach by Blomberg and Hess (2009) and two other related studies. 7. We say “typical” because most indirect effects of behavioral responses to disasters are not themselves the product of any special behavioral considerations but simply supply-chain effects. This is not to say that the behavior of one individual cannot affect others, as in actions inciting a mob; however, we chose to consider this group action to be a group direct effect. 8. Other examples of behavioral overreactions can have positive social benefits. For example, voluntary social distancing during the COVID pandemic beyond mandatory levels, as well as voluntary avoidance and aversion behavior in general, have reduced infection rates and provided a partially offsetting positive effect on GDP. 9. We do not delve further into measuring tempered risk perceptions versus exaggerated ones, except for the following observations: the price-responsiveness of aversion behavior provides a handle by which to measure the direct cost of behavioral changes. The results could be compared with science-based estimates of the probability and consequences of additional harm to distinguish category 2 and category 3. 10. Risk attitude is typically defined in microeconomics, and decision analysis is defined in terms of the shape of the utility function: concave is risk averse, linear is risk neutral, and convex is risk seeking. It specifies a transformation on an outcome scale that reflects preferences for gambles, assuming rational (maximization of expected utility) preferences. Risk attitude is operationalized as a parameter defining the concavity of the function, for example, risk tolerance. Note that emotional, behavioral, and cognitive responses to disasters can differ significantly across individual, socioeconomic, political orientation, and locational characteristics (McArdle et al., 2012). 11. We omit consideration of dread, which is often referred to as a feeling of great fear, and usually in an anticipatory fashion. Moreover, the term “dread” has more of a passive connotation in terms of internal feelings, while “fear” typically refers to an actionable emotion. The extreme actions of disaster victims are the focus of this chapter. 12. There is an extensive literature on “near misses” of catastrophic events that captures some behavioral responses closely related to those discussed in this chapter. The literature characterizes two types of such events, “resilient” and “vulnerable,” which have opposite impacts on those who experience them. Resilient (with the term used differently than elsewhere in this chapter) near misses are perceived as evidence that the threat is not so great, thereby reducing the perceived threat. In contrast, vulnerable near misses increase the perceived threat (see Dillon & Tinsley, 2008; Dillon et al., 2014; Tinsley et al., 2012). Cui et al. (2017) investigated differences in how people perceive near-miss events and developed a scale to measure “near-miss appraisal” as a psychological trait. They also found that this appraisal is as much a product of the perceptions of the person experiencing the near miss as it is of the specific details of the event. 13. Note that psychologists make the distinction between cognitive and affective (emotional) determinants of risk perception and avoidance behavior. Risk attitude is generally a cognitive component, while the aversion behavior discussed here is an affective, or emotional, component. Slovic (1987) and others have posited the affect heuristic, in which risk perception is based on emotions, such as fear or anger, rather than rational components, such as probability and magnitude of loss. This is sometimes referred to as “risk (perception) as feelings.”

Behavioral economic consequences of disasters  67 14. At the same time, both may lead to moral hazard in terms of building structures in low-lying coastal areas or building in a floodplain. 15. There is another important aspect of individual behavior that is beyond the scope of this chapter. This applies to strategic behavior among individuals, such as in social distancing during a pandemic or survivalist behavior in cases where disasters cause losses so great that they breach the subsistence level. Overall, the standard framework for strategic analysis in the case with most disasters is a game against nature (though feedback affects can complicate the analysis), while strategic behavior among individuals and groups opens up a much broader realm. 16. Individual responses to evacuation orders relate to characteristics of warnings, risk perceptions, and expected consequences (Dash & Gladwin, 2007). One factor often cited as leading to lack of compliance is the high percentage of false alarms, which can reach as up to 76% in the case of tornadoes (Barnes et al., 2007). 17. The cost of evacuation is lower than often specified (including the costs of housing and feeding the displaced population) as are the economic consequences (Whitehead, 2003). One aspect of resilience that is typically omitted in such studies is the “recapture factor” (Park et al., 2011; Rose, 2009), which refers to the potential to make up lost production in many sectors by working overtime or extra shifts once the population returns. 18. This is an interesting instance of influencing fear and quelling it (the latter ironically being done in an “exaggerated” way). This line of inquiry proceeds further down an interesting path by concluding, consistent with long-standing findings, that socioeconomic status influences individual responses to threats. 19. Related questions include “in what ways are people affected and to what extent,” but we subsume these into the measurement of consequences. 20. The reason is that the value of an asset is typically considered to be the net present value of future returns on it—essentially normal (plus economic) profits derived from it—which would be a portion of GDP; hence, also including measures of the flow of goods and services emanating from the asset would involve some double counting. One exception would be the justified inclusion of loss of production of goods and services while repair and reconstruction are taking place (Rose, 2015). 21. We note at the outset that we have made the distinction between avoidance and aversion solely on the basis of price sensitivity. This is not the only factor that can warrant distinctions. For example, avoidance can be reduced by “nudges” (Sunstein & Thaler, 2008) or risk communication, both of which are low cost and effective up to a point. However, further inquiry is beyond the scope of this chapter. 22. We do not go into details about how one translates information on behavioral responses into supply or demand changes. Refer to Giesecke et al. (2012) for a detailed account. Note also that this approach has been validated by Nassios and Giesecke (2018) in comparing the methodology to a macroeconometric analysis of the economic consequences of the 9/11 WTC attacks by Blomberg and Hess (2009). 23. Note that Giesecke et al. (2012) model these as vertical shifts in the demand curve, for example, because they relate to price, rather than the more typical horizontal demand shifts relating to non-price aspects (shifters) of demand functions, such as income, population, or regulations. 24. Investment does stimulate additional production, but in future time periods, rather than the one under study in a static modeling formulation. 25. Rose (2009) makes the distinction between static and dynamic economic resilience, where the latter refers to investment in repair and reconstruction to promote recovery. Resilience tactics differ somewhat in this case, in that they are geared primarily to actions that facilitate investment, such as outside assistance, cutting red tape, and so on (see Xie et al., 2018). 26. Note that in both of these studies, and some others noted in this section, both avoidance and aversion behavior and resilience were included. For example, in the 9/11 study, avoidance was simulated by an estimate of the decline in airline travel, but the workings of the computable general equilibrium model then included aspects of the transportation mode shift as an offsetting increase in overall transportation demand. 27. Another major example of reducing losses from disasters, especially pertinent to behavioral impacts, and basically parallel to the examples of resilience presented above, pertains to the area of risk communication (see, e.g., Rosoff et al., 2012). This has the ability to quell fears after the disaster, and even to anticipate fears through inoculation messaging prior to disasters (Ivanov et al., 2018). But see also Fischhoff et al. (2003) on perception problems people have in processing such communications. 28. Demand curves, for example, can be derived from a set of indifference curves in utility space, but such a derivation is not necessary for the exposition presented here. In addition, change in welfare is typically measured by equivalent variation or compensating variation and can be exhibited in utility space as well, but not as transparently as the supply/demand exposition discussed here. 29. See von Winterfeldt et al. (2020) and Farrow (2020) for examples of the state-of the-art BCA for evaluating DHS projects. 30. A noted exception to broad inclusion is to exclude preventative or remedial actions in BCA. 31. For an alternative view which suggests that the time horizon should end at the point at which the situation approaches a new equilibrium, see, for example, Bier & Nosek, 2017. 32. We emphasize losses in terms of BI, but acknowledge the prominent role of behavioral responses in exacerbating the losses to households as well. The latter are typically nonmarket impacts and often included in BCA, which

68  Handbook on the economics of disasters is especially well-suited to estimate impacts of this kind. They are not typically part of standard ECA metrics, though there is no reason they should not be included (see, e.g., Rose & Oladosu, 2009). 33. We confined our analysis to natural disasters, technical accidents, and terrorism, the behavioral economic impacts of war could readily be even much larger.

REFERENCES Albala-Bertrand, J. M. (1993). Natural disaster situations and growth: A macroeconomic model for sudden disaster impacts. World Development, 21(9), 1417–1434. https://doi.org/10.1016/0305-750X(93)90122-P Barnes, L. R., Gruntfest, E. C., Hayden, M. H., Schultz, D.M., & Benight, C. (2007). False alarms and close calls: A conceptual model of warning accuracy. Weather and Forecasting, 22(5), 1140–1147. https://doi.org/10.1175/ WAF1031.1 Beine, M., Charness, G., Dupuy, A., & Joxhe, M. (2020). Shaking things up: On the stability of risk and time preferences. CESifo Working Paper No. 8187. https://www.econstor.eu/bitstream/10419/216583/1/cesifo1_wp8187. pdf Bier, V., & Nosek, A. (2017). Mass relocations after nuclear disasters [Proceedings]. American Nuclear Society International Topical Meeting on Probabilistic Safety Assessment and Analysis, Pittsburgh, PA. Blomberg, S. B., & Hess, G. D. (2009). Estimating the macroeconomic consequence of 9/11. Peace Economics, Peace Science, and Public Policy, 15(2), 307–330. https://doi.org/10.2202/1554-8597.1167 Boardman, A., Greenberg, D., Vining, A., & Weimer, D. (2018). Cost-benefit analysis: Concepts and practice (5th ed.). Cambridge University Press. https://doi.org/10.1017/9781108235594 Burns, W., & Slovic, P. (2010). Predicting and modeling public response to a terrorist strike. In P. Slovic (Ed.), The feeling of risk: New perspectives on risk perception. Routledge. Byrd, K., & John, R. (2021). Survey of avoidance behavior during COVID-19. Center for Risk and Economic Analysis of Terrorism Events (CREATE), University of Southern California, Los Angeles, CA. Cui, J., Rosoff, H., & John, R. S. (2017). A polytomous item response theory model for measuring near-miss appraisal as a psychological trait. Decision Analysis, 14(2), 75–86. https://doi.org/10.1287/deca.2017.0345 Cutter, S. (2016). The landscape of disaster resilience indicators in the USA. Natural Hazards, 80, 741–758. https:// link.springer.com/article/10.1007%2Fs11069-015-1993-2 Dacy, D., & Kunreuther, H. (1969). The economics of natural disasters. Free Press. Dash, N., & Gladwin, H. (2007). Evacuation decision making and behavioral responses: Individual and household. Natural Hazards Review, 8(3), 69–77. https://doi.org/10.1061/(ASCE)1527-6988(2007)8:3(69) de Vericourt, F., Gurkan, H., & Wang, S. (2021). Informing the public about a pandemic. Management Science, electronic preprint. https://doi.org/10.1287/mnsc.2021.4016 Dillon, R. L., & Tinsley, C. H. (2008). How near-misses influence decision making under risk: A missed opportunity for learning. Management Science, 54(8), 1425–1440. https://pubsonline.informs.org/doi/abs/10.1287/ mnsc.1080.0869 Dillon, R. L., Tinsley, C. H., & Burns, W. J. (2014). Evolving risk perceptions about near-miss terrorist events. Decision Analysis, 11(1), 27–42. https://pubsonline.informs.org/doi/abs/10.1287/deca.2013.0286 Dixon, P., Jerie, M., Rimmer, M., & Wittwer, G. (2019). Rapid assessments of the economic implications of terrorism events using a regional CGE model: Creating GRAD-ECAT (generalized, regional and dynamic economic consequence analysis tool). In Y. Okuyama & A. Rose (Eds.), Advances in spatial and economic modeling of disaster impacts (pp. 121–161). Springer Nature. https://doi.org/10.1007/978-3-030-16237-5_6 Dixon, P., & Rimmer, M. (2010). Optimal tariffs: Should Australia cut automotive tariffs unilaterally? Economic Record, 86(273), 143–161. https://doi.org/10.1111/j.1475-4932.2009.00599.x Dormady, N., Rose, A., Roa-Henriquez, A., & Morin, B. (2022). The cost-effectiveness of economic resilience. Under review by The International Journal of Production Economics, 244, 108371. Dormady, N., Szelazek, T., & Rose, A. (2014). The potential impact of an anthrax attack on real estate prices and foreclosures in Seattle. Risk Analysis, 34(1), 187–201. https://doi.org/10.1111/risa.12059 Elbakidze, L., & McCarl, B. (2006). Animal disease pre-event preparedness versus post-event response: When is it economic to protect? Journal of Agricultural and Applied Economics, 38(2), 227–236. https://doi.org/10.1017/ S1074070800022367 Fairchild, A. L., Colgrove, J., & Jones, M. M. (2006). The challenge of mandatory evacuation: Providing for and deciding for. Health Affairs, 25(4), 958–967. https://doi.org/10.1377/hlthaff.25.4.958 Farrow, S., & Rose, A. (2018). Welfare analysis: Bridging the partial and general equilibrium divide for policy analysis. Journal of Benefit-Cost Analysis, 9(1), 67–83. https://doi.org/10.1017/bca.2017.29 Fischhoff, B., Gonzalez, R., Small, D., & Lerner, J. (2003). Evaluating the success of terror risk communications. Biosecurity and Bioterrorism: Biodefense Strategy, Practice, and Science, 1(4), 1–4. https://doi. org/10.1089/153871303771861450

Behavioral economic consequences of disasters  69 Gertz, A., Davies, J., & Black, S. (2019). A CGE framework for modeling the economics of flooding and recovery in a major urban area. Risk Analysis, 39(6), 1314–1341. https://onlinelibrary.wiley.com/doi/abs/10.1111/risa.13285 Giesecke, J., Burns, W., Barrett, A., Bayrak, E., Rose, A., Slovic, P., & Suher, M. (2012). Assessment of the regional economic impacts of catastrophic events: A CGE analysis of resource loss and behavioral effects of a radiological dispersion device attack scenario. Risk Analysis, 32(4), 583–600. https://doi.org/10.1111/j.1539-6924.2010.01567.x Gordon, P., Moore, J., Park, J., & Richardson, H. (2007). The economic impacts of a terrorist attack on the US commercial aviation system. Risk Analysis, 27(3), 505–412. https://doi.org/10.1111/j.1539-6924.2007.00903.x Greenstone, M., & Gallagher, J. (2008). Does hazardous waste matter? Evidence from the housing market and the Superfund program. Quarterly Journal of Economics, 123(3), 951–1003. https://doi.org/10.1162/ qjec.2008.123.3.951 Ivanov, B., Sellnow, T., Getchell, M., & Burns, W. (2018). The potential for inoculation messages and post inoculation talk to minimize the social impact of politically motivated acts of violence. Journal of Contingencies and Crisis Management, 26(4), 414–424. https://onlinelibrary.wiley.com/doi/full/10.1111/1468-5973.12213 Kajitani, Y., & Tatano, H. (2009). Estimation of lifeline resilience factors based on empirical surveys of Japanese industries. Earthquake Spectra, 25(4), 755–776. https://journals.sagepub.com/doi/abs/10.1193/1.3240354 Kamenica, E., & Gentzkow, M. (2011). Bayesian persuasion. American Economic Review, 101(6), 2590–2615. https:// doi.org/10.1257/aer.101.6.2590 Kapucu, N., & Van Wart, M. (2008). Making matters worse: An anatomy of leadership failures in managing catastrophic events. Administration and Society, 40(7), 711–740. https://doi.org/10.1177/0095399708323143 Kasperson, R., Renn, O., Slovic, P., Brown, H., Emel, J., Goble, R., Kasperson, J., & Ratick, S. (1988). The social amplification of risk: A conceptual framework. Risk Analysis, 8, 177–187. https://onlinelibrary.wiley.com/doi/ abs/10.1111/j.1539-6924.1988.tb01168.x McArdle, S., Rosoff, H., & John, R. (2012). Identifying moderators of changing worries, concerns, and behavioral avoidance following 9/11. Risk Analysis, 32(4), 744–761. https://doi.org/10.1111/j.1539-6924.2012.01814.x Nagamatsu, S., Rose, A., & Eyer, J. (2020). Return migration and decontamination after the 2011 Fukushima nuclear power plant accidents. Risk Analysis, 40(4), 800–817. https://doi.org/10.1111/risa.13432 Nassios, J., & Giesecke, J. (2018). Informing ex ante event studies with macro-econometric evidence on the structural and policy impacts of terrorism. Risk Analysis, 38(4), 804–825. https://doi.org/10.1111/risa.12874 Niskanen, W. (1971). Bureaucracy and representative government. Aldine. https://www.taylorfrancis.com/books/ mono/10.4324/9781315081878/bureaucracy-representative-government-william-niskanen Oladosu, G., Rose, A. & Lee, B. (2013). Economic impacts of potential foot and mouth disease agro-terrorism in the United States: A computable general equilibrium analysis. Journal of Biothreats and Biodefense, S12, 001. https:// doi.org/10.4172/2157-2526.S12-001 Orbach, B. (2013). What is government failure? Yale Journal on Regulation Online, 30, 44–56. https://www.yalejreg. com/bulletin/what-is-government-failure/ Park, J., Cho, J., & Rose, A. (2011). Modeling a major source of economic resilience to disasters: Recapturing lost production. Natural Hazards, 58(1), 163–182. https://doi.org/10.1007/s11069-010-9656-9 Pidgeon, N., Kasperson, R., & Slovic, P. (2003). The social amplification of risk. Cambridge University Press. https:// doi.org/10.1017/CBO9780511550461 Prager, F., Wei, D., & Rose, A. (2017). Total economic consequences of an influenza outbreak in the United States. Risk Analysis, 37(1), 4–19. https://doi.org/10.1111/risa.12625 Rose, A. (2004). Economic principles, issues, and research priorities in natural hazard loss estimation. In Y. Okuyama & S. Chang (Eds.), Modeling the spatial economic impacts of natural hazards (pp. 13–36). Springer. https://doi. org/10.1007/978-3-540-24787-6_2 Rose, A. (2007). Economic resilience to disasters: Multidisciplinary origins and contextual dimensions. Environmental Hazards: Human and Social Dimensions, 7(4), 383–398. https://doi.org/10.1016/j.envhaz.2007.10.001 Rose, A. (2009a). A framework for analyzing and estimating the total economic impacts of a terrorist attack and natural disaster. Journal of Homeland Security and Emergency Management, 6(1), Article 9. doi. org/10.2202/1547-7355.1399 Rose, A. (2009b). Economic resilience to disasters [Report no. 8]. Community & Regional Resilience Institute, Oak Ridge National Laboratory, Oak Ridge, TN. https://s31207.pcdn.co/wp-content/uploads/2019/09/EconomicResilience-to-Disasters.pdf Rose, A. (2015). Macroeconomic consequences of terrorist attacks: Estimation for the analysis of policies and rules. In C. Mansfield & V. K. Smith (Eds.), Benefit transfer for the analysis of DHS policies and rules (pp. 172–200). Edward Elgar. https://doi.org/10.4337/9781784711085.00016 Rose, A. (2017). Benefit-cost analysis of economic resilience actions. In S. Cutter (Ed.), Oxford research encyclopedia of natural hazard science (pp.  1–32). Oxford University Press. https://doi.org/10.1093/ acrefore/9780199389407.013.69 Rose, A. (2021). Behavioral Economic Consequences of Disasters: A Basis for Inclusion in Benefit-Cost Analysis. Working Paper, Center for Risk and Economic Analysis of Threats and Emergencies (CREATE), University of California, Los Angeles, CA.

70  Handbook on the economics of disasters Rose, A., Avetisyan, M., & Chatterjee, S. (2014). A framework for analyzing the economic tradeoffs between urban commerce and security. Risk Analysis, 34(5), 1554–1579. https://doi.org/10.1016/j.tra.2015.04.027 Rose, A., Avetisyan, M., Rosoff, H., Burns, W., Slovic, P., & Chan, O. (2017a). The role of behavioral responses in the total economic consequences of terrorist attacks on US air travel targets. Risk Analysis, 37(7), 1403–1418. https:// doi.org/10.1111/risa.12727 Rose, A., & Oladosu, G. (2009). Regional economic impacts of natural and man-made hazards disrupting utility lifeline services to households. In H. Richardson, P. Gordon, & J. Moore (Eds.), Economic impacts of Hurricane Katrina (pp. 921–923). Edward Elgar. https://doi.org/10.1080/01900690903004056 Rose, A., Oladosu, G., Lee, B., & Beeler Asay, G. (2009). The economic impacts of the 2001 terrorist attacks on the World Trade Center: A computable general equilibrium analysis. Peace Economics, Peace Science, and Public Policy, 15(2), Article 6. https://doi.org/10.2202/1554-8597.1161 Rose, A., Prager, F., Chen, Z., & Chatterjee, S. (2017b). Economic consequence analysis of disasters: The E-CAT software tool. Springer. https://doi.org/10.1007/978-981-10-2567-9 Rose, A., Wei, D., Byrd, K., & John, R. (2021). Economic impacts of spillover effects of terrorism countermeasures at public assembly sites. Under review by the Journal of Homeland Security and Emergency Management. Rose, A., Wei, D., & Chen, Z. (2020). Estimating U.S. anti-dumping/countervailing (AD/CVD) duty enforcement benefits. In Z. Chen et al. (Eds.), Development studies in regional science: Essays in honor of Kingsley Haynes (pp. 449-474). Springer. https://link.springer.com/chapter/10.1007/978-981-15-1435-7_22 Rosoff, H., John, R., & Prager, F. (2012). Flu, risks, and videotape: Escalation of fear and avoidance behavior. Risk Analysis, 32(4), 729–743. https://doi.org/10.1111/j.1539-6924.2012.01769.x Schulze, W., & Wansink, B. (2012). Toxics, Toyotas, and terrorism: The behavioral economics of fear and stigma. Risk Analysis, 32(4), 678–694. https://doi.org/10.1111/j.1539-6924.2011.01748.x Shupp, R., Loveridge, S., Skidmore, M., Lim, J., & Rogers, C. (2017a). Risk, loss, and ambiguity aversion after a natural disaster. Economics of Disasters and Climate Change, 1(2), 121–142. Shupp, R., Loveridge, S., Skidmore, M., Lim, J., & Rogers, C. (2017b). Trust and patience after a tornado. Weather, Climate, and Society, 9(4), 659–668. Skidmore, M., & Toya, H. (2002). Do natural hazards promote long-run growth? Economic Inquiry, 40(4), 664–687. https://onlinelibrary.wiley.com/doi/full/10.1093/ei/40.4.664 Slovic, P. (1987). Perception of risk. Science, 236(4799), 280–285. https://doi.org/10.1126/science.3563507 Straub, A. M. (2020). “Natural disasters don’t kill people, governments kill people”: Hurricane Maria, Puerto Rico– recreancy, and “risk society.” Natural Hazards, 105, 1603–1621. https://doi.org/10.1007/s11069-020-04368-z Sunstein, C., & Thaler, R. (2008). Nudge: Improving decisions about health, wealth, and happiness. Yale University Press. Tinsley C. H., Dillon, R. L., & Cronin, M. A. (2012). How near-miss events amplify or attenuate risky decision making. Management Science, 58(9), 1596–1613. https://doi.org/10.1287/mnsc.1120.1517 US EPA Science Advisory Board. (2017). Economy-wide modeling of environmental regulations.US Environmental Protection Agency, Washington, DC. Viscusi, W. K. (2020). Pricing the global health risks of the COVID-19 pandemic. Journal of Risk Uncertainty, 61(2): 101–128. https://link.springer.com/article/10.1007/s11166-020-09337-2 von Winterfeldt, D., Farrow, S., John, R., Eyer, J., Rose, A., & Rosoff, H. (2020). Assessing the benefits and costs of homeland security research: A risk-informed methodology with applications for the US Coast Guard. Risk Analysis, 40(3), 450–475. https://doi.org/10.1111/risa.13403 von Winterfeldt, D., & O’Sullivan, T. M. (2006). Should we protect commercial airplanes against surface-to-air missile attacks by terrorists? Decision Analysis, 3(2), 63–75. https://doi.org/10.1287/deca.1060.0071 Walmsley, T., Rose, A., & Wei, D. (2021a). The impacts of the coronavirus on the economy of the United States. Economics of Disasters and Climate Change, 5(1), 1–52. https://link.springer.com/article/10.1007/ s41885-020-00080-1 Walmsley, T., Rose, A., & Wei, D. (2021b). Impact on the US macroeconomy of mandatory business closures in response to the COVID-19 pandemic. Applied Economics Letters 28/1528(15), 1293–1300. https://www.tandfonline.com/doi/full/10.1080/13504851.2020.1809626 Walmsley, T., Wei, D., John, R., Hlavka, J., Rose, A., & Byrd, K. (2021c). Impacts of the coronavirus on the economy of the United States: In-depth considerations of health effects, avoidance behavior, resilience, and international trade linkages [Final report to the Center for Accelerating Operational Efficiency (CAOE)]. Center for Risk and Economic Analysis of Terrorism Events (CREATE), University of Southern California, Los Angeles, CA. Waugh Jr., W. L. (2006). The political costs of failure in the Katrina and Rita disasters. Annals of the American Academy of Political and Social Science, 604(1), 10–25. https://doi.org/10.1177/0002716205284916 Wei, D., Chen, Z., & Rose, A. (2020). Evaluating the role of resilience in recovering from major port disruptions: A multi-regional analysis. Papers in Regional Science, 99(6), 1691–1722. https://rsaiconnect.onlinelibrary.wiley. com/doi/10.1111/pirs.12553 Whitehead, J. C. (2003). One million dollars per mile? The opportunity costs of hurricane evacuation. Ocean and Coastal Management, 46(11–12), 1069–1083. https://doi.org/10.1016/j.ocecoaman.2003.11.001

Behavioral economic consequences of disasters  71 Xie, W., Rose, A., Li, S., He, J., Li, N., Ali, T., & Wei, W. (2018). Dynamic economic resilience and economic recovery from disasters: A quantitative assessment. Risk Analysis, 38(6), 1306–1318. https://doi.org/10.1111/risa.12948 Zhao, M., Rosoff, H., & John, R. (2019). Media disaster reporting effects on public risk perception and response to escalating tornado warnings: A natural experiment. Risk Analysis, 39(3), 535–552. https://doi.org/10.1111/ risa.13205 Zhou, L., & Chen, Z. (2021). Are CGE models reliable for disaster impact analyses? Economic Systems Research, 33(1), 20–46. https://www.tandfonline.com/doi/full/10.1080/09535314.2020.1780566

5. The role of biases and heuristics in addressing natural disasters Howard Kunreuther and Wouter Botzen*

1. INTRODUCTION Many individuals who reside in hazard-prone areas would prefer not to think about possible natural disasters. Those at risk exhibit a set of biases that cause them to avoid undertaking protective measures. In the 1940s, Herbert Simon (1947) coined the term satisficing, observing that we rarely make decisions optimally. He noted that because of our cognitive limitations, we utilize heuristics or rules of thumb that are adequate for solving most of the decisions that we face daily but are far from optimal in other circumstances. This idea was extended in experiments by Daniel Kahneman and Amos Tversky in the 1970s that revealed two biases that influenced judgments on the likelihood that an outcome will occur: • Availability: The ease in which similar instances of the event can be recalled. For example, if homeowners residing in hazard-prone areas have not experienced a disaster, they may prefer not to reflect on its future occurrence until after experiencing damage. Then they may focus on the consequences of a future event without reflecting on the likelihood that they may experience another loss (Tversky & Kahneman, 1973). • Representativeness: The degree to which an event is stereotypically associated with a particular category. For example, individuals may underestimate the risk of flooding from a hurricane if they think of these disasters as primarily wind-related events (Kahneman & Tversky, 1972). Over the past 50 years, many controlled experiments and field studies in psychology and behavioral economics have further developed these ideas, with a particular focus on how individuals react under conditions of risk and uncertainty (e.g., Dormady et al., 2021). Findings from these studies are summarized in Daniel Kahneman’s book, Thinking, Fast and Slow (2011), which highlights two modes of thinking. Intuitive thinking (System 1) operates automatically and quickly with little or no effort and no voluntary control. In contrast, deliberative thinking (System 2) allocates attention to effortful and intentional mental activities where individuals undertake trade-offs, such as comparing the costs of investing in protective measures with its potential benefits. If one has considerable experience with certain events, such as knowing when to brake when driving a car, intuitive thinking will generally lead to good decisions. On the other hand, if one has not experienced a natural disaster, then those at risk are likely to not pay attention to the consequences of disasters by assuming they will not experience a loss. 72

The role of biases and heuristics in addressing natural disasters   73

2. BIASES AND HEURISTICS INFLUENCING DISASTER PREPAREDNESS DECISIONS We now turn to the cognitive biases and heuristics that influence individuals’ decisions on whether to purchase insurance or invest in cost-effective disaster risk reduction measures:1 • Myopia: There is a tendency to focus on overly short future time horizons that are typically shorter than the planning horizons required to determine the long-term value of protective investments against natural disasters. Controlled experiments and field surveys reveal that this behavior can be partially explained by high discount rates, so the expected benefits of risk reduction measures are underestimated (Gneezy & Potters, 1997). This results in underinvestment in measures that lower natural disaster damage to properties, such as elevating homes. They are often characterized by relatively high up-front costs and somewhat lower annual risk reduction benefits that accrue over the lifetime of a home. A related factor that characterizes myopia is narrow framing, which means that people isolate their current decision from its effect on other actions that may occur in the future (Kahneman & Lovello, 1993; Redelmeier & Tversky, 1992). For example, homeowners in flood and hurricane-prone areas may not consider how reducing future damage to their property, such as elevating a house, may enable them to remain in their current location, given the increased likelihood of flood-related losses from climate change due to sea level rise (SLR) and more intense hurricanes over the next 20 years. Empirical studies in the natural disaster domain support the importance of time preferences and myopia in under-preparedness for disasters (e.g., Botzen et al., 2019b; Gelino & Reed, 2020; Mol et al., 2018). • Amnesia: A variety of studies have shown that emotions such as worry for natural disasters, anxiety, or regret about uninsured losses influence uninsured individuals to purchase coverage following a disaster (e.g., Kunreuther & Pauly, 2017; Robinson & Botzen, 2018, 2019); however, feelings associated with the event fade quickly over time (Atreya et al., 2015). This appears to explain why many homeowners decide not to renew their flood insurance policy if they do not experience another disaster in the next few years (Michel-Kerjan et al., 2012). They do not appreciate that the best return on an insurance policy is no return at all. • Optimism: There is a tendency to underestimate the likelihood that losses will occur from future hazards. One reason for this misperception is that we base the likelihood of a specific event on our own personal experiences rather than experts’ estimate of the risk based on statistical data (Botzen et al., 2015; Mol et al., 2020a). More specifically, there is a tendency to underweight the probability of a disaster if one has not recently experienced the event (Robinson & Botzen, 2020). To illustrate, a longitudinal study of homeowners in Germany found that the likelihood of adopting risk mitigation measures is about 11% higher for households who experienced high flood damages (Osberghaus, 2017). • Inertia: A principal reason why we do not undertake protective measures to reduce future natural disaster losses is that we often prefer to stick with the status quo rather than forging new paths of action. This decision saves both time and energy given that one does not collect information on the costs and benefits of new alternatives (Samuelson

74  Handbook on the economics of disasters & Zeckhauser, 1988). Maintaining the current situation is the easy option when there is uncertainty, as illustrated by such aphorisms as “better the devil you know than the devil you don’t” and “when in doubt, do nothing.” This behavior is consistent with experimental evidence from the Netherlands showing that individuals prefer remaining uninsured against flood risk rather than examining different flood insurance options (Botzen et al., 2013). • Simplification: Individuals are likely to focus on a subset of the relevant facts when making choices involving risk. We are likely to pay attention to either the low probability of the event occurring or its potential consequences without recognizing that both dimensions are relevant. For example, prior to experiencing a loss, there is a tendency for many to view the event’s likelihood as falling below a threshold level of concern, so they do not pay attention to its consequences. If one is worried about the impact of a disaster, the reverse is true: a focus on its consequences without considering the probability of a future occurrence (Meyer & Kunreuther, 2017). ○ With respect to insurance purchase, McClelland et al. (1993) in a controlled experiment found a bimodal distribution with respect to the willingness to pay for coverage when the probability of a loss was 1 in 100. Most individuals (88%) either dismissed the likelihood of the event occurring by indicating their willingness to pay (WTP) for insurance was $0, while others indicated a WTP considerably more than the expected loss because they were worried about the consequences of the disaster. Similar patterns have been observed in studies on the demand for flood insurance in the Netherlands and the United States, where empirical evidence reveals that the lack of interest in purchasing coverage can be explained by a threshold model where individuals perceive the likelihood of a loss to be so low that they do not consider the potential impacts should a disaster occur (Botzen et al. 2019a; Robinson & Botzen, 2019). • Herding: Individuals’ choices are often influenced by other people’s actions especially under conditions of uncertainty. In an early survey of residents in flood- and earthquakeprone areas in the United States, one of the most important factors determining whether a homeowner purchased earthquake or flood insurance was discussion with friends and neighbors rather than considering the likelihood and consequences of a future disaster occurring (Kunreuther et al., 1978). More recently, in a study of households residing along the Rhine River in Germany, Bubeck et al. (2013) found a positive relationship between homeowners’ decisions to invest in flood risk mitigation and learning that their friends and neighbors had adopted loss prevention measures. Poussin et al. (2014) reported similar findings for residents in France. A field study analyzing factors that caused Queenslanders in Australia to buy flood insurance found that ownership was unrelated to perceptions of the probability of floods, but highly correlated with whether residents believed there was a social norm for the insurance (Lo, 2013). • Prominence: In determining whether to prepare for future disasters, the more prominent attributes are weighted more heavily in choices than in judgments. This behavior, uncovered in an experiment by Slovic (1975), is termed the prominence effect and forms the basis of a general theory of choice called the contingent weighting model (Tversky et al., 1988). Unlike expressed values, chosen actions need to be justified, and decisions congruent with prominent attributes are inherently more defensible. One reason that uninsured homeowners purchase coverage following severe damage from a

The role of biases and heuristics in addressing natural disasters   75 disaster is that their loss is the prominent attribute relative to the perceived likelihood of the event. This may explain the large increase in earthquake insurance purchases by homeowners impacted by the Loma Prieta earthquake in 1989 (Palm, 1995) and the Northridge quake of 1994 (Palm & Carroll, 1998). Most residents in California, when queried on the likelihood of another quake in their area after experiencing a severe loss, correctly responded that it was lower than before the disaster. They still wanted to purchase insurance after suffering a loss because avoiding the financial consequences of being uninsured following the next quake was prominent in their mind.

3.  ROLE OF A BEHAVIORAL RISK AUDIT A behavioral risk audit is an approach which recognizes biases that lead individuals to underprepare for future disasters. Rather than trying to eliminate the biases noted previously, individuals and policy makers should instead accept them as inherent constraints in decisionmaking and develop policies that work with, rather than against, them (Meyer & Kunreuther, 2017). The behavioral risk audit involves the following steps: • Context specification: Identify a hazard that poses a risk to life or property, and the situational context of that threat. For example, a threat could be wildfires, and the context could be homeowners living in exposed communities. • Bias manifestation: Identify how each of the decision biases would manifest in that context. For example, optimism would be reflected in a homeowner’s perception that their house would not suffer any damage from a future wildfire in an area subject to this hazard. • Implication for under-preparedness: For each of the biases, indicate how it could lead to a specific form of under-preparedness. For example, believing that one’s home will not suffer damage from a wildfire would cause the homeowner to underestimate the value of investing in fireproofing materials and designs. • Identification of remedies: Address the problem of under-preparedness by developing potential remedies for each of the biases. To avoid optimism, insurers and policy makers could develop graphic communication plans that increase the ease with which the impact of a future fire is conveyed and provide financial incentives that make investments in loss reduction measures worthwhile even under optimistic assessments of the fire risk. • Prioritization: After determining possible remedies for under-preparedness tied to each of the biases, prioritize their implementation based on synergies between the remedies and available resources. For example, communication plans that increase the ease with which the risk of fire is recalled can be inexpensive and may address optimism, amnesia, and myopia. Using the principles of choice architecture (Thaler & Sunstein, 2021), communications about risk can be framed to address the preceding biases so that decision makers are more likely to pay attention to them in the following ways: • Stretching time horizons: Due to the optimism and simplification biases, decision makers are likely to ignore the consequences of a disaster if they perceive its likelihood

76  Handbook on the economics of disasters

•

•

•

•

of occurrence to be below their threshold level of concern. Stretching the time horizon over which probabilities are presented will motivate them to consider undertaking protective measures. Botzen et al. (2016) applies this concept to increase demand for flood risk reduction measures for certain subgroups of the population. Web-based experiments with high stakes revealed that presenting the likelihood of a flood as “greater than 1-in-4” over a 30-year period instead of an annual probability of “1-in-100” significantly increases demand for flood insurance (Chaudhry et al., 2021). Similarly, Bradt (2019) conducted a web-based experiment and found that hypothetical WTP for flood insurance is higher when participants were given information that the probability of experiencing flooding is about 26% over a 30-year period, compared with information stating an annual flood probability of 1%. Robinson et al. (2021a) show that the effects of reframing flood probabilities over a long time horizon on flood insurance demand are especially strong among younger homeowners who are more concerned about the consequences of climate change. Bundling risks: Another way to get individuals to pay attention to low probability risks is to bundle different types of disasters into one insurance policy (Slovic et al., 1977). It is important that such an insurance product provides transparent information about the various risks in the bundle, so the policyholder is aware of the type of coverage being purchased. Correcting distorted memories: To address the amnesia bias, targeted risk communication campaigns can keep the memory of past natural disasters alive by stressing negative emotions that people experienced during these events (Garde-Hansen et al., 2017). Such communication messages are likely to be more effective if they are combined with concrete suggestions about cost-effective risk reduction measures that individuals can take to limit impacts of future disasters. Focusing on worst case scenarios: Due to the simplification bias and prominence effect, there is a tendency to focus on the low probability of a disaster and to ignore or underestimate its consequences prior to its occurrence. To address these biases, scenarios should be designed that highlight the financial impact of experiencing severe damage from a disaster if homeowners were uninsured and had not invested in loss-reduction measures. By switching their attention to the outcome, people may consider investing in protective measures rather than treating the disaster as below their threshold level of concern. Bradt (2019) and Botzen et al. (2013) show that a communication message that focuses on the consequences of flooding increases demand for flood insurance in the United States and the Netherlands, respectively. Using default options: The inertia bias causes individuals to maintain the status quo due to the time, costs, uncertainty, and attention associated with change. Providing property owners with the opportunity to opt out of specific coverage (e.g., earthquake protection) appended to their homeowners insurance policy, rather than offering this coverage as a separate policy, should increase the demand for this insurance (Kunreuther, 2018). This strategy of default nudges can be combined with the bundling of risks described earlier.2 Robinson et al. (2021b) present empirical evidence supporting this default effect in demand for natural disaster insurance in the United Kingdom and the Netherlands. They note that offering natural disaster coverage by default increases the uptake of insurance among consumers who have not experienced disaster losses and have not purchased disaster insurance. Homeowners in the provinces of British Columbia and Quebec

The role of biases and heuristics in addressing natural disasters   77 were much more interested in insuring themselves against earthquakes when they were provided with a policy and given the opportunity to opt out of this coverage than when they were asked whether they would like to purchase earthquake insurance (Kunreuther et al., 2021). This new evidence on the positive effects of default options on demand for natural disaster insurance should give a profit incentive to insurance companies to adopt these defaults more widely. • Triggering social norms: To address the herding bias, social norm nudges could be applied to improve the uptake of protection measures. For instance, homeowners who adopt protective measures could be given a seal of approval based on a certified inspection; others may then follow suit because it signals socially desirable behavior, and they may perceive an increase to their property values (Bicchieri & Dimant, 2019). Alternatively, social norms may be triggered by informing individuals about the uptake of protection measures by others or by communicating that being well prepared for disasters is the right thing to do.

4.  INCENTIVIZING INVESTMENTS IN PROTECTION Individuals at risk may be reluctant to invest in cost-effective loss reduction measures if they involve a high up-front cash outlay. Given budgetary constraints and individuals’ focus on short time horizons, it is difficult to convince them that the expected discounted benefits of the investment over the expected life of the property exceeds the cost of the measure. Decision makers’ resistance is likely to be compounded if they perceive the risk to be below their threshold level of concern. Residents in hazard-prone areas may also be concerned that if they move in the next few years, the property value of their home will not reflect the expected benefits from investing in loss reduction measures because the new owner will not be concerned about the risk of a disaster (Kunreuther et al., 2013). We now discuss ways to incentivize individuals to undertake these measures. 4.1  Insurance Premium Reductions To address the myopia bias, premium discounts for risk reduction can be coupled with lowinterest loans that spread the costs of installing damage mitigation measures over time. For cost-effective mitigation measures, the annual premium discount will exceed the annual loan costs, meaning that the homeowner receives net financial benefits every year (Kunreuther, 2015). Based on a survey and incentivized economic experiments, Botzen et al. (2009) and Mol et al. (2018, 2020b) show that individuals in the Netherlands are more likely to invest in flood damage mitigation measures if they are rewarded with annual discounts on their insurance policy. In general, economic incentives to invest in loss reduction measures and purchase natural disaster insurance coverage go hand in hand. Recent surveys in the United States and Germany reveal that homeowners with natural disaster insurance coverage are more likely to undertake risk reduction measures than uninsured individuals, thus reducing the moral hazard associated with purchasing coverage (Botzen et al., 2019a; Carson et al., 2013; Hudson et al., 2017; Petrolia et al., 2015). Using an economic experiment in the Netherlands, Mol et al. (2018) show that individual risk aversion can explain why homeowners prefer to purchase natural disaster insurance and

78  Handbook on the economics of disasters invest in risk reduction measures. Taken together, these findings suggest that these individuals view insurance coverage and risk reduction activities as complements instead of substitutes, suggesting that insurance companies have a good reason to encourage their clients to mitigate natural disaster damage. Reducing their premiums if they undertake these measures will provide an economic rationale for making their house safer. In practice, there are relatively few examples of insurance companies that provide financial incentives to encourage investments in risk reduction activities by policyholders (Hudson et al., 2020). Hudson et al. (2019) estimate that flood risk can be reduced between 20% and 26% across the European Union (EU) if premium discounts are given to policyholders who undertake measures that mitigate flood damage to their property. These findings are further supported by agent-based models (ABMs) that capture boundedly rational adaptation decisions by heterogenous households. Haer et al. (2017) developed such an ABM for Rotterdam in the Netherlands, which shows that future flood risk under SLR scenarios may be reduced by almost 30% by providing insurance premium discounts for risk reduction. Another ABM for Europe indicates that a 38% reduction in future flood risk can be obtained through similar economic incentives (Haer et al., 2019). Taken together, these findings illustrate that insurance premium reductions for investing in cost-effective loss reduction measures can be an important component of a broader natural disaster risk management strategy. Kunreuther et al. (2002) recommends insurance premium reductions for those who adopt loss reduction measures with an accompanying seal of approval to encourage others to invest in similar measures. For instance, homeowners in flood-prone areas who elevate their homes and obtain elevation certificates should be eligible for a lower premium under the National Flood Insurance Program. When others in their community learn of this, they may then follow suit due to the herding bias. 4.2  Mitigation Grants and Loans The Federal Emergency Management Agency (FEMA) created the Flood Mitigation Assistance (FMA) program in 1994 to reduce flood insurance claims. FMA is funded by premiums received by the National Flood Insurance Program (NFIP) to support loss reduction measures, such as elevation or relocation of property, floodproofing commercial structures, or demolition and rebuilding of property that has received significant damage from a severe flood. Long-term loans to homes and businesses for mitigation would encourage individuals to invest in cost-effective risk reduction measures. Consider a property owner who could pay $25,000 to elevate his coastal property from three feet below Base Flood Elevation (BFE) to one foot above BFE to reduce storm surge damage from hurricanes (Aerts et al., 2013).3 If flood insurance premiums are based on risk, then the annual premium would decrease by $3,480 from $4,000 to $520. A 15-year loan for $25,000 at an annual interest rate of 2¾% would result in annual payments of $2,036, so the savings to the homeowner each year would be $1,444 (i.e., $3,480–$2,036). 4.3  Means-Tested Vouchers A 2018 report from FEMA revealed that many low-income homeowners in high-risk floodprone areas in the United States may not be able to afford risk-based insurance premiums.

The role of biases and heuristics in addressing natural disasters   79 V Zone Property 20 Cost in Thousands of Dollars

18 16

Cost to Federal Government Cost to Homeowner

14 12 10 8 6

A Zone Property

4 2 0

r r 0 0 20+ 20+ che che 0–2 0–2 ars ars ears ears Vou Vou e e e e Y Y Y Y c c , , r r r, ran ran he he her, che ouc ouc Insu Insu ouc Vou nV nV nV n a a a a o o o o e/L e/L e/L e/L ranc ranc ranc ranc Insu Insu Insu Insu

Source:  Kousky and Kunreuther (2014).

Figure 5.1  Cost of program to the federal government and a hypothetical homeowner

Hudson et al. (2016) show that many low-income households in Germany and France would not be able to afford risk-based flood insurance premiums and invest in cost-effective loss reduction measures. Several reports and papers have proposed and examined possible federal policy solutions, all centered around some form of means-tested assistance for insurance and hazard mitigation investments (Dixon et al. 2017; Kousky & Kunreuther, 2014; National Research Council, 2015, 2016). One way to charge risk-based premiums while at the same time addressing issues of affordability is to offer means-tested vouchers that cover part of the cost of protection. Several existing programs could serve as models for developing such a voucher system in the United States: the Food Stamp Program, the Low-Income Home Energy Assistance Program (LIHEAP), and Universal Service Fund (USF) (Kunreuther & Michel-Kerjan, 2011). The amount of the voucher would be based on current income and determined by a specific set of criteria as outlined in a report by the National Research Council (2015). If the property owners were offered a multiyear loan to invest in mitigation measure(s), the voucher could cover not only a portion of the resulting risk-based insurance premium but also the annual loan cost to make the package affordable. As a condition for the voucher, the property owner could be required to invest in mitigation. An empirical study of homeowners in Ocean County, New Jersey, who were subject to flood-related damage from hurricanes reveals that the amount of the voucher is likely to be reduced significantly from what it would have been had the structure not been mitigated, as shown in Figure 5.1 for property in a high hazard flood area (the V Zone) and a lower hazard area (the A Zone) (Kousky & Kunreuther, 2014).

80  Handbook on the economics of disasters

5. REDUCING FUTURE DISASTER LOSSES BY ADDRESSING CLIMATE CHANGE According to a 2019 survey undertaken by the Yale Program on Climate Change Communication, a majority of Americans are worried about climate change.5 Nonetheless, concern does not drive most people to take positive action. Although the perceived future consequences, such as extreme weather events, may be harmful, they are familiar and perceived as controllable (Kunreuther & Slovic, 2021). It is important that political leaders and decision makers in the private and public sectors recognize the existence of these cognitive biases and turn to experts for advice on climate change. Climate scientists have long recognized that CO2 emissions and their effects are increasing exponentially. CO2 emissions and concentrations will be considerably higher in the coming years unless we take strong measures to reduce them. Figure 5.2 shows the increase in inflation-adjusted natural disaster losses that occurred from 1970 to 2020. The trend in these historical losses can be mainly explained by an increase in population and economic activities in areas that are prone to natural hazards (Hoeppe, 2016), but climate change may have contributed to some proportion of this trend already (Estrada et al., 2015). In the future, natural disaster losses are expected to rise even further because climate change is projected to increase the frequency and/or severity of several extreme weather events. For instance, the recent Intergovernmental Panel on Climate Change (IPCC, 2021) report contains empirical evidence that climate change will increase future heat waves, droughts, extreme precipitation, floods, and tropical cyclones. 500 450 400 350 300 250 200 150 100 50 0 1970

1975

1980

Insured losses

1985

1990

1995

Uninsured losses

2000

2005

2010

2015

2020

10-year moving average insured losses

10-year moving average economic losses Economic losses = insured + uninsured losses Source:  Swiss Re Institute.

Figure 5.2  Overall and insured worldwide losses from natural disasters

The role of biases and heuristics in addressing natural disasters   81 One way to reduce carbon emissions is to move from fossil fuels, such as coal and oil, to energy-efficient technologies, but relatively few homeowners have made this switch because of myopia. If homeowners focus on the long-term benefits of solar on their energy bills, they will see that the expected benefits are likely to exceed the up-front costs. In many areas of the country, the initial cost of adopting these measures is considerably less than the projected savings in energy costs over time. A recent study indicates that by 2030, solar will be cheaper than any other form of power generation in every state (Manghani, 2021). People also are reluctant to alter their current behavior due to the inertia bias. This tendency is reinforced by the herding bias when people interact with friends and neighbors who feel as they do, and by the prominence effect, in which they are unwilling to give up existing comforts and conveniences such as their accustomed level of heating and air conditioning, that loom important in their decision process. Moreover, we experience psychic numbing, in which numerical projections of CO2 concentrations fail to stimulate the emotional reactions necessary to motivate action. Additionally, pseudoinefficacy makes us feel that any personal contributions we make toward reducing a catastrophic threat will be insignificant and thus ineffective (Slovic & Västfjäll, 2019). People are also moving into harm’s way by not considering the severe damage climate change might inflict upon them in the coming years. From 1980 to 2018, the population of Florida’s hurricane-prone counties increased by 163% from 3.7 million to 9.8 million people. The population of the United States as a whole rose by only 61% during the same period. A 2013 analysis of 136 major coastal cities around the world revealed that SLR of an optimistic 20 cm (7.9 inches) by 2050 will cause the average annual flood losses in those cities to increase to $1.2 trillion in that year, compared to $52 billion in 2005. A more pessimistic scenario in which SLR reaches 40 cm (15.7 inches) by 2050 would bring average annual flood losses of $1.6 trillion. Houston, Texas, was among the 20 most vulnerable coastal cities in the study. Its average annual damage according to the optimistic scenario would increase by 78%, from $5.1 billion in 2005 to $9.1 billion in 2050 (Hallegatte et al., 2013). If CO2 emissions continue to grow exponentially, most of the United States could see 20 to 30 more days each year with maximum temperatures above 90 degrees Fahrenheit, and in the Southeast, 40 to 50 more such days.5 Such extreme heat poses serious health risks, especially for the very young and the very old, as well as construction and agricultural workers, and those living in urban cores. A study by researchers at the Earth Institute of Columbia University has found that, over the next 40 years, the rising temperatures associated with climate change could cause wildfires in California to continue to grow exponentially (Williams et al., 2019).

6. CONCLUSION Decades of research in psychology and behavioral economics has revealed that individual decision making with regard to low-probability/high-impact risks, such as natural disasters, is guided by heuristics and biases that prevent optimal protective responses that are inherent constraints in decision-making under risk and uncertainty. Policy makers and other key decision makers must develop a risk management strategy that accepts these biases as part of an individual’s choice process. We propose framing options using the principles of choice architecture that will help people pay closer attention to natural disaster risks. These strategies include bundling various

82  Handbook on the economics of disasters low-probability risks into a single insurance policy, stretching disaster probabilities over long time horizons, focusing on worst-case scenarios, correcting for distorted memories of past disasters, and the use of default options. Such strategies are likely to lead to adoption of protective measures by providing premium discounts to insurance policyholders who undertake risk reduction measures. To address the myopia bias, low-interest home improvement loans will spread the high up-front costs of these investments over time, so they are financially attractive. We recommend future studies to evaluate how well effective combinations of these measures fare in helping individuals prepare for the risks of natural disasters.

NOTES * Thanks to an anonymous reviewer and Carol Heller for helpful comments and suggestions on an earlier draft of this chapter. Support for this research comes from a grant from the Alfred P. Sloan Foundation (G-2018- grant f11100/SUB18-04), the Travelers–Wharton Partnership for Risk Management, National Science Foundation (NSF) grant (EAR-1520683) through Princeton University, and the Wharton Risk Management and Decision Processes Center. Wouter Botzen received financial support from the Netherlands Organisation for Scientific Research (NWO) VIDI Grant No. 452.14.005. 1. The materials included in this and the subsequent sections build upon Botzen and Kunreuther (2022). Another overview of biases in individual decision and risk analysis can be found in Montibeller and von Winterfeldt (2015). 2. This use of defaults does not limit consumer choice, so it is unlikely to be seen as unethical. Nevertheless, behavioral interventions may have regressive effects (e.g., White and Sintov, 2020). This means it is important to examine how these interventions impact low-income groups and to address any affordability issues that may arise with additional policies as we discuss in Section 4. 3. https://climatecommunication.yale.edu/about/projects/global-warmings-six-americas/ 4. https://www.ucsusa.org/resources/heat-waves-and-climate-change

REFERENCES Aerts, J. C. J. H., Botzen, W. J. W., de Moel, H., & Bowman, M. (2013). Cost estimates for flood resilience and protection strategies in New York City. Annals of the New York Academy of Sciences, 1294, 1–104. Atreya, A., Ferreira, S., & Michel-Kerjan, E. (2015). What drives households to buy flood insurance? New evidence from Georgia. Ecological Economics, 117, 153–161. Bicchieri, C., & Dimant, E. (2022). Nudging with care: The risks and benefits of social information. Public Choice, 191:443–464. Botzen, W. J. W., Aerts, J. C. J. H., & van den Bergh, J. C. J. M. (2009). Willingness of homeowners to mitigate climate risk through insurance. Ecological Economics, 68(8–9), 2265–2277. Botzen, W. J. W., de Boer, J., & Terpstra, T. (2013). Framing of risk and preferences for annual and multi-year flood insurance. Journal of Economic Psychology, 39, 357–375. Botzen, W. J. W., & Kunreuther, H. C. (2022). Role of insurance in reducing losses from natural disasters. Handbook of environmental hazards and society. Routledge. Botzen, W. J. W., Kunreuther, H. C., Czajkowski, J., & de Moel, H. (2019b). Adoption of individual flood damage mitigation measures in New York City: An extension of Protection Motivation Theory. Risk Analysis, 39(10), 2143–2159. Botzen, W. J. W., Kunreuther, H. C., & Michel-Kerjan, E. (2019a). Protecting against disaster risks: Why insurance and prevention may be complements. Journal of Risk and Uncertainty, 59, 151–169. Botzen, W. J. W., Kunreuther, H. C., & Michel-Kerjan, E. (2015). Divergence between individual perceptions and objective indicators of tail risks: Evidence from floodplain residents in New York City. Judgment and Decision Making, 10(4), 365–385. Botzen, W. J. W., Michel-Kerjan, E., Kunreuther, H. C., de Moel, H., & Aerts, J. C. J. H. (2016). Political affiliation affects adaptation to climate risks: Evidence from New York City. Climatic Change, 138(1), 353–360. Bradt, J. (2019). Comparing the effects of behaviorally-informed interventions on flood insurance demand: An experimental analysis of “boosts” and “nudges.” Behavioural Public Policy, forthcoming. https://ssrn.com/ abstract=3424279.

The role of biases and heuristics in addressing natural disasters   83 Bubeck, P., Botzen, W. J. W., Kreibich, H., & Aerts, J. C. J. H. (2013). Detailed insights into the influence of floodcoping appraisals on mitigation behaviour. Global Environmental Change, 23(5), 1327–1338. Carson, J. M., McCullough, K. A., & Pooser, D. M. (2013). Deciding whether to invest in mitigation measures: Evidence from Florida. Journal of Risk and Insurance, 80(2), 309–327. Chaudhry, S. J., Hand, M., & Kunreuther, H. (2021). Broad bracketing for low probability events. Journal of Risk and Uncertainty, 61, 211–244. Dixon, L., Clancy, N., Miller, B., Hoegberg, S., Lewis, M. M., Bender, B., Ebinger, S., Hodges, M., Syck, G. M., Nagy, C., & Choquette, S. R. (2017). The cost and affordability of flood insurance in New York City. Santa Monica, CA: RAND Corporation. Dormady, N. C., Greenbaum, R. T., & Young, K. A. (2021). An experimental investigation of resilience decision making in repeated disasters. Environment Systems Decisions, 41, 556–576. Estrada, F., Botzen, W. J. W., & Tol, R. (2015). Economic losses from US hurricanes consistent with an influence from climate change. Nature Geoscience, 8, 880–884. Federal Emergency Management Agency. (2018). An affordability framework for the national flood insurance. Office of Homeland Security. Washington, DC. https://www.fema.gov/sites/default/files/2020-05/Affordability_ april_2018.pdf. Garde-Hansen, J., McEwen, L., Holmes, A., & Jones, O. (2017). Sustainable flood memory: Remembering as resilience. Memory Studies, 10(4), 384–405. Gelino, B. W., & Reed, D. D. (2020). Temporal discounting of tornado shelter-seeking intentions amidst standard and impact-based weather alerts: A crowdsourced experiment. Journal of Experimental Psychology: Applied, 26(1), 16–25. Gneezy, U., & Potters, J. (1997). An experiment on risk taking and evaluation periods. The Quarterly Journal of Economics, 112(2), 631–645. Haer, T., Botzen, W. J. W., & Aerts, J. C. J. H. (2019). Advancing disaster policies by integrating dynamic adaptive behaviour in risk assessments using an agent-based modelling approach. Environmental Research Letters, 14, 044022, 1–9. Haer, T., Botzen, W. J. W., de Moel, H., & Aerts, J. C. J. H. (2017). Integrating household risk mitigation behaviour in flood risk analysis: An agent-based model approach. Risk Analysis, 37(10), 1977–1992. Hallegatte, S., Green, C., Nicholls, R. J., & Corfee-Morlot, J. (2013). Future flood losses in major coastal cities. Nature Climate Change, 3, 802–806. Hoeppe, P. (2016). Trends in weather related disasters: Consequences for insurers and society. Weather and Climate Extremes, 11, 70–79. Hudson, P., Botzen, W. J. W., & Aerts, J. C. J. H. (2019). Flood insurance arrangements in the European Union for future flood risk under climate and socio-economic change. Global Environmental Change, 58, 101966. Hudson, P., Botzen, W. J. W., Czajkowski, J., & Kreibich, H. (2017). Moral hazard in natural disaster insurance markets: Empirical evidence from Germany and the United States. Land Economics, 93(2), 179–208. Hudson, P., Botzen, W. J. W., Feyen, L., & Aerts, J. C. J. H. (2016). Incentivising flood risk adaptation through risk based insurance premiums: Trade-offs between affordability and risk reduction. Ecological Economics, 125, 1–13. Hudson, P., Ruiter, M. C., De Ruig, L. T., Kuik, O. J., Botzen, W. J. W., Le Den, X., Persson, M., Benoist, A., & Nielsen, C. N. (2020). An assessment of best practices of extreme weather insurance and directions for a more resilient society. Environmental Hazards, 19(3), 301–321. IPCC. (2021). Climate Change 2021: The physical science basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom, and New York, NY. In press. doi:10.1017/9781009157896. Kahneman, D. (2011). Thinking fast and slow. New York: Farrar, Straus and Giroux. Kahneman, D., & Lovallo, D. (1993). Timid choices and bold forecasts: A cognitive perspective on risk taking. Management Science, 39(1), 17–31. Kahneman, D., & Tversky, A. (1972). Subjective probability: A judgment of representativeness. Cognitive Psychology, 3(3), 430–454. Kousky, C., & Kunreuther, H. (2014). Addressing affordability in the National Flood Insurance Program. Journal of Extreme Events, 1(01). Kunreuther, H. (2015). The role of insurance in reducing losses from extreme events: The need for public-private partnerships. The Geneva Papers on Risk and Insurance–Issues and Practice, 40(4), 741–762. Kunreuther, H. (2018). Improving the National Flood Insurance Program. Behavioural Public Policy, 1–15. Kunreuther, H., Conell-Price, L., Kovacs, P., & Goda, K. (2021). The impact of a government risk pool and an opt-out framing on demand for earthquake protection (No. w29144). National Bureau of Economic Research. Cambridge, MA. Kunreuther, H., McNulty, P., & Kang, Y. (2002). Improving environmental safety through third party inspection. Risk Analysis, 22, 309–318. Kunreuther, H., Meyer, R., & Michel-Kerjan, E. (2013). Overcoming decision biases to reduce losses from natural catastrophes. In The behavioral foundations of public policy (pp. 398–414). Princeton University Press, Princeton, NJ.

84  Handbook on the economics of disasters Kunreuther, H., & Michel-Kerjan, E. (2011). At war with the weather: Managing large-scale risks in a new era of catastrophes. Cambridge: MIT Press. Kunreuther, H., & Pauly, M. (2018). Dynamic insurance decision-making for rare events: The role of emotions. The Geneva Papers on Risk and Insurance–Issues and Practice, 43(2), 335–355. Kunreuther, H., & Slovic, P. (2021). Learning from the COVID-19 pandemic to address climate change. Management and Business Review, 1(1). Lo, A. (2013). The role of social norms in climate adaptation: Mediating risk perception and flood insurance purchase. Global Environmental Change, 23(5), 1249–1257. Manghani, R. (2021). Total eclipse: How falling costs will secure solar’s dominance in power. Wood Mackenzie. January. https://www.woodmac.com/horizons/how-falling-costs-will-secure-solars-dominance-in-power. McClelland, G. H., Schulze, W. D., & Coursey, D. L. (1993). Insurance for low-probability hazards: A bimodal response to unlikely events. Journal of Risk and Uncertainty, 7(1), 95–116. Meyer, R., & Kunreuther, H. (2017). The ostrich paradox: Why we underprepare for disasters. Philadelphia, PA: Wharton School Press. Michel-Kerjan, E., Lemoyne de Forges, S., & Kunreuther, H. (2012). Policy tenure under the U.S. National Flood Insurance Program (NFIP). Risk Analysis, 32, 644–658. Mol, J. M., Botzen, W. J. W., & Blasch, J. E. (2020a). Behavioral motivations for self-insurance under different disaster risk insurance schemes. Journal of Economic Behavior & Organization, 180, 967-991. Mol, J. M., Botzen, W. J. W., & Blasch, J. (2020b). Risk reduction in natural disaster insurance: Experimental evidence on moral hazard and financial incentives. Journal of Behavioral and Experimental Economics, 84, 101500. Mol, J. M., Botzen, W. J. W., Blasch, J., & de Moel, H. (2020c). Insights into flood risk misperceptions of homeowners. Risk Analysis, 40(7), 1450–1468. Montibeller, G., & von Winterfeldt, D. (2015). Cognitive and motivational biases in decision and risk analysis. Risk Analysis, 35, 1230–1251. National Research Council. (2015). Affordability of national flood insurance program premiums [Report 1]. Washington, DC: The National Academies Press. National Research Council. (2016). Affordability of national flood insurance program premiums [Report 2]. Washington, DC: The National Academies Press. Osberghaus, D. (2017). The effect of flood experience on household mitigation: Evidence from longitudinal and insurance data. Global Environmental Change, 43, 126–136. Palm, R. (1995). Earthquake insurance: A longitudinal study of California homeowners. Boulder, CO: Westview Press. Palm, R., & Carroll, J. (1998). Illusions of safety: Cultural and earthquake hazard response in California and Japan. Boulder, CO: Westview Press. Petrolia, D. R., Hwang, J., Landry, C. E., & Coble, K. H. (2015). Wind insurance and mitigation in the coastal zone. Land Economics, 91(2), 272–295. Poussin, J., Botzen, W. J. W., & Aerts, J. C. J. H. (2014). Factors of influence on flood damage mitigation behaviour by households: Literature review and results from a French survey. Environmental Science and Policy, 40, 69–77. Redelmeier, D. A., & Tversky, A. (1992). On the framing of multiple prospects. Psychological Science, 3(3), 191–193. Robinson, P., & Botzen, W. J. W. (2018). The impact of regret and worry on the threshold level of concern for flood insurance demand: Evidence from Dutch homeowners. Judgment and Decision Making, 13(3), 237–245. Robinson, P., & Botzen, W. J. W. (2019). Determinants of probability neglect and risk attitudes for disaster risk: An online experimental study of flood insurance demand among homeowners. Risk Analysis, 39(11), 2514–2527. Robinson, P., & Botzen, W. J. W. (2020). Flood insurance demand and probability weighting: The influences of regret, worry, locus of control and the threshold of concern heuristic. Water Resources and Economics, 30, 100144. Robinson, P. J., & Botzen, W. J. W., Duijndam, D., Molenaar, A. (2021a). Risk communication nudges and flood insurance demand. Climate Risk Management, 34, 100366. Robinson, P., Botzen, W. J. W., Kunreuther, H., & Chaudhry, S. (2021b). Default options and insurance demand. Journal of Economic Behavior & Organization, 183, 39–56. Samuelson, W., & Zeckhauser, R. (1988). Status quo bias in decision making. Journal of Risk and Uncertainty, 1, 7–59. Simon, H. (1947). Administrative behavior. New York, NY: Macmillan. Slovic, P. (1975). Choice between equally valued alternatives. Journal of Experimental Psychology: Human Perception and Performance, 1(3), 280. Slovic, P., Fischhoff, B., Lichtenstein, S., Corrigan, B., & Combs, B. (1977). Preference for insuring against probable small losses: Insurance implications. Journal of Risk and Insurance, 44(2), 237–258. Slovic, P., & Västfjäll, D. (2019). The arithmetic of compassion and the future of risk management. In H. Kunreuther, R. J. Meyer, & E. O. Michel-Kerjan (Eds.), The future of risk management. Philadelphia, PA: University of Pennsylvania Press. Thaler, R. H., & Sunstein, C. R. (2021). Nudge: The final edition. Penguin. London, England.

The role of biases and heuristics in addressing natural disasters   85 Tversky, A., & Kahneman. D. (1973). Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 5, 207–232. Tversky, A., Sattath, S., & Slovic, P. (1988). Contingent weighting in judgment and choice. Psychological Review, 95(3), 371. White, L. V., & Sintov, N. D. (2020). Varied health and financial impacts of time-of-use energy rates across sociodemographic groups raise equity concerns. Nature Energy, 5, 16–17. Williams, A. P., Abatzoglou, J. T., Gershunov, A., Guzman-Morales, J., Bishop, D. A., Balch, J. K., & Lettenmaier, D. P. (2019). Observed impacts of anthropogenic climate change on wildfire in California. Earth’s Future, 7(8), 892–910.

6. Risk preferences and natural disasters: a review of theoretical and empirical themes Laura Bakkensen and Marc N. Conte

1. INTRODUCTION Experiences around the globe in the recent past have served as devastating reminders of the role that natural disasters play in shaping human existence. In 2020, overall global losses resulting from natural catastrophes totaled $210 billion across 980 different events, a notable increase from the $170 billion incurred across 860 events in 2019 (losses in 2020 USDs). The top five events based on insured losses in 2020 all occurred in the United States, where there were 22 weather and climate disasters that each resulted in at least $1 billion of losses, leading to roughly $95 billion in total damages across the 22 events. That said, the year’s costliest natural disaster was severe flooding in China during the summer monsoon rains, which led to damages of $17 billion. In addition to the financial damages incurred, the catastrophic events of 2019 and 2020 led to a combined total of 17,635 fatalities. Not all regions of the world experienced catastrophic natural disasters in 2020, yet no continent has been immune from experiencing earthquakes, drought, flooding, heat waves, severe weather, tornadoes, tropical cyclones, tsunamis, or wildfires in the first two decades of the new millennium.1 Understanding decision-making under uncertainty is critical as societies around the world attempt to strike the right balance with policies and market behavior that lead to investments in capital with different exposure to the risk of natural disasters. Given the well-known risks of such catastrophic events in particular locations around the globe, it might appear surprising that individuals, corporations, and government agencies choose to invest in locations at risk of experiencing such disasters when safer alternatives exist. This empirical fact raises several questions about how individuals make decisions to manage these risks, including their decisions surrounding housing location, insurance coverage, and adaptation, which we will explore in some detail in this chapter. Economists are generally comfortable assuming that individual choices are made to maximize the expected utility of the potential outcomes. For this assumption to hold in the context of natural disasters, individuals must have access to information about the risks and potential damages from different disaster events, this information must be sufficiently salient, subjective beliefs about risk exposure must be correlated with actual exposure levels, and this risk-exposure information must enter the decision-making process. One implication of the preceding assumptions and underlying conditions is that market prices should convey information about disaster risk to rational and attentive agents. Then, we should expect that the value of land (and other related market products such as insurance coverage) in a given period should already reflect the risk of damages from natural disasters facing the land. While empirical support for this theory exists in the market for agricultural land in the United States (see, e.g., Mendelsohn et al., 1994, and Severen et al., 2018), data 86

Risk preferences and natural disasters  87 in other markets is less supportive of the idea that all of these conditions hold in practice, making a thorough understanding of the relation between risk preferences and preparation for/ reactions to natural disasters critical to minimize the social costs associated with such events. In this chapter, we first present a brief overview of the foundational theory on risk preferences and empirical approaches to elicit risk preferences from individuals. We then turn to three key decisions surrounding natural disaster risk management: insurance, location choice, and adaptation. We review the current state of each general literature with regard to natural disasters, present stylized models of optimal decision-making to highlight the relevance of risk preferences in each context, and lastly review the state of the empirical risk-preferences literature in each realm. We also note key research gaps and frontiers for future work. We conclude with final thoughts and implications for policy.

2.  ECONOMIC THEORY OF RISK PREFERENCES Economics is rooted in the philosophy of utilitarianism and generally asserts that individuals make consumption decisions to maximize their utility subject to the bundle of goods they are able to afford. Therefore, consumption is determined by individual preferences and wealth. Under the assumption that wealth is exhausted in pursuit of consumption (where savings can be thought of as consumption in future periods), an individual’s utility as a function of wealth in period t is often written as follows:

U = U (Wt ) (6.1)

It is frequently assumed that utility increases with consumption of a good, although the rate of increase decreases with additional consumption, namely the utility from consuming the first unit of the good exceeds the marginal utility of the second unit of consumption. This assumption can be carried through to the shape of the utility function with regard to wealth so that (U(W)′ > 0, U″ (W) < 0), implying that the utility function is concave in wealth, which has direct ramifications for how economists think about attitudes toward risk. The following subsections will present the dominant model of behavior under risk, expected utility theory, and, briefly, prospect theory, an alternative to expected utility theory. After these alternative models of decision-making are presented, this section will conclude with a brief discussion of a fundamental, though increasingly questioned, assumption about the stability of preferences as well as an overview of empirical methods to elicit risk preferences in the real world. 2.1  Expected Utility Theory Expected utility theory posits that even though individuals may claim to be motivated by the pursuit of wealth, in reality, utility derives from what the wealth is used to purchase. So, when a lottery is presented, individuals will not evaluate the lottery based solely on the dollar prize being offered but also on how much they personally value the prize. This fact suggests that expected utility, rather than the expected value of the lottery, will determine how an individual values a lottery, based on her attitude toward risk.

88  Handbook on the economics of disasters U(Wealth)

U(c2) U(E[c]) E[U(c)]

U(c1) c1

E[c]

c2

Wealth

U–1(E[U(c)]) Note:  Figure 6.1 depicts an individual’s utility (y-axis) given different wealth levels (x-axis). The individual is risk averse because the utility function is concave: the individual would prefer a known payout relative to a lottery with equal expected payout that would pay out either a higher or lower value with equal probability.

Figure 6.1  Risk aversion Figure 6.1 depicts how an individual’s utility varies with wealth when facing a lottery that provides an equal chance of winning one of two cash prizes (c1 < c2). The figure illustrates several key values that are important to consider as an individual evaluates her willingness to pay to participate in the lottery. The value E[c] on the x-axis depicts the expected value of the lottery, where E[c] = 1/2c1 + 1/2c2. E[c] would be the maximum amount that an individual would be willing to pay for the chance to play the lottery, if she were risk neutral, meaning that she cared only about the expected monetary payoff associated with the lottery and derived no utility or disutility from the risk in the lottery. The y-axis measures the utility associated with different wealth levels. As the individual considers her willingness to pay to participate in the lottery, expected utility theory assumes that she will make her decision based on the lottery’s expected utility (E[U(c)]), which is simply the expected value of the utilities associated with the two possible lottery outcomes: E[U (c)] = 12 U (c1 ) + 12 U (c2 ) . The concave shape of the utility function in Figure 6.1 leads to the result that U(E[c]) > E[U(c)]. In words, this equation states that the utility of having the wealth level of E[c] with certainty is greater than the expected utility associated with the lottery that pays off c1 and c2 with equal probability. This result follows from Jensen’s inequality. The preceding result illustrates that an individual with a concave utility function would not be willing to pay any more than the value leading to the utility level given by U(E[c]). This value is given by U–1 (E[U(c)]), and this amount is referred to as the certainty equivalent associated with the lottery, as it represents the guaranteed amount of money that would make the individual just as well off as she would be if she were to participate in the lottery. And we see that for risk-averse individuals who have concave utility functions, the certainty equivalent is less than the expected prize from the lottery. For an individual with a linear utility function, the certainty equivalent would equal the expected prize from the lottery (U–1 (E[U(c)]) = E[c]). These types of individuals are known

Risk preferences and natural disasters  89 as risk neutral. For individuals with convex utility functions, the certainty equivalent is greater than the expected prize from the lottery (U–1 (E[U(c)]) < E[c]), as the risk of the gamble generates utility for these individuals, who are known as risk loving. Two important quantities related to the certainty equivalent have to do with the risk premium associated with a particular lottery. The absolute risk premium, πA, represents the difference between the expected value of the lottery and the certainty equivalent: πA = E[c] – U–1 (E[U(c)]). The relative risk premium, πR, depicts the size of the absolute risk premium −1 −1 πA ( E[U (c)]) E[U (c)] = E[c] − UE[c] = 1 − U (E[c] relative to the expected value of the lottery: π R = E[c] . These two measures of risk premia in turn can be expressed as functions of absolute risk aversion, −U ʹʹ(c) c! (c) A(c) = −UUʹʹʹ(c) , or relative risk aversion, R(c) = U ʹ(c) , respectively (Pratt, 1964; Arrow, 1965). U”(c) represents the concavity of the utility function, while U’(c) represents its slope. These measures of risk aversion can be incorporated into utility functions to describe individual attitudes toward risk. Common specifications of utility function that capture risk aversion are the constant absolute risk-aversion (CARA) utility and constant relative risk-aversion − ax (CRRA) utility. CARA utility is given by U (x) = 1− ae for a ≠ 0, where a is the coefficient of −U ʹʹ( x) absolute risk aversion (a = U ʹ( x) ) . If a = 0, then U(x) = x to capture risk neutrality. The CRRA 1−γ utility is given by U ( x) = x1− −γ 1 for γ ≠ 1, where γ is called the coefficient of CRRA. If γ = 0, then U(x) = x – 1 to capture risk neutrality (Kreps, 1990). Expected utility theory is broadly embraced and applied in economics, although the axioms upon which the theory is built have been challenged given observed behavior, notably in experimental settings. A classic example challenging the independence axiom of expected utility theory comes from Allais (1953). The independence axiom (Samuelson, 1952) states that if the lottery P′ is preferred to the lottery P, then the mixture αP′ + (1 – α)P″ will be preferred to the mixture αP + (1 – α)P″ ∀ α > 0 and P″. The Allais paradox arises from a scenario in which a person chooses between lotteries a1 and a2 and between lotteries a3 and a4, where a1, a2, a3, and a4 are as follows: a1, 100% chance of $1,00,000; a2, 10% chance of $5,000,000, 89% chance of $1,000,000, and 1% chance of $0; a3, 10% chance of $5,000,000, 89% chance of $1,000,000, and 1% chance of $0; and a4, 11% chance of $1,000,000 and 89% chance of $0. In this example, an agent preferring a1 to a2 and preferring a4 to a3 (or to a1 and a3 to a4) exhibits behavior consistent with the agent having indifference curves that are parallel straight lines as surmised by expected utility theory (Machina, 1987). Laboratory experiments have shown that agents commonly choose a1 and a3, violating the independence assumption. The paradox of these observed choices emerges when considering that the expected value of a1 is $1 million and the expected value of a2 is $1.39 million. By preferring a1 to a2, it appears that the agent is maximizing expected utility rather than expected value. If a1 > a2, then u(1) > 0.1u(5) + 0.89u(1) + 0.01u(0), meaning that 0.11u(1) > 0.1u(5) + 0.1u(0), which then implies (as seen by adding 0.89u(0) to each side) that 0.11u(l) + 0.89u(0) > 0.1u(5) + 0.90u(0), meaning that an expected utility maximizing agent must prefer a4 to a3. Given that the expected value of a4 is higher than that of a3, then an agent maximizing expected value should prefer a4 to a3, but the choice in the first lottery pair is inconsistent with the choice in the second stage (List and Haigh, 2005). While the implications of the above example are still being explored (see, e.g., Andreoni and Sprenger, 2010; Gneezy et al., 2006), the results might suggest that individuals are not

90  Handbook on the economics of disasters fully rational maximizers of expected utility. To more accurately depict human behavior might require modeling individuals as concerned with their own self-interest but unable to achieve the best outcome due to constraints imposed by their limitations and biases. 2.2  Prospect Theory Prospect theory is a response to the gap between observed behavior and the predictions of expected utility theory. While there are similarities between the two approaches, the fundamental divergence is that prospect theory acknowledges limitations and biases that prevent individuals from making fully rational choices. One key difference is that prospect theory creates separate expected utility functions for gains and for losses (Kahneman and Tversky, 1979) based on a reference point. Each of these functions is motivated by loss aversion, which describes the tendency of individuals to prefer avoiding losses rather than acquiring equivalent gains, and it relates to the idea of an endowment effect, with the value of an object varying based on the framing of a transaction (Kahneman et al., 1990). In describing how the utility function changes with gains in wealth, prospect theory assumes that the willingness to pay for access to a lottery associated with potential gains displays risk aversion, meaning that the function has a concave shape as the wealth associated with the lottery increases. Simultaneously, the willingness to accept needed to sell the lottery when facing certain losses is assumed to display risk-loving behavior, meaning that as the potential gains from the lottery increase, the utility of the lottery increases at a faster rate. The implication is that the utility effect of a loss of a particular amount of wealth is greater than that of an equivalent gain in wealth. See Figure 6.2 for a depiction of this S-shaped relationship between the utility from gains versus the disutility from losses. Utility

Losses

Gains

Reference point Note:  Figure 6.2 depicts an individual’s utility (y-axis) given different levels of gains versus losses (x-axis). Prospect theory describes an S-shaped utility function where the absolute value of utility from a gain is less than the absolute value of utility from an equivalent loss.

Figure 6.2  Prospect theory

Risk preferences and natural disasters  91 Another difference between prospect theory and expected utility theory is the use of decision weights on probabilities rather than the use of probabilities of occurrence. This difference is based on the observation that people typically focus more on very rare events rather than on the common risks of everyday life. Third-generation prospect theory differs from first-generation (Kahneman and Tversky, 1979) and second-generation (Tversky and Kahneman, 1992) prospect theory by allowing for uncertain reference points and different decision weights for gains and losses that sum to one in order to account for the preferencereversal phenomenon that cannot be explained by expected utility theory (Schmidt and Hey, 2004, Schmidt et al., 2008). 2.3  Stability of Risk Preferences In consumer choice theory, preferences are typically assumed to be stationary, so that observed differences in the choices of goods purchased over time are attributed to changes in the budget constraint. This assumption is useful in allowing economists to model causal relationships between changes in the opportunity sets available to consumers and their subsequent choices of bundles of goods via the use of comparative statics (Andersen et al., 2008). The seminal defense of this approach is offered by Stigler and Becker (1977), who argue that greater insights into the behavior responsible for changing consumption decisions over time are available via changes in the bundles of goods that are affordable (the opportunity set) than would be associated with the relaxation of the preference-stationarity assumption. At first glance, this assumption may seem too disconnected from reality. However, the idea of preference stationarity can be relaxed by making the arguments of the utility function state contingent, meaning that choices can vary across different states of nature so long as these states are exogenous to consumer choices. This possibility can be used to explain apparent differences in risk preferences across lottery pairs, whose differences might relate to different states of nature, in a way that is consistent with expected utility theory (Andersen et al., 2008). While the early defenses of preference stability were made when data on preferences were sparse and challenging to collect, empirical work in personality psychology in subsequent years has led to a consensus on a wide range of topics, including empirical support for preference stability, as reviewed in Caplan (2003). Still, stability of preferences, and the exogeneity of different states of nature are empirical questions (Schildberg-Hörisch, 2018), and the measurement of risk preferences, including changes in these preferences associated with natural disasters, is an active area of research. See Section 3.4 for a review of the empirical literature examining risk preference stability in the case of natural disasters. It must be noted that there are theoretical concerns about the researchers’ ability to accurately capture risk attitudes in lab experiments or administrative data (Rabin and Thaler, 2001). This concern arises from the fact that the lotteries presented to elicit risk preferences typically involve payments in a relatively narrow range (possibly due to budget constraints for the researchers) over which expected utility theory would predict risk neutrality. Findings of risk aversion over this range of wealth levels would imply extreme concavity of the utility function that would lead to ludicrous outcomes. For example, a riskaverse individual who always turns down a 50–50 gamble of losing $10 or gaining $11 would never accept a 50–50 gamble of losing $100 or gaining $Y, no matter the value of Y (Rabin and Thaler, 2001).

92  Handbook on the economics of disasters 2.4  Eliciting Risk Preferences In tandem with the theoretical literature, a vibrant empirical literature aims to elicit risk preferences across a variety of settings. We provide a brief and nonexhaustive review of types of empirical methods to elicit risk preferences, including both stated and revealed preference approaches, using surveys, artifactual field experiments, and observational data. A commonly used elicitation method is the survey, which is advantageous because it is typically less resource intensive to implement than data collection in the field and gives the researcher the ability to carefully craft the exact questions and controls needed rather than having potentially incomplete information sometimes found in observational data sets such as administrative data or proprietary data sets from the real world. Survey-based elicitation questions often contain a lottery or risky decision upon which individuals select their preferred option. Charness et al. (2013) provide a nice review of specific experimental methods in economics and psychology to elicit risk preferences, highlighting the advantages and disadvantages of a variety of methods. Dave et al. (2010) examine how the level of difficulty of the elicitation method needs to be appropriate for the level of numerical skill of the respondent. They find that simpler questions are less noisy for low-skilled respondents, and complex questions are more accurate with sufficient skill. While advantageous in some contexts, experimental survey data is not without critics. Along with biases and errors that can manifest in surveys in general (Groves et al., 2011), risk preference elicitation can also suffer from additional concerns. A core decision researchers make early on surrounds respondents’ incentives to participate. Namely, whether real money payoffs should be used for respondents. If real payoffs are not used, concerns over hypothetical bias leading to behaviors diverging from real-world decisions may occur (Harrison, 2006). Even if real money is used in the lottery setting, payoffs may be too small to be meaningful, while higher payoffs might preclude adequate sample size. Indeed, Rabin (2013) (among several papers) highlights that the scale of lottery payoffs can change the degree of risk aversion elicited, cautioning the assumption of an expected utility theory framework. Davis and Holt (1993) offer a potential solution for researchers to use larger bets with a lower probability of winning as people may seem less risk averse if stakes are too low. To increase the real-world relevance while still maintaining control over design, researchers have increasingly turned to artifactual field experiments—similar to conventional lab experiments except deployed in nontraditional real-world respondent pools (Harrison and List, 2004)—to elicit preferences. While often costlier to conduct than laboratory experiments, researchers are able to tap larger populations of respondents across different economic and social settings. An advantage of artifactual field experiments in contexts where the cost of living is low is that lottery payoffs are more significant, allowing for larger samples and examination of expanded theoretical perspectives (Tanaka et al., 2010). However, Harrison et al. (2020) highlight that classic survey issues of sample selection and attrition can have parallel consequences in risk preference elicitation approaches using longitudinal data. Thus, best practices for fieldwork must still be utilized (Harrison and List, 2004), including ethical considerations (Desposato, 2015). While there are many specific protocols to elicit risk preferences in conventional and artifactual lab experiments, two main protocols used in both settings are the unitary (or single) lottery choice and the multiple price list (Charness et al., 2013; Holt and Laury, 2014). The unitary (or single) lottery choice elicits risk preferences through a single question for the

Risk preferences and natural disasters  93 respondent to answer, for example, allocation of assets across risky versus safe domains (Gneezy and Potters, 1997) or other binary gambles (Dave et al., 2010). Unitary lotteries are advantageous given their simplicity and are still able to differentiate risk preferences across individuals. In contrast, and based on seminal work by Holt and Laury (2002), multiple price list experiments have respondents choose between multiple consecutive binary outcomes where the expected payoff of one of the outcomes increases faster than the other. Risk preferences can be determined based on the point at which the respondent switches from preferring one outcome for the other (Drichoutis and Lusk, 2016). While these more complex methods of elicitation methods allow the researcher to examine more advanced risk preference questions relative to simpler methods, they demand higher sophistication from the respondent that, if not present, may lead to measurement error and limited predictive power (Charness et al., 2013). In addition, care must be taken with multiple price lists to accurately identify the utility function curvature versus nonlinear probability weighting by the respondents (Drichoutis and Lusk, 2016). Overall, there is no superior method, and the appropriate elicitation method depends on the context and researcher objectives (Charness et al., 2013). Lastly, researchers also turn to revealed preference evidence using observational data, with the advantage that these decisions represent real-world choices with meaningful consequences. However, since researchers often use already existing behavioral data, a potential limitation is the reduction in researcher ability to control the setting or collect all relevant control variables. The latter is particularly important, as risk preferences have been found to correlate with a variety of other factors (Dohmen et al., 2010). Nonetheless, researchers have long utilized this data to examine decisions surrounding risk in many real-world contexts, including insurance and investment decisions (e.g., Einav et al., 2012; Szpiro, 1986).

3.  THEMES IN RISK PREFERENCES AND NATURAL DISASTERS Having reviewed foundational concepts and definitions in risk preference theory as well as briefly overviewed empirical methods to assess risk preferences, we now apply these concepts to the case of natural disasters. In particular, we examine the cases of insurance, location choice, and adaptation. We first briefly review frontiers that are not based on risk preferences in each case. We then present a simple theoretical model for each case to highlight the contribution of a risk preferences lens. We then review the empirical risk preferences literature in each case as well as highlight key research gaps and opportunities. Lastly, we examine the empirical evidence surrounding the stability of risk preferences in the context of natural disasters. Altogether, this section motivates and examines key research themes and opportunities relating to risk preferences and natural disasters. 3.1 Insurance In theory, insurance could play a prominent role in mitigating the economic damages caused by natural disasters, either through the simple reimbursement of claims in the wake of catastrophic events or through increased costs of building or purchasing property in areas at risk of disaster events. In practice, some features of insurance markets might impede insurance policies from conveying the true risk of disaster events to policyholders.

94  Handbook on the economics of disasters The total economic damages from natural disaster events have been shown to be drawn from a fat-tailed distribution (see, e.g., Conte and Kelly, 2018, regarding tropical cyclone damages). Without going into too much detail, the probability of extreme events in a fat-tailed distribution decays at a much slower rate than in a normal distribution. One implication of this result is that extreme events (e.g., storms that occur once in 200 years) are much more likely in a fat-tailed distribution than in a thin-tailed distribution, like the normal. Another challenging implication is that it is difficult to learn about the potential impacts of extreme events in a fat-tailed distribution, as the damages associated with a 1% tail event might be orders of magnitude less than the damages associated with a 0.5% tail event. See Conte and Kelly (2021) for more details about fat tails and natural disasters. These features of damages from natural disasters mean that firms offering disaster insurance are forced to hold costly financial reserves in order to maintain solvency in the wake of catastrophic events. As a result of these costly reserves, insurers will not be able to provide coverage at actuarially fair prices (see Section 3.1.1 for a definition of actuarially fair insurance). A direct implication of this fact is that risk-averse homeowners will not fully insure. An additional implication arises when we consider the salience of risk posed by natural disasters in practice. Because natural disasters are relatively rare events, it is quite possible that the risk of catastrophic disasters may not be salient to economic agents, although the intensity of natural disasters appears to be increasing with climate change (Coronese et al., 2019). Evidence to support this possibility comes from flooding risk and unexpectedly low participation levels among eligible properties in the National Flood Insurance Program (NFIP) in the United States (Kriesel and Landry, 2004; Petrolia et al., 2013). While it has been shown that policy uptake increases in the immediate wake of flood events (Gallagher, 2014), much of this increase has been shown to be due to regulatory requirements for disaster aid eligibility (Kousky, 2017), which might be taken as further evidence that this source of risk is not driving the decisionmaking process for homeowners. Insurance policies that reflect the true risk of damages could serve to increase the salience of disaster risk and potentially reduce the damages caused by disaster by promoting adaptations to such risk or by affecting the location choice of homeowners and firms. As mentioned previously, the fat-tailed nature of disaster damages makes this possibility less likely. Further limiting the potential role of insurance in motivating more efficient responses to disaster risk is the political economy of the situation that finds many politicians and regulators interested in maintaining robust property values in at-risk areas (e.g., coastal zones), meaning that they might apply pressure to keep insurance rates low. The fact that the state-run insurance firm increased insurance rates across the state of Florida in response to the 2004 hurricanes, while private insurers tended to only increase rates in those counties with the greatest claims, might be taken as evidence of this political-economy mechanism (Conte and Kelly, 2020). 3.1.1  Model of insurance demand As mentioned in previous sections, natural disasters place people at risk of potentially catastrophic damages in many locations around the globe. And yet, many people live in places that are at disproportionately higher risk of being impacted by natural disasters (e.g., coastal Florida). Embracing the assumption that economic agents are fully rational actors, this fact suggests that the amenities associated with living in these locations must be greater than the costs, including those associated with the risk of natural disasters.

Risk preferences and natural disasters  95 The results described in the preceding section suggest that perhaps economic agents are not capable of fully rational decision-making or that the information necessary to make such decisions is not always available (or seems too costly to obtain). The latter possibility is particularly worthy of consideration given the previous discussions of when and how property value and the price of disaster insurance fail to provide adequate signals of the risk of damages due to natural disasters. Interestingly, as noted earlier, the literature exploring this topic, a fundamental example of decision-making under uncertainty, has generally proceeded under the assumption that households are risk neutral. To highlight the importance of accounting for risk preferences when attempting to explain behavior in markets related to natural disaster impacts (e.g., housing, insurance), we will now present simple models of decision-making under uncertainty. The models in this section will also allow us to briefly comment on the importance of insurance pricing in providing accurate risk information under real-world conditions. We begin with a canonical example of decision-making under uncertainty: demand for insurance (see, e.g., Deaton and Muellbauer, 1980; Kreps, 1990). Let an agent with wealth W face a financial loss L with probability p. The agent can protect himself against this loss through the purchase of insurance, with a policy that will provide payment in the event that a loss occurs. Let a policy associated with a payment of X dollars in the event of a loss cost qX dollars. The optimal amount of insurance for this agent can be determined by viewing this as an optimization problem under uncertainty, where the agent must choose how much coverage to purchase to maximize expected utility, as given by the following:

max pu(W − qX − L + X ) + (1 − p)u(W − qX ) (6.2) X

If we allow U(X) to represent the agent’s objective function, then the first-order condition is

dU = p(1 − q)uʹ(W − qX − L + X ) − (1 − p)quʹ(W − qX ) = 0 (6.3) dX

This condition is necessary and sufficient for an interior solution if the utility function u is concave. Rearranged, the condition reveals that the marginal benefit of an extra dollar of insurance in the loss state multiplied by the probability of loss is equal to the marginal cost of the extra dollar of insurance in the no loss state. Actuarially fair insurance provides coverage at a price such that the expected payout for the insurer just equals the cost of insurance, that is, p = q. Under this condition, the first-order condition simplifies to

uʹ(W − qX − L + X ) = uʹ(W − qX ) (6.4)

The agent will set X = L, fully insuring against the loss. A concave utility function corresponds to an agent who is risk averse, so (6.4) reveals the well-known result that risk-averse agents will fully insure when offered actuarially fair insurance. This result is potentially quite significant because for insurance to play a role in providing signals about risk from natural disasters that might be more salient than previous experience, households must purchase insurance.

96  Handbook on the economics of disasters Of course, it is also well known that risk-neutral agents will be indifferent between insurance purchase and no insurance when insurance is actuarially fair. The prevalence of different attitudes toward risk in communities threatened by damages from natural disasters will impact the ability of insurance to increase the salience of these intermittent events whose damages may be drawn from fat-tailed distributions, making it even more difficult to learn about extreme events. Even among risk-averse agents, demand for insurance falls when prices are no longer actuarially fair (namely q > p). To see this, recall that for interior solutions, the optimal level of coverage X satisfies the first-order condition:

uʹ(W − qX − L + X ) (1− p)q = > 1 (6.5) p(1− q) uʹ(W − qX )

And we see that even a risk-averse agent will not fully insure (X* < L) under these conditions. Given the costly transfer of wealth to the loss state, the agent prefers to live with less wealth there and more wealth in the no loss state. Insurance priced above actuarially fair levels is likely to occur for several reasons, including the need for insurers to hold costly reserves to cover claims in the wake of tail events. On the other hand, there are also many conditions under which the price of insurance will understate the true risk facing the agent. For example, NFIP prices have long been criticized for being discounted for some policies, and plans by the Federal Emergency Management Agency (FEMA) to raise rates recently have come under fire from politicians because they do not want to harm their constituents in the short term, which aligns with election cycles. The impact of offering coverage at artificially low prices on policy uptake depends on the risk preferences of the agent. Specifically, the results depend on whether the individual has increasing or decreasing absolute risk aversion. Absolute risk aversion is a measure of the local curvature of the utility function, as defined in Section 2.1. Specifically, for a twice-differentiable utility function u(), the Arrow-Pratt coefficient of absolute risk aversion is given by the following:

A(x) =

−uʹʹ(x) (6.6) uʹ(x)

A key concern is how an individual’s risk aversion varies with her wealth level. The utility function u() is said to have decreasing (constant, increasing) absolute risk aversion if A(x) is a decreasing (constant, increasing) function of x. The implications of decreasing absolute risk aversion (DARA) are that an individual’s tolerance for risk is increasing in her wealth levels. In the case of demand for insurance, we would expect that an individual with DARA preferences would prefer to purchase less insurance as her wealth increases. To see this, we turn to comparative statics, noting the following:

d 2U = p(1 − q)uʹʹ(W − qX − L + X ) − (1 − p)quʹʹ(W − qX ) (6.7) dXdW

While the sign of the preceding expression is indeterminate, we can use the first-order condid 2U > 0 tion for an optimal amount of coverage to help illustrate that it must be the case that if dXdW

Risk preferences and natural disasters  97 for X*(W), then, if W′ > W, it must be that X* (W′) > X* (W). To see this, note that when X = X *(W),

(1 − p)q =



p(1 − q)uʹ(W − qX − L + X ) . (6.8) uʹ(W − qX )

Then,

⎛ uʹʹ(W − qX − L + X ) uʹʹ(W − qX ) ⎞ d 2U | X = X *(W ) = p(1 − q)uʹ(W − qX − L + X ) ⎜ = ⎟ (6.9) dXdW ⎝ uʹ(W − qX − L + X ) uʹ(W − qX ⎠ And, we see that

d 2U | dXdW X = X *(W )

sign =

[ A(W − qX ) − A(W − qX − L + X )] (6.10)

The final expression allows us to explore how coverage varies with wealth, and we see that when p = q, X = L ∀W, so the choice of coverage is independent of wealth. When p < q, we know that X < L, and we see that the purchase of coverage decreases with more wealth for individuals with DARA preferences and that the coverage purchased increases with additional wealth for individuals with increasing absolute risk averse (IARA) preferences, while remaining unchanged for those with CARA preferences. Having considered how the demand for insurance varies with regard to wealth based on different risk preferences, we now consider how the optimal amount of coverage varies in response to the price of coverage, q. To do so, we will adopt the same approach that we used to explore how demand for coverage varied with wealth. We begin by differentiating the first-order condition for the optimal level of coverage, evaluated at the optimal level of coverage, X*, with respect to q:



d 2U = (− p)uʹ(W − qX − L + X ) + P(1 − q)(− X )uʹʹ(W − qX − L + X ) (6.11) dXdq − [(1 − p)uʹ(W − qX ) − (1 − p)q(− X )uʹʹ(W − qX )]

d 2U = p(1 − q)uʹʹ(W − qX − L + X ) − (1− p)quʹʹ(W − qX ) , we can rearrange (6.11) Noting that dXdW to yield the following:



−[ puʹ(W − qX − L + X ) + (1 − p)uʹ(W − qW )] −

d 2U (6.12) dXdW

The bracketed term in (6.12) represents the substitution effect associated with an increase in q. The effect on coverage demanded is negative due to the increase in the price of insurance. With insurance now relatively more expensive, an individual can be made better off by reducing the amount of coverage purchased in order to buy other goods. The second term in (6.12) represents the income effect of the increase in q, as a higher price for coverage would decrease overall wealth, ceteris paribus.

98  Handbook on the economics of disasters Assuming a positive level of coverage purchased, the income effect will have the opposite d 2U . Accordingly, we see that the sign of this effect will depend on an individual’s sign of dXdW attitude toward risk. For an individual with DARA preferences, the reduction in wealth due to the price increase makes the individual more risk averse, leading to a positive income effect, as the price increase results in an increased purchase of coverage. If this positive income effect is sufficiently strong to outweigh the negative substitution effect, then insurance can be considered a Giffen good. We see that the income effect is nonpositive for individuals with IARA or CARA preferences, as the reduction in wealth associated with the increase in the price of insurance either makes the individual less risk averse (IARA preferences) or has no effect on the individual’s attitude toward risk (CARA preferences). In these cases, the second term will be nonpositive, and the overall effect of the increase in the price of coverage will be to reduce the amount of coverage purchased. See Schlesinger (2013) for a more thorough treatment of this subject in the context of loading factors and varying degrees of coinsurance. 3.1.2  Empirical evidence In contrast to the wide and active literature on risk preference and insurance in risk settings outside of natural disasters—for example, health (Kairies-Schwarz et al., 2017), retirement savings (Einav et al., 2012), and home and auto insurance (Barseghyan et al., 2013; Cohen and Einav, 2007)—a small but important strand of literature examines the impact of risk preferences on demand for natural disaster insurance. The main results are consistent with the hypotheses generated from our model of insurance demand presented earlier. Attanasi and Karlinger (1979) provide both theoretical and empirical evidence on disaster insurance and risk preferences using data from four towns in New Jersey. They find that as risk aversion increases, demand for insurance shifts outward and becomes more price-inelastic, implying that optimal insurance coverage increases are conditional on a given premium. In their sample, they observe average risk preferences parameters across townships consistent with CARA preferences. In addition, Petrolia et al. (2013) examine household demand for flood insurance coupled with experimental-based estimates of risk preferences across a sample of US residents across the Gulf Coast and Florida. They find that risk-averse individuals are significantly more likely to have flood insurance policies. In an agricultural context, Jianjun et al. (2015) and Jin et al. (2016) both find that among their sample of farmers in Yongqian, China, the average farmer was risk averse, and the level of risk aversion was significantly and positively related to the purchase of weather-indexed crop insurance. Lastly, Hellerstein et al. (2013) use a sample of 68 farmers from the US Corn Belt (Indiana, Ohio, and Iowa) to examine the impact of risk preferences in farming decisions, including insurance. They find the unexpected result that risk aversion is negatively related to the likelihood that farmers have crop insurance policies. However, they explain this not as a contradiction to the theory but rather as a consequence of the lottery-choice experiment that they used to elicit risk preferences in the lab.2 They conclude that this commonly used laboratory measure of risk preferences may not perform well as a proxy for real-world preferences. Table 6.1 displays a summary of the theoretical and empirical results surrounding the quantity of insurance purchased across moderating variables, as well as key research opportunities. Despite the theoretical rationale, a large amount of literature examines why farmers purchase below-optimal levels of insurance, even in cases when it is priced far below the actuarially fair rates and therefore should appeal to a wide range of risk preferences. Alternative products

Risk preferences and natural disasters  99 Table 6.1  Summary of theoretical and empirical insurance results Variable

Theoretical Direction

Empirical Direction

Risk Aversion

+

+

Wealth, DARA Preferences

−

−, RO

Wealth, IARA Preferences

+

+, RO

Premium Price, DARA Preferences

+

+, RO

Premium Price, IARA Preferences

−

−, RO

Note:  Table 6.1 displays the relationship (positive [+] or negative [-]) between each variable and the amount of insurance purchased. RO: Research opportunity given limited empirical evidence.

have been developed to increase insurance take-up among farmers, including in developing country contexts. Index insurance, for example, sets payouts based on easily observable metrics (e.g., rainfall levels or average yields in a location) rather than individual-level losses. This is especially appealing to small farms as it lowers the costs in verifying claims that might make traditional insurance premiums too costly (Basis, 2021). Microinsurance is another tool offering small policies to poor households at low cost (Janzen and Carter, 2019). The literature offers a variety of plausible explanations as to why low take-up occurs, despite innovations (Carter et al., 2014). Related to risk preferences, Clarke (2016) highlights that highly riskaverse individuals may not purchase insurance contracts if the basis risk—that is, the risk that the payout index may not correlate with farm-level conditions, leading to the possibility that farmers pay for insurance and have a bad crop but receive no payout—is high. In addition, other preferences, such as for time and ambiguity, and behavioral factors can explain why expected utility theory fails to predict insurance take-up (Clarke et al., 2012). Lastly, risk preferences may not be known to the policy maker or researcher, thereby limiting evaluation of the optimal level of insurance. Even when risk preferences are elicited, difficulty in estimation may lead to measurement error that distorts third-party evaluation of the optimality of decision-making. Regardless of the insurance type, care must be taken to overcome these challenges to insurance take-up, including those driven by risk preferences. 3.2  Location Choice Since Tiebout (1956), researchers have been interested in how individuals sort over heterogeneous local (dis)amenities, including a growing literature on sorting and migration in response to natural disaster risk and events (Boustan et al., 2012; Hornbeck, 2012). A growing literature examines sorting over climatological natural disaster risks such as floods (Bakkensen and Ma, 2020), hurricanes (Fan and Bakkensen, 2021), sea level rise (Bakkensen and Barrage, 2022), and other types of extreme weather (Fan et al., 2018), as well as local disaster mitigation efforts (Fan and Davlasheridze, 2016). A parallel literature examines housing market impacts from disaster events.3 Broadly, the literature finds heterogeneity in disaster risk beliefs characterized by a majority of residents underestimating the true probability of event occurrence (Bakkensen and Barrage, 2022; Bernstein et al., 2019) that leads to volatility in housing market prices following events as disaster risk salience increases (Bakkensen et al., 2019; Bin and Landry, 2013; McCoy and Walsh, 2018). A comparative dearth of empirical

100  Handbook on the economics of disasters literature examines how risk preferences impact location choice surrounding both long-run disaster risk as well as disaster events, the latter of which can lead to risk belief updating if individuals were not perfectly attentive (Gallagher, 2014). In this subsection, we examine the theoretical rationale for why risk preferences may matter in location-choice decisions as well as review the empirical literature. 3.2.1  Location choice as an application of portfolio theory The risk of damage from natural disasters varies across space, with certain locations being more at risk to damages from tropical cyclones (e.g., coastal properties), earthquakes (e.g., properties on fault lines), and wildfires (e.g., properties at the wildland-urban interface). In making their decision about where to purchase property, individuals have an opportunity to manage their level of risk from natural disasters. While the papers mentioned previously have presented evidence of changes in prices and beliefs in the wake of natural disasters, they have tended to do so without accounting for the impact of risk preferences on these outcomes. In this subsection, we use the framework of portfolio theory to explore how people’s attitudes toward risk, in conjunction with their preferences for the amenities associated with property purchase, will affect the composition of communities in areas at risk from natural disasters. The results here are motivated by descriptions of consumer choice across financial assets with varying degrees of risk (see, e.g., Deaton and Muellbauer, 1980; Kreps, 1990). Consider an individual facing a decision about how to allocate his wealth, W, between the purchase of property in two locations. Property purchased in the safe location provides a constant, known stream of benefits, r. Property purchased in the at-risk location offers amenities beyond those available in the safe location (e.g., coastal living, serenity in remote locations, etc.) but also faces the risk of damages from natural disasters. Let the random variable z with CDF F reflect the stream of benefits associated with property purchased in the at-risk location. Assume that the individual’s utility is described by U(), an increasing and concave utility function. Given his initial wealth of W, the individual is capable of purchasing X units of property in the at-risk location and W – X units of property in the safe location.4 Then, the individual’s objective function is given by

max ∫ U ( Xz + (W − X )r) dF(z) (6.13) X

This objective function leads to the following first-order condition:

∫ U ʹ( Xz + (W − X )r)(z − r) dF(z) = 0 (6.14)

If we begin by considering a risk-neutral individual, meaning that U(X) = αX for some constant a, we see that the returns to the purchase of property are given by α[Wr + X(E[z] – r)]. This quantity will either always be positive (E[z] > r) or always negative. From the perspective of a risk-neutral individual, the purchase of property is determined purely by the expected rate of return, so that he will put all of his wealth toward the purchase of property in the at-risk location (if E[z] > r) or in the safe location. Now, if we assume the individual is risk averse, so that U′() < 0, then the first-order condition shown in (6.14) is necessary and sufficient to identify the optimal quantity of property purchased. In this case, we see that a risk-averse investor will allocate some of his wealth to

Risk preferences and natural disasters  101 the purchase of property in the at-risk location, as long as property in the at-risk location has a positive return. Another way to think about this problem is to split the returns to property in the at-risk location into two components. The first piece is the nonmonetary amenity value of living in the at-risk location, ϕ(i), which we can allow to vary across individuals, indexed by i. The second piece is the potential damages caused by a natural disaster striking the property in the at-risk location, which we leave to be denoted by random variable z with CDF F. With this modification, the individual’s objective function becomes

max ∫ U ( Xz + (W − X )r) dF(z) + v(φ (i) X ) (6.15) X

This new objective function leads to the following first-order condition:

∫ U ʹ( Xz + (W − X )r)(z − r) dF(z) + vʹ(φ (i) X )φ (i) = 0 (6.16)

This leads to the following condition defining the optimal purchase of property in the at-risk location (assuming X > 0):

− ∫ U ʹ( Xz + (W − X )r)(z − r) dF(z) = vʹ(φ (i) X )φ (i) (6.17)

The left-hand side of (6.17) represents the marginal cost of an additional unit of property in the at-risk location. Now that we are letting z represent the damages from a natural disaster, these expected damages might be less than r. The right-hand side of (6.17) shows the increase in utility, due to increased nonmonetary amenities, associated with the purchase of an additional unit of property in the at-risk location. Within this modified framework, we see that the purchase of property in the at-risk location is dependent on an individual’s attitude toward risk, as well as his appreciation of the amenities associated with living in the at-risk location. So, for individuals who do not value these amenities (i.e., φ(i) = 0) the risk of damages from natural disasters may be sufficient to preclude any purchase of property in the at-risk location. On the other hand, individuals with very strong preferences for these amenities may decide to purchase property uniquely in the at-risk location, despite their attitude toward risk. Having established a framework for thinking about an individual’s location-choice problem, we can now turn to the question of how an individual’s attitude toward risk impacts the decision about how much property in the at-risk location to purchase. Consider individuals a and b who both face the decision about how to allocate their wealth across property in the at-risk and safe locations. Their objective functions are given by the following:

max ∫ U a ( Xz + (W − X )r) dF(z) + v(φ (i) X ) (6.18)



max ∫ U b ( Xz + (W − X )r) dF(z) + v(φ (i) X ) (6.19)

X

X

102  Handbook on the economics of disasters To simplify things, we assume that the two individuals have the same preferences for the nonmonetary amenities of life in the at-risk location, so we focus our attention on the first term in each objective function. Then, a sufficient condition for individual b to invest more wealth than a in property in the at-risk location is as follows:

∫ U aʹ ( Xz + (W − X )r)(z − r) dF(z) = 0 ⇒ ∫ U bʹ ( Xz + (W − X )r)(z − r) dF(z) ≥ 0 (6.20) This condition could follow from the concavity of Ub. If a is more risk averse than b, then Ub can be expressed as a function of Ua and a nondecreasing convex function h (i.e., Ub = h ○ Ua). Then, (6.19) can be rewritten as follows:

∫ hʹ(u( Xz + (W − X )r))uʹ( Xz + (W − X )r)(z − r) dF(z) ≥ 0 (6.21)

It turns out that (6.21) will always hold. The first term h’() is positive and increasing in z. The second set of terms is negative when z < r and positive when z > r, so multiplying the second set of terms by h’() places more weight on draws for which z > r. The result of this fact is that the product of these two terms h’() and u’() will integrate to a greater amount than the value of the second term alone, or 0. So, for two individuals with the same preferences for the nonmonetary amenities of living in the at-risk location, with one individual more risk averse than the other, it turns out that the more risk-averse individual will optimally purchase less property in the at-risk location than the other individual, for any initial wealth level. Given the preceding result, it is natural to inquire about how the amount of property purchased in the at-risk location by a risk-averse individual will change in response to an increase in wealth. To do so, we turn to the comparative statics of the first-order condition for our location-choice problem presented as an application of portfolio theory. Then, we want to d 2U ≥ 0 . We see that know if dXdW

d 2U = dXdW



∫ rU ʹʹ( X (z − r) + Wr)(z − r) dF(z) (6.22)

which we can rewrite as

d 2U = dXdW

U ʹʹ( X (z − r) + Wr)

∫ r U ʹ( X (z − r) + Wr)U ʹ( X (z − r) + Wr)(z − r)dF(z) (6.23)

We see that, as before, the second term is negative when z < r and positive when z > r. Now, the first term is negative. So, the preceding relation will hold when the first term is decreasing in z. This condition is the definition of DARA preferences. So, we see that the amount of property purchased in the at-risk location at higher levels of wealth will increase for individuals with DARA preferences, remain constant for individuals with CARA preferences, and fall for individuals with IARA preferences. 3.2.2  Empirical evidence The theoretical underpinnings of heterogeneous risk preferences and location choice are clear. Indeed, Sheldon and Zhan (2019) note (although do not empirically explore) that individuals

Risk preferences and natural disasters  103 Table 6.2  Summary of theoretical and empirical location choice results Variable

Theoretical Direction

Empirical Direction

Risk Aversion

−

−

Wealth, DARA Preferences

+

RO

Wealth, IARA Preferences

−

RO

Note:  Table 6.2 displays the relationship (positive [+] or negative [−]) between each variable and the amount of property purchased in a risky location. RO: Research opportunity given limited empirical evidence.

sort based on preferences for disaster risk, leading risk-averse households to be more likely to live in safer areas. A comparatively small amount of literature explores empirical evidence of sorting over natural disaster risk based on risk preferences. However, clear evidence in related applications highlights the likelihood of this phenomena, including in the labor sector. Bellemare and Shearer (2010) find that workers in British Columbia who face more significant daily income risk in their jobs are more risk tolerant than the average population, indicative of sorting over occupational risk. Similar sorting over risk preferences in labor markets is found by Bonin et al. (2007) in the case of German occupational sectors and by DeLeire and Levy (2004) in the case of on-the-job mortality risk in the United States. Risk preferences also matter in deciding to migrate. Jaeger et al. (2010) find a negative relationship between risk aversion and likelihood of migration in Germany. In the case of Indonesia and Ghana, Goldbach and Schlüter (2018) also find migrants are less likely to be risk averse relative to nonmigrants. Similarly, Arcand and Mbaye (2013) find that risk aversion is negatively correlated with willingness to pay for illegal migration in Senegal. However, a fruitful area of future research could expand empirical tests of these observations in a natural disaster context. Table 6.2 displays a summary of the theoretical and empirical results surrounding high-risk, location-choice decisions across key moderating variables, as well as key research opportunities. 3.3 Adaptation We lastly consider the important case of adaptation as a final lens to examine risk preferences and natural disasters. Adaptation to natural disaster is a key private mechanism and policy lever to increase resilience and decrease disaster harm. Adaptation to natural disasters can be broadly conceptualized as activities or adjustments in response to current or future disaster risks in order to reduce harm (Field et al., 2012).5 Indeed, adaptation can take many forms across both private and public actors, including physical strengthening and public works interventions, natural capital, as well as evacuations and migration (Bakkensen and Blair, 2021).6 The effectiveness of adaptation in reducing natural disaster losses remains an active area of research (Bakkensen and Mendelsohn, 2016); however, it is commonly found that individuals tend to be underprepared for natural disasters (Meyer and Kunreuther, 2017), leaving them more vulnerable to disaster impacts than is efficient. An active research area examines how risk preferences impact adoption of natural disaster adaptation technologies, especially in agricultural risk settings. This subsection provides a theoretical motivation as to why risk preferences might matter for adaptation decisions and follows with a review of relevant themes in the literature.

104  Handbook on the economics of disasters 3.3.1  Model of risk preferences and adaptation We define adaptation as any activities to reduce the expected damages from a disaster, conditional on a disaster event occurring. For the purposes of this chapter, we abstract away from the decision of optimal levels of insurance versus adaptation versus location choice. Lewis and Nickerson (1989) provide a theoretical treatment of optimal adaptation (called “selfinsurance”) to natural disasters in the context of other factors such as the level of risk aversion, uncertainty, post-disaster aid, and individual wealth. However, the interaction between optimal adaptation, insurance, and location choice remains an open area of future research. To explore the impact of risk aversion on adaptation decisions, assume an individual with wealth, w, is at risk of a natural disaster with probability p. She can install adaptive technology, A, at cost c per unit of A. If a disaster hits with no adaptation, it will lead to disaster losses equal to L. If adaptation is employed, it will lead to losses equal to L – A, where losses L are reduced from their no-adaptation baseline by magnitude A. We also assume utility is concave. Thus, if there is no adaptation, expected utility of the individual is

E(U | A = 0) = (1 − p)U (w) + pU (w − L) (6.24)



If the individual has adaptation, expected utility is instead

E(U | A = 1) = (1 − p)U (w − cA) + pU (w − L + A − cA) (6.25)

The individual’s objective to maximize expected utility is thus

max E(U ) = (1 − p )U ( w − cA) + pU (w − L + A − cA) (6.26) A

The first-order condition is thus

−c(1 − p)U ʹ(w − cA) + (1 − c) pU ʹ(w − L + A − cA) = 0 (6.27)

Rearranging, it appears as

(1 − c) pU ʹ(w − L + A − cA) = c(1 − p)U ʹ(w − cA) (6.28)

Assuming that the individual is risk averse and that c ≤ p, the individual will fully adapt until L = A. This ensures that her utility is constant regardless of the state of the world. Similar to the insurance case, a risk-neutral individual would be indifferent to full adaptation or no adaptation if c = p. This finding is highlighted in previous literature, including that the risk premium increases in the level of risk aversion (Pratt, 1978).7 However, if adaptation leads to greater reductions in expected losses than its cost, then a risk-neutral or even risk-loving individual may still engage in adaptation to reduce expected losses if the technology were effective enough relative to its cost. Empirical literature has highlighted that many such adaptive technologies exist. However, while not formally derived here, the optimal level of adaptation when insurance is available is an important area of active research, including how insurance policies can incentivize adaptation if premium rates are reduced to account for lower expected losses following adaptation (Hudson et al., 2016).

Risk preferences and natural disasters  105 Unlike insurance, adaptive technologies have some probability of failure (e.g., hurricane shutters can break, levees can be breached). In addition, some newer technologies may have greater uncertainty regarding their performance or failure rates. Assume with some probability, π, the new adaptive technology, An, will fail. Expected utility in the case of no adaptation remains the same:

E(U | An = 0) = (1 − p)U (w) + pU (w − L) (6.29)



However, expected utility in the case of the risky adaptation is

E(U | An = 1) = (1 − p)U (w − cAn ) + p(1 − π )U (w − L + An − cAn ) + p( π )U (w − L − cAn ).

(6.30)

The individual’s objective function thus becomes

max E(U ) = (1 – p )U ( w – cAn ) + p (1 – π )U ( w – L + An – cAn ) An

+ p ( π )U ( w – L – cAn ) ,

(6.31)

with a first-order condition of

(1 − p)(−c)U ʹ(w − cAn ) + p(1 − π )(1 − c)U ʹ(w − L + An − cAn ) + p( π )(1 − c)U ʹ(w − L − cAn ) = 0.

(6.32)

Rearranged, this becomes

(1 − p)(c)U ʹ(w − cAn ) = p(1 − π )(1 − c)U ʹ(w − L + An − cAn ) + p( π )(1 − c)U ʹ(w − L + An − cAn ).

(6.33)

Compared with the risky technology and all else equal, all individuals would prefer the sure option regardless of risk preferences. However, new risky technologies often have a higher benefit from utilization. Thus, there would exist some minimum higher level of adaptation efficacy, An, where An > A, that would entice individuals to use the riskier technology. Given their risk preferences, someone with a higher degree of risk aversion would need a higher effectiveness of the technology to overcome the reduction in utility from the adaptation technology’s inherent riskiness. Thus, all else equal, we can expect to see heterogeneity in utilization of a new high risk, high reward adaption technology across user risk preferences, with more risk-averse individuals holding out adoption (i.e., preferring proven but lower return technologies) unless either the benefit of the adaptation is clearly high enough or the riskiness is low enough. 3.3.2  Empirical evidence Relative to insurance and location choice, a comparatively large and active literature examines the impact of risk preferences on the adoption of adaptation technologies and other risk reduction strategies, especially in agricultural settings. Consistent with our simple stylized

106  Handbook on the economics of disasters model presented previously, Bozzola (2014) finds empirical evidence in the case of Italian cereal producers that individuals with higher aversion to downside risk are more willing to adopt irrigation and other adaptive technologies to reduce profit variance and/or downside risk exposure, even if some profit is sacrificed.8 Similarly, Asravor (2019) finds that in Northern Ghana, risk-averse farmers are significantly more likely to have greater crop diversification. As highlighted by our theoretical model earlier, risk aversion does not always correlate with higher levels of adaptation technology adoption. Liu (2013) examines how risk preferences across cotton farmers in China impact the adoption of the newer, genetically modified Bt cotton that promised higher yields and less loss from pests. She finds that individuals with higher levels of risk or loss aversion are slower to adopt the Bt cotton, highlighting that uncertainty surrounding technological performance will encourage more risk-loving individuals to take the gamble on a newer or less-proven adaptation technology. In a follow-up paper, Liu and Huang (2013) find that risk-averse farmers are more likely to use higher amounts of pesticides to reduce crop loss from pests. However, loss-averse individuals used less quantities, prioritizing losses to health from pesticides over the financial losses from the increased consequences of pests. Connecting adaptation with insurance, as expected by theory and the previous findings, Jianjun et al. (2015) find that risk-averse Chinese farmers were less likely to adopt climate innovations, such as new technologies or crop types, and were instead significantly more likely to purchase weather index crop insurance. Using the case of South African farmers, Brick and Visser (2015) also find that risk-averse individuals are more likely to use traditional agricultural products instead of newer high-yield varieties that necessitate financing even despite insurance availability, highlighting that insurance is not necessarily a cure-all to increase technological adoption and combat poverty traps. However, some countervailing findings exist. Ross et al. (2012) show that among rice farmers in Laos, ambiguity aversion and not risk aversion inhibits the adoption of new technologies despite the potential for higher profits, highlighting the importance of controlling for other correlated factors. Similarly, He et al. (2020) use the case of Chinese farmers and find that while risk aversion is significantly related to adaptation cognition, it does not have a significant impact on adaptive behaviors. Instead, they find loss aversion explains risk and adaptation cognition as well as adaptation. Vieider et al. (2019) explore evidence amongst Vietnamese farmers, finding that risk aversion does not necessarily reduce new technology adoption. They note that their sample is closer to risk neutral than typical Western subject pools, thus highlighting the importance of context in analyzing risk decisions. Finally, Vieider et al. (2019) also find that risk aversion does not correlate with wealth (although it does correlate with income). This is in contrast to existing literature that finds risk aversion can be a key mechanism in the poverty trap where poorer individuals may also be risk averse and thus less likely to adopt new innovations that could raise them from poverty (Giné and Yang, 2009; Mosley and Verschoor, 2005). Outside of agriculture, de Blasio et al. (2020) examine risk preferences among Italian households, finding that increased risk aversion following a disaster event inhibits entrepreneurship among Italian households. Wibbenmeyer et al. (2013) examine how risk preferences impact fire management and suppression strategies among US federal wildfire managers. They find overall risk decisions are better modeled on nonexpected utility decision-making as managers tend to over-allocate resources when the probability or likely magnitude of fires is low rather than minimizing expected losses. Table 6.3 displays a summary of the theoretical and empirical results surrounding the quantity of adaptive technology employed across moderating variables.

Risk preferences and natural disasters  107 Table 6.3  Summary of theoretical and empirical insurance results Variable

Theoretical Direction

Empirical Direction

Risk Aversion

+

+

High Risk, High Return Technology

−

−

Insurance

−

−, 0

Note:  Table 6.3 displays the relationship (positive [+], negative [−], or no relationship [0]) between each variable and the amount of adaptation.

3.4  Risk Preference Stability As explored in Section 2.3, researchers long believed preferences to be permanent, including Stigler and Becker’s famous paper “De gustibus non est disputandum” (1977), meaning “in matters of tastes, there can be no disputes.” More recently, however, a growing body of theoretical and empirical evidence has shown that risk preferences may be malleable given a large enough shock (Schildberg-Hörisch, 2018). If true, this can have significant implications for welfare. If monumental negative events lead to reductions in risk tolerance, this can lead to adoption of less-risky behaviors that, in some cases, may be suboptimal, including in investment decisions, reduced entrepreneurship, increased skepticism of new technologies, and distortions in migrations decisions (Schildberg-Hörisch, 2018).9 As suggestive evidence, Falk et al. (2018) find that countries with higher levels of risk aversion also have lower total factor productivity. Empirical findings highlight that tail-event natural disasters are often sufficient shocks to observably impact risk preferences.10 In addition, given the arguably exogenous nature of their realizations, natural disasters are often used as natural experiments in identifying the impacts of changes in risk preference. For example, Cassar et al. (2017) examine the case of the 2004 tsunami in Thailand, finding that individuals from villages hardest hit by the tsunami had long-lasting increases in risk aversion as well as prosocial behavior and impatience. Bchir et al. (2013) find that a massive earthquake in Peru led to an increase in risk aversion. They then exploit the earthquake as an instrumental variable to overcome the potential endogeneity of risk preferences and labor sector to examine the arguably causal effect of risk aversion on entrepreneurship. Bourdeau-Brien and Kryzanowski (2020) find that extreme weather events in the United States led to significant increases in risk aversion, which, in turn, had consequences for investors in municipal bonds. They also highlight that these changes in investment behavior led to larger macroeconomic and fiscal consequences as it reduced the effectiveness of post-disaster aid and slowed disaster recovery. In turn, Chuang and Schechter (2015) conduct a thorough literature review on risk preferences after natural disasters, highlighting some disagreement in empirical literature findings, with some studies finding post-disaster increases in risk aversion (e.g., Cameron and Shah, 2015), decreases in risk aversion (e.g., Eckel et al., 2009), and no change in risk aversion (e.g., Becchetti et al., 2012). Given the occasionally contradictory findings, additional literature has parsed these conclusions through examination of mediating response channels that can explain how and why risk preferences shift, although a comprehensive theory is still an active area of research. Using the

108  Handbook on the economics of disasters case of a hypothetical disaster shock on specialty producers in Indiana, Wahdat et al. (2021) find income to be a key risk aversion mechanism: farmers who receive a randomly assigned (hypothetical) disaster shock have an increase in absolute (but not relative) risk aversion risk premium relative to the nontreated group. Brown et al. (2018) use the case of the December 2012 Cyclone Evan that devastated Fiji, finding the event increased subjective risk perception and levels of risk aversion, but only for the Indo-Fijian residents and not Indigenous Fijian residents. They propose moderating mechanisms: Indigenous Fijians have a comparatively more collectivist social structure, and these stronger social networks helped to better insulate them from the disaster shock, thus leading to insignificant changes in risk aversion following the event. Finally, Shupp et al. (2017) use a combination of survey and experimental approaches to study how risk, loss, and ambiguity aversion changed after a large 2013 tornado in Oklahoma City, Oklahoma. They find heterogeneity in impacts: individuals who were injured, themselves, had an increase in risk aversion. However, individuals who knew a friend or neighbor who perished became less risk averse, highlighting that the impact of disasters on risk preferences can be different and need to be carefully studied. Instead of looking at the impact of disaster events on risk preferences, a second strand of literature examines the correlation between risk preferences and background disaster risk. As Gollier and Pratt (1996) show theoretically (called “risk vulnerability”) and Lee (2008) shows empirically, even in the absence of an event, background risk can make risk-averse individuals more risk averse. In the context of natural disasters, Bchir et al. (2013) find a positive correlation between the background risk of lahars (volcanic landslides) in Peru and risk seeking, but the relationship is mitigated among higher-income individuals. Thus, while a preponderance of the literature has focused on the relationship between disaster strikes and risk preferences, a fruitful area of future inquiry could expand the understanding and empirical identification of the endogenous relationship between disaster risk and risk preferences, as individuals may sort over natural disaster risk based on their risk preferences but also have preferences shaped by disaster events and even underlying disaster risk in risky areas. This has important implications for how to define a counterfactual empirical control group as groups in, for example, low-risk or less-impacted regions may be ex ante different from impacted groups.

4. CONCLUSION Natural disasters lead to devastating losses to humans around the globe, and these impacts are expected to increase with upcoming changes to climate and human communities (Field et al., 2012). Thus, it is critical to understand why and how humans cope with these risks, as well as how response patterns may differ based on context and preferences. While comparatively more is known about the impact of risk perception and salience on decision-making surrounding natural disasters, much less is known about how risk preferences may moderate decisionmaking surrounding natural disasters. This chapter summarizes key themes and knowledge gaps through a review of the relevant theoretical and empirical literature. After providing a brief summary of the theory and empirical methods behind risk preferences, we examine three disaster-relevant cases: insurance, location choice, and adaptation decisions. For each, we provide both theoretical and empirical motivation for the relevance of a risk preference lens in each context, as well as highlight key gaps, including in the empirical literature. The difficulty in eliciting risk preferences in survey and real-world data—given that it is not typically

Risk preferences and natural disasters  109 included in many experimental and administrative data sets due in part to the potential limitations of hypothetical risk decisions or real payoffs that are trivially small as well as the high cost to implement a large lottery payment across a big sample—is a hindrance to expanding empirical evidence. Finally, we examine the stability of risk preferences, including the recent evidence that these preferences may be impacted by natural disaster events. Risk preferences also hold key implications for natural disaster policy. Risk preferences constitute another form of heterogeneity across individuals in society and help to explain why otherwise similar individuals may make different decisions surrounding management of disaster risk mitigation and recovery. Thus, understanding how risk preferences impact decision-making across a variety of settings can allow policy makers to craft better policy to achieve policy goals. From an insurance perspective, risk-loving individuals would need additional incentive to take up a policy relative to risk-averse individuals, so tools such as insurance mandates or price regulation can help increase insurance adoption and may explain, in part, why disaster insurance take-up remains low in some countries, such as the United States, despite a history of insurance premiums set at below actuarially fair rates. Coupling the findings from location-choice and adaptation theory, which show that riskloving individuals are more likely to live in risky areas and less likely to adapt, ceteris paribus, risk preferences highlight an additional reason why individuals in harm’s way may be less prepared for natural disasters. This can motivate policy intervention from both a paternalistic perspective to encourage additional protection, and also an externality perspective to mitigate the broader societal costs of natural disasters that may be larger due to lower levels of individual protection. The findings that risk preferences may not be stable offers an additional avenue for policy levers. This highlights how natural disasters can be windows of opportunity to achieve policy goals to, for example, increase adaptation, insurance, or population in lower risk settings, when society may be more risk averse. In addition, public support for disaster policy reform may also be greater following a significant disaster due to even temporary changes in levels of risk aversion in society. A key obstacle that policy makers must overcome to better leverage risk preferences as a tool remains lack of data on individuals’ risk preferences. Indeed, few administrative surveys contain risk preference elicitation questions. Even a simple hypothetical risk preference lottery embedded in existing administrative survey instruments could lead to valuable information for policy makers to better craft effective disaster policy. In addition, given that risk preferences are often correlated with other individual characteristics (e.g., risk perception, preferences for other (dis)amenities, time preferences, cognition), disentangling the role of risk preferences versus other factors will be important for future research to better ascertain the causal mechanisms as well as provide better policy recommendations. Natural disasters have long imposed terrible hardship and damage around the world. Better understanding the arguably understudied role of risk preferences is thus critical in mitigating disaster losses and improving societal resilience that will have consequences for decades to come.

NOTES 1. These statistics were all provided by Munich Re, Geo Risks Research, NatCatSERVICE, 2021, and were accessed via https://www.iii.org/fact-statistic/facts-statistics-global-catastrophes. 2. This result is based on a coarser measure of risk preferences. Using a finer measure, they find no relationship between lab-elicited risk preferences and real-world insurance take-up.

110  Handbook on the economics of disasters 3. See review of how, for example, flood risk is capitalized in house prices by Beltrán et al. (2018). 4. Note that, in this setup, the individual will own property in both locations. This outcome may not be consistent with the experience of most individuals in practice. It is possible to separate those who buy property in the at-risk locations from those who buy property in the safe location by introducing an individual-specific enjoyment of the amenities in the at-risk location; however, this comes at some cost of distraction from the main intention of this section. See Conte and Kelly (2018) for details. 5. Adaptation can also be used to amplify benefits, but we focus on harm reduction here. 6. Insurance is also considered adaptation, but we treat the two separately in this chapter (Thomas and Leichenko, 2011). 7. We note that while this relationship holds for the risk premium, this property does not always hold for willingness to pay for risk reduction across risk preferences (Eeckhoudt et al., 1997; Langlais, 2005). 8. Downside risk is a risk that can lead to losses greater than the critical threshold even if there is potential for gains (Menezes et al., 1980). 9. Of course, if individuals were under-perceiving the risks of natural disasters, they may have been taking on a suboptimally low level of preparedness in advance of a disaster. Therefore, in some cases, the disaster shock could lead to decisions closer to the optimal if individuals then take on less risky endeavors. The level of ex ante risk perception is therefore an important ingredient in the welfare impacts. 10. A parallel literature similarly has found that war and large-scale violence are significant shocks sufficient to impact risk preferences (e.g., Callen et al., 2014).

REFERENCES Allais, M. (1953). Le comportement de 1’homme rationnel devant le risque: critique des postulats et axiomes de l’école américaine. Econometrica: Journal of the Econometric Society, 21(4):503–546. Andersen, S., Harrison, G. W., Lau, M. I., and Elisabet Rutström, E. (2008). Lost in state space: Are preferences stable? International Economic Review, 49(3):1091–1112. Andreoni, J., and Sprenger, C. (2010). Certain and uncertain utility: The allais paradox and five decision theory phenomena. Working Paper. Arcand, J.-L. and Mbaye, L. (2013). Braving the waves: The role of time and risk preferences in illegal migration from Senegal. IZA Discussion Paper No. 7517. Arrow, K. J. (1965). Aspects of a theory of risk bearing, Yrjo Jahnsson lectures, Helsinki. Reprinted in essays in the theory of risk bearing (1971). Asravor, R. K. (2019). Farmers’ risk preference and the adoption of risk management strategies in Northern Ghana. Journal of Environmental Planning and Management, 62(5):881–900. Attanasi, E. D., and Karlinger, M. R. (1979). Risk preferences and flood insurance. American Journal of Agricultural Economics, 61(3):490–495. Bakkensen, L., and Blair, L. (2021, under review). The economics of tropical cyclone adaptation. In Handbook of behavioural economics and climate change. Edward Elgar Publishing. Bakkensen, L. A., and Barrage, L. (2022). Flood risk belief heterogeneity and coastal home price dynamics: Going under water? Review of Financial Studies. https://doi.org/10.1093/rfs/hhab122. Bakkensen, L. A., Ding, X., and Ma, L. (2019). Flood risk and salience: New evidence from the sunshine state. Southern Economic Journal, 85(4):1132–1158. Bakkensen, L. A., and Ma, L. (2020). Sorting over flood risk and implications for policy reform. Journal of Environmental Economics and Management, 104:102362. Bakkensen, L. A., and Mendelsohn, R. O. (2016). Risk and adaptation: Evidence from global hurricane damages and fatalities. Journal of the Association of Environmental and Resource Economists, 3(3):555–587. Barseghyan, L., Molinari, F., O’Donoghue, T., and Teitelbaum, J. C. (2013). The nature of risk preferences: Evidence from insurance choices. American Economic Review, 103(6):2499–2529. Basis. (2021). How agricultural index insurance can promote risk management and resilience in developing countries. https://basis.ucdavis.edu/agricultural-index-insurance-economic-development. Bchir, M. A., and Willinger, M. (2013). Does the exposure to natural hazards affect risk and time preferences? Some insights from a field experiment in Perú. Unpublished manuscript. Becchetti, L., Castriota, S., and Gonzo, P. (2012). Calamity, aid and indirect reciprocity: The long run impact of tsunami on altruism. CEIS Tor Vergata, research paper series No. 239. Bellemare, C., and Shearer, B. (2010). Sorting, incentives and risk preferences: Evidence from a field experiment. Economics Letters, 108(3):345–348. Beltrán, A., Maddison, D., and Elliott, R. J. (2018). Is flood risk capitalised into property values? Ecological Economics, 146:668–685.

Risk preferences and natural disasters  111 Bernstein, A., Gustafson, M. T., and Lewis, R. (2019). Disaster on the horizon: The price effect of sea level rise. Journal of Financial Economics, 134(2):253–272. Bin, O., and Landry, C. E. (2013). Changes in implicit flood risk premiums: Empirical evidence from the housing market. Journal of Environmental Economics and Management, 65(3):361–376. Bonin, H., Dohmen, T., Falk, A., Huffman, D., and Sunde, U. (2007). Cross-sectional earnings risk and occupational sorting: The role of risk attitudes. Labour Economics, 14(6):926–937. Bourdeau-Brien, M., and Kryzanowski, L. (2020). Natural disasters and risk aversion. Journal of Economic Behavior & Organization, 177:818–835. Boustan, L. P., Kahn, M. E., and Rhode, P. W. (2012). Moving to higher ground: Migration response to natural disasters in the early twentieth century. American Economic Review, 102(3):238. Bozzola, M. (2014). Adaptation to climate change: Farmers’ risk preferences and the role of irrigation. Technical report. Brick, K., and Visser, M. (2015). Risk preferences, technology adoption and insurance uptake: A framed experiment. Journal of Economic Behavior & Organization, 118:383–396. Brown, P., Daigneault, A. J., Tjernström, E., and Zou, W. (2018). Natural disasters, social protection, and risk perceptions. World Development, 104:310–325. Callen, M., Isaqzadeh, M., Long, J. D., and Sprenger, C. (2014). Violence and risk preference: Experimental evidence from Afghanistan. American Economic Review, 104(l):123–148. Cameron, L., and Shah, M. (2015). Risk-taking behavior in the wake of natural disasters. Journal of Human Resources, 50(2):484–515. Caplan, B. (2003). Stigler-Becker versus Myers-Briggs: Why preference-based explanations are scientifically meaningful and empirically important. Journal of Economic Behavior & Organization, 50(4):391–405. Carter, M., De Janvry, A., Sadoulet, E., Sarris, A., et al. (2014). Index-based weather insurance for developing countries: A review of evidence and a set of propositions for up-scaling. Fondation pour les études et recherches sur le développement international. Development Policies Working Paper, 111. Cassar, A., Healy, A., and Von Kessler, C. (2017). Trust, risk, and time preferences after a natural disaster: Experimental evidence from Thailand. World Development, 94:90–105. Charness, G., Gneezy, U., and Imas, A. (2013). Experimental methods: Eliciting risk preferences. Journal of Economic Behavior & Organization, 87:43–51. Chuang, Y., and Schechter, L. (2015). Stability of experimental and survey measures of risk, time, and social preferences: A review and some new results. Journal of Development Economics, 117:151–170. Clarke, D. J. (2016). A theory of rational demand for index insurance. American Economic Journal: Microeconomics, 8(l):283–306. Clarke, D. J., Clarke, D., Mahul, O., Rao, K. N., and Verma, N. (2012). Weather based crop insurance in India. World Bank Policy Research Working Paper, No. 5985. Cohen, A., and Einav, L. (2007). Estimating risk preferences from deductible choice. American Economic Review, 97(3):745–788. Conte, M. N., and Kelly, D. L. (2018). An imperfect storm: Fat-tailed tropical cyclone damages, insurance, and climate policy. Journal of Environmental Economics and Management, 92:677–706. Conte, M. N., and Kelly, D. L. (2020). How beliefs change following a natural disaster: Evidence from wind insurance data. University of Miami Working Paper. Conte, M. N., and Kelly, D. L. (2021). Understanding the improbable: A survey of fat tails in environmental economics. Annual Review of Resource Economics, 3:1–23. Coronese, M., Lamperti, F., Keller, K., Chiaromonte, F., and Roventini, A. (2019). Evidence for sharp increase in the economic damages of extreme natural disasters. Proceedings of the National Academy of Sciences, 116(43):21450–21455. Dave, C., Eckel, C. C., Johnson, C. A., and Rojas, C. (2010). Eliciting risk preferences: When is simple better? Journal of Risk and Uncertainty, 41(3):219–243. Davis, D. D., and Holt, C. A. (1993). Experimental economics. Princeton University Press. de Blasio, G., De Paola, M., Poy, S., and Scoppa, V. (2020). Massive earthquakes, risk aversion, and entrepreneurship. Small Business Economics, 57(1):1–28. Deaton, A., and Muellbauer, J. (1980). Economics and consumer behavior. Cambridge University Press. DeLeire, T., and Levy, H. (2004). Worker sorting and the risk of death on the job. Journal of Labor Economics, 22(4):925–953. Desposato, S. (2015). Ethics and experiments: Problems and solutions for social scientists and policy professionals. Routledge. Dohmen, T., Falk, A., Huffman, D., and Sunde, U. (2010). Are risk aversion and impatience related to cognitive ability? American Economic Review, 100(3):1238–60. Drichoutis, A. C., and Lusk, J. L. (2016). What can multiple price lists really tell us about risk preferences? Journal of Risk and Uncertainty, 53(2–3):89–106.

112  Handbook on the economics of disasters Eckel, C. C., El-Gamal, M. A., and Wilson, R. K. (2009). Risk loving after the storm: A Bayesian-network study of Hurricane Katrina evacuees. Journal of Economic Behavior & Organization, 69(2):110–124. Eeckhoudt, L., Godfroid, P., and Gollier, C. (1997). Willingness to pay, the risk premium and risk aversion. Economics Letters, 55(3):355–360. Einav, L., Finkelstein, A., Pascu, L, and Cullen, M. R. (2012). How general are risk preferences? Choices under uncertainty in different domains. American Economic Review, 102(6):2606–38. Falk, A., Becker, A., Dohmen, T., Enke, B., Huffman, D., and Sunde, U. (2018). Global evidence on economic preferences. Quarterly Journal of Economics, 133(4):1645–1692. Fan, Q., and Bakkensen, L. (2021). Household sorting as adaptation to hurricane risk in the United States. Land Economics. doi: 10.3368/le.98.2.111319-0162R1. Fan, Q., and Davlasheridze, M. (2016). Flood risk, flood mitigation, and location choice: Evaluating the national flood insurance program’s community rating system. Risk Analysis, 36(6):1125–1147. Fan, Q., Fisher-Vanden, K., and Klaiber, H. A. (2018). Climate change, migration, and regional economic impacts in the United States. Journal of the Association of Environmental and Resource Economists, 5(3):643–671. Field, C. B., Barros, V., Stocker, T. F., and Dahe, Q. (2012). Managing the risks of extreme events and disasters to advance climate change adaptation: Special report of the intergovernmental panel on climate change. Cambridge University Press. Gallagher, J. (2014). Learning about an infrequent event: Evidence from flood insurance take-up in the United States. American Economic Journal: Applied Economics, 6(3):206–233. Giné, X., and Yang, D. (2009). Insurance, credit, and technology adoption: Field experimental evidence from Malawi. Journal of Development Economics, 89(1):1–11. Gneezy, U., List, J. A., and Wu, G. (2006). The uncertainty effect: When a risky prospect is valued less than its worst possible outcome. Quarterly Journal of Economics, 121(4):1283–1309. Gneezy, U., and Potters, J. (1997). An experiment on risk taking and evaluation periods. The Quarterly Journal of Economics, 112(2):631–645. Goldbach, C., and Schlüter, A. (2018). Risk aversion, time preferences, and out-migration: Experimental evidence from Ghana and Indonesia. Journal of Economic Behavior & Organization, 150:132–148. Gollier, C., and Pratt, J. W. (1996). Risk vulnerability and the tempering effect of background risk. Econometrica: Journal of the Econometric Society, 64(5):1109–1123. Groves, R. M., Fowler Jr., F. J., Couper, M. P., Lepkowski, J. M., Singer, E., and Tourangeau, R. (2011). Survey methodology, volume 561. John Wiley & Sons. Harrison, G. W. (2006). Hypothetical bias over uncertain outcomes. In Using experimental methods in environmental and resource economics, Edward Elgar Publishing. Northampton, MA. pp. 41–69. Harrison, G. W., Lau, M. I., and Yoo, H. I. (2020). Risk attitudes, sample selection, and attrition in a longitudinal field experiment. Review of Economics and Statistics, 102(3):552–568. Harrison, G. W., and List, J. A. (2004). Field experiments. Journal of Economic Literature, 42(4):1009–1055. He, R., Jin, J., Kuang, F., Zhang, C., and Guan, T. (2020). Farmers’ risk cognition, risk preferences and climate change adaptive behavior: A structural equation modeling approach. International Journal of Environmental Research and Public Health, 17(1):85. Hellerstein, D., Higgins, N., and Horowitz, J. (2013). The predictive power of risk preference measures for farming decisions. European Review of Agricultural Economics, 40(5):807–833. Holt, C. A., and Laury, S. K. (2002). Risk aversion and incentive effects. American Economic Review, 92(5):1644–1655. Holt, C. A., and Laury, S. K. (2014). Assessment arid estimation of risk preferences. Handbook of the Economics of Risk and Uncertainty, 1:135–201. Hornbeck, R. (2012). The enduring impact of the American dust bowl: Short- and long-run adjustments to environmental catastrophe. American Economic Review, 102(4):1477–1507. Hudson, P., Botzen, W. W., Feyen, L., and Aerts, J. C. (2016). Incentivising flood risk adaptation through risk based insurance premiums: Trade-offs between affordability and risk reduction. Ecological Economics, 125:1–13. Jaeger, D. A., Dohmen, T., Falk, A., Huffman, D., Sunde, U., and Bonin, H. (2010). Direct evidence on risk attitudes and migration. Review of Economics and Statistics, 92(3):684–689. Janzen, S. A., and Carter, M. R. (2019). After the drought: The impact of microinsurance on consumption smoothing and asset protection. American Journal of Agricultural Economics, 101(3):651–671. Jianjun, J., Yiwei, G., Xiaomin, W., and Nam, P. K. (2015). Farmers’ risk preferences and their climate change adaptation strategies in the Yongqiao District, China. Land Use Policy, 47:365–372. Jin, J., Wang, W., and Wang, X. (2016). Farmers’ risk preferences and agricultural weather index insurance uptake in rural China. International Journal of Disaster Risk Science, 7(4):366–373. Kahneman, D., Knetsch, J. L., and Thaler, R. H. (1990). Experimental tests of the endowment effect and the Coase theorem. Journal of Political Economy, 98(6):1325–1348. Kahneman, D., and Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47(2):263–292.

Risk preferences and natural disasters  113 Kairies-Schwarz, N., Kokot, J., Vomhof, M., and Weßling, J. (2017). Health insurance choice and risk preferences under cumulative prospect theory: An experiment. Journal of Economic Behavior & Organization, 137:374–397. Kousky, C. (2017). Disasters as learning experiences or disasters as policy opportunities? Examining flood insurance purchases after hurricanes. Risk Analysis, 37(3):517–530. Kreps, D. M. (1990). A course in microeconomic theory. Princeton University Press. Kriesel, W., and Landry, C. (2004). Participation in the national flood insurance program: An empirical analysis for coastal properties. Journal of Risk and Insurance, 71(3):405–420. Langlais, E. (2005). Willingness to pay for risk reduction and risk aversion without the expected utility assumption. Theory and Decision, 59(l):43–50. Lee, J. (2008). The effect of the background risk in a simple chance improving decision model. Journal of Risk and Uncertainty, 36(1):19–41. Lewis, T., and Nickerson, D. (1989). Self-insurance against natural disasters. Journal of Environmental Economics and Management, 16(3):209–223. List, J. A., and Haigh, M. S. (2005). A simple test of expected utility theory using professional traders. Proceedings of the National Academy of Sciences, 102(3):945–948. Liu, E. M. (2013). Time to change what to sow: Risk preferences and technology adoption decisions of cotton farmers in China. Review of Economics and Statistics, 95(4):1386–1403. Liu, E. M., and Huang, J. (2013). Risk preferences and pesticide use by cotton farmers in China. Journal of Development Economics, 103:202–215. Machina, M. J. (1987). Choice under uncertainty: Problems solved and unsolved. Journal of Economic Perspectives, 1(1):121–154. McCoy, S. J., and Walsh, R. P. (2018). Wildfire risk, salience & housing demand. Journal of Environmental Economics and Management, 91:203–228. Mendelsohn, R., Nordhaus, W. D., and Shaw, D. (1994). The impact of global warming on agriculture: A Ricardian analysis. American Economic Review, 84(4):753–771. Menezes, C., Geiss, C., and Tressler, J. (1980). Increasing downside risk. American Economic Review, 70(5):921–932. Meyer, R., and Kunreuther, H. (2017). The ostrich paradox: Why we underprepare for disasters. Wharton School Press. Mosley, P., and Verschoor, A. (2005). Risk attitudes and the “vicious circle of poverty.” The European Journal of Development Research, 17(l):59–88. Petrolia, D. R., Landry, C. E., and Coble, K. H. (2013). Risk preferences, risk perceptions, and flood insurance. Land Economics, 89(2):227–245. Pratt, J. W. (1964). Risk aversion in the small and in the large. Econometrica, 32:122–136. Pratt, J. W. (1978). Risk aversion in the small and in the large. In P. Diamond and M. Rothschild (Eds.), Uncertainty in economics, pp. 59–79. Elsevier. Rabin, M. (2013). Risk aversion and expected-utility theory: A calibration theorem. In L. MacLean and W. Ziemba (Eds.), Handbook of the Fundamentals of Financial Decision Making: Part I, pp. 241–252. World Scientific. Rabin, M., and Thaler, R. H. (2001). Anomalies: Risk aversion. Journal of Economic Perspectives, 15(1):219–232. Ross, N., Santos, P., and Capon, T. (2012). Risk, ambiguity and the adoption of new technologies: Experimental evidence from a developing economy. Working Paper. https://ageconsearch.umn.edu/record/126492/. Samuelson, P. A. (1952). Probability, utility, and the independence axiom. Econometrica: Journal of the Econometric Society, 20(4):670–678. Schildberg-Hörisch, H. (2018). Are risk preferences stable? Journal of Economic Perspectives, 32(2):135–54. Schlesinger, H. (2013). The theory of insurance demand. Handbook of insurance, pp. 167–184. Springer, New York, NY, 2013. 167–184. Schmidt, U., and Hey, J. D. (2004). Are preference reversals errors? An experimental investigation. Journal of Risk and Uncertainty, 29(3):207–218. Schmidt, U., Starmer, C., and Sugden, R. (2008). Third-generation prospect theory. Journal of Risk and Uncertainty, 36(3):203–223. Severen, C., Costello, C., and Deschenes, O. (2018). A forward-looking Ricardian approach: Do land markets capitalize climate change forecasts? Journal of Environmental Economics and Management, 89:235–254. Sheldon, T. L., and Zhan, C. (2019). The impact of natural disasters on US home ownership. Journal of the Association of Environmental and Resource Economists, 6(6):1169–1203. Shupp, R., Loveridge, S., Skidmore, M., Lim, J., and Rogers, C. (2017). Risk, loss, and ambiguity aversion after a natural disaster. Economics of Disasters and Climate Change, 1(2):121–142. Stigler, G. J., and Becker, G. S. (1977). De gustibus non est disputandum. American Economic Review, 67(2):76–90. Szpiro, G. G. (1986). Measuring risk aversion: An alternative approach. Review of Economics and Statistics, 61(1):156–159. Tanaka, T., Garnerer, C. F., and Nguyen, Q. (2010). Risk and time preferences: Linking experimental and household survey data from Vietnam. American Economic Review, 100(l):557–71.

114  Handbook on the economics of disasters Thomas, A., and Leichenko, R. (2011). Adaptation through insurance: Lessons from the NFIP. International Journal of Climate Change Strategies and Management, 3(3):250–263. Tiebout, C. M. (1956). A pure theory of local expenditures. Journal of Political Economy, 64(5):416–424. Tversky, A., and Kahneman, D. (1992). Advances in prospect theory: Cumulative representation of uncertainty. Journal of Risk and Uncertainty, 5(4):297–323. Vieider, F. M., Martinsson, P., Nam, P. K., and Truong, N. (2019). Risk preferences and development revisited. Theory and Decision, 86(1):1–21. Wahdat, A. Z., Gunderson, M. A., and Lusk, J. L. (2021). Farm producers’ household consumption and individual risk behavior after natural disasters. Agricultural and Resource Economics Review, 50(l):127–149. Wibbenmeyer, M. J., Hand, M. S., Calkin, D. E., Venn, T. J., and Thompson, M. P. (2013). Risk preferences in strategic wildfire decision making: A choice experiment with us wildfire managers. Risk Analysis, 33(6):1021–1037.

PART II EVALUATION OF DISASTER CONSEQUENCES

Section I Economic Impacts

7. Economic consequences of pre-COVID-19 epidemics: a literature review Ilan Noy and Tomáš Uher

1. INTRODUCTION In December 2019, a new strain of a coronavirus, SARS-CoV-2, was identified in Wuhan, China. It was causing a respiratory illness now known as COVID-19. In the following months, the virus rapidly spread, causing a global pandemic with significant health, economic, political, and societal implications. As a response to the threat of contagion, most countries implemented dramatic interventions that reduced economic activity by curtailing the movement of people, enforcing social distancing, shutting or heavily restricting border crossings, and prohibiting various activities. These measures resulted in deep recessions in many economies, and their precipitousness has not before been seen in the modern era. Recent studies such as Béland et al. (2020), McKibbin and Fernando (2021), Chudik et al. (2020), Brodeur et al. (2020), Nicola et al. (2020), and Maliszewska et al. (2020) represent some of the early attempts to inform us about COVID-19’s economic consequences. Now, in mid-2022, in high- and middle-income countries, a vaccination campaign has largely run its course, though most are still very far from herd immunity because of a persistent refusal by some to vaccinate; in low-income countries, vaccination rates are still very low, largely because of difficulties in delivery rather than in provisioning doses of the vaccine. As such, the economic consequences of this ongoing pandemic are still being acutely felt in many (most) countries, and it is still too early to summarize what we know about the pandemic’s overall impact on the global economy. Therefore, past epidemic events can potentially provide us useful information about the plausible impact of local epidemics and global pandemics on both the local and global economies. Except for the 1918 H1N1 influenza pandemic (often and erroneously known as the Spanish Flu), modern epidemics and pandemics were not severe enough to enable us to fully compare them to the current crisis. In terms of scale, global reach, and mortality, the 1918 event does offer us sobering lessons. However, several factors make learning from this past event challenging. Its concurrence with the end of World War I, and the enormous changes the end of the “War to End All Wars” brought about have hidden many of the impacts of the 1918 pandemic. Yet even though subsequent epidemics were of a smaller scale and severity, we believe they can also still provide us some useful data points with respect to the ways in which these events impacted our economies. The economic research about the impact of these past epidemic events, including the 1918 pandemic, is the focus of this survey. An epidemic can be viewed as a negative health shock adversely affecting the economies in which it emerges and to which it spread. In an attempt to avoid some of the health consequences of the spread of the infectious disease, populations also significantly change their 117

118  Handbook on the economics of disasters behavior to reduce the possibility of infection. This behavioral shift happens as a consequence of individual decisions based on their own risk perceptions or because of policy decisions and mandates—the non-pharmaceutical interventions (NPIs) enacted by governments. These NPIs typically impose various restrictions on selected types of movement and social interactions. In many cases, for example, during the SARS epidemic of 2003, it is possible to conclude that this aggregate behavioral reaction constituted the main source of the economic impact associated with the epidemic (Noy & Shields, 2019). The increasing connectedness of the world’s societies and economies accelerates the spread of contagions, as well as their economic impacts because of these cascading behavioral reactions. The increasing disease transmission linkages also result in these cascading effects, further worsening the severity of the economic downturn. Conversely, however, global economic connectedness has also led to increasing development and increasing health-care standards around the world, which can potentially reduce the societal and economic vulnerability to epidemics’ adverse consequences (Barro et al., 2020; Bloom & Canning, 2004). For example, we have seen both channels operate during the COVID crisis, with global connectedness both increasing and decreasing the economic impact of the pandemic. As for any large shock, including those associated with natural hazards, the economic impacts of epidemics are heterogeneous across countries, across economic sectors, and across socioeconomic classes. While research suggests that many economic sectors are affected adversely from an adverse event, epidemic occurrence may also be beneficial for certain industries (Hassan et al., 2020). One sector that can benefit, for example, is health-care manufacturing (e.g., for personal protective equipment), in the same way that the construction sector typically experiences a boost from earthquakes. There are other potential positive impacts. In the case of the current COVID-19 pandemic, the resulting global lockdown led to some temporary positive environmental effects by reducing air pollution and greenhouse gases emissions (Khan et al., 2021; Muhammad et al., 2020). However, these positive impacts are typically degrees of magnitude smaller than the damage from these events. Forster et al. (2020), for example, estimated only a negligible cooling effect from the COVID-19 lockdown. Similarly, one positive potential impact is a renewed commitment to reduce vulnerability to future epidemics—this is the public health version of the Build Back Better aspiration common in post-disaster recoveries (Noy et al., 2020b). Whether indeed an occurrence of an epidemic serves as an alarm bell for the affected society in terms of formulating better pandemic preparedness for the future and potentially more severe events (and therefore reducing their consequences) remains an unanswered question, although some circumstantial observations suggest that the countries exposed to SARS in 2003 performed better in 2020 because of that past exposure. This review is structured as follows. Section 2 discusses economic impact pathways and distinguishes between supply- and demand-side channels of impact. Section 3 describes macroeconomic impacts such as on gross domestic product (GDP), consumption, and investment. Section 4 details sectoral impacts and notes the heterogeneity of economic impacts across different industries. Section 5 discusses effects on microeconomic and socioeconomic indicators such as household consumption, income inequality, and unemployment. Section 6 documents longer-term impacts. Section 7 considers the determinants of economic impacts, and Section 8 provides a conclusion.

Economic consequences of pre-COVID-19 epidemics: a literature review  119

2.  IMPACT PATHWAYS Epidemics affect economic activity negatively through both demand-side and supply-side effects, and several researchers have attempted to establish the degree of importance for these differing channels of effects. In the case of the Black Death pandemic in the 14th century, for example, its long-term economic impacts are assumed to have been caused mainly by a significant supply-side reduction in the available labor force (Jedwab et al., 2020). Similarly, when analyzing the 1918 pandemic, some researchers estimate its economic consequences to be brought about mostly by supply-side channels, as the pandemic negatively affected primarily the working-age population and thus especially labor-intensive manufacturing that was often done in very crowded conditions (Beach et al., 2020; Garrett, 2009; Noy et al., 2020a; Velde, 2020). Along with these substantial supply-side negative effects, Correia et al. (2020) also find evidence for demand-side influence due to the negative impacts on the stock of durable goods and bank assets in the United States (creating decreases in demand through the loss of collateral value or tightening budget constraints). A relatively different perspective is offered by Basco et al. (2020). Focusing on Spain, Basco et al. interpret the short-term decrease in real wages they observe to argue it mostly represents a negative demand shock in labor markets. Jinjarak et al. (2020) point to both supply-side and demand-side effects in the case of the 1968 H3N2 pandemic. In analyzing the 2003 SARS epidemic, Siu and Wong (2004) conclude that SARS constituted a negative demand shock with widespread lockdowns imposed or voluntarily adopted, and that the supply side of the affected economies in East Asia remained relatively unaffected. Dixon et al. (2010) also find the severity of the economic impacts to be more sensitive to demand-side effects (e.g., reduction in tourism and leisure activities) when they simulate a hypothetical H1N1 pandemic in the United States using a computable general equilibrium (CGE) modeling approach. These discussions around the nature of the shock (supply or demand) are important from a policy perspective. Dixon et al. (2010), for example, propose that their demand-focused finding suggests that demand-increasing policies may be an effective response to mitigate the adverse economic repercussions of the hypothetical pandemic they investigate.

3.  MACROECONOMIC IMPACTS The macroeconomic consequences of epidemics and pandemics are typically examined using the standard aggregate macro indicators. Predominantly, the variables that are examined are GDP, consumption, investment, labor market indicators, and stock markets. 3.1  GDP and Consumption By far, the most extensive literature on the impact of pandemics (as is the case for natural hazards and disasters more generally) is an examination of GDP. The majority of researchers conclude that epidemics have a negative effect on GDP (or its growth trajectory) and often use this indicator to proxy the severity of the economic impact (Barro & Ursúa, 2008; Barro et al., 2020; Beach et al., 2020; Bloom et al., 2005; Carillo & Jappelli, 2020; De Santis & Van der

120  Handbook on the economics of disasters Veken, 2020; Doan et al., 2020; Huber et al., 2018; James & Sargent, 2006; Kennedy et al., 2006; Keogh-Brown & Smith, 2008; Keogh-Brown et al., 2010; Kirigia et al., 2015; Lee & McKibbin, 2004; Ma et al., 2020a; Ma et al., 2020b; McKibbin & Sidorenko, 2006; Prager et al., 2017; Rodríguez-Caballero & Vera-Valdés, 2020; Smith et al., 2009; Smith et al., 2011; Smith & Keogh-Brown, 2013). In the case of the 1918 pandemic, Barro and Ursúa (2008) estimate its substantial adverse effect on both GDP and consumption across different countries but are constrained in their ability to control for the effects of World War I and its end. A subsequent paper from Barro et al. (2020) attempts to better separate the impacts of the pandemic from the effects of the war; the authors estimate that the death rate caused by the pandemic led to declines of 6% in GDP and 8% in private consumption as well as a temporary increase in the inflation rate for the sample of 42 countries for which they have data. De Santis and Van der Veken (2020) expand on the research of Barro et al. by applying a nonlinear quantitative model to the same data set and construct conditional distribution of real GDP per capita growth estimating a cumulative 7% decrease in economic activity. Investigating the impact of the 1968 H3N2 influenza pandemic, Jinjarak et al. (2020) estimate a statistically significant GDP decrease (along with productivity, investment, and consumption decrease) in a sample of 52 countries. For the 2003 SARS epidemic, Doan et al. (2020) find only a very short-run negative impact on per capita GDP in the four heavily affected countries (China, Hong Kong, Taiwan, and Singapore) followed by a very quick recovery. Similar results are estimated by Smith and Keogh-Brown (2013), who use a CGE model calibrated for the 2009 H1N1 influenza pandemic in lower- and middle-income countries (Thailand, South Africa, and Uganda). Smith and Keogh-Brown find that the economic impact was most severe for Uganda, the country with the lowest income in this group. These results also correspond with the findings of De Santis and Van der Veken (2020), which indicate a higher GDP drop for lower-income countries during the 1918 pandemic. Ma et al. (2020a) use local projections and panel regressions methods for a large set of countries and discover that the GDP growth falls by 3 percentage points on average during the year of the disease outbreak but recovers relatively quickly. Ma et al. (2020b) estimate a real GDP decrease of 2.6 percentage points, on average, analyzing a set of five epidemic and pandemic events of the 21st century preceding the COVID-19 pandemic using data across 210 countries. A significant decline in economic output, although not quantified, is also documented for the case of the 14th-century Black Death (Dols, 2019; Pamuk, 2007). From a theoretical perspective, Boucekkine and Laffargue (2010) estimate a permanent negative effect of epidemics on the level of output in the economy using an overlapping generations model. Short-term consumption reduction is another common phenomenon observed in connection with epidemic occurrence. Researchers estimated this decline for the 1918 pandemic (Barro & Ursúa, 2008; Barro et al., 2020), the 2003 SARS epidemic (Siu & Wong, 2004), the 1968 H3N2 influenza pandemic (Jinjarak et al., 2020), and for a set of six modern epidemics and pandemics (Ma et al., 2020a). While this seems to be a reliable short-term consequence, there does not seem to be enough evidence to conclude anything about long-run effects on consumption. Even for the medium run (up to ten years), there are mixed results, with Obrizan et al. (2020) finding a positive effect on per capita consumption in Sweden following the 1918 pandemic, and Beutels et al. (2009) estimating only a brief postponement of consumption in the aftermath of the 2003 SARS epidemic.

Economic consequences of pre-COVID-19 epidemics: a literature review  121 Using nightlight data as a proxy for economic activity, Doan et al. (2020) discover partial evidence for the relatively more recent 2003 SARS epidemic having a persistent impact on the Chinese economy in the heavily affected regions. However, these results are not conclusive due to the rapid development of the country’s economy at the time. 3.2  Investment and Labor Markets Several papers conclude that epidemics and pandemics lead to a decrease in investment. Jinjarak et al. (2020) observe a short-term reduction of investment during the 1968 H3N2 influenza pandemic and similarly Keogh-Brown and Smith (2008) estimate the same effect for the 2003 SARS epidemic. Obrizan et al. (2020) find a mild negative impact on investment in the medium to long term for the 1918 pandemic. Ma et al. (2020a) report a sharp drop in investment spending across six modern epidemic and pandemic events, and Jordà et al. (2020) uncover a long-term reduction of investment opportunities following a set of historical pandemics of the last millennium. McKibbin and Sidorenko (2006) argue that a pandemic event can lead to a reevaluation of a country’s risk profile and, consequently, to a rise in the cost of borrowing that decreases investment. Multiple studies estimate a negative labor supply shock to be an important mechanism behind an epidemic’s influence on the economy (Beach et al., 2020; Bodenhorn, 2020; Garrett, 2009; Jedwab et al., 2022; Jordà et al., 2020; Kennedy et al., 2006; McKibbin & Sidorenko, 2006; Ojo, 2020). In the case of the 1918 pandemic, the negative labor supply shock seems to be a primary driver of the subsequent economic contraction, as the virus affected the workingage population most severely (Beach et al., 2020; Noy et al., 2020a). Apart from the labor force being reduced by the direct health effects of the disease among the population, labor markets can also be negatively affected via behavioral channels. Based on a cross-country panel of historical epidemics from 1970 onward, Yu et al. (2020) estimate the behavioral response of individuals to have a strong adverse impact on labor force participation in the affected countries. Furthermore, this effect appears to be stronger for countries with high uncertainty avoidance, which is typical in lower-income settings. Bootsma and Ferguson (2007) find a similar dynamic when they estimate a behavioral shift of individuals to limit interaction during the 1918 pandemic, and De La Fuente (2020) report the reluctance to gather into worker groups a primary cause of agricultural labor shortage experienced during the 2014 Ebola epidemic. Several papers indicate that the reductions in labor supply observed during an epidemic lead to an increase in wage rates in the medium to long run. Jordà et al. (2020) analyze long-run economic consequences of historical pandemics and observe an increase in wage rates as a consequence of the labor supply decrease caused by pandemic mortality and morbidity. The long-term effect of wage rate and per capita income increase is also relatively well documented for the case of the Black Death (14th century) in studies by Borsch (2005), Voigtländer and Voth (2013), Jedwab et al. (2022), and Pamuk and Shatzmiller (2014). Even earlier, the Justinianic plague during the 6th century appears to have led to a similar effect (Findlay & Lundahl, 2017; Pamuk & Shatzmiller 2014). In regards to the 1918 pandemic, Garrett’s (2009) findings are along similar lines. Garrett specifically focuses on the US manufacturing labor market and estimates that the labor force reduction caused the marginal product of labor and capital per worker to rise, eventually increasing wage rates. Brainerd and Siegler (2003) report similar results of an increase in per capita income in the years following the pandemic.

122  Handbook on the economics of disasters On the other hand, some studies suggest that a pandemic can lower wages and incomes in the shorter run. Basco et al. (2020) find that in Spain, the excess mortality of the 1918 pandemic led to a significant, and possibly counterintuitive, short-term decrease in real wages. They estimate this impact was heterogeneous between occupations with jobs such as tailors and shoemakers being the most affected, as the demand for their products collapsed, while metalworkers were relatively unaffected. Similarly, Dahl et al. (2020) find a short-term decrease in household income in the heavily affected municipalities in Denmark. Epidemics also appear to have a negative effect on labor productivity, either directly due to its adverse health effects or indirectly by impeding human capital acquisition. Jinjarak et al. (2020) estimate that the excess mortality caused by the 1968 H3N2 influenza led to a substantial decrease in productivity (1.9 percentage points). Guimbeau et al. (2020) study the short- and long-term impacts of the 1918 Spanish flu on the demographic, human capital, and productivity trajectories in Brazil’s Sao Paulo. Using detailed district-level historical data on health, productivity, and education, Guimbeau et al. report both short- and long-run negative impacts on agricultural productivity. Céspedes et al. (2020a) find that an adverse economic shock caused by a pandemic can lead to an unemployment and asset price deflation vicious cycle, suggesting a cyclical causal chain of decreasing productivity, collateral value, and company’s borrowing possibility. In a subsequent study, the authors estimate unconventional policies such as loan guarantees, wage subsidies, or equity injections to be an effective response to preserve a high-productivity economic equilibrium (Céspedes et al., 2020b). Evidence suggests that adverse economic consequences of epidemics are not limited to the regions directly affected by the contagion itself but can spill over to other countries through trade connections or supply chains. Kostova et al. (2019) analyze the consequences of the 2014 West Africa Ebola outbreak and estimate a negative economic effect on trade partners of affected countries. Based on a difference-in-differences model, Kostova et al. estimate a $1.08 billion relative decrease in US exports to the three Ebola-affected West African countries, which is consequently associated with a certain level of job losses in these export sectors. Ma et al. (2020a) similarly estimate a decline in international trade as a consequence of modern pandemics and describe a negative indirect effect of affected trading partners on own-country GDP. 3.3  Financial Markets Several researchers analyze the impacts of epidemics on the stock market, and on financial markets more generally. Barro et al. (2020) find evidence that the 1918 pandemic event caused a significant decrease in real returns on short-term government bills but failed to find a statistically significant effect on the real returns of stocks using annual data. Burdekin (2020) uses higher-frequency monthly data and finds that the pandemic mortality rates are associated with a significant reduction in stock market prices using a panel regression for the United States and nine European countries. On the other hand, Velde (2020) argues that “the US stock market did quite well during the epidemic” (p. 36) and estimates that US stocks were relatively unaffected. Karlsson et al. (2014), in contrast, conclude that the 1918 pandemic led to a decrease in capital returns. Regarding the 2003 SARS epidemic, Siu and Wong (2004) report that the event had only mild negative consequences based on the 1.8% drop of the Hong Kong Hang Seng Index. Chen et al. (2007) finds a temporary but statistically significant decline in hotel stock prices (maybe

Economic consequences of pre-COVID-19 epidemics: a literature review  123 not surprisingly, given the temporary collapse of international travel in the region). Ma et al. (2020b) estimate a short-run overreaction of stock markets when compared to the severity of the economic impacts, when analyzing a set of recent epidemic events.

4.  SECTORAL IMPACTS Evidence indicates that the economic repercussions of epidemic and pandemic events are heterogenous across different sectors and vary based on the characteristics of the epidemic, the contagiousness of the pathogen, and the nature of production in each sector (e.g., whether it is labor intensive in crowded conditions). These heterogenous sectoral impacts can also be distinct because they can originate from changes in demand, which in itself can be heterogeneously impacted by the shock. Using a theoretical framework, McKibbin and Sidorenko (2006) provide an insight into the dynamics of sectoral impacts by estimating a behavioral shift of consumer preferences away from the sectors exposed to the risk of infection during a pandemic. Assuming that the most infection-exposed sectors are more dependent on the movement and interaction of people, it is possible to connect this finding with the results of Keogh-Brown and Smith (2008), who show that these sectors suffer more severe adverse economic impacts. Several sectors have been investigated to provide further details about these heterogeneities. 4.1  Leisure and Hospitality Sector Behavioral change by consumers and mandatory restrictions introduced during an epidemic typically lead to significant demand-driven negative impacts on the leisure and hospitality sector. Steep but temporary declines in tourism were estimated for the 2003 SARS epidemic (Au et al., 2005; Beutels et al., 2009; Cooper, 2006; Keogh-Brown & Smith, 2008; Kuo et al., 2008; Siu & Wong, 2004; Zeng et al., 2005), the 2009 H1N1 pandemic (Haque & Haque, 2018; Page et al., 2012; Rassy & Smith, 2013), and using a generic epidemic modeling (Dixon et al., 2010). International travel in particular was negatively affected during the 2003 SARS epidemic (Beutels et al., 2009; Noy & Shields, 2019; Siu & Wong, 2004), and the hotels and restaurants sector experienced a decline both during the 2003 SARS and 2014 Ebola epidemics (Bowles et al., 2016; Chen et al., 2007; Keogh-Brown & Smith, 2008; Kim et al., 2005) in the affected countries. 4.2  Insurance and Finance A sudden increase in mortality and morbidity rates caused by an epidemic results in possibly unanticipated large losses for the life insurance sector. Weisbart (2006) speculates that a moderate pandemic (similar to the 1957 H2N2 or 1968 H3N2 pandemics) could lead to additional insurance claims of $15 billion, and a severe pandemic (similar to the 1918 pandemic) could cause $155 billion in additional life insurance claims in the United States. Weisbart further argues that financially weaker life insurance companies would not survive a severe pandemic event and would be taken over by their state’s insurance regulator. Similar bailouts will probably occur even in countries in which there is no explicit obligation of the state to take over the liabilities of insolvent insurers.

124  Handbook on the economics of disasters Negative repercussions for the insurance sector are also estimated by Keogh-Brown et al. (2010), who simulate a pandemic using a multisector CGE model and find a significant reduction in domestic output for the insurance sector. Researchers, however, have failed to find significant impacts on the insurance sector in the aftermath of the 1918 pandemic. Cortes and Verdickt (2020) analyze the pandemic’s effect on the insurance industry in the United States and find no change in insurers’ profitability before and after the pandemic. Similarly, Phillips (1984) finds that the insurance sector was not negatively affected in South Africa. Both studies report the post-pandemic increase in demand to be the primary explanation for this relatively surprising finding. However, the demand increase in the United States does not appear to be directly caused by the pandemic mortality itself, but potentially by the increasing rates of income, to which the pandemic may have partially contributed through reductions in labor supply (Short, 2019). Anderson et al. (2020) estimate a short-term impact on the financial sector during the 1918 pandemic and report that the banks in the heavily affected regions in the United States experienced deposit withdrawals. Furthermore, Anderson et al. find that the banks that were members of the Federal Reserve were able to access its liquidity and sustain lending, in contrast with the nonmember banks, which did not borrow on the interbank market and were forced to consequently suspend lending. Gong et al. (2020) identified a short-run increase in the cost of bank loans and a decrease in bank loan volumes during the 2009 H1N1 influenza pandemic based on a sample of 37 countries and their differential exposure to the influenza shock. 4.3  Other Sectors Apart from economic sectors directly associated with face-to-face interactions, a severe epidemic can cause disruptions to many other industries. Partial insight into sector-specific impacts is provided by studies based on modeled pandemics. In this regard, Keogh-Brown et al. (2010) estimate severe impacts for labor-intensive sectors such as health, social services, and education. Smith and Keogh-Brown (2013) report similar findings, predicting significant impact on the service sector with capital-intensive sectors relatively less affected. Several papers attempt to differentiate between the sectoral impacts of the 1918 pandemic. As a consequence of the relatively short time during which NPIs were implemented, the dislocation associated with the end of the war, and the contagion affecting primarily the working-age population, these effects appear to be different from more recent epidemics. Although some of the impacts were significant in scale, most researchers conclude that they were fairly temporary, and industries were able to recover quickly once the pandemic subsided. Velde (2020) estimates an industrial output drop of 20% in the United States during a period of several months of the pandemic’s peak. Similarly, Correia et al. (2020) find a reduction in manufacturing output by 18%. Garrett (2008), Bodenhorn (2020), and Velde (2020) find a demand-driven short-run decrease in retail sales in the United States. Bodenhorn also observes declines in the coal, lumber, and textile industries’ outputs, further estimating that the effects on the lumber industry were both supply and demand driven while the coal and textile industries were primarily affected through a negative labor supply shock. The effects on the textile industry were also investigated for Japan by Noy et al. (2020), and they report similar findings. There also appears to be some evidence that epidemics can negatively impact agricultural production. In the case of the 2014 Ebola epidemic, De La Fuente et al. (2020) estimate a negative effect on rice production caused mainly by a behaviorally induced labor shortage. De Kadt

Economic consequences of pre-COVID-19 epidemics: a literature review  125 (2020) provides suggestive evidence for the 1918 pandemic event to cause a mild but prolonged decrease in maize production in South Africa. Guimbeau et al. (2020) report a decrease in agricultural productivity in both short run and long run as a consequence of the 1918 pandemic.

5.  MICROECONOMIC IMPACTS To understand the impact of pandemics on well-being, it is imperative that one also examines the microeconomic impacts. This literature is significantly sparser, but several papers do examine household consumption and savings, firms’ trading activity, and firms’ profitability. Apart from the adverse effects on aggregate consumption in general, discussed earlier, Norling (2020) and Basco et al. (2020) identify a reduction of household consumption as a consequence of the 1918 pandemic, and Jordà et al. (2020) estimate a behavioral effect in which individuals increase household savings. For firm-level analysis, Fernandes and Tang (2020) focus on the 2003 SARS epidemic’s effect on firms’ import/export growth rates in the affected countries. Even though the SARS epidemic lasted only a few months, Fernandes and Tang find a medium-term negative effect of 4 to 6 percentage points’ decrease in trade from the pre-epidemic trend two years after the epidemic. Furthermore, their results suggest that smaller exporters were more likely to exit the market as a consequence of the economic disruption the epidemic caused. They also conclude that the export of goods, which are located downstream within the supply chain and are readily substitutable with foreign alternatives, was more affected and experienced a slower recovery. In terms of other firm-level economic consequences more broadly, Ma et al. (2020b) identifies a drop in firms’ profitability and firms’ debt increase as a consequence of recent epidemics. 5.1  Socioeconomic Impacts Besides the standard microeconomic measurements for firms and households, socioeconomic indicators related to well-being were also found to be affected by epidemics. These effects include impacts on income inequality, unemployment, poverty and literacy, and public trust. Currently, there does not appear to be a consensus on the way epidemics affect income inequality in general. If an epidemic results in higher mortality among those with low income—possibly because their general health is frequently worse and access to health care more restricted—this will in theory lower income inequality (of those who survived). On the other hand, if the income of low-income populations is decreased disproportionately more than the income of higherincome populations due to job loss or illness, an epidemic will cause inequality to increase. A perspective from the pre-industrial era suggests that historical pandemics lead to a decrease in income inequality. Jedwab et al. (2022) find this to be the case for the Black Death, and Alfani and Murphy (2017), studying historically high mortality crises such as plagues and epidemics, find this to be true more generally. Similarly, Sayed and Peng (2020) find that income inequality decreased as a consequence of four major pandemics of the past century. However, Galletta and Giommoni (2020) report an opposite effect, estimating that income inequality has increased as a consequence of the 1918 pandemic in Italian municipalities. Sayed and Peng (2020) indeed note that an epidemic’s effect on income inequality may strongly depend on the characteristics of the disease and its differential effects on labor supply, productivity, and consumption.

126  Handbook on the economics of disasters This heterogeneity is also a function of the nature of labor markets in the epidemic-affected economies. Evidence suggests that epidemic events have a negative effect on employment generally, but they may hit specific sectors especially hard. Ma et al. (2020a) estimates an unemployment rate rise of nearly 1% during the year of an epidemic outbreak and a recovery that lasts two years. They also describe a heterogeneous effect, as it appears that the epidemics they investigated disproportionately affected less educated workers and females. Noy et al. (2020) find exactly that heterogeneity in their investigations of employment of women in the Japanese textile sector during the 1918 influenza pandemic. Using a difference-in-differences model and regional data from Sweden, Karlsson et al. (2014) conclude that the 1918 pandemic led to an increase in poverty rates in subsequent years. Guimbeau et al. (2020) further expand the analysis of the socioeconomic consequences of the 1918 pandemic and estimate it had both short- and long-term adverse effects on literacy in Brazil. There is also some evidence that epidemic events can lead to a decrease of public and social trust in populations; similar evidence exists for disasters more broadly, though this literature is contested. Aksoy (2020) estimates that epidemic exposure since 1970 led to a permanent decrease of public trust in political institutions and public health systems. Maybe not surprisingly, Aksoy suggests this effect is connected to the public health policies enacted during the epidemic. Using data from the General Social Survey (GSS) in the United States, an analysis by Aasve et al. (2020) indicates that the 1918 pandemic had a negative effect on social trust of the descendants of the survivors. This finding may have possible economic implications, considering that a reduction of social trust in the population can lead to negative long-term consequences for economic development (Tabellini, 2010). 5.2  In Utero Exposure Several papers have examined the connection between in utero exposure to the influenza virus during the 1918 pandemic and long-term negative health and economic impacts for these in utero–exposed individuals (Almond, 2006; Beach et al., 2018; Chul Hong & Yun, 2017; Lin & Liu, 2014; Percoco, 2016). This literature ties in more broadly to the “fetal origins” hypothesis, which places a strong emphasis on the importance of in utero experience in determining adult outcomes in various dimensions (Almond & Currie, 2011). Richter and Robling (2013) and Cook et al. (2019) further explore this causal impact and provide evidence that the adverse effects can be multigenerational. Enami (2016) focuses on the in utero exposure to the 1957 H2N2 pandemic and finds a statistically significant negative effect on the future earnings of non-white females, in contrast with non-white males whose earnings were increased as a consequence of the pandemic.

6.  LONGER-TERM IMPACTS Estimating the long-term economic consequences of epidemics is challenging due to the difficulty of constructing an appropriate and reliable counterfactual scenario and the necessity to focus on relatively nonrecent historical events, for which there can be limited data availability.

Economic consequences of pre-COVID-19 epidemics: a literature review  127 In this regard, several researchers turn their focus onto the 14th-century Black Death, the 1918 pandemic, and a small set of other historical pandemics. For the case of the 1918 pandemic, many researchers describe a V-shaped recovery, in which the economy was able to return to its previous level of activity and subsequent growth trajectories relatively shortly after the pandemic subsided (Beach et al., 2020; Bodenhorn, 2020; Carillo & Jappelli, 2020; Dahl et al., 2020; Garret, 2007; Velde, 2020). As discussed in Section 5.2, Almond (2006), Beach et al. (2018), Chul Hong and Yun (2017), Lin and Liu (2014), and Percoco (2016) all find negative socioeconomic impacts for in utero–exposed individuals (identified by their date of birth). Richter and Robling (2013) and Cook et al. (2019) expand on this by identifying the transmission of these adverse impacts on subsequent generations. Using aggregate data, Guimbeau et al. (2020) estimate a long-term decrease in agricultural productivity and literacy as a consequence of the same pandemic in Brazil. Obrizan et al. (2020) use an overlapping generations model to identify plausible long-term pandemic repercussions in a version of the model calibrated with data from Sweden and find a significant negative effect on economic output. Gao (2020) studies whether the 1918 pandemic had an effect on tax revenue in the United States. Based on two difference-in-differences models, Gao does not find a lasting impact of the pandemic on tax revenue growth in the medium term (seven years). Blickle (2020) discovers that, in the most heavily affected regions of Germany, the pandemic caused a reduction in per capita public spending and increased the share of votes for the extremist political parties in the following decade. In somewhat more ambitious investigations, going back centuries, several researchers attempt to examine the long-run economic effects of the 6th-century Justinianic and the 14th-century Black Death plagues. As discussed in Section 3.2, both of these events caused an increase in wage rates and per capita incomes in most affected countries (Borsch, 2005; Findlay & Lundahl, 2017; Jedwab et al., 2022; Pamuk & Shatzmiller, 2014; Voigtländer & Voth, 2013). This effect was accompanied by a decrease in land prices and interest rates, consequently leading to a reduction in income inequality (Jedwab et al., 2022; Pamuk & Shatzmiller, 2014). However, these impacts were heterogeneous across different regions, as the increased wages that followed the Black Death were only sustained in the long run (over several centuries) in Northwestern Europe. Scholars such as Jedwab et al. (2022) and Pamuk (2007) argue that this contributed to the so-called “Great Divergence” between Europe and the rest of the world and the “Little Divergence” between northwestern and southern Europe. As discussed previously, the central hypothesis in this literature is that epidemics can cause wage rates to increase because of decreases in labor supply associated with a pandemic and thus change other parameters, such as the incentives and investments in technological progress (Jordà et al., 2020; Jedwab et al., 2022). Jordà et al. (2020) attempt to assess medium- and long-run economic consequences of pandemics using the rates of return on assets for several major pandemics in the past 700 years. They focus on the United Kingdom, Spain, Germany, France, Italy, and the Netherlands. Jordà et al. were able to estimate that these historical pandemics were followed by low returns on assets as well as lower investment opportunities persisting, possibly, for several decades. Similarly, the study by Rodríguez-Caballero and Vera-Valdés (2020) of the long-run economic consequences of historical pandemics using time-series data from the United Kingdom discovered a long-lasting negative effect on growth and unemployment.

128  Handbook on the economics of disasters

7.  THE DETERMINANTS OF IMPACTS Economic consequences of epidemics are determined by various factors. While reviewing the evidence on the macroeconomic effects of epidemics of the late 20th and 21th centuries, Bloom et al. (2020) found that the economic consequences are generally dependent on three main dimensions: (1) the pathogen’s characteristics (e.g., mortality, morbidity, and infectiousness), (2) population attributes (e.g., susceptible demographics, socioeconomic stratification), and (3) country-level attributes (e.g., institutional environment, public health capacity, and governance). Other studies provide more specific insights into the factors that influence the degree to which the economy is affected and are discussed next. Even though epidemic events can have economic implications even in countries unaffected by the disease itself, evidence suggests that the countries/regions most affected by the pathogen are typically those that experience the most severe economic consequences too (Doan & Noy, 2021; Keogh-Brown & Smith, 2008). The main characteristics of the transmission pathways of the pathogen define another set of important determinants. Especially problematic in this context are pathogens that can transmit through aerosols from asymptomatic or presymptomatic individuals. Social disruption is the greatest in these cases. A counter example is the Ebola virus, which is infectious through bodily contact only in violently symptomatic patients. In this case, the economic dislocation associated with social distancing requirements is much more relaxed compared to a pathogen that allows for a relatively easier environmental spread and may cause a significantly more severe economic shock affecting export-related sectors as well as e-commerce or tourism, for example, SARS (Noy & Shields, 2019; Rassy & Smith, 2013). The location of a country’s production in the global value chains is important in shaping the impact of an epidemic; both Fernandes and Tang (2020) and George et al. (2020) estimate that the importance of China in the global value chains is positively correlated with the severity of the economic impacts of the 2003 SARS epidemic as it was felt in the affected countries (all directly tied to China’s supply chains). Equally important is a country’s fiscal space, that is, its ability to aggressively respond by increasing government spending (Ma et al., 2020b). This finding aligns well with other research that examines the role of fiscal space in a government’s ability to buffer a variety of shocks. Government public health mandates and prohibitions can also play a role, but Correia et al. (2020) find that these NPIs imposed in some US cities did not exacerbate the economic impacts compared to cities where no such restrictions were put in place during the 1918 pandemic (Noy et al., 2020a, provide a similar finding for Japanese prefectures). Using higher-frequency data, Velde’s (2020) results fall along similar lines as he estimates that the limited economic cost of these restrictions was outweighed by their positive impact through a reduction in the labor supply shock associated with the disease itself. In contrast, both Chapelle (2020) and Barro (2020) fail to find any link between the impositions of these restrictions and the economic performance of the cities in the United States during the same pandemic.

8. CONCLUSION This review summarized the contemporary state of knowledge regarding the economic impacts of past epidemics and pandemics, pre-COVID-19, with particular focus on highly infectious rapidly spreading diseases such as influenza and Ebola. One catastrophic pandemic that we

Economic consequences of pre-COVID-19 epidemics: a literature review  129 excluded from our analysis is the HIV/AIDS pandemic that has extracted a very heavy toll globally, but in particular in southern Africa. The main reason for this exclusion is that, in important ways, the AIDS pandemic is different in its persistent economic impact than the sudden onset (and eventual dissipation) of the epidemics we examined. We examined the macroeconomic, microeconomic, socioeconomic, sectoral, and long-term impacts of these events, and we provided some descriptions of the possible pathways and determinants of these impacts. Considering that epidemics mainly represent a significant negative health shock for the populations affected, the resulting economic consequences are predominantly adverse; though certain sectors and certain demographics might benefit from them. The aggregate behavioral response of the population to the threat of infection appears to be one of the primary determinants of the economic damages. However, other factors, such as the severity of the epidemic, transmission pathway of the pathogen, or the specific forms of NPIs that are implemented, also play an important role in determining the economic repercussions. Ultimately, many of these factors are interdependent. To summarize, most researchers suggest that epidemics and pandemics negatively affect economic output, investment, consumption, international trade, and employment rate as well as other economic aggregates, and, consequently, they decrease both local and, in the case of pandemics, global GDP. The intensity of the impacts is heterogeneous across different sectors and demographics, disproportionately affecting sectors dependent on the movement and interaction of individuals, such as tourism or hospitality, and more vulnerable demographic groups. The existing research also indicates that an epidemic can cause long-lasting disruptions, especially in the case of severe events such as the 14th-century Black Death or the 1918 pandemic. Acknowledging the complexity of a reliable estimation of long-run consequences, further analysis of these questions may prove a promising and valuable area of future research. More detailed economic analyses of 1957 H2N2 and 1968 H3N2 influenza pandemics appear both feasible and useful for assessing any potential future impact of similar events. Surprisingly, these events have been studied a lot less than one would have expected. These events occurred in circumstances that are more similar to our current global economy than the 1918 pandemic, so understanding their impact may be more helpful in assisting us in dealing with our current COVID-19 catastrophe. We also believe it would be beneficial to gain more understanding of the effects of countrylevel economic impact determinants such as epidemic preparedness, previous epidemic exposure, or population-level genetic resistance, as these factors may also serve an important function in determining the gravity of economic impacts or explaining differential outcomes across countries or regions. There is anecdotal evidence to support various hypotheses about these issues. For example, we know that the people of the Americas lacked genetic resistance and were therefore almost wiped out by the new pathogens that the Europeans brought with them in the 15th and 16th centuries (and later in the Pacific). We don’t really have any understanding of the long-run economic implications of these catastrophes; nor do we know if, for example, genetic resistance can explain some of the divergence in the path of COVID-19 in 2020 and its economic consequences. Many other questions about the economics of epidemics remain unanswered, either because of, until recently, a lack of interest, or because of a lack of data. Unfortunately, the pandemic that started in late 2019 in Wuhan will provide us enough data to try and research these questions for many years to come.

130  Handbook on the economics of disasters

REFERENCES Aassve, A., Alfani, G., Gandolfi, F., & Le Moglie, M. (2021). Epidemics and trust: The case of the Spanish Flu. Health Economics, 30(4), 840–857. Aksoy, C. G., Eichengreen, B., & Saka, O. (2020). The political scar of epidemics. IZA Discussion Paper No. 13351. Institute of Labor Economics, Bonn, Germany. Alfani, G., & Murphy, T. E. (2017). Plague and lethal epidemics in the pre-industrial world. Journal of Economic History, 77(1), 314–343. Almond, D. (2006). Is the 1918 influenza pandemic over? Long-term effects of in utero influenza exposure in the post-1940 US population. Journal of Political Economy, 114(4): 672–712. Almond, D., & Currie, J. (2011). Killing me softly: The fetal origins hypothesis. Journal of Economic Perspectives, 25(3), 153–172. Anderson, H., Chang, J. W., & Copeland, A. M. (2020). The effect of the Central Bank liquidity support during pandemics: Evidence from the 1918 influenza pandemic. FRB of New York Staff Report, (928). Au, A. K., Ramasamy, B., & Yeung, M. C. (2005). The effects of SARS on the Hong Kong tourism industry: An empirical evaluation. Asia Pacific Journal of Tourism Research, 10(1), 85–95. Barro, R. J. (2020). Non-pharmaceutical interventions and mortality in US cities during the great influenza pandemic, 1918–1919. Working Paper 27049. National Bureau of Economic Research. Barro, R. J., & Ursúa, J. F. (2008). Macroeconomic crises since 1870. Brookings Papers on Economic Activity, 39(1), 255–350. Barro, R. J., Ursúa, J. F., & Weng, J. (2020). The coronavirus and the great influenza pandemic: Lessons from the “Spanish flu” for the coronavirus’s potential effects on mortality and economic activity. Working Paper 26866. National Bureau of Economic Research. Basco, S., Domenech, J., & Rosés, J. R. (2020). The redistributive effects of pandemics: Evidence on the Spanish Flu. CEPR Discussion Papers 14753. Beach, B., Clay, K., & Saavedra, M. H. (2020). The 1918 influenza pandemic and its lessons for COVID-19. Journal of Economic Literature, 60(1), 41–84. Beach, B., Ferrie, J., & Saavedra, M. (2018). Fetal shock or selection? The 1918 influenza pandemic and human capital development. Working paper 24725. National Bureau of Economic Research. Béland, L. P., Brodeur, A., & Wright, T. (2020). The short-term economic consequences of COVID-19: Exposure to disease, remote work and government response. IZA Discussion Paper 13159. Beutels, P., Jia, N., Zhou, Q. Y., Smith, R., Cao, W. C., & De Vlas, S. J. (2009). The economic impact of SARS in Beijing, China. Tropical Medicine & International Health, 14, 85–91. Blickle, K. (2020). Pandemics change cities: municipal spending and voter extremism in Germany, 1918–1933. FRB of New York Staff Report, (921). Bloom, D. E., & Canning, D. (2004). Epidemics and economics. Harvard School of Public Health. Program on the Global Demography of Aging. Working Paper No. 9. Retrieved May 23, 2022, from https://cdn1.sph.harvard.edu/ wp-content/uploads/sites/1288/2013/10/BLOOM_CANNINGWP9.2006.pdf Bloom, D. E., Kuhn, M., & Prettner, K. (2020). Modern infectious diseases: Macroeconomic impacts and policy responses. Working paper 27757. National Bureau of Economic Research. Bloom, E., De Wit, V., & Carangal-San Jose, M. J. (2005). Potential economic impact of an avian flu pandemic on Asia. Asian Development Bank. Economics and Research Department. ERD Policy Brief Series No. 42. Retrieved May, 23 2022, from https://www.think-asia.org/bitstream/handle/11540/2165/pb042.pdf?sequence=1 Bodenhorn, H. (2020). Business in a time of Spanish Influenza. Working paper 27495. National Bureau of Economic Research. Bootsma, M. C., & Ferguson, N. M. (2007). The effect of public health measures on the 1918 influenza pandemic in US cities. Proceedings of the National Academy of Sciences, 104(18), 7588–7593. Borsch, S. J. (2005). The Black Death in Egypt and England: A comparative study. University of Texas Press. Boucekkine, R., & Laffargue, J. P. (2010). On the distributional consequences of epidemics. Journal of Economic Dynamics and Control, 34(2), 231–245. Bowles, J., Hjort, J., Melvin, T., & Werker, E. (2016). Ebola, jobs and economic activity in Liberia. J Epidemiology Community Health, 70(3), 271–277. Brainerd, E., & Siegler, M. V. (2003). The economic effects of the 1918 influenza epidemic. CEPR Discussion Paper No. 3791. Brodeur, A., Gray, D. M., Islam, A., & Bhuiyan, S. (2020). A literature review of the economics of COVID-19. Journal of Economic Surveys, 35, 1007–1044. Burdekin, R. C. (2020). Death and the stock market: International evidence from the Spanish Flu. Applied Economics Letters, 28(17), 1512–1520. Carillo, M. F., & Jappelli, T. (2020). Pandemic and local economic growth: Evidence from the Great Influenza in Italy. CEPR Discussion Paper No. DP14849. Available at SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3628169

Economic consequences of pre-COVID-19 epidemics: a literature review  131 Céspedes, L. F., Chang, R., & Velasco, A. (2020a). The macroeconomics of a pandemic: A minimalist model. Working paper 27228. National Bureau of Economic Research. Céspedes, L. F., Chang, R., & Velasco, A. (2020b). Macroeconomic policy responses to a pandemic. COVID-19 in Developing Economies, 1, 175–186. Chapelle, G. (2020). The medium run impact of nonpharmaceutical interventions. Evidence from the 1918 influenza in US cities (No. 2020-04). THEMA, Université de Cergy-Pontoise. Chen, M. H., Jang, S. S., & Kim, W. G. (2007). The impact of the SARS outbreak on Taiwanese hotel stock performance: An event-study approach. International Journal of Hospitality Management, 26(1), 200–212. Chudik, A., Mohaddes, K., Pesaran, M. H., Raissi, M., & Rebucci, A. (2020). A counterfactual economic analysis of COVID-19 using a threshold augmented multi-country model. Working paper 27855. National Bureau of Economic Research. Chul Hong, S., & Yun, Y. (2017). Fetal exposure to the 1918 influenza pandemic in colonial Korea and human capital development. Seoul Journal of Economics, 30(4), 353–383. Cook, C. J., Fletcher, J. M., & Forgues, A. (2019). Multigenerational effects of early-life health shocks. Demography, 56, 1855–1874. Cooper, M. (2006). Japanese tourism and the SARS epidemic of 2003. Journal of Travel and Tourism Marketing, 19(2–3), 117–131. Correia, S., Luck, S., & Verner, E. (2020). Pandemics depress the economy, public health interventions do not: Evidence from the 1918 flu. SSRN 3561560. Cortes, G., & Verdickt, G. (2020). Did the 1918–19 influenza pandemic kill the US life insurance industry? SSRN 3697832. Dahl, C. M., Hansen, C. W., & Jense, P. S. (2020). The 1918 epidemic and a V-shaped recession: Evidence from municipal income data. Covid Economics, (6), 137–162. De Kadt, D., Fourie, J., Greyling, J., Murard, E., & Norling, J. (2020). The causes and consequences of the 1918 influenza in South Africa. Working paper 12/2020. Stellenbosch University. De La Fuente, A., Jacoby, H. G., & Lawin, K. G. (2020). Impact of the West African Ebola epidemic on agricultural production and rural welfare: Evidence from Liberia. Journal of African Economies, 29(5), 454–474. De Santis, R. A., & Van der Veken, W. (2020), Macroeconomic risks across the globe due to the Spanish Flu. ECB Working Paper Series, No. 2466. Dixon, P. B., Lee, B., Muehlenbeck, T., Rimmer, M. T., Rose, A., & Verikios, G. (2010). Effects on the US of an H1N1 epidemic: Analysis with a quarterly CGE model. Journal of Homeland Security and Emergency Management, 7(1), 64–65. https://doi.org/10.2202/1547-7355.1769 Doan, N., Nguyen, C., Noy, I., & Sawada, Y. (2020). The economic impacts of a pandemic: What happened after SARS in 2003? CESifo Working Paper 8687. Doan, N., & Noy, I. (2021). A comprehensive measure of lifeyears lost due to COVID-19 in 2020: A comparison across countries and with past disasters. Global Policy, 12(4), 553–561. Enami, A. (2016). The effect of in utero exposure to Asian flu (1957–58) on future earnings. SSRN 2711764. Fernandes, A., & Tang, H. (2020). How did the 2003 SARS epidemic shape Chinese trade? SSRN 3603699. Findlay, R., & Lundahl, M. (2017). Demographic shocks and the factor proportions model: From the plague of Justinian to the Black Death. In R. Findlay, R. Henriksson, H. Londgern, & M. Lundahl (Eds.), The economics of the frontier (pp. 125–172). Palgrave Macmillan, London. Forster, P. M., Forster, H. I., Evans, M. J., Gidden, M. J., Jones, C. D., Keller, C. A., . . . & Turnock, S. T. (2020). Current and future global climate impacts resulting from COVID-19. Nature Climate Change, 10(10), 913–919. Galletta, S., & Giommoni, T. (2020). The effect of the 1918 influenza pandemic on income inequality: Evidence from Italy. SSRN 3634793. Gao, T. (2020). How pandemics impact tax revenues and inequality: Evidence from the 1918 Spanish Flu. Mimeo. Garrett, T. A. (2008). Pandemic economics: The 1918 influenza and its modern-day implications. Federal Reserve Bank of St. Louis Review, 90(March/April 2008), 75–94. Garrett, T. A. (2009). War and pestilence as labor market shocks: US manufacturing wage growth 1914–1919. Economic Inquiry, 47(4), 711–725. George, A., Li, C., Lim, J. Z., & Xie, T. (2020). Propagation of epidemics’ economic impacts via production networks: The cases of China and ASEAN during SARS and COVID-19. SSRN 3641263. Gong, D., Jiang, T., & Lu, L. (2020). Pandemic and bank lending: Evidence from the 2009 H1N1 pandemic. Finance Research Letters, 39(1). https://doi.org/10.1016/j.frl.2020.101627 Guimbeau, A., Menon, N., & Musacchio, A. (2020). The Brazilian bombshell? The long-term impact of the 1918 influenza pandemic the South American way. Working Paper 26929. National Bureau of Economic Research. Haque, T. H., & Haque, M. O. (2018). The swine flu and its impacts on tourism in Brunei. Journal of Hospitality and Tourism Management, 36, 92–101. Hassan, T. A., Hollander, S., van Lent, L., & Tahoun, A. (2020). Firm-level exposure to epidemic diseases: Covid-19, SARS, and H1N1. Working Paper 26971. National Bureau of Economic Research.

132  Handbook on the economics of disasters Huber, C., Finelli, L., & Stevens, W. (2018). The economic and social burden of the 2014 Ebola outbreak in West Africa. Journal of Infectious Diseases, 218(Supplement 5), S698–S704. James, S., & Sargent, T. C. (2006). The economic impact of an influenza pandemic. Canada Department of Finance. Finance Canada. Working Paper 2007-04. Jedwab, R., Johnson, N., & Koyama, M. (2022). The economic impact of the Black Death. Journal of Economic Literature, 60(1), 132–178. Jinjarak, Y., Noy, I., & Ta, Q. (2021). Pandemics and economic growth: Evidence from the 1968 H3N2 influenza. Economics of Disasters and Climate Change, 6(1), 73–93. Jordà, Ò., Singh, S. R., & Taylor, A. M. (2022). Longer-run economic consequences of pandemics. Review of Economics and Statistics, 104(1), 166–175. Karlsson, M., Nilsson, T., & Pichler, S. (2014). The impact of the 1918 Spanish flu epidemic on economic performance in Sweden: An investigation into the consequences of an extraordinary mortality shock. Journal of Health Economics, 36, 1–19. Kennedy, S., Thomson, J., & Vujanovic, P. (2006). A primer on the macroeconomic effects of an influenza pandemic (No. 2006-01). The Treasury, Australian Government. Keogh-Brown, M. R., & Smith, R. D. (2008). The economic impact of SARS: How does the reality match the predictions? Health Policy, 88(1), 110–120. Keogh-Brown, M. R., Smith, R. D., Edmunds, J. W., & Beutels, P. (2010). The macroeconomic impact of pandemic influenza: Estimates from models of the United Kingdom, France, Belgium and The Netherlands. European Journal of Health Economics, 11(6), 543–554. Khan, I., Shah, D., & Shah, S. S. (2021). COVID-19 pandemic and its positive impacts on environment: An updated review. International Journal of Environmental Science and Technology, 18(2), 521–530. Kim, S. S., Chun, H., & Lee, H. (2005). The effects of SARS on the Korean hotel industry and measures to overcome the crisis: A case study of six Korean five-star hotels. Asia Pacific Journal of Tourism Research, 10(4), 369–377. Kirigia, J. M., Masiye, F., Kirigia, D. G., & Akweongo, P. (2015). Indirect costs associated with deaths from the Ebola virus disease in West Africa. Infectious Diseases of Poverty, 4(1), 1–10. Kostova, D., Cassell, C. H., Redd, J. T., Williams, D. E., Singh, T., Martel, L. D., & Bunnell, R. E. (2019). Longdistance effects of epidemics: Assessing the link between the 2014 West Africa Ebola outbreak and US exports and employment. Health Economics, 28(11), 1248–1261. Kuo, H. I., Chen, C. C., Tseng, W. C., Ju, L. F., & Huang, B. W. (2008). Assessing impacts of SARS and Avian Flu on international tourism demand to Asia. Tourism Management, 29(5), 917–928. Lee, J., W., & McKibbin, W., J. (2004). Estimating the global economic costs of SARS. In S. Knobler, A. Mahmoud, S. Lemon, A. Mack, L. Sivitz, & K. Oberholtzer (Eds.), Learning from SARS: Preparing for the next disease outbreak. Washington, DC: National Academies Press. Lin, M., & E. Liu (2014). Does in utero exposure to illness matter? The 1918 influenza epidemic in Taiwan as a natural experiment. Journal of Health Economics, 37, 152–163. Ma, C., Rogers, J. H., & Zhou, S. (2020a). Modern pandemics: Recession and recovery. International Finance Discussion Papers 1295. Washington: Board of Governors of the Federal Reserve System. Ma, C., Rogers, J. H., & Zhou, S. (2020b). Global economic and financial effects of 21st century pandemics and epidemics. Covid Economics, (5), 56–78. Maliszewska, M., Mattoo, A., & Van Der Mensbrugghe, D. (2020). The potential impact of COVID-19 on GDP and trade: A preliminary assessment. World Bank Policy Research Working Paper 9211. McKibbin, W., & Fernando, R. (2021). The global macroeconomic impacts of COVID-19: Seven scenarios. Asian Economic Papers, 20(2), 1–30. McKibbin, W. J., & Sidorenko, A. (2006). Global macroeconomic consequences of pandemic influenza. Sydney, Australia: Lowy Institute for International Policy. Muhammad, S., Long, X., & Salman, M. (2020). COVID-19 pandemic and environmental pollution: A blessing in disguise? Science of The Total Environment, 728(August), 138820. Nicola, M., Alsafi, Z., Sohrabi, C., Kerwan, A., Al-Jabir, A., Iosifidis, C., . . . & Agha, R. (2020). The socio-economic implications of the coronavirus and COVID-19 pandemic: A review. International Journal of Surgery, 78(June), 185–193. Noy, I., De Alwis, D., Ferrarini, B., & Park, D. (2020b). Defining build-back-better after disasters with an example: Sri Lanka’s recovery after the 2004 Tsunami. International Review of Environmental and Resource Economics, 14(4), 349–380. Noy, I., Okubo, T., & Strobl, E. (2020a). The Japanese textile sector and the influenza pandemic of 1918–1920. CESifo Working Paper 8651. Noy, I., & Shields, S. (2019). The 2003 severe acute respiratory syndrome epidemic: A retroactive examination of economic costs. Asian Development Bank Economics Working Paper #591. Obrizan, M., Karlsson, M., & Matvieiev, M. (2020). The macroeconomic impact of the 1918–19 influenza pandemic in Sweden. MPRA Working Paper 98910.

Economic consequences of pre-COVID-19 epidemics: a literature review  133 Ojo, O. B. (2020). Socio-economic impacts of 1918–19 influenza epidemic in Punjab. Journal of Asian and African Studies, 55(7), 1023–1032. Page, S., Song, H., & Wu, D. C. (2012). Assessing the impacts of the global economic crisis and swine flu on inbound tourism demand in the United Kingdom. Journal of Travel Research, 51(2), 142–153. Pamuk, Ş. (2007). The Black Death and the origins of the “Great Divergence” across Europe, 1300–1600. European Review of Economic History, 11(3), 289–317. Pamuk, Ş., & Shatzmiller, M. (2014). Plagues, wages, and economic change in the Islamic Middle East, 700–1500. Journal of Economic History, 74(1), 196–229. Phillips, H. (1984). Black October: The impact of the Spanish influenza epidemic of 1918 on South Africa. (Thesis). University of Cape Town, Faculty of Humanities, Department of Historical Studies. Retrieved on May 23, 2022, from http://hdl.handle.net/11427/7852 Percoco, M. (2016). Health shocks and human capital accumulation: The case of the Spanish Flu in Italian regions. Regional Studies, 50(9), 1496–1508. Prager, F., Wei, D., & Rose, A. (2017). Total economic consequences of an influenza outbreak in the United States. Risk Analysis, 37(1), 4–19. Rassy, D., & Smith, R. D. (2013). The economic impact of H1N1 on Mexico’s tourist and pork sectors. Health Economics, 22(7), 824–834. Richter, A., & Robling, P. (2013). Multigenerational effects of the 1918–19 influenza pandemic in Sweden. Swedish Institute for Social Research, 5. Rodríguez-Caballero, C. V., & Vera-Valdés, J. E. (2020). Long-lasting economic effects of pandemics: Evidence on growth and unemployment. Econometrics, 8(3), 37. Sayed, A., & Peng, B. (2020). Pandemics and income inequality: A historical review. SN Business & Economics, 1(4), 1–17. Short, D. (2019). The effect of the 1918 influenza pandemic on US life insurance holdings. Augustana College Working Paper. Siu, A., & Wong, Y. R. (2004). Economic impact of SARS: The case of Hong Kong. Asian Economic Papers, 3(1), 62–83. Smith, R. D., & Keogh-Brown, M. R. (2013). Macroeconomic impact of a mild influenza pandemic and associated policies in Thailand, South Africa and Uganda: A computable general equilibrium analysis. Influenza and Other Respiratory Viruses, 7(6), 1400–1408. Smith, R. D., Keogh-Brown, M. R., & Barnett, T. (2011). Estimating the economic impact of pandemic influenza: An application of the computable general equilibrium model to the UK. Social Science & Medicine, 73(2), 235–244. Smith, R. D., Keogh-Brown, M. R., Barnett, T., & Tait, J. (2009). The economy-wide impact of pandemic influenza on the UK: A computable general equilibrium modelling experiment. British Medical Journal, 339, b4571. Tabellini, G. (2010). Culture and institutions: Economic development in the regions of Europe. Journal of the European Economic Association, 8(4), 677–716. Velde, F. R. (2020). What happened to the US economy during the 1918 influenza pandemic? A view through highfrequency data. FRB of Chicago Working Paper No. 2020-11. Voigtländer, N., & Voth, H. J. (2013). The three horsemen of riches: Plague, war, and urbanization in early modern Europe. Review of Economic Studies, 80(2), 774–811. Weisbart, S. (2006). Pandemic: Can the life insurance industry survive the avian flu? New York Times, A10. Retrieved May, 23, 2022, from https://www.iii.org/sites/default/files/Bird_Flu.pdf Yu, Z., Xiao, Y., & Li, Y. (2020). The response of the labor force participation rate to an epidemic: Evidence from a cross-country analysis. Emerging Markets Finance and Trade, 56(10), 2390–2407. Zeng, B., Carter, R. W., & De Lacy, T. (2005). Short-term perturbations and tourism effects: The case of SARS in China. Current Issues in Tourism, 8(4), 306–322.

8. Natural disasters and economic growth: revisiting the evidence Jesús Crespo Cuaresma*

1. INTRODUCTION From a theoretical point of view, the effect of natural disasters on economic growth is uncertain. Standard neoclassical models with exogenous technological growth predict that, to the extent that they erode the capital stock, natural disasters will tend to move economies away from their long-run equilibrium. After a fall in income per capita, the neoclassical modelling framework predicts that capital accumulation dynamics will lead to an increase in the economic growth as the country affected by the catastrophic event adjusts to its steady state. Theoretical models, including endogenous technological progress, acknowledge the potential of natural catastrophes to lead to updates in the capital stock and thus cause substantially higher income per capita growth in the aftermath of the disaster (see, for example, Hallegatte & Dumas, 2009). Reconstruction efforts after a disaster may thus act as a mechanism to improve the technological content of capital goods and result in sustained income growth increases (see Crespo Cuaresma et al., 2008, for evidence based on the R&D content of imported goods after natural catastrophes). Exposure to natural disaster risk can also have an effect on the long-run accumulation patterns of production factors by changing their relative return and thus may affect income per capita trajectories. To the extent that the accumulation path of factors of production are affected by the occurrence of catastrophic events, different income per capita trajectories across economies could be at least partly explained by the frequency and intensity of natural disaster events. If disaster risk increases the uncertainty of the returns on physical capital, investments in human capital may become more attractive, and investments in education may increase. The existing empirical evidence for such an effect is mixed, however, with Skidmore and Toya (2002) finding a positive correlation effect of risk measures related to climatic disasters and educational attainment measures, while Crespo Cuaresma (2010) finds a negative correlation that is robust to specification uncertainty. The different mechanisms implied by theoretical models point towards net effects of natural disasters on gross domestic product (GDP) per capita and its growth rate that are country-specific and depend on the particular characteristics of the economy considered. In particular, the existing theoretical frameworks often predict different reactions of income per capita dynamics to natural disaster risk depending on whether the focus is on short-term and medium-term dynamics, or the effects are to be examined in the long run. Empirical studies dealing with the validation of such theoretical predictions exploit variation of economic growth and disaster risk over time and across countries to assess the quantitative role played by natural disasters as a determinant of income per capita changes. Studies addressing responses in the short and medium run tend to rely on panel data modelling tools and multivariate time series specifications to assess the link between catastrophic risk and income dynamics. 134

Natural disasters and economic growth: revisiting the evidence   135 Raddatz (2007) and Loayza et al. (2012) are relevant examples of this branch of the literature, which tends to use dynamic panel data models to quantify the immediate reaction of GDP after catastrophic events take place. The results of the existing empirical studies dealing with the short-run effects of disaster emphasize the existence of heterogeneous responses across countries, economic sectors and natural disaster types. Such response heterogeneity implies that the average effects found for samples, including different countries or types of disasters, are not necessarily very informative about the actual GDP shock caused by the catastrophe. Although the effect of disasters on economic growth in the short to medium run is generally found to be larger in developing countries than in developed ones, the direction and size of the GDP response shows a large degree of variation depending on the characteristics of the economy affected (see also Fomby et al., 2013, and Felbermayr & Gröschl, 2014). The heterogeneous nature of the macroeconomic response to disasters is also highlighted in the results of Noy (2009), which imply that the effect of catastrophes differs depending on human capital, institutions, economic development and openness. Making use of a different methodological framework based on the construction of synthetic controls, Cavallo et al. (2013) are able to identify short-run and long-run effects of disasters on GDP per capita growth. The study finds that these are not significant once we consider exceptional cases of very large disasters whose political consequences affected output negatively. Long-term effects of disaster risk on income per capita growth are routinely assessed empirically, making use of cross-country differences in GDP per capita dynamics, and usually making use of standard regression models based on cross-sectional data that summarize information on long-run economic growth and its potential determinants, including exposure to disaster risk. In such a setting, the focus on economic growth differences across economies that persist over long periods of time allows for an interpretation of the partial correlation between disaster frequency or intensity and income dynamics as a long-term effect. The seminal piece by Skidmore and Toya (2002) constitutes the first thorough analysis of the role of disasters as determinants of economic growth trends in the framework of cross-country regressions. The results in Skidmore and Toya (2002) give evidence that the frequency of climatic disasters correlates positively with long-run economic growth, as well as with the accumulation of human capital and with total factor productivity. The meta-analysis of studies on the nexus between natural disasters and economic growth provided by Klomp and Valckx (2014) confirms that studies using panel data (and thus usually concentrating on short-run and medium-run economic growth impacts) tend to find effect estimates that are more negative than those employing cross-sectional data. In this contribution, we assess the evidence for an effect of natural disasters on the observed differences in long-term GDP per capita growth rates across countries explicitly considering the dimension of uncertainty embodied in the choice of a particular econometric specification. For that purpose, we start by estimating the basic specifications in Skidmore and Toya (2002) making use of an updated data set that spans the period of 1970–2019 and contains information for 123 economies. Our results indicate that there is no statistically significant (linear) association between natural disaster frequency and long-run economic growth. To address simultaneously the uncertainty surrounding the specification of the regression model and the possibility of effects of natural disasters that differ across subsamples of countries, we employ Bayesian model averaging (BMA) techniques. BMA allows us to carry out inference about the effect of natural disaster frequency and intensity which is robust to model uncertainty and thus evaluate the importance of natural disasters as a determinant of differences in long-run

136  Handbook on the economics of disasters economic growth rates across countries. In addition, by including specifications with interaction terms into our space of models, we are able to assess the robustness of context-specific effects of natural disasters on economic development. Our results indicate that the frequency and intensity of natural disasters are not robust predictors of long-run economic growth differences across countries over the past half a century. Our assessment of effect heterogeneity does not unveil robust effects that vary over the level of development of the economies considered or across continents. In addition, the partial correlation between disaster risk and human or physical capital accumulation does not appear robust in our sample of countries. The chapter is structured as follows. In Section 2, we revisit the analysis carried out by Skidmore and Toya (2002), making use of the same type of specifications but an updated data set that expands the information to the period 1970–2019. Section 3 presents the results of applying BMA methods to the econometric framework proposed by Skidmore and Toya (2002) and thus examines the robustness of natural disaster variables as predictors of long-run growth in the global sample of countries under scrutiny. Finally, Section 4 concludes and proposes potentially fruitful avenues of further research that build on the analysis carried out.

2. DISASTERS AND ECONOMIC GROWTH: REVISITING SKIDMORE AND TOYA (2002) The empirical evidence linking natural disasters to long-run economic growth presented in Skidmore and Toya (2002) builds on the estimates of a cross-country regression model that associates average GDP per capita growth over a period of 30 years (1960–1990) with a group of controls and that includes measures of natural disaster risk as independent variables. The specifications estimated in Skidmore and Toya (2002) are therefore of the type

Δy = α + Xβ + Disγ + ε , (8.1)

where Δy is a vector containing observations of the average yearly growth rate of GDP per capita for N different countries over the given period, X is a matrix summarizing the information of K covariates considered to be necessary controls to explain differences in economic growth, Dis is a matrix including observations on variables related to disaster frequency and ε is a vector or random shocks assumed to be uncorrelated and homoscedastic, with variance σ2. The parameter vectors β and γ quantify the effects of the different variables on economic growth. We assume the error term ε to be uncorrelated with the explanatory variables, and we evaluate some of the independent covariates summarized in the matrix X at the initial period used to compute the growth rate of GDP per capita to minimize potential endogeneity problems caused by the feedback effects of income growth on the potential determinants of economic development that are controlled for in the model. The natural disaster variables employed in Skidmore and Toya (2002) are log-transformed disaster frequency measures, based on the total number of disaster events in the period considered or the number of disasters per squared kilometre. Figure 8.1 presents scatterplots that depict the unconditional association between the growth rate of GDP per capita in the period 1970–2019 (purchasing power parity [PPP] adjusted, sourced from the Penn World Table 10.0, see Feenstra et al., 2015) and natural disaster frequency (sourced from the Emergency Events

Natural disasters and economic growth: revisiting the evidence   137 Database [EM-DAT] database, Guha-Sapir et al., 2021) for the sample of 123 countries for which the information required to perform the regressions is available. We present the data for the total number of disaster occurrences, as well as disaggregated in climatic (flood and storms) and geologic (landslides, earthquakes and volcanic activity) disasters, and transform the frequency as log(1+frequency) to accommodate a concave relationship between the variables, as is common in the literature.

Annual GDP per capita growth, 1970–2019

8% 6% 4% 2% 0%

0

1

2

3

4

5

6

7

5

6

7

5

6

7

–2% –4% –6% –8%

–10%

log(1 + # total natural disasters)

Annual GDP per capita growth, 1970–2019

8% 6% 4% 2% 0%

0

1

2

3

4

–2% –4% –6% –8%

–10%

log(1 + # total climatic disasters)

Annual GDP per capita growth, 1970–2019

8% 6% 4% 2% 0%

0

1

2

3

4

–2% –4% –6% –8%

–10%

log(1 + # total geologic disasters)

Figure 8.1  Annual GDP per capita growth vs. natural disaster frequency, 1970–2019

138  Handbook on the economics of disasters Table 8.1  Bivariate regression results Parameter Estimate

R-squared

Total disasters

0.002* (1.71)

0.023

Climatic disasters

0.002 (1.66)

0.022

Geologic disasters

0.002 (1.55)

0.019

Total disasters per sq. km

0.206** (2.03)

0.033

Climatic disasters per sq. km

0.208** (2.02)

0.033

Geological disasters per sq. km

2.544 (0.74)

0.004

Observations

123

Note:  The dependent variable is the annual growth rate of GDP per capita, 1970–2019. Each row corresponds to a bivariate regression model, t statistics in parentheses. * p < 0.10,** p < 0.05,*** p < 0.01

The scatterplots in Figure 8.1 show a slight positive unconditional association between disaster frequency and economic growth, which is only marginally statistically significant for the case of total disasters and roughly resembles the descriptive results that Skidmore and Toya (2002) present for the period 1960–1990. Table 8.1 shows the estimates of linear regression models where the growth of GDP per capita is regressed on single (log transformed) disaster frequency variables, both unadjusted and normalized by country area. The estimates are systematically positive and statistically significant for the case of total disasters, total disasters per squared kilometre and climatic disasters per squared kilometre. A one standard deviation change in the disaster frequency variable is associated with an annual growth rate of GDP per capita, which is, on average, 0.26 percentage points higher for the model with the unadjusted disaster frequency variable and 0.34 percentage points higher for the variables normalized by country size. The basic specification of the multivariate regression model in equation (8.1) entertained by Skidmore and Toya (2002) includes the initial level of income per capita, the fertility rate, the investment rate, the share of trade on total GDP, the share of government expenditure on total GDP and the initial rate of secondary school attainment as controls in the matrix X. We collect data for the period 1970–2019 for these covariates, sourced from the Penn World Table 10.0 (for GDP per capita and the shares of investment, total trade and government expenditure on GDP), the World Bank’s World Development Indicators (for the fertility rate) and the Barro-Lee data set (for secondary school attainment, see Barro & Lee, 2013). For the disaster variables, we estimate regression models based on the overall number of disaster occurrences as well as for disasters per squared kilometre for the period 1970–2019.1 The parameter estimates of the corresponding models of the type presented in equation (8.1) are presented in Table 8.2. The first column presents the estimation results of a specification

Natural disasters and economic growth: revisiting the evidence   139 Table 8.2  Multivariate cross-country growth regressions Initial income per capita, logged Initial secondary schooling, logged Fertility rate

(1)

(2)

(3)

(4)

(5)

–0.014*** (–9.89)

–0.015*** (–9.81)

–0.015*** (–9.76)

–0.014*** (–9.94)

–0.014*** (–9.92)

0.004* (1.78)

0.004* (1.85)

0.003 (1.60)

0.003 (1.64)

0.003 (1.64)

–0.007*** (–5.57)

–0.008*** (–5.77)

–0.008*** (–5.74)

–0.007*** (–5.20)

–0.007*** (–5.09)

Investment share

0.043** (2.04)

0.043** (2.07)

0.043** (2.02)

0.046** (2.20)

0.043** (1.99)

Government expenditure share

0.005 (0.27)

–0.002 (–0.12)

–0.002 (–0.12)

0.011 (0.54)

0.009 (0.47)

Trade share of GDP

0.004 (1.30)

0.001 (0.45)

0.002 (0.50)

0.003 (1.03)

0.004 (1.15)

Disasters, total

–0.002 (–1.46)

Disasters, climatic

–0.002 (–1.23)

Disasters, geologic

–0.000 (–0.01)

Disasters per sq. km, total

0.107 (1.46)

Disasters per sq. km, climatic

0.119 (1.55)

Disasters per sq. km, geologic

–1.347 (–0.51)

Intercept

0.153*** (11.58)

Observations

123

R2

0.567

0.170*** (9.68) 123 0.575

0.169*** (9.69) 123 0.575

0.150*** (11.29)

0.151*** (11.21)

123

123

0.575

0.576

Note:  The dependent variable is the annual growth rate of GDP per capita, 1970–2019. All disaster variables transformed as log(1+x), t-statistics in parenthesis. * p < 0.10,** p < 0.05,*** p < 0.01

including only the variables in the X matrix and revealing significant positive effects of investment and negative effects of fertility and the initial level of income per capita on economic growth, thus indicating conditional convergence dynamics and positive growth effects of physical capital accumulation such as those predicted by the Solow model of economic growth and other neoclassical economic growth models. These partial correlations are unchanged when adding disaster variables to the model, and the schooling variable becomes statistically significant in the specifications that include disaster variables normalized by area. In contrast to the empirical results in Skidmore and Toya (2002), none of the estimated effects of natural disasters in the multivariate regressions is statistically significant at any reasonable signifi-

140  Handbook on the economics of disasters cance level, although the signs of the parameter estimates in the model including climatic and geologic variables normalized by area coincide with those reported in their contribution. These results are partly determined by the particular period employed to assess the effects. Estimating the specifications for the period up to 1990, which resembles more faithfully the exercise in Skidmore and Toya (2002), results in similar insights as those reported in Table 8.1, but renders a significantly positive estimate for the total disaster variable if it is computed using occurrence data ranging back to 1900. Such results cast doubts on the robustness of the partial correlation between natural disaster risk and long-run economic growth across countries of the world and over long periods of time. The existing evidence on heterogeneous effects across countries and over time (see the review in Klomp & Valckx, 2014) calls for a thorough assessment of the robustness of this partial correlation where the possibility of different effects across economies is explicitly acknowledged. In the next section, we address the role played by model uncertainty and parameter heterogeneity in the relationship between natural disasters and long-term economic growth by making use of Bayesian methods.

3. ARE NATURAL DISASTERS A ROBUST DETERMINANT OF ECONOMIC GROWTH? 3.1  Model Uncertainty, Economic Growth and Natural Disasters Starting with the contribution by Levine and Renelt (1992), the empirical literature on economic growth has incorporated methods to assess the robustness of determinants of GDP per capita growth differences across countries to specification uncertainty. In particular, model averaging methods have been used to obtain estimates of the effects of potential determinants of economic growth differences across countries that incorporate the uncertainty embodied in the choice of particular specifications (as defined by the nature of the covariates that are controlled for in the framework of linear regression models). The contributions of Sala-i-Martin (1997), Fernandez et al. (2001a) and Sala-i-Martin et al. (2004) are representative examples of this branch of the literature, which has been summarized by Steel (2020) in an extensive and systematic manner. In this section, we assess the robustness of natural disaster variables as determinants of economic growth, integrating the uncertainty about additional controls and about potential heterogeneous effects across countries into the inference. For that purpose, we employ BMA methods in specifications that account for interaction terms between natural disaster variables and other covariates. We therefore consider models of the class given by

L

Δy = α + X k β k + Dis jγ j + ∑ Dis j z lθ l + ε , (8.2) l=1

where Xk contains observations on k different variables of the pool of total potential controls X , Disj contains observations on j disaster covariates from the pool Dis and the interacting variables z l belong to a pool Z ⊆ X . Our aim is to carry out inference on the effect of particular disaster covariates on economic growth, taking into account that the true model linking these two variables is actually unknown and that the inferential step should thus embody

Natural disasters and economic growth: revisiting the evidence   141 the uncertainty surrounding specification choice. Additionally, effects of natural disasters that vary across economies, depending on the level of the variables summarized in z l , are explicitly allowed for in the space of entertained model specifications. Assuming that we are interested in the parameter attached to a particular variable, δ, its posterior distribution p(δ|y) can be written as

p (δ |y ) = ∑

card ( M ) m=1

p (δ |y, M m ) p( M m | y), (8.3)

with p(δ|y, Mm) denoting the posterior distribution of δ in model Mm, which is defined by the choice of a particular group of covariates and p(Mm | y) denoting the posterior model probability (PMP) for specification Mm. Using Bayes’ theorem, PMPs can be shown to be proportional to the product of the marginal likelihood of the model (p(Mm|y)) and its corresponding prior probability (p(Mm)). An analytical expression for the marginal likelihood of a given specification can be obtained by combining an improper diffuse prior on α and σ2 with a conditional Gaussian prior on the rest of the parameters of the model (Zellner, 1986), so that β k , γ j ,θ l | α ,σ 2 ~ N (0,(gF ʹF )−1 ) , where F is a matrix containing all the covariates of the model as columns. The prior parameter g can be elicited so that Bayes factors between models replicate well-known model selection criteria. Setting g = 1/N, where N is the number of observations, corresponds to comparing models using the Bayesian information criterion (Kass & Raftery 1995; Kass & Wasserman, 1995), while g = 1/K2, where K is the total number of covariates that are considered as potential determinants, mirrors the risk inflation criterion in Foster and George (1994). 3.2  Data, Variables and Models In addition to the control variables included in the benchmark specification presented in Section 2, we also include other potential variables as determinants of economic growth, which correspond to additional controls used for robustness analysis in Skidmore and Toya (2002): the urbanization rate (in the initial year, 1970, sourced from the World Bank’s World Development Indicators), the country’s latitude, a dummy variable identifying countries in the tropics, dummy variables for continents and a dummy variable for landlocked economies. We consider the disaster frequency variables per squared kilometre as the main variables of interest, and we also include disaster intensity variables (natural disaster casualties by category: climatic, geologic and total). For the basic BMA exercise, we focus on assessing the importance of natural disaster variables based on the frequency of total, climatic and geological disasters per square kilometre. To assess the possibility of heterogeneous responses to natural disasters across subsamples, we also consider as additional covariates all the interactions between the continent dummies and the disasters variables, as well as the interaction of the disaster risk covariates and the initial income per capita covariate. Therefore, our set of specifications nests models where the effect of natural disasters on economic growth differs not only across continents of the world but also depending on the economic development level of the country. Several theoretical channels and empirical results related to heterogeneous responses to disaster risk justify the use of such interactions. Crespo Cuaresma et al. (2008), for instance, report different effects of disaster occurrence on the technological content of imports in the reconstruction phase, which in turn

142  Handbook on the economics of disasters may affect the economic growth potential of countries following a disaster episode. Many other empirical studies report differences in the GDP response to disasters depending on the income per capita level of the economy affected (e.g., see Noy, 2009). The benchmark setting based on all the additional controls included in the specifications by Skidmore and Toya (2002) and the interaction terms of disaster frequency variables with continent dummies and initial income per capita leads to a pool of 46 potential covariates that can be used in the specification given by equation (8.2) and therefore a model space composed by 246 (= 70,368,744,177,664) models. Since the estimation of all these specifications is computationally unfeasible, we employ Markov Chain Monte Carlo model composition (MC3) methods to calculate the PMPs and thus compute the model-averaged statistics. While an uninformative prior over specifications, where p(Mm) = 2−46 would appear as a natural choice, such a prior choice leads to a large mass of prior probability being assigned to models that include around 23 independent variables (e.g., see Sala-i-Martin, 2004). Alternative hyperpriors, where the prior inclusion probability of the individual covariates is modelled as a beta distribution, can be exploited to obtain uninformative distributions over model size, which will be used in our application (see Ley & Steel, 2009). In our application, we employ such a binomial-beta hyperprior and thus assume a flat probability distribution across model size instead of across actual models. The inclusion of interaction terms between disaster variables and other controls as covariates in the modelling framework may have important consequences in the design of the prior over the model space. The difficulty of interpreting the parameter estimates in models where the interaction term is included but the parent variables that compose the interaction are not have led to proposals of assigning zero probability to such specifications (see Chipman, 1996; Crespo Cuaresma, 2011). We apply BMA using both a standard prior over the model space, which does not exclude such models, and the strong heredity prior, which assigns them a zero probability a priori and thus de facto excludes them from the set of specifications considered. 3.3  Bayesian Model Averaging Results The results of the BMA exercise, based on two million MC3 model composition steps after 10,000 burn-in steps, are presented in Table 8.3. The BMA exercise makes use of a unit information prior, which, in this case, corresponds to g-prior with g = 1/123, and a binomialbeta prior for model size, which is elicited by imposing an expected prior mean of 23 for the number of included regressors, thus implying a flat prior over model size (see Ley & Steel, 2009). We entertain two different priors over the model space concerning how specifications, including interaction terms, are treated. Under the “no heredity” prior, we treat interaction terms as standard covariates and therefore do not penalize models that include these variables but exclude parent variables, whereas the “strong heredity” prior sets a prior probability of zero to such specifications. For each variable, we report the PMP of specifications in which it is included. This statistic, the posterior inclusion probability (PIP) is routinely used in the empirical literature on model uncertainty to quantify the robustness of a covariate as a determinant of the phenomenon under study. In general, variables that have a PIP that is larger than their corresponding prior inclusion probability tend to be labelled robust, indicating that after observing the data, the support for the inclusion of the covariate has increased. We also present the mean of the model-averaged posterior distribution of the parameters and its corresponding standard deviation.

Natural disasters and economic growth: revisiting the evidence   143 Table 8.3  Bayesian model averaging results No Heredity Prior

Strong Heredity Prior

PIP

PM

PSD

PIP

PM

PSD

Fertility rate

1.0000

–0.0091

0.0009

1.0000

–0.0091

0.0009

Initial income

1.0000

–0.0118

0.0012

1.0000

–0.0118

0.0012

Asia

0.6621

0.0057

0.0045

0.6648

0.0058

0.0046

America × Disasters per sq. km, climatic

0.1223

–0.1850

12.7656

0.0003

–0.0003

0.0411

America × Total disasters per sq. km

0.1099

–0.1355

12.7816

0.0003

–0.0003

0.0371

America

0.1082

–0.0008

0.0025

0.1315

–0.0010

0.0028

Trade over GDP

0.0599

0.0003

0.0014

0.0517

0.0003

0.0013

Investment share

0.0496

0.0020

0.0100

0.0487

0.0020

0.0100

America × Disasters deaths, climatic

0.0496

–0.0033

0.0167

0.0001

0.0000

0.0005

America ×Disasters per sq. km, geologic

0.0466

–0.2546

13.0322

0.0001

–0.0010

0.1110

Initial income × Disasters deaths, geologic

0.0312

0.0000

0.0006

0.0004

0.0000

0.0005

Disasters death, geologic

0.0310

–0.0002

0.0044

0.0319

–0.0003

0.0042

Asia × Total disasters per sq. km

0.0301

0.0141

0.1820

0.0013

0.0013

0.0410

Initial schooling

0.0279

0.0001

0.0006

0.0216

0.0001

0.0005

America × Disasters deaths, geologic

0.0275

–0.0002

0.0019

0.0000

0.0000

0.0005

Disaster deaths per sq. km, climatic

0.0262

–0.0002

0.0077

0.0231

–0.0003

0.0039

Initial income × Disasters death, climatic

0.0250

–0.0001

0.0011

0.0002

0.0000

0.0003

Tropics dummy

0.0248

–0.0001

0.0007

0.0224

–0.0001

0.0007

Asia × Disasters per sq. km, climatic

0.0233

0.0066

0.1608

0.0012

0.0013

0.0433

Africa

0.0230

–0.0001

0.0011

0.0223

–0.0001

0.0011

Total disasters per sq. km × Total disaster deaths

0.0211

0.0039

0.0398

0.0000

0.0000

0.0010

Asia × Disaster deaths, total

0.0204

0.0000

0.0004

0.0003

0.0000

0.0001

Note:  The dependent variable is the annual growth rate of GDP per capita, 1970–2019. All disaster variables transformed as log(1+x), PIP: Posterior inclusion probability, PM: Mean of the posterior distribution of the parameter, PSD: Standard deviation of the posterior distribution of the parameter. Results based on five million MC3 replications after 10,000 burn-in steps. Bold figures for variables with PIP > 0.5. Variables with PIP > 0.02 presented.

144  Handbook on the economics of disasters Independently of the prior used over the model space, the model averaging exercise identifies only three robust covariates as determinants of economic growth in our sample of countries: the fertility rate, the initial income per capita and the dummy variable for Asia. The negative effect of fertility on economic growth and the speed of conditional income convergence (embodied in the parameter attached to the initial income per capita level) are estimated with a high degree of precision, and the mean of the posterior distribution of the parameter of the Asian dummy indicates a higher growth rate of income in this world region. In the most conservative prior setting, corresponding to model priors without heredity penalties, the disaster covariates that achieve the highest PIP correspond to the interaction between the American continent dummy and both total and climatic disasters.2 Although the mean of the posterior distribution of both of these interactions is negative, the large uncertainty surrounding these estimates and their relatively low PIP indicate a lack of robustness of the effect. For the BMA exercise with a strong heredity prior, none of the disaster variables reaches a PIP above 0.05. The lack of robustness of natural disaster variables as robust determinants of long-run economic growth differences across countries may be related to the particular definition of disaster events in the EM-DAT database. A natural disaster is included in the EM-DAT database if at least one of the following criteria is fulfilled: (1) 10 or more people reported killed, (2) 100 or more people reported affected, (3) declaration of a state of emergency or (4) call for international assistance. Given this relatively lax definition of natural catastrophe, the variables used to measure disasters and their intensity are composed by aggregates of very different types of events, which may have disparate macroeconomic effects. In addition, since the availability of reliable information on natural catastrophes has changed during the past five decades, the reliability of the disaster variables may change over time. To ensure that our results are not driven by these characteristics of the data source, we performed additional robustness checks concentrating exclusively on the period 1970–1990 (thus getting closer to the original data used by Skidmore & Toya, 2002) and using more stringent definitions of disasters. The results based on the sample 1970–1990 are qualitatively similar to those presented in Table 8.3, with none of the disaster variables or their interactions achieving PIP values above 0.15. We repeated the BMA exercise using alternative natural disaster variables that only include natural catastrophes resulting in more than 500 casualties, thus concentrating on the most destructive disaster occurrences. Although the PIP of the interaction term between the Asian dummy and the variable measuring total deaths by disasters significantly increased in this setting, its value was still below 0.25, and the model-averaged effect (which had a negative posterior mean) was estimated with large uncertainty. Since the lack of robustness of the natural disaster variables as a determinant of economic growth differences may be driven by differential responses to particular types of disasters, in the next step, we consider covariates based on more disaggregated disaster definitions. Instead of clustering the observed disaster occurrences into climatic and geologic disasters, we consider variables based on the frequency and intensity of storms, floods, earthquakes, landslides and volcanic eruptions. The results of the BMA exercise for the set of disaggregated disaster categories are presented in Table 8.4 and resemble those obtained for aggregated disaster categories. None of the disaster variables or their interaction with continent or income variables appears as a robust determinant of long-run economic growth differences across countries. The results presented in Tables 8.3 and 8.4 point towards a systematic lack of robustness of natural disaster variables as empirical determinants of economic growth over long periods of time. Given the setting used to carry out inference in the presence of specification uncertainty

Natural disasters and economic growth: revisiting the evidence   145 Table 8.4  Bayesian model averaging results: disaggregated disaster categories No Heredity Prior PIP

PM

Strong Heredity Prior PSD

PIP

PM

PSD

Fertility rate

1.0000

–0.0091

0.0009

1.0000

–0.0091

0.0009

Initial income

1.0000

–0.0119

0.0012

1.0000

–0.0119

0.0013

Asia

0.6789

0.0059

0.0046

0.7133

0.0064

0.0046

Disaster deaths per sq. km, flood

0.1687

–0.0237

0.0700

0.1668

–0.0195

0.0554

America × Total occurrence per sq. km, floods

0.1339

–0.8141

2.3419

0.0002

–0.0011

0.0822

America

0.1020

–0.0007

0.0024

0.1212

–0.0009

0.0027

America × Total occurrence per sq. km, storms

0.1004

–0.1940

0.7160

0.0003

–0.0005

0.0313

Trade

0.0864

0.0005

0.0018

0.0785

0.0005

0.0018

America × Total occurrence per sq. km, earthquakes

0.0477

–0.3318

3.3835

0.0001

–0.0011

0.2104

Investment share

0.0439

0.0017

0.0091

0.0472

0.0018

0.0096

Disaster deaths, storms

0.0434

0.0014

0.0092

0.0299

0.0006

0.0063

Asia × Total occurrence per sq. km, floods

0.0286

0.0827

0.7288

0.0039

0.0226

0.3894

Disaster deaths per sq. km, earthquakes

0.0260

–0.0002

0.0016

0.0278

–0.0002

0.0017

Asia × Total occurrence per sq. km, earthquakes

0.0257

0.1708

1.5295

0.0021

0.0294

0.7362

Tropics dummy

0.0247

–0.0001

0.0007

0.0269

–0.0001

0.0008

Initial schooling

0.0232

0.0001

0.0005

0.0214

0.0001

0.0005

Africa

0.0213

–0.0001

0.0011

0.0229

–0.0001

0.0011

Initial income × Total occurrence per sq. km, earthquake

0.0203

–0.0351

0.6940

0.0039

–0.0392

0.6835

Note:  The dependent variable is the annual growth rate of GDP per capita, 1970–2019. All disaster variables transformed as log(1+x), PIP: Posterior inclusion probability, PM: Mean of the posterior distribution of the parameter, PSD: Standard deviation of the posterior distribution of the parameter. Results based on one million MC3replications after 10,000 burn-in steps. Bold figures for variables with PIP > 0.5. Variables with PIP > 0.02 presented.

making use of BMA, the resulting effects of natural catastrophes are to be interpreted as conditional on production factor accumulation (since the physical and human capital investment rates are controlled for as part of the set of potential covariates). If natural disaster risk determines the speed of accumulation of production factors, economic growth would be affected by disaster frequency and intensity, but the current setting would not be able to identify such

146  Handbook on the economics of disasters an effect. To explore the possibility of natural disasters robustly affecting economic growth via factor accumulation, we apply BMA to models with the investment rates of physical and human capital as dependent variables. For the case of the physical investment rate, we employ the same set of covariates used for the cross-country growth regressions and substitute the dependent variable with the investment covariate evaluated at the year 2019. For human capital, we use the human capital index provided by the Penn World Table 10.0, which combines average years of schooling data from Barro and Lee (2013) with return to education estimates based on wage regressions. For the BMA setting in the human capital regressions, we use the 2019 value of the index as the dependent variable and include as a potential explanatory covariate its initial value in 1970. The results for the most conservative prior elicitation, corresponding to the use of a unit information prior on the parameters combined with non-hereditary priors on the model space, are presented in Table 8.5. The resulting PIP estimates indicate that the disaster variables are not robust correlates of factor accumulation once specification uncertainty is integrated into the inference in the context of cross-country regressions covering the past five decades. Table 8.5  Bayesian model averaging results: factor accumulation Investment Share

Human Capital Index

PIP

PM

PSD

PIP

PM

PSD

Africa

0.5745

–0.0473

0.0437

0.0213

–0.0024

0.0236

America

0.5105

–0.0370

0.0389

0.0140

0.0008

0.0132

Asia × Disaster deaths, total

0.2948

0.0113

0.0188

0.0175

–0.0009

0.0111

Trade over GDP

0.2065

0.0107

0.0228

0.2366

0.0479

0.0938

Tropics dummy

0.1950

–0.0093

0.0204

0.0191

0.0015

0.0155

Asia

0.1278

0.0072

0.0204

0.0968

0.0152

0.0520

Disasters per sq. km, geological × Disaster deaths, geological

0.1007

23.1087

82.0216

0.0123

1.2627

45.5770

Initial income

0.0959

0.0020

0.0068

0.0181

0.0001

0.0072

Fertility rate

0.0773

–0.0011

0.0043

0.9946

–0.1557

0.0369

Initial investment share

0.0607

0.0094

0.0419

0.2561

0.1953

0.3634

Initial human capital index

0.0475

0.0018

0.0090

0.9999

0.4887

0.0941

Disasters per sq. km, climatic × Disaster deaths, climatic

0.0327

–0.1823

1.7822

0.0152

–0.1072

2.7484

Africa × Disaster deaths, geological

0.0255

–0.1783

1.3321

0.0114

–0.0953

2.2171

Total disasters per sq. km × Total disaster deaths

0.0232

–0.0120

0.5076

0.0126

–0.0142

0.5815

Initial income × Disasters per sq. km, climatic

0.0228

–0.0083

0.4896

0.0183

–0.0093

1.4022

Asia × Disasters per sq. km, climatic

0.0223

–0.2882

25.9278

0.0181

0.0617

11.2153

Natural disasters and economic growth: revisiting the evidence   147 Table 8.5  (Continued) Investment Share

Human Capital Index

PIP

PM

PSD

PIP

PM

PSD

Total disasters per sq. km

0.0220

0.0638

14.2890

0.0189

–0.0597

10.2391

Initial income × Disasters per sq. km, total

0.0219

–0.0076

0.4547

0.0190

–0.0189

1.3952

Asia × Disasters per sq. km, total

0.0218

0.2790

25.9201

0.0175

0.1334

11.2334

Disaster per sq. km, climatic

0.0215

0.0278

14.3830

0.0188

–0.0578

10.2625

Asia ×Disasters per sq. km, geologic

0.0212

–0.2737

26.0191

0.0143

0.5877

17.7968

Africa × Disaster deaths, climatic

0.0209

–0.0994

0.8579

0.0145

–0.1427

1.9995

Urbanization rate

0.0146

0.0000

0.0001

0.5232

0.0029

0.0031

Disaster deaths, total

0.0135

0.0002

0.0042

0.0236

–0.0010

0.0094

America × Disaster deaths, geologic

0.0132

0.0005

0.0167

0.0231

–0.0044

0.0409

Disaster deaths, geologic

0.0132

0.0012

0.0505

0.0230

–0.0047

0.1007

Initial income × Disaster deaths, geologic

0.0130

0.0000

0.0070

0.0225

–0.0005

0.0136

Africa × Total disaster deaths

0.0107

–0.0010

0.0155

0.0211

0.0117

0.1045

America ×Disasters per sq. km, geologic

0.0102

–0.1520

42.5408

0.0374

–7.6998

203.1145

America × Disaster deaths, total

0.0101

0.0000

0.0014

0.0215

–0.0009

0.0087

America × Disasters per sq. km, climatic

0.0098

0.1705

42.7722

0.0456

–0.8561

196.7974

America × Disasters per sq. km, total

0.0094

–0.1300

42.7691

0.0470

–1.6646

197.1610

Note:  The dependent variable is the investment share in 2019 for the first set of columns and the human capital index in 2019 for the second one. All disaster variables transformed as log(1+x), PIP: Posterior inclusion probability, PM: Mean of the posterior distribution of the parameter, PSD: Standard deviation of the posterior distribution of the parameter. Results based on one million MCMC replications after 10,000 burn-in steps. Bold figures for variables with PIP > 0.5. Variables with PIP > 0.02 presented.

4. CONCLUSIONS From a theoretical perspective, the net effect of natural disaster risk and the occurrence of catastrophic disasters on long-run economic growth trends is uncertain, and the empirical literature is plagued with contradictory results and context-specific differences in the estimates of such impacts. In this contribution, we present a rigorous assessment of the partial correlation between natural disaster occurrence and intensity measures and GDP per capita growth

148  Handbook on the economics of disasters over the period 1970–2019 for a cross-section of 123 countries. The explicit assessment of specification uncertainty and parameter heterogeneity allows us to obtain estimates of the effect that are not necessarily constant across income levels or continents and that are robust to the set of controls used in the regression model. Our results present strong evidence against natural disasters being a robust determinant of long-term income growth or the accumulation of aggregate factors of production in the context of cross-country regressions. The analysis indicates that the uncertainty surrounding existing empirical estimates of partial correlations between natural disaster risk and GDP per capita growth may have been underestimated due to the lack of assessment of specification uncertainty and potential effect heterogeneity. This contribution stresses the difficulty of obtaining precise estimates based on global data sets, and emphasizes the need to address the question of the economic effects of natural disasters making use of data at a higher level of granularity than that usually employed in the macroeconomic literature (a conclusion also reached by Osberghaus, 2019, in his review of the effects of natural disasters on trade and financial flows). Some recent contributions using big data methods to quantify the effect of natural disaster occurrence on economic activity and human mobility such as Boakye et al. (2019) or Yabe et al. (2020) point towards possible methodological frameworks that may prove useful in this endeavour.

NOTES * The author thanks Benjamin Bruce for research assistance, and Mark Skidmore and an anonymous reviewer for useful comments on an earlier version of the chapter. 1. We also estimate regression models based on disaster frequency ranging back to the year 1900, which does not qualitatively affect the main results presented. 2. The unit information prior employed embodies a less stringent penalty to model complexity (as measured by the number of variables included in the specification) as compared to the risk inflation criterion. Fernandez et al. (2001b) propose a benchmark prior for BMA based on eliciting g in the g-prior as g = 1/max(N, K2), and thus predicate the use of a unit information prior in settings where the number of observations is relatively large compared to the number of potential covariates, and the risk inflation prior if the number of variables is relatively large. The sample size and number of variables in our empirical exercise would call for the use of the risk inflation criterion, but we chose the more conservative setting provided by the unit information prior, which leads to larger posterior inclusion probabilities for all potential covariates. The results presented in this section remain qualitatively identical if the risk inflation prior is used instead of the unit information prior.

REFERENCES Barro, R., & Lee, J. W. (2013). A new data set of educational attainment in the world, 1950–2010. Journal of Development Economics, 104, 184–198. Boakye, J., Gardoni, P., & Murphy, C. (2019). Using opportunities in big data analytics to more accurately predict societal consequences of natural disasters. Civil Engineering and Environmental Systems, 36(1), 100–114. Cavallo, E., Galiani, S., Noy, I., & Pantano, J. (2013). Catastrophic natural disasters and economic growth. Review of Economics and Statistics, 95(5), 1549–1561. Chipman, H. (1996). Bayesian variable selection with related predictors. Canadian Journal of Statistics, 24(1), 17–36. Crespo Cuaresma, J. (2010). Natural disasters and human capital accumulation. The World Bank Economic Review, 24(2), 280–302. Crespo Cuaresma, J. (2011). How different is Africa? A comment on Masanjala and Papageorgiou. Journal of Applied Econometrics, 26(6), 1041–1047. Crespo Cuaresma, J., Hlouskova, J., & Obersteiner, M. (2008). Natural disasters as creative destruction? Evidence from developing countries. Economic Inquiry, 46(2), 214–226.

Natural disasters and economic growth: revisiting the evidence   149 Feenstra, R. C., Inklaar, R., & Timmer, M. P. (2015). The next generation of the Penn World Table. American Economic Review, 105(10), 3150–3182. Felbermayr, G., & Gröschl, J. (2014). Naturally negative: The growth effects of natural disasters. Journal of Development Economics, 111, 92–106. Fernandez, C., Ley, E., & Steel, M. F. (2001a). Model uncertainty in cross-country growth regressions. Journal of Applied Econometrics, 16(5), 563–576. Fernandez, C., Ley, E., & Steel, M. F. (2001b). Benchmark priors for Bayesian model averaging. Journal of Econometrics, 100(2), 381–427. Fomby, T., Ikeda, Y., & Loayza, N. V. (2013). The growth aftermath of natural disasters. Journal of Applied Econometrics, 28(3), 412–434. Foster, D. P., & George, E. I. (1994). The risk inflation criterion for multiple regression. The Annals of Statistics, 22(3), 1947–1975. Guha-Sapir, Below, R., & Hoyois, Ph. (2021). EM-DAT: The CRED/OFDA International Disaster Database – www. emdat.be – Université Catholique de Louvain – Brussels – Belgium. Hallegatte, S., & Dumas, P. (2009). Can natural disasters have positive consequences? Investigating the role of embodied technical change. Ecological Economics, 68(3), 777–786. Kass, R. E., & Raftery, A. E. (1995). Bayes factors. Journal of the American Statistical Association, 90, 773–795. Kass, R. E., & Wasserman, L. (1995). A reference Bayesian test for nested hypotheses and its relationship to the schwarz criterion. Journal of the American Statistical Association, 90, 928–934. Klomp, J., & Valckx, K. (2014). Natural disasters and economic growth: A meta-analysis. Global Environmental Change, 26, 183–195. Levine, R., & Renelt, D. (1992). A sensitivity analysis of cross-country growth regressions. The American Economic Review, 82(4), 942–963. Ley, E., & Steel, M. F. (2009). On the effect of prior assumptions in Bayesian model averaging with applications to growth regression. Journal of Applied Econometrics, 24(4), 651–674. Loayza, N. V., Olaberria, E., Rigolini, J., & Christiaensen, L. (2012). Natural disasters and growth: Going beyond the averages. World Development, 40(7), 1317–1336. Noy, I. (2009). The macroeconomic consequences of disasters. Journal of Development Economics, 88(2), 221–231. Osberghaus, D. (2019). The effects of natural disasters and weather variations on international trade and financial flows: A review of the empirical literature. Economics of Disasters and Climate Change, 3(3), 305–325. Raddatz, C. (2007). Are external shocks responsible for the instability of output in low-income countries? Journal of Development Economics, 84(1), 155–187. Sala-i-Martin, X. (1997). I just ran two million regressions. American Economic Review, 87(2), 178–183. Sala-i-Martin, X., Doppelhofer, G., & Miller, R. I. (2004). Determinants of long-term growth: A Bayesian Averaging of Classical Estimates (BACE) Approach. American Economic Review, 94(4), 813–835. Skidmore, M., & Toya, H. (2002). Do natural disasters promote long-run growth? Economic Inquiry, 40(4), 664–687. Steel, M. F. (2020). Model averaging and its use in economics. Journal of Economic Literature, 58(3), 644–719. Yabe, T., Zhang, Y., & Ukkusuri, S. V. (2020). Quantifying the economic impact of disasters on businesses using human mobility data: A Bayesian causal inference approach. EPJ Data Science, 9(1), 36. Zellner, A. (1986). On assessing prior distributions and Bayesian Regression Analysis with g-prior distributions. In P. K. Goel & A. Zellner (Eds.), Bayesian inference and decision techniques: Essays in honour of Bruno de Finetti. Amsterdam: North Holland.

9. The impact of natural disasters on economic growth

Eduardo A. Cavallo, Oscar Becerra and Laura Acevedo

1. INTRODUCTION Natural disasters can have severe economic and human consequences, depriving people of livelihoods and assets, as well as lives. Yet despite their potentially enormous costs, an understanding of the economic effects of natural disasters—their frequency, their duration, their severity—remains a work in progress. This chapter focuses on the impacts of natural disasters on economic growth. Economic theory offers competing hypotheses as to the possible impacts of natural disasters on growth. Models rooted in a Shumpeterian tradition would predict output falling in the aftermath of a shock that depletes labor and capital, subsequently unleashing the forces of creative destruction in the economy, leading to higher productivity and growth.1 The Solow (1956) model with production functions that exhibit diminishing marginal productivity of capital, would predict higher growth rates in the aftermath of a shock that reduces the capital-to-labor ratio below the steady-state level. Instead, in learning-by-doing models, a shock that destroys human and physical capital has negative effects on productivity and growth.2 The competing theoretical predictions suggest that assessing the impact of natural disasters on economic growth is ultimately an empirical question. However, the task of providing conclusive empirical evidence is elusive. A main challenge is that this assessment requires a counterfactual that is not observable: What would have happened absent the shock? The empirical work relies on two alternative approaches: cross-country regressions and comparative case studies. Both approaches provide useful insights to understand dimensions of the size and direction of the effect of natural disasters on growth, and they have their own advantages and limitations. Cross-country analyses typically follow the growth regressions tradition (Barro, 1991). The effect of natural disasters on growth is estimated in a regression in which the dependent variable is the annual growth rate of gross domestic product (GDP) (or GDP per capita), and the explanatory variables include an indicator of the occurrence, or a measure of intensity of the natural disaster, and other determinants of economic growth (Cavallo & Noy, 2011).3 An advantage of this approach is that the estimated effect can be interpreted as the impact of a disaster on growth for the average country. Using the terminology of the treatment effects literature, the estimate is the average treatment effect of a natural disaster on economic growth.4 The use of this approach generates a trade-off between the gains of generality in the interpretation and weak identification due to possible endogeneity issues that arise, for example, from the definition of natural disasters. This is so because not all natural disasters are the same, nor is there a convention, or a clear definition of what a natural disaster is. One thing is the “hazard”—which is who and what geographic areas are at risk; another thing is the 150

The impact of natural disasters on economic growth   151 “incidence”—which is whether the hazard materializes, when and where; and yet another thing is the “impact”—who gets affected when the disaster strikes, and how. To become a disaster, a hazard must generate destruction of human and physical capital. While many countries are exposed to natural hazards, the incidence of the disaster depends on the capacity of a society to mitigate those hazards, making those impacts endogenous to economic development and growth. Even though these considerations can be mitigated by including control variables, unless all factors are controlled for, the estimates may be biased.5 The second methodological approach is comparative case studies. In them, the analysis focuses on the effect of one or more large and catastrophic disasters. The analysis is typically carried out on longitudinal data sets. The effects on economic growth are measured by using event studies, in which the effects are estimated by comparing the average growth rates before and after the disasters.6 The loss in generality of this approach is offset by gains in identification because it requires less stringent assumptions. Following the analogy to the treatment effects literature, the estimated effect of natural disasters on economic growth in the comparative studies approach emulates an average treatment effect for the treatment group; that is, it is the estimated effect of the natural disaster on economic growth for the group of countries severely affected by a disaster. This chapter adds to the literature by providing new estimates of the impacts of large, catastrophic natural disasters on economic growth at the country level in the short term (i.e., on the year of the disaster) and the medium term (i.e., a few years after the disaster) using comparative case studies. The methodology builds on Borensztein et al. (2017), who study the behavior of the average level and growth rate of real GDP per capita six years before and six years after the 50 natural disasters with highest mortality in a sample going up to 2008. They find that output drops the year of the disaster between 2 and 4 percentage points on average. In this chapter, we update their results, extending the sample 11 years to 2019 and considering additional disasters. Extending the sample permits including all catastrophic events that materialized post 2008, for example, the 2010 earthquake that struck Haiti, which has the highest mortality rate on record (222,170 deaths according to the Emergency Events Database [EM-DAT]). Considering other types of events permits drawing broader conclusions regarding the impacts of natural disasters. In adopting the Borensztein et al. (2017) methodology, we opted for a simple but transparent event study approach pooling across different types of natural disaster episodes. We build counterfactuals using pre-disaster trends and use them to assess the impact of natural disasters on real GDP per capita growth and levels. We find that during the year of the disaster, real GDP per capita growth declines by between 3.7 and 2.1 percentage points (p.p.) on average, vis-àvis the average pre-disaster growth for the most catastrophic disasters in the sample, where the severity of the disasters is determined based on the number of fatalities per million people. The estimated negative impacts decline as less-severe disasters, and/or disasters materializing in more advanced economies, are included in the samples. The pre- and post-disaster average growth rates are not statistically different, suggesting that the occurrence of the natural disaster does not affect real GDP per capita growth in the medium term. However, the fact that the post-disaster growth rate is not higher than the pre-disaster average suggests that the output lost during the disaster is never fully recovered. To identify natural disasters with the potential of having aggregate impacts on the economy, we follow Borensztein et al. (2017) and select episodes based on their associated mortality

152  Handbook on the economics of disasters rate. This in turn implies that the resulting sample for the event studies includes predominantly developing countries where the mortality associated to disasters is higher. Although the sample is ultimately determined by the selection criterion chosen, using mortality as selection criteria guarantees that the resulting sample includes countries with similar institutions, financial markets, and insurance markets, thereby enabling us to draw policy conclusions that are generalizable to countries that share these characteristics. Developing economies tend to have a lower capacity to mitigate the effects of natural disasters, and, even worse, they can amplify them (e.g., by having lenient building codes or by building in high-risk areas, shallower credit and insurance markets, weaker institutional capabilities, and poor health-care systems). In additional analyses, we use monetary damages as the measure of the severity of a natural disaster (that include more developed economies) and have found neither short- nor long-term effects of the disaster on economic growth. Taken together, our results suggest that the incidence of natural disasters on growth is mostly an economic development issue. Section 2 provides a brief and selective overview of the related literature to ascertain how this chapter fits into the broader picture. Section 3 is devoted to defining the shock, beginning with the information of natural disasters at the individual events’ level, and then aggregating them up to the country/year dimension, such that the defined unit of observation can be then used to shed light on the empirical question of the chapter. Section 4 presents the empirical strategy, the main results, and sensitivity analyses. Finally, Section 5 presents some final remarks and policy implications.

2.  RELATED LITERATURE The literature analyzing the effects of natural disasters on economic growth is diverse, yet there is no consensus about the effects of a natural disaster on economic growth. From a theoretical standpoint, the path to recovery from the output lost amidst the disaster could yield differential effects in the medium and long runs (Cavallo et al., 2013). Thus, the empirical literature has focused on assessing the effects of natural disasters on economic growth in the short run or in the long run, trying to disentangle which factors may amplify or mitigate the effects. The variation in the estimated effects of natural disasters on economic growth is sizable. In a meta-analysis comparing 750 estimates reported in 22 studies for the early 2000s, Klomp and Valckx (2014) find that most of the studies tend to find negative effects of natural disasters on economic growth in the short run (at the year of the disaster), yet an important fraction of those estimates is not significant. The negative effects of disasters on economic growth are concentrated on developing countries and related to disasters triggered by hydro-meteorological and climatic events. Recent advances using new sources of data (for both disasters and economic development) suggest that very large natural hazards have a negative contemporaneous effect on economic growth and highlight the fact that a natural hazard can have devastating effects on a subnational level, which is particularly difficult to cope with in smaller economies (Bertinelli & Strobl, 2013; Felbermayr & Gröschl, 2014; Klomp, 2016). The approach used in this chapter—as well as in most papers that use either cross-country regressions or comparative studies—relies on the occurrence of natural disasters. However, some papers use information on the hazard. Because the timing, location, and intensity of a hazard can be considered orthogonal to the determinants of the economic growth, the estimated

The impact of natural disasters on economic growth   153 effect of the natural hazard on economic growth is less likely to be affected by endogeneity issues. Moreover, as Cavallo et al. (2013) and Felbermayr and Gröschl (2014) show, there is a positive correlation between the physical magnitude of a disaster and the direct impacts of a natural disaster in terms of number of people killed and pecuniary damages to structures. Thus, the estimated effect of natural hazards is an indirect estimate of the effect of natural disasters on economic growth. But using information on hazards rather than on disasters has two limitations from an empirical view. First, because different types of hazards are measured in different scale units (e.g., magnitude for earthquakes, wind speed for storms, etc.), the analysis requires a focus on only one type of disaster or to use an aggregation method that may compromise on the interpretation. Second, because not all hazards materialize in natural disasters, the interpretation of the estimated effect is different. Using the terminology of the treatment effects literature, the estimated effect resembles an intention to treat, that is, the average effect that a natural hazard has on economic growth, considering that not all hazards become natural disasters. A promising avenue of research in this field is to explore the factors for why a natural hazard becomes a natural disaster and use this information to estimate the overall effect of natural disasters on economic growth. An additional dimension in the related analyses is the locational impact of natural disasters. Hazards, incidences, and impacts vary according to the unit of analysis: whether it is individuals, a neighborhood or a community, a province within a country, or the country itself. When assessing the impacts of natural disasters on economic growth, the unit of analysis is usually the country. Focusing on the country level may, however, lead to missing significant events. Disasters tend to occur at a subnational level, which means that some natural hazards materialize over smaller units creating significant impacts on those affected, yet minimal impacts on the aggregate economy (this is especially true in large countries). The fact that those events do not show up in GDP figures does not mean they do not matter. They may matter greatly—they may even be lethal—on those directly impacted, but do not have significant aggregate effects, especially in large countries. The main challenge to exploit subnational-level variation of disasters is the lack of systematic information on economic activity at the subnational level, especially for low-income countries. Some papers have analyzed the effects of natural disasters on economic growth at the subnational level in the United States and other advanced economies, where the information is available. Thanks to this approach, results have uncovered mechanisms used by populations to self-protect from the negative effects of disasters that usually are not observable at the country level, such as internal migration (Boustan et al., 2012; Boustan et al., 2020). In developing economies, though, the information at the subnational level is scarcer and, in many cases, unreliable. New approaches to overcome this limitation have used satellite data of night-time light intensity as an indirect measure of economic activity (Bertinelli & Strobl, 2013; Klomp, 2016). The increasing availability of high frequency data and the use of new techniques can shed light on relevant factors that make some regions resilient to the force of nature, thus constituting another promising avenue for research. Considering the available evidence, the emerging consensus in the literature—which is still evolving—is that natural disasters have, on average, a negative impact on short-term economic growth, while the medium- to long-run effects remain elusive. This chapter provides new estimates of the short- and medium-term effects focusing on catastrophic natural disasters. In what follows, we characterize the data set used to define natural disasters and to implement the empirical strategy.

154  Handbook on the economics of disasters

3.  STYLIZED FACTS ABOUT NATURAL DISASTERS The source of data for natural disasters used in this chapter is EM-DAT, an online emergency disaster database of the Center for Research on the Epidemiology of Disasters (CRED).7 The EM-DAT database has worldwide coverage and reports data on the occurrence and effects of different types of disasters from 1900 to the present.8 EM-DAT defines a natural disaster as a situation or event that overwhelms local capacity, necessitating a request for external assistance. For a disaster to be entered into the database, at least one of the following criteria must be fulfilled: (1) 10 or more people reported killed, (2) 100 people reported affected, (3) declaration of a state of emergency, or (4) call for international assistance. Among the group labeled “Natural disasters,”9 we focus on the four subgroups that are the most common and for which there is more information available:10 • Geophysical: a hazard originating from solid earth, such as volcanic eruptions and earthquakes. • Meteorological: a hazard caused by short-lived, extreme weather and atmospheric conditions that last from minutes to days, such as extreme temperatures and storms. • Hydrological: a hazard caused by the occurrence, movement, and distribution of surface and subsurface freshwater and saltwater, such as landslides and floods. • Climatological: a hazard caused by long-lived, atmospheric processes ranging from intra-seasonal to multi-decadal climate variability, such as wildfires and droughts. The left panel of Table 9.1 presents the frequency distribution of the four types of natural disasters since 1970.11 There are a total of 12,377 unique events in the sample between 1970 and 2019. About 45% of them are hydrological (mainly floods and landslides), 35% are meteorological (mainly storms and extreme temperatures), and 11% are geophysical (earthquakes, tsunamis, volcanic activity, and mass movements). These three types of disasters account for 91% of the events in the sample; they share the characteristic that they can cause great damage in a short period of time (typically less than ten days), and the timing, location, and duration are not exactly predictable, making them shock-like events for the affected communities. The rest are climatological events comprising droughts and wildfires. Droughts account for 5.3% of the episodes in the sample; they last significantly longer than any of the other types of events (232 days on average). The longer duration makes them less shock-like, more slow-moving processes than the other types of natural disasters. The central panel of Table 9.1 provides the same information for the subset of events that have data on the associated mortality. The total number of natural disasters is reduced by about one-third to 8,500 unique events. Of those, 97% are either hydrological, meteorological, or geophysical. There are only 52 droughts (0.6% of the total events) with information on fatalities. Figure 9.1 shows the average number of natural disasters in each country by region, using the World Bank country classification (World Bank, 2021). Southeast Asia and the Pacific are the regions with highest average incidence of disasters per country in every decade. The worst decade on record was 2000–09, when the average number of events per country was 38 in Southeast Asia and the Pacific, and 12 in the Middle East and Africa.

The impact of natural disasters on economic growth   155

Number of events per country (mean)

40

38.1

36.6

30 25.4 22.7 19.9

20 14.4 10

7.4 2.8

0

11.9

4.1

1970–79

4.0

6.3

1980–89

20.5

18.0

10.2

15.0 12.1

11.8 10.1

6.2

1990–99

Middle East & Africa North America & Latin America and the Caribbean

2000–09

2010–19

Europe & Central Asia Southeast Asia & Pacific

Source:  Authors’ calculations based on EM-DAT.

Figure 9.1  Average number of natural disasters in each country by region Figure 9.1 also shows that there seems to be an increasing trend in the incidence of natural disasters reported in EM-DAT, which may be related to climate change and/or to improved reporting in the data set. Figure 9.2, panel A, shows the total number of events over time by type pooling across regions. There was a significant increase in hydrological (×5.9 since 1970), meteorological (×4.7), and climatological events (×6), all of which may be influenced by climate change.12 However, there was also a significant increase (×3 since 1970) in the number of geophysical events that are not influenced by climate change, which suggests that at least part of the increasing trend of natural disasters is due to improved reporting. To probe deeper on the latter point, we check whether the increasing trend in the number of disasters also holds for large disasters, which, due to their magnitude, are more likely to be reported. We determine the magnitude of the disasters based on the human mortality rate rather than on the estimated direct economic damages because the latter is more likely to be biased toward advanced economies in the database.13 Panel B shows the total number of events (left axis) over time pooling across regions and the number of events above a mortality threshold (right axis). The mortality threshold set is equal to the average mortality rate (per million people) in the sample, covering all episodes since 1970 (which is 17.4 death per million inhabitants; see Table 9A.1 in the appendix to this chapter). The figure shows that while there is an increasing trend in the overall number of events, there is no trend for “above-the-mean mortality” events, providing confirmation that the increasing trend on the former may be driven in part by improved reporting. Once the smaller—conceivably less-catastrophic events—are filtered out of the sample, there does not seem to be an increasing trend for remaining events; instead, there are years with peaks and troughs in the number of events around a mean of about 7.1 occurrences per year.

156  Handbook on the economics of disasters

Number of disasters

400

112

300

218

200

100

0

24 12 1965

36

37

30

5 1975 Meteorological

1985

1995 Year

Hydrological

2005

2015

Geophysical

2025

Climatological

A. Number of Natural Disasters by Subgroup

Number of events

15

250 200

10

150 5

100 50

Number of large events

20

300

0 1970

1980

1990

2000

2010

2020

Year All

Large (Right scale)

Note: Large events are defined as those with a mortality rate above 17.4 B. Increasing Prevalence of Natural Disasters (the world-wide mean before country-year aggregation). Note:  Large events are defined as those with a mortality rate above 17.4 (the worldwide mean before country-year aggregation). Source:  Authors’ calculations based on EM-DAT.

Figure 9.2  Reported natural disasters, 1970–2019

157

61

Drought

12,377

Storm

Total

100

30.5

4.7

39.6

5.5

8.9

0.3

1.7

5.3

3.5

17.8

1.8

13.3

9.0

1.2