Population Change and Impacts in Asia and the Pacific (New Frontiers in Regional Science: Asian Perspectives, 30) 9811002290, 9789811002298

Citation preview

New Frontiers in Regional Science: Asian Perspectives 30

Jacques Poot Matthew Roskruge  Editors

Population Change and Impacts in Asia and the Pacific

New Frontiers in Regional Science: Asian Perspectives Volume 30

Editor-in-Chief Yoshiro Higano, University of Tsukuba, Tsukuba, Ibaraki, Japan

New Frontiers in Regional Science: Asian Perspectives This series is a constellation of works by scholars in the field of regional science and in related disciplines specifically focusing on dynamism in Asia. Asia is the most dynamic part of the world. Japan, Korea, Taiwan, and Singapore experienced rapid and miracle economic growth in the 1970s. Malaysia, Indonesia, and Thailand followed in the 1980s. China, India, and Vietnam are now rising countries in Asia and are even leading the world economy. Due to their rapid economic development and growth, Asian countries continue to face a variety of urgent issues including regional and institutional unbalanced growth, environmental problems, poverty amidst prosperity, an ageing society, the collapse of the bubble economy, and deflation, among others. Asian countries are diversified as they have their own cultural, historical, and geographical as well as political conditions. Due to this fact, scholars specializing in regional science as an inter- and multi-discipline have taken leading roles in providing mitigating policy proposals based on robust interdisciplinary analysis of multifaceted regional issues and subjects in Asia. This series not only will present unique research results from Asia that are unfamiliar in other parts of the world because of language barriers, but also will publish advanced research results from those regions that have focused on regional and urban issues in Asia from different perspectives. The series aims to expand the frontiers of regional science through diffusion of intrinsically developed and advanced modern regional science methodologies in Asia and other areas of the world. Readers will be inspired to realize that regional and urban issues in the world are so vast that their established methodologies still have space for development and refinement, and to understand the importance of the interdisciplinary and multidisciplinary approach that is inherent in regional science for analyzing and resolving urgent regional and urban issues in Asia. Topics under consideration in this series include the theory of social cost and benefit analysis and criteria of public investments, socio-economic vulnerability against disasters, food security and policy, agro-food systems in China, industrial clustering in Asia, comprehensive management of water environment and resources in a river basin, the international trade bloc and food security, migration and labor market in Asia, land policy and local property tax, Information and Communication Technology planning, consumer “shop-around” movements, and regeneration of downtowns, among others. Researchers who are interested in publishing their books in this Series should obtain a proposal form from Yoshiro Higano (Editor in Chief, [email protected]) and return the completed form to him.

Editor in Chief Yoshiro Higano, University of Tsukuba Managing Editors Makoto Tawada (General Managing Editor), Aichi Gakuin University Kiyoko Hagihara, Bukkyo University Lily Kiminami, Niigata University Editorial Board Yasuhiro Sakai (Advisor Chief Japan), Shiga University Yasuhide Okuyama, University of Kitakyushu Zheng Wang, Chinese Academy of Sciences Hiroyuki Shibusawa, Toyohashi University of Technology Saburo Saito, Fukuoka University Makoto Okamura, Hiroshima University Moriki Hosoe, Kumamoto Gakuen University Budy Prasetyo Resosudarmo, Crawford School of Public Policy, ANU Shin-Kun Peng, Academia Sinica Geoffrey John Dennis Hewings, University of Illinois Euijune Kim, Seoul National University Srijit Mishra, Indira Gandhi Institute of Development Research Amitrajeet A. Batabyal, Rochester Institute of Technology Yizhi Wang, Shanghai Academy of Social Sciences Daniel Shefer, Technion - Israel Institute of Technology Akira Kiminami, The University of Tokyo Jorge Serrano, National University of Mexico Binh Tran-Nam, UNSW Sydney, RMIT University Vietnam Ngoc Anh Nguyen, Development and Policies Research Center Thai-Ha Le, RMIT University Vietnam Advisory Board Peter Nijkamp (Chair, Ex Officio Member of Editorial Board), Tinbergen Institute Rachel S. Franklin, Brown University Mark D. Partridge, Ohio State University Jacques Poot, University of Waikato Aura Reggiani, University of Bologna

More information about this series at http://www.springer.com/series/13039

Jacques Poot • Matthew Roskruge Editors

Population Change and Impacts in Asia and the Pacific

Editors Jacques Poot Department of Spatial Economics Vrije Universiteit Amsterdam Amsterdam, The Netherlands

Matthew Roskruge School of Economics and Finance Massey University Auckland, New Zealand

National Institute of Demographic and Economic Analysis (NIDEA) University of Waikato Hamilton, New Zealand

ISSN 2199-5974 ISSN 2199-5982 (electronic) New Frontiers in Regional Science: Asian Perspectives ISBN 978-981-10-0229-8 ISBN 978-981-10-0230-4 (eBook) https://doi.org/10.1007/978-981-10-0230-4 © Springer Nature Singapore Pte Ltd. 2020 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

Preface

This book has had a long gestation period. The idea for the book emerged when the editors organised a set of six special sessions on ‘regional science perspectives on population change and impacts’ at the 2014 World Congress of the Regional Science Association International (RSAI) that was to be held May 27–29 in Ayutthaya, Thailand. A total of 18 papers were offered but, due to domestic political unrest in Thailand that week, the conference was cancelled. However, the subsequent 2014 North American Regional Science Conference (NARSC) held in Bethesda, Washington DC, November 12–15, offered an additional opportunity to organise nine sessions on population, labour and regional science. Conference participants who had either presented a paper, or intended to present a paper, at one of these two conferences and who had conducted research that focussed on the Asia-Pacific region were subsequently invited to contribute to this book. A contract was signed with Springer in August 2015. Each submitted paper was subsequently reviewed by two anonymous reviewers. After feedback and revision, this resulted in a total of 15 contributions that were accepted for publication by early 2017. Unfortunately, the editors then each faced a change in personal circumstances that seriously impacted on their ability to complete and edit the manuscript for publication. This significantly delayed the completion of the book. We would like to thank all contributors for their patience and understanding and, of course, for their contributions without which this book could not have come to fruition. In the end, we hope that—the long gestation period has yielded a publication that will satisfy both contributors and readers. The first editor would like to acknowledge the support of the Department of Spatial Economics at the Vrije Universiteit Amsterdam where, as visiting professor, he was offered the required facilities to complete the editing of the book. The editors are especially grateful to Elfie Bonke and Jenny Wiersema in the Department of Spatial Economics for assisting in the production of a manuscript that satisfies the Springer style and formatting guidelines as much as possible.

vii

viii

Preface

Finally, we would like to thank Yoshiro Higano, Editor in Chief of the New Frontiers in Regional Science: Asian Perspectives series, and Selvakumar Rajendran, Production Editor for Springer Nature, for their ongoing encouragement and support. Amsterdam, The Netherlands Auckland, New Zealand November 2019

Jacques Poot Matthew Roskruge

Contents

1

Population Change in the Asia-Pacific Region: Trends, Issues and Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jacques Poot and Matthew Roskruge

Part I 2

Population Distribution

Pareto’s Law and City Size in China: Diverging Patterns in Land and People . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . John Gibson and Chao Li

3

City Size Distribution in Colombia and Its Regions, 1835–2005 . . . . Gerson Javier Pérez-Valbuena and Adolfo Meisel-Roca

4

Exploring Economic Futures for Japan Under Rapid Depopulation: A Dynamic Regional CGE Model Approach . . . . . . Suminori Tokunaga and Mitsuru Okiyama

5

1

29 49

77

Using Spatial Microsimulation to Derive a Base File for a Spatial Decision Support System . . . . . . . . . . . . . . . . . . . . . . . 107 Robert Tanton and Yogi Vidyattama

Part II

Migration and Development

6

The Drivers of Long-Distance Commuting in Chile: The Role of the Spatial Distribution of Economic Activities . . . . . . . . . . . . . . 123 Francisco Rowe and Martin Bell

7

Can Regionally-Targeted Temporary Visas Be an Effective Policy Instrument? A General Equilibrium Analysis . . . . . . . . . . . . 151 Nhi H. Tran, Elizabeth L. Roos, James A. Giesecke, and John R. Madden

ix

x

Contents

8

Modeling the Dynamics of Circular Migration . . . . . . . . . . . . . . . . 183 Natasha T. Duncan, Jacques Poot, and Brigitte Waldorf

9

The Changing Composition and Fortunes of Overseas Graduates in Australia: The Case of Chinese and Indian Graduates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 Angelina Zhi Rou Tang, Francisco Rowe, Jonathan Corcoran, and Alessandra Faggian

10

Effective Work Experience and Labour Market Impacts of New Zealand Immigration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221 Sholeh A. Maani and Michael M. H. Tse

11

Migration and Regional Development in Timor-Leste . . . . . . . . . . . 247 Tomaz Ponce Dentinho, Pedro Damião, and Maria da Conceição Rego

Part III

Population Age Composition and Impacts

12

China’s Ageing Population: The Present Situation and Prospects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269 Guoping Mao, Fuzhong Lu, Xuchun Fan, and Debiao Wu

13

Population Aging in India: Facts, Issues, and Options . . . . . . . . . . . 289 Arunika Agarwal, Alyssa Lubet, Elizabeth Mitgang, Sanjay Mohanty, and David E. Bloom

14

The Child Deficit and the Changing Value of Children in Asia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313 Philip S. Morrison

15

Situational Stressors Among Caregivers of Older Persons in Thailand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341 Amara Soonthorndhada, Denis Akankunda Bwesigye, Jeerawan Hongthong, and Wannee Hutaphat

Contributors

Arunika Agarwal Department of Global Health and Population, Harvard T.H. Chan School of Public Health, Harvard University, Boston, MA, USA Martin Bell Queensland Centre for Population Research, School of Earth and Environmental Sciences, University of Queensland, Brisbane, QLD, Australia David Bloom Department of Global Health and Population, Harvard T.H. Chan School of Public Health, Harvard University, Boston, MA, USA Denis Akankunda Bwesigye Makerere University College of Health Sciences, Kampala, Uganda Maria da Conceição Rego Department of Economics and CEFAGE, University of Evora, Évora, Portugal Jonathan Corcoran Queensland Centre for Population Research, School of Geography, Planning and Environmental Management, University of Queensland, Brisbane, QLD, Australia Pedro Damião Department of Economics and CEFAGE, University of Evora, Évora, Portugal Tomaz Ponce Dentinho Centre for Studies of Applied Economics of the Atlantic, University of the Azores, Ponta Delgada, Portugal Natasha T. Duncan Honors College, Purdue University, West Lafayette, IN, USA Alessandra Faggian Social Sciences, Gran Sasso Science Institute, L’Aquila, Italy Xuchun Fan School of Business, Jiaxing University, Jiaxing City, Zhejiang Province, People’s Republic of China John Gibson Department of Economics, School of Accounting, Finance and Economics, University of Waikato, Hamilton, New Zealand James A. Giesecke Centre of Policy Studies, Victoria University, Melbourne, VIC, Australia xi

xii

Contributors

Jeerawan Hongthong Institute for Population and Social Research, Mahidol University, Salaya, Thailand Wannee Hutaphat Institute for Population and Social Research, Mahidol University, Salaya, Thailand Chao Li Department of Economics, School of Accounting, Finance and Economics, University of Waikato, Hamilton, New Zealand Fuzhong Lu Nanhu College, Jiaxing University, Jiaxing City, Zhejiang Province, People’s Republic of China Alyssa Lubet Department of Global Health and Population, Harvard T.H. Chan School of Public Health, Harvard University, Boston, MA, USA Sholeh A. Maani Graduate School of Management, University of Auckland, Auckland, New Zealand IZA, Bonn, Germany John R. Madden Centre of Policy Studies, Victoria University, Melbourne, VIC, Australia Guoping Mao School of Mathematical and Information Sciences, Jiaxing University, Jiaxing City, Zhejiang Province, People’s Republic of China Adolfo Meisel-Roca Universidad del Norte, Barranquilla, Colombia Elizabeth Mitgang Department of International Health, Johns Hopkins Bloomberg School of Public Health, John Hopkins University, Baltimore, MD, USA Sanjay Mohanty International Institute for Population Sciences, Mumbai, Maharashtra, India Philip S. Morrison School of Geography, Environment and Earth Sciences, Victoria University of Wellington, Wellington, New Zealand Mitsuru Okiyama Reitaku Institute of Political Economies and Social Studies, Reitaku University, Kashiwa, Chiba, Japan Gerson Javier Pérez-Valbuena Center for Regional Economics Studies, Banco de la República, Cartagena de Indias, Colombia Jacques Poot Department of Spatial Economics, Vrije Universiteit Amsterdam, Amsterdam, The Netherlands National Institute of Demographic and Economic Analysis (NIDEA), University of Waikato, Hamilton, New Zealand E. Louisa Roos Centre of Policy Studies, Victoria University, Melbourne, VIC, Australia Matthew Roskruge School of Economics and Finance, Massey University, Auckland, New Zealand

Contributors

xiii

Francisco Rowe Queensland Centre for Population Research, School of Geography, Planning and Environmental Management, University of Queensland, Brisbane, QLD, Australia Department of Geography and Planning, School of Environmental Sciences, University of Liverpool, Liverpool, UK Amara Soonthorndhada Institute for Population and Social Research, Mahidol University, Salaya, Thailand Angelina Zhi Rou Tang Queensland Centre for Population Research, School of Geography, Planning and Environmental Management, University of Queensland, Brisbane, QLD, Australia Institute for Social Science Research, University of Queensland, Brisbane, QLD, Australia Robert Tanton National Centre for Social and Economic Modelling, Institute for Governance and Policy Analysis, University of Canberra, Canberra, ACT, Australia Suminori Tokunaga Faculty of Economics and Business Administration, Reitaku University, Kashiwa, Chiba, Japan Chikuro Hiroike School of Graduate Studies, Reitaku University, Kashiwa, Chiba, Japan Nhi H. Tran Centre of Policy Studies, Victoria University, Melbourne, VIC, Australia Michael M. H. Tse Fonterra Centre, Auckland, New Zealand Yogi Vidyattama National Centre for Social and Economic Modelling, Institute for Governance and Policy Analysis, University of Canberra, Canberra, ACT, Australia Brigitte Waldorf Department of Agricultural Economics, Purdue University, West Lafayette, IN, USA Debiao Wu School of Mathematical and Information Sciences, Jiaxing University, Jiaxing City, Zhejiang Province, People’s Republic of China

Chapter 1

Population Change in the Asia-Pacific Region: Trends, Issues and Models Jacques Poot and Matthew Roskruge

Abstract This chapter provides an introduction to the discussion of demographic changes, issues and models in the Asia-Pacific region that are the focus of this book. The Asia-Pacific region represents 4.9 billion people or roughly 63% of the world’s population. This region is hugely diverse: the countries and subregions vary demographically, geographically, economically, culturally and institutionally. We examine salient features of population change over the last three decades and draw conclusions on the underlying factors which influence population dynamics. We also consider population projections until the middle of the twenty-first century and briefly review some salient impacts of demographic change. Population growth is declining and populations are ageing, both numerically and structurally, everywhere in the region. Fertility rates are converging to replacement levels in many countries or remain below replacement. Large differences in life expectancies, urbanisation and international migration remain. Given the large spatial variations, we argue for a greater emphasis on subnational and multiregional population analysis and projections. Keywords Population trends · Projections · Ageing · Migration · Asia-Pacific

J. Poot (*) Department of Spatial Economics, Vrije Universiteit Amsterdam, Amsterdam, The Netherlands National Institute of Demographic and Economic Analysis (NIDEA), University of Waikato, Hamilton, New Zealand e-mail: [email protected]; [email protected] M. Roskruge School of Economics and Finance, Massey University, Auckland, New Zealand e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 J. Poot, M. Roskruge (eds.), Population Change and Impacts in Asia and the Pacific, New Frontiers in Regional Science: Asian Perspectives 30, https://doi.org/10.1007/978-981-10-0230-4_1

1

2

1.1

J. Poot and M. Roskruge

Introduction

In our turbulent world—with large geopolitical change, climate change, environmental damage, growing inequality and changing patterns of trade among the issues of concern—one might be tempted to think of demographic trends as relatively straightforward and predictable. Yet the size, composition, distribution and change of today’s global population can have profound impacts on the world for many years to come. Additionally, the underlying demographic processes of fertility, mortality and migration are not static but dynamic and are influenced by a range of socioeconomic, cultural and environmental factors. While not going as far as saying that ‘demography is destiny’, the underlying premise of this book is that population matters and is subject to important spatial-temporal variation. The focus of this book is not the world as a whole but a large part of the world that has become increasingly important—demographically, politically and economically—during the last half century: the Asia-Pacific region. We define this region here as comprising the countries of eastern, southern and south-eastern Asia, Oceania and the countries of the Americas with a Pacific coast. The Asia-Pacific region, thus defined, accounts for about 63% of the estimated world population of 7.8 billion in 2020 and exhibits great variety in key demographic outcomes, such as population size, growth, composition and distribution.1 This variety is the result of spatially varying processes of fertility, mortality and migration, combined with spatially varying population structures. This volume brings together a range of contributions that jointly provide contemporary regional science perspectives on population change, and its socioeconomic consequences, in the Asia-Pacific region. Regional science is a multidisciplinary field in the social sciences that is concerned with analytical approaches to regional, urban or rural issues. The contributions to this volume are from a range of disciplines, but have in common that geography matters—both within countries and in cross-country or cross-region comparisons. In this introductory chapter, we first review in the next section current population trends and prospect in the Asia-Pacific region and its subregions. This is followed by a brief discussion of potential impacts in Sect. 1.3. We provide an outline of the case studies that are included in this book in Sect. 1.4. The final section sums up and provides some thoughts on the future direction of regional science research to study population change and impacts in the Asia-Pacific region.

1

The discussion of Asia-Pacific demographic changes and their consequences must remain by necessity brief in this introductory chapter. For a comprehensive review with respect to Asia, see, for example, Zhao and Hayes (2017).

1 Population Change in the Asia-Pacific Region: Trends, Issues and Models

1.2

3

Population Trends and Prospects

Using the classification of the United Nations (United Nations 2019a), we define the Asia-Pacific region in this chapter as including all countries in Eastern Asia, SouthEastern Asia, Australasia and Oceania; plus all countries in Southern Asia (excluding the Islamic Republic of Iran); as well as the countries on the Pacific rim of the Americas.2 Table 1.1 lists the 2020 population estimates (in thousands) for each of the countries in the eight subregions across this Asia-Pacific region, based on the 2019 revision of world population prospects (United Nations 2019a). The Asia-Pacific region’s population of 4.9 billion accounts for 63% of the world’s population of about 7.8 billion in 2020. Of the 10 countries with the largest populations in the world (China, India, the United States, Indonesia, Brazil, Pakistan, Nigeria, Bangladesh, Russia and Mexico), seven are located in the AsiaPacific region. Table 1.1 highlights the demographic diversity across the eight subregions in terms of population size. The smallest of these subregions is Oceania, which includes a relatively large Papua New Guinea (population about nine million) but also many small island nations with populations down to one to two thousand (Tokelau and Niue). Oceania accounts for only 0.2% of the world’s population, but many small island nations are disproportionally of global concern due to the expected sea level rise as a consequence of global warming. Four atoll nations in the Asia-Pacific region (Kiribati, Tuvalu and the Marshall Islands in Oceania; and the Maldives in Southern Asia) are at greatest risk (Letman 2018). The two largest subregions, Southern Asia (of which India alone represents three quarters of the population) and Eastern Asia (with China contributing 85% of the population), account each for more than one fifth of the world population. While the United States remains for now the third most populous country in the world (with Nigeria expected to overtake it by the middle of the century), the Pacific countries of the Americas account for only 8.5% of the estimated global population in 2020. Figure 1.1 displays the average annual population growth rates of the subregions over the last 30 years (1990–2020) and compares those with the projected population growth rates over the next 30 years (2020–2050), using the medium variant projections of the United Nations (2019a). The lowest average growth rate (0.6%) over the 1990–2020 period was observed in Eastern Asia, the highest (1.9%) in Oceania. Growth is expected to decline everywhere and to become even negative in Eastern Asia ( 0.1% per annum).3 2

United Nations Population Division tabulations include the Islamic Republic of Iran in the group of countries of South Asia. We have excluded Iran from this group because this country in the Middle East is more naturally grouped with the countries that are formally referred to as Western Asia. 3 In this chapter, we use exclusively population projections produced by the United Nations Population Division. Although there is consensus among demographers about the broad trends, different assumptions may lead to somewhat different results. For example, population projections of the International Institute for Applied Systems Analysis (IIASA) expect a faster decline in fertility than the UN projections as a consequence of accelerated improvements in the education of young women in developing countries (Lutz et al. 2018).

4

J. Poot and M. Roskruge

Table 1.1 Estimated population (in thousands) of countries in the Asia-Pacific region in 2020 Southern Asia Afghanistan Bangladesh Bhutan India Maldives Nepal Pakistan Sri Lanka Eastern Asia China China, Hong Kong SAR China, Macao SAR China, Taiwan Province of China Dem. People’s Republic of Korea Japan Mongolia Republic of Korea South-Eastern Asia Brunei Darussalam Cambodia Indonesia Lao People’s Democratic Republic Malaysia Myanmar Philippines Singapore Thailand Timor-Leste Viet Nam Pacific Central America Costa Rica El Salvador Guatemala Honduras

23.8% 38,928 164,689 772 1,380,004 541 29,137 220,892 21,413 21.5% 1,439,324 7497

Pacific South America Chile Colombia Ecuador Peru Australasia Australia New Zealand Oceania (excl. Austr & NZ) Melanesia Fiji

1.5% 19,116 50,883 17,643 32,972 0.4% 25,500 4822 0.2%

896

649 23,817

New Caledonia Papua New Guinea

25,779

Solomon Islands

687

Vanuatu Micronesia Guam Kiribati Marshall Islands Micronesia (Fed. States of) Nauru Northern Mariana Islands Palau

307

126,476 3278 51,269 8.6% 437 16,719 273,524 7276 32,366 54,410 109,581 5850 69,800 1318 97,339 2.3% 5094 6486 17,916 9905

Polynesia American Samoa Cook Islands French Polynesia Niue Samoa Tokelau Tonga Tuvalu Wallis and Futuna Islands Pacific North America Canada

285 8947

169 119 59 115 11 58 18

55 18 281 2 198 1 106 12 11 4.7% 37,742 (continued)

1 Population Change in the Asia-Pacific Region: Trends, Issues and Models

5

Table 1.1 (continued) Mexico Nicaragua Panama

128,933 6625 4315

United States of America

331,003

Total Population – AsiaPacific Region Total Population – World

4,914,395

63%

7,794,799

Note: The subregional percentages refer to the subregion’s share of the world population Source: United Nations (2019a)

Eastern Asia Pacific North America Australasia South-Eastern Asia Pacific South America Pacific Central America Southern Asia Oceania -0.5

0

0.5 1990-2020

1

1.5

2

2.5

2020-2050

Note: The figure displays estimated and projected average annual growth rates calculated as ln(pop(2020) / pop(1990)) / 30 * 100% and ln(pop(2050) / pop(2020)) / 30 * 100% respectively. Projected population growth is based on the medium population projection. Source: United Nations (2019a)

Fig. 1.1 Past and projected population growth in the Asia-Pacific region, 1990–2050

The ranking of the subregions in terms of population growth is expected to change somewhat, with Australasia notably projected to have relatively higher population growth over 2020–2050 than all other subregions except Oceania. This is due to high levels of net inward international migration in Australia and New Zealand expected to continue in the coming decades. The United States and Canada (Pacific North America) experienced lower growth than Australasia over the 1990–2020 period and are expected to grow less than Australasia over the 2020–2050 period as well. The populations of mainland China, Taiwan, Japan and South Korea are all expected to decline. Especially for China, the contrast is huge: going from a positive average annual growth rate of 0.7% between 1990 and 2020 to a negative growth rate of 0.1% for 2020–2050. One of the main causes is the implementation of the one-child policy in China in the 1970s when the total fertility rate (TFR, the average

6

J. Poot and M. Roskruge

number of children is expected to have over a lifetime) was still greater than five. While the one-child policy was replaced by a two-child policy in 2016 and limits may soon be removed altogether, China’s TFR is in 2020 only 1.7 and not expected to increase much, leading to negative growth and notable population ageing. By 2050, India’s population is projected to be 1.6 billion in the medium projection. By then, India will be the country with the largest population in the world, after overtaking China’s population in the late 2020s. However, its annual population growth rate is also declining: from 1.5% for 1990–2020 to 0.6% for 2020–2050. Indonesia is expected to see a population growth rate decline of similar magnitude. The population growth across all eight Asia-Pacific subregions (as well as in other parts of the world) over the past 30 years has not been uniformly distributed across age groups, with most countries experiencing structural and numerical ageing, some more rapidly than others. This trend is projected to continue. Figure 1.2 shows the total dependency ratio (number aged 0–14 plus number aged 65+ per 100 persons aged 15–64) for each of the eight regions in 1990, 2020 and 2050 (medium variant projections). The total dependency ratio is the sum of the child dependency ratio (number aged 0–14 per 100 aged 15–64), which is displayed in Fig. 1.3, and the old-age dependency ratio (number aged 65+ per 100 aged 15–64), shown in Fig. 1.4. The total dependency ratio is a purely demographic and approximate measure of the number of individuals likely to be economically ‘dependent’ on the working-age population. It should be noted that this ratio does not account for seniors who are

90.0 80.0 70.0 60.0 50.0 40.0 30.0 20.0 10.0 0.0

1990

2020

2050

Note: The figure displays the estimated and projected population aged 0 to 14 plus those aged 65 years and over per 100 population aged 15-64. Projected ratios are based on the medium population projections. Source: United Nations (2019a)

Fig. 1.2 Total dependency ratios in the Asia-Pacific region, 1990–2050

1 Population Change in the Asia-Pacific Region: Trends, Issues and Models

7

80.0 70.0 60.0 50.0 40.0 30.0 20.0 10.0 0.0

1990

2020

2050

Note: The figure displays the estimated and projected population aged 0-14 years per 100 population aged 15-64. Projected ratios are based on the medium population projections. Source: United Nations (2019a)

Fig. 1.3 Child dependency ratios in the Asia-Pacific region, 1990–2050

50.0 45.0 40.0 35.0 30.0 25.0 20.0 15.0 10.0 5.0 0.0

1990

2020

2050

Note: The figure displays the estimated and projected population aged 65 years and over per 100 population aged 15-64. Projected ratios are based on the medium population projections. Source: United Nations (2019a)

Fig. 1.4 Old-age dependency ratios in the Asia-Pacific region, 1990–2050

8

J. Poot and M. Roskruge

working, nor for working-age people who are unemployed or not in the labour force. Changes in the dependency ratios provide an indication of the changing age structure of a population and the potential support requirements likely to arise as a result. Such support involves mostly private transfers from the middle aged to the young and a mixture of public and private transfers from the middle aged to the old. At the macro level, these transfers are formally estimated and projected by means of a methodology called National Transfer Accounts (Lee and Mason 2010). While the threshold age of 65 for calculating the old-age dependency ratio is still the most common choice (given the availability of public pensions at that age in many countries), it should be noted that particularly among those aged 65–75 (i.e. the ‘young-old’), a growing proportion are in relatively good health and are able to work longer. Figure 1.2 shows that the highest total dependency ratio among the subregions in 1990 was observed in Pacific Central America: 79.9. Total dependency has been declining between 1990 and 2020 in all subregions except in Australasia and Pacific North America. A decline in total dependency usually takes place towards the end of the demographic transition—the change from high to low rates of mortality and fertility (e.g. Lee 2003)—when fertility is already low but the labour force is large due to higher past fertility and the number of older persons may still be relatively small due to higher mortality in the past. A lower total dependency ratio implies a higher overall labour force participation rate, which may be seen as a ‘demographic dividend’ (Bloom et al. 2003). The ‘demographic dividend’ provides a window of opportunity for countries to capitalise on a bigger proportion of population at working ages. Over the next three decades, this window of opportunity still exists for Oceania and Southern Asia, subregions with relatively younger populations and higher fertility rates. In terms of a changing age structure, a huge contrast can be observed over the coming three decades between Southern Asia (dominated by India), where the total dependency ratio will continue to decline, down to a projected 48.1 in 2050, and Eastern Asia (dominated by China), where rapid population ageing is leading to a sharp increase in the total dependency ratio to 69.7 in 2050. The Eastern Asia population is expected to be more aged and having lower fertility than Australasia and Pacific North America in the future. As noted by Bloom and Williamson (1998), the demographic transition has been more dramatic in Eastern Asia during the twentieth century than in any other region or historical period, resulting in more rapidly declining child dependency ratios and increasing old-age dependency ratios in the region. The child dependency ratios were still very high in 1990 in Pacific Central America and Oceania (72.4 and 73.3 respectively). As can be seen from Fig. 1.3, all subregions experienced a decline in child dependency between 1990 and 2020 and are expected to continue to do so until 2050, although the change is relatively minor in Australasia and Pacific North America. Similarly, Fig. 1.4 shows that all subregions experienced a structurally ageing population over the last three decades and will continue to do so over the next 30 years. The increase in the old-age dependency ratio in Southern Asia and Oceania is relatively small, but huge in Eastern Asia.

1 Population Change in the Asia-Pacific Region: Trends, Issues and Models

9

Fertility levels have declined rapidly across almost all regions in the Asia-Pacific (and elsewhere in the world) in the last few decades and are projected to decline further, with the number of countries with below-replacement level fertility (2.1 births per woman, the level required for a population to maintain its size in the long run) likely to increase (see Fig. 1.5). Conversely, with continuously falling fertility rates and increasing proportions at older ages, the share of the working-age population is set to decline, resulting in a substantial projected increase in total dependency ratios over the 2020–2050 period in Eastern Asia, Pacific North America and Australasia (Lee and Mason 2011). A high total dependency ratio means a larger share of resources will be needed for a relatively less productive segment of the population, which in turn may inhibit economic growth (Bloom et al. 2003). By 2020, fertility rates have fallen below replacement level 2.1 in Eastern Asia, Australasia, Pacific North America and Pacific South America. The lowest TFR (1.1) can be found in South Korea, but with UN (2019a) expecting a small increase to 1.4 by 2050. The TFR lies between 2.2 and 2.4 in Southern Asia, South-Eastern Asia and Pacific Central America. In comparison, Oceania still has a much higher TFR of 3.5 live births per women. By 2050, it is likely that the TFR for all the Asia-Pacific subregions, except Oceania (with an expected TFR of 2.6), will have converged to below-replacement fertility between 1.7 and 1.9.

4.5 4 3.5

Southern Asia Eastern Asia South-Eastern Asia Pacific Central America Pacific South America Australasia Oceania Pacific North America

3 2.5 2 1.5

Note: Projected rates are based on the medium population projections. Source: United Nations (2019a)

Fig. 1.5 Total fertility rates in the Asia-Pacific region, 1990–2050

10

J. Poot and M. Roskruge

Given that it accounts for a large share of Eastern Asia’s population, the sharp fall in the TFR for Eastern Asia seen in Fig. 1.5 is a direct result of China’s one-child policy. China’s fertility rate of around 5.9 live births per woman in the early 1970s has dropped to a level of 1.7 by 2020, which is similar or lower than that of many developed countries, including the United States, Britain, and France and among the lowest in the world. This issue of low fertility is compounded by the social bias which favours male children in China (but also in some other Asian and Middle Eastern countries), resulting in one of the world’s most skewed sex ratios at birth by 2010 (Feng 2010). The natural sex ratio of births is 105 boys per 100 girls. United Nations (2019a) estimates suggest that China’s sex ratio at birth rose from 108 boys per 100 girls in 1990 to 117 in 2010, which, in conjunction with the one-child policy, impacted severely on the total fertility rate of the country. The infant mortality rate (IMR), measured as the number of infants dying before reaching 1 year of age per 1000 live births in a given period, is an important indicator of the health of a population. It reflects the quality of maternal and child health and medical services. It is also associated with socioeconomic conditions and public health practices. Countries with higher levels of economic development have lower IMRs as seen in Fig. 1.6, with Australasia and Pacific North America having the lowest rates in the Asia-Pacific region (less than 6 per 1000 births). Infant mortality rates have fallen dramatically in Asia over the last 30 years, with many countries, including China, India and Indonesia, experiencing declines of greater than 50% (OECD/WHO 2012). Eastern Asia saw a very substantial decline in infant mortality over the 1990–2020 period, with the IMR projected to reach similar levels to those expected in Australasia and Pacific North America by 2050. The IMR in Southern Asia and Oceania remains in 2020 the highest in the AsiaPacific region with around 35–36 infant deaths on average per 1000 live births, followed by South-Eastern Asia (23.5 deaths per 1000). The mortality rates in Southern Asia have fallen much more dramatically than in Oceania with some countries in the region—Maldives, Nepal, Bangladesh, Bhutan and Sri Lanka— reducing their IMRs by over 60% since 1990. The IMRs in Brunei Darussalam, Timor-Leste, Singapore, Thailand, Cambodia and Indonesia in South-Eastern Asia have also seen a similar decline over the last 25 years. Life expectancy at birth indicates the number of years a new-born infant would live if prevailing patterns of mortality at the time of its birth were to stay the same throughout its life. It should be noted that it is not possible to accurately predict the age-specific death rate of any birth cohort in the future, and therefore, if death rates continue to fall (as is expected), actual life spans will continue to increase. Life expectancy at birth reflects the overall mortality of a population and is a proxy measure of the overall quality of life and health status of a country. Figure 1.7 shows that life expectancy at birth increased remarkably across all subregions over the 1990–2020 period, especially in Southern Asia, which had the lowest life expectancy in 1990 (59.3 years) but which increased by 19% over the following three decades. Pacific Central America, Pacific North America and Australasia experienced the lowest 1990–2020 increase in life expectancy (7, 5 and 4% respectively), but these subregions had already high life expectancy in 1990. By 2020, life

1 Population Change in the Asia-Pacific Region: Trends, Issues and Models

90.0 80.0

11

Southern Asia Eastern Asia South-Eastern Asia

70.0 60.0 50.0

Pacific Central America Pacific South America Australasia Oceania Pacific North America

40.0 30.0 20.0 10.0 0.0

Note: Projected rates are based on the medium population projections. Infant mortality rates are defined as the number of infant deaths per 1,000 live births, 1q0. Source: United Nations (2019a)

Fig. 1.6 Infant mortality rates in the Asia-Pacific region, 1990–2050

expectancy at birth reached 83.8 years in Australasia (compared with 79.5 years in Pacific North America). This increase in life span is mainly due to rising living standards, better nutrition, water and sanitation, increased education and better access to quality health services (OECD/WHO 2012). Eastern Asia and Pacific Central and Southern America are likely to reach a life expectancy well above 80 years by 2050. In contrast, life spans are much lower in countries in Oceania (lowest in Melanesia), Southern Asia (lowest in Afghanistan, Pakistan, India and Bangladesh) and South-Eastern Asia (lowest in Lao People’s Democratic Republic, Myanmar, Cambodia and Timor-Leste). Lifespan improvement is expected to be the slowest in Oceania, with this region possibly having the lowest life expectancy in the region by 2050. Life expectancy is strongly linked to economic development, which is—in turn— affected by urbanisation, the increase in proportion of a country’s population living in urban areas, predominantly driven by net rural to urban migration. There is considerable evidence that, in the modern knowledge and service-driven economy, cities offer many spillover benefits to resident firms and households (so-called

12

J. Poot and M. Roskruge

100.0 90.0 80.0 70.0 60.0 50.0 40.0 30.0 20.0 10.0 0.0

1990-1995

2020-2025

2050-2055

Note: Projected life expectancies are based on the medium population projections. Source: United Nations (2019a)

Fig. 1.7 Life expectancy at birth in the Asia-Pacific region, 1990–2050

agglomeration advantages, see Glaeser (2011) and de Groot et al. (2016)). Urbanisation can be linked to economic growth, a better educated and more productive labour force, greater environmental sustainability through more efficient use of resources and effective regulation, as well as improved social welfare through better access to services. Many cities are also expected to become even more ethnically diverse (Poot and Pawar 2013). On the other hand, urbanisation does not necessarily result in an equitable distribution of wealth and well-being. Additionally, there are considerable challenges with respect to the availability of housing and shared resources like transportation, schools, etc. (e.g. Tacoli 2012). Since 2007, more than half of the world’s population has been living in urban areas. In 1950, 30% of the world’s population was urban, and by 2050, 68% of the world’s population is projected to be urban (United Nations 2018). Asia—particularly Eastern, South-Eastern and Southern Asia—has urbanised faster than any other regions of the world (ESCAP 2013). Given cross-country differences in the organisation and boundaries of local government, cities are best compared in terms of metropolitan or functional areas.

1 Population Change in the Asia-Pacific Region: Trends, Issues and Models

13

Even so, estimates of city populations vary considerably. Nonetheless, Asian cities dominate any list of large metropolitan areas (led by Tokyo, with Shanghai a close second, and this order is likely to reverse—see, e.g., Poot and Vloeimans 2015). In contrast, Oceania still has a relatively much smaller proportion (less than one-quarter) of its population living in urban areas, and this is unlikely to change in any substantial away over the next 30 years (see Fig. 1.8). Australasia, along with the Pacific rim of the Americas, is already highly urbanised, and projections show that the urban population in these regions is likely to increase to more than four fifths of the population living in urban areas by 2050. However, the most striking feature of Fig. 1.8 is the urbanisation of Eastern Asia: from 34% in 1990 to more than 80% by 2050. With the majority of the countries in Eastern Asia (Hong Kong SAR, Macao SAR, Japan, and Republic of Korea) already highly urbanised, the change in the population-weighted average for this subregion is mainly driven by the extremely rapid urbanisation of China. In other parts of Asia, there are also large differences in urbanisation between countries. For example, parts of South-Eastern Asia (namely Singapore, Malaysia and Brunei Darussalam) are predominantly

100.0 90.0 80.0 70.0 60.0 50.0 40.0 30.0 20.0 10.0 0.0 1990 1995 2000 2005 2010 2015 2020 2025 2030 2035 2040 2045 2050 Southern Asia Eastern Asia South-Eastern Asia

Pacific Central America

Pacific South America

Australasia

Oceania

Pacific North America

Source: United Nations (2018)

Fig. 1.8 Percentage of the population residing in urban areas in the Asia-Pacific region, 1990–2050

14

J. Poot and M. Roskruge

urban, while countries like Cambodia, Timor-Leste, Viet Nam and Myanmar still have two thirds or more of the population living in rural areas. The final topic in this review of demographic trends in the Asia-Pacific region is international migration. The United Nations Population Division provides estimates of the migrant stock (number of people born in a country other than that in which they live) for each country based on official statistics on the foreign-born or foreign population (including foreign nationals born in the host country). The foreign-born population covers all people who have ever migrated from their country of birth to their current country of residence. The difference across countries between the size of the foreign-born population and that of the foreign population depends on the rules governing the acquisition of citizenship or permanent residency in each country. The international migrant stock in 2019 is estimated to be about 270 million people globally, of whom about 100 million reside in the Asia-Pacific region (United Nations 2019b). Table 1.2 shows the foreign migrant stock as a percentage of the population in the eight Asia-Pacific subregions in 1990, 2005 and 2019. Australasia has the highest proportion of its population being foreign born (29% in 2019), followed by Pacific North America (16%). In both regions, this proportion has increased considerably over the last three decades. Although the share of migrants in the other regions of the Asia-Pacific is small, there is considerable variation among countries within these regions in terms of foreign or foreign-born residents among the population. In Eastern Asia, for example, well over one third of the population of Macao (China) and Hong Kong (China) are born overseas, whereas this proportion is negligible in other countries across this region. Similarly, Maldives in Southern Asia, and Brunei Darussalam and Singapore in South-Eastern Asia, have much higher proportions of migrant populations compared to the neighbouring countries in their region. In Oceania, only 1.1% of the population of Melanesia are migrants, whereas this proportion is much higher in Micronesia (21.7%) as well as Polynesia (10.1%). It is also notable that the immigrant share of population has been declining in Oceania, as well as in Southern Asia. In general, the share of migrants as a proportion of the total population is higher in richer countries in the Asia-Pacific region, exceptions being countries like Japan and Republic of Korea where laws around immigration and citizenship remain Table 1.2 Migrant stock as a percentage of the population in the Asia-Pacific region, 1990, 2005 and 2019

Subregion Eastern Asia Pacific South America South-Eastern Asia Southern Asia Pacific Central America Oceania Pacific North America Australasia Source: United Nations (2019b)

1990 0.29 0.44 0.71 1.46 1.69 3.12 9.89 22.04

2005 0.40 0.65 1.23 0.75 0.92 2.91 13.88 23.56

2019 0.48 2.72 1.54 0.62 1.05 2.54 16.00 28.74

1 Population Change in the Asia-Pacific Region: Trends, Issues and Models

15

restrictive. It is remarkable that Eastern Asia, which has the fastest ageing of the population in the region (recall Fig. 1.4) and is experiencing or expecting population decline (see Fig. 1.1), has also the lowest rates of immigration, that is, immigration is not yet seen as a means to lower the total dependency ratio at least temporarily.

1.3

Impacts of Demographic Change in the Asia-Pacific Region

The changes in size, age composition and spatial distribution of subregional populations in the Asia-Pacific region that have been outlined in the previous section have wide-ranging economic, social and environmental consequences. It is impossible to review all of these in the limited space available in this introductory chapter. Instead, we will simply highlight some selected issues that have been brought to the fore in the media and in policy debates in recent years. Even when restricting the impacts of demographic change to just economic ones, one would still have to distinguish between demand-side effects (e.g. on specific markets and public services) and supply-side effect (on labour supply). Additionally, a distinction between macro-level effects (such as on savings, investment and public debt) and micro-level effects (e.g. differences in consumption patterns across age groups and varying locations) is also essential. Furthermore, short-run effects on markets and the public sector are likely to be very different from long-term effect on technological change and economic growth. Finally, one might want to bring into the analysis effects on the environment such as land use, pollution and congestion, and effects on health and well-being.4 While in principle possible, it remains a huge challenge to bring all, or even just some, of these effects together in some form of integrated modelling, such as by means of computable general equilibrium (CGE) models. Recent examples of using CGE models to assess the impacts of the rapid population ageing in China are Wang et al. (2017) and Wei et al. (2018). In the present book, Tokunaga and Okiyama use in Chap. 4 a dynamic regional CGE model to assess the impacts of rapid depopulation in some regions of Japan. Tran et al. apply in Chap. 7 a multiregional CGE model to test the effectiveness of a regional-specific immigration policy in Australia. The impacts of population ageing, the dominant feature of Asia-Pacific population change—as we saw in the last section—are generally considered to be problematic. One of the main economic impacts is the need for increased public funding of the consumption of those no longer working, given that private savings are unlikely to be adequate in many circumstances and familial support is also becoming less (particularly where family members have migrated to large cities). Despite (or perhaps because of) continued improvements in healthy life expectancies, there will also be a growing need for care and carers of the aged (e.g., Phillips 2018). 4

See, for example, Hayes (2013, 2017) on sustainable development in the Asia-Pacific context.

16

J. Poot and M. Roskruge

Additionally, a considerable reallocation of public funding to the healthcare sector is inevitable. Population ageing is also contributing to depopulation of rural and peripheral urban areas, when the young in those areas are migrating to the metropolitan areas that benefit from agglomeration advantages. This is having undesirable consequences for provincial centres and rural areas. This is particularly felt in Japan,5 but also in other countries in the Asia-Pacific region, including in North America and Australasia. While depopulation is generally seen as very problematic, it has been argued that there are also ways to benefit from this situation, in the form of a ‘depopulation dividend’ (Matanle 2017). Li et al. (2019) argue that rural areas with population shrinkage can ‘reinvent’ themselves by outward orientation (towards urban markets) of new economic activities, local entrepreneurship and social capital. Another expected impact of ageing is that of lower productivity and economic growth. A shortage of labour may encourage employers to invest in labour-saving technological change (Acemoglu and Restrepo 2018). This may lower returns on capital investment, and real interest rates are then also expected to decline (Arslanalp et al. 2018). Additionally, innovation may be less as well given that it is particularly the young who contribute most to R&D. However, the evidence on productivity effects is mixed (for a survey see, e.g., Poot 2008). It is nonetheless also possible to identify some benefits from population ageing. For example, as long as they remain healthy, older people are known to have a relatively high level of subjective well-being, that is, they are happier (e.g., Steptoe et al. 2015). It can also be argued that aged societies may have lower levels of conflict and crime and higher levels of social cohesion, given that large cohorts of young people may, particularly when unemployment is high and incomes are low, exacerbate conflict and unrest. Countering this argument is that older people have often small personal network and a higher incidence of loneliness, which may lower an area’s social cohesion. Older populations are also less likely to protest against corruption (Farzanegan and Witthuhn 2017). With respect to potential policy responses that might mitigate the negative impacts of population ageing, it is unlikely that demographic solutions, that is, policies that increase fertility and/or migration, are going to be effective (Parsons and Gilmour 2018). Fertility rates in Eastern Asia are so low that pro-birth policies in the form of tax incentives or better childcare facilities are unlikely to bring back fertility rates to close to the replacement level. Migration can slow down the ageing of the population, given that migrants are predominantly young single adults or young families. However, this effect is only temporary unless migrants have much higher fertility than the host population. If their fertility converges to that of the host

5 See, for example, the article ‘Rural Japan: old, shrinking and broke’ in The Economist of June 29, 2019. It is important to note that even some suburban areas in otherwise still growing metropolitan areas may also be ageing rapidly, see ‘Demography in Japan: a negative-sum game’ in The Economist of January 7, 2017 (www.economist.com, accessed 11 Nov 2019).

1 Population Change in the Asia-Pacific Region: Trends, Issues and Models

17

population, which is often the case, a migration wave can accelerate population ageing in the long run. The benefits of migration would then need to be seen in terms of the contribution migrants can make to economic growth through diversity and knowledge transfer (Brunow et al. 2015). Instead of policies aiming at fertility and migration, the negative economic impact of population ageing can be more effectively addressed by policies that encourage a prolonging of the working life of those at work (but in meaningful and suitable employment), an increase in female labour force participation,6 and gains in labour productivity that at least partially offset the increasing old-age dependency ratio. Recent research suggests that, in terms of productivity growth, countries with large populations (hence specifically China and India in the Asia-Pacific region) may be at an advantage in the future. Certainly, the scale of the US population has contributed to its technological dominance over the last century. Desmet et al. (2018) argue that more populous countries should be more innovative than less populous ones because the returns to developing a new technology are greater in the former countries. Over the last millennium, however, it has often been relatively smaller countries and their colonies (think, e.g., the Netherlands in the seventeenth century) that have had the highest technological change and, therefore, incomes. Net migration has often been from populous countries with low per capita income to the small but rich ones. Desmet et al. (2018) expect that if migration from poor countries to rich ones remains restricted, Asia and sub-Saharan Africa might become the engines of innovation in the very long run (but requiring half a millennium or so to reach that stage in their model). However, according to these authors, the best global growth path would be obtained when all barriers to migration would be eliminated, which might raise global welfare threefold. Ironically, however, countries that are ageing the fastest and facing population decline are also often the most opposed to immigration. While such countries may permit some temporary migration and return migration of their diaspora, the main solution to labour shortages is often seen in the form of labour-saving technological change in industrial production but also in caring for the large number of older people.7 Acemoglu and Restrepo (2018) find that ageing of the work force (resulting in relatively more older workers relative to middle-aged workers) increases the use of robots and other automation technologies. The active encouragement of international migration is still mostly restricted to North America and Australia, as we saw in Table 1.2.8 Future levels of international migration remain difficult to forecast as they are strongly dependent on policy development and on responses of populations detrimentally affected by climate

6 Note that in India the trend is actually the opposite: the proportion of women working full-time has been sharply declining, see, for example, Desai and Joshi (2019). 7 See, for example, Flandorfer (2012) for a survey on the growing use of robots for the care of older people. 8 However, as noted in the previous section, the subregional averages do hide large differences between individual countries. For example, about 40% of Singapore’s population is foreign born.

18

J. Poot and M. Roskruge

change (e.g. Beine and Parsons 2015). As noted above, the removal of restrictions on international migration is likely to have large global economic benefits, but the external economic effects and noneconomic effects are complex (see, e.g., Nijkamp et al. 2012), and a review of these is beyond the scope of this chapter.

1.4

Outline of the Book

The chapters in this book offer, by means of case studies, various perspectives on population change and impacts in the Asia-Pacific region. The contributions also offer several methodological approaches ranging from descriptive accounts to formal modelling. The adopted approaches have in common that they are predominantly quantitative rather than qualitative. Following this introductory chapter, the 14 case studies in this book have been grouped into three parts. The four chapters of Part I of the book are concerned with issues of the distribution of population and economic activity. Part II of the book consists of six chapters that focus on causes and consequences of internal mobility and cross-border migration. The four chapters of Part III of the book focus on aspects on population ageing in the two most populous countries in the region, China and India, as well as on elder care and on causes of low fertility. In the remainder of this section, we provide an outline of each chapter.

1.4.1

Population Distribution

As described earlier in this chapter, the populations of the subregions of the AsiaPacific region have been changing rapidly through spatially specific trends in fertility, mortality and migration. The four chapters in part I of this book focus on patterns of population distribution within selected countries: China, Colombia, Japan and Australia respectively. A common theme through all contributions is that of urbanisation and agglomeration. Chapters 2 and 3 are concerned with the distribution of cities in terms of their populations and how these distributions change over time. For a survey, see, for example, Gabaix (2016). Of particular interest is Zipf’s law that suggests that city populations fit a Pareto distribution with a parameter equal to one. In practice, this means that the biggest city should be twice as big as the second largest city, three times the third one and so on. Another law, Gibrat’s law, says that city populations grow randomly at growth rates which have the same mean and variance, and that the population growth rate and city size are independent of each other. In that case, small and big cities grow on average at the same rate. Of course, both ‘laws’ are statistical properties that may or may not fit the actual distribution and dynamics of city populations in any given country.

1 Population Change in the Asia-Pacific Region: Trends, Issues and Models

19

The study by John Gibson and Chao Li in Chap. 2 presents the evolution of the population and area size distribution of Chinese cities, using population census data from 2000 to 2010, combined with remote sensing data. Using Pareto’s law as the benchmark, the authors suggest that the very largest cities in China appear to have scope to absorb more migrants, in contrast with the pro-small bias in urban policy during the last decade. Gibson and Li find that migrants without local hukou registration increasingly congregate in a few large cities. This means that previous studies that rely on the count of local hukou holders suggest that the city size distribution is more even than it actually is. Comparisons of the distributions over time show that the city size distribution is diverging in terms of the urban resident population but converging in terms of land area per city. These patterns suggest that growth in the resident population of large cities is not being assisted by fast enough area expansion, while area expansion of less populous cities is too fast for their slower growth in resident numbers. Gerson Javier Pérez-Valbuena and Adolfo Meisel-Roca continue in Chap. 3 with this theme by exploring urban population growth in Columbia. By using census data over a very long period, namely from 1835 to 2005, their chapter studies the urban hierarchy in Colombia and its main regions. They also investigate whether there is coincidence between national and regional population patterns, as suggested by Gabaix (1999). Interestingly, Pérez-Valbuena and Meisel-Roca find that the urban population growth process changed since the 1950s. Before then, small cities grew at different speeds compared to medium and big ones, violating Zipf’s law and Gibrat’s law. Since the 1950s, Columbian cities grow at the same expected rate irrespective of their size. However, the national trend is reflected in only two out of the four regions analysed. The stochastic population growth model suggests that policies are needed to influence city population growth, for example, policies that improved quality of life and urban amenities. Chap. 4, by Suminori Tokunaga and Mitsuru Okiyama, investigates the types of policy measures needed to revitalise Japan’s economy in an era of depopulation, but without exacerbating disparities among regional economies. The authors employ a recursive-dynamic six-region computable general equilibrium model (D6SCGE) to draw a picture of the future of regional economies by 2040 when taking Japan’s demographic projections of continued population ageing, decline and regional depopulation into account. Tokunaga and Okiyama find that it is essential that Japan raises it productivity growth to combat the adverse demographic trends. Their modelling suggests that new industrial clusters and new infrastructure will be needed. Additionally, targeted policies will have to address the increasing gap between the metropolitan core regions and the peripheral regions. In Chap. 5, Robert Tanton and Yogi Vidyattama explore the integration of demographic data into spatial decision-making. They note that regional population forecasts are normally conducted using a purely demographic model that can only generate forecasts of size and age-sex structure of the population. Such demographic models do not usually incorporate spatial information that might indicate the capacity of an area to support differing population sizes and densities depending on physical, environmental and economic factors. Tanton and Vidyattama, therefore,

20

J. Poot and M. Roskruge

argue in favour of the use of a Spatial Decision Support System (SDSS) for forecasting urban populations at a much more detailed level, incorporating some of these conditions, boundaries and limits. The chapter describes how a spatial microsimulation method can be used as an input into the SDSS. The authors provide a potential application for Australia’s capital city, Canberra. They find that spatial microsimulation models can be used to create the synthetic person-level base file that can then be used in the SDSS. By operating at the person level, the SDSS can implement a number of complex boundaries, including physical, environmental and economic boundaries.

1.4.2

Migration and Development

Part II of this book, containing six chapters, is concerned with causes and consequences of international migration and internal mobility. Francisco Rowe and Martin Bell investigate in Chap. 6 a less studied aspect of population flows, namely the drivers of long-distance commuting. Long-distance commuting has emerged as an alternative to migration to equilibrate imbalances across spatial labour markets. Coupled to changes in the labour and housing market, technological advances (including much cheaper air travel) have promoted long-distance commuting. This changes the links between the spatial distribution of population and regional economies. Rowe and Bell focus on Chile. They examine micro-census data and use regression analysis to show how contextual factors have shaped long-distance commuting in Chile. They find that the nature and spatial distribution of mining and construction activities have played a key role in promoting long-distance commuting in Chile. They also conclude that rising female labour force participation and home ownership have exerted an influence that is different from what is found in various European and North American countries. Nhi Tran, Elizabeth Roos, James Giesecke and John Madden focus in Chap. 7 on the role of temporary visas as a policy instrument for encouraging population growth in regions of Australia that face demographic stagnation. Preferential treatment in granting temporary visas to migrants agreeing to work in nonmetropolitan regions has become an important instrument of Australia’s regional development policy. This policy is analysed in this chapter with a 15-region dynamic CGE model, which identifies workers by skill class, employment status, region, age group and visa status. The model captures intertemporal transitions between labour categories and the interaction between regional labour demand and supply in determining interregional migration and regional employment. Simulation results show that targeting the temporary visa programme at nonmetropolitan regions does result in an improvement in regional development indicators for the targeted regions. However, labour-market displacement effects—principally increased net interregional emigration of existing workers—substantially diminishes the regional development benefits over the 4-year period for which the temporary visas are granted.

1 Population Change in the Asia-Pacific Region: Trends, Issues and Models

21

Nevertheless, some regional stimulus effects do persist over subsequent years with small positive effects remaining at the end of the 10-year simulation period. Natasha Duncan, Jacques Poot and Brigitte Waldorf explore in Chap. 8 the dynamics of migration between Australia and New Zealand. The authors note that international migration is no longer unidirectional and permanent, but often circular: expatriates are returning to home countries and/or moving on to other destinations. Circular migration flows are influenced by host and home countries’ migration policies and efforts at (re)attracting migrants. Increasingly, governments aim to enact policies that facilitate the (re)entry of high-skilled migrants, thereby encouraging circulation. Duncan, Poot and Waldorf model emigration and return migration when diaspora has a policy-influenced propensity to return, but when the size of the diaspora population in the host country acts as an attractor for further emigration from the home country. This type of model generates nonlinear dynamics in the size of the home population and the diaspora. The model is calibrated with more than a century of data on migration between New Zealand and Australia. It is possible to broadly replicate the historical evolution of the population and migrant stocks and to show how policies can influence the future trajectories. Chapter 9 by Angelina Tang, Francisco Rowe, Jonathan Corcoran and Alessandra Faggian adds further to our understanding of the role of international migration in the Asia-Pacific region by exploring how Indian and Chinese students of Australian tertiary institutions (the largest groups of international students in that country) fare following graduation. The authors use data from the 2008–2012 Australian Graduate Surveys. They find that, compared to their domestic counterparts, both Chinese and Indian graduates have relatively poor labour-market outcomes, that is, they experience relatively lower salaries and higher levels of job-education mismatch. However, this situation has steadily improved over the 15-year period to 2012, which the authors attribute to shifts in immigration policies. Sholeh Maani and Michael Tse undertake in Chap. 10 an analysis of the impacts of immigration on the New Zealand labour market. This is an important topic for New Zealand, given that net immigration rates remain high and more than one-quarter of the population was born abroad. Maani and Tse use a methodology introduced by Borjas (2003). Applying a panel analysis of skill groups, they estimate the effects of immigrant supply shocks on native earnings and employment. Individual-level data were obtained from the New Zealand Income Surveys 2002–2007. Defining skill groups based on education and work experience, changes in the economic opportunities of native skill groups are examined that are due to the supply of immigrant workers. The results indicate that immigration has little impact on the economic opportunities of native workers. However, as has been found in related literature, the biggest impact of a new wave of immigrants is felt by earlier immigrants of similar skills and origin. In the last chapter of Part II, Chap. 11, Tomaz Dentinho, Pedro Damião and Maria da Conceição Rego focus on the interaction between migration and regional development by means of a case study of Timor-Leste. The authors formulate a development-migration model that explains development as a function of tradable and non-tradable basic employment and its productivity. Cumulative migration is

22

J. Poot and M. Roskruge

modelled as a function of differences between the origin and destination regions in terms of population and basic employment, output per capita and the distance between regions. The authors show that Dili, the capital of East Timor, is attracting much of the employment and growth of the country. This is not only due to the concentration of basic employment in Dili as a result of the rents associated with the oil industry, but also due to Dili’s absorption of most of the non-tradable basic employment that can be linked to the spatial allocation of public expenditure.

1.4.3

Population Age Composition and Impacts

Part III focuses on population ageing in the Asia-Pacific region. Chap. 12 by Guoping Mao, Fuzhong Lu, Xuchun Fan and Debiao Wu explores the ageing population in China. The median age of the population of China (38 years) will overtake the median age in the United States in 2020. Over the 2020–2045 period, the percentage of China’s population over the age of 65 will more than double from 12 to 25%. The chapter by Mao et al. presents the current demographic situation in China at the national and regional (31 regions) level in terms of descriptive statistics. Regional data are also provided on health and financial support of the elderly. Mao et al. identify a considerable range of problems to be addressed as a result of population ageing. A key issue is the need for the state and/or private markets to take over the traditional roles of family support and community care. The systems of pension provision, healthcare and medical care will all need to be reviewed, particularly in the light of the future trends that are sketched in the chapter. Arunika Agarwal, Alyssa Lubet, Elizabeth Mitgang, Sanjay Mohanty and David Bloom provide in Chap. 13 an in-depth analysis of population ageing in India— currently still the second of the world’s demographic superpowers, but overtaking China within a decade. As in China, increasing longevity and falling fertility have resulted in rapid numerical and structural population ageing (although fertility rates do remain much higher than in China). This change triggered complex health, social and economic challenges to which this diverse and heterogeneous country must rapidly adapt. As in the chapter on China, Agarwal and co-authors first describe demographic changes and then review the major challenges ahead. They point to the interconnected areas of health (especially a growing burden of non-communicable diseases), gender (viz. the needs and vulnerabilities of an increasingly female older adult population) and income security. The authors provide an overview of initiatives to adapt to population ageing and provide support to older adults and their families. They provide some policy recommendations but warn that to successfully address the challenges that population ageing pose for India will require ambitious changes and innovations in health, fiscal and social policies. Philip Morrison looks in Chap. 14 at the distinction between the actual number of children Asian couples have and the desired number of children. The latter tends to be greater than the former, implying a so-called child deficit. Morrison measures this child deficit by means of data from successive World Values Surveys (WVSs) that

1 Population Change in the Asia-Pacific Region: Trends, Issues and Models

23

have been conducted since the 1980s in 14 Asian countries. WVSs include questions on family values, fertility intentions and number of children. Morrison argues that the values which support the ideal family size change more slowly than the externally imposed constraints on the number actually born. He estimates logistic regression models to identify the determinants of the number of children an Asian couple would ideally have. He finds that, while demographic and socioeconomic variables play a predictable role in the generation of the ideal and actual number of children, the characteristics of countries themselves matter considerably. Specifically, the contrast between East & South-East Asia versus South Asia is great in this respect. Therefore, the geographical, historical and institutional context remain fundamental in understanding the level and dynamics of the child deficit across Asia. The final chapter of this book, Chap. 15 by Amara Soonthorndhada, Denis Akankunda Bwesigye, Jeerawan Hongthong and Wannee Hutaphat, focuses on elder care by means of a case study of (potential) elder abuse and neglect in Thailand. This is a rather sensitive topic everywhere and perhaps even more so in Asia. It is hard to obtain objective data and for policymakers to acknowledge that there is a significant problem. Among the many problems that vulnerable older people may face are abuse or neglect. In Thailand, the most common cause of elder abuse is psychological abuse, followed by intentional neglect. The chapter explores the role of the health status of caregivers by means of a survey of 348 caregivers looking after impaired adults aged 60 and over in Thailand. Soonthorndhada et al. find that around 40% of caregivers experienced some stress within the month prior to the survey, which has the potential to lead to elder abuse or neglect. However, there are notable gender differences. The authors conclude that there is a need for more community-based research that can highlight the needs of both caregivers and care recipients. There should also be more technical training available to long-term caregivers.

1.5

Final Thoughts

We have shown in this chapter that the Asia-Pacific region is demographically very diverse—with populations ranging from small island microstates to the two most populous countries in the world. Yet, there are also several common trends. Population growth is declining everywhere (and negative in Eastern Asia) and populations are numerically and structurally ageing everywhere, reflected in declining child dependency ratios and increasing old-age dependency ratios—which together imply increasing total dependency ratios, except in Southern Asia, where a demographic dividend can still be reaped. The underlying drivers include declining fertility everywhere in the region (but remaining relatively high in Oceania), increasing life expectancy and declining infant mortality (particularly in Southern Asia). Continuing urbanisation, accelerated in Eastern Asia, is also a dominant feature in the region. Finally, a relatively large

24

J. Poot and M. Roskruge

demographic impact of international migration remains for now largely restricted to Australasia and North America. Given that it aims to provide a multidisciplinary and analytical social science approach to regional, urban and rural issues, there is much scope for further work in regional science on trends and impacts of demographic change in the Asia-Pacific region. For example, it is increasingly popular to replace conventional cohortcomponent population projections with classic or Bayesian probabilistic projections. Of particular interest would be the development of multiregional and multigroup projections at different spatial scales that go beyond the classic age–sex composition but also include marital status, ethnicity, education, labour force and household types. A greater emphasis on families and their networks is also beneficial (e.g. Schoen 2019). The modern projections approach and that of microsimulation and agent-based modelling would be natural techniques for those in regional science to apply. When moving in this direction, the regional science approach to population research has much in common with what Raymer et al. (2019) recently advocate as the task of spatial demography: a multiregional modelling approach in which age, cohort and location are central. Regional science has in fact stood at the birth of this approach with the seminal work by Andrei Rogers published in Papers in Regional Science in 1966, before his influential book was published 2 years later (Rogers 1968). Exciting new developments in data availability, such as geo-referenced integrated administrative data and ‘big data’ generated by internet use, social media and location monitoring, provide exciting opportunities for innovative work. In statistical modelling, spatial econometrics of demographic data is also increasingly popular (e.g. Firmino Costa da Silva et al. 2017). For policy development, effective integrated modelling of geophysical, environmental, demographic and economic subsystems offers both opportunities and challenges (Rutledge et al. 2008). In all of these modelling developments, we argue for a greater emphasis on subnational and multiregional population analysis and projections. Acknowledgement We thank Shefali Pawar of the National Institute of Demographic and Economic Analysis at the University of Waikato in New Zealand for research assistance with writing Sect. 1.2.

References Acemoglu D, Restrepo P (2018) Demographics and Automation. NBER Working Paper 24421. National Bureau of Economic Research, Cambridge, MA Arslanalp S, Lee J, Rawat U (2018) Demographics and Interest Rates in Asia. IMF Working Paper WP/18/172. International Monetary Fund, Washington, DC Beine M, Parsons C (2015) Climatic factors as determinants of international migration. Scand J Econ 117(2):723–767 Bloom DE, Williamson JG (1998) Demographic transitions and economic miracles in emerging Asia. World Bank Econ Rev 12(3):419–455

1 Population Change in the Asia-Pacific Region: Trends, Issues and Models

25

Bloom D, Canning D, Sevilla J (2003) The demographic dividend: a new perspective on the economic consequences of population change. Population Matters (A RAND Program of Policy-Relevant Research Communication), Santa Monica, CA Borjas GJ (2003) The labour demand curve is downwards sloping: re-examining the impact of immigration on the labor market. Q J Econ 118(4):1335–1374 Brunow S, Nijkamp P, Poot J (2015) The impact of international migration on economic growth in the global economy. In: Chiswick BR, Miller PW (eds) Handbook on the Economics of International Migration, vol 1B. Elsevier, Amsterdam De Groot HLF, Poot J, Smit MJ (2016) Which agglomeration externalities matter most and why? J Econ Surv 30(4):756–782 Desai S, Joshi O (2019) The paradox of declining female work force participation in an era of economic growth. Indian J Labour Econ 62(1):55–71 Desmet K, Nagy DK, Rossi-Hansberg E (2018) The geography of development. J Polit Econ 126 (3):903–983 ESCAP (2013) Urbanization trends in Asia and the Pacific. Economic and Social Commission for Asia and the Pacific. United Nations, New York Farzanegan MR, Witthuhn S (2017) Corruption and political stability: does the youth bulge matter? Eur J Polit Econ 49:47–70 Feng W (2010) China’s population destiny: the looming crisis. Curr Hist 109(728):244–251 Firmino Costa da Silva D, Elhorst JP, da Mota Silveira Neto R (2017) Urban and rural population growth in a spatial panel of municipalities. Reg Stud 51(6):894–908 Flandorfer P (2012) Population ageing and socially assistive robots for elderly persons: the importance of sociodemographic factors for user acceptance. Int J Popul Res 2012:829835. https://doi.org/10.1155/2012/829835 Gabaix X (1999) Zipf’s law for cities: an explanation. Q J Econ 114(3):739–767 Gabaix X (2016) Power laws in economics: an introduction. J Econ Perspect 30(1):185–205 Glaeser EL (2011) Triumph of the city: how our greatest invention makes us richer, smarter, greener, healthier and happier. Penguin Press, London Hayes A (2013) Population dynamics and sustainable development in Asia and the Pacific. Asia Pac Popul J 28(1):57–83 Hayes AC (2017) Population and environment in Asia. Chapter 25. In: Zhao Z, Hayes AC (eds) Routledge handbook of Asian demography. Routledge, London Lee R (2003) The demographic transition: three centuries of fundamental change. J Econ Perspect 17:167–190 Lee R, Mason A (2010) Some macroeconomic aspects of global population aging. Demography 47 (S1):S151–S172 Lee S-H, Mason A (2011) International migration, population age structure and economic growth in Asia. Asian Pac Migr J 20(2):195–213 Letman J (2018) Rising seas give island nation a stark choice: relocate or elevate. Natl Geogr November 19 Li Y, Westlund H, Liu Y (2019) Why some rural areas decline while some others not: an overview of rural evolution in the world. J Rural Stud 68:135–143 Lutz W, Goujon A, KC S, Stonawski M, Stilianakis N (2018) Demographic and human capital scenarios for the 21st century: 2018 Assessment for 201 Countries. Publications Office of the European Union, Luxembourg Matanle P (2017) Towards an Asia-Pacific ‘depopulation dividend’ in the 21st century: regional growth and shrinkage in Japan and New Zealand. Asia Pac J Jpn Focus 5(6):5 Nijkamp P, Poot J, Sahin M (eds) (2012) Migration impact assessment: new horizons. Edward Elgar, Cheltenham OECD/WHO (2012) Infant mortality. In: Health at a glance: Asia/Pacific. OECD, Paris Parsons AJQ, Gilmour S (2018) An evaluation of fertility- and migration-based policy responses to Japan’s ageing population. PLoS One 13(12):e0209285. https://doi.org/10.1371/journal.pone. 0209285

26

J. Poot and M. Roskruge

Phillips DR (2018) Asia-Pacific and global population ageing. In: Klassen TR, Higo M, Dhirathiti NS, Devasahayam TW (eds) Ageing in Asia-Pacific: Interdisciplinary and comparative perspectives. Routledge, London, pp 25–45 Poot J (2008) Demographic change and regional competitiveness: the effects of immigration and ageing. Int J Foresight Innov Policy 4(1/2):129–145 Poot J, Pawar S (2013) Is demography destiny? Urban population change and economic vitality of future cities. J Urban Manage 2(1):5–23 Poot J, Vloeimans EE (2015) ‘Urban World Idol’: could Shanghai rank #1 by 2050? Int J Glob Environ Iss 14(1/2):40–54 Raymer J, Willekens F, Rogers A (2019) Spatial demography: a unifying core and agenda for further research. Popul Space Place 25:e2179. https://doi.org/10.1002/psp.2179 Rogers A (1966) Matrix analysis of interregional population growth and distribution. Pap Reg Sci Assoc 18:177–196 Rogers A (1968) Matrix analysis of interregional population growth and distribution. University of California Press, Berkeley Rutledge DT, Cameron M, Elliot S, Fenton T, Huser B, McBride G, McDonald G, O’Connor M, Phyn D, Poot J, Price R, Scrimgeour F, Small B, Tait A, van Delden H, Wedderburn ME, Woods RA (2008) Choosing regional futures: challenges and choices in building integrated models to support long-term regional planning in New Zealand. Reg Sci Policy Pract 1 (1):85–108 Schoen R (ed) (2019) Analytical family demography. Springer, Berlin Steptoe A, Deaton A, Stone AA (2015) Subjective wellbeing, health, and ageing. Lancet 385 (9968):640–648 Tacoli C (2012) Urbanization, gender and urban poverty: paid work and unpaid carework in the city. Population and Development Branch, United Nations Population Fund (UNFPA), New York United Nations (2018) 2018 revision of world urbanization prospects. Department of Economic and Social Affairs, Population Division Online Edition. Rev. 1. https://population.un.org/wup/. Accessed 4 Nov 2019 United Nations (2019a) 2019 Revision of world population prospects. Department of Economic and Social Affairs, Population Division Online Edition. Rev. 1. https://population.un.org/wpp/. Accessed 4 Nov 2019 United Nations (2019b) 2019 Revision of the International Migrant Stock. Department of Economic and Social Affairs, Population Division. https://www.un.org/en/development/desa/population/ migration/data/estimates2/estimates19.asp. Accessed 7 Nov 2019 Wang X, Chen KZ, Robinson S, Huang Z (2017) Will China’s demographic transition exacerbate its income inequality? CGE modeling with top-down microsimulation. J Asia Pac Econ 22 (2):227–252 Wei T, Zhu Q, Glomsrød S (2018) Ageing impact on the economy and emissions in China: a global computable general equilibrium analysis. Energies 11(4):817. https://doi.org/10.3390/ en11040817 Zhao Z, Hayes AC (eds) (2017) Routledge handbook of Asian demography. Routledge, London

Part I

Population Distribution

Chapter 2

Pareto’s Law and City Size in China: Diverging Patterns in Land and People John Gibson and Chao Li

Abstract Using Pareto’s law as a benchmark, the largest cities in China appear to have scope to absorb more migrants, contrary to the pro-small bias in urban policy. We use population census data from 2000 and 2010, and remote sensing data, to study the evolution of the size distribution of Chinese cities in terms of land and people. Migrants without local hukou registration increasingly congregate in a few large cities. Hence, previous studies that rely on the count of local hukou holders make the city size distribution seem more even than it actually is. Temporal comparisons show that the city size distribution is diverging in terms of the urban resident population but converging in terms of land area. These divergent patterns suggest that growth in the resident population of large cities is not being assisted by fast enough area expansion, while area expansion of less populous cities is too fast, given their slow growth in resident numbers. Keywords Agglomeration · City size · Hukou · Migration · Pareto’s law · China

2.1

Introduction

China’s urban population is forecast to be one billion by 2030, an increase of 350 million from 2010 (MGI 2009). This urban expansion may occur by existing big cities joining Shanghai as a mega-city but also by China’s central and local governments growing new cities in currently less urbanized areas. There may be quite different effects on wages and productivity, house prices, land and water use, food security, and environmental stress of taking one path versus the other. Consequently, there is debate about whether small and medium-sized cities in China should be favored over expansion of big cities. J. Gibson (*) · C. Li Department of Economics, School of Accounting, Finance and Economics, University of Waikato, Hamilton, New Zealand e-mail: [email protected]; [email protected] © Springer Nature Singapore Pte Ltd. 2020 J. Poot, M. Roskruge (eds.), Population Change and Impacts in Asia and the Pacific, New Frontiers in Regional Science: Asian Perspectives 30, https://doi.org/10.1007/978-981-10-0230-4_2

29

30

J. Gibson and C. Li

Efforts to limit China’s big cities have a long history. The 1990 “City Planning Law” (Zhonghua Renmin Gongheguo Chengshi Guihua Fa) mandated “strictly controlling the size of large cities and developing medium-sized and small cities” (Xu 2009). New cities in this era were often just counties with new labels. This unsuccessful experiment of creating cities was ended in the late 1990s (Fan et al. 2012). More even-handed policy followed, with the Tenth Five-Year Plan (2001–2005) seeking balanced development of large, medium-sized, and small cities and the Eleventh Five-Year Plan (2006–2010) emphasizing development of metropolitan regions. In line with this balanced approach, the “Urban and Rural Planning Law” (Zhonghua Renmin Gongheguo Chengxiang Guihua Fa) of 2008 dropped the key phrase “strictly controlling the size of large cities” that had been part of the 1990 “City Planning Law” (Fan et al. 2012). Since then, the policy pendulum moved again in the direction of limiting growth of the biggest cities. In 2014, President Xi Jinping announced reforms that would fully remove hukou restrictions in towns and small cities, and gradually ease restrictions in medium-sized cities. Additionally, reasonable conditions would be imposed on individuals who would settle in big cities while the population of megacities would be strictly controlled. Big city growth may also be limited by land-use controls. Citing food security concerns, land on the outskirts of the biggest cities like Beijing and Shanghai is being classified as “permanent basic farmland” to be used only for cultivation. In announcing these controls, the Minister for Land and Resources claimed that good farmland had been eaten by steel and cement. Conversely, land-use controls for small urban areas are less strictly enforced and fiscal decentralization creates incentives for local officials to convert more farmland to industrial or residential use than is actually needed (Lichtenberg and Ding 2009). What is missing in this swing back to a pro-small bias is consideration of the evidence on China’s evolving urban system. Using Pareto’s law as a benchmark, we find that the very largest cities in China have scope to absorb more migrants, contrary to the pro-small bias in urban policy. One of the most robust empirical facts about the relative size of cities in market economies is that they follow either Pareto’s law or— more specifically—Zipf’s law (Gabaix 1999).1 While economists and demographers tend to apply these rank-size laws to population, new research that builds a statistical representation of cities from the bottom up shows that Zipf’s law holds both for population and for area, to a good approximation, in Great Britain and the United States (Rozenfeld et al. 2011). The negative relationship between the logarithm of city size (in terms of people or land) and city rank, expected according to Pareto’s law, is used here to identify the cities with too many migrants and those with scope to take more. We also use changes in this relationship over time to contrast trends in the distribution of city area and city population. A key feature of our analysis is the use of census data from 2000 and 2010. Previous studies may mislead because they use annual data that measure how many people have hukou household registration for each place, not who live

1

Zipf’s law is a special case of the Pareto distribution, with a Pareto exponent equal to one.

2 Pareto’s Law and City Size in China: Diverging Patterns in Land and People

31

there. The census counts show that non-hukou migrants increasingly congregate in a few larger cities; for example, there are urban districts in each of China’s 287 prefectures, but just 27 are home to 71 million of the 117 million non-hukou migrants residing in urban districts in the 2010 census.2 Thus, if cities are measured by the count of local hukou holders, as in prior studies, the size distribution seems more even than it actually is, because it ignores the funneling of migrants into a few big cities. The claim by prior studies that, over time, China’s cities became more evenly distributed in terms of population size, may therefore not be reliable because those studies do not use data that count people where they actually live. In fact, contrary to existing claims, the census counts of residents in 2000 and 2010 show that the Pareto coefficient is falling, which implies a less even distribution (larger changes in city size are needed to change city rank). In fact, the city size distribution is moving closer to the unitary value implied by Zipf’s law. Yet, when we study city area, whether measured more finely using Landsat or more coarsely using night-time lights, there is a clear trend for the Pareto coefficients to be rising, implying that cities are becoming more equally sized in area. These divergent trends may reflect a mismatch between migrants funneling into a few large cities and governments trying to steer them into smaller cities. One aspect of these smaller cities expanding in area is “ghost towns” where empty new housing units sit on recently converted farmland. Another problem with prior studies is that county-level cities (xianji shi) are often included in samples along with urban districts (shiqu). But some county-level cities are just relabeled counties and do not differ from rural counties in economic performance (Li 2011; Fan et al. 2012). More reason for doubt about county-level cities is that they lack urbanization externalities (as do rural counties). Such externalities are only found in urban districts (Li and Gibson 2014b). Thus, a focus of some studies on the apparent growth of small cities (e.g., Anderson and Ge 2005) may be misplaced since county-level cities should be excluded (as is done here) when studying Pareto’s law. Another set of studies that may give a misleading picture of China’s evolving urban system are studies based on employment statistics. Some of these, such as Au and Henderson (2006a), come to similar conclusions as the current study: China’s big cities are too small to reap all agglomeration economies. However, the data Au and Henderson (2006a) use are not reliable. For example, they estimate an inverted U-shaped relationship between output per worker and city scale, in terms of 1997 employment, but the Yearbook data that they use exclude most private sector workers. Similarly, Au and Henderson (2006b) use employment data in urban yearbooks from 1991 to 1998 to estimate the effect of city scale (employment) on per-worker productivity and to simulate the effect of doubling urban agglomerations,

2 Non-hukou migrants are people who move somewhere other than where their hukou registration is from without converting either their type of hukou (agricultural or nonagricultural) or their place of registration (hukou suozaidi). The problems for the interpretation of China’s statistics due to these migrants are discussed in Li and Gibson (2013).

32

J. Gibson and C. Li

where this increased scale would follow from relaxing migration restrictions. It is surely true that the restrictions lower productivity, but the yearbook employment data that are used cover only subsets of total employment in “directly reporting industrial enterprises” whose share of total employment varies from about 67% in some years down to 40% in others. Holz (2013) shows that all-sector employment, which should be a more accurate measure of city scale, is only available in the decadal population census. In the next section, we describe our data and methods and pay attention to the restrictions on the sample (based on size thresholds and data availability). In Sect. 2.3, we report several results: firstly, that measuring cities by how many hukou registrations they have makes cities seem more evenly sized; secondly, that the city size distribution is becoming less even over time in terms of population but more even over time in terms of land area; thirdly, that it is the moving away of cities from the middle of the population size distribution that is causing the divergence in city size; and, fourthly, that very large cities appear to have the most scope for absorbing more migrants. Our conclusions are in Sect. 2.4.

2.2

Data and Model Specification

A review of data quality issues showed that the most reliable information on city population is from the Population Census in 2000 and in 2010. The census counts residents of an area as those living there at least 6 months, giving a more realistic measure of a city’s size than the count of people whose hukou registration is from that place but who may live elsewhere. Existing studies rely on the China City Statistical Yearbook (NBS 2011), which ignores non-hukou residents. When we need hukou counts, to contrast with results using counts of residents, we take them from the Ministry of Public Security (MPS 2001, 2011). A further problem with Yearbook data is that they do not report on urban cores, while the census reports on each individual district within a prefectural city, and contiguous districts are the best proxy for an urban core in China (Roberts et al. 2012).3 We start with urban districts of all 287 prefectural cities in the 2010 census. Among these are 24 that were classified as Leagues, Regions, or Autonomous Prefectures in 2000. To keep the same geographical coverage, we treat these as if they were prefectural cities in 2000. The size thresholds set on the estimation samples (see below) exclude 20 of these 24, so our inclusive approach to treating them as cores of prefectural cities in 2000 should not matter. 3 The urban core of a prefectural city is made up of adjacent districts (shiqu). The exceptions in our data are ten districts of Chongqing (Puling, Wansheng, Shuangqiao, Changshou, Jiangjin, Hechuan, Yongchuan, Nanchuan, Wanzhou, and Qianjiang) that are excluded due to being largely nonurbanized and only recently upgraded from county-level city or county status, plus four districts of Wuhan (Caidian, Jianxia, Huangpi, and Xinzhou) and one from Kunming (Dongchuan) that are similar to county-level cities or counties.

2 Pareto’s Law and City Size in China: Diverging Patterns in Land and People

33

Our interest is in using Ordinary Least Squares (OLS) to estimate: ln ðRankÞ ¼ α  β ln ðSizeÞ þ ε

ð2:1Þ

with β the Pareto exponent, ε a random error, and Size and Rank may be in terms of people or land. The special case of b β ¼ 1 is Zipf’s law. Prior studies for China use mainly this specification (e.g., Anderson and Ge 2005; Liang 2010; Li and Sui 2013) or else a specification that shifts city ranks by 0.5, but gives similar results (Xu and Zhu 2009; Chen et al. 2013). When considering population, we estimate Eq. (2.1) with four measures of city size and city rank: the nonagricultural hukou population (NA) and the urban resident population (U) for 2000 and for 2010. Comparing results for NA and U helps to assess possible bias in prior studies that only use NA to measure city size. Comparing 2000 and 2010 shows if the city size distribution is converging. For city area, we use remote sensing data from Landsat, which is more precise, but available for limited years, and night-time lights observed by the Defense Meteorological Satellite Program (DMSP), which give coarser resolution annual measures, and are used by Gibson et al. (2014) to measure the area of China’s cities.4 A typical pattern that emerges when predictions from Eq. (2.1) are compared with a scatter plot is for the lower tail to flatten out due to “cities” too small to distinguish from rural areas (Brakman et al. 1999). Studies set thresholds to exclude these small cities (Giesen et al. 2010). For example, thresholds for China range from 80,000 (Xu and Zhu 2009) or 100,000 (Anderson and Ge 2005; Liang 2010) to 200,000 and 500,000 (Chen et al. 2013), and also use relative values such as the smallest city in the top 70% of cities (Li and Sui 2013) and rolling sample approaches that constantly change the threshold (Peng 2010). Thresholds such as 80,000 once coincided with official city size definitions, but are less relevant now due to growth in average city size. Our approach is to set a threshold for the most reliable measure (the urban resident count) and hold constant the proportion of cities below that threshold in the other samples. A threshold of 0.3 million for U in 2010 excludes n ¼ 36 (one-eighth) of the total sample (only 2% of urban residents are in these small cities). In order to also drop the smallest one-eighth of cities for each of the other three samples, thresholds of 0.204 million for urban residents in 2000, and 0.147 million and 0.197 million for the nonagricultural hukou in 2000 and 2010, respectively, are used. These four estimation samples, each of n ¼ 251, are used when we focus just on trends in city size in terms of people. Fewer cities have their area separately distinguished, since adjacent cities that are separated by administrative boundaries may agglomerate into a single unit when 4

The two sources of remote sensing data on city size are highly correlated, with a correlation coefficient of 0.86 for comparing Landsat and night light-derived measures of city area in 2000. The other source of information on city area is from Yearbook reports of built-up areas, but comparisons with remote sensing data show that this information is unreliable (Gibson et al. 2014; Liu et al. 2012) since local governments have incentives to under-report land conversions.

34

J. Gibson and C. Li

viewed from space. This clumping together is especially apparent for cities in the Pearl River delta and Yangtze delta (Gibson et al. 2014). Any cities that joined together at any time from 1992 to 2012, according to night lights, are treated as a single unit for all years, by merging the land area and population data for the separate cities covered by the agglomeration. After making these merges and also dropping any cities with lit area less than 1 km2 (the spatial resolution of the DMSP data), a sample of 205 cities was available. This is the sample used when studying trends in the distribution of city area and when comparing trends in the Pareto coefficients for land and people. Comparing Pareto exponents shows the dynamics of the city size distribution and whether sizes are converging but does not show where change occurs. Kernel density plots of relative city size can show this, and these are presented for all four population samples. The Markov transition matrix is another nonparametric approach that we use, dividing cities in the four population samples into six groups defined by cut-points at 0.4, 0.6, 1.2, 2, and 4 times the population of the average city in a particular sample.5 The higher the Markov transition probability for moving into a new group between 2000 and 2010, the less stable are cities in the original size range. We also consider deviations of actual city size from the size predicted by Pareto’s law, given a particular city rank: d ¼ exp Size

! ln ðRank Þ  b α b β

ð2:2Þ

Equation (2.2) is applied to each of our four population samples. These deviations show where, from the standpoint of Pareto’s law, cities are made “too large” or “too small” by either policy (showing up when using the NA hukou count to measure cities) or by the choices of migrants (showing up when using the urban resident count). We also see where these “oversized” or “undersized” cities move in the distribution over time and consider what may cause these moves. For example, if small cities are made too large by policy biases, it should show up when using the NA hukou count. Additionally, the Kernel density and Markov transition matrix would show movement of small cities into the middle-sized groups. This would be a case of planning causing a more even city size distribution, as suggested by existing studies that show Pareto exponents rising over time. This is attributed either to the growth of small cities (e.g., Anderson and Ge 2005; Xu and Zhu 2009) or to restrictions on large cities (e.g., Chen et al. 2013; Li and Sui 2013). But if this pattern is not apparent when using data on the urban resident population, it

5

For the urban resident population (U) in the 2010 census, these relative cut-points correspond to 0.56 million (m), 0.84 m, 1.69 m, 2.80 m, and 5.63 m. These ranges are similar to those announced by the State Council in 2014 for adjusting standards for categorizing city sizes; 0.5 million, 1 m, 3 m, 5 m, and ten million except that the very small cities are divided into two groups with another threshold at 0.2 million in the State Council guidelines.

2 Pareto’s Law and City Size in China: Diverging Patterns in Land and People

35

suggests a statistical artifact in prior studies due to these studies measuring city size in terms of hukou registrations rather than by the count of how many people actually live in each city. The comparison between planned city sizes (shown by the nonagricultural hukou count) and the distribution of city sizes resulting from decisions of migrants can also be examined more directly. The number of non-hukou migrants (M) in each city in 2010 can be calculated as: M 2010 ¼ U 2010  NA2010

ð2:3Þ

and we can compare the actual stock of incoming and outgoing migrants in 2010 with the movement of people needed in order for the city size distribution to exactly follow Pareto’s law. That is, we can use Eq. (2.2) to calculate the predicted size for each city in terms of both urban residents and nonagricultural hukou holders and then examine hypothetical flows based on the deviation of the two actual population series from these two predictions: d Md 2010 ¼ Size  Size

ð2:4Þ

The comparison of the actual stock of migrants with the hypothetical number that would hold under Pareto’s law can identify whether it is large or small cities that have already taken in enough migrants (for their rank) and what type of cities can potentially take in more.

2.3

Results

The results of estimating Eq. (2.1) on the four samples are reported in Table 2.1, with the raw data and fitted trends shown in Fig. 2.1. Measuring city size by the number of hukou registrations (NA) makes Pareto exponents significantly ( p < 0.001) larger than if cities are measured by their number of residents. Larger Pareto exponents imply a more even distribution of city sizes (since small changes in city size are associated with larger changes in rank). Intuitively, ignoring non-hukou residents, as existing studies have done, leads one to miss the fact that many migrants congregate into a few large cities, and being blind to this pattern wrongly implies a more even city size distribution. But even with the smaller Pareto coefficients for city size measured in terms of residents, the null hypothesis of β ¼ 1 is rejected, so Zipf’s law does not hold exactly. Likewise, prior studies for China find evidence against the parallel growth of cities (Chen et al. 2013). The second finding in Table 2.1 is that Pareto exponents fell significantly from 2000 to 2010, regardless of whether city population is measured by counting residents or by counting the number of nonagricultural hukou registrations. Thus, the population size distribution of China’s cities became less even over the decade

36

J. Gibson and C. Li

Table 2.1 Rank-size regressions for urban resident and nonagricultural Hukou registered population in 2000 and 2010 Ln(NA2000)

Ln(Rank NA2000) 1.251 (0.011)

Ln(U2000)

Ln(Rank U2000)

Ln(Rank NA2010)

1.207 (0.011)

Ln(NA2010)

1.169 (0.013)

Ln(U2010) Constant Observations Adjusted R2

Ln(Rank U2010)

3.542 (0.012) 251 0.98

4.002 (0.010) 251 0.98

1.147 (0.009) 4.386 (0.008) 251 0.98

4.026 (0.012) 251 0.97

ln(200)

Notes: Standard errors are in parentheses.  Significantly different from 0 (constant) or 1 (for the Pareto exponent) at p ¼ 0.01 confidence level. NA is the nonagricultural hukou registered population and U is the urban resident population. See Appendix Table 2.5 for details

ln(10) ln(20)

NA Predicted

ln(1)

ln(Rank)

ln(50)

U 2010 U 2000 NA 2010 NA 2000 U Predicted

ln(.15) ln(.2)

ln(.3) ln(.4)

ln(.6) ln(.8)

ln(1.2) ln(1.7) ln(2.5)

ln(4)

ln(6)

ln(10)

ln(15) ln(20)

ln(30)

ln(45)

ln(Size in millions)

Note: The number of observations is 251 for all four samples (NA2000, U2000, NA2010 and U2010). Details on these four population sub-samples can be found in Appendix Table 2.5.

Fig. 2.1 Rank-size plot

and this divergence is not something noted in the prior literature. Indeed, some studies claim the reverse, that is, a convergence in the city size distribution (e.g., Anderson and Ge 2005; Xu and Zhu 2009). While Pareto coefficients are falling over time for city size in terms of population, they are getting larger for city size in terms of land area (Table 2.2). The Pareto

7.149 (0.083) 0.860 0.000

0.905 (0.011)

2008

3.496 (0.017) 0.963 0.000

7.905 8.911 (0.069) (0.059) 0.933 0.967 0.000 0.000 Ln(rank NA) 2000 1.056 (0.014)

Ln(Rank DMSP) 1995 2000 0.705 (0.020) 0.821 (0.015)

1.011 (0.014) 3.845 (0.014) 0.963 0.436

2010

0.943 (0.010) 9.574 (0.057) 0.978 0.000

2010

Adj R2 p-value

Constant

2010

Ln(U) 2000

Adj R2 p-value

Constant

2010

2008

2000

Ln(Landsat) 1995

8.947 (0.134) 0.857 0.654

3.844 (0.012) 0.973 0.000

9.083 (0.134) 0.864 0.792 Ln(rank U) 2000 1.051 (0.012)

Ln(Rank Landsat) 1995 2000 0.987 (0.028) 1.007 (0.028)

1.031 (0.010) 4.140 (0.010) 0.980 0.003

2010

9.303 (0.128) 0.883 0.504

1.017 (0.026)

2008

2010

Notes: Standard errors in parentheses.  Significantly different from 0 at 1% confidence level. The p-value is for testing the null hypothesis that the Pareto exponent equals 1 (Zipf’s Law). DMSP is city area using night lights from satellites F12 (1995), F15 (2000), F16 (2008), and F18 (2010) with a 50% luminosity threshold. NA is the nonagricultural hukou population and U is the urban resident population. Number of observations is 205 excluding areas that are missing data or are less than 1 km2 according to DMSP in all 4 years 1995, 2000, 2008 and 2010, and includes adjacent cities (according to administrative boundaries) that merged into single agglomerations according to DMSP (and the same merges are applied to data from Landsat, NA, and U to maintain spatial consistency)

Adj R2 p-value

Constant

2010

Ln(NA) 2000

Adj R2 p-value

Constant

2010

2008

2000

Ln(DMSP) 1995

Table 2.2 Rank-size regressions using city area in 1995, 2000, 2008 and 2010, and city population in 2000 and 2010

2 Pareto’s Law and City Size in China: Diverging Patterns in Land and People 37

38

J. Gibson and C. Li

.6

U 2010 U 2000 NA 2010

0

.2

Kernel Density

.4

NA 2000

ln(1/5)

ln(2/5)

ln(3/5) ln(4/5) ln(1)ln(1.2)ln(1.5)

ln(2)

ln(4)

ln(6)

ln(8) ln(10)

ln(14)

ln(Relative Size)

Note: City size is relative to the mean of each distribution, as described in Appendix Table 2.5.

Fig. 2.2 Relative city size distribution

coefficient for city area measured by night lights rose from 0.705 in 1995 to 0.943 in 2010.6 City area data from Landsat follow Zipf’s law always more closely (results in the top right-hand panel of Table 2.2 show no rejection of the hypothesis that β ¼ 1), but the Pareto coefficients also rise over time. Yet, when the same sample of 205 cities is used, that is, combining population of nearby cities whose lights merge into one agglomeration, the Pareto coefficients on city size according to population fall over time (bottom panel of Table 2.2) in line with what Table 2.1 shows. Thus, there are contrasting changes in the distribution of city size according to land and people, with cities becoming more equal in area and less equal in population. Regressions do not show where the changes in the distribution of city size by population occur but the kernel densities in Fig. 2.2 give some clues. Between 2000 and 2010, there was a move away from the middle of the distribution, with the proportion of cities from 0.4 to two times the mean falling, and the proportion that are either smaller than 0.4 of the mean or larger than twice the mean rising. More evidence of dispersion for medium sizes is in the Markov transition matrices for city population groups reported in Table 2.3 according to residents (top panel) or nonagricultural hukou holders (lower panel). Cut-points of 0.4, 0.6, 1.2, 2, and 4 times the mean city size define groups in the table. The small-medium, medium, and large-medium cities have the lowest odds of staying in the same resident size ranging from 2000 to 2010, with probabilities from 47% to 65%. Usually it is a move downward, indicating a fall in size relative to the mean. In

6

Landsat data are only available for 1995, 2000, and 2008 (we thank Dr. Xiangzheng Deng for these). DMSP data are used for the same years, and for 2010, to compare with the population estimates.

2 Pareto’s Law and City Size in China: Diverging Patterns in Land and People

39

Table 2.3 Markov transition matrices for city size groups Urban residents 2010 Small (89) Small-medium (59) Medium (58) Large-medium (16) Large (17) Very large (12) Total (251) NA 2010 Small (80) Small-medium (49) Medium (67) Large-medium (31) Large (14) Very large (10) Total (251)

Urban residents 2000 S SM 79.2% 41.9% (57) (26) 12.5% 46.8% (9) (29) 9.7% (6) 1.6% (1)

M

31.8% (21) 65.2% (43) 1.5% (1)

1.5% (1) 91.7% 100% 100% (72) (62) (66) Nonagricultural (NA) Hukou 2000 S SM M 78.5% 37.3% (51) (25) 12.3% 43.3% 17.1% (8) (29) (12) 3.1% 13.4% 71.4% (2) (9) (50) 1.5% 4.5% 10% (1) (3) (7) 1.4% (1)

95.4% (65)

98.5% (67)

100% (70)

LM

32.1% (9) 50% (14) 17.9% (5)

100% (28)

L

VL

85.7% (12) 14.3% (2) 100% (14)

100% (9) 100% (9)

LM

L

VL

24% (6) 60% (15) 12% (3) 4% (1) 100% (25)

29.4% (5) 58.8% (10) 11.8% (2) 100% (17)

100% (7) 100% (7)

Notes: Transition probability (in %) is calculated by dividing the number of cities that move to a size range in 2010 by the total of cities in the range they left in 2000. The number of cities in each cell is in (), with 251 cities in total. The abbreviations S, SM, M, LM, L, and VL are for small, smallmedium, medium, large-medium, large and very large, and are based on cut-points of 0.4, 0.6, 1.2, 2, and 4 times the mean city size (as measured by either U or NA, in either 2000 or 2010)

contrast, 79% of small cities, and (100%) 86% of (very) large cities stay in the same size range between the two censuses. Upward moves are less common, but they include two instances of moves to nonadjacent groups in Guangdong province: Foshan went from medium size to very large and Huizhou from small-medium size to large-medium. The example of Foshan shows the error in using the nonagricultural hukou holders as a measure of

40

J. Gibson and C. Li

city size; in 2010, there were 3.7 million people with nonagricultural hukou registration from Foshan, but the census counted 6.8 million residents. Obscuring city size dynamics by using the count of local hukou holders to measure city size is seen by comparing the two panels in Table 2.3. When the hukou count is used, there is no clear pattern of lower transition probabilities along the main diagonal, below the very large size group. In contrast, resident size classes from small-medium to large-medium show a substantial probability of dispersion. In other words, cities in the middle of the distribution had lower odds of staying in the same size groups between 2000 and 2010 than did cities who started in the tails of the distribution. The different patterns for the two transition matrices reflect the fact that the number of nonagricultural hukou holders registered in a particular city evolves only slowly over time since hukou status is inherited and is hard to convert, and so this indicator misses the rapid changes in size that can come from migrants voting with their feet. For example, an average of 35% of the cities that were large-medium, medium, or small-medium in 2000 fell into the next lower size category by 2010. In contrast, when the size groupings are according to the hukou count, just 26% of cities in these size ranges, on average, had fallen into a lower size range by 2010 due to the slower evolution of the spatial pattern of hukou registrations. Migrants voting with their feet may change the city size distribution, but they still may be too few, since the largest cities remain undersized in terms of Pareto’s law. Consider the top seven cities by urban resident population in 2010; all are well below the Fig. 2.1 trend line showing the size needed to fit an exact Pareto distribution. The extra people needed to shift the data points for those seven cities onto the trend line is one indicator of how many more migrants it would take to have cities of the right size for their rank. Why do these cities have too few migrants? Some may be deterred by high housing costs, exacerbated by restricted urban land expansion in the largest cities. Li and Gibson (2014a) use a hedonic model to compare apartment prices across Chinese cities. They find that for a city like Beijing (ranked second by residents), prices are 230% higher than for a city like Changsha (ranked 27th). It may be that those wanting to move into very large Chinese cities are deterred by the high housing costs and instead go into large and large-medium cities like Changsha. In fact, most cities in Fig. 2.1 of between two million and five million residents (including Changsha) seem larger than what the Pareto distribution would predict (so are above the trend line). The deviation above the trend line for this group of cities was less apparent in 2000, when house prices were more equal. The destination choices of non-hukou migrants, and the scope for cities to absorb more, are shown by applying Eq. (2.3) and Eq. (2.4) and comparing the results to the size of each city. In Fig. 2.3, this is done for all 244 cities that are common to the NA and U samples for 2010 (so each city is in the largest 87.5% of cities according to both counts). The predicted stock of non-hukou migrants (from Eq. (2.3)) ranges from over seven million for Shanghai and Shenzhen (ranked first and third by residents) to 1.5 million for Shantou (ranked 18th by residents, but fifth in terms of hukou registrations), and these stocks are shown by the red squares in the figure.

41

Predicted in NA 2010 Actual Predicted in U 2010

9 7 -1 0 1 2 3

5

Migration 2010 (million)

20

25

2 Pareto’s Law and City Size in China: Diverging Patterns in Land and People

ln(2/5)

ln(3/5)

ln(1.2)

ln(2)

ln(4)

ln(Relative Size in U 2010)

Notes: Relative size is according to the mean city size by urban residents in 2010. Vertical lines correspond to the size classes in Tables 2.3 and 2.4. The stocks of actual and hypothetical migrants are based on Eqs. (2.3) and (2.4). The number of observations is 244, which are the cities that are common to both the NA 2010 and U 2010 samples.

Fig. 2.3 Stock of non-hukou migrants 2010, and the hypothetical stock needed for exact Pareto distributions, versus city population

The hypothetical number of extra non-hukou migrants to give an exact Pareto distribution in terms of the number of residents is shown by the green triangle markers; and the hypothetical number to give an exact Pareto distribution in terms of nonagricultural hukou registrations is shown by the blue circles. These hypothetical values come from Eq. (2.4) and they can be negative, which corresponds to cities that were larger in 2010 than what would be predicted from the rank of that city. This exercise suggests that the scope to absorb more non-hukou migrants is mainly limited to the very large cities, defined as those with four or more times the mean number of residents. For example, to achieve the exact Pareto distribution, Shanghai would need to absorb sufficient migrants to get the total resident population to just over 40 million people, making it slightly larger than Tokyo. Similarly, Beijing would need to get to a resident population of almost 30 million, making it slightly larger than Delhi but a little less populous than Jakarta. Beyond these specific examples, we consider the six size groups (from “very large” to “small”) shown by the vertical lines in Fig. 2.3, with results for these size classes in Table 2.4. Consider the 12 “very large” cities; in 2010, these were home to 47 million non-hukou migrants. In the next class of cities, the migrants total 19 million and are funneled into 16 cities. One city in this size range (Shantou) has 1.5 million out-migrants. The large-medium and medium classes each have 14–15 million migrants, while the small-medium and small size classes each hold 8–9 million migrants, and these are spread over many cities. The three smallest size classes also include 22 cities that are the source of 3.5 million out-migrants.

42

J. Gibson and C. Li

Table 2.4 Actual migrants and hypothetical to get an exact Pareto distribution

City size groups (based on 2010 resident count) Very large Large Large-medium Medium Small-medium Small Total

Actual In 46.58 (12) 19.21 (16) 13.63 (16) 15.12 (53) 8.77 (52) 8.28 (72) 111.59 (221)

Out

1.52 (1)

0.94 (5) 1.74 (7) 0.83 (10) 5.04 (23)

Hypothetical extra migration Urban Residents Nonagricultural hukou 2010 2010 In Out In Out 52.63 0.64 41.02 0.07 (10) (2) (11) (1) 0.42 4.41 3.59 0.71 (3) (14) (9) (8) 6.26 0.05 1.12 (16) (1) (15) 0.38 0.98 0.02 4.51 (25) (33) (1) (57) 0.03 1.13 0.09 2.75 (8) (51) (10) (49) 1.31 0.38 1.92 0.89 (44) (38) (53) (29) 54.77 13.80 46.69 10.05 (90) (154) (85) (159)

Notes: The number of migrants is in millions, with the number of cities in (). N ¼ 244, based on cities common to the samples for NA 2010 and U 2010

When attention shifts from actual migration to the hypothetical pattern needed to produce exact Pareto distributions for city size and rank, the relevant values are shown in the last four columns of Table 2.4. Considering the results for urban residents in 2010, an extra 52.6 million migrants could go into ten of the very large cities while 0.6 million could leave the other two very large cities and the resulting size distribution would sit exactly on the trend line shown in Fig. 2.1. The other main change to get an exact Pareto distribution is for the large and large-medium cities to have about ten million fewer migrants (by moving them into the very large size class). This result is just another way of noting the pattern from Fig. 2.1; cities between 1.2 and 4 times the mean city size (in terms of residents, these are cities of from two million to five million people in 2010) seem larger than what would be predicted from their rank under an exact Pareto distribution, while the very largest cities are smaller than what is predicted. Finally, for the three smallest size groups in Table 2.4, the extra inward or outward migration needed to get to an exact Pareto distribution never amounts to more than one million people per group. An exact Pareto distribution for the nonagricultural hukou population of each city requires less migration, with 46.7 million coming in and ten million going out. Measuring cities by their hukou registrations leads to a seemingly more even distribution than is truly the case since the funneling of most non-hukou migrants into just a few host cities is ignored. Furthermore, since the Pareto distribution for the non-hukou count is more evenly spread than is the one for the resident count (Table 2.1 and Fig. 2.1), it takes less movement to get to this target. The second

2 Pareto’s Law and City Size in China: Diverging Patterns in Land and People

43

difference between using the hukou count and the resident count, in terms of hypothetical migration to get cities the right size for their rank, is a lack of apparent “queuing” for the very large cities. When cities are measured in terms of residents, the large and large-medium size groups seem to have “too many” migrants; removing ten million of them and transferring them into the very large cities would give sizes more consistent with the Pareto distribution. But this pattern is not apparent when city size is measured by the hukou count, and it is the small-medium and medium-sized cities that appear to be the most oversized (for their rank). Despite differences between the two sets of results for hypothetical migration, a key point from Table 2.4 is their similarity in showing that it is only the very large cities with scope to accept many more non-hukou migrants, in terms of having city sizes that follow a Pareto distribution more closely. This finding contrasts with the views of leaders such as Xi Jinping and potentially informs about China’s evolving urban system. The limited capacity of small and medium-sized cities to absorb migrants, either in terms of the actual stock in 2010 or the hypothetical number of extra migrants to get to an exact Pareto distribution, shows that an urbanization process of nearby rural–urban migrants going to live in small local cities is unlikely to succeed in transforming China into a fully urbanized country. Instead, it is the agglomeration processes that are a key to China’s urban transition. These processes not only absorb nearby rural–urban migrants but also take in inter-regional rural– urban and urban–urban migrants. Moreover, it is the very large cities, and not the small towns, that provide agglomeration-related productivity advantages. These advantages appear to operate only in the tertiary sector and not in the secondary sector activities like construction and manufacturing that increasingly left urban districts and moved into smaller towns and counties between 2000 and 2010 (Li and Gibson 2014b). Thus, a focus on directing migrants into small cities will not put them into the places where they are likely to be the most productive. This misallocation will be especially costly as China rebalances the economy by developing the undersized services sector and reducing reliance on the oversized manufacturing sector (Ghani 2012), since it is the services sector that benefits the most from locating in larger agglomerations. A related issue concerns population density, which may contribute to the agglomeration effects discussed above. Some existing studies already note that the density of China’s cities is falling relative to comparator cities elsewhere (Du et al. 2014) and that urban area expansion can shift from being land saving to land using as patterns of urban development become less dense (Deng et al. 2015). The finding of diverging patterns in the Pareto coefficients for land and people that are described here implies that the trends in population density will vary along China’s city size distribution.

44

2.4

J. Gibson and C. Li

Conclusions

There is ongoing debate about China’s evolving urban system and especially about the policy question of whether small and medium-sized cities should be favored over expansion of big cities. In this chapter, we uncover three facts that were missed in prior studies. First, if the population of cities is measured by their number of nonagricultural hukou registrations – as in previous studies – Pareto exponents seem larger than they actually are. These prior studies miss the funneling of non-hukou migrants into just a few large cities. Once this fact is missed, statistical inquiries into Pareto and other distributions tend to find a more even city size distribution than truly exists. This bias is exacerbated by studies that include county-level cities in their samples, despite such “cities” lacking an urban core (Roberts et al. 2012). Second, the population size distribution of China’s cities has become less even over time, with movement out of the middle of the distribution between 2000 and 2010. In contrast, previous studies have highlighted an apparent move toward a more even population distribution for cities in China. The final new fact that is revealed here is that while the city size distribution is becoming less even over time in terms of population, it is becoming more even over time in terms of land area. These three facts are found mainly by regressing city rank on city size, in terms of land and people, where we use both the de jure population from the hukou registration system and the de facto population from resident-based census counts. The Pareto exponents from such regressions provide one benchmark for evaluating an urban system, with the special case of a Pareto coefficient of unity (Zipf’s law) seeming to hold in some market economies (Gabaix 1999).7 Moreover, these relationships are not just empirical regularities; theoretical models of city growth can generate Zipf’s law for both land and people by assuming Cobb-Douglas preferences for goods and housing, random growth with small frictions, and small urbanization externalities (Rozenfeld et al. 2011). These relationships appear to hold more for geography-based definitions of cities rather than for legal-based ones, which is consistent with the evidence in Table 2.2 that Pareto coefficients are closer to the Zipf’s law value of unity when cities are defined according to the measurements from space rather than according to administrative boundaries. In addition to uncovering these facts, we also use the predictions from the estimated Pareto distribution as a normative standard to judge China’s city size distribution. We are relying on this as a proxy for what city sizes look like in societies with a less distorted urban past than China. The logic of this exercise is that the parts of China’s urban system deviating from a Pareto distribution may be a legacy of the command economy era and of restrictions on labor movement and land supply that are less apparent in market economies. When viewed from this normative Some studies suggest that Pareto’s law only fits the upper tail (e.g., Eeckhout 2004, 2009) and propose the lognormal distribution as a better representation of the city size distribution, but this may be because they use legal definitions of cities rather than geography-based definitions. 7

2 Pareto’s Law and City Size in China: Diverging Patterns in Land and People

45

standard, even though China’s largest cities look gigantic relative to other cities (e.g., Shanghai with 20 million residents in 2010 is almost double the size of the third largest city and triple the size of the seventh largest), they actually are too small; according to the size, they ought to be as the top-ranked cities in the urban hierarchy. Thus, it is the very large cities that appear to have too few people and that have the most scope for absorbing more migrants in order to become the right size for their rank. In contrast, there is almost no scope for small-to-medium cities to absorb more migrants when using this benchmark, notwithstanding the policy biases and statements from political leaders in favor of directing migrants toward these smaller cities. These normative results suggest that the focus of the Eleventh Five-Year Plan (2006–2010) on the development of metropolitan regions was broadly appropriate, while the swing toward a pro-small policy bias in the 12th five-year plan (2011–2015) was not. For example, rather than the policy to set land outside of the largest cities aside as “permanent basic farmland,” a better focus for rules on urban land expansion might be to more strictly regulate farmland conversion by small and medium-sized Chinese cities. The finding that the trend in the Pareto exponent for the distribution of city size in terms of land area is moving in the opposite direction to that for city size in terms of population supports this suggestion. These divergent patterns suggest that growth in the resident population of large cities is not being assisted by fast enough area expansion, while area expansion of less populous cities is too fast for their slow growth in resident numbers. Also, reforms to governance and public finance that see less tax revenue transferred from local to central government – or fewer responsibilities left for local government to fund from their own budgets – may reduce the reliance of these smaller cities on revenue from land auctions. With less need for land conversion in small cities, the diverging patterns of cities becoming more equally sized in area and less equally sized in population may be reversed.

Appendix Table 2.5 Details on the four population subsamples used

Share of population in the subsample Actual city size Minimum (threshold for inclusion in subsample) Median Mean (used to form six size classes in Table 2.3) Maximum Predicted city size (Eq. (2.2)) Minimum

Subsample name NA2000 U2000 97.8% 97.9%

NA2010 98.0%

U2010 97.8% 0.301 0.698 1.407 20.218

0.147 0.386 0.666 9.382

0.204 0.552 0.975 13.460

0.199 0.571 0.979 12.286

0.205

0.283

0.277

0.370 (continued)

46

J. Gibson and C. Li

Table 2.5 (continued)

Median Mean Maximum Stock of non-hukou migrants in 2010 (Eq. (2.3)) Minimum Median Mean Maximum Hypothetical extra migrants for exact Pareto Minimum Median Mean Maximum

Subsample name NA2000 U2000 0.356 0.501 0.722 1.074 16.953 27.558

NA2010 0.500 1.126 31.274

U2010 0.676 1.571 45.794

1.523 0.166 0.437 7.931 0.176 0.037 0.150 18.988

0.613 0.008 0.168 25.576

Notes: NA is the nonagricultural registered population and U is the urban resident population. Each subsample has 251 observations and represents the largest 87.5% of cities for each indicator in each year, except for migrant calculations that are based on the union of NA and U (n ¼ 244). Numbers are in millions Sources: MPS (2001, 2011) and NBS (2003, 2011, 2012)

References Anderson G, Ge Y (2005) The size distribution of Chinese cities. Reg Sci Urban Econ 35 (6):756–776 Au C, Henderson V (2006a) Are Chinese cities too small? Rev Econ Stud 73(3):549–576 Au C, Henderson V (2006b) How migration restrictions limit agglomeration and productivity in China. J Dev Econ 80(2):350–388 Brakman S, Garretsen H, Van Marrewijk C, Van Den Berg M (1999) The return of Zipf: towards a further understanding of the rank-size distribution. J Reg Sci 39(1):183–213 Chen Z, Fu S, Zhang D (2013) Searching for the parallel growth of cities in China. Urban Stud 50 (10):2118–2135 Deng X, Huang J, Rozelle S, Zhang J, Li Z (2015) Impact of urbanization on cultivated land changes in China. Land Use Policy 45:1):1–1):7 Du S, Shi P, Van Rompaey A (2014) The relationship between urban sprawl and farmland displacement in the Pearl River Delta, China. Land 3(1):34–51 Eeckhout J (2004) Gibrat’s law for (all) cities. Am Econ Rev 94(5):1429–1451 Eeckhout J (2009) Gibrat’s law for (all) cities: reply. Am Econ Rev 99(4):1676–1683 Fan S, Li L, Zhang X (2012) Challenges of creating cities in China: lessons from a short-lived county-to-city upgrading policy. J Comp Econ 40(3):476–491 Gabaix X (1999) Zipf’s law for cities: an explanation. Q J Econ 114(3):739–767 Ghani E (2012) Are China and India converging?. http://www.voxeu.org/article/are-china-andindia-converging Gibson J, Li C, Boe-Gibson G (2014) Economic growth and expansion of China’s urban land area: evidence from administrative data and night lights, 1993–2012. Sustainability 6(11):7850–7865

2 Pareto’s Law and City Size in China: Diverging Patterns in Land and People

47

Giesen K, Zimmermann A, Suedekum J (2010) The size distribution across all cities–double Pareto lognormal strikes. J Urban Econ 68(2):129–137 Holz C (2013) Chinese statistics: classification systems and data sources. Eurasian Geogr Econ 54 (5–6):532–571 Li L (2011) The incentive role of creating cities in China. China Econ Rev 22(1):172–181 Li C, Gibson J (2013) Rising regional inequality in China: fact or artifact? World Dev 47(1):16–29 Li C, Gibson J (2014a) Spatial price differences and inequality in the People’s Republic of China: housing market evidence. Asian Dev Rev 31(1):92–120 Li C, Gibson J (2014b) Urbanization economies in China: nature, location, and effects. Working Paper No. 14/02. Department of Economics, University of Waikato Li S, Sui D (2013) Pareto’s law and sample size: a case study of China’s urban system 1984-2008. GeoJournal 78(3):615–626 Liang S (2010) Research on China’s urban development strategy based on arable land preservation. J Urban Plann Dev 137(3):329–336 Lichtenberg E, Ding C (2009) Local officials as land developers: urban spatial expansion in China. J Urban Econ 66(1):57–64 Liu Z, He C, Zhang Q, Huang Q, Yang Y (2012) Extracting the dynamics of urban expansion in China using DMSP-OLS nighttime light data from 1992 to 2008. Landsc Urban Plan 106 (1):62–72 McKinsey Global Institute (MGI) (2009) Preparing for China’s Urban Billion. http://www. mckinsey.com/insights/urbanization/preparing_for_urban_billion_in_china Ministry of Public Security (MPS) (2001) Population statistics of the People’s republic of China by county 2000 (Zhonghua Renmin Gongheguo Quanguo Fenxianshi Renkou Tongji Ziliao 2000 Nian). Qunzhong Press, Beijing Ministry of Public Security (MPS) (2011) Population statistics of the People’s republic of China by county 2010 (Zhonghua Renmin Gongheguo Quanguo Fenxianshi Renkou Tongji Ziliao 2010 Nian). Qunzhong Press, Beijing National Bureau of Statistics (NBS) (2003) Tabulation on the 2000 population census of the People’s republic of China by county (Zhongguo 2000 Nian Renkou Pucha Fenxian Ziliao). China Statistics Press, Beijing National Bureau of Statistics (NBS) (2011) China City statistical yearbook 2011 (Zhongguo Chengshi Tongji Nianjian 2011). China Statistics Press, Beijing National Bureau of Statistics (NBS) (2012) Tabulation on the 2010 population census of the People’s republic of China by county (Zhongguo 2010 Nian Renkou Pucha Fenxian Ziliao). China Statistics Press, Beijing Peng G (2010) Zipf’s law for Chinese cities: rolling sample regressions. Phys A Stat Mech Appl 389(18):3804–3813 Roberts M, Deichmann U, Fingleton B, Shi T (2012) Evaluating China’s road to prosperity: a new economic geography approach. Reg Sci Urban Econ 42(4):580–594 Rozenfeld H, Rybski D, Gabaix X, Makse H (2011) The area and population of cities: new insights from a different perspective on cities. Am Econ Rev 101(5):2205–2225 Xu Z (2009) Productivity and agglomeration economies in Chinese cities. Comp Econ Stud 51 (3):284–301 Xu Z, Zhu N (2009) City size distribution in China: are large cities dominant? Urban Stud 46 (10):2159–2185

Chapter 3

City Size Distribution in Colombia and Its Regions, 1835–2005 Gerson Javier Pérez-Valbuena and Adolfo Meisel-Roca

Abstract By using census data from 1835 to 2005, this chapter studies the urban hierarchy in Colombia and its regions. The chapter focuses on three issues: firstly, the city size distribution by means of Zipf’s law and Gibrat’s law; secondly, evolution in the population growth models; and, thirdly, the empirical validation of the point made by Gabaix (Q J Econ 114(3):739–767, 1999b) on the coincidence between national and regional population patterns. Using the adjusted rank–size relationship and non-parametric techniques, we find that city size distributions follow Zipf’s power law, and also that Gibrat’s law holds at the national level and partially for the regions over the second half of the twentieth century. These results are consistent with changes in the population growth model from the mid-fifties at national and regional levels. Keywords City size distribution · Zipf’s law · Gibrat’s law · Colombia

3.1

Introduction

In urban economics, the city size distribution has attracted much attention in recent decades. The literature has focused on empirically testing whether or not Zipf’s law and Gibrat’s law hold for a group of cities within a country. The former specifies that city size follows a Pareto distribution with the parameter equal to 1. In practice, this

G. J. Pérez-Valbuena (*) Center for Regional Economics Studies, Banco de la República, Cartagena de Indias, Colombia e-mail: [email protected] A. Meisel-Roca Universidad del Norte, Barranquilla, Colombia e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 J. Poot, M. Roskruge (eds.), Population Change and Impacts in Asia and the Pacific, New Frontiers in Regional Science: Asian Perspectives 30, https://doi.org/10.1007/978-981-10-0230-4_3

49

50

G. J. Pérez-Valbuena and A. Meisel-Roca

means that the biggest city (the capital city Bogotá) should be twice as big as the second largest city (Medellín), three times the third one (Cali), and so on.1 Gibrat’s law on the other hand indicates that cities’ populations grow randomly with the same mean and variance, and that the population growth rate and city size are independent of each other. In other words, small and big cities grow at the same rate on average. Why is it so important to investigate these empirical regularities? The answer to this question is that analyzing the historical dynamics of population lets us know whether or not urban areas, from a region or a country, follow a common growth path and if there is a unique rank–size relationship between them. This also allows us to predict with a high reliability short- and long-run urban development, and how cities can be affected by exogenous factors. This is especially relevant for developing young economies that are still on their way to the long-run population growth rates expected in developed countries. In this respect, it was argued in recent years that: “. . .we have much improved our understanding of the origin of the Zipf’s law, which has forced a great rethinking about the origins of cities. . .” (Gabaix 2016: 188). If the historical time series of city size turns out to be stationary in terms of city growth, it is expected that external shocks have only temporary effects and that city size will eventually go back to its long-term path. This will not be the case with a non-stationary growth process in which case external shocks may lead to permanent or long-run effects. In terms of characterizing city size growth, Schaffar and Dimou (2012) mention two theories. First is the random population growth theory in which cities grow randomly and, in the long run or steady state, Zipf’s and Gibrat’s laws hold. According to this theory, cities are characterized by having high labor mobility and constant returns to scale. Specifically, people’s location decisions do not depend on the location of firms but on the quality of life offered in the cities. The authors also refer to the random process as closely related to the generation of amenities by means of public policies and/or natural or historical shocks. Gabaix (2016) relates the joint observation of Zipf’s and Gibrat’s laws to economic theories, such as constant returns to scale, from which economic models for random growth of cities can be derived. Duranton (2006, 2007), using micro-theory modeling, also showed that the emergence and disappearance of cities could be affected by innovation shocks faced by firms and by the cities themselves. The second theory is the deterministic population growth theory, according to which the size of the cities depends on firms’ size and location. Thus, city size growth depends on externalities, human capital, and location decisions. In particular, according to Schaffar and Dimou (2012: 709): “Firms concentrate geographically in order to take advantage of conglomeration effects, linked either to specialization [. . .] or to diversification [. . .], but suffer, on the other hand, from congestion and the costs of commuting.”

1 Zipf (1949) was the first to formally suggest that city size follows a Pareto distribution. Nevertheless, Auerbach (1913) was the first to note this empirical regularity.

3 City Size Distribution in Colombia and Its Regions, 1835–2005

51

According to this theory, Zipf’s and Gibrat’s laws are not strictly enforced.2 The aim of this chapter is to characterize the temporal dynamics of the population in Colombia and its regions during the last two centuries, in particular, it is intended to explore the random growth hypothesis. It also aims to establish whether or not, as theoretically argued by Gabaix (1999b), the individual compliance with Zipf’s law in each region will result in the same outcome for the national aggregate. Giesen and Südekum (2011) empirically tested for the first time the theoretical postulations made by Gabaix (1999b) for Germany. There have been a large number of studies on the city size distribution in the USA, but only recently, the upper tail was analyzed separately from the whole city size distribution. Accordingly, González-Val (2010) found for the USA, throughout the twentieth century, weak evidence in favor of Gibrat’s law and also found that Zipf’s law is restricted to the upper tail of the city size distribution. This chapter makes three contributions to the literature. First, it is the first time that Colombian population data from the nineteenth century are used for the purpose of analyzing the city size distribution. Second, it is the first time that Gibrat’s law is tested with Colombian population data, and it is also the first time that non-parametric approaches are used and linked to the rank–size results. Third, no other study of Colombian population data has analyzed the city size distribution at the regional level before. In this respect, the chapter tests whether or not population growth follows a dynamic process similar to that theoretically predicted by Gabaix (1999b). His theory implies that if Zipf’s law holds in individual regions within a country, then the law is also fulfilled for the country as a whole. The literature has focused on establishing the dynamics of the urban hierarchy in developed countries, but to date little has been said about emerging economies. One exception is China, which has been the focus of several individual studies (Song and Zhang 2002; Zhou and Ma 2003; Ye and Xie 2012; Li and Gibson, this volume), and also part of cross-country studies (Soo 2005; Schaffar and Dimou 2012). Another interesting case is Israel (Benguigui and Blumenfeld-Lieberthal 2011). Examples for developing countries include those for the Balkans, Indonesia, India, and Brazil. Dimou and Schaffar (2009) analyze the hierarchy and urban growth in the Balkans. Firdaus and Fitria (2010) and Basu and Bandyapadhyay (2009) analyze the city size distribution and its potentially related factors for Indonesia and India, respectively. For Brazil, Moura and Ribeiro (2006) found estimates supporting Zipf’s power law by means of census data for the last quarter of the twentieth century and a definition of cities as urban areas with 30,000 inhabitants or more. A more recent study for Brazil found that its population has been evolving in such a way that Zipf’s law and Gibrat’s law are becoming a better way to describe how the cities will grow in the future (Matlaba et al. 2013). Duranton (2006, 2007) develops a mechanism through which the agglomeration of firms is related to Zipf’s law. The argument is based on the proportional relationship between migration and the quantity of goods produced in a city or region, and also the relationship between investment in innovation and the number of firms. Under this scenario, small and discreet innovations will result in proportional population growth which, in turn, generates a Pareto distribution. 2

52

G. J. Pérez-Valbuena and A. Meisel-Roca

For the case of Colombia, there are two earlier studies related to the city size distribution. Using census data from 1918 to 1993, Bernal and Nieto (2006) found that Zipf’s law does not hold in any of the years, even though the rank–size coefficient decreases throughout the century to settle to around 1 in 1985. On the other hand, Pérez (2006) used census data from 1912 to 1993 to analyze population dynamics of the urban areas in Colombia as a whole and in the Caribbean region. The results showed that Zipf’s law does not hold in the Caribbean region in any of the periods but it does hold in the country as a whole from 1985. At the national level, these results are consistent with Soo (2005) in his multi-country analysis, in which Colombian population data for 1993 and 1999 were used. The results show a statistically significant rank–size coefficient equal to 1. All these studies face at least two limitations. First, they do not totally cover the history of the demographic dynamics of the country since they are all based only on the twentieth century data. Second, they do not include a regional analysis for the whole country. This is a limitation for a country like Colombia, which has significant regional disparities. This chapter shows that since the second half of the twentieth century there has been a change in the dynamics of population growth in Colombia. Until then, the growth of the population followed what is called a deterministic model, in which the location of the firms, as well as the endowment and distribution of natural resources, are key factors in people’s location decisions. From that moment on, the country seems to have started the transition to a stochastic growth model that is in accordance with Zipf’s and Gibrat’s laws. Under these circumstances, the model predicts constant population’s growth rates, which are linked to exogenous shocks generating amenities and improving people’s quality of life.3 Additionally, results at the regional level are mostly consistent with findings for the country as a whole, thereby supporting the theoretical argument given by Gabaix (1999b). This chapter is divided into six sections, including this Introduction. Section 3.2 presents a brief historical context of Colombia in the nineteenth and twentieth centuries. Section 3.3 introduces a description of the theoretical and empirical characteristics of Zipf’s and Gibrat’s laws and the relationship with each other. Section 3.4 describes the data and provides the main descriptive statistics. Section 3.5 reports and analyzes the main results at national and regional levels. The sixth section concludes.

3

The mechanism through which the deterministic growth model is related to the quality of life is given by the relationship between the agglomeration and the cities’ growth. The agglomeration of firms within particular cities or regions implies positive effects on income and employment.

3 City Size Distribution in Colombia and Its Regions, 1835–2005

3.2 3.2.1

53

Historical Context Second Half of the Nineteenth Century

During this period, the population in Colombia was primarily rural: agricultural and livestock activities were prevalent. For example, during the period 1850–1870 there is no evidence about any ongoing urbanization process (Melo 1994). Nevertheless, there is a turning point during the rest of the nineteenth century due to an increase in trade with neighboring countries. This situation was prevailing in the border and coastal cities as well as in the main production centers and in the coffee regions, located in the northeast of the country. However, these emerging urbanization points were isolated from each other.4 In terms of road infrastructure, the country was mainly connected by bridle paths. The primary means through which goods were carried into the country was the Magdalena River, which crosses the country from south to north up to the Caribbean Sea. As it is to be expected, trading was facing high transportation costs preventing the specialization of the regions. The railway development started only in 1870, leaving aside the still premature road infrastructure development.

3.2.2

First Half of the Twentieth Century

The first decades of the twentieth century began with a series of events that changed the social and economic structure of the country. The first event was the civil war, called the “one thousand days war.” This gave rise to a second event, the relocation of the coffee crops from the northeast to the middle-west of the country, where they still remain nowadays. During the first two decades of the century, this meant a major development of rail infrastructure with the main purpose of getting the coffee production out to the external markets. Also, production centers were connected with some of the main consumption areas. Bejarano (1994) argued that, whereas the railroad network was 600 km in 1900, 22 years later it was about 1600 km. About 90% of the coffee production was transported by rail. The beginning of the industrial activities promoted a wider urbanization process, especially within the main cities: Bogotá, Medellín, Cali, and Barranquilla. These events, together with infrastructure investment, turned out to be the main determinants of rural-to-urban migration processes during these decades. On the other hand, in 1930 and 1940 the two main events were the international financial crisis and the Second World War. The first had an impact on the coffee exports, and the second contributed to an increase in the inflation rate once the USA decided to get involved in the conflict. 4 Almost half of the territory, the southeast of Colombia has historically been occupied by rain forest and jungle, where only about 4% of the population lives.

54

3.2.3

G. J. Pérez-Valbuena and A. Meisel-Roca

Second Half of Twentieth Century and the Beginning of the Twenty-First Century

During the first decades, there was evidence of economic recovery of the country, which was getting along well by keeping a moderate but constant annual growth rate of around 5%. Also, population growth was accelerating, which can be explained by improvements in medicine and the quality of life (Ocampo et al. 1994). Agricultural and livestock activities faced a reduction in their share of output, while there was a strengthening of road infrastructure and public services, especially around the main urban areas, leading to greater migration out of the countryside. These events gave rise to a regional recovery in which industrialized cities were not the only recipients of growth. During the 1960s and 1970s, the first decentralization attempts were undertaken. The last decades of the twentieth century, and the first couple of years of the twenty-first, were characterized by high levels of violence and insecurity. It was the result of the coexistence of guerrillas, paramilitaries, and drug trafficking groups with a major presence in the countryside, and resulting in a massive rural population exodus. All these historical facts in Colombia are related to population dynamics and, consequently, to the empirical Zipf’s and Gibrat’s regularities. As mentioned before, the relocation decisions of firms at the beginning of Colombia’s economic development were the main cause of migration, resulting in the producing cities to face disproportionate growth rates. As the economy developed over time, with better road infrastructure, peoples’ location decisions were more determined by the quality of life offered in the cities than them being “hot spots” for production.

3.3

Zipf’s and Gibrat’s Laws

Zipf’s law states that city size is inversely proportional to its rank, implying that the relationship fits a power law with an exponent equal to 1. Eeckhout (2004) shows that this relationship can be seen from a different perspective. Let us assume there is a group of cities, each with size Si and in descending order according to their size, each one with rank ri. Thus, city size Si is equal to 1/ri times the biggest city C, Si ¼ r1i C. If this relationship holds, the resulting graph between the rank and the city size (in logs) would be a straight line with slope equal to 1. The author also shows that Zipf’s law is a representation of the Pareto distribution r i ðSi Þ ¼ bSa i , which relates the city size Si and the rank ri(Si). The law holds when estimating the log-linear version of the relationship gives a slope coefficient a ¼ 1: ln ðr i Þ ¼ ln ðbÞ  a ln ðSi Þ þ εi

ð3:1Þ

3 City Size Distribution in Colombia and Its Regions, 1835–2005

55

which has been widely used in empirical studies of urban economics.5 Recently, Gabaix and Ibragimov (2011) showed that estimating this relationship in small samples gives a biased coefficient of a. In order to solve this problem, the authors suggest using r i  12 instead of ri:
 1 ln r i  ¼ ln ðbÞ  a ln ðSi Þ þ εi 2

ð3:2Þ

Additionally, the standard error of the slope parameter a has to be replaced by ð2=nÞ1=2b a, where n corresponds to the sample size. So far, Zipf’s law seems to imply a static analysis of the city size distribution. Nevertheless, the results derived from this law correspond to a long-run equilibrium which, in fact, suggests a dynamic examination. This fact led some authors to focus more on the dynamic aspect of the cities. In particular, Gibrat (1931) observed in a micro-study about firms that their growth was size independent, implying the firms’ size following a lognormal distribution. These results led to Gibrat’s law, also called the law of proportional growth. In urban analysis, this indicates that, independent of their size, cities grow randomly with the same mean and variance. These findings led some authors (Champernowne 1953; Simon 1955) to relate Zipf’s and Gibrat’s laws. They found a clear relationship between cities’ growth rates and the Pareto distribution. The latter arises in a natural way conditional upon the time series of the population growth following Gibrat’s law.6 Gabaix (1999a) argues that Zipf’s law should be a prerequisite for introducing a local growth model. The author refers to two means through which Zipf’s law may hold. First, while medium and big cities are different in many aspects compared to the smaller ones, there are yet some externalities affecting them both in the same way, meaning that they are actually not so different. Second, externalities affecting cities are important for defining their growth and hence it makes them different in their population dynamics. However, the advantage of large cities in terms of productivity vanishes when we start taking into account negative externalities such as traffic, pollution, and criminality, making them in fact similar to the small cities. Levy (2009) brings to the discussion the importance of the analysis of the upper tail of the city size distribution. This is because it represents the large cities and a high proportion of the national population. The author argues that if the upper tail of the distribution follows a lognormal density this does not mean that at the same time

5 This expression indicates that it is possible to empirically estimate, through a linear regression, the coefficient a, from which we can then derive whether or not the Zipf’s law holds. 6 In this case, if the growth of all cities is proportional, such as predicted by Gibrat’s law, the straight line with slope equal to 1 predicted by Zipf’s Law should move in a parallel fashion (Goerlich and Mas 2010).

56

G. J. Pérez-Valbuena and A. Meisel-Roca

it follows a Pareto distribution. This is because in the limit the upper tail of the lognormal distribution tends to follow a Pareto power law.7 Following Ioannides and Overman (2003) and Eeckhout (2004), the growth rate gi of city i can be expressed as: gi ¼ mðSi Þ þ εi

ð3:3Þ

where m(Si) ¼ E[g| Si]. For this purpose, the population growth rate is first normalized, which implies that the mean is subtracted from the population growth rate of city i in year t, and the result is then divided by the standard deviation of the growth rates of the corresponding reference group. The expression m(Si) indicates that the population growth depends on the city size (in logs). In order to not restrict the functional form in m and to provide greater functionality, the conditional mean and variance of the population growth rate are computed non-parametrically. The Nadaraya–Watson (Nadaraya 1964; Watson 1964) approach is used for this, so that estimates of the conditional mean and variance are given by:8 n1 b h ð SÞ ¼ m

n1 b σ 2h ðSÞ ¼

n P

K h ðS  Si Þgi i¼1 n P n1 K h ðS  Si Þ i¼1

n P

ð3:4Þ

b ðSÞÞ2 K h ðS  Si Þðgi  m

i¼1

n1

n P

K h ðS  S i Þ

ð3:5Þ

i¼1

In this case, Kh refers to the kernel function (Epanechnikov) whose bandwidth h is equal to 0.5.9 In terms of the regional analysis, Gabaix (1999b) demonstrated theoretically that, if in a country with heterogeneous regions Zipf’s law holds for all of them individually, it also holds for the country as a whole.

7

These arguments lead the author to assert that, when analyzing the upper tail of the city size distribution, the hypothesis testing power is low when the intention is to distinguish between a lognormal and a Pareto distribution (see Eeckhout’s (2009) response to Levy (2009)). 8 Härdle (1990) presents a detailed description on the computation of non-parametric estimates of the mean and variance. 9 In order to confirm the robustness of the results with respect to different bandwidths, additional exercises were carried out using the Silverman’s (1986) optimal bandwidth and similar results were found. For Germany, Giesen and Südekum (2011) also found, in a similar exercise, similar results using these two bandwidths.

3 City Size Distribution in Colombia and Its Regions, 1835–2005

3.4

57

Data and Descriptive Statistics

The main data source is the census population corresponding to the following years: 1835, 1843, 1851, 1870, 1905, 1912, 1918, 1938, 1951, 1964, 1973, 1985, 1993, and 2005. The 1928 census is not included because the information was not approved by the national authorities. It is important to observe that the frequency of the census data varies, especially between 1870–1905 and 1918–1938. So far, there is no clear indication in the literature about how to deal with this heterogeneity and the effects it may have on the city size analysis. The geographic units in this chapter are the municipalities, defined by the national authorities as the fundamental political and administrative division in the country.10 It is important to mention that in Colombia border expansions, especially of big cities, are often the result of mergers of some administrative units. The potential effects from this situation may be the differential population growth rates that may be observed during the period of the merger. However, the effects on the city size distribution are small as long as the number of mergers is small. For the purpose of this chapter, regions are defined as follows: Caribe (La Guajira, Magdalena, Atlántico, Bolívar, Cesar, Córdoba, and Sucre); Central (Antioquia, Caldas, Caquetá, Huila, Quindío, Risaralda, and Tolima); Eastern (Boyacá, Cundinamarca, Meta, Norte de Santander, and Santander); Pacífico (Cauca, Chocó, Nariño, and Valle del Cauca); New Departments (Amazonas, Arauca, Casanare, Guainía, Guaviare, Putumayo, Vaupés, and Vichada). This corresponds to the official geopolitical division of the country. The main reason for carrying out a regional analysis is that these regions are all different in terms of their social and economic characteristics (Pérez 2005). One important issue in city size distribution, about which there has always been controversy, is how to define a city. Some authors have defined cities as administrative units with more than 100,000 inhabitants (Rosen and Resnick 1980; Cheshire 1999; Soo 2005; Giesen and Südekum 2011), others use metropolitan areas (Dobkins and Ioannides 2001; Ioannides and Overman 2003), or consider those belonging to the upper tail of the city size distribution. Recently, there have been some authors using the full population distribution (Anderson and Ge 2005; Nota and Song 2006; González-Val 2010). The main criticism on this approach is that it might blur the definition of a city as an urban area may be compared with a rural one.11 These arguments show that it is reasonable to think that nowadays, in order to understand the urbanization processes, it is necessary to define population thresholds that better describe the urban hierarchy (Schaffar and Dimou 2012).12 In terms of the results obtained using a left-truncated distribution, Giesen and Südekum (2011) state 10

Other administrative units are the metropolitan areas. However, in Colombia they are only 10. Some recent studies have argued that using a truncated population distribution may give rise to biased Pareto coefficients (Giesen et al. 2010; Ioannides and Skouras 2013; González-Val et al. 2013). 12 In Colombia in 2015, approximately 76% of the population was living in the urban areas. 11

58

G. J. Pérez-Valbuena and A. Meisel-Roca

a

b

Note: The figure on the left (a) represents the distribution for the total population. The figure on the right (b) represents the distribution for the 50% of the population residing in the larger cities. Source: Authors’ calculations based on the National Archive and National Department of Statistics.

Fig. 3.1 Kernel densities for the city size (Colombian cities, 1835–2005)

that: “It is well known that left-truncation does not change the power law properties of this distribution, i.e. if the 100 largest cities [. . .] follow a Zipfian power law, so should the 50 largest cities” (pp. 682). In order to approach the concept of urban areas, this chapter considers the larger municipalities that together capture 50% of the population. This definition is also used to select the municipalities in each of the regions defined above.13 Fig. 3.1 shows the evolution of the city size distribution at the national level: panel (a) is for the total population, it is without any restrictions on the selection criteria. As mentioned before, it implies that at each of the census years the whole population (urban and rural) is considered. On the other hand, panel (b) shows how the city size distribution changes over time for the urban definition chosen in this chapter. If we only consider the first and the final years of the series, we can observe that the total population gradually evolved in such a way that there was only a shift in the distribution due to the natural population growth. We can also note the presence of big cities (Bogotá, Medellín, Cali, Barranquilla, and Cartagena), as shown in the right tail of the distribution in panel (a). Also, it is interesting to see the increase in the concentration as the distribution moves right, reaching the highest point in 1912. In later years, the distribution keeps moving forward and the concentration gradually goes down again, reaching in 2005 similar levels as in 1835. This behavior could be explained by the inclusion of new cities, which seems to be supported by the shape of the distribution’s upper tail (panel b). For big cities (panel b), we observe bigger changes over the nearly 200 years. When comparing the two extreme years, we can see that in 1835 population was not

13

In order to test the robustness of the results, two alternatives were also used: those including the municipalities within 90 and 95 percentiles of the population distribution. The results were similar in magnitude and significance.

3 City Size Distribution in Colombia and Its Regions, 1835–2005

59

Table 3.1 Descriptive statistics of the Colombian cities, 1835–2005 (Municipalities gathering 50% of the population) Year 1835 1843 1851 1870 1905 1912 1918 1938 1951 1964 1973 1985 1993 2005

Number of cities 149 161 166 175 163 180 178 161 124 82 60 42 33 28

Mean 5.186 5.486 6.309 7.625 13.094 13.800 15.932 26.969 46.487 106.461 190.391 357.279 500.291 762.797

Standard deviation 3.321 3.201 2.772 3.701 9.298 10.700 13.050 33.202 77.596 214.350 405.843 687.004 881.159 1.292.902

Median 4.376 4.782 5.446 6.823 10.858 10.935 12.757 19.110 26.225 45.060 75.219 167.832 248.525 380.061

Minimum size 3.199 3.402 4.034 4.878 7.304 8.110 9.032 13.722 17.975 30.637 42.131 79.893 109.115 187.249

Maximum size 39.442 40.086 29.649 40.883 100.000 121.257 143.994 355.502 715.250 1.697.311 2.861.913 4.236.490 4.945.448 6.840.116

Gini index 0.213 0.196 0.189 0.185 0.254 0.248 0.262 0.350 0.460 0.560 0.606 0.581 0.554 0.535

Source: Authors’ calculations based on the National Archive and National Department of Statistics

only smaller but also much more concentrated in medium-sized cities (with an average of 5186 inhabitants and standard deviation of 3321). Subsequently, the urban population distribution changed to one in which there is a higher prevalence of big cities. Tables 3.1 and 3.2 summarize the information of selected population for the criteria defined previously, at national and regional levels. Appendices 1 and 2 show descriptive statistics for both national and regional levels. At both national and regional levels, we observe the primacy of big cities, which is consistent with the results described before. This pattern of urban dynamics can be explained inter alia by historical, geographical, institutional, and political factors. The effect of big cities on the dynamics of the population can be seen by the primacy indicator, which in Colombia went from 1.22 in 1835 to 1.25 in 2005. When computing this indicator for the regions, we observe all of them increasing over time but with a high dispersion. The only exception is the Caribbean region for which primacy looks quite stable (changing from 0.59 to 0.65 between 1835 and 2005). The rest of the regions are characterized by an increase in the primacy index: Central (0.42–1.68), Eastern (1.37–4.54), and Pacific (0.43–2.13). These results are consistent with the Gini inequality index (last columns in Tables 3.1 and 3.2), which steadily increased over time. Figure 3.2 shows the spatial population distribution of municipalities with over 7000 inhabitants in Colombia in the nineteenth and twentieth centuries.14 The results reveal at least three characteristics. The two most evident ones are the ones related

14 An arbitrary city size is taken in order to facilitate visualization of changes in the number and size of cities over time.

60

G. J. Pérez-Valbuena and A. Meisel-Roca

Table 3.2 Descriptive statistics (Colombian regions, 1835–2005) (Municipalities gathering 50% of the population) Number of Year cities (a) Caribbean region 1835 58 1843 61 1851 61 1870 52 1905 54 1912 49 1918 45 1938 40 1951 39 1964 40 1973 47 1985 43 1993 40 2005 40 (b) Central region 1835 51 1843 59 1851 64 1870 67 1905 76 1912 81 1918 86 1938 88 1951 82 1964 78 1973 79 1985 74 1993 65 2005 53 (c) Eastern region 1835 124 1843 128 1851 131 1870 140 1905 141 1912 146 1918 144 1938 136 1951 119 1964 85

Mean

Standard deviation

Median

Minimum size

Maximum size

Gini index

2.711 2.713 2.905 4.357 6.823 10.507 12.976 24.989 34.567 56.854 68.796 101.955 119.867 158.421

1.879 1.701 1.725 2.140 6.062 7.938 10.922 26.051 46.277 82.268 109.939 158.681 181.150 228.043

2.132 1.996 2.300 3.416 4.916 7.798 10.001 15.631 19.064 31.685 34.125 50.298 53.968 70.611

1.240 1.234 1.278 2.383 3.138 5.311 5.797 10.111 12.713 20.834 24.086 32.142 36.474 42.542

11.929 10.145 9.896 11.595 40.115 48.907 64.543 152.348 283.238 498.301 703.488 927.233 993.759 1.146.498

0.312 0.299 0.294 0.245 0.342 0.313 0.328 0.396 0.437 0.464 0.495 0.512 0.526 0.533

3.994 4.241 4.869 6.557 12.148 13.063 14.262 21.809 31.285 47.446 56.543 76.487 93.312 143.159

1.888 1.775 2.156 3.485 6.983 8.420 9.439 19.633 42.221 90.933 134.003 177.522 209.432 310.588

3.390 3.494 4.106 5.538 10.224 10.586 11.586 16.193 20.516 25.695 27.940 32.298 38.660 53.507

2.261 2.509 2.738 3.872 6.312 6.834 7.122 11.603 14.429 17.831 19.042 24.264 27.038 2.214.494

10.280 9.118 13.755 29.765 53.936 71.004 79.146 168.266 363.865 772.887 1.163.868 1.480.382 1.630.009 2.214.494

0.237 0.217 0.223 0.211 0.252 0.257 0.269 0.298 0.375 0.476 0.526 0.556 0.581 0.597

4.590 4.883 5.703 6.553 8.524 9.057 10.320 14.835 21.030 45.215

3.493 3.552 3.044 3.750 8.391 9.847 11.833 30.202 65.479 183.941

3.894 4.279 5.022 5.842 6.639 7.435 8.368 10.516 11.958 16.467

2.571 2.794 3.172 3.709 4.840 5.011 5.672 6.896 8.026 11.677

39.442 40.086 29.649 40.883 100.000 121.257 143.994 355.502 715.250 1.697.311

0.251 0.234 0.224 0.223 0.266 0.260 0.268 0.360 0.477 0.639 (continued)

3 City Size Distribution in Colombia and Its Regions, 1835–2005

61

Table 3.2 (continued) Number of Year cities 1973 60 1985 37 1993 20 2005 14 (d) Pacific region 1835 41 1843 43 1851 45 1870 54 1905 43 1912 53 1918 55 1938 56 1951 48 1964 44 1973 39 1985 30 1993 32 2005 30

Mean 85.360 186.449 382.818 715.672

Standard deviation 368.247 689.286 1.080.676 1.770.276

Median 22.516 41.464 89.766 172.068

Minimum size 15.436 23.874 45.696 107.417

Maximum size 2.861.913 4.236.490 4.945.448 6.840.116

Gini index 0.734 0.760 0.732 0.714

3.445 4.204 4.664 5.151 11.234 10.301 11.637 19.202 32.204 51.271 75.877 122.421 129.778 172.622

1.443 1.790 2.034 2.499 6.825 5.276 7.112 14.226 40.897 94.717 155.686 254.493 288.482 379.397

3.027 3.709 4.034 4.324 9.500 8.918 9.109 14.084 19.093 25.500 35.401 48.511 50.903 63.131

1.987 2.392 2.754 2.543 5.695 5.659 6.114 10.294 14.844 17.836 22.007 30.803 28.978 34.710

8.173 10.376 11.848 12.743 30.835 27.760 45.525 101.883 284.186 637.929 991.549 1.429.026 1.666.468 2.119.843

0.212 0.209 0.218 0.248 0.291 0.248 0.276 0.291 0.400 0.499 0.541 0.583 0.603 0.620

Source: Authors’ calculations based on the National Archive and National Department of Statistics

with the increase in the number of cities and the corresponding city size growth. As described by González-Val (2010), these are typical features found in relatively young countries, even in the USA. In contrast, European countries show signs of having achieved their population’s long-term growth path, not only in terms of the number of cities but also in size. A third characteristic is the location of the main urban centers. It is noticeable that geographical factors have been crucial in determining the structure of population settlements, in particular those along the Andean mountain range, which crosses South America from south to north and, once the Colombian territory is reached, splits out into three different mountain chains. Historically, the main urban centers have been mostly located side to side of the three mountain chains and close to seaports, especially those in the Caribbean Sea. In order to better understand the population distribution in Colombia and its evolution over time, the link with the distribution of natural resources and the geographical characteristics has to be analyzed. Colombia has access to the sea from both the Caribbean in the northwest and the Pacific in the west. Moreover, the country is endowed with two rivers that run from south to north, Magdalena and Cauca. These two rivers were determining factors in the evolution of the population distribution, not only because of the relationship with agricultural and livestock activities but also because historically these were the only means for cargo

62

G. J. Pérez-Valbuena and A. Meisel-Roca

(b) 1951 (a) 1835

(c) 2005

Source: Authors’ calculations based on the National Archive and National Department of Statistics.

Fig. 3.2 Spatial population distribution in Colombian municipalities

transportation from the hinterland to the coasts. These factors contributed to the decisions of the firms to locate close to the water, which facilitated the transportation of the products to the external markets.

3 City Size Distribution in Colombia and Its Regions, 1835–2005

63

Table 3.3 Pareto coefficients (national city size distribution, 1835–2005) Year 1835 1843 1851 1870 1905 1912 1918 1938 1951 1964 1973 1985 1993 2005

Number of cities 149 161 166 175 163 180 178 161 124 82 60 42 33 28

Pareto coefficient 2.767 2.946 3.074 3.104 2.397 2.479 2.363 1.878 1.488 1.217 1.092 1.104 1.133 1.168

Standard deviation 0.321 0.328 0.337 0.332 0.266 0.261 0.250 0.209 0.189 0.190 0.199 0.241 0.279 0.312

t-statistic Ho: β ¼ 1 5.511 () 5.926 () 6.147 () 6.340 () 5.262 () 5.659 () 5.441 () 4.194 () 2.583 () 1.140 0.463 0.433 0.476 0.539

Note: Null hypothesis is rejected at 5% () and at 1% ()

3.5

Results

In this section, we analyze the characteristics of the rank–size relationship at the national and regional levels; and test if Zipf’s law holds. Then, with the purpose of characterizing the urban population growth (Gibrat’s law) and the city size growth for Colombia and its regions, the population distribution is analyzed over 10-year periods.

3.5.1

National Analysis

Table 3.3 shows the rank–size, or Pareto, coefficients for the population censuses corresponding to the 1835–2005 period, and Fig. 3.3 shows the corresponding rank– size population relationship (in logs) for the first and the last available census data. In all cases, ordinary least squares (OLS) was run with the Gabaix and Ibragimov’s (2011) small size bias correction for both the coefficients and the standard errors. These results show evidence of Zipf’s law, but only from the second half of the twentieth century, 1964 in particular. Even though the Pareto coefficient has fallen over time, from this year onwards there seems to be a stabilization pattern around the optimal value of 1. For those cases for which the null hypothesis (Pareto coefficient ¼ 1) is not rejected, the coefficient remains a little larger than unity. Giesen and Südekum (2011) argue that this situation is likely without leading to a rejection of Zipf’s law. They explain that the optimal value of 1 is only obtained in the limit. In general, one of the factors leading to deviations of the Pareto coefficient from unity is the sample size. González-Val (2010) states that empirical research has proven the

64

G. J. Pérez-Valbuena and A. Meisel-Roca

(a) 1835

(b) 2005

Source: Authors’ calculations based on the National Archive and National Department of Statistics.

Fig. 3.3 Rank–size relationships (Colombia, 1835–2005)

high sensitivity of the results to two factors: the spatial unit and the sample size. Eeckhout (2004) combines this characterization into just one issue: the selected truncation point in the distribution of city sizes. Given the relationship between Zipf’s and Gibrat’s laws, an approximation of the rank–size coefficient to 1 would also mean that Gibrat’s law, in which cities, would have similar growth rates regardless of their size, holds.15 A quick look at the behavioral time pattern of the Pareto coefficient and the number of municipalities included seems to show a causal relationship. The coefficient decreases over time as the number of municipalities reduces. Nevertheless, additional robustness exercises, such as those using the alternative selection criteria mentioned before, showed similar coefficients and patterns even for those cases where the number of municipalities increased over time. These results offer evidence about robustness of the criteria used for choosing the set of municipalities. Considering the urban growth theories, the country as a whole would be facing a transition period from deterministic to stochastic growth. In other words, the growth of cities is no longer determined mostly by firms’ location decisions and the externalities of human capital. The new urban growth model is characterized by a higher mobility of the labor force and the amenities offered in the urban centers, most of them coming from the implementation of public policies (Duranton 2006, 2007). According to Gibrat’s law, cities grow, irrespective of their size, randomly with common mean and variance. The verification of this empirical regularity would then imply a formal test of the random growth hypothesis of the cities. If a group of cities follow this rule, it implies that the scale effect does not play a central role in the cities’ growth pattern (Schaffar and Dimou 2012). Additionally, Giesen and Südekum (2011) and Härdle (1990) offer some arguments in favor of Gibrat’s test instead of Zipf’s test. The first argument is that carrying out the test through

15

This relationship holds as long as particular statistical conditions holds (Gabaix 1999b; Skouras 2009).

3 City Size Distribution in Colombia and Its Regions, 1835–2005

65

non-parametric methods does not lead to a specific functional form, and then offers major flexibility in terms of the hypothesis testing. The second argument deals with the outcome of the test, which is not only based on a single statistic. Rather, the results considered are along the city size distribution. To empirically verify both Gibrat’s and Zipf’s laws, Fig. 3.4 shows the results for the non-parametric estimates of the mean and variance of the population growth in Colombia as a whole during the last two centuries. To facilitate the reading of the results, the period of analysis 1835–2005 was divided into three sub-periods as follows: 1835–1870, 1905–1951, and 1964–2005. It is important to remember again that Gibrat’s law implies that cities grow with the same mean and variance, and that their growth should be independent of the size. Because the population series were standardized (zero mean and unit variance), it is expected that any deviation from a mean of zero and any deviation from a variance of unity imply rejection of Gibrat’s law. One plausible explanation for the change in the population dynamics from the first to the second half of the twentieth century is the development of the road infrastructure in the country. It is precisely between 1920 and 1960 when Colombia faced one of the most ambitious transformations in the road infrastructure so far (Bonet and Meisel 1999). For the first time, the country succeeded in connecting the main regions throughout the country by means of railroad and road infrastructure, with huge economic and demographic effects. Some of the economic growth theories argue that one of the expected effects from greater regional integration is an increase in economic disparities and strengthening of a center–periphery model (Krugman 1990). Under this approach, there is a clear socioeconomic and demographic differentiation between the predominant city (center) and the remaining dependent cities or regions (periphery). The following section tests these arguments for the population of the regions in Colombia.

3.5.2

Regional Analysis

This subsection has two objectives. First, it characterizes the city size distribution of the Colombian regions since the implementation of the Republic regime, establishing whether or not the growth patterns follow the empirical regularities predicted by Zipf’s and Gibrat’s laws. Secondly, it tests the theoretical prediction of Gabaix (1999b) that if Zipf’s law holds for the regions individually, it also holds for the country. As mentioned before, five regions were defined: Caribbean, Central, East, Pacific, and New Departments. This last region was not included in the analysis due to the small sample size. Table 3.4 shows the results for the rank–size estimations of the remaining four regions. The results are interesting in several ways. First, the urban population during the first decades of the Republic (panel a) showed a deterministic type of growth, with some big cities growing at different rates compared to other big cities. This is not in accordance with Gibrat’s law. Secondly, the city size growth inherited from the nineteenth century seems to still be present in the first decade of the twentieth century. Panel b shows a similar pattern to that in the nineteenth century where the

66

G. J. Pérez-Valbuena and A. Meisel-Roca

Fig. 3.4 Non-parametric results for the Nadaraya–Watson estimates. Mean and variance of the population growth at national level

3 City Size Distribution in Colombia and Its Regions, 1835–2005

67

Table 3.4 Pareto coefficients for the city size distribution (Colombian regions, 1835–2005) Year Number of cities (a) Caribbean region 1835 58 1843 61 1851 61 1870 52 1905 54 1912 49 1918 45 1938 40 1951 39 1964 40 1973 47 1985 43 1993 40 2005 40 (b) Central region 1835 51 1843 59 1851 64 1870 67 1905 76 1912 81 1918 86 1938 88 1951 82 1964 88 1973 79 1985 74 1993 65 2005 53 (c) Eastern region 1835 124 1843 128 1851 131 1870 140 1905 141 1912 145 1918 144 1938 136 1951 119 1964 85 1973 60 1985 37 1993 20 2005 14

Pareto coefficient

Standard deviation

t-statistic Ho: β ¼ 1

1.852 1.901 1.897 2.287 1.780 1.889 1.816 1.546 1.459 1.391 1.322 1.247 1.183 1.139

0.344 0.344 0.343 0.449 0.342 0.382 0.383 0.346 0.330 0.311 0.273 0.269 0.264 0.255

2.476 () 2.616 () 2.611 () 2.869 () 2.276 () 2.329 () 2.132 () 1.579 1.389 1.258 1.181 0.918 0.691 0.545

2.343 2.536 2.521 2.676 2.287 2.291 2.186 2.092 1.754 1.456 1.350 1.251 1.159 1.089

0.464 0.467 0.446 0.462 0.371 0.360 0.333 0.315 0.274 0.233 0.215 0.206 0.203 0.211

2.894 () 3.289 () 3.412 () 3.625 () 3.469 () 3.586 () 3.557 () 3.462 () 2.752 () 1.955 1.628 1.220 0.783 0.419

2.390 2.526 2.605 2.586 2.323 2.383 2.341 1.954 1.634 1.269 1.065 0.925 0.852 0.806

0.304 0.316 0.322 0.309 0.277 0.280 0.276 0.237 0.212 0.195 0.194 0.215 0.270 0.305

4.579 () 4.832 () 4.986 () 5.131 () 4.781 () 4.942 () 4.860 () 4.026 () 2.991 () 1.382 0.333 0.348 0.547 0.638 (continued)

68

G. J. Pérez-Valbuena and A. Meisel-Roca

Table 3.4 (continued) Year Number of cities (d) Pacific region 1835 41 1843 43 1851 45 1870 54 1905 43 1912 53 1918 55 1938 56 1951 48 1964 44 1973 39 1985 30 1993 32 2005 30

Pareto coefficient

Standard deviation

t-statistic Ho: β ¼ 1

2.574 2.666 2.563 2.215 1.926 2.286 2.116 2.052 1.578 1.333 1.237 1.106 1.077 1.010

0.569 0.575 0.540 0.426 0.415 0.444 0.403 0.388 0.322 0.284 0.280 0.286 0.269 0.261

2.768 () 2.897 () 2.892 () 2.850 () 2.229 () 2.896 () 2.765 () 2.712 () 1.794 1.172 0.846 0.371 0.286 0.039

Note: Null hypothesis is rejected at 5% () and at 1% () Source: Authors’ calculations based on Census data

mean and variance of the population growth clearly depended on the city size. Nevertheless, the variance seems to move within the confidence intervals along the whole city size distribution, showing some stabilization pattern. Thirdly, the situation changed since the second half of the twentieth century and the first years of the twenty-first century. During the latter period, there appears to be evidence of population growth as predicted by Gibrat’s law. For both the mean and variance, we cannot reject the hypothesis of independence between city size and population growth. This is consistent with the estimated rank–size coefficients. Three conclusions can be drawn from these results. First, all regions’ patterns are similar to the national ones. During the whole nineteenth century and the first decade of the twentieth century, the dynamics of population growth is different from growth during the second half of the twentieth century and the first years of the twenty-first century. The dynamics of population growth in the regions started to change, each at their own speed. They moved from a deterministic to a random growth model. In the first model, population’s growth was conditioned to geographical locations where production activities were carried out. In the second, population grows faster in those places where a better quality of life and amenities are offered, which can be provided by the local authorities.16

16

Notice, for instance, that in the Caribbean region Zipf’s Law began to hold before it did in the other regions. Although this chapter does not pretend to answer the question why this is so, one possible explanation is that, historically, this region was one of the first to be populated, mainly due to its importance as a seaport and a “hot spot” for international trade.

3 City Size Distribution in Colombia and Its Regions, 1835–2005

69

The second conclusion is related to the value of the estimates. Although the null hypothesis (β ¼ 1) cannot be rejected, in some cases the coefficients substantially deviate from unity. This situation occurs even though Zipf’s law continues to hold formally. According to the literature, the causes of these deviations include the sample size, the geographical unit, the truncation point of the distribution, and the emergence of new urban centers, provided the rates at which the cities are created are not greater than growth rates of the existing cities (Gabaix 1999b). All this is consistent with the fact that Colombia is a developing country, where population dynamics is still in transition and new urban centers still emerge along with significant city size growth of the existing ones. This situation contrasts with that of several European countries where the number and size of urban centers seems to have reached the long-term steady state several decades ago (González-Val 2010). The third conclusion, which should be analyzed together with the national results, is that there seems to be evidence for Colombia in favor of the theoretical argument given by Gabaix (1999b). The results show a temporal coincidence of Zipf’s law at both individual regions and for the country as a whole, similar to what Giesen and Südekum (2011) found by means of using German data. The next exercise, for each of the four regions, requires computing the non-parametric estimators for the mean and variance of the population growth. The results can be observed in Fig. 3.5, where only for two of the regions, the Caribbean and the Eastern regions, there is evidence supporting the argument that urban areas within their own regions grew with constant means and variances between 1964 and 2005.17 For the Central and Pacific regions, and for the whole 1964–2005 period, there is evidence, at the 99% level of confidence, of differential population growth and city size dependence over different sections of the distribution. These results are likely to be affected by the reasons mentioned before about the potential deviations of the rank–size parameters. One possible additional explanation for the regions’ differential behavior is offered by Bonet and Meisel (1999). The authors found evidence that the regions faced since 1960 regional economic disparities. The possible explanations include firstly, the impact of the industrialization policies given by the import substitution model; secondly, the social and economic strengthening of the existing cities, especially the capital city Bogotá; and thirdly, the weakening of the relative importance of the Caribbean region’s departments. These factors are likely to have played a part in favoring some regions rather than the others, leading to a corresponding economic growth and demographic differential between them.

17 The results are only given for this particular period because for all regions and for the rest of the series the hypotheses of constant means and variances were rejected. Additionally, it is of major interest to check the results for Gibrat’s and Zipf’s laws only for the periods where the latter law holds.

70

G. J. Pérez-Valbuena and A. Meisel-Roca

(a) Caribbean Region

(c) Eastern Region

(b) Central Region

(d) Pacific Region

Fig. 3.5 Non-parametric estimators (Nadaraya–Watson) for the mean and variance of the population growth (1964–2005). Regions of Colombia

3.6

Conclusions

This chapter has two main findings. First, the growth of the Colombian cities changed during the second half of the twentieth century. During the Republican period, in the nineteenth century and the first half of the twentieth century, the urban growth was characterized by city size dependence patterns. Small cities grew at different speeds compared to the medium and the big ones, violating Zipf’s law and Gibrat’s law. According to theory, such observed behavior is described by the so-called deterministic population growth model, whereby the demographic dynamics are mainly influenced by the location of production centers. During the second half of the twentieth century, the population growth model started a new transition in which Zipf´s law and Gibrat’s laws hold. Cities grew at the same rate, irrespective of their size. This is known as the stochastic population growth model, which is affected by exogenous shocks, especially policies that improved quality of life and urban amenities.

3 City Size Distribution in Colombia and Its Regions, 1835–2005

71

Secondly, there is a coincidence, in terms of Zipf’s law, between national and regional city sizes’ dynamics. City size distributions exhibit a linear rank–size relationship in each of the regions individually and the country as a whole, as theoretically demonstrated by Gabaix (1999b). Nevertheless, when the Gibrat’s tests are compared, the national trend is coincident with only two out of the four regions analyzed, which can be explained by particular characteristics in Colombia at the time, such as the consolidation of the transportation infrastructure developed during the first half of the twentieth century. In short, from the second half of the twentieth century, the government represented by the local authorities had a bigger influence in urban dynamics. Colombia and its regions depend nowadays, as predicted by the stochastic growth approach, much more on public policies to change the speed of growth of the city population. From this point onwards, local governments are able to affect the longterm urban population growth by means of public policies. If mayors and governors have not done anything to make their cities attractive, they will have to face being part of the “slow growth” club. Disclaimer A Spanish version of this research was published in Revista de Historia Económica/ Journal of Iberian and Latin American Economic History (New Series), Volume 32/Issue 02/September 2014, pp 247–286.

Appendix 1 Descriptive statistics of the Colombian cities, 1835–2005 (Municipalities representing 100% of the population) Year 1835 1843 1851 1870 1905 1912 1918 1938 1951 1964 1973 1985 1993 2005

Number of cities 749 751 802 739 762 767 806 809 827 879 1020 1024 1061 1113

Mean 2.070 2.360 2.617 3.624 5.616 6.486 7.056 10.745 13.944 19.875 22.403 29.352 31.206 38.524

Standard deviation 2.270 2.355 2.461 3.045 6.004 6.759 7.955 17.079 33.148 71.058 106.622 153.862 175.391 233.820

Median 1.512 1.846 2.010 2.845 4.175 5.054 5.408 7.309 8.330 10.093 10.118 12.000 11.343 12.626

Source: Authors’ calculations based on Census data

Minimum size 34 63 37 58 97 33 160 399 347 294 85 797 78 225

Maximum size 39.442 40.086 29.649 40.883 100.000 121.257 143.994 355.502 715.250 1.697.311 2.861.913 4.236.490 4.945.448 6.840.116

Gini index 0.468 0.445 0.459 0.392 0.423 0.387 0.411 0.442 0.508 0.581 0.635 0.669 0.701 0.726

72

G. J. Pérez-Valbuena and A. Meisel-Roca

Appendix 2 Descriptive statistics of the Colombian regions, 1835–2005 (Municipalities representing 100% of the population) Number of Year cities (a) Caribbean region 1835 182 1843 187 1851 189 1870 126 1905 145 1912 118 1918 119 1938 113 1951 117 1964 128 1973 161 1985 161 1993 162 2005 194 (b) Central region 1835 127 1843 135 1851 156 1870 159 1905 176 1912 184 1918 203 1938 211 1951 217 1964 236 1973 270 1985 273 1993 273 2005 278 (c) Eastern region 1835 315 1843 306 1851 321 1870 314 1905 322 1912 323 1918 324

Mean

Standard deviation

Median

Minimum size

Maximum size

Gini index

1.236 1.269 1.344 2.595 3.641 6.277 7.018 12.707 16.541 25.490 28.693 38.933 42.404 46.861

1.480 1.416 1.475 2.051 4.466 6.270 8.202 17.967 29.526 50.441 64.513 89.988 99.832 117.517

765 761 791 1.950 2.506 4.473 4.472 8.177 10.663 14.614 15.206 18.962 19.631 19.065

74 70 93 390 97 184 726 1.412 1.970 2.318 2.624 3.676 3.840 2.721

11.929 10.145 9.896 11.595 40.115 48.907 64.543 152.348 283.238 498.301 703.488 927.233 993.759 1.146.498

0.512 0.497 0.498 0.387 0.462 0.403 0.448 0.481 0.512 0.535 0.562 0.580 0.595 0.628

2.300 2.661 2.876 3.949 7.516 8.257 8.665 13.004 16.970 22.435 23.692 29.656 31.748 39.136

1.895 1.887 2.229 3.258 6.191 7.109 7.896 14.809 28.290 55.056 75.289 96.443 107.418 143.948

1.911 2.276 2.344 3.336 5.566 6.396 6.454 9.755 11.615 13.303 12.897 14.922 14.858 16.079

79 66 198 281 464 547 276 474 1.255 2.297 918 2.005 2.329 2.690

10.280 9.118 13.755 29.765 53.936 71.004 79.146 168.266 363.865 772.887 1.163.868 1.480.382 1.630.009 2.214.494

0.418 0.370 0.397 0.388 0.375 0.368 0.390 0.408 0.461 0.526 0.566 0.593 0.623 0.663

2.590 2.923 3.338 4.174 5.350 5.870 6.555

2.813 2.891 2.847 3.358 6.268 7.261 8.618

2.006 2.454 2.766 3.446 4.280 4.732 5.163

71 63 51 58 262 567 524

39.442 40.086 29.649 40.883 100.000 121.257 143.994

0.425 0.399 0.407 0.362 0.379 0.359 0.373 (continued)

3 City Size Distribution in Colombia and Its Regions, 1835–2005 Number of Year cities 1938 334 1951 336 1964 345 1973 383 1985 384 1993 391 2005 396 (d) Pacific region 1835 98 1843 94 1851 103 1870 130 1905 107 1912 118 1918 127 1938 131 1951 136 1964 141 1973 153 1985 153 1993 163 2005 177

73

Mean 8.656 10.666 15.937 19.140 25.672 28.064 36.264

Standard deviation 19.937 39.638 92.455 147.547 217.821 252.483 346.825

Median 6.111 6.379 7.396 7.029 7.767 6.982 7.503

Minimum size 681 785 965 833 797 270 885

Maximum size 355.502 715.250 1.697.311 2.861.913 4.236.490 4.945.448 6.840.116

Gini index 0.428 0.500 0.615 0.691 0.743 0.786 0.822

2.067 2.782 2.948 3.059 6.481 6.669 7.247 11.804 16.388 22.892 27.733 34.534 36.422 41.955

1.546 1.832 2.080 2.423 5.888 4.893 6.128 11.393 26.951 55.991 82.912 119.513 134.530 165.143

1.675 2.320 2.377 2.176 4.656 5.272 5.634 9.209 10.184 13.374 13.040 14.119 15.329 15.696

156 179 199 480 325 938 261 1.109 954 2.149 1.938 3.661 2.063 3.481

8.173 10.376 11.848 12.743 30.835 27.760 45.525 101.883 284.186 637.929 991.549 1.429.026 1.666.468 2.119.843

0.389 0.335 0.359 0.391 0.414 0.351 0.379 0.391 0.485 0.544 0.593 0.643 0.648 0.672

Note: Regions are defined as follows: Caribbean (La Guajira, Magdalena, Atlántico, Bolívar, Cesar, Córdoba, and Sucre); Central (Antioquia, Caldas, Caquetá, Huila, Quindío, Risaralda, and Tolima); Eastern (Boyacá, Cundinamarca, Meta, Norte de Santander, and Santander); Pacific (Cauca, Chocó, Nariño, and Valle); New Departments (Amazonas, Arauca, Casanare, Guainía, Guaviare, Putumayo, Vaupés, and Vichada) Source: Authors’ calculations based on Census data

References Anderson G, Ge Y (2005) The size distribution of Chinese cities. Reg Sci Urban Econ 35 (6):756–776 Auerbach F (1913) Das gesetz der bevölkerungskonzentration. Petermanns Geographische Mitteilungen 59:74–76 Basu B, Bandyapadhyay S (2009) Zipf’s law and distribution of population in Indian cities. Indian J Phys 83(11):1575–1582 Bejarano J (1994) El despegue cafetero (1900-1928). In: J. A. Ocampo (compilador) Histórica Económica de Colombia. TM Editores-Fedesarrollo, Bogotá, pp 173–207 Benguigui L, Blumenfeld-Lieberthal E (2011) The end of a paradigm: is Zipf’s law universal? J Geogr Syst 13(1):87–100 Bernal G, Nieto C (2006) Evolución del coeficiente de Zipf para Colombia en el siglo XX, Documentos de economía no. 5, Universidad Javeriana-Departamento de Economía

74

G. J. Pérez-Valbuena and A. Meisel-Roca

Bonet J, Meisel A (1999) La convergencia regional en Colombia: una visión de largo plazo, 1926–1995. Documentos de trabajo sobre economía regional no. 8, Banco de la RepúblicaCartagena Champernowne DG (1953) A model of income distribution. Econ J 63(250):318–351 Cheshire P (1999) Trends in sizes and structures of urban areas. In: Cheshire PC, Mills ES (eds) Handbook of regional and urban economics, vol 3. North-Holland, Amsterdam, pp 1339–1373 Dimou M, Schaffar A (2009) Urban hierarchies and city-growth in the Balkans. Urban Stud 46 (13):2891–2906 Dobkins L, Ioannides Y (2001) Spatial interactions among US cities: 1900-1990. Reg Sci Urban Econ 31(6):701–731 Duranton G (2006) Some foundations for Zipf’s law: product proliferation and local spillovers. Reg Sci Urban Econ 36(4):542–563 Duranton G (2007) Urban evolutions: the fast, the slow, and the still. Am Econ Rev 97(1):197–221 Eeckhout J (2004) Gibrat’s law for (all) cities. Am Econ Rev 94(5):1429–1451 Eeckhout J (2009) Gibrat’s law for (all) cities: reply. Am Econ Rev 99(4):1676–1683 Firdaus M, Fitria A (2010) Does the rank-size rule matter in Indonesia? Determinants of the size distribution of cities. J Indones Econ Business 25(1):114–120 Gabaix X (1999a) Zipf’s law and the growth of cities. Am Econ Rev 89(2):129–132 Gabaix X (1999b) Zipf’s law for cities: an explanation. Q J Econ 114(3):739–767 Gabaix X (2016) Power laws in economics: an introduction. J Econ Perspect 30(1):185–205 Gabaix X, Ibragimov R (2011) Rank-1/2: a simple way to improve the OLS estimation of tail exponents. J Business Econ Stat 29(1):24–39 Gibrat R (1931) Les inégalités économiques. Recueil Sirey, Paris Giesen K, Südekum J (2011) Zipf’s law for cities in the regions and the country. J Econ Geogr 11 (4):667–686 Giesen K, Zimmermann A, Suedekum J (2010) The size distribution across all cities—double Pareto lognormal strikes. J Urban Econ 68:129–137 Goerlich F, Mas M (2010) La distribución empírica del tamaño de las ciudades en España, 19002001. Quién verifica la Ley de Zipf, Revista de Economía Aplicada XVIII 54:133–159 González-Val R (2010) The evolution of U.S. city size distribution from a long-term perspective (1900–2000). J Reg Sci 50(5):952–972 González-Val R, Ramos A, Sanz-García F, Vera-Cabello M (2013) City distribution for all cities: which one is best. Pap Reg Sci 94(1):177–196 Härdle W (1990) Applied nonparametric regression. Cambridge University Press, Cambridge Ioannides Y, Overman H (2003) Zipf’s law for cities: an empirical examination. Reg Sci Urban Econ 33(2):127–137 Ioannides Y, Skouras S (2013) US city size distribution: robustly Pareto, but only in the tail. J Urban Econ 73:18–29 Krugman P (1990) Increasing returns and economic geography. Working paper no. w3275. National Bureau of Economic Research Levy M (2009) Gibrat’s law for (all) cities: comment. Am Econ Rev 99(4):1672–1675 Matlaba VJ, Holmes MJ, Mccann P, Poot J (2013) A century of the evolution of the urban system in Brazil. Rev Urban Reg Dev Stud 25(3):129–151 Melo J (1994) Las vicisitudes del modelo liberal (1850-1899). In: J. A. Ocampo (compilador) Histórica Económica de Colombia. TM Editores-Fedesarrollo, Bogotá, pp 119–172 Moura NJ, Ribeiro MB (2006) Zipf’s law for Brazilian cities. Phys A Stat Mech Appl 367:441–448 Nadaraya E (1964) On estimating regression. Theor Probab Appl 9:141–142 Nota S, Song F (2006) Further analysis of the Zipf’s law: does the rank-size rule really exist, UNR Joint Economics Working Paper Series, No. 07-2006. University of Nevada, Reno Ocampo J, Bernal J, Avella M, Errázuriz M (1994) La consolidación del capitalismo moderno (1945-1986). In: J. A. Ocampo (compilador) Histórica Económica de Colombia. TM EditoresFedesarrollo, Bogotá, pp 243–334

3 City Size Distribution in Colombia and Its Regions, 1835–2005

75

Pérez GJ (2005) Dimensión espacial de la pobreza en Colombia. Revista ensayos sobre política económica 48:234–293 Pérez GJ (2006) Población y Ley de Zipf en Colombia y la Costa Caribe, 1912–1993, Documentos de trabajo sobre economía regional no. 71, Banco de la República–CEER Rosen K, Resnick M (1980) The size distribution of cities: an examination of the Pareto law and primacy. J Urban Econ 8(2):165–186 Schaffar A, Dimou M (2012) Rank-size city dynamics in China and India, 1981–2004. Reg Stud 46 (6):707–721 Silverman B (1986) Density estimation for statistics and data analysis. Chapman and Hall, New York Simon H (1955) On a class of skew distribution functions. Biometrika 42(3-4):425–440 Skouras S (2009) Explaining Zipf’s law for US cities. Available at SSRN 1527497 Song S, Zhang K (2002) Urbanization and city-size distribution in China. Urban Stud 39 (12):2317–2327 Soo K (2005) Zipf’s law for cities: a cross-country investigation. Reg Sci Urban Econ 35 (3):239–263 Watson G (1964) Smooth regression analysis. Sankhya 26:359–376 Ye X, Xie Y (2012) Re-examination of Zipf’s law and urban dynamic in China: a regional approach. Ann Reg Sci 49(1):135–156 Zhou Y, Ma L (2003) China’s urbanization levels: reconstructing a baseline from the fifth population census. China Q 173:176–196 Zipf G (1949) Human behavior and the principle of least effort. Addison-Wesley, Cambridge, MA

Chapter 4

Exploring Economic Futures for Japan Under Rapid Depopulation: A Dynamic Regional CGE Model Approach Suminori Tokunaga and Mitsuru Okiyama

Abstract In this chapter, we empirically investigate the impact of future population decline on the Japanese regional economy, using a multi-regional computable general equilibrium model. In this model, we combine forecasts of potential economic growth with a scenario of labor force reduction due to population decline. We project the labor force up to 2040. We find that unless economic productivity in Japan’s “depopulating society” improves, economic growth will be virtually zero initially and turn to negative by 2030. However, economic growth could reach 0.7–0.8% if policies are implemented to increase productivity, but such economic growth will be mainly driven by the growth in urban areas and it will increase the disparities in growth among regions. Consequently, the simulation results indicate that incorporating the impact of industrial clusters on regional areas yields growth that is on a par with that of urban areas. The construction of wide-ranging industrial clusters will be effective in preventing the exacerbation of inter-regional economic welfare disparities. Keywords Population decline · Labor force · CGE model · Japan

S. Tokunaga (*) Chikuro Hiroike School of Graduate Studies, Reitaku University, Kashiwa, Chiba, Japan Faculty of Economics and Business Administration, Reitaku University, Kashiwa, Chiba, Japan e-mail: [email protected] M. Okiyama Reitaku Institute of Political Economies and Social Studies, Reitaku University, Kashiwa, Chiba, Japan e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 J. Poot, M. Roskruge (eds.), Population Change and Impacts in Asia and the Pacific, New Frontiers in Regional Science: Asian Perspectives 30, https://doi.org/10.1007/978-981-10-0230-4_4

77

78

4.1

S. Tokunaga and M. Okiyama

Introduction

The purpose of this chapter is to measure the impacts of rapid population decline on the Japanese regional economy and clarify the types of policy measures needed to revitalize Japan’s economy in a depopulating society. We consider introducing broad-range industrial clusters with coagglomeration in the mega-region (Porter 1998, 2000). The Japanese economy has experienced low growth and deflation for more than 20 years. The Abe administration’s policy focus has been to exit deflation and revitalize the economy. The government has been implementing new economic policies known as “Abenomics” that aim at bold monetary policy, flexible fiscal policy, and a growth strategy that promotes private investment (Cabinet Office 2013a). The targets proposed in “Basic Policies for Economic and Fiscal Management and Reform: Ending Deflation and Revitalizing the Economy” (see Cabinet Office 2013a) include a nominal GDP growth rate of 3% and a real GDP growth rate of approximately 2% on average, for the new decade of revival (from fiscal 2013 through fiscal 2022). In addition, the “Japan Revitalization Strategy” (see Cabinet Office 2013b) entails the creation of three plans, namely the “Plan for the Revitalization of Japanese Industry,” “Strategic Market Creation Plan,” and “Strategy of Global Outreach” with clear key performance indicators (KPI) set for each policy goal. The plans contained a numerical target for key performance indicators. The “Japan Revitalization Strategy” was amended in 2014. Cabinet Office (2014) stated “However, it is not easy to put the Japanese economy back onto a full growth path at a time when Japan’s economy and society face the advent of a population fall through a declining birthrate and aging population.” Therefore, the revitalization strategy emphasizes that boosting productivity and strengthening earnings capacity is vital for the Japanese economy as a whole.1 A major premise for these economic policies is Japan’s depopulating society, caused by a declining birth rate and an aging population. Furthermore, the changing demographics—in which the core labor force population, i.e., those aged 20–64, is shrinking, while the population aged 65 and over is growing—are not uniform across Japan. Considerable inter-regional disparities exist. These population dynamics are troubling because they do not only restrain the Japanese economy and impede the revitalization of regional economies, but they could also lead to increased disparities among regions. Thus, the present study reveals what kind of policy is necessary in order to achieve the KPI targets. Next, assuming that the KPI targets are being achieved in the future, we forecast Japan’s medium- to long-term economic growth rate that is consistent with this by using a recursive dynamic six-region structural computable general equilibrium model, which we will refer to as the D6SCGE model. And we 1 Mitsubishi Research Institute (2014) and the Japan Center for Economic Research (2015) are forecasting a medium- and long-term Japanese economic outlook under a depopulation society.

4 Exploring Economic Futures for Japan Under Rapid Depopulation: A Dynamic. . .

79

will draw a picture of the future of regional economies in 2040, based on the depopulating society. This chapter is organized as follows. Section 4.2 provides the assumptions for building the scenario for the simulation. Section 4.3 explains the structure and features of a recursive dynamic six-region computable general equilibrium model. Section 4.4 discusses the scenario assumptions and the results of the simulations based on these scenarios. Finally, Sect. 4.5 presents our conclusions and discusses the policy implications.

4.2 4.2.1

Assumptions in Setting the Scenarios Assumptions of a Depopulating Society

The population forecasts frequently used in surveys and papers regarding the depopulating society can be found for example in “Population Projections for Japan (January 2012 projections),” published by the National Institute of Population and Social Security Research (2012). These projections use data from the 2000 Population Census of Japan and the Vital Statistics of Japan. The projections are based on three projection variants for the national population until 2060 that incorporate the trend of social mobility in regional populations. According to these projections, Japan’s population is expected to decrease by approximately an average annual 970,000 people during the 5-year period from 2035 to 2040. By then, the proportion of older persons (aged 65 and over) will then be approximately 39% of the total population, as shown in Fig. 4.1.

millions 140 120

29.5

34.0

100

36.1

36.6

36.8

37.4

80 60

75.6

70.9

67.8

65.6

62.8

40

59.1

38.7

53.9

20 22.9

21.8

20.1

18.5

17.0

15.6

14.7

2010

2015

2020

2025

2030

2035

2040

0 0-19 age

20-64 age

65 age and oiver

Source: Original data based on Population Projections for Japan (January 2012:IPSS)

Fig. 4.1 Trends in the youth, working-age, and elderly population in Japan

80

S. Tokunaga and M. Okiyama

The Institute of Population and Social Security (IPSS) released “Population Projections for Japan by Region” (see National Institute of Population and Social Security Research 2013), in which it projects the population of prefectures and municipalities by gender and age group until 2040, based on national mediumvariant fertility and mortality assumptions.2 Figure 4.2 illustrates the information, gathered for regional classifications, based on these regional population projections. As can be seen from the trends for the population aged 20–64 and the population aged 65 and above by region (indexed to 2010 ¼ 100), the former is declining in all regions and the decline becomes more pronounced year after year. This is particularly noticeable in Hokkaido and Tohoku by 2040, where the population aged 20–64 is projected to decline by 40% relative to 2010. Even in Kanto, the population of this group is projected to decline by approximately 30%. On the other hand, the population aged 65 and above is projected to rise in all regions, with regional disparities becoming pronounced after 2025. The high growth rate in this segment of the population is particularly pronounced in Okinawa, because the birth rate of Okinawa Prefecture remained relatively high for longer than in each province of the mainland. There is also only moderate growth in the population aged 65 and above in Kanto until 2030, which is then followed by a sharp increase. On the other hand, the population aged 65 and above in Tohoku is projected to peak around 2025 before declining, while the population aged 65 and above is projected to remain virtually flat in Hokkaido. This study uses the IPSS population projections for populations through to 2040 for each prefecture by gender and for five age groups spanning ages 20–64. In the D6SCGE model that incorporates the labor force as an exogenous variable, we estimate this variable in each period by taking the forecasted labor force participation rate for each of the five age groups and then multiply each rate by the corresponding projected population. This model assumes that there will not be full employment in the labor market in each region. Hence, unemployment becomes an endogenous variable which is determined from the changes in wages and prices in each region. Therefore, the projected labor force, that is the labor supply, is not assumed affected by endogenous unemployment.3 For the forecasted labor force participation rate for each of the five age groups, we refer to the labor force participation rates for each of the five age groups and for each gender through to 2030 by means of two scenarios, the “economic revival/ progressive labor participation scenario” and the “zero growth/unchanged labor participation scenario.” These populations can be found in the “Labor Supply and Demand Estimates” released by the Japan Institute for Labour Policy and Training 2

Kuwahata (2012) reported that regional and prefectural results of depopulation scenarios suggest that the population growth rate will even decline in major urban areas that have experienced relatively stable growth until now. 3 Giesecke and Madden (2013) argue that a common long-run labor market assumption in regional CGE modeling can allow for endogenous regional populations (constrained by a national population target), subject to some independent conditions about regional wages or income relativities.

4 Exploring Economic Futures for Japan Under Rapid Depopulation: A Dynamic. . . 2010 year = 100

81

20-64 years old

110.0 100.0 90.0

80.0 70.0 60.0 50.0 2010

2015

2020

2025

2030

2035

2040

Hokkaido

Tohoku

Kanto

Chubu,Kinki,Chugoku and Shikoku

Kyushu

Okinawa

65 year old and over

2010 year = 100

180.0

160.0

140.0

120.0

100.0

80.0 2010

2015

2020

2025

2030

2035

2040

Hokkaido

Tohoku

Kanto

Chubu,Kinki,Chugoku and Shikoku

Kyushu

Okinawa

Source: Original data based on Population Projections by regions (March 2013:IPSS)

Fig. 4.2 Trends in population aged 20–64 and 65-year old and above by regions

(JILPT 2014) in May 2014. The former scenario assumes that the economic and employment policies indicated in the previously mentioned “Japan Revitalization Strategy” and elsewhere are appropriately implemented. This is a scenario for progressing toward 2% real growth and labor force participation by the young, women, the elderly, and so on. In contrast, the latter scenario features close to zero

82

S. Tokunaga and M. Okiyama

economic growth and labor force participation rates by gender and age group that are at the same levels as in 2012. The difference between the two scenarios in the labor force of 2030 is that the number of males and females in the labor force is projected to be higher by 2.33 million and 3.69 million, respectively, in the former scenario than in the latter scenario. Our simulation scenarios correspond to these two labor force scenarios.

4.2.2

Additional Demand Generated by the Japan Revitalization Strategy

The aforementioned “Japan Revitalization Strategy” contains a diverse range of policies. This study considers the same approach as that of the JILPT and focuses on some KPIs in the form of policy targets that have been quantified. These are shown in Table 4.1 in the three action plans of the revitalization strategy. We review the JILPT estimates for 2020 and 2030 shown in Table 4.1 about the attainment level Table 4.1 Japan revitalization strategy

Target level in value for market development globally (unit: billion yen) Agricultural, fishery, foods, and beverage Manufacturing industry Tertiary industry

Staying healthy longer health and nursing care delivery and advanced medical technologies 2020 2030 0 0

Carrying out clean and economic energy demand and supply 2020 2030 0 0

0 11,000

13,860 4190

Target level in value for market development globally (unit: billion yen) Agricultural, fishery, foods, and beverage Manufacturing industry Tertiary industry Source: Original data based on Cabinet office

0 23,000

23,850 6150

Constructing the next-generation infrastructure 2020 2030 0 0

5,800 11,600 15,000 33,000 Taking Creating vitality for advantage of agricultural, forestry and potential fishery industries, and rural tourism communities resources Increasing Increasing Developing the number exports of the sixth of foreigners agricultural industry out of visiting products agriculture Japan 2020 2030 2020 2030 2020 2030 1000 5000 9000 16,000 20 40 0 0 0 0 120 260 0 0 0 0 920 2000

4 Exploring Economic Futures for Japan Under Rapid Depopulation: A Dynamic. . .

83

of additional demand (export and domestic demand) that would be generated in various industries in order to be successful in the revitalization strategy. Table 4.1 indicates an increase in production by agriculture-related industries of approximately 10 trillion yen by 2020 due to the growth in new demand for agriculture, forestry, and fishery-related products, both domestically and overseas, if policies relevant to agriculture-related industries, including food and beverages, can be effectively implemented. Furthermore, this demand increase is expected to double to 21 trillion yen by 2030. In addition, the increase in additional demand in the manufacturing sector from new markets for clean and economical energy and the construction of the next-generation infrastructure is expected to boost additional demand in manufacturing-related industries by 14 trillion yen by 2020 and by 24 trillion yen by 2030. Furthermore, in tertiary industries such as transportation and services, additional demandsare projected to increase by approximately 31 trillion yen by 2020 and then double to 64 trillion by 2030 due to increased demand in the health care, prevention, and lifestyle support industries, the transportation and information and communications sectors, and for services that deal with foreign visitors to Japan. We have incorporated the additional demand for each of the industries in our model.4 We now explain how to deal with the above additional demand in the D6SCGE model. We need to find the extent of increased domestic production in order to meet the additional demand. We do this by performing some simulations. However, these variables, i.e., additional demand and domestic production in this model, are endogenous variables. We adopt the method of varying two parameters so as to reach the level of additional demand when performing the simulation. The two parameters are a parameter that indicates productivity and a parameter of the production subsidies ratio in the industry in which additional demand is generated.

4.3 4.3.1

Structure of the Dynamic Six-Region CGE Model Structure of the SAM and 6SCGE Model

The database used for the D6SCGE model consists of an inter-regional social accounting matrix (SAM) for each of the six regions. The SAM encompasses the inter-regional input–output table of the competitive imports model for the six regions of Hokkaido, Tohoku, Kanto (the seven prefectures of the Kanto area, plus the four prefectures of Niigata, Nagano, Yamanashi, and Shizuoka), Chubu/Kinki/Chugoku/ 4 Besides this, there have been policy proposals for further strengthening the competitiveness of location, including specific measures targeted at achieving the establishment of “National Strategic Special Zones,” revitalization of the financial and capital markets, and the revitalization of regional economies and structural reforms. As it is difficult to reflect all these measures when setting the scenarios in this study, we assume these to have been incorporated in the additional demand for each of the aforementioned industries.

84

S. Tokunaga and M. Okiyama

Fig. 4.3 Map of the six regions of Japan used in this study Hokkaido

Okinawa prefecture

Tohoku region

Kanto region

Chubu/Kinki/ Chugoku/Shikoku Kyushu region

Shikoku, Kyushu, and Okinawa. Figure 4.3 shows the position of the six regions of Japan. The areas including Hokkaido, Tohoku, Kyushu, and Okinawa are defined as rural areas in this study, and Kanto and Chubu/Kinki/Chugoku/Shikoku are defined as urban areas. The data sources for the regional SAMs are the 2005 Ministry of Internal Affairs and Communications’ national input–output table, the 2005 Ministry of Economy, Trade and Industry’s nine inter-regional input–output tables along with its nine intraregional input–output tables of the competitive imports model, and the Cabinet Office’s fiscal 2005 Prefectural Accounts for 47 prefectures.5 We constructed the D6SCGE model by means of a six-region computable general equilibrium (6SCGE) model with the addition of a recursive dynamic dimension. The 6SCGE model comprises of 20 agents (1 household, 16 industries, 1 company, 1 regional government, and 1 investment bank) in each of the six regions, 16 commodity markets, and the two production factor markets for labor and capital. To this, we add two agents, i.e., the central government and the overseas sector. We assume that total factor endowments for labor and capital are exogenously fixed; and that both labor and capital cannot move outside the region, although both labor and

5 We used the Ministry of Economy, Trade and Industry’s nine-region database for the five regions of Hokkaido, Tohoku, Kanto, Kyushu, and Okinawa. However, we constructed the data for Chubu/ Kinki/Chugoku/Shikoku by deducting both the SAM for the above five regions and the export/ import sector between the above five regions from the national SAM.

4 Exploring Economic Futures for Japan Under Rapid Depopulation: A Dynamic. . .

Government Block Government revenue

Trade Block

Government savings

Government

Import supply

consumption

Export supply

Composite goods

Savings -Investment Block

Savings

85

Goods purchased by each block

Subsidy

Armington's composite goods

Investment Domstic trade between regions

Household Block

Demestic Production Block Indirect tax Domestic Output

Household savings

Household consumption

Composite intermediate goods Direct tax

Capital

Composite factor input

Household income

Labor,Capital

Labor,Capital Production Factor Market

Wage rate,Return to Capital

Fig. 4.4 Structure of the D6SCGE model

capital can move between industries within the region.6 In addition, we make unemployment an endogenous variable in the 6SCGE model. Specifically, we incorporated a Phillips curve into the 6SCGE model. The 6SCGE model is specified in seven blocks: (1) domestic production, (2) households, (3) savings and investment, (4) composite goods market, (5) production factor market, (6) trade, and (7) regional and central government, as in Fig. 4.4.7 We explain only the structure of the production block in the case of one region as an example, since we will conduct simulation of the improving productivity in the next section. First, the

6 As the Japanese labor market is characterized by low rates of inter-regional migration at the megaregion level (not the prefecture level), we assume in the D6SCGE model that total factor endowments for labor and capital are exogenously fixed, and that labor and capital cannot move outside the region. However, these assumptions may be relaxed in future versions of the model. 7 The equations and definitions of parameters and variables for other blocks in the 6SCGE model are in Okiyama et al. (2015) and Tokunaga et al. (2017). For the spatial CGE model, see Tokunaga et al. (2003), EcoModModeling School (2012), and Tokunaga and Okiyama (2014).

86

S. Tokunaga and M. Okiyama

domestic production block has a nested structure of production. Each
production sector a (a2A) in region o (o2S) is assumed to produce XDoa ¼ XDoc of a commodity c (c2C) and to maximize profit in a multi-level setting. At level 1, an industry in sector a, constrained by a Leontief technology, produces output by using each intermediate input good XCoca aggregated from 16 commodities. The added value is KLoa . Since the zero profit condition is assumed satisfied, the producer price PDoa of sector a in region o is derived from the equality of income and production costs. At level 2, the intermediate goods aggregated for the 16 commodities are derived from the Armington composite good XXdo ca imported from region d(d2R) under the constraint of constantreturns-to-scale CES-type technology, and the Armington composite good XXoo ca within region o. In addition, the price of the intermediate goods PXCocaaggregated from sector a of region o is derived by the definition of demand–supply balance for intermediate goods. In addition, even the added value portion is derived from the labor Loa and capital K oa from sector a of consumption site o under the constraint of constant-returns-to-scale CES-type technology in the same manner as the intermediate goods sector. Then, the producer price PDoa of sector a in region o results from the “zero profit condition.” Thus, this price can be derived from income being equal to production costs. Due to the mobility of capital and labor, the return to capital PKo, and the wage rate PLo for the consumption region o are equalized across all industries in region o. In addition, the price of the intermediate goods PXCoa aggregated from sector a in the region is derived by the assumption of demand–supply balance for intermediate goods.

4.3.2

Recursive Dynamic Dimension and Setting Parameters

We now explain the addition of a recursive dynamic dimension and the setting of the elasticity of substitution and other parameters of the model.8 The dynamic period in this study extends to 2040, as the IPSS’s population estimates by region are provided up to that year. We have taken 2010 as the base year (T ¼ 0) and incrementally added a recursive dynamic dimension to the model for 30 periods (T ¼ 30) until 2040.9 The procedure for adding a recursive dynamic dimension to the model is that it consists of calculating the initial value of the total capital stock KTo by dividing the steadystate growth rate (a 1.0% economic growth rate growth for each region) by the initial value of total investment demand for the six-region SAM ITo. 8 The method of adding a recursive dynamic dimension to the model is largely based on the GAMS code of Recursive Dynamics provided by the EcoModModeling School (2012) and Dixon and Rimmer (2002). We would like to note our appreciation for having been able to use this code. 9 Since we have constructed the six inter-regional SAM using the 2005 dataset as noted above, the CGE model would normally use 2005 as the benchmark equilibrium year. However, even when adding a recursive dynamic dimension to the model with 2005 as the initial period, it is not possible to accurately trace the period from 2005 to 2010 because the Global Financial Crisis occurred during that time. Therefore, while there are issues in replacing the benchmark year 2005 with 2010, we take 2010 to be the benchmark year in this study when adding a recursive dynamic dimension to the model.

4 Exploring Economic Futures for Japan Under Rapid Depopulation: A Dynamic. . .

87

We then use this total capital stock to derive the capital stock for each industry K ao from the percentage of capital in each industry according to the SAM. Multiplying the capital stock derived in this way for each industry by the steady-state growth rate growth results in the initial value of actual investment for each industry INVa0 . In addition, we derive the initial value of the return to capital for each industry PKa0 by dividing the initial value of the capital demand (capital services) for each industry according to the SAM by the capital stock for each industry calculated by the above process.10 An investment agent determines the total investment demand for each period IT by applying a certain percentage aIT of its own utility for each period UI. We calculate the investment INVat made by each industry in period t by multiplying the initial value of the actual investment for each industry INVa0 by the square root of the rate of the return to capital for each industry, PKat , and the average return to capital, PKAVGt.11 We then add capital stock K at1 in period t21 to the actual investment INVat in period t. By definition, this equals the capital stock in period t, K at . In addition, the inter-regional intermediaries are the aggregated goods, ICct , derived from the composite commodity according to the Armington assumption XIct imported and exported from both regions, the intra-regional composite commodity and the inter-regional current account balance that affects the savings St of both regions. The capital stock for each industry in each period derived this way is fixed in a similar manner as the labor supply LS and foreign savings SF for each period when multiplied by the steady-state growth rate growth. These procedures add a recursive dynamic dimension to the 6SCGE model. The simulations are conducted with the wage rates of all the six regions incorporated into the formula for the endogenous variables in the labor supply and demand equations.12 We estimated the parameters for each sector with a calibration method that used the six inter-regional SAM data with 2005 as the benchmark year. For the setting of parameters other than the parameters mentioned above, we used the values of GTAP7.1, specifically for the elasticity of substitution for labor and capital in the production block and the elasticity of substitution for the CES-type (Armington) function of the trade block. Then, in accordance with previous studies by Ban (2007), Hayashiyama et al. (2011), and others, we set the inter-regional elasticity of substitution for the production sector, the inter-regional elasticity of substitution for the household and investment blocks, and the elasticity of substitution of the CET-type function for the trade block. These values are summarized in Table 4.2. In addition, to setting the Phillips parameter for each region, we used time-series data from the past and used the figures shown in Table 4.3.

We refer to this as ‘payment to capital’ KPAY a0 in the dynamic model. The square root of the product of two returns means that the elasticity of the return to capital to investment demand is one. 12 We fixed the wage rate for one of the six regions as the numeraire and checked whether or not the labor market for the fixed region satisfied the equilibrium condition according to Walras’ law. However, Walras’ law did not hold true for that region, so we used this method as an alternative. 10 11

Production sector Crop cultivation Livestock Forestry Fisheries Foods Beverage Feeds and fertilizer Petroleum and coal products Textile products, Pulp, paper, and wooden products, chemical products, etc. Iron and steal, non-ferrous metals, and metal products General machinery, electrical machinery, and transportation equipment, etc Construct Electricity, gas, and heat-water supply Commerce Transport Financial and insurance, real estate, communication, Public administration, Education, and services, etc

Elasticity of transformation in CET function 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0

2.0 2.0

2.0 2.0 2.0 2.0 2.0

Elasticity of substitution between capital-labor in the CES function 0.6 0.6 0.6 0.6 1.2 1.2 1.2 1.3 1.2

1.3

1.3

1.4 1.3

1.3 1.7 1.3

Table 4.2 List of the elasticity of substitution assumptions in this model

1.9 1.9 1.9

1.9 2.8

3.2

3.5

Elasticity of substitution of ARMINGTON function 2.4 2.0 2.7 1.2 2.0 2.0 3.0 2.5 3.2

2.0 2.0 2.0

2.0 2.0

2.0

2.0

Elasticity of substitution between intermediate goods of different origin in the CES function 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0 2.0

0.5 0.5 0.5

0.5 0.5

0.5

0.5

Elasticity of substitution between final goods of different origin in the CES function 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5

88 S. Tokunaga and M. Okiyama

4 Exploring Economic Futures for Japan Under Rapid Depopulation: A Dynamic. . . Table 4.3 Phillips parameter Hokkaido Tohoku Kanto Chubu, Kinkr,Chugoku, and Shikoku Kyushu Oknawa

89

Phillips parameter 0.85 0.75 0.55 0.55 0.60 0.55

Table 4.4 Projection of population in the labor force by region Population in labor force (index of 2010 ¼ 100) Whole country Hokkaido Tohoku Kanto Chubu, Kinki, Chugoku, and Shikoku Kyushu Okinawa Population in labor force (index of 2010 ¼ 100) Whole country Hokkaido Tohoku Kanto Chubu, Kinki, Chugoku, and Shikoku Kyushu Okinawa

Base scenario (population decline) 2010 2015 2020 2025 2030 100.0 94.6 91.2 88.2 84.0 100.0 93.2 87.9 83.5 78.4 100.0 91.7 85.3 79.6 74.0 100.0 95.6 93.0 90.5 86.3 100.0 94.3 91.3 88.6 84.5 100.0 93.8 88.8 85.0 81.3 100.0 99.4 97.6 96.4 95.1 Scenario A and Scenario B 2010 2015 2020 2025 2030 100.0 96.0 95.1 94.1 91.2 100.0 94.5 91.3 88.4 84.0 100.0 93.0 88.9 84.8 80.0 100.0 97.0 96.9 96.6 93.9 100.0 95.8 95.4 94.8 92.1 100.0 95.1 92.3 90.0 87.3 100.0 100.3 100.2 100.4 99.9

2035 78.6 72.4 68.3 80.7 79.2 77.0 92.8

2040 72.1 65.1 61.9 74.0 72.7 71.4 88.9

2035 86.5 78.2 74.6 89.1 87.6 83.6 98.3

2040 80.1 70.5 67.9 82.5 81.1 78.1 94.6

Source: Original data based on IPSS population projections for Japan and labor supply-and-demand estimation (JILPT)

4.4 4.4.1

Simulations and the Results Setting the Simulation Scenarios

In this study, our base case scenario (Base Scenario) assumes a scenario based on the projected rate of decline in the labor force by region, and our alternative-case scenarios call for an increasing labor participation rate due to various policies associated with economic revitalization. For the latter, we assume two scenarios: Scenario A of the “Japan Revitalization Strategy” and Scenario B of broad-range industrial clusters with improving productivity in the manufacturing sector. Table 4.4 shows the rates of change in the labor force among the six regions for the Base Scenario and for the other two scenarios. The Base Scenario assumes that the labor force in 2040 declines to 72.1 (2010 ¼ 100), which declines to 65.1 for

90

S. Tokunaga and M. Okiyama

Table 4.5 Setting the change in subsidy provided to each industry in common to Scenarios A and B Ratio of subsidy provided to industry (annual % points of increase within each period of time) Hokkaido Tohoku Kanto Chubu, Kinki, Chugoku, and Shikoku Kyushu Okinawa

Crop cultivation and food Common

General machinery, electrical machinery and transportation equipment etc Scenario A Scenario B

Transport Common

T ¼ 1– 5 2.0% 2.0% 0.0% 0.0%

T ¼ 6– 10 3.0% 3.0% 0.0% 0.0%

T ¼ 1– 5 0.0% 0.0% 0.7% 0.7%

T ¼ 6– 10 0.0% 0.0% 0.7% 0.7%

T ¼ 1– 5 0.0% 0.4% 0.7% 0.7%

T ¼ 6– 10 0.0% 1.0% 0.7% 0.7%

T ¼ 1– 5 0.0% 0.0% 1.5% 1.5%

T ¼ 6– 10 0.0% 0.0% 1.5% 1.5%

2.0% 2.0%

3.0% 3.0%

0.0% 0.0%

0.0% 0.0%

0.4% 0.0%

1.0% 0.0%

0.0% 1.5%

0.0% 1.5%

Hokkaido, 61.9 for Tohoku, 74.0 for Kanto, 72.7 for Chubu/Kinki/Chugoku/Shikoku, 71.4 for Kyushu, and 88.9 for Okinawa. In contrast, the other two scenarios assume that the rates of decline will be approximately 5–8% less than in the Base Scenario. Let us now explain the two alternative-case scenarios that differ from the Base Scenario. Scenario A features increased production to meet the increased exports and domestic supply attributed to the “Japan Revitalization Strategy” mentioned above. The specific conditions for inserting this into the D6SCGE model are given in Tables 4.5 and 4.6. This scenario clearly incorporates targets of the “Japan Revitalization Strategy” such as increasing exports of agricultural, forestry, and fishery products and foods (i.e., sectors for generating income from regional resources) and increasing exports of manufactured goods and transportation services in the fields of energy and infrastructure. To this end, as shown in Table 4.5, we assumed that production subsidies would be raised for crop farming and the food industry in the regions of Hokkaido, Tohoku, and Kyushu for the former scenario, whereas we assumed that production subsidies would be raised for the manufacture of general machinery and durable goods and for the transport industry in the urban areas of Kanto and Chubu/Kinki/Chugoku/ Shikoku for the latter scenario. Second, we assumed an increase in industrial productivity to supply the goods and services needed to meet additional demand generated through the development of new markets, with the construction of the next-generation infrastructure and growth of the so-called “sixth industry,” the medical and healthcare sector, and the energy sector following that same strategy. Table 4.6 indicates the magnitude and approximate timing of productivity improvements by industry and regions. As was the case in Table 4.5, the productivity improvements with respect to the agriculture, forestry and fishery industries, and the food industry are concentrated in regional areas, whereas productivity improvements in the manufacturing and service industries are concentrated in urban areas.

4 Exploring Economic Futures for Japan Under Rapid Depopulation: A Dynamic. . .

91

Table 4.6 Setting the change in productivity in main industries in Scenario A and Scenario B Efficiency parameter in the fiirm’s production function (Annual rate of time period) Hokkaido Tohoku Kanto Chubu, Kinki, Chugoku, and Shikoku Kyushu Okinawa Efficiency parameter in the firm’s production function (Annual rate of time period) Hokkaido Tohoku Kanto Chubu, Kinki, Chugoku, and Shikoku Kyushu Okinawa Efficiency parameter in the firm’s production function (Annual rate of time period) Hokkaido

Crop cultivation, livestock and fisheries

Foods, beverage

Common

Common

T ¼ 1– 5 3.0% 3.0% 0.0% 0.0%

T ¼ 6– 10 5.0% 5.0% 0.0% 0.0%

T ¼ 11– 20 5.0% 5.0% 0.0% 0.0%

T ¼ 20– 30 5.0% 5.0% 0.0% 0.0%

T ¼ 1– 5 1.5% 1.0% 0.0% 0.0%

T ¼ 6– 10 2.5% 2.0% 0.0% 0.0%

T ¼ 11– 20 3.0% 2.5% 0.0% 0.0%

T ¼ 20– 30 3.5% 3.0% 0.0% 0.0%

3.0% 5.0% 5.0% 5.0% 1.0% 2.0% 2.5% 3.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% General machinery, electrical machinery, and transportation equipment, etc

Scenario A T ¼ 1– 5 0.0% 0.0% 0.05% 0.05%

T ¼ 6– 10 0.0% 0.0% 0.10% 0.10%

Scenario B T ¼ 11– 20 0.0% 0.0% 0.15% 0.15%

T ¼ 20– 30 0.0% 0.0% 0.20% 0.20%

T ¼ 1– 5 0.00% 0.50% 0.05% 0.05%

T ¼ 6– 10 0.00% 0.75% 0.10% 0.10%

T ¼ 11– 20 0.00% 2.00% 0.15% 0.15%

T ¼ 20– 30 0.00% 3.00% 0.20% 0.20%

0.0% 0.0% 0.0% 0.0% 0.50% 0.75% 2.00% 3.00% 0.0% 0.0% 0.0% 0.0% 0.00% 0.00% 0.00% 0.00% Tertiary industry excluding electricity, gas and heat-water supply, commerce and transport

Scenario A T ¼ 1– 5 0.0%

T ¼ 6– 10 0.0%

Scenario B T ¼ 11– 20 0.0%

T ¼ 20– 30 0.0%

T ¼ 1– 5 0.3%

T ¼ 6– 10 0.5%

T ¼ 11– 20 0.8%

T ¼ 20– 30 1.0% (continued)

92

S. Tokunaga and M. Okiyama

Table 4.6 (continued) Efficiency parameter in the firm’s production function (Annual rate of time period) Tohoku Kanto Chubu, Kinki, Chugoku, and Shikoku Kyushu Okinawa

Tertiary industry excluding electricity, gas and heat-water supply, commerce and transport

Scenario A

Scenario B

T ¼ 1– 5 0.0% 0.8% 0.8%

T ¼ 6– 10 0.0% 0.8% 0.8%

T ¼ 11– 20 0.0% 1.0% 1.0%

T ¼ 20– 30 0.0% 1.5% 1.5%

T ¼ 1– 5 0.3% 0.8% 0.8%

T ¼ 6– 10 0.5% 0.8% 0.8%

T ¼ 11– 20 0.8% 1.0% 1.0%

T ¼ 20– 30 1.0% 1.5% 1.5%

0.0% 0.5%

0.0% 1.0%

0.0% 1.5%

0.0% 1.5%

0.3% 0.5%

0.5% 1.0%

0.8% 1.5%

1.0% 1.5%

We now elaborate on the Scenario B of the development of broad-range industrial clusters with coagglomeration in the mega-regional areas of Tohoku and Kyushu, following Porter (2000).13 As shown in Table 4.6, we assume in this scenario improving productivity in the manufacturing sector in the regional areas of Tohoku and Kyushu, which will have the impact of improving productivity by an annual rate of 2% and 3% in the 2020s and the 2030s, respectively. Plans for automotive industrial clusters in such local areas as Kita-Kyushu, Miyagi, and Iwate serve as the basis for this assumption. In addition, this scenario assumes the development of broad-range industrial clusters following the fiscal support and corporate tax cuts by (local) governments according to Porter (2000). Regarding the specific settings applied to this scenario, we first assume that fiscal support will be offered in the form of production subsidies for the manufacturing sectors of general machinery, electrical machinery, and transportation equipment. We assume that production subsidies in Tohoku and Kyushu, as shown in Table 4.5, will be raised to 7% by 2020, the percentage identical to the one applied to the same industries in Kanto and Chubu/Kinki/Chugoku/Shikoku regions in Scenario A. Second, we assume that corporate tax cuts are designed to entice companies to Tohoku and Kyushu, as shown in Table 4.7. We have set the rates of reductions to be applied by 2020 at 30% for Tohoku and at 16% for Kyushu, based on the data obtained through the calibration method. This policy will leave the corporate tax rates for the both regions at near-equal levels, which is almost 3 percentage points lower than those for Kanto and Chubu/

13 For the economics of clusters and agglomeration, see Ellison and Glaeser (1997), Fujita et al. (1999, 2002, 2013), Duranton et al. (2010), and Tokunaga et al. (2012).

4 Exploring Economic Futures for Japan Under Rapid Depopulation: A Dynamic. . .

93

Table 4.7 Corporate tax by region

(Annual % points of increase or decrease within each period of time) Hokkaido Tohoku Kanto Chubu,Kinki,Chugoku, and Shikoku Kyushu Okinawa

Corporate tax Scenario B T ¼ 1–5 T ¼ 6–10 0.00% 0.00% 0.20% 0.20% 0.00% 0.00% 0.0% 0.0% 0.10% 0.10% 0.00% 0.00%

Kinki/Chugoku/Shikoku. We assume that these two policies will be funded by transferring part of the local allocation tax grants distributed by the central government to Kanto. While Scenario A assumes productivity improvements in tertiary industries located in Kanto, Chubu /Kinki/Chugoku/Shikoku, and Okinawa, Scenario B assumes productivity improvements in service industries located in Hokkaido, Tohoku, and Kyushu as well. The characteristics of the three scenarios are summarized in Table 4.8. In this table, each factor with a “o” is incorporated in the above Scenarios.

4.4.2

Simulation Results

4.4.2.1

Evaluation of Scenario A

Here, we will verify whether the simulation results under the conditions set for Scenario A, in which productivity improves due to higher production subsidies for particular industries by region, achieve the targets for these industries set in the “Japan Revitalization Strategy” mentioned in Sect. 2.2. Comparing the difference (deducting the Base Scenario from Scenario A) shown in Table 4.9 with the targets of the “Japan Revitalization Strategy” that are aggregated in Table 4.1, all the results exceed the target values, except the results for the agriculture and fishery industries and the food and beverage industries. The latter are below the target values. Therefore, the results of the simulation for Scenario A can be considered to generally reflect the results if the policies proposed by the “Japan Revitalization Strategy” are implemented. We consider the simulation results for the three scenarios next.

Region Base Scenario All regions Scenario A All regions Hokkaido Tohoku Kanto Scenario B Chubu, Kinki, Chugoku, and Shikoku Kyushu Okinawa

Increasing labor participation rate

Depopulating society

○ ○ ○ ○ ○ ○ ○

○ ○ ○ ○

○

○ ○

○

Projection based on JILPT

IPSS population projection

Table 4.8 Features of each scenario

○ ○

○

○ ○ ○ ○

○ ○

○

○ ○ ○ ○

Japan Revitalization Strategy Increase in Increase in production industrial subsidies productivity

○

○

○

○ ○

Broad-range industrial clusters Increase in Increase in production industrial subsidies productivity

○

○

Corporate tax

94 S. Tokunaga and M. Okiyama

The total amount of export and domestic supply (unit:billion yen) Agricultural, fishery, foods, and beverage Manufacturing industry Tertiary industry Total

Base Scenario (population decline) 2020 2030 48,689 49,851 189,038 188,540 544,013 542,261 781,740 780,652

Table 4.9 Simulation result of Scenario A compared with the Japan revitalization strategy

Scenario A 2020 2030 56,104 66,106 214,398 222,044 580,984 637,640 851,485 925,791

Difference (Scenario A–Base Scenario.P.D.) 2020 2030 7415 16,255 25,360 33,505 36,971 95,379 69,745 145,139

Target level of the Japan revitalization strategy 2020 2030 10,020 21,040 13,980 24,110 31,110 64,150 55,110 109,300

4 Exploring Economic Futures for Japan Under Rapid Depopulation: A Dynamic. . . 95

96

4.4.2.2

S. Tokunaga and M. Okiyama

Economic Effects of the Formation of Industrial Clusters

Evaluation of the Scenario in Terms of GRP Here, we report real GRP (Gross Regional Product) for the whole country and the six regions for each scenario, as shown in Table 4.10. First, in the Base Scenario, the national economic growth rate will be approximately zero by 2030 because of the decline in the labor force by region. After 2030, the Japanese economy will fall into a negative growth of about 0.4% per annum during the 10-year period from 2030 to 2040. In contrast, for Scenario A of the “Japan Revitalization Strategy,” we find that the growth rate is in the range of 0.6% per annum until 2020 and 0.8% in the 2020s, then slowing down to the 0.5% range in the 2030s. However, as the pushing effect of the real growth rate is weak and the interregional economic gap is wide in this Scenario A, we also consider the Scenario B of the development of broad-range industrial clusters with coagglomeration in Tohoku and Kyushu. For Scenario B, we find that the growth rate is in the range of 0.7% per annum until 2020 and 0.9% in the 2020s, slowing down to 0.7% in the 2030s. We thus find that Scenario B is effective in pushing up real economic growth rate. Second, we look at Fig. 4.5 to evaluate Scenario A and Scenario B by using the index of the percentage deviation in cumulative real GRP from Base Scenario. We find that in Scenario A the policy of the “Japan Revitalization Strategy” has contributed significantly to recovery economic growth rate for the urban areas of Kanto and Chubu/Kinki/Chugoku/Shikoku more than for the rural area of Hokkaido, Tohoku, and Kyushu. On the other hand, we also find that the transitions in percentage deviation of Scenario B in Hokkaido, Tohoku, and Kyushu have the same shape and size as the transitions in Kanto and Chubu/Kinki/Chugoku/Shikoku that approximately overlap on the transitions in Scenario A. Therefore, if the manufacturing sectors in the Tohoku and Kyushu areas will improve their productivity, the regional economic growth of rural areas in Tohoku and Kyushu is expected to be on a par with that of urban areas in Kanto and Chubu/Kinki/ Chugoku/Shikoku. In addition, although the decline in the labor force, as was shown in Table 4.4, will exert downward pressure on economic growth in the Tohoku region compared to other regions from 2030 to 2040, Tohoku will be able to maintain economic growth of 0.6%, as shown in Table 4.10. As evident, the simulation results show that besides the demographic change in the labor force, the inter-regional economic gap between urban and regional areas will spread unless there is the development of broad-range industrial clusters.

(Annual rate of 10 years) Base Scenario Scenario A Scenario B

(Annual rate of 10 years) Base Scenario Scenario A Scenario B

(Annual rate of 10 years) Base Scenario Scenario A Scenario B

Whole country Hokkaido 2010–2020 2020–2030 2030–2040 2010–2020 2020–2030 2030–2040 0.04% 0.02% 0.41% 0.27% 0.20% 0.57% 0.64% 0.80% 0.55% 0.22% 0.35% 0.23% 0.69% 0.91% 0.70% 0.44% 0.78% 0.33% Kanto Chubu/Kinki/Chugoku/Shikoku 2010–2020 2020–2030 2030–2040 2010–2020 2020–2030 2030–2040 0.06% 0.02% 0.42% 0.05% 0.01% 0.41% 0.81% 0.92% 0.72% 0.67% 0.86% 0.62% 0.82% 0.94% 0.75% 0.67% 0.87% 0.65% Okinawa 2010–2020 2020–2030 2030–2040 0.30% 0.29% 0.05% 1.11% 1.58% 1.28% 1.12% 1.60% 1.31%

Table 4.10 Simulation results on real GRP by scenario Tohoku 2010–2020 2020–2030 2030–2040 0.43% 0.35% 0.56% 0.01% 0.16% 0.24% 0.28% 0.77% 0.61% Kyushu 2010–2020 2020–2030 2030–2040 0.18% 0.04% 0.27% 0.20% 0.40% 0.06% 0.48% 0.99% 0.85%

4 Exploring Economic Futures for Japan Under Rapid Depopulation: A Dynamic. . . 97

98

S. Tokunaga and M. Okiyama

The whole country

% 16 14

14

12

12

10

10

Scenario A

8

6

4

4

2

2 2015

2020

2025

2030

2035

2040

Tohoku

% 16

0 2010

14

14 12

6

6

4

4

2

2

0 2010

0 2010

% 16

2020

2025

2030

2035

2040

Chubu/Kinki/Chougoku/Shikoku

14

14 12

6

6

4

4

2

2

0 2010

2015

2020

2015

2025

2030

2035

2040

2030

2035

2040

0 2010

2035

2040

2020

2025

2030

2035

2040

2030

2035

2040

Kyushu

Scenario A

8

Scenario B

2030

Scenario B

10

Scenario A

8

2025

Scenario A

% 16

12 10

2020

Kanto

8

Scenario B

2015

2015

10

Scenario A

8

Scenario B

% 16

12 10

Scenario A

8

Scenario B

6

0 2010

Hokkaido

% 16

Scenario B

2015

2020

2025

Okinawa

% 1.4 1.2 1.0 0.8

Scenario A

0.6

Scenario B

0.4 0.2 0.0 2010

2015

2020

2025

Fig. 4.5 Transition of the percentage deviation in cumulative real GRP results from Base Scenario

4 Exploring Economic Futures for Japan Under Rapid Depopulation: A Dynamic. . .

99

Evaluation of the Scenario in Terms of an Equivalent Variation Next, we examine the equivalent variation (EV) for the whole country and the six regions, as shown in Fig. 4.6a, b.14 All the figures in Fig. 4.6a show the cumulative equivalent valuation on the Base Scenario. All the figures in Fig. 4.6a show the transition of cumulative equivalent valuation on the Base Scenario. On the other hand, all the figures in Fig. 4.6b show the transition of the deviation in cumulative equivalent valuation from the Base Scenario on a comparative bar graph for each of the two scenarios (the graph of left-hand side is the Scenario A and right-hand side is the Scenario B). Each figure highlights the following. First, as for the whole country and the six regions in Fig. 4.6a, the trend of each cumulative EV is negative in the Base Scenario because of the deviation from the consumption budget for household on a steadystate growth path. That is, such a negative EV reflects the Japanese economy in a “depopulating society.” Second, for Scenario A and Scenario B in Fig. 4.6b, each deviation in cumulative EV as for the whole country and the six regions continues to increase because of the impact of the policies for the “Japan Revitalization Strategy.” As for Kanto and Chubu/Kinki/Chugoku/Shikoku and Okinawa, there is not much difference in cumulative EV between Scenario A and Scenario B because of the same settings. For Kanto and Chubu/Kinki/Chugoku/Shikoku, Scenario B exceeds Scenario A by 20.2 trillion yen and 9.8 trillion yen in 2040, respectively. On the other hand, as for Hokkaido, Tohoku, and Kyushu, Scenario B exceeds Scenario A by 14.3 trillion yen, 28.2 trillion yen, and 42.5 trillion yen in 2040, respectively. The aforementioned simulation results indicate that from the perspective of the economic welfare for the whole country that Scenario B exceeds Scenario A by 114.9 trillion yen in 2040, policies of the development of manufacturing industry in rural areas and the formation of broad-range industrial clusters with coagglomeration will enhance the impact of the policy of the “Japan Revitalization Strategy” and contribute to reduce regional disparities in the economic welfare for residents.

Evaluation of the Scenario in Terms of the Unemployment Rate Next, we examine the unemployment rate of the whole country and the six regions by means of Table 4.11. First, in the Base Scenario, the national unemployment rate is approximately 5% in the 2030s because of the decline in the labor force by region. For Scenario A of the “Japan Revitalization Strategy,” we find that the national unemployment rate is 4.5% in the 2020s, then slowing down to the 2.5% range in the 2030s. However, as the improving effect of the unemployment rate is weak for this 14

This study uses the equivalent variation as an index that measures the economic welfare. According to EcoModModeling School (2012), the equivalent variation is the difference between the consumption budget for the household of the ‘proposed change’ deflated by the price index that represents the change in prices induced by the ‘proposed change’ and the consumption budget of the ‘benchmark equilibrium’.

100

S. Tokunaga and M. Okiyama The whole country

Billion yen

0

-200,000

-10,000

-400,000

-20,000

Base Scenario

-600,000

-30,000

-800,000

-40,000

-1,000,000

-50,000

-1,200,000

-60,000

-1,400,000 2010

2015

2020

2025

2030

2035

2040

Tohoku

Billion yen

-70,000 2010

Base Scenario

2015

2020

2025

2030

2035

2040

2030

2035

2040

2030

2035

2040

Kanto

Billion yen

0

0

-20,000

-100,000

-40,000 -60,000

Hokkado

Billion yen

0

-200,000 Base Scenario -300,000 Base Scenario

-80,000

-400,000

-100,000

-500,000

-120,000 2010

2015

2020

2025

2030

2035

2040

Chubu/kinki/Chugoku/Shikoku

Billion yen

-600,000 2010

2015

2025

Kyushu

Billion yen

0

2020

0

-50,000

-20,000

-100,000 -40,000

-150,000 -200,000

Base Scenario

-60,000

-250,000

-80,000

Base Scenario

-300,000 -100,000

-350,000

-120,000

-400,000 -450,000 2010

2015

2020

2025

2030

2035

2040

2030

2035

2040

-140,000 2010

2015

2020

2025

Okinawa

Billion yen 0 -1,000 -2,000 -3,000

Base Scenario -4,000 -5,000 -6,000 2010

2015

2020

2025

Fig. 4.6 (a) Transition of the cumulative equivalent variation results from Base Scenario. (b) Transition of the deviation in cumulative equivalent variation results from Base Scenario

4 Exploring Economic Futures for Japan Under Rapid Depopulation: A Dynamic. . .

The whole country

Billion yen

40,000 35,000

1,200,000 1,000,000

Scenario A

30,000

Scenario B

25,000

800,000

20,000

600,000

Scenario A

15,000 400,000

2010

Scenario B

10,000

200,000 0

Hokkado

Billion yen

1,400,000

101

5,000 2015

2020

2025

2030

2035

2040

Tohoku

Billion yen

0

2010

2015

2025

2030

2035

2040

2030

2035

2040

2030

2035

2040

Kanto

Billion yen

70,000

2020

600,000

60,000

500,000

50,000 400,000

Scenario A

40,000

300,000

Scenario B

30,000

200,000

20,000

2010

Billion yen

2015

2020

2025

2030

2035

2040

Chubu/kinki/Chugoku/Shikoku

0

2010

100,000

400,000

90,000

2020

2025

Kyushu

80,000

350,000

70,000

300,000 250,000 200,000

Scenario A

60,000

Scenario B

50,000 40,000

150,000

30,000

100,000

Scenario A Scenario B

20,000

50,000

10,000

2010

2015

2020

2025

2030

2035

2040

Okinawa

Billion yen 10,000 9,000 8,000 7,000 6,000 5,000

Scenario A

4,000 3,000

Scenario B

2,000 1,000 0

2015

Billion yen

450,000

0

Scenario B

100,000

10,000 0

Scenario A

2010

2015

2020

Fig. 4.6 (continued)

2025

2030

2035

2040

0

2010

2015

2020

2025

102

S. Tokunaga and M. Okiyama

Table 4.11 Simulation results on unemployment rate by scenario (Annual rate of 10 years) Base Scenario Scenario A Scenario B

Whole country 2010– 2020– 2020 2030 6.6% 5.8% 6.1% 4.5% 6.0% 4.3%

(Annual rate of 10 years) Base Scenario Scenario A Scenario B

Kanto 2010– 2020 6.0% 5.4% 5.4%

(Annual rate of 10 years) Base Scenario Scenario A Scenario B

2020– 2030 5.3% 3.9% 3.9%

Hokkaido Tohoku 2010– 2020– 2030– 2010– 2020 2030 2040 2020 7.2% 6.3% 5.3% 7.4% 7.1% 6.0% 4.7% 7.3% 7.0% 5.6% 3.8% 7.1% Chubu/Kinki/Chugoku/ Shikoku Kyushu 2030– 2010– 2020– 2030– 2010– 2040 2020 2030 2040 2020 4.4% 6.6% 5.8% 4.8% 7.6% 1.9% 6.0% 4.3% 2.2% 7.5% 1.8% 6.0% 4.2% 2.1% 7.3% Okinawa 2010–2020 2020–2030 13.7% 12.8% 12.7% 9.6% 12.7% 9.5%

2030– 2040 4.8% 2.5% 2.2%

2020– 2030 6.2% 5.9% 5.2%

2030– 2040 5.0% 4.4% 3.0%

2020– 2030 6.5% 6.2% 5.3%

2030– 2040 5.4% 4.9% 3.0%

2030–2040 11.7% 6.4% 6.2%

Scenario A, we consider the Scenario B of the development of broad-range industrial clusters with coagglomeration in Tohoku and Kyushu. For Scenario B, we find that the national unemployment rate will be 4.3% in the 2020s, then slowing down to 2.2% in the 2030s. This tendency of impacts is the same in the six regions. Thus, we find that Scenario B is effective in improving the unemployment rate. Second, we find that in Scenario A and B, the policy has contributed to the improvement of the unemployment rate for the urban areas of Kanto and Chubu/Kinki/Chugoku/Shikoku more than for the rural area of Hokkaido, Tohoku, and Kyushu (see Table 4.11).

Economic Assessment of the Formation of Broad-Range Industrial Clusters In regards to the assessment of the formation of broad-range industrial clusters with coagglomeration in the regional areas of Tohoku and Kyushu and its impact on regional economies, we note the following three points based on our analysis of real GRP, as shown in Table 4.10, and equivalent variations, as shown in Fig. 4.6b, through our comparison with scenario A. First, our observation of real GRP for the regional areas of Tohoku and Kyushu shows that the formation of broad-range industrial clusters with coagglomeration will have a significant impact on economic growth over time. For the period of 2020–2030, the differences in percentage points from scenario A stand at 0.61% for Tohoku and 0.59% for Kyushu. For the period of 2030–2040, these differences widen to about 0.8% for the both regions.

4 Exploring Economic Futures for Japan Under Rapid Depopulation: A Dynamic. . .

103

Second, based on our observation of changes in real GRP, the regional areas of Kanto and Chubu/Kinki/Chugoku/Shikoku will suffer very little, or may even slightly benefit, from having part of their local allocation tax grants transferred to Tohoku and Kyushu. This is because Kanto will likely see an increased inflow of goods and services to Tohoku and Kyushu due to a surge in their respective economic activities. Such an increase will be large enough to offset the negative economic impact that Kanto will suffer from declines in their local allocation tax grants. Our observation also shows that changes in real GRP for the whole country in scenario B will exceed its counterparts in scenario A by 0.05–0.15% points. Third, our observation of the deviation in cumulative equivalent variation for Tohoku and Kyushu—observed in scenario B and shown in Fig. 4.6b—shows that the formation of broad-range industrial clusters will make a substantial contribution to the expansion trend in the two regions’ equivalent variations. This is because of the impact of the policy for the development of manufacturing industry and the formation of broad-range industrial clusters with coagglomeration. In addition, Kanto will get the deviation in cumulative equivalent variation more than that in scenario A, though this region will bear the burden of creating broad-range industrial clusters.

4.5

Conclusion and Policy Implications

In this study, we have measured the impacts of rapid population decline on the Japanese regional economy and clarified the types of policy measures needed to revitalize Japan’s economy in a depopulating society by introducing broad-range industrial clusters with coagglomeration in the mega-region (Porter 1998, 2000). This study has used a recursive dynamic regional CGE model. The following three points were highlighted through the simulations. First, unless the productivity of production activities in a “depopulating society” improves, Japan’s economic growth will be virtually zero initially and decline to a negative growth of 0.4% by 2030. However, economic growth could reach a level of approximately 0.7–0.8% if policies are taken to increase productivity and raise production subsidies to create the production volume required to meet the additional domestic and overseas demand generated by the Abe administration’s “Japan Revitalization Strategy.” However, such economic growth will be supported by growth in urban areas and it will increase the disparities in growth among regions. Consequently, the simulation results indicate that incorporating the impact of industrial clusters on regional areas yields growth that is on a par with that of urban areas. Considering this point, some form of preferential policies to support industry (e.g., policies for varying corporate taxes by region), such as in the industrial clusters focused on the manufacturing sector (as is currently evident in automotive industrial clusters in the Tohoku and Kyushu areas), would revitalize economies of regional areas and ultimately correct inter-regional economic disparities.

104

S. Tokunaga and M. Okiyama

Second, the construction of wide-ranging industrial clusters—those which go beyond the current automobile clusters in the regional areas of Tohoku and Kyushu—to incorporate a wider range of manufacturing sectors, such as general machinery and electrical machinery, will be effective in preventing the exacerbation of inter-regional economic welfare disparities. The simulations conducted in this analysis suggest that the construction of widerange industrial clusters with coagglomeration will invigorate regional economies and will lead to narrower inter-regional economic disparities. Funds will be secured by changing the allocation ratio of local allocation tax grants in favor of regional areas rather than urban areas. In order to resolve inter-regional economic disparities, it is necessary to build a new infrastructure by utilizing those funds to provide production subsidies for the member industries of broad-range industrial clusters with coagglomeration, as well as to offer corporate tax cuts to lure industries to regional areas. Finally, the trend of overconcentration in the Kanto region is inevitable in the global economy, and this is occurring at the same time that Japan is rapidly becoming a “depopulating society.” From the economic welfare perspective, such circumstances require a constant correction of inter-regional economic and social disparities. Simulations that can clearly identify effective and efficient economic policies to deal with this will, therefore, become increasingly important.

References Ban K (2007) Development of a multiregional dynamic applied general equilibrium model for the Japanese economy - Regional economic analysis based on a forward-looking perspective (in Japanese). RIETI Discussion Paper Series Cabinet Office (2013a) Basic policies for the economic and fiscal management and reform–ending deflation; and revitalization of the economy. http://www5.cao.go.jp/keizai1/2013/20130614_ 2013_basicpolicies_e.pdf Cabinet Office (2013b) Basic framework for fiscal consolidation: medium-term fiscal Plan. http:// www5.cao.go.jp/keizai1/2013/20130808_medium_term.pdf Cabinet Office (2014) Japan revitalization strategy as revised in 2014 (in Japanese). http://www5. cao.go.jp/keizai-shimon/kaigi/minutes/2014/0624/shiryo_02_1.pdf Dixon PB, Rimmer MT (2002) Dynamic general equilibrium modelling for forecasting and policy. North-Holland, Amsterdam Duranton G, Martin P, Mayer T, Mayneris F (2010) The economics of clusters. Oxford University Press, Oxford EcoMod Modeling School (2012) Advanced techniques in CGE modeling with GAMS. Global Economic Modeling Network Ellison G, Glaeser EL (1997) Geographic concentration in U.S. manufacturing industries: a dartboard approach. J Polit Econ 105(5):898–927 Fujita M, Thisse JF (2002, 2013) Economics of agglomeration: cities, industrial location, and regional Growth. Cambridge University Press, New York, NY Fujita M, Krugman P, Venables AJ (1999) The spatial economy: cities, regions and international trade. MIT Press, Cambridge, MA Giesecke JA, Madden JR (2013) Regional CGE modelling. In: Dixon P, Jorgenson D (eds) Handbook of computable general equilibrium modelling. North-Holland, Amsterdam

4 Exploring Economic Futures for Japan Under Rapid Depopulation: A Dynamic. . .

105

Hayashiyama Y, Abe M, Muto S (2011) Evaluation of GHG discharge reduction policy by 47 Prefectures multi-regional CGE. J Appl Reg Sci 16:67–91 Japan Center for Economic Research (2015) The Japanese economy’s challenge in achieving fiscal reconstruction and the transition of demographic trends: consideration of future generationsupporting social security reforms (in Japanese). In: 41st Medium-Term Forecast for the Japanese Economy 2014-2025. http://www.jcer.or.jp/eng/economic/medium.html JILPT(2014) Labor supply and demand estimates—policy simulations based on the labor supply and demand model (2013) (in Japanese), Research Material Series, 129 Kuwahata S (2012) Impact of progress of low birthrate and aging on regional economy growth (in Japanese). Gerontology Journal, NLI Research Institute Mitsubishi Research Institute (2014) Medium-term forecast for domestic and overseas economy: FY2014-2030 (in Japanese), PRESS RELEASE. http://www.mri.co.jp/news/press/teigen/ 015411.html National Institute of Population and Social Security Research (IPSS) (2012) Population projections for Japan: 2011 to 2060 (in Japanese). http://www.ipss.go.jp/site-ad/index_english/esuikei/ gh2401e.asp National Institute of Population and Social Security Research in Japan (IPSS) (2013) Regional Population Projections for Japan: 2010-2040 (in Japanese). Population Research Series. http:// www.ipss.go.jp/pp-shicyoson/j/shicyoson13/6houkoku/houkoku.pdf Okiyama M, Tokunaga S, Ikegawa M (2015) An impact analysis of fiscal measures to promote selfsustained economic development in okinawa using a multi-regional CGE model (in Japanese). RIETI Discussion Paper Series Porter M (1998) On competition. Harvard Business School Press, Cambridge, Mass Porter M (2000) Location, competition, and economic development: local clusters in a global economy. Econ Dev Q 14(1):15–34 Tokunaga S, Okiyama M (2014) Reconstruction the disaster-affected region of the Great East Japan earthquake and recovery of the regional economy (in Japanese). Bunshindou, Tokyo Tokunaga S, Resosudarmo BP, Wuryanto LE, Dung NT (2003) An inter-regional CGE model to assess the impacts of tariff reduction and fiscal decentralization on regional economy. Stud Reg Sci 33(2):1–25 Tokunaga S, Kageyama M, Akune Y, Nakamura R (2012) Empirical analysis of agglomeration economies in Japanese assembly-type manufacturing industry for 1985-2000. Rev Urban Reg Dev Stud 26(1):57–79 Tokunaga S, Okiyama M, Ikegawa M (2017) Impact of climate change on regional economies through fluctuations in Japan’s rice production: using dynamic panel data and spatial CGE models. In: Shibusawa H et al (eds) Socioeconomic environmental policies and evaluations in regional science. New frontiers in regional science: Asian perspective, vol 24. Springer Science +Business Media, Singapore

Chapter 5

Using Spatial Microsimulation to Derive a Base File for a Spatial Decision Support System Robert Tanton and Yogi Vidyattama

Abstract This chapter describes how a spatial microsimulation method could be used as input in a Spatial Decision Support System (SDSS) to provide simulations for planners and communities, showing what their community might look like given different economic, social and demographic parameters and constraints. We find that spatial microsimulation models can be used to create the synthetic person-level base file that can then be used by the SDSS to project potential future scenarios. The advantage of using spatial microsimulation to create the base file is that it can include other variables through imputation or statistical matching techniques based on the individual-level data. The SDSS with a synthetic population from a spatial microsimulation model can implement a number of complex boundaries, including physical boundaries, environmental boundaries and economic boundaries. Further, indicators like educational attainment, income and household type can also be projected, allowing for complex simulations of potential futures for a city under different scenarios. Keywords Microsimulation · SDSS · Population projections · Scenarios

5.1

Introduction

Demographic forecasts are important for planners in deciding what buildings and infrastructure are required in a city. Planners have long benefited from age/sex projections to inform the planning process. Cities where high growth is predicted will need new housing developments, more shops, roads and public transport to cater for the increased population. As more people in the world move to cities for work, the planning problems are getting more complex. Different priorities R. Tanton (*) · Y. Vidyattama National Centre for Social and Economic Modelling, Institute for Governance and Policy Analysis, University of Canberra, Canberra, ACT, Australia e-mail: [email protected]; [email protected] © Springer Nature Singapore Pte Ltd. 2020 J. Poot, M. Roskruge (eds.), Population Change and Impacts in Asia and the Pacific, New Frontiers in Regional Science: Asian Perspectives 30, https://doi.org/10.1007/978-981-10-0230-4_5

107

108

R. Tanton and Y. Vidyattama

(environmental, economic and social) need to be managed simultaneously by planners and decision-makers. Demographic forecasts are traditionally developed by using what is called the cohort component method. With this method, births, deaths and migration rates are projected for the city, usually using past trends and these are combined with an ageing process to develop projections (Booth 2006). Small area forecasts of population, produced by the Australian Bureau of Statistics (ABS) for the Commonwealth and published by the Australian Institute of Health and Welfare (AIHW 2016), use a cohort component method and information from the planning areas of each State. Recent developments in this type of modelling have attempted ‘smarter’, i.e. more refined, methods in projecting births, deaths and migration (Booth 2006). An assessment of cohort component methods is provided by Wilson (2016) and a review of population projection methods is provided by Wilson (2015). Wilson also refers to microsimulation models in his 2015 paper, but does not assess these. While demographic projections are essential for planners and the cohort component method (with developments in estimating births, deaths and net migration) is an obvious method to generate age–sex population projections, the considerations of most city planners now go beyond simple population projections. Infrastructure, transport, environmental sustainability and ‘liveability’ (quality of life—it includes the built and natural environments, economic prosperity, social stability and equity, educational opportunity and cultural, entertainment and recreation possibilities) are all considerations that modern city planners need to take account of in making planning decisions. There are also concerns over unlimited population growth in an area that may be constrained politically or physically—for example, an area surrounded by mountains, oceans or state boundaries. While the obvious solution to going out is going up and building denser areas, the original Limits to Growth modelling also identified constraints in terms of food availability internationally (Meadows et al. 1972). While recent critics of this approach suggest that the world has changed to adapt to a larger population (Bloom 2011; Dorling 2013), this is an international perspective, rather than a local perspective. Recent research in Australia has recognized that there are limits to how many people can live in an area (Lowe 2012) due to a wider range of factors than just food sustainability—hence this research incorporates factors such as liveability and environmental sustainability. In many ways, the questions may be ‘what will the future look like, and how will we plan for this future’? rather than ‘can we fit this number of people in’? As noted above, Wilson (2015) included microsimulation models in his paper on population projections. Static spatial microsimulation has been used for population projections, as described in Vidyattama and Tanton (2010). Van Imhoff and Post (1998) outline a dynamic microsimulation approach to population forecasting (with births, deaths and migration endogenous). Dynamic models implicitly include models of births, deaths and migration. This is a more advanced approach than the static approach used by Vidyattama and Tanton (2010), which used cohort component population projections with labour force projections to align their projections

5 Using Spatial Microsimulation to Derive a Base File for a Spatial Decision. . .

109

to. Work in Canada using a dynamic microsimulation model has developed population projections for small areas (see Marois and Bélanger 2014). While population projections are important for planners, it has long been recognized that planning relies on much more than just population projections. Geographic Information Systems (GIS) have been around since the early 1960s to help planners, but most Geographical Information Systems do not have the analytical and statistical systems required to help planners assist the decision-makers. Decision support systems, which have evolved out of Management Information Systems designed in the 1960s, do have these analytical and statistical systems but are not spatial. It has only been in more recent years that these two fields have been brought together to form the area of spatial decision support systems (Densham and Armstrong 1994). The main role of a Spatial Decision Support System (SDSS) is to provide data to support the decision-making process and allow the decision-makers to resolve semistructured or ill-structured spatial decision problems. The solutions to these types of problems are usually so complex that computer programs and algorithms can help by informing the solution, but the final solution is generally not given by the computer program (Chakroun and Benie 2005). The SDSS allows planners to derive different scenarios, which can then be presented to decision-makers or local communities for assessment and a decision. So, the SDSS provides suggestions on what is required to service the growing population of an area in terms of transport routes, infrastructure required, environmental sustainability, etc., but the planners take the final decisions in consultation with the affected communities. One of the methods for creating an SDSS is to use a spatial microsimulation model (Birkin and Clarke 1988; Birkin et al. 2009). The spatial microsimulation models used for the SDSS usually define a synthetic population for a small area. This resolves one of the key limitations of small-area SDSS, which usually requires extensive small-area data that are often simply not available. The advantage of using a synthetic population for an SDSS is that the synthetic population for each small area can be projected by age, sex, family type, etc., using a dynamic model. The contribution of this chapter is to suggest a way to extend this spatial microsimulation SDSS by incorporating economic and environmental constraints imposed by other models and also show how other indicators can be incorporated into the spatial microsimulation base file by means of imputation and data-matching techniques. This then provides a very powerful tool for simulating different scenarios, as the base file includes a number of not previously available social, economic and demographic data as well as linked models to provide further data for the base file. These other models can inform transport planning, infrastructure, environmental considerations, etc. This is a large and challenging piece of work and there is currently no application available to demonstrate how the new indicators could be imputed and how these models could be linked. However, a number of linked models and imputed indicators, outlined in this chapter, indicate that the methods have now developed sufficiently to bring them together to form a complex SDSS. A case study of an Australian territory where this type of model could be useful, given the population

110

R. Tanton and Y. Vidyattama

projections and the physical and political constraints around the territory, is also given in this chapter. The Australian Bureau of Statistics has projected (using a mid-range estimate of births, deaths and migration—called ‘Series B’), that the population of the Australian Capital Territory (the ACT) will increase from the level in 2014 of 390,000 to a population of 682,000 in 40 years’ time (2054) (Australian Bureau of Statistics 2013a). The ACT is a territory entirely contained within the area of the state of New South Wales. This means that space is at a premium and fitting these people in will require increased population density in many areas. It will also have implications for the water supply, employment, infrastructure and transport networks both in the territory itself and in the wider region. These are all factors that need to be considered in planning decisions. This chapter aims to show how spatial microsimulation could potentially be used to provide the basis for an SDSS and to build the initial framework for a case study of the ACT. While Spatial Decision Support Systems have been developed using spatial microsimulation overseas (Birkin et al. 2009; Wu and Birkin 2013), these have not incorporated other models to implement constraints to the population. This chapter is the first that suggests the idea that spatial microsimulation, together with other imputed variables and other linked models, can jointly provide the population constraints as well as potentially other regional indicators. This chapter is organized in a number of sections. The next section reviews the literature on SDSS and shows how spatial microsimulation can be used in an SDSS. Section 5.3 introduces the specific issues of planning in the ACT. The fourth section shows how the different modules of an SDSS in the ACT might fit together and the final section highlights the data issues and the importance of collaboration with the local government.

5.2

Spatial Decision Support Systems

More than a decade ago, Chakroun and Benie (2005) outlined a number of components that would need to be considered in designing an SDSS. It is useful to consider these components individually when designing the SDSS. The first is a database management system (DBMS), which has functions for spatial information. There are a number of database systems that have spatial extensions, including Microsoft SQLServer and PostgreSQL, which has the spatial extension PostGIS. The second component is the model base management system (MBMS), which contains the functions to manage the models. The models can be embedded in the DBMS or may just receive and pass data through to the DBMS. Thirdly, the dialog generation and management system (DGMS) manages the interface between the user and the rest of the system. It is the graphical interface between the models (the MBMS) and the data (the DBMS). A diagram showing these components and how they relate to each other is shown in Fig. 5.1.

5 Using Spatial Microsimulation to Derive a Base File for a Spatial Decision. . . Fig. 5.1 Diagram of SDSS components

111

DBMS

MBMS

DGMS

User

The DBMS is usually a fairly simple choice of which spatially enabled database to use and this will usually come down to the scale of databases and cost. The MBMS can be as simple or as complex as required, depending on the questions that the user wants answered. If the user seeks answers to questions about what buildings to build on particular pieces of land, then the user may want to incorporate models of site preparation costs (levelling sloping land, etc.) or risk of natural disasters (e.g. if the land is on a fault line or in an earthquake- or fire-prone area). Alternatively, if the question is about how many people could potentially live in a certain area, then models of transport and road use may be required. The MBMS may incorporate spatial information from spatial databases including remote sensing data and GPS information, but may also incorporate non-spatial information like environmental reports and impact statements. This qualitative information can be turned into quantitative scores through either utility functions or scores entered by the user into the DGMS. One of the models in the MBMS could potentially be a spatial microsimulation model. Both SDSS and spatial microsimulation have been around since the 1990s and some of the early spatial microsimulation models were developed to assist urban planning (Birkin and Clarke 1988). However, it is really only very recently that spatial microsimulation has started to make an impact on SDSS, as spatial researchers have started to recognize the power of bringing these two methods together. One initiative in the UK is a collaboration between the Centre for Applied Spatial Analysis at the University College London and the University of Leeds for developing a model called Talisman. This model emphasizes the potential flows between different locations with methods based on spatial interaction, agent-based models (ABM), cellular automata (CA) and microsimulation (ESRC 2016). The Talisman model is one large model. Another approach in designing an SDSS is to bring together a number of different, already developed models rather than

112

R. Tanton and Y. Vidyattama

designing one complex model that attempts to answer all questions. This is probably the easiest way to bring a spatial microsimulation model into an SDSS. There are a number of advantages to this approach: The existing models are tested and working in their fields; New models do not have to be developed, thereby saving time; The models can use global best practice approaches for their field; When new methods are developed in particular areas, changes can easily be incorporated into the overall model; and 5. Each model can be written in a different computer language.

1. 2. 3. 4.

This approach has worked successfully in some recent research that brought together Computable General Equilibrium (CGE) and microsimulation models (see Rao et al. 2013). The disadvantage of this approach is that feedback and interactions are difficult to incorporate. An iterative approach can be used whereby models are re-run and the output from one model is used as an input for another, but this can be complicated. In conclusion, this section has identified the components of an SDSS and then argued that spatial microsimulation could be part of the MBMS, along with other models. There is then no requirement to have one large model in the MBMS, as long as each model can be linked to the others in some way.

5.3

Planning in a Region

This section introduces the issue of planning for population growth in a region, with physical and political borders on a number of sides (as many areas have). We also consider how an SDSS may assist in planning in any region. Our case study is the Australian Capital Territory (ACT), which is the region that contains Australia’s capital Canberra. While this chapter considers an SDSS for Canberra as a case study, the methods and systems described here can be used for town planning in any city or town that has physical, political, social and economic boundaries. The ACT is a small self-governing territory in Australia within the much larger state of New South Wales. The ACT has one city in it, Canberra and is bordered by mountains to the south and west. The ACT has an area of 2400 km2, but over half of this consists of national parks. Due to the political borders around the ACT, Canberra has a very restricted footprint for population growth. Because it is the national capital, many areas are owned and managed by the Commonwealth Government of Australia, rather than the ACT Government, so cannot be built on. Building on hills and ridges in the ACT is also banned, which again limits the land available for building. There are also mountains to the south and west and another city (Queanbeyan) to the North-East. Further, there is community resistance to high-rise buildings. There are therefore limited areas for new developments in Canberra through physical limits—mainly areas to the North (which have already reached a political ACT-NSW border) and

5 Using Spatial Microsimulation to Derive a Base File for a Spatial Decision. . .

113

the South-East and East (which have hit a political ACT-NSW border or a border with a national park). Areas to the West are being developed but are approaching national park which cannot be built on. Community resistance and less demand for high-density units have also affected any infill. Despite the physical limitations, the ACT has released a plan for development as part of its planning processes (ACT Government 2014). The plan includes some new areas for residential building, mainly the western edge and Eastern broadacre areas, plus some urban infill in the town centres (mainly in the City, but also in the town centres of Belconnen, Gungahlin, Woden and Tuggeranong). However, the serious question remains of where Canberra will fit another 292,000 people in by 2054, along with additional industrial areas to provide jobs, transport corridors for roads and retail space for shops. There are a number of other issues that the Government will need to think about apart from just where to fit the people, including water, jobs, transport, liveability and recreation areas. All these factors need to be considered in the assessment of how many houses can fit in the new areas, whether plot sizes need to be smaller to fit more houses in and what minimum level of population density is required to fit in the extra people. This is all information that could be provided by means of an SDSS. In this section, we have identified the important planning issues in the case study of the ACT in Australia. The main issue we have identified is that there are a number of physical constraints in a city that has political and physical borders, as well as environmental, economic and social constraints. All these constraints need to be considered in developing the potential future population scenarios.

5.4

Designing an SDSS with a Spatial Microsimulation Model

As outlined in the first section, spatial microsimulation has been used previously for SDSS, but spatial microsimulation models have evolved extensively since this early work (see Tanton and Edwards 2013, for a review of recent developments in spatial microsimulation models). This section considers how these developments in spatial microsimulation methods could be incorporated into an SDSS. We also review the potential problems and advantages of using a spatial microsimulation model in an SDSS. In thinking of how a system of models could be used for an SDSS, a dynamic spatial microsimulation model is particularly suitable for demographic projections, through modelling a synthetic population for each small area. Other models can then be incorporated to account for economic, environmental and social constraints, or to add other indicators needed in the planning process. The other models would use the synthetic population for each small area derived by the spatial microsimulation model. The synthetic population would provide the advantage of not having to acquire data for everyone in the area, which is usually not available.

114

R. Tanton and Y. Vidyattama

Infrastructure

Environment

Economic Income, House Prices)

Land Use

Employment (Industry)

Demographics (Spatial Microsimulation Base FIle)

Livability

Fig. 5.2 System of models for an SDSS

The starting point for the spatial microsimulation model is a database drawn from a household survey. The data would include variables like family type and housing expenditure and for all family members, variables such as education level, income and income source. All these variables have already been incorporated in a spatial microsimulation model, as they can be derived from the Australian Census and from survey data (see Tanton et al. 2011). The resulting base data set has already more information than the data for a traditional cohort component model, which consists of just age and sex assigned to each person. Additional information can be added to the data set for spatial microsimulation by means of imputation or data matching. This spatial microsimulation approach can also be used for generating population projections, as outlined in the introductory section. For our purposes, these projections are superior to the simple age/sex projections from a cohort component model for two reasons. The first reason is that they have many other variables assigned to individuals apart from just age and sex. The second reason is that the projections can be constrained, using either other social and economic models (as shown below) or using some external information. The spatial microsimulation can also produce projections based on the type and amount of housing planned to be built. In particular, it can produce projections that contain the characteristics of people that could come and fill the housing, including the household composition and the occupations of the household members. One proposed system that brings together all the areas for an SDSS identified above is shown in Fig. 5.2. It can be seen that the synthetic demographic data set that includes a number of new indicators, derived by imputation and matching methods, is central to this model. All the other modules can use these new indicators to derive better projections, or can even use the results from another model in the overall framework. For example, the labour force model may use results from an education model. This system also recognizes that the estimates from the other models may

5 Using Spatial Microsimulation to Derive a Base File for a Spatial Decision. . .

115

affect the demographic variables in the synthetic demographic data set, so there is a two-way relationship between all the models. This can be achieved through an iterative system—so, for each year, the system runs until a stable population is achieved given the set of parameters (regarding employment, infrastructure, etc.) and the next year is then started with the dynamic spatial microsimulation module, which would age people by 1 year and then make demographic changes using models of marriage, family formation, births, deaths, migration, education, etc. The process starts with a synthetic population for each small area in the SDSS. This synthetic population can assign people and families to different houses in the area by means of a dynamic spatial microsimulation model. This base data set can then be updated each year with the simulated births, deaths, migration, marriages, education changes, etc. (the dynamic component of the model). The approach normally used for dynamic microsimulation is a Monte Carlo approach, which assigns a certain percentage of people of a certain age and, for example a particular educational qualification, based on probabilities from an external source. Other indicators can be imputed by means of other methods (for example, statistical matching—see Namazi-Rad et al. 2017). If required, a limit to the population growth in an area could be imposed by means of land use information, but assigning families to households also places a constraint on the synthetic population, because the number of households must be assumed to be equal to the assumed number of houses in the area. If there are more people in Canberra generated by the model than the houses can support, then scenarios using higher-density housing can be considered. Alternatively, it can be assumed that the borders move into areas of NSW. As outlined at the beginning of this chapter, this model is for simulation, to inform decision-makers and to identify different options. The model does not make a value judgement as to whether the results are the best option or not. The latter is for planners and politicians to decide. Households can also be split when young people reach a certain age and will leave the household. Some of these young people will migrate out of the ACT and some will stay in the city and start families. These proportions could be informed by using a longitudinal data set like the Australian Household, Income and Labour Dynamics in Australia (HILDA) data set, a household-based panel survey that collects information about economic and subjective well-being, labour market dynamics and family dynamics in Australia (Watson and Wooden 2001) or the ABS longitudinal census data set (ABS 2013b). The land use data will inform this household selection process. When more land is released for residential use, or land is re-zoned to allow multiple stories, these houses or units can then be filled with more people from the base file, allowing greater population growth. As described in Norman et al. (2012), the ACT government already has a system that allows them to update this information every 24 h according to the latest planning decision in the territory. Economic data can be extracted from a tax/transfer microsimulation model called STINMOD which is available for use in this system. The model can provide updated income information using the latest tax and public transfer rules. This would be part of the initial information created for each household in the small area. This is an

116

R. Tanton and Y. Vidyattama

example of where model linking can provide additional data not normally available to the base data set. In future, given that microsimulation and CGE models have already been linked, some of the economic variables can also be estimated by means of a CGE model. Additional information from other sources can also be used to assist the simulation. For example, house prices in the area and household incomes can be used to assign specific households to an area. Incomes would only be used to help inform how much new inward migration there might be in an area. For example, this can be done by not assigning a household earning a low income into a very high-houseprice neighbourhood. This kind of assignment can only be very general because, while we know a household’s income, it does not take into account the wealth of the household. A high level of wealth could make it possible for a low-income family to be living in a very expensive house. Estimates of the number of motor vehicles in the area, based on assumptions about the number of people of driving age, the availability of public transport in the area and the types of motor vehicles driven by different age groups and income profiles, can then be derived. This can then lead to estimates of CO2 emissions from cars. Estimates of water use and electricity can also be calculated, using figures provided by the ABS in their publication Household Energy Consumption Survey, Australia (ABS 2014) and from their Water Account, Australia (ABS 2015). Data on water availability are available for most cities through public reports from water authorities. Employment estimates could be based on the number of people of working age in the area and unemployment rates for the area (from the small-area unemployment estimates; see Australian Government Department of Employment 2013) combined with area data on employment by industry. Such data are available from the 2011 Census. Data on industry by occupation are also available for each person in the base data set of the spatial microsimulation model. In the model, we know that the industries are already in the area (from the Census data), so the starting level of employment is exogenous to our model. The model then assigns people to areas based on employment growth. Liveability in the area is difficult to model, but from other research we know that liveability is complex and associated with many variables. Some of these include the availability and quality of recreational facilities, transportation, the built environment and the design of communities for physical activity (Dannenberg et al. 2011). Some of this information could be available in the base data set (e.g. availability of recreational facilities) and some can be modelled separately.

5 Using Spatial Microsimulation to Derive a Base File for a Spatial Decision. . .

5.5

117

Data and Local Government Support

As outlined in the framework above, a spatial microsimulation model can be brought into an SDSS by using the model to create the base file that the SDSS will start with. This provides a much more detailed data set than the age–sex information provided by a cohort component model. Additionally, the spatial microsimulation model generates unit record data, whereas a cohort component model only provides aggregate data. However, there is a significant disadvantage to microsimulation modelling and this is that it is highly data-intensive. Ideally, birth and death rates would be required for each small area, so that the dynamic model can apply these to the population in that area. However, such data are rarely available at a very fine spatial scale. Because of this, most dynamic spatial microsimulation models assume that the area is homogenous, so that these parameters do not vary spatially. In the case of Canberra, it is a reasonable assumption that this is a fairly homogenous area, because we would expect similar birth and mortality rates to apply across the area. One example of a successful dynamic spatial microsimulation model is SVERIGE, a spatial microsimulation model for Sweden which combines a deterministic model of individual behaviour with Monte Carlo simulation. SVERIGE uses the patterns of emigration, immigration, employment and earnings, education, leaving home, divorce, cohabitation and marriage, as well as mortality and fertility, as the basis for modelling the dynamic behaviour of individuals in the model. The Monte Carlo simulation picks experiences for each individual in the microdata to subsequently experience based on probabilities and, hence, updates the individual characteristics in the microdata. SVERIGE uses a huge amount of data, derived from extensive data sets available in Sweden through the Swedish Government. Another dynamic spatial microsimulation model is SMILE, which was built as both a static and dynamic spatial microsimulation model (Ballas et al. 2005). It was constructed to estimate and project small-area population statistics in Ireland. The model starts as a static model using an iterative proportional fitting (IPF) method to spatially disaggregate the aggregate microdata. Once this has been done, the demographic processes of mortality, fertility and migration are simulated. The mortality process is simulated by using the probability of death based on age, gender and location while the probability of birth is simulated based on age, marital status and location. The simulation of the migration process uses random sampling from calculated migration probabilities derived from the 1991 and 1996 Census of Population. These data provide migration probabilities from one area to another by age, gender and location. Again, the amount of data required is immense, but is available for Ireland. The importance of detailed data also means that strong links to the local government are advantageous for the modelling. Given that the local government would be the main users of the simulations, their involvement would normally already be sought at the stage of developing this type of model.

118

R. Tanton and Y. Vidyattama

The information provided by the local government can include three types of information: 1. information about the topological structure or building cadastral of the area; 2. information about population projections; and 3. information on future development planning of the area. While the dynamic microsimulation model will be deriving its own constrained population projections, the local government may be able to provide other data that may contribute to the modelling. For example, the cadastre-level data for specific areas will allow the model to assign families to actual households; future developments may already be planned that could be incorporated into the model; and there may be information on migration or fertility rates that the local government knows that can be fed into one of the models. All this means that a close relationship with the local government is essential for the success of the modelling and their data supply is paramount.

5.6

Conclusions

In this chapter, we have outlined the advantages of a Spatial Decision Support System (SDSS) to assist planning decisions compared to a cohort component population projection method that provides simple unconstrained age–sex projections. The main advantages of the SDSS are that, firstly, it can provide other indicators besides just the population size and distribution by age and sex and secondly that these indicators can then inform the planning process and thirdly feed into other models in the SDSS, which can apply demographic, economic and social constraints. We have shown that unconstrained population growth is a particular issue for the ACT, a land-locked area in Australia in which the population is already approaching its borders. We propose a way to model population growth that recognizes economic, environmental and social limits by means of a framework for an SDSS that will be able to provide scenarios to better inform planners and residents about what a potential future ACT may look like. The framework is innovative because it uses a synthetic population in each small area based on a spatial microsimulation methodology, but then uses a number of other models to bring in economic, social and environmental indicators and limits. This spatial microsimulation model would ideally be a dynamic model. Hence, rather than using population projections from another source, it would derive population projections internally, but with physical and environmental limits. We have also highlighted one of the limitations of this type of model, which is its reliance on very detailed data. Extensive data would be required for each small area, which means that strong links need to be established with the local government to access such data. Applying the model to a homogeneous area, like Canberra, reduces some of this need for data as it is reasonable to assume that fertility rates are constant

5 Using Spatial Microsimulation to Derive a Base File for a Spatial Decision. . .

119

across Canberra. However, even in the case of Canberra, there is still a need for significant amounts of data. Nonetheless, we have identified and reviewed two overseas models that have overcome this spatial data hurdle. Hence, the full development of the model is possible as long as the researcher has good links to government data. What we have been able to show in this chapter is that a spatial microsimulation model can make a valuable contribution to an SDSS by creating a synthetic population in every small area. This synthetic population will have much more information than a cohort component model can provide. Being at the individual level means other information can be imputed onto the synthetic population. The resulting base data set can then be used by a number of different models to develop estimates of economic indicators, infrastructure need, environmental impacts, employment, industry impacts and liveability. These models can then be used to provide scenarios for the future of a city—based on different population densities and new developments in different areas, how many jobs will be required, new road and public transport requirements and a range of other planning needs, depending on the models that have been incorporated.

References ACT Government (2014) Territory plan. ACT Government, Canberra, ACT Australian Bureau of Statistics (2013a) Population projections Australia: 2012 (Base) to 2101. ABS, Canberra Australian Bureau of Statistics (2013b) Microdata: Australian Census Longitudinal Dataset, 20062011. Cat # 2080.0. ABS, Canberra Australian Bureau of Statistics (2014) Household Energy Consumption Survey, Australia. Cat # 4670.0. ABS, Canberra Australian Bureau of Statistics (2015) Water Account, Australia, 2013-14. Cat # 4610.0. ABS, Canberra Australian Government Department of Employment (2013) Small area labour markets Australia. Canberra Australian Institute of Health and Welfare (AIHW) (2016) Customized projections prepared for the Australian Government. Department of Social Services by the Australian Bureau of Statistics, AIHW, Canberra. http://www.aihw.gov.au/nacdc/population-projections/. Accessed 12 Aug 2016 Ballas D, Clarke GP, Wiemers E (2005) Building a dynamic spatial microsimulation model for Ireland. Popul Space Place 11(3):157–172 Birkin M, Clarke M (1988) SYNTHESIS—a synthetic spatial information system for urban and regional analysis: methods and examples. Environ Plann A 20(12):1645–1671 Birkin M, Turner A, Wu B, Townend P, Arshad J, Xu J (2009) MoSeS: a grid-enabled spatial decision support system. Social Sci Comput Rev 27(4):493–508 Bloom DE (2011) 7 billion and counting. Science 333(6042):562–569 Booth H (2006) Demographic forecasting: 1980 to 2005 in review. Int J Forecast 22(3):547–581 Chakroun H, Benie GB (2005) Improving spatial decision support systems. Appl GIS 1(1):1–27 Dannenberg A, Frumkin H, Jackson R (eds) (2011) Making healthy places: designing and building for health, well-being and sustainability. Island Press, Washington

120

R. Tanton and Y. Vidyattama

Densham P, Armstrong M (1994) A heterogeneous processing approach to spatial decision support systems. In: Waugh T, Healey RG (eds) Advances in GIS research. Proceedings of the Sixth International symposium on spatial data handling, pp 29–45 Dorling D (2013) Population 10 billion: the coming demographic crisis and how to survive it. Constable and Robinson, London ESRC (2016) TALISMAN: geospatial data analysis and simulation. http://www.researchcatalogue. esrc.ac.uk/grants/RES-576-25-0039/read. Accessed 13 Sep 2016 Lowe I (2012) Bigger or better: Australia’s population debate. University of Queensland Press, Brisbane Marois G, Bélanger A (2014) Microsimulation model projecting small area populations using contextual variables: an application to the Montreal Metropolitan Area, 2006–2031. Int J Microsimul 7(1):158–193 Meadows DH, Meadows DL, Randers J, Behrens WW (1972) The limits to growth. Potomac Associates—Universe Books, New York Namazi-Rad M, Tanton R, Steel D, Mokhtarian P, Dasc S (2017) An unconstrained statistical matching algorithm for combining individual and household level geo-specific census and survey data. Comput Environ Urban Syst 63(May):3–14 Norman BWS, Heath L, Vidyattama Y, Gilmour C, Zhang W, Weir B (2012) Scoping study for a knowledge portal for the Australian Capital Region (ACR). Report to the ACT Government October 2012, Canberra Rao M, Tanton R, Vidyattama Y (2013) A systems approach to analyse the impacts of water policy reform in the murray-darling basin: a conceptual and an analytical framework. NATSEM Working Paper 13/22, Canberra Tanton R, Edwards KL (eds) (2013) Spatial microsimulation: a reference guide for users. Springer, Heidelberg Tanton R, Vidyattama Y, Nepal B, McNamara J (2011) Small area estimation using a reweighting algorithm. J R Stat Soc Ser A (Statistics in Society) 174(4):931–951. https://doi.org/10.1111/j. 1467-985X.2011.00690.x Van Imhoff E, Post W (1998) Microsimulation methods for population projection. Population 10 (1):97–138 Vidyattama Y, Tanton R (2010) Projecting small area statistics with an Australian spatial microsimulation model (SpatialMSM). Aust J Reg Stud 16(1):99–126 Watson N, Wooden M (2001) The household, income and labour dynamics in Australia (HILDA) survey: an introduction. Aust Soc Policy 2:79–99 Wilson T (2015) New evaluations of simple models for small area population forecasts. Popul Space Place 21(4):335–353 Wilson T (2016) Evaluation of alternative cohort-component models for local area population forecasts. Popul Res Policy Rev 35(2):241–261 Wu B, Birkin M (2013) Moses: a dynamic spatial microsimulation model for demographic planning. In: Tanton R, Edwards KL (eds) Handbook of spatial microsimulation. Springer, Berlin

Part II

Migration and Development

Chapter 6

The Drivers of Long-Distance Commuting in Chile: The Role of the Spatial Distribution of Economic Activities Francisco Rowe

and Martin Bell

Abstract Long-distance commuting has emerged as an alternative to migration to equilibrate spatial labour markets. Coupled to changes in the labour and housing markets, technological advances have promoted long-distance commuting by reshaping the links between the spatial distribution of population and regional economies. While previous research has examined these links in developed countries, less is known about how these changes have played out in developing economies. Using micro-census data and regression analysis, this chapter addresses how contextual factors have shaped long-distance commuting in Chile. Our results reveal that the nature and spatial distribution of mining and construction activities have been the primary drivers of long-distance commuting in Chile. This contrasts with developed countries where, along with these activities, factors associated with the new service economy also comprise predominant forces encouraging long-distance commuting, particularly for those in high-skilled occupations. Keywords Long-distance commuting · Spatial labour markets · Extractive industries · Chile · Census

F. Rowe (*) Geographic Data Science Lab, Department of Geography and Planning, University of Liverpool, Liverpool, UK e-mail: [email protected] M. Bell Queensland Centre for Population Research, School of Earth and Environmental Sciences, University of Queensland, Brisbane, QLD, Australia e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2020 J. Poot, M. Roskruge (eds.), Population Change and Impacts in Asia and the Pacific, New Frontiers in Regional Science: Asian Perspectives 30, https://doi.org/10.1007/978-981-10-0230-4_6

123

124

6.1

F. Rowe and M. Bell

Introduction

Labour migration and long-distance commuting interact to regulate discrepancies in regional labour markets. Individually, these processes reflect spatial flexibility; jointly, they shape the way labour supply adjusts to demand (Yapa et al. 1971; Evers and Van Der Veen 1985). Over the last three decades, substantial attention has been devoted to the forces driving long-distance commuting (Rouwendal and Rietveld 1994; Green et al. 1999; Helminen and Ristimäki 2007), showing that telecommunication and transport technologies (Hardill and Green 2003), new and flexible forms of work organisation (Houghton 1993), along with rising female labour force participation and home ownership (Green et al. 1999; Oswald 1999), and persistent spatial differentials in living costs (Cameron and Muellbauer 1998) play significant roles. To date, however, research has concentrated on developed nations and focused on commuting within metropolitan areas, rather than between more extensive labour market regions (Eliasson et al. 2003; Melo et al. 2012). Less is known about how these developments have played out in developing economies, partly due to the lack of appropriate data. In practice, however, there are strong reasons to suggest that the impacts of some of the above developments may be even more pronounced in the developing world. In Latin American (LA), for example, following a period of import substitution and population concentration from the 1930s to the 1970s, several economies experienced tumultuous transformations, are characterised by economic liberalisation and export-oriented development, improved access to telecommunication and transport technologies and increased inflows of Foreign Direct Investment (FDI) (Economic Commission for Latin America and the Caribbean (ECLAC) 2012). As a result, many large-scale natural resource-based projects developed in locations distant from major population centres (ECLAC 2012). These projects reshaped the economic landscape of LA countries and triggered substantial adjustment in labour supply. Aroca and Atienza (2008, 2011) revealed the negative economic impacts of long-distance commuting on the Chilean mining region of Antofagasta, but to date, there has been no attempt at a more systematic, nation-wide analysis of the extent and driving forces of long-distance commuting in Chile. This chapter aims to redress this deficit by examining how the changing spatial configuration of economic activities and human settlement pattern have shaped longdistance commuting in Chile. In particular, the chapter seeks to determine the key contextual forces that have shaped long-distance commuting in Chile. Using confidentialised micro-data of the 2002 Census of the population, this paper first analyses the patterns of commuting between Chilean regions and then applies a linear regression model to identify the factors driving long-distance commuting. The results seek to establish whether long-distance commuting in Chile has responded in a similar way to the set of forces found to trigger the emergence of long-distance commuting in developed countries.

6 The Drivers of Long-Distance Commuting in Chile: The Role of the Spatial. . .

125

The next section reviews the contextual forces identified as shaping long-distance commuting in developed countries and argues how these forces may have unfolded in Chile as a result of a period of remarkable economic and political transformation following the country’s rapid economic liberalisation. Section 6.3 then examines the literature on the individual factors associated to long-distance commuting and establishes a number of hypotheses to be tested in the Chilean context. Section 6.4 describes the data and methodology used before the results are presented and discussed in Sect. 6.5. Section 6.6 summarises the key findings and discusses policy implications.

6.2

Contextual Forces Promoting Long-Distance Commuting

Long-distance commuting is generally defined as the journey from a place of residence to a place of employment over a certain distance threshold.1 Unlike migration, long-distance commuting does not involve a ‘permanent’ change of residence and is highly variable in duration (Bell and Ward 2000). In spatial labour markets, long-distance commuting is an alternative to migration as a mechanism to meet labour demand. People may decide to commute, migrate or a combination of both in order to take up a job opportunity in a distant location. Individual decisions, thus, connect these two forms of mobility, with the implication that changes affecting migration may also impact long-distance commuting or vice versa. Mobility transition theory suggests that as countries develop, levels of migration are likely to decline due to increasing long-distance commuting (Zelinsky 1971). As nations develop, various forces operate to increase commuting levels. With improvements in transportation infrastructure and Information and Communication Technologies (ICTs), ‘new’ and more flexible work practices have emerged. Various forms of remote working, including home working and teleworking, have become increasingly common, giving rise to a greater spatial separation between home and work (Hardill and Green 2003; Janelle 2004; Helminen and Ristimäki 2007), while the introduction of more advanced transport technologies has reduced travel times and costs (Rouwendal and Rietveld 1994; Rietveld and Vickerman 2004; Lafourcade and Thisse 2011). Coupled to these developments, rising female labour force participation and home ownership have also fostered long-distance commuting. Rising female labour force participation has entailed the formation of dual-earner and dual-career households, promoting a greater substitution of long-distance commuting for migration for as a mobility and earnings strategy and vice versa (Green 1995). These households seek to avoid migration as it is likely to imply sacrificing one partner’s career. Rising

1

Various thresholds have been used. Section 6.4 provides a short discussion of these thresholds and the definition of long-distance commuting.

126

F. Rowe and M. Bell

home ownership may also involve substitution of long-distance commuting for migration for as owner-occupiers tend to move less due to the high transaction costs and strong emotional attachment to their homes (Oswald 1999).

6.2.1

Contextual Forces in Chile

These developments have also occurred in the developing world and, like urbanisation and the demographic transition, they are likely to have unfolded more rapidly than in developed countries (United Nations 2004). In Chile, the path to modernisation appears to have further fostered the growth of long-distance commuting through a unique sequences of economic and political developments. From the 1930s to 1973, Chile adopted an Import Substitution Industrialisation (ISI) strategy of economic development and a closed, centrally planned economic regime. This economic model promoted migration to the main centre of employment of the country, Santiago, resulting in increasing population concentration in the Metropolitan Region (MR) during the 1960s (Herrick 1965). This was largely because, under the ISI model, companies located close to areas with large populations, in order to sell to the domestic consumer market (Gilbert 1993; Uribe-Echevarría 1995, 1996). As the MR had a privileged location in the middle of the country, the largest consumer market and centralised governmental and business activity, this region experienced rapid population growth and clustering of firms. In 1973, Chile underwent a military coup commanded by Pinochet which terminated the ISI model. Between 1975 and 1982, the military government implemented a series of structural reforms to stabilise and transform the Chilean economy into a free-market system (Corbo et al. 1997). These reforms achieved fruition in the early 1990s, when Chile transitioned into a globalised, market-oriented economy, with rapid economic growth and increasing FDI (see Fig. 6.1). Figure 6.2 shows the distinctive spatial distribution of population, mining and construction activities in Chile characterised by high concentration of population in the MR, clustering of mining in northern regions and wide distribution of construction activity. From the early 1990s, regions outside the MR have experienced consistently high levels of FDI, generating large increases in labour demand for construction workers, mining operators and technical experts. However, since human capital in Chile is concentrated in the MR, and the main cites in mining regions are unattractive places to live (Aroca and Atienza 2008; Lagos and Blanco 2010; Rowe 2013b), companies have sought to source labour through the use of Fly-in/Fly-Out (FIFO) and Drive-in/Drive-out (DIDO) operation arrangements, and it is this discrepancy that is argued to have been pivotal in promoting long-distance commuting (Rowe 2013a). In developed countries, the nature of mining and construction is often identified as a major trigger of long-distance commuting. In Australia and Canada, most longdistance commuting is driven by the use of shift rosters to manage mining and

6 The Drivers of Long-Distance Commuting in Chile: The Role of the Spatial. . .

Pre-military coup

127

Post-military coup

Source: Authors’ elaboration using GDP data from Díaz et al. (2010) and FDI data from Foreign Investment Commitee (www.foreigninvestment.ck).

Fig. 6.1 Trends in FDI and GDP growth, pre- and post-military coup in 1973

construction operations, promoting extended schedules away from home (Storey and Shrimpton 1988; Houghton 1993). In the construction sector, these work arrangements are common because construction projects have tight schedules and involve short-term contracts (Storey and Shrimpton 1988). In the mining sector, they are often used because mining sites are located in remote and inhospitable areas far from major cities and services. In an effort to reduce operating costs, companies prefer to finance temporary accommodation and commuting, rather than maintain mining towns adjacent to mineral deposits (Storey and Shrimpton 1988; Houghton 1993; Storey 2001). In industrialised countries, the nature of mining and construction activities have not acted in isolation. Key aspects associated to the new service economy have also been pivotal in increasing the level of long-distance commuting.

6.3

Who Commutes Long Distances?

Prior work in industrialised countries has identified six broad categories of factors that promote the likelihood of long-distance commuting. These are labour market characteristics, geographical context, housing market circumstances, demographic characteristics, household composition and migration status. The influence of economic activities in promoting long-distance commuting is reflected in individual labour market characteristics. For instance, Storey and Shrimpton (1988) found high propensities for long-distance commuting among Canadian mining and construction workers, reflecting the remote location of mining sites and the use of FIFO in mining operations, and also the temporary nature and short duration of construction projects. Long-distance commuting has also been

128

F. Rowe and M. Bell

Source: Authors’ elaboration using the 2002 census. Note: the map was simplified using the DouglasPeucker algorithm.

Fig. 6.2 Spatial distribution of population, mining and construction activity in Chile

linked to the agriculture and transportation sectors, reflecting the seasonal nature of agricultural work (Bell and Brown 2006; Hanson and Bell 2007; Silva et al. 2011) and transport system jobs, such as aircrew and drivers (Storey and Shrimpton 1988). Particular occupations and work classes foster long-distance commuting. Individuals in high-skilled occupations are more likely to commute over longer distances because of the prestige, financial benefits and intrinsic satisfaction of their jobs and also because these occupations enjoy greater flexibility in working conditions (Green et al. 1999). Companies are well disposed to provide greater flexibility to highly skilled, managerial and professional employees (Green et al. 1999; Hardill and Green 2003). Employment type also matters. The self-employed and employers are not likely to commute over long distances (Giuliano 1998), because they enjoy greater freedom, enabling them to migrate, rather than commute (Van Ham and Hooimeijer 2009). Geographical context of residential location is a major factor promoting long commutes in Canada (Axisa et al. 2012). People living in rural areas tend to have longer commutes than those in densely populated regions, as employment opportunities tend to concentrate in metropolitan areas. On the other hand, greater job

6 The Drivers of Long-Distance Commuting in Chile: The Role of the Spatial. . .

129

accessibility in metropolitan centres may also be associated with long-distance commuting (Van Ham and Hooimeijer 2009). Compared to residents in areas with low job accessibility, those in metropolitan areas can reach a large number of potential jobs by making small investments in commuting distance. This, thus, promotes commuting and reduces incentives to migrate for employment reasons (Van Ham and Hooimeijer 2009). In capturing these opposing effects of the geographical context, the effects of urban hierarchy and job accessibility on commuting should be considered together. Less densely populous areas do not necessarily have low job accessibility. Indeed, Chilean census data show that various municipalities in northern and central-southern Chile with low population density have high job accessibility. Labour market effects operate simultaneously with the influence of housing market conditions. High housing prices often promote long-distance commuting in flows because of their deterrent effect on in-migration (Cameron and Muellbauer 1998). Similarly, housing ownership is argued to promote commuting by deterring migration. High transaction costs related to home ownership often discourage migration, and hence, it is argued to promote long-distance commuting (Van Ommeren et al. 2000). In fact, owner-occupiers have been consistently found to be less likely to migrate than renters (Clark and Dieleman 1984; Helderman et al. 2004, 2006). Migration is likely to lead to long-distance commuting. In Canada and England, recent migrants were found to commute over longer distances than longer term residents (Champion et al. 2009; Axisa et al. 2012). This often reflects commuting back to the place of work prior to relocation, before adjusting work location (Champion et al. 2009), or because the destination region lacks suitable housing (Green et al. 1999). Unlike for migration, there is no consensus on the effect of age on long-distance commuting. For migration, age represents key events in the life cycle (Rogers et al. 1978), while propensities for long-distance commuting have been reported to be decreasing with age in the Netherlands (Van Ham et al. 2001) and peaking at mid-working ages in Spain (Artís et al. 2000). Higher educational attainment and being male appear to promote long-distance commuting, with males travelling longer distances than females—reflecting the greater household responsibilities of the latter (Madden 1981; Hanson and Pratt 1995). Highly educated individuals usually display longer commutes because after investment in education they are willing to commute over long distances to secure suitable jobs where necessary (Van Ham et al. 2001; Eliasson et al. 2003). Highly educated workers are also overrepresented in highly skilled occupations possessing greater flexibility, which enables them to alternate days at the office and at home. Household composition also fosters long-distance commuting. Reflecting the effect of rising female labour participation on spatial choices of families, dualearner households tend to have higher commutes and lower migration intensities (Agnes 1974; Shick and Tryon 1982; Green 1997). This is because migration may imply for the partner the sacrifice of a career and promotion prospects, disruption to contact networks and adaptation to a new routine (Mincer 1978; Green et al. 1999). Dual-earner households, thus, tend to have higher propensities for long-distance

130

F. Rowe and M. Bell

Table 6.1 Variables shown in previous research to be positively associated with long-distance commuting Category Labour market

Geographical context

Housing tenure Recent migration status Demographic characteristics Demographic characteristics Household composition

Variable Employment type Industry sector Occupation Population density Job accessibility Housing tenure Migration status Age Educational level Gender Household structure

Longer commuting journeys expected for Full-time employees (excluding the self-employed and employers) Workers in mining, agriculture, construction and transportation Workers in highly skilled occupations Workers living in densely populated locations Workers living in locations with higher job accessibility Owner-occupiers Those who have previously migrated No consensus Workers with higher educational levels Males Dual-earner households, especially those with children

Source: Based on previous research (see references in text)

commuting than one-earner households (Green 1997). Households with children face further restrictions to migration since moving disrupts education and friendship networks (Green 1997). A lengthening of commuting distance and time, thus, represents a strategy for dual-earner households, especially for those with children, to avoid these disruptions (Green 1997; Green et al. 1999). From this review, a set of hypotheses concerning the characteristics of longdistance commuters can be derived (Table 6.1). The next section describes the data and methodology to empirically test their effect on long-distance commuting in Chile.

6.4

Data and Methodology

To determine the influence of the factors that may promote long-distance commuting, this chapter uses confidentialised micro-data on the 2002 Chilean Census of Population and Housing. The data were extracted from the Chilean Migration (CHIM) database assembled by Rowe and Bell (2013). This database contains temporally harmonised data on migration for three successive census periods in 1982, 1992 and 2002, and data on commuting for 2002. In 2002, for the first time, the Chilean census collected information on individuals’ place of employment, affording the derivation of commuting data by comparing this information to the

6 The Drivers of Long-Distance Commuting in Chile: The Role of the Spatial. . .

131

place of usual residence. Rowe (2017) provides a full description of the CHIM database. Commuting data can be obtained at three hierarchical levels of geography, representing the administrative boundaries of Chile: municipalities (304 minor spatial units), provinces (51) and regions (13 major spatial units). It is important to note that the number of municipalities and regions contained in the CHIM database differs from their current numbers, 346 and 15, respectively, due to the creation of four municipalities and two regions, and adjustments to account for boundaries changes over time. As reported by Casado-Díaz et al. (2017a, b), transcription errors in census records, and due to wrong recording of place of employment and place of usual residence data, were amended. The analysis focuses on the employed segment of the workforce aged 15–64, excluding unemployed individuals and those not in the labour force (i.e. students, retirees and housewives). The data comprise 4,484,953 individual records, equivalent to 79% of the Chilean workforce in 2002. The analysis consists of two parts. The first part analyses the patterns of commuting between the 13 major regions of Chile. The second part comprises a multivariate analysis which seeks to determine the influence of the above range of variables on commuting distance. The model is specified as follows: D ¼ Xβ þ e

ð6:1Þ

where D is a n x 1 vector with elements di representing the natural logarithm of commuting distance for an individual i. X is a n x K matrix containing information on various labour market, geographical context, housing market, demographic, previous migration experience and household characteristics. ß is a K x 1 vector of parameters to be estimated. e is a n x 1 vector with elements ei representing unobserved error terms, which are assumed to be independent and normally distributed. To estimate Eq. (6.1), commuting distance (in km) was log transformed to normalise its right-skewed distribution, which reflects the distance decay effect on geographical mobility. This transformation is convenient because Eq. (6.1) can, thus, be estimated by using ordinary least squares (OLS), and coefficients can be interpreted as percentage change effects: when multiplied by 100, a ß coefficient indicates the percentage change in commuting distance associated with a unit change in the corresponding independent variable. It is important to note that our analyses are based on a large data set of over four million observations. Working with large data sets entails careful interpretation of the results to draw conclusions from estimates. With large data sets, standard statistical significance testing becomes less relevant as any null hypothesis will almost certainly be rejected (Granger 2003). Rather than based on statistical significance, evaluation of model estimation should be in terms of economic significance or relationships that are theoretically relevant (Kashyap and Stein 2000). We, thus, focus on theoretically significant relationships in the data, rather than on statistical tests.

132

6.4.1

F. Rowe and M. Bell

Dependent Variable

Unlike most previous research, this chapter uses the natural logarithm of commuting distance as the dependent variable. Previous work often defines long-distance commuting by means of a dichotomous split, in which moves over a certain distance or time threshold, or crossing certain geographical boundary, are considered longdistance commuting moves and recoded into a binary variable for logistic regression analyses. For instance, Van Ham and Hooimeijer (2009) defined long-distance commuting as work trips that involved journey times longer than 75 min (return), while Öhman and Lindgren (2003) considered long-distance commuting moves to be those over 200 km. OECD (2005) defined long-distance commuting as moves that cross a boundary between large administrative areas. These definitions are rather arbitrary and can bias the results by focusing only on one segment of the distance distribution. Following the recent work by Axisa et al. (2012), we argue that using the entire distance distribution of commuting moves is likely to provide more sensible assessment of the factors that influence long-distance commuting. In this context, the magnitude of a regression coefficient captures the length of commuting distance. If positive, a larger regression coefficient indicates that an individual commutes a longer commuting distance for a one unit increase in the explanatory variable. Measurement of commuting distance is based on distance between centroids of municipal areas, which represent the smallest geographical units at which information on place of residence and employment is collected in Chile. Previous analyses have commonly ignored moves within the same spatial unit, but omitting these moves may bias results by truncating the distance distribution (Bell et al. 2002). Some intraregional moves involve longer distances than those between regions (Boyle et al. 1998). Moreover, excluding intraregional moves reduces the total number of movers retained in the analysis, and this influences the resulting parameter estimates (Boyle and Flowerdew 1997). To address this issue, we borrow from the literature on internal migration by estimating intra-municipal distances as equivalent to half the radius of a circle of each municipality’s equivalent area (Long et al. 1988; Bell 1995). As Rogerson (1990) indicated, this is a more accurate estimate than the full radius of a geographical circle. Figure 6.3 shows the distribution of moves according to commuting distance. As expected, commuting decreases sharply with distance. Of the more than 4.4 million people in the workforce, 50% commute less than 12.4 km and only over 3% (107 thousand people) commute more than 100 km from their municipality of residence. In addition, Fig. 6.4 displays a cartogram map of the area size and degree of mining specialisation of municipalities. It shows that northern municipalities specialised in mining tend to cover larger geographical areas than non-mining municipalities in central and central-southern Chile. Because mining deposits are usually located far from urban areas, this evidence indicates that individuals living and working in mining municipalities may commute over long distances—maybe covering more extensive areas than those commuting between central and central-southern

6 The Drivers of Long-Distance Commuting in Chile: The Role of the Spatial. . .

133

All distances 30.0

Percent

20.0

5,000,000 4,500,000 4,000,000 3,500,000 3,000,000 2,500,000 2,000,000 1,500,000 1,000,000 500,000 0

25.0 20.0

15.0

15.0 10.0

10.0

5.0 0.0 05- 5 10 10 15 1 20 20 25 25 30 30 35 35 40 40 45 45 50 50 55 55 -6 60 0 65 65 70 70 75 75 80 80 85 85 90 90 95 -95 -1 00

5.0

4,500,000

Distances under 100 km 30.0

3,500,000 3,000,000 2,500,000 2,000,000 1,500,000 1,000,000 500,000 0

070 5 14 -75 021 145 028 215 035 285 042 355 049 425 056 495 063 565 070 635 077 705 084 775 091 845 098 915 10 0-9 50 85 11 -10 20 55 11 -11 90 25 12 -11 60 95 13 -12 30 65 14 -13 00 35 14 -14 70 05 15 -14 40 75 16 -15 10 45 16 -16 80 15 17 -16 50 85 18 -17 20 55 -1 82 5

0.0

4,000,000 Number of commuters

25.0

5,000,000

Distance (five-km interval) Percent of all moves

Cumulative distribution

Source: Authors’ calculations, using the 2002 census. Distance corresponds to the Euclidian distance between centroids of an individual’s municipalities of residence and employment.

Fig. 6.3 Distribution of commuting moves by distance (km)

municipalities. Excluding intra-municipal commuting distances is, thus, likely to distort estimates of the relationship between mining activity and commuting distance.

6.4.2

Independent Variables

To examine the influence of the various factors in Table 6.1, a suite of independent variables was included in the regression model. Age, gender and education variables were considered to capture the effects of demographic characteristics. Since age has been found to have a non-linear effect on commuting, we also included age squared in our regression model. Household composition variables were also considered to represent the negative impact of being a member of a dual-earner household and having children. Housing tenure variables were considered to assess the negative effect of being a home owner. A migration indicator variable was used to evaluate the effect of recent migration on commuting.2 Variables indicating the industry, occupation and class of employment of individuals were considered to assess the

2

Our model does not suffer from endogeneity problems. Long-distance commuting may be seen as a response to, and a cause of, migration. Our analysis seeks to estimate the influence of a past migration move on ‘present’ long-distance commuting. To this end, we capture migration by comparing census information on usual place of residence in 1997 and 2002; that is, prior to the date of the commuting data (i.e. the census date).

134

F. Rowe and M. Bell

Tarapacá Antofagasta

Atacama Coquimbo Valparaíso

MR

O´higgins Del Maule Bío-bío Araucanía Los Lagos

Aysén

Quarles and outliers: Mining Locaon Quoent (Number of municipalies)

Magallanes

Fig. 6.4 Cartogram of the geographical size of municipalities in Chile (in km2). Colours represent the geographical size distribution across quartiles based on their specialisation in mining

influence of labour market factors, and four variables were included to account for the geographical context: population density, job accessibility, mining specialisation and geographical centrality of residential location. The variable mining specialisation accounts for the fact that individuals living near to mining areas may be less likely to be long-distance commuters. However, this variable may also capture differences in geographical area size. This is likely to be the case because distances for intra-municipal commuters are directly dependant

6 The Drivers of Long-Distance Commuting in Chile: The Role of the Spatial. . .

135

on the size of their residential areas, and as shown by Fig. 6.3, mining municipalities tend to be larger than non-mining municipalities. We also use the variable geographical centrality to capture variations in commuting distances according to individuals’ municipality of residence. Workers living in central Chile have a theoretically shorter distance to commute compared to individuals in the extreme north and south. Appendix 1 provides detailed information on the definition of these variables and presents descriptive statistics of the data used in the analysis. This study is particularly interested in the role of the spatial distribution of economic activity and the pattern of human settlement on promoting long-distance commuting. Hence, particular attention is placed on examining the coefficients of the variables representing the influences of industry sector and geographical context on commuting distance. We expect large and statistically significant coefficients for variables associated with working in mining and construction and living in areas with high population density and high specialisation in mining.

6.5

Results and Discussion

Before discussing the regression analysis, we establish a general understanding of the inter-regional pattern of commuting. Although these moves include some shortdistance mobility, most involve fairly long distances and provide an accurate representation of the forces shaping labour commuting in Chile (Rowe 2014).

6.5.1

Patterns of Inter-Regional Commuting

Census data reveal that 2.6% of the employed workforce (114,900 workers) commuted to work in a region is different from their region of residence in 2002. While census data indicate that this is smaller than the percentage of workers who migrate between regions (6.7%), it is important to point out that these statistics are not readily comparable. Migration can only be measured over a 5-year transition interval using 2002 census data (1997–2002), while commuting is measured at the census day. To assess the role of commuting in regional labour markets, Table 6.2 presents the net commuting outcomes together with in- and out-commuting flows. The broad pattern that emerges is that commuting has operated ‘efficiently’ by removing labour from north-central regions (Coquimbo and Valparaíso) and agricultural regions in south-central Chile (O’higgins, Del Maule, Bío-bío and Araucanía) and by expanding labour supply in mining regions in northern Chile (Tarapacá, Antofagasta and Atacama), remote southern regions (Los Lagos, Aysén and Magallanes) and the MR. Within this broad pattern, four features stand out. The first is the modest role of commuting in the labour supply of the MR. Despite representing the main centre of national employment and attracting the largest commuting inflow, this region saw an

136

F. Rowe and M. Bell

Table 6.2 Statistics of inter-regional commuting Geographic location North

Central

South

Regions Tarapacá Antofagasta Atacama Coquimbo Valparaíso Metropolitan Region O’Higgins Maule Bío-bío Araucanía Lagos Aysén Magallanes

Commuting In Out 4651 2774 16,396 2027 4486 3941 4069 8012 10,387 21,356 33,715 29,332

Net 1877 14,369 545 3943 10,969 4383

Net of the work for CE (%) 1.2 8.3 0.6 2.0 2.1 0.2

10,353 6059 9170 4433 7295 1974 1947

749 1649 6870 3657 3339 1800 1524

0.3 0.5 1.1 1.4 0.9 5.2 2.5

11,102 7708 16,040 8090 3956 174 423

Source: Authors’ elaboration using 2002 census data

expansion in its workforce through commuting of only 0.2%, as a result of a large offsetting outflow of workers, principally to its adjacent regions, Valparaíso and O’higgins. Movements from/to the MR to/from Valparaíso and O’higgins account for more than 28% of the commuting flows in the entire system and often take place over long distances, involving a distance of more than 80 km, about 1.5-h travel time. Commuting flows from the MR to adjacent regions appear to reflect the temporary nature of the construction sector as they are dominated by workers in this sector (accounting for 18% of the outflows), while commuting flows from Valparaíso and O’Higgins to the MR seem to respond to their geographical proximity to the main employment centre of the country, the Santiago Metropolitan area. Workers can access a large and diverse pool of jobs in the MR by commuting over moderate distances, while they avoid the high housing costs in this densely populated area. A second major feature is the role of commuting in removing labour from agricultural regions. There are large commuting flows from agricultural regions to the MR. They account for 10% of the flows in the system and are dominated by workers employed in the construction sector, suggesting that the temporary character of construction activities is a major factor promoting these moves (Rowe 2013a). There is another network of commuting flows between agricultural regions which appear to reflect the seasonal nature of agricultural work. This is because commuting flows between agricultural regions are dominated by moves between adjacent regions with high specialisation in agriculture and by movements of people employed in the agricultural sector (over 33%), echoing the temporary circuits of agricultural employment identified in Australia (Hanson and Bell 2007).

6 The Drivers of Long-Distance Commuting in Chile: The Role of the Spatial. . .

137

The third feature is the significant role of commuting in Aysén and Magallanes. Despite their remote location at more than 2000 km from the MR, fragmented geographical territory and absence of land connection within the country, commuting operates to greatly expand the local labour supply of these regions. This expansion is more pronounced than for most regions. While it is magnified by the small population size of these regions, the regular posting of military personnel appears to be a key factor promoting commuting to these remote locations (Rowe 2013a). The fourth feature is the role that commuting plays in supplying labour to the mining regions, especially to Antofagasta. Commuting contributes over 8% of total labour supply in this region, with attractive power that extends beyond adjacent regions to Bío-bío more than 1500 km south (Aroca and Atienza 2011). As described in Sect. 6.2.1, mining and construction activities appear to be key forces shaping this pattern. Following the large rise in FDI in mining activities in the early 1990s, there has been an increase in the number of large-scale mining and construction projects, leading to sustained employment growth (Robles 2010). While this in turn has promoted the settlement of internal migrants in mining regions (Rowe 2013a), it triggered large commuting flows due to the remote location of mining sites, unattractive social and environmental conditions of nearby cities and the use of FIFO and DIDO in mining operations, in combination with the temporary nature of construction activity. To assess the strength and significance of these connections, we now quantify the relative influence of the various factors described in Sect. 6.4.2 on longdistance commuting, using a linear regression model.

6.5.2

Determinants of Long-Distance Commuting

Three different model specifications were estimated. Table 6.3 and Fig. 6.5 report the regression results. Model 1 includes only individual labour market characteristics. Model 2 extends this model by adding demographic characteristics, household composition, housing attributes and migration history. In addition to these variables, Model 3 includes a set of geographical contextual factors. These models enable identification of the personal characteristics and geographical factors that lead workers to commute longer distances, assessing the relative influence of labour market, demographic, household composition, housing market, migration status and geographical context variables. Supporting our expectations, the results reveal that working in mining and construction activities are the most significant factors fostering long-distance commuting in Chile. Coefficients for mining and construction workers are consistently the largest across the range of variables in all three models. These coefficients in Model 1 indicate that working in mining and construction increases commuting distance by 130.8% and 43.7% relative to working in the trade sector. Including geographical contextual variables in Model 3 reduces these coefficients by 10%

138

F. Rowe and M. Bell

Table 6.3 Regression models of the natural logarithm of commuting distance in km Variable Model 1 Labour market characteristics Industry sector (Ref¼Trade) Agriculture 0.2002 Mining 1.3079 Manufacturing 0.0675 Utilities 0.2341 Construction 0.4365 Transport and communication 0.0782 Finance 0.0202 Business services 0.0749 Public administration 0.0095 Community and personal services 0.0674 Occupation (Ref¼Tradespeople) Managers 0.0227 Professionals 0.0540 Technicians 0.0799 Clerks 0.0207 Agriculture and fishery workers 0.0746 Craft workers 0.0587 Plant operators 0.0749 Labourers 0.0472 Armed forces 0.2511 Employment class (wage earner) Domestic worker 0.1046 Self-employed and employers 0.2614 Unpaid family workers 0.2057 Demographic characteristics Age Age squared Female Education (preschool or non-formal education) Primary education Secondary education Higher education Household composition (Ref¼Alternative household structure) One-earner household Two-earner household, without children Two-earner household, with children Housing tenure (rent-free occupier) Renter Owner-occupier

Model 2

Model 3

0.1742 1.1352 0.0680 0.2124 0.3361 0.0604 0.0122 0.0456 0.0266 0.0236

0.0159 0.5172 0.0188 0.1309 0.3091 0.0652 0.0842 0.1005 0.0521 0.0460

0.0560 0.0101 0.0479 0.0225 0.0324 0.0031 0.0168 0.0551 0.1418

0.0713 0.0022 0.0563 0.0525 0.0499 0.0367 0.0417 0.0871 0.1340

0.0523 0.2240 0.1636

0.0027 0.2377 0.2211

0.0094 0.0149 0.1139

0.0064 0.0094 0.1104

0.0018 0.0346 0.0706

0.0110 0.0623 0.1013

0.0373 0.0092 0.0123

0.0273 0.0207 0.0113

0.0396 0.0438

0.1056 0.0359 (continued)

6 The Drivers of Long-Distance Commuting in Chile: The Role of the Spatial. . .

139

Table 6.3 (continued) Variable Migration status (stayer) Migrant Geographical context Population density Job accessibility Centrality Mining specialisation Constant Summary statistics N R2 Adjusted R2 AIC

Model 1

Model 2

Model 3

0.0090

0.1161

2.4451

2.2823

0.1722 0.0336 0.4478 0.6257 2.4269

4408384 0.05 0.05 12695909

4298672 0.05 0.05 12046542

4298672 0.26 0.26 10972296

Note: Robust standard errors were used. Significance: p