Essays on internal migration and housing markets of urban India

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ESSAYS ON INTERNAL MIGRATION AND HOUSING MARKETS OF URBAN INDIA by Arnab Dutta A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (URBAN PLANNING AND DEVELOPMENT) May 2023 Copyright 2023 Arnab Dutta Dedication Dedicated to Baba, Ma, Bon, Choto Dida, and Sevanti ii Acknowledgments I want to take this opportunity to acknowledge everyone who has influenced me throughout my life; who made it possible for me to be where I am. At the outset, I want to mention that my Ph.D. journey at USC has been incredibly rewarding, both intellectually and on a personal level. I am especially grateful to the USC Price School and the Lusk Center for Real Estate for providing me with critical funds for my dissertation research. I thank my advisor and coauthor Richard Green for his contributions to my intellectual growth. Richard was precisely the advisor I needed when I started working with him in Oc- tober 2019. The fact that I was able to initiate several research projects in the last three years is almost entirely because of the encouraging environment of free thinking that Richard fosters. His kindness showed me that academic research could be a rewarding experience if we take care of those we work with. Richard made me a better scholar and a better person. My other committee members—Jorge De La Roca, Marlon Boarnet, and Matt Kahn—have been mentors to me in various ways. Jorge always helps me push my abilities. Whenever I discuss research with him, Jorge asks me questions that make me constructively rethink my approach. He is the person I can always rely on to help me with an identification issue. Marlon has been a constant source of support for me almost since the beginning of my Ph.D. His constant encouraging words keep me going. If I am about to lose all hope about pursuing a research idea, I can talk to Marlon, and he will renew my faith in the idea with a completely new perspective. Last but not the least, Matt will always have something for me to read. If I need to know whether there is an article on a research question, he will be the one I talk to first. Occasionally, when I am being lazy, Matt will check on me and ask me about my research progress. Without this amazing team, I would not have done half the work I did during my Ph.D. My dissertation will have been incomplete without my two friends-turned-coauthors, Sahil Gandhi and Greg Randolph. I owe a good share of my dissertation to them, not just because they are my coauthors in two chapters of this dissertation, but because I have also learned from them in various ways. They taught me a great deal about the existing research on migration and housing and on how to navigate the labyrinth of academics. Above all, they showed me how to be a better researcher. iii I thank Antonio Bento and Gen Giuliano for advising me during the early days of my Ph.D. A special thanks to Lisa Schweitzer, who was a source of support for me during a challenging time. I also thank my past advisors—Ashwini Chhatre, Chetan Ghate, Abhiroop Mukhopadhyay, and Venky Panchapagesan—for mentoring me and helping me get admitted into a Ph.D. program. Among those who have helped me complete this journey are the friends that I have made over the last several years. I thank all of them. But a few of them were particularly influential to me. I thank Rajat Kochhar and Ruozi Song for always willing to discuss research with me. I am lucky that I had Subhasish Sutradhar as a roommate and a friend. He helped me cultivate scientific and rigorous thinking. Moshiul Alam (Zubin) will never fail to support and help whenever he can. I learned mathematics and optimism from him. Sounak Thakur has been an academic and emotional mentor. I have seldom come across a person as caring and compassionate as him. He also taught me how to be mildly funny. A very special thanks to Nirvana Mitra. He is the reason I took a serious interest in pursuing an academic career back in college. Finally, I thank my childhood friend Arka Choudhury for instilling confidence and faith in me. Now, I want to talk about the five most influential people in my life—my parents, Manas Dutta and Krishna Dutta; my sister Ananya Dutta; my grandmother, Tripti Dutta; and my partner Sevanti Nag. Even though my parents had financial struggles, they never, for once, stopped prioritizing my education. They sacrificed in every possible way to pay for my tuition and books, even if they could not spend on personal necessities. My sister has been a source of joy for me from the very day she was born. She helped me get through my darkest days and continues to support me. My grandmother’s faith in me never waned. Even if the whole world disagrees with me, I can always rely on her to be on my side. Last but not the least, I am grateful to my partner and fiancée, Sevanti, for being with me throughout my Ph.D. Through thick and thin, she has constantly supported me in every possible way, doing chores, running errands, and teaching me social skills wherever I lack them. She is everything one can look for in a companion. I owe my Ph.D. in large part to these five people. iv Contents Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Chapter 1: Telecom Growth and Rural-urban Migration . . . . . . . . . . . . . . . . . . . . . 7 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 1.2 India’s telecom sector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 1.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 1.3.1 Aggregate data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 1.3.2 Microdata . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 1.3.2.1 National Sample Survey . . . . . . . . . . . . . . . . . . . . . . . . . 17 1.3.2.2 India Human Development Survey . . . . . . . . . . . . . . . . . . . 17 1.4 The impact of India’s telecom expansion on internal mobility . . . . . . . . . . . . . 18 1.4.1 District-level telecom growth and rural-urban migration . . . . . . . . . . . . 18 1.4.1.1 Extensive margins using difference-in-difference . . . . . . . . . . . 19 1.4.1.2 Intensive margins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 1.4.1.2.1 Using interactions terms . . . . . . . . . . . . . . . . . . . . 21 1.4.1.2.2 Continuous treatment . . . . . . . . . . . . . . . . . . . . . . 22 1.4.1.3 Inter-state and inter-circle migration . . . . . . . . . . . . . . . . . . 22 1.4.1.4 Parallel trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 1.4.1.4.1 Pre-treatment trends . . . . . . . . . . . . . . . . . . . . . . 24 1.4.1.4.2 Falsification tests . . . . . . . . . . . . . . . . . . . . . . . . 24 1.4.1.5 Summarizing the effects . . . . . . . . . . . . . . . . . . . . . . . . . 26 v 1.4.2 Telecom shock and individual mobility . . . . . . . . . . . . . . . . . . . . . . 27 1.4.2.1 Definition of distant telecom shock . . . . . . . . . . . . . . . . . . . 27 1.4.2.2 Rural-urban migration . . . . . . . . . . . . . . . . . . . . . . . . . . 28 1.4.2.3 Urban-rural migration . . . . . . . . . . . . . . . . . . . . . . . . . . 31 1.4.2.4 Net effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 1.5 Channels of effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 1.5.1 Rural-urban migration premium . . . . . . . . . . . . . . . . . . . . . . . . . . 34 1.5.1.1 Telecom shock-induced migration premium . . . . . . . . . . . . . . 35 1.5.1.2 Migration premium in the pre-telecom era . . . . . . . . . . . . . . . 37 1.5.1.3 Comparing migration premia . . . . . . . . . . . . . . . . . . . . . . 38 1.5.2 Improved information flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 1.5.3 Robustness checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 1.6 Heterogeneity in migrants destinations . . . . . . . . . . . . . . . . . . . . . . . . . . 45 1.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 1.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 Chapter 2: Distant Shocks, Migration, and Housing Supply . . . . . . . . . . . . . . . . . . 50 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 2.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 2.2.1 Informality of housing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 2.2.2 State-level data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 2.2.3 District-level data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 2.3 Stylized facts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 2.4 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 2.4.1 Spatial equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 2.4.2 Spatial disequilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 2.4.3 Mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 2.4.4 Housing demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 2.4.5 Housing supply . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 2.5 Empirical implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 vi 2.5.1 Defining instruments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 2.5.2 Distant shocks, inter-state migration, and urban population growth . . . . . 70 2.5.2.1 Empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 2.5.2.2 Covariates in estimation . . . . . . . . . . . . . . . . . . . . . . . . . 71 2.5.2.3 Effect of inter-state migration on local urban population growth . . 72 2.5.3 Urban population growth and housing demand . . . . . . . . . . . . . . . . . 74 2.5.3.1 Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 2.5.3.2 Effect of local urban population growth on housing demand . . . . 75 2.5.4 Demand shifters and housing supply elasticity estimation . . . . . . . . . . . 76 2.5.4.1 Estimating equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 2.5.4.2 Housing supply elasticity estimates . . . . . . . . . . . . . . . . . . . 78 2.6 Discussion on instruments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 2.6.1 Validity of instruments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 2.6.1.1 Relevance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 2.6.1.2 Endogeneity concerns . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 2.6.2 Robustness checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 2.6.2.1 Non-contiguous state shocks . . . . . . . . . . . . . . . . . . . . . . . 85 2.6.2.2 Long-term migration . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 2.7 Policy implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 2.7.1 State-level Elasticities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 2.7.2 Land use regulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 2.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 Chapter 3: Land Use Deregulation and Housing Supply . . . . . . . . . . . . . . . . . . . . 93 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 3.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 3.2.1 The ULC act . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 3.2.2 Land acquisition under the ULC act . . . . . . . . . . . . . . . . . . . . . . . . 98 3.2.3 Legal issues in the ULC repeal act . . . . . . . . . . . . . . . . . . . . . . . . . 99 3.2.4 Anecdotal evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 vii 3.3 Data and descriptives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 3.3.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 3.3.2 Data sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 3.3.3 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 3.4 Impact of ULC repeal on housing and non-residential supply . . . . . . . . . . . . . 107 3.4.1 Empirical implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 3.4.2 Decomposing the housing stock . . . . . . . . . . . . . . . . . . . . . . . . . . 109 3.4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 3.4.3.1 Formal housing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 3.4.3.2 Informal housing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 3.4.3.3 Non-residential stock . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 3.4.3.4 Redevelopment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 3.4.4 Robustness checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 3.4.5 Falsification tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 3.4.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 3.5 Legal proceedings in ULC enacting cities . . . . . . . . . . . . . . . . . . . . . . . . . 121 3.5.1 Legal challenges and housing markets . . . . . . . . . . . . . . . . . . . . . . 122 3.5.2 Estimating equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 3.5.3 Empirical estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 3.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 3.6.1 Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 3.6.1.1 Set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 3.6.1.1.1 Bid-rent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 3.6.1.1.2 Housing supply . . . . . . . . . . . . . . . . . . . . . . . . . 129 3.6.1.1.2.1 Land values outside the city . . . . . . . . . . . . . . 129 3.6.1.1.2.2 Building technologies . . . . . . . . . . . . . . . . . . 129 3.6.1.1.2.3 Informal housing production . . . . . . . . . . . . . 130 3.6.1.1.2.4 Formal housing production . . . . . . . . . . . . . . 130 3.6.1.1.3 Land development . . . . . . . . . . . . . . . . . . . . . . . 131 3.6.1.2 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 viii 3.6.1.2.1 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 3.6.1.2.2 Land development without regulations . . . . . . . . . . . 132 3.6.1.2.3 Land development under regulations . . . . . . . . . . . . 132 3.6.1.2.3.1 Defining the ULC act . . . . . . . . . . . . . . . . . . 133 3.6.1.2.3.2 Land development under the ULC act . . . . . . . . 133 3.6.1.2.4 Land development under the ULC repeal act . . . . . . . . 134 3.6.1.3 Reconciling model implications with empirical estimates . . . . . . 135 3.6.2 Implications for policies and future research . . . . . . . . . . . . . . . . . . . 136 3.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140 Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 Appendix A: Supplementary Figures and tables . . . . . . . . . . . . . . . . . . . . . . . . . 154 Appendix B: Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176 B.1 Urban administration in India . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176 B.2 Data processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178 B.3 ULC data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180 Appendix C: Derivations and proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182 ix List of Tables 1.1 Summary statistics of aggregate variables . . . . . . . . . . . . . . . . . . . . . . . . . 15 1.2 Pre-2000 balance test of aggregate variables . . . . . . . . . . . . . . . . . . . . . . . . 16 1.3 Characteristics of working-age individuals in the NSS sample survey data . . . . . . 18 1.4 Telecom auction effect on district-level telecom expansion and migration . . . . . . 20 1.5 Effect of telecom expansion on individual migration . . . . . . . . . . . . . . . . . . 30 1.6 Migration premia estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 1.7 Telecom expansion-induced follower migration . . . . . . . . . . . . . . . . . . . . . 41 1.8 Comparing coincident events’ impact on rural-urban migration . . . . . . . . . . . . 44 2.1 Summary statistics of state-level variables . . . . . . . . . . . . . . . . . . . . . . . . . 58 2.2 Summary statistics of district-level variables . . . . . . . . . . . . . . . . . . . . . . . 59 2.3 Distant shock-induced migration and local urban population growth . . . . . . . . . 73 2.4 Distant shock-induced local urban population growth and housing demand . . . . 76 2.5 Housing demand shifters and inverse supply elasticity estimation . . . . . . . . . . 79 2.6 Distant non-contiguous states’ shock-induced migration and local urbanization . . 86 2.7 Distant shock-induced long-term migration and local urbanization . . . . . . . . . . 88 2.8 State-level formal housing supply elasticities . . . . . . . . . . . . . . . . . . . . . . . 90 3.1 Summary statistics of city-level variables . . . . . . . . . . . . . . . . . . . . . . . . . 103 3.2 Characteristics of households surveyed by IHDS . . . . . . . . . . . . . . . . . . . . . 104 3.3 Balance test of city-level variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 3.4 ULC repeal impact on city-level formal residential housing . . . . . . . . . . . . . . 111 3.5 ULC repeal impact on city-level informal residential housing . . . . . . . . . . . . . 112 3.6 ULC repeal impact on city-level non-residential stock . . . . . . . . . . . . . . . . . . 114 3.7 ULC repeal impact on redevelopment from informal to formal housing . . . . . . . 116 3.8 Isolating JNNURM in the ULC repeal impact on city-level supply . . . . . . . . . . . 119 3.9 ULC act and land-related legal proceedings . . . . . . . . . . . . . . . . . . . . . . . . 126 A.1 Constituents and telecom operators by circles . . . . . . . . . . . . . . . . . . . . . . 161 A.2 Summary statistics of aggregate variables by treatment and control districts . . . . . 162 x A.3 Summary statistics of IHDS variables . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 A.4 NSS microdata subsamples and their experienced telecom shock . . . . . . . . . . . 164 A.5 Mobile phone ownership and migration among rural households . . . . . . . . . . . 165 A.6 Golden Quadrilateral and urban commodity prices . . . . . . . . . . . . . . . . . . . 166 A.7 Local urban population growth and housing demand without median rooms . . . . 167 A.8 Formal and vacant housing supply elasticity estimation with full sample . . . . . . 168 A.9 Housing supply elasticity estimation with non-contiguous state shocks . . . . . . . 169 A.10 Summary statistics of city-level variables for early repeal cities . . . . . . . . . . . . 170 A.11 Summary statistics of city-level variables for late repeal cities . . . . . . . . . . . . . 171 A.12 Summary statistics of city-level variables for no-ULC cities . . . . . . . . . . . . . . . 172 A.13 ULC repeal impact on city-level supply excluding Tamil Nadu . . . . . . . . . . . . 173 A.14 ULC repeal impact on city-level supply with decomposed late repeal treatment . . 174 A.15 ULC act and legal proceedings with 1971 population percentile fixed effects . . . . 175 B.1 Composition of UAs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177 B.2 UA-level data collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 B.3 State-level ULC repeal data collection . . . . . . . . . . . . . . . . . . . . . . . . . . . 181 xi List of Figures 1 Internal migration to cities of India (1991-2011) . . . . . . . . . . . . . . . . . . . . . . 2 2 Housing stock in urban India (2001-2011) . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.1 Number of telecom operators and rural-urban migration trends (1991-2005) . . . . . 8 1.2 Telecom spectrum auction bids (2001) . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 1.3 Rural-urban migration trends (1981-2005) . . . . . . . . . . . . . . . . . . . . . . . . . 25 1.4 Telecom growth impact on rural-urban migration by destination districts’ incomes . 46 2.1 Shock-induced migration’s impact on housing demand . . . . . . . . . . . . . . . . . 51 2.2 Inter-state migration and housing in urban India . . . . . . . . . . . . . . . . . . . . . 61 2.3 Share of decadal migrants in urban population of India . . . . . . . . . . . . . . . . . 62 3.1 City-level growth rates of houses and non-residential buildings (2001-2011) . . . . . 105 3.2 Land-related legal proceedings vs. 1971 city population quintiles . . . . . . . . . . . 123 A.1 Telecom circles in India (1996-2005) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 A.2 Map of Golden Quadrilateral recipient states in India . . . . . . . . . . . . . . . . . . 155 A.3 Map of India with geocoded cities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156 A.4 Impact of false telecom shock on local telecom operators in counterfactual districts 157 A.5 Impact of false telecom shock on rural-urban migration in counterfactual districts . 158 A.6 Density of p-values from false treatment I falsification tests . . . . . . . . . . . . . . 159 A.7 Density of p-values from false treatment II falsification tests . . . . . . . . . . . . . . 160 xii Abstract This dissertation explores how Indian cities’ housing supply responds to urban growth, which is increasingly driven by migration. The first chapter shows that technological growth-induced rural-urban migration accounted for a quarter of India’s urban population growth during the early 2000s. Rural residents moved to urban areas in response to India’s mobile phone service expansion because of higher urban wages and relatively seamless transmission of distant labor market information. The effects were, however, primarily concentrated in the wealthiest urban areas that drew the most migrants. The second chapter shows that droughts and highway in- frastructure investments induced interstate migration in India and increased housing demand in Indian cities during the 2000s. This chapter then estimates the slopes of urban India’s formal, in- formal, and vacant housing supply curves. The estimates indicate that while the formal housing supply in India is inelastic, developers engage in speculative construction as well. A negative informal housing supply elasticity estimate confirms the existence of gentrification. The third chapter shows that the repeal of India’s urban land ceiling laws, which allowed governments to acquire privately owned vacant land through ceiling limits, did not lead to an expected growth in formal housing supply during the 2000s. Disputes in the ownership of vacant land parcels led to legal battles between governments and landowners after the reform, which delayed the new construction of formal houses. The dissertation’s findings indicate that while Indian cities are expanding, their housing markets are not growing fast enough because of land use regulations and institutional frictions. xiii Introduction India is set to be the world’s largest country in2023, and its current urban population is estimated to be around half a billion, or 11% of the global urban population. 1 In fact, over the past three decades, the urban population of India has been growing at the decadal rate of 32%. 2 This growth in India’s urban population is increasingly accompanied by its growth in internal migration. Historically, India has had low levels of permanent mobility, which is an active topic of academic research (Bhavnani and Lacina, 2017; Kone et al., 2018; Munshi and Rosenzweig, 2016). But that seems to be changing in recent times as more Indians are moving towards and between cities. Figure 1a indicates that internal migration to cities of India, which includes those moving from rural and between urban areas, grew by more than 70% from 36 million to 62 million between the 1990s and the 2000s. Furthermore, figure 1b shows that a large portion of this growth comes from the movement of people from rural to urban areas. Are there enough houses in cities of India, growing increasingly through migration? A quick glance at figure 2a, which presents the stock of residential housing units in urban India in 2001 and 2011, tells us that urban India’s housing markets are growing as well. As with any de- veloping country, a large share of India’s housing stock consists of informal houses made from non-durable substances, such as thatch, mud, plastic, etc. Figure 2a indicates that about 15% of urban India’s housing stock in 2011 consisted of informal houses. The remaining portion of the overall stock consisted of formal houses built from more durable substances, such as concrete and metal. Figure 2b shows that, between 2001 and 2011, the total number of occupied residential hous- ing units went up by50%. A bulk of this growth was because of the growth in the formal housing stock by 61%. The informal housing stock, on the other hand, grew by less than 10%. At the same time, the number of vacant formal housing units increased by more than 80%. What do these growth patterns mean in the broader context of India’s urbanization? Does a slow growth in the informal housing stock combined with a rising vacant housing stock imply that the formal housing market supply is elastic? Or do these numbers hide deeper institutional frictions? 1 Most recent data, in 2021, from the World Bank puts the urban population of India at roughly 498 million. For details, visit: https://data.worldbank.org/indicator/SP .URB.TOTL. 2 In 2001, India’s urban population was 286 million, and in 2011, it was 377 million, a growth of 32%. 1 Figure 1: Internal migration to cities of India (1991-2011) 36 62 11 16 25 46 0 25 50 75 Migration (millions) All internal Inter−state Intra−state 1991−2001 2001−2011 (a) Rural-urban and urban-urban migration 21 30 6 8 14 22 0 25 50 75 Migration (millions) All internal Inter−state Intra−state 1991−2001 2001−2011 (b) Rural-urban migration Source: Authors’ calculations based on Census of India. Note: Figure presents internal migration patterns to urban India between 1991 and 2011. Panel (a) presents the total number of rural-urban and urban-urban migrants who moved to cities between 1991-2001 and between 2001-2011. Panel (b) presents only the number of rural-urban migrants who moved during the same time. Bars labeled with their corresponding values. 2 Figure 2: Housing stock in urban India (2001-2011) 52 78 41 66 11 12 6 11 0 25 50 75 100 Residential housing units (millions) All units Formal units Informal units Vacant units 2001 2011 (a) Housing stock 50 61 9 83 0 25 50 75 100 Residential housing stock growth (%) All units Formal units Informal units Vacant units (b) Housing stock growth Source: Authors’ calculations based on Census of India. Note: Figure presents the composition and growth of housing stock in urban India. Panel (a) presents the number of housing units, and panel (b) the growth in the number of units between 2001 and 2011. Formal housing units’ roofs and walls made of galvanized iron, metal, asbestos sheets, burnt bricks, stone, and concrete. Informal housing units’ roofs or walls made of grass, thatch, bamboo, plastic, polythene, mud, unburnt brick, and wood. All vacant houses are formal units. “All units” or overall stock does not include vacant formal units. Bars labeled with their corresponding values. 3 This dissertation aims to understand urban India’s housing markets. On the one hand, there clearly seems to be a surge in demand for housing, driven in large part by increasing migration, which merits investigation. On the other hand, examining the nature of housing supply in Indian cities is critical towards understanding the reason behind the slow growth in informal housing and the rising vacant housing stock. The first two chapters of the dissertation tackle the question of increasing migration towards urban areas of India, how such migration affects the demand for housing in cities, and how well cities’ markets supply housing in response to migration-induced demand. Three major factors have driven up internal mobility in India in recent times: technological growth, droughts, and highway infrastructure investments. As in much of the world, there was huge technological growth in India during the 2000s. The technological expansion was particularly successful in the telecommunications sector of India, as can be seen from the fact that mobile phone usage became widespread during the 2000s. By the end of that decade, India had more than 600 million mobile phone subscribers compared to just a few million in 2001 (Census of India, 2001b, 2011b). In the first chapter, coauthored with Gre- gory F. Randolph, I show that this growth in mobile phone usage led to about 12 million people moving from rural to urban areas. Mobility increased because, on the one hand, telecommuni- cations expansion led to a growth in the urban wage premia obtained by rural-urban migrants. On the other hand, the availability of mobile phones made it possible for a relatively seamless transmission of information about distant labor markets. At the same time, the benefits from telecommunications expansion may have been primarily concentrated in the affluent urban areas of India since rural-urban migrants were predominantly headed toward the wealthiest 20% of cities. The fact that climate shocks increase migration in India is well-known (Bhavnani and Lacina, 2017; Jayachandran, 2006; Rosenzweig and Stark, 1989; Rosenzweig and Udry, 2014). The second chapter’s empirical findings that droughts lead to more inter-state migration into urban areas essentially confirm the existing academic literature’s take on this issue. However, while recent research has shown that highway infrastructure investment led to the movement of firms to- wards regions that received highway investments (Ghani et al., 2016), a thorough investigation of such investments’ impact on the movement of people has not been done. The second chapter, 4 coauthored with Sahil Gandhi and Richard K. Green, examines this issue and concludes that re- gions that were beneficiaries of highway improvements both received and sent migrants. On the one hand, migration towards such regions increased because of firm concentration-induced labor demand shocks (Ghani et al., 2016). On the other hand, out-migration from such regions also increased because individuals were possibly earning more, and higher earners are better insured against risky migration outcomes (Bryan et al., 2014; Morten, 2019; Munshi and Rosenzweig, 2016). The second chapter further examines the responsiveness of the housing market supply in urban India by estimating the housing supply elasticity. The formal housing supply elasticity of urban India is estimated to be 1.62, slightly less than the national-level elasticity of 1.75 in the United States (Saiz, 2010). Given that economists consider the United States to be a largely supply-inelastic country (Berry and Glaeser, 2005; Glaeser and Gyourko, 2005; Glaeser et al., 2006; Green et al., 2005), an elasticity of 1.62 in India is lower still. At the same time, the supply of vacant formal units responds to rents more than occupied formal units, confirming earlier research suggesting that there was a lot of speculative construction in India during the 2000s (Gandhi et al., 2022). Finally, the informal housing supply elasticity is estimated to be -0.49. The negative informal elasticity indicates the existence of gentrification—slum houses are converted into formal commercial and residential buildings through demolitions (Bhan, 2009) and in situ redevelopments (Rains and Krishna, 2020; Rains et al., 2019). The low formal supply elasticity combined with a much higher elasticity of vacant houses indicates that there may be underlying regulatory and institutional frictions. The third chapter investigates how land use regulations and institutional frictions, such as ill-defined property rights and poorly-maintained transaction records, affect housing supply by conducting an event study of a large land use deregulatory reform implemented in India during the 2000s. The regulations studied consist of a set of urban land ceiling laws, which were enacted in 70 of the largest urban areas of India during the 1970s. These laws placed ceiling limits on privately owned vacant land. Excess vacant land was meant to be acquired and redistributed to the urban poor by governments. But, the laws did not achieve their objectives and were repealed during the 2000s. Even though the reform was expected to enhance housing market growth, the third chapter finds that the reform had no impact on the formal housing supply. This surprising 5 finding masks deep-seated institutional issues. Land acquisition under the urban land ceiling laws was slow, sometimes spanning decades. As a result, when the laws were repealed, a large number of land parcels were not fully acquired by governments. The repeal act had no provisions for what would happen with these land parcels in the acquisition process. This, in turn, led to confusion in the ownership of under-acquisition land parcels, and individual landowners and governments initiated legal battles to contest ownership rights over such parcels. As long as legal challenges over ownership rights of under-acquisition land continue, these parcels cannot be used for any new construction. And, settling land-related legal disputes in India could take a long time (Gandhi et al., 2021). Therefore, the benefits of land use deregulation under institutional frictions may only be realized in the long run. Even though the main motivation of this dissertation is to gain an insight into how the urban- ization processes have been unfolding in India, the individual chapters make theoretical contri- butions of their own to the existing academic literature. All three chapters also have implications for policy-making. These additional contributions are discussed within each chapter. Concluding remarks are provided at the end of the dissertation. 6 Chapter 1 Telecom Growth and Rural-urban Migration Arnab Dutta Gregory F. Randolph 1 1.1 Introduction Does the growth of virtual communication technologies enhance or diminish the pull of cities? While early reactions to the boom in telecommunication famously predicted the “death of dis- tance” (Cairncross, 1997), Gaspar and Glaeser (1998) argue that telecommunications technologies and face-to-face interactions in cities are complementary. They show that improving telecom- munications technology increases the need for interpersonal interactions, which outweighs the substitution of face-to-face interactions with virtual meetings. Storper and Venables (2004) echo this idea and theorize that urban concentration increases as an economy relies more on high-tech firms, which require face-to-face contact to innovate. Besides increasing the benefits of agglomeration to urban residents, technology can make cities more attractive to rural residents through rising urban wages, better opportunities, and hu- man capital accumulation (Lucas, 2004). However, the impact of improved telecommunications technologies on rural-urban migration in developing countries has not been studied extensively. Moreover, while the academic literature has explored the agglomeration benefits of technology in developed countries like the United States (US), we know less about how technology affects cities in developing countries like India. We address this literature gap by studying the impact of mo- bile phone infrastructure expansion (hereafter referred to interchangeably as telecom expansion, telecom growth, and telecom shock) on internal migration in India. In the early 2000s, dramatic growth in telecom infrastructure in India coincided with increas- ing internal migration. Figure 1.1 shows that the annual average number of mobile phone service providers (hereafter referred to as telecom operators) at the district level grew by 457%, from 0.7 to 3.9, between the 1990s and the early 2000s. 2 During the same time, inter-district rural-urban migration in India grew by 36%. The telecom boom was enabled by a 2001 auction that allocated 1 Ph.D. candidate in Urban Planning and Development, University of Southern California. 2 A district is an administrative unit in India similar to that of a county in the US. 7 Figure 1.1: Number of telecom operators and rural-urban migration trends (1991-2005) 0 1 2 3 4 5 Telecom operators and rural−urban migration 1991−2000 2001−2005 Telecom operators Migration (’000) 95% C.I. Source: Authors’ calculations based on Census of India and Sridhar (2007). Note: Figure presents the evolution of the number of district-level telecom operators and the stock of inter-district rural-urban migrants in India in the pre-2000 and post-2000 periods. Pre-2000 period is 1991-2000, and post-2000 is 2001-2005. Bars represent means, and vertical lines represent 95% confidence intervals. Stock of migrants living in a district in 2011 who moved during 1991-2000 and during 2001-2005 is used to construct the migration variable. Both the number of telecom operators and the migration stock figures are annualized. Migration figures are given in thousands. All regions had zero telecom operators before 1996. licenses to telecom operators bidding for the 1800 MHz band of spectrum. While telecom oper- ators bid for licenses to operate in most regions of India, parts of 13 states and union territories covering 112 districts did not receive any bid. We exploit the bid status of regions resulting from the auction in our identification strategy. We use data from the Census of India, the National Sample Survey Organization, and Srid- har (2007) to show that India’s telecom expansion between the 1990s and the early 2000s led to higher levels of internal migration, especially rural-urban migration. 3 Using district-level data in a difference-in-difference framework, we find that telecom growth spurred by the 2001 spectrum auction led to a 15% increase in inter-district rural-urban migration between the 1990s and the early 2000s. Since Census datasets potentially undercount migrants, we use individual-level data 3 The 1990s period covers the years 1991-2000 and the early 2000s covers the period 2001-2005. For brevity, we refer to the period 1991-2000 as pre-2000 and 2001-2005 as post-2000 throughout the rest of the chapter. The pre-2000 period includes the year 2000. 8 in a two-way fixed effects model to estimate the impact of telecom growth on migration. Our results show a net increase in rural-urban migration of about 12 million individuals, of which 4 million working-age individuals (aged 18-52 years at the time of the shock in the year 2000) moved for employment. Therefore, the telecom shock spurred mobility for purposes beyond em- ployment. Our estimates indicate that the telecom expansion-induced net rural-urban migration explains more than a quarter of India’s urban population growth during the early 2000s. Next, we find that rural-urban migrants who moved due to the telecom shock earned a 10% higher migration premium than rural-urban migrants who had moved before the shock. We also find that rural residents were more likely to move out in response to the telecom shock if their households had sent a former member out to another district before the shock. Building on previous literature (Lu et al., 2016; Munshi, 2003), we interpret that previous migrants act as sources of information to aid the migration process and that these information flows were improved by telecom expansion. These channels help to explain the growth of rural-urban labor migration resulting from the telecom shock. We conduct some robustness checks and show that other confounding events, such as high- way investments and international trade shocks coinciding with the telecom shock, do not explain the effect of telecom on migration. Furthermore, we find an association between the ownership of mobile phones and out-migration among rural residents. And finally, we show that the wealth- iest 20% of urban areas in India received the most migrants as a result of the telecom shock. Hence, the impact of telecom growth on rural-urban migration primarily arises from the pull effect of a select number of cities that were already wealthy. Drawing from quantitative and qualitative literature on migration and labor, we posit that the relationship between telecom growth and internal migration operates through multiple channels. First, telecom expansion acts as a positive labor demand shock in potential destinations of mi- grants, inducing labor mobility (Topel, 1986). Individuals would move from rural to urban areas with the expectation of a wage premium (Harris and Todaro, 1970; Todaro, 1969). The urban pull effect of telecom would be bolstered by better opportunities in urban areas and the possibility of human capital accumulation by rural-urban migrants working in cities (Lucas, 2004). Second, telecom expansion can make migration less risky. Information asymmetry produces friction in the spatial allocation of labor (Katz and Stark, 1986). People may be less likely to mi- 9 grate when they are uncertain of conditions in distant labor markets, such as wages, availability of work (Lall and Selod, 2006), cost of living (Schultz, 1982), cultural and linguistic obstacles (Kone et al., 2018), or their ability to access social welfare programs (Imbert and Papp, 2020). Virtual communication technologies can enable migrants to gather more information about dis- tant labor markets prior to leaving home, reducing uncertainties in migration outcomes. This channel could increase migration in both directions between rural and urban areas, given that improved information flows also enable urban dwellers to better understand rural labor markets (Boas, 2020; Young, 2013). We would expect the expansion of information via mobile phones to enable the most growth in migration to places where friends and family members are already living since people tend to move along established migration routes (Munshi, 2003). Indeed, previous empirical studies have shown that growth in access to telecommunications can spur out-migration from rural areas through access to labor market information and contact with migrants’ family members (Aker et al., 2012; Lu et al., 2016; Muto, 2009). A third channel of effect may be that telecom expansion enables migrants to retain their attachments to home, reducing an “affective” barrier to migration. Qualitative evidence shows that information and communication technologies enable people to be highly mobile while still maintaining their place attachments (Boas, 2020; Gustafson, 2001, 2014; Skop and Adams, 2009). As telecom expands, the inframarginal migrant who was earlier reluctant to move due to loss of contact with loved ones may become more comfortable with out-migration. And finally, given that the telecom shock affected non-employment mobility more than la- bor migration, there could be other non-economic channels, such as the movement of women from their rural ancestral homes to their spouses’ homes in urban areas after marriage (Rao and Finnoff, 2015; Rosenzweig and Stark, 1989). Our empirical findings contribute to three strands of academic literature. First, our results relate to a classic economic geography question dealing with how technological change reshapes the balance between forces of dispersion and concentration, helping to further disprove the Cairncross (1997) “death of distance” hypothesis and to substantiate claims made by Gaspar and Glaeser (1998) and Storper and Venables (2004), who argue that technology and face-to-face interactions are complementary. Based on our results, technology appears to enhance the pull of 10 cities rather than diminish the benefits of agglomeration. Second, we show that the telecommunication growth enhanced the pull of the wealthiest urban areas in India. This is in contrast to what recent literature has shown in the case of developed countries like the US, where there has been a decline in migration to high-income cities with technological growth (Ganong and Shoag, 2017). Finally, we contribute to the existing literature on internal migration in India. Previous studies have shown that, historically, India has had some of the lowest levels of permanent internal migration in the world because of social networks in villages (Munshi and Rosenzweig, 2016), social protection programs in rural areas (Imbert and Papp, 2020), regionally exclusive social insurance schemes (Kone et al., 2018), fiscal federalism (Bhavnani and Lacina, 2017), and urban housing regulations (Dutta et al., 2022). Recent studies have shown that highway investments (Dutta et al., 2021a) and the IT boom during the 1990s and 2000s led to higher internal migration in India (Ghose, 2021). This chapter contributes to this new body of literature that examines the recent growth in internal mobility in India. More broadly, the implications of our findings also connect to a bigger debate about the relationship between internal migration and spatial inequality, a point we get back to in section 1.7. We organize this chapter as follows. In section 1.2, we provide a brief background of the telecommunication sector of India, followed by a discussion of the data in section 1.3. In sec- tion 1.4, we show the impact of telecom expansion on migration; in section 1.5, we explore the channels of effect of telecom growth on migration. After investigating migrants’ top destinations in section 1.6, we discuss the implications of our findings in section 1.7 and provide concluding remarks in section 1.8. 1.2 India’s telecom sector India’s telecom infrastructure growth began with a series of reforms during the 1990s. Under the central government, the Department of Telecommunications (DOT) divided India into 23 geographically distinct regions called “circles” in the National Telecom Policy of 1994. These circles are roughly equivalent to state boundaries and were categorized into four zones based 11 on their revenue-generating potential at the time of inception. 4 In figure A. 1, we show a map of India with the 23 circles demarcated and labeled. Mobile phone service providers were allocated licenses to operate within each circle based on first-price sealed-bid auctions (Chattopadhyay and Chatterjee, 2014). The auction to allocate licenses for operating in the 1800 MHz band of spectrum in India was held in 2001. The introduction of the higher band of spectrum was an improvement on the 800 MHz band of spectrum available in India before 2001. The DOT implemented a three-stage auction, wherein each service provider could bid for service provision rights in a circle at each stage and continue or decline their bids after observing the ongoing bid price at the beginning of the next stage (Sridhar, 2007). While 18 of the 23 circles received a bid in the 2001 auction, most parts of 13 states and union territories (UTs) covered by the circles of Assam, Bihar, North East, Orissa, and West Bengal did not receive any bid. 5 In figure 1.2, we show a telecom circle map of India marking regions that received a bid and regions that did not receive any bid in the auction. The reason for these circles not receiving bids could be manifold, including the uncertainties associated with revenue generation, the social and political climate, and the cost of setting up businesses in these circles. The spectrum auction of 2001 was, in many ways, a pivotal moment for India’s telecommuni- cation infrastructure growth. Not only did services improve with the higher band of spectrum, but the average number of telecom operators providing services in a circle went up from about 0.7 during 1991-2000 to 3.9 in the period 2001-2005, a growth of 457% (see table A.1). The telecom expansion continued through the rest of the 2000s, as is evident from the fact that the share of landline and mobile phone users in India went up from 3% in 2001 to 57% in 2011, equivalent to 685 million additional subscribers (Census of India). 6 We exploit this pivot in India’s telecom 4 Uttar Pradesh is divided into two smaller circles, and smaller states in the North-east were combined to form one circle. The four largest metropolitan cities of Delhi, Mumbai, Kolkata, and Chennai, had their own telecom circles. See table A.1 for more details on the composition of circles. The Chennai circle was merged with the Tamil Nadu circle after 2005, leading to 22 circles thereafter. 5 With the exception of the Kolkata metropolitan circle, no region in the states and UTs of the Andaman and Nicobar Islands, Arunachal Pradesh, Assam, Bihar, Jharkhand, Manipur, Meghalaya, Mizoram, Nagaland, Orissa, Sikkim, Tripura, and West Bengal received a bid. 6 The telecom expansion during the second half of 2000s can be partly attributed to auctions held after 2005, which includes the 2G spectrum auction of 2008. The auction of 2008 ended up in a major scandal of bribery and corruption, which implicated several high-profile central government officials and politicians. Popularly known as the 2G scam, the press widely covered it at the time. For example, see this New York Times article for details: https://www.nytimes.com/2010/12/14/world/asia/14india.html. 12 Figure 1.2: Telecom spectrum auction bids (2001) No data Jammu & Kashmir Himachal Pradesh Punjab Haryana Delhi Rajasthan Uttar Pradesh (W) Uttar Pradesh (E) Bihar North East Assam West Bengal Kolkata Orissa Madhya Pradesh Gujarat Maharashtra Mumbai Andhra Pradesh Karnataka Kerala Tamil Nadu Chennai Did not receive any bid Received a bid No data Data source: Sridhar (2007). Note: Figure presents a map of India with the 23 telecom circles demarcated and labeled with their respective names. The circles of Assam, Bihar, North east, Orissa, and West Bengal, covering 112 districts in our analysis, did not receive any bid in India’s 2001 auction of the 1800 MHz band of spectrum. Remaining circles covering 307 districts received at least one bid. expansion by comparing the pre-2000 period’s growth with the growth in the post-2000 period in our analysis. Table A.1 shows the constituent districts and the average number of telecom operators in each of the 23 telecom circles in India during the pre- and the post-2000 periods. 1.3 Data We draw data from the National Sample Survey Organization (NSS) and the Census of India to conduct our analysis of the impact of telecom expansion on migration. Our analysis consists of 13 two major components. First, we use aggregated figures from the Census to conduct a district- level analysis. And second, we analyze individual-level data from the sample survey conducted by the NSS during 2007-2008. We use annual data on the number of telecom service providers by different circles in India during 1996-2006 from Sridhar (2007). In addition, we use data from other sources to conduct robustness checks. 1.3.1 Aggregate data The Census of India provides aggregated decennial in-migration figures at the district level. These aggregated figures provide information about migrants’ time of movement (i.e., less than a year ago, 1-4 years ago, 5-9 years ago, and so on), the distance migrants traveled from their last place of residence (inter-district, inter-state, etc.), their sector of origin (rural or urban), and their current place of residence (rural or urban). For our analysis, we use data on the aggregate number of rural-urban migrants who moved to urban areas of a district from rural areas of all other districts in India between 1991 and 2005. We construct a panel of 419 districts across India observed during 1991-2000, which we call the pre-2000 period, and again during 2001-2005, or the post-2000 period. There are variations in the numbers of districts in the original datasets. District boundaries in India change consid- erably over time due to the realignment of existing districts’ boundaries to carve out new ones. Between 2001 and 2011, the number of districts went up from 593 to 640. We create 479 hypo- thetical districts with time-consistent boundaries by combining all contiguous districts affected by boundary changes and leaving districts unaffected by boundary realignment unchanged. We lose 60 districts due to missing data. Out of this new sample of 419 districts, 112 belong to circles that did not receive a bid in the 2001 auction, and 307 belong to circles that received a bid. Using data from the Census, we construct annualized migration figures for pre- 2000 based on the stock of inter-district rural-urban migrants who were living in a district in 2011 and had moved during 1991-2000; for post-2000, we use the stock of migrants who moved during 2001-2005. We use district-level data on population, employment, and literacy from Census 2011 for post-2000 and data from Census 2001 for pre-2000. 7 The NSS consumption and employment 7 Since we use some of these additional variables as controls in our aggregate analysis, the mismatch in the time frame of migration figures ( 2001-2005) and the demographic variables (2001-2011) for post-2000 is not a major threat to our analysis. We report results from regressions with and without the controls in table 1.4 and find that the results 14 Table 1.1: Summary statistics of aggregate variables Pre-2000 Post-2000 Mean SD Mean SD Variable (1) (2) (3) (4) Local telecom operators 0.67 0.22 3.95 1.02 Annualized inter-district rural-urban migrants ('000) 2.60 8.35 3.53 11.18 Urban population (millions) 0.67 1.15 0.89 1.44 Share literate urban 0.68 0.08 0.73 0.07 Share employed urban 0.31 0.04 0.34 0.04 Labor demand shock 9.39 0.22 10.35 0.24 Mean monthly per capita consumption (INR) 959 261 1,016 338 Mean weekly wages (INR) 185 114 186 98 GQ 0 0 0.19 0.39 N 419 419 419 419 Source: Authors’ calculations based on Census of India, Labor Bureau of India, National Sample Survey Organization, and Sridhar (2007). Note: Table presents summary statistics of district-level aggregate variables for a panel of 419 districts. The pre- 2000 period is between 1991-2000, and the post-2000 period is 2001-2005. Migration figures are given in thousands. Population figures are given in millions. Literate and employed shares are the shares of urban residents in a district who are literate and employed, respectively. Consumption and wage earning figures based on urban households. Labor demand shock is calculated by first multiplying the 2001 share of employment in four sectors–—agriculture, manufacturing, mining and construction, and services—–with the log volume of the corresponding sector’s exports (in USD) in 2001 (for pre-2000) and 2005 (post-2000), and then adding the products over the four sectors for each period. GQ is a dummy equal to one if the district was a recipient of the Golden Quadrilateral highway upgrade project after 2001, and zero otherwise. Consumption and wage values rounded off to the nearest integers. All other values are rounded off to two decimal places. Real mean monthly per capita consumption and mean weekly wages in 2001 Indian National Rupees (INR) are calculated using the Consumer Price Index (CPI) data from the Labor Bureau of India. In PPP terms, $1 = 10 INR in 2001. For exchange rates, see: https://data.oecd.org/conversion/purchasing- power-parities-ppp.htm. surveys enable us to calculate district-level mean monthly per capita consumption and mean weekly wage earning figures for the pre- and post- 2000 periods. We obtain data on the Golden Quadrilateral highway upgrade coverage from Ghani et al. (2016) while using data from the DGCIS (2020) and World Trade Organization (2016) to compute district-level economic shocks. We calculate all aggregated variable values for urban areas of a district. And finally, we allocate each district in our analysis to one of the 23 telecom circles in India and calculate the population-weighted average number of telecom operators for each district for the given period. The summary statistics of all aggregated variables for the full sample are given in table 1.1. Detailed summary statistics by samples of treatment and control districts are presented in table A.2. in both cases are similar. 15 Table 1.2: Pre-2000 balance test of aggregate variables Pre-2000 samples of districts that Received a bid Did not receive a bid Difference (1)-(2) Variable (1) (2) (3) Local telecom operators 0.77 0.39 0.38 (0.01) (0.01) [0.00] Log annualized inter-district rural-urban migration 7.08 6.24 0.84 (0.07) (0.11) [0.00] Log urban population 13.04 11.98 1.06 (0.06) (0.10) [0.00] Share literate urban 0.68 0.68 0.00 (0.00) (0.01) [0.94] Share employed urban 0.31 0.30 0.01 (0.00) (0.00) [0.02] Log consumption urban 6.84 6.79 0.05 (0.02) (0.03) [0.10] Log weekly wage urban 5.03 4.92 0.11 (0.04) (0.10) [0.30] N 307 112 Source: Authors’ calculations. Note: Table presents two-sample t-test statistics for pre-2000 values of district-level aggregate variables. The treatment group of districts received a bid in India’s 2001 auction of 1800 MHz spectrum band, and the control group did not. Variable means presented in columns (1) and (2), and mean differences between treatment and control groups in column (3) without parentheses or brackets. Standard errors are given in parentheses below variable means in columns (1) and (2). The p-values of two-sample t-tests are given in brackets below mean differences in column (3). All values are rounded off to two decimal places. In table 1.2, we present the pre-2000 balance tests for district-level variables. We observe that districts that received a bid in the 2001 spectrum auction were significantly more populated with higher volumes of rural-urban migrants, had marginally higher shares of employed individu- als, and had higher numbers of telecom operators in the pre-2000 period. However, districts that did and did not receive a bid were almost identical in terms of literacy, per capita con- sumption, and wage earnings—factors that matter more for telecom shock-induced migration. In section 1.4.1.4.2, we show that the volume of migrants in districts that received a bid increased more in the post-2000 period. 16 1.3.2 Microdata We use individual- and household-level data from the National Sample Survey and the India Human Development Survey to conduct our microdata analysis. 1.3.2.1 National Sample Survey The National Sample Survey Organization (NSS) conducted an employment and migration sur- vey of 672,901 individuals belonging to 125,578 households between July 2007 and June 2008. The survey asked individuals whether they moved to their present location at some point in the past, and if they moved, their reason for movement (employment, education, etc.), their year of movement, and their place of last residence (state and sector, rural or urban, of origin). The survey reports details like the gender, age, education, employment status, and earnings of all individuals. In addition, the survey records whether the surveyed individuals lived in rural or urban areas. We use the NSS survey data to identify subsamples of individuals who never migrated or were continuously living in rural or urban areas since the year 2001, rural-urban migrants who moved before and after 2000, urban-rural migrants who moved after 2001, and out-migrants who left rural areas before and after 2001. In table A.4, we provide a breakdown of the subsample sizes based on the age of migrants, migrants’ reason for movement (employment or others), the time of movement, and the size of the received telecom shock (based on the number of operators). We use these subsamples to construct panels of individuals whose locations are observed in the year 2000 and the post-2000 period. 8 In addition, we use information on wage earnings, age, gender, and educational qualifications of labor migrants to conduct our analysis of migration wage premia. The summary statistics of these variables are provided in table 1.3. 1.3.2.2 India Human Development Survey We use household-level survey data from the first wave of the India Human Development Survey (IHDS) conducted by the University of Maryland and the National Council of Applied Economic Research during November 2004-October 2005 (Desai et al., 2015). The IHDS I survey records 8 We exclude individuals for whom we did not have any documented migration histories, those born after 2000, return migrants, and households that migrated in the past year. 17 Table 1.3: Characteristics of working-age individuals in the NSS sample survey data Rural-urban labor migrants Rural resident Moved 2001-2005 Moved before 2001 non-migrants Mean SD Mean SD Mean SD Variable (1) (2) (3) (4) (5) (6) Daily wage 151 168 156 216 33 79 Female dummy 0.08 0.28 0.06 0.25 0.34 0.47 Age at the time of survey 34 8 42 9 40 10 School graduate and above 0.45 0.50 0.40 0.49 0.19 0.39 Years spent in urban areas 4 1 19 8 0 0 N 1,050 1,050 2,712 2,712 96,116 96,116 Source: Authors’ calculations based on National Sample Survey Organization. Note: Table presents characteristics of working-age individuals in the employment and migration data from the NSS survey conducted during July 2007-June 2008. Individuals are divided into three subsamples. The first subsample (columns (1) and (2)) consists of inter-district rural-urban migrants who moved between 2001 and 2005. The second subsample (columns (3) and (4)) consists of inter-district rural-urban migrants who moved any time before 2001. The third subsample (columns (5) and (6)) consists of individuals who were living in rural areas in the year 2000 and did not migrate out after 2000. All individuals in the three subsamples were between the ages of 18 and 52 in 2000. Daily wage figures are based on earnings observed at the time of the survey during 2007-2008. Years spent in urban is equal to the number of years since a rural-urban migrant moved to an urban area. The female dummy and school graduate dummy variables are rounded off to two decimal places. Daily wage, age, and years spent in urban areas rounded off to the nearest integer values. In PPP terms, $1 = 12 INR during 2007-2008. For exchange rates, see: https://data.oecd.org/conversion/purchasing-power-parities-ppp.htm. information about non-resident members of26,472 rural households. The survey also asks house- holds whether they own mobile phones. We use this data to examine the link between mobile phone ownership and migration. The summary statistics of IHDS variables are given in table A.3. 1.4 The impact of India’s telecom expansion on internal mobility In this section, we estimate the impact of telecom growth on inter-district migration using ag- gregate and microdata. First, we estimate the impact of telecom expansion on the stock of inter- district rural-urban migrants using district-level data. Then, we estimate the impact of telecom growth on net rural-urban migration using microdata. 1.4.1 District-level telecom growth and rural-urban migration We estimate the impact of telecom expansion on the stock of inter-district rural-urban migrants who moved before and after 2001 and were living in their destination districts in 2011. We use a difference-in-difference strategy in a two-way fixed effects framework exploiting the 2001 auction 18 of the 1800 MHz band of spectrum in which 112 districts out of 419 did not receive a bid. In the following sections, we discuss the identification strategy and the results. 1.4.1.1 Extensive margins using difference-in-difference The two dependent variables of interest are the log of inter-district rural-urban migrant stock and the number of telecom operators in district j at time t. 9 For our treatment, we define a dummy variable Auction jt , which is equal to one for the post-treatment period if district j received a bid in the 2001 telecom spectrum auction, and zero otherwise. Hereon, we also refer to Auction jt as the auction bid dummy. Our goal is to estimate the impact of receiving a bid in the auction on the number of telecom operators and the stock of rural-urban migrants in a district. We can set this as a difference-in-difference estimation problem using the following two-way fixed effects (TWFE) equation: y jt = β Auction jt +κ j +η t +ϵ jt (1.1a) where, y jt = {Operators jt , Log Migration jt }, κ j and η t are district fixed effects and the time trend, respectively, and ϵ jt is the error term. The coefficient β is the average treatment effect of the spectrum auction. We present the estimates of the auction bid dummy coefficient from equation ( 1.1a) in columns (1) and (2) of table 1.4. First, districts that received a bid in the 2001 spectrum auction had, on average, 1.2 additional telecom operators in the post-2000 period compared to districts that did not receive any bid. The average number of telecom operators in districts that did not receive a bid in the 2001 auction went up by 2.4 between the pre- and post-2000 periods (see panel (b) of table A.2), implying that receiving a bid in the 2001 auction accounted for 50% of the growth in telecom operators in India during this time. Next, districts that received a bid witnessed 5.3 percentage points higher growth in the stock of annualized inter-district rural-urban migrants than districts that did not receive a bid. The average control district’s stock of annualized inter-district rural-urban migrants in India increased by 20% between the pre- and post-2000 periods (see panel (b) of table A.2). This implies that the 9 We annualize the stock of migrants and the number of telecom operators for the given period. 19 Table 1.4: Telecom auction effect on district-level telecom expansion and migration Dependent variable = Telecom operators and log migration Log migration Operators Inter-district Inter-circle (1) (2) (3) (4) (5) (6) Auction 1.202*** 0.053*** -0.169** -0.124* -0.064 (0.068) (0.017) (0.070) (0.065) (0.119) Auction× Operators 0.051*** 0.045*** 0.052** (0.016) (0.014) (0.026) Operators 0.044*** (0.011) Log urban population 0.290*** (0.076) Share literate urban 0.101 (0.535) Share employed urban 0.039 (0.672) Log consumption urban -0.031 (0.024) District fixed effects ✓ ✓ ✓ ✓ ✓ ✓ Time trend ✓ ✓ ✓ ✓ ✓ ✓ N 838 838 838 838 838 838 R 2 0.959 0.723 0.734 0.768 0.732 0.594 Source: Authors’ calculations. Note: Table presents two-way fixed effects regression estimates using a balanced panel of 419 districts observed during 1991-2000 or the pre-2000 period and during 2001-2005 or the post-2000 period. Dependent variable in column (1) is the average number of telecom operators in a district during each period. Dependent variable in columns (2)-(5) is the log of annualized stock of inter-district rural-urban migrants who were living in a district in 2011 and had moved during the pre- and post-2000 periods from rural areas of all other districts of India. Dependent variable in column (6) is the log of annualized stock of inter-district rural-urban migrants living in a district in 2011 who had moved from rural areas outside the state and the telecom circle in which the district was located. Auction is a dummy equal to one if a district’s telecom circle received a bid in India’s 1800 MHz spectrum band auction held in 2001, and zero otherwise. Standard errors are clustered at the district level and reported in parentheses. * p< 0.10, ** p< 0.05, *** p < 0.01. spectrum auction of 2001 led to a 26% increase in rural-urban migration in India during this time. 10 We use equation (1.1a) to estimate the extensive margin of the pull effect of telecom growth in drawing rural residents to Indian cities. However, the number of telecom operators grew across both treatment and control districts of India between pre- and post-2000 (table A.2). So, what was the intensive margin of the pull effect of telecom growth? In other words, what was the 10 Control districts’ mean stock of annualized inter-district rural-urban migrants increased from 1,074 to 1,293 be- tween pre- and post-2000, which is a growth of 20.4%. We divide the treatment effect of 5.3 percentage points by the baseline growth of 20.4% to get the figure of 26% 20 marginal impact of an additional telecom operator in urban areas on rural-urban migration? 1.4.1.2 Intensive margins We estimate the intensive margin of the pull effect of telecom growth on inter-district rural- urban migration to Indian cities in two ways. First, we modify equation (1.1a) by including an interaction of the auction bid dummy and the number of telecom operators in a district. This provides us with the intensive margin of the pull effect of telecom growth in districts that received a bid in the2001 spectrum auction. And second, we estimate the marginal impact of an additional telecom operator on inter-district rural-urban migration by treating the number of operators as a continuous variable. We present these estimation results in the following subsections. 1.4.1.2.1 Using interactions terms We include an interaction of the auction bid dummy and the number of telecom operators in equation (1.1a) and rewrite it as follows: Log Migration jt = α 0 Auction jt +α 1 Auction jt × Operators jt +σx jt +κ j +η t +ϵ jt (1.1b) where, x jt consists of a vector of covariates that affect rural-urban migration to j such as popu- lation, employment, literacy, and consumption, κ j and η t are district fixed effects and the time trend, respectively, and ϵ jt is the error term. The coefficient α 1 provides us with the marginal impact of an additional telecom operator on inter-district rural-urban migration to districts that received a bid in the 2001 auction. We present the estimates of α 1 with and without the vector of controls, respectively, in columns (3) and (4) of table 1.4. The estimates with and without controls are very similar. Controlling for population, literacy, employment, and consumption, districts that received a bid experienced 4.5% growth in the stock of inter-district rural-urban migrants for every additional telecom operator (column (4) of table 1.4). The estimate without controls is slightly higher at 5%. We also see that the estimate of α 0 or the impact of the auction bid dummy on inter-district rural-urban migrant stock is negative in columns (3) and (4) of table 1.4. This implies that the treated districts would have experienced a decline in migration if the number of operators was 21 set to zero. In other words, the positive impact of the auction on migration is due to the growth in the number of telecom operators. 1.4.1.2.2 Continuous treatment We estimate the marginal impact of an additional telecom operator in the full sample of districts with the following regression: Log Migration jt = α ′ Operators jt +κ j +η t +ϵ jt (1.1c) where, κ j and η t are district fixed effects and the time trend, respectively, and ϵ jt is the error term. The coefficient α ′ provides us with the marginal effect of an additional telecom operator on migration in the full sample. We provide the estimate of α ′ in column (5) of table 1.4. Every additional telecom operator increased the stock of inter-district rural-urban migrants in a district by 4.4%. This implies that the average treatment effect of telecom operator growth on rural-urban migration in the entire sample of districts is about 15%. 11 This result indicates that the telecom growth in India during the early 2000s led to an increase in rural-urban migration across the board. The average treatment effect estimate in column (5) of table 1.4 is only slightly lower than that in column (4). However, the number of telecom operators grew faster in districts that received a bid compared to the full sample of districts. 12 Hence, the effect of telecom expansion on migration was higher by two percentage points among districts that received a bid in the 2001 auction compared to the full sample of districts. 13 1.4.1.3 Inter-state and inter-circle migration There is a problem with interpreting the impact of the telecom shock on rural-urban migration in equation (1.1b) as a pull effect. A telecom circle, on average, encompasses 28 districts in India. Even though it is unlikely that the impact of the number of telecom operators in urban and 11 We get this figure by multiplying the estimate of 4.4% with 3.3, which is the average change in the number of operators in the full sample of districts (see table 1.1). 12 Districts that received a bid had 3.6 additional telecom operators in the post-2000 period (panel (a) of table A.2). In the full sample of districts, operator growth was 3.3 (see table 1.1). 13 Multiple 3.6 and 3.3 with 4.5% and 4.4% to get telecom expansion’s gross impact of 16.2% and 14.5%, respectively, on the rural-urban migrant stock in the sample of bid districts and the full sample. 22 rural areas would be exactly the same, migrants moving from one district’s rural area to another district’s urban area within the same circle would have experienced similar levels of the telecom shock in their origin and their destination districts. 14 Unfortunately, our data does not allow us to directly disentangle the shock’s push and pull effects within a circle. Hence, we run regressions after changing the definition of migration in equation (1.1b) to include only inter-district rural-urban migrants who moved outside their state and circle-of-origin. 15 Since there is considerable heterogeneity in the number of telecom operators across circles of India in the post-2000 period (see table A.1), the estimates we get by restricting migration to inter-state and inter-circle movements come closer to isolating a pull factor impact than the migration variable that includes all inter-district rural-urban migrants. The impact of telecom operator growth on the stock of inter-state and inter-circle rural-urban migrants in districts with an auction bid is given in column (6) of table 1.4. The inter-state and inter-circle rural-urban migrant stock grew by 5.2% for every additional telecom operator in treated districts. This effect is almost identical to the marginal impact of an additional operator on overall (intra- and inter-state combined) inter-district rural-urban migration among treated districts (column (3) of table 1.4). This finding indicates that the telecom shock affected inter- district rural-urban migration primarily through the urban pull channel, and any rural push effect of the telecom shock within a district on rural-urban migration was negligible. 1.4.1.4 Parallel trends We must satisfy the parallel trends assumption to get consistent estimates of α and β using the difference-in-difference estimation set-up in equations (1.1a) and (1.1b). A recent body of literature has highlighted the problems with the TWFE estimation of difference-in-difference co- efficients when the treatment effect is not equal across groups and time, which can happen when the treatment is staggered (Athey and Imbens, 2022; Baker et al., 2022; Callaway and Sant’Anna, 14 The growth in telecommunication technologies in urban and rural areas within a circle would be a function of pre-existing infrastructure. Telecom infrastructure in most regions of India was clearly better in urban than in rural areas before the spectrum auction. This is evident from the fact that while about 19% of the urban population in India had access to landline phones in 2001, less than 5% of the rural population had a landline connection. 15 As an example, consider the state of Tripura, which is part of the North-East circle along with five other states in India. In our new definition of the migration variable, individuals would be considered migrants only if their desti- nations are both outside the state of Tripura and the entire circle of North East that includes the states of Arunachal Pradesh, Manipur, Meghalaya, Mizoram, and Nagaland. 23 2021; De Chaisemartin and d’Haultfoeuille, 2020; Goodman-Bacon, 2021). 16 However, satisfy- ing the parallel trends assumption should be enough to get consistent difference-in-difference estimates in our case because our set-up involves a binary treatment and all treated districts re- ceived the treatment at the same time. The parallel trends assumption in our case would imply that districts that received a bid in the 2001 spectrum auction would have had a similar trend in the number of telecom operators and the stock of rural-urban migrants as districts that did not receive any bid. 17 1.4.1.4.1 Pre-treatment trends We present the trend of annualized rural-urban migration in India between 1981-2005 by districts that received a bid and districts that did not receive any bid in figure 1.3. Even before the 2001 auction, districts that received a bid had, on average, almost three times more inter-district rural- urban migrants than districts that did not receive a bid. However, districts that did and did not receive a bid had similar rural-urban migration growth patterns between the 1980s and the 1990s. Rural-urban migration to districts that received a bid went up by 28% and by 22% to districts that did not receive a bid between the 1980s and the 1990s. 18 In contrast, districts that received a bid witnessed a 34% growth in rural-urban migration between the 1990s and the early 2000s, whereas, districts without a bid saw a decline in the rural-urban migration growth rate from 22% to 18% during the same time. 19 These figures indicate that the growth rate of migration between the two regions, those that did and did not receive a bid, diverged in the post-treatment period. 1.4.1.4.2 Falsification tests We further conduct falsification tests to assess the credibility of the parallel trends assumption for both the number of telecom operators and the stock of rural-urban migrants living in a district. We do this by first randomly assigning treatment to some of the 112 districts that did not receive any bid in the 2001 auction. We use a pseudo-random variable following a Bernoulli distribution 16 See De Chaisemartin and d’Haultfoeuille (2022) for a review of the literature. 17 We are unable to show any pre-treatment trends of the number of telecom operators in treatement and control districts because we have only two data points on the number of operators. Before 1996, no region had any mobile phone service provider. 18 Annualized migration in thousands increased from 2.5 to 3.2 in districts that received a bid. In districts that did not receive a bid, migration increased from 0.9 to 1.1. 19 In bid-receiving districts, migration increased from 3.2 to 4.3; in control districts, migration increased from 1.1 to 1.3. 24 Figure 1.3: Rural-urban migration trends (1981-2005) 0 1 2 3 4 5 6 Inter−district rural−urban migration (’000) 1981−1990 1991−2000 2001−2005 Did not receive a bid Received a bid 95% C.I. Source: Authors’ calculations based on Census of India. Note: Figure presents the evolution of district-level inter-district rural-urban migration in India between 1981 and 2005. Bars represent means, and vertical lines represent 95% confidence intervals. Means and confidence intervals for each period are paired. District-level figures for districts in telecom circles that did not receive a bid in India’s 2001 auction of the 1800 MHz spectrum band are given by the left-hand bars and lines, and for districts in circles that received at least one bid are given by the right-hand bars and lines. Migration figures are annualized and given in thousands. Data on the stock of migrants living in a district in 2011 is used for the periods 1991-2005, and for 1981-1991 the stock of migrants in 2001 is used. with a success probability of 0.73 to assign the fake treatment to the untreated districts. The success probability is based on the share of districts in the full analysis sample that received treatment (302 divided by 414). We replace the Auction jt dummy with the Fake auction jt dummy, and rewrite equation (1.1a) as follows: y jt = β ′ Fake auction jt +σx jt +κ j +η t +ϵ jt (1.1d) where, y jt = {Operators jt , Log Migration jt }, x jt consists of a vector of controls, κ j and η t are district fixed effects and the time trend, respectively, and ϵ jt is the error term. The coefficient β ′ provides an estimate of the fake treatment effect on the number of telecom operators and rural-urban migration. We run 1,000 regressions using equation (1.1d). In figures A. 4 and A.5, we present the kernel 25 densities of the estimates of β ′ and their corresponding p-values. Even though there is a large spike at the zero value, the densities of coefficient estimates in figures A. 4a and A.5a are not informative. This is because there are a large number of non-zero values in figures A. 4a and A.5a that are similar in magnitude to the estimates presented in columns (1) and (2) of table1.4. There- fore, we turn to the p-values to get a sense of the statistical significance of the β ′ estimates. The kernel densities of p-values in figures A. 4b and A.5b clearly indicate that most regressions yield estimates of the fake treatment that are below the conventional threshold of 90% significance. These findings provide evidence in support of the parallel trends assumption in our difference- in-difference identification. In section 1.5.3, we go even further and test alternative explanations for the telecom shock-induced increase in rural-urban migration in the post-2000 period. 1.4.1.5 Summarizing the effects Our findings indicate that India’s telecom expansion during the early 2000s resulting from the telecom spectrum auction of 2001 increased rural-urban migration. While the impact of the spectrum auction on rural-urban migration was 26%, the telecom operator growth, in general, led to a 15% growth in rural-urban migration. In the full sample of districts, the average stock of rural-urban migration during this time grew by 36%. Thus, the telecom growth accounted for 42% of the growth in inter-district rural-urban migration in India between the 1990s and the early 2000s. These figures translate to 0.5 million additional rural-urban migrants at the national level between 2001 and 2005. 20 There is a problem with the estimates from our aggregated district-level analysis. We use the stock of migrants that had moved in the 1990s and early 2000s and were still living in a district in 2011 to construct our migration variables. Since the Census of India does not document detailed migration histories or multiple movements of individuals, the migration stock figures may not accurately capture the actual flow of migrants during the 1990s and the early 2000s. On the one hand, many rural individuals who moved to an urban area during 2001-2005 could have moved back to a rural area or another urban area between 2006-2011. On the other hand, it is also possible that annual average rural-urban migration did not change or declined between the 1991- 20 The number of annualized rural-urban migrants moving across Indian districts between1991-2000 was0.6 million, and between 2001-2005 it was 0.8 million. We multiply the growth of 0.2 million by the effect of telecom growth and further multiply those figures by five to obtain the total number of people that moved between 2001 and 2005. 26 2000 and 2001-2005 periods, but more of the earlier migrants had left or returned to rural areas by 2011. In the former case, our migration stock variable would be underestimating the number of individuals who moved as a result of the telecom shock. In the latter case, our migration variable would be overestimating the number of rural-urban migrants moving in response to the telecom shock. In the next section, we address this problem by using microdata and find that the Census figures significantly underestimate the number of migrants who moved in response to the telecom shock. 1.4.2 Telecom shock and individual mobility We use a two-way fixed effects framework to estimate the impact of the telecom shock of the early 2000s on individual-level migration in India. First, we estimate the impact of the shock on inter-district rural-urban migration. Then, we estimate the shock’s impact on inter-district urban-rural migration. We construct a migration share-weighted telecom operator variable to quantify the telecom shock experienced by rural residents in the year 2000. For urban residents, the shock is the number of local telecom operators in the district. Finally, we estimate the effect of telecom expansion on net inter-district rural-urban migration in India. 1.4.2.1 Definition of distant telecom shock Consider an individual i living in region j at time t− 1. Suppose that at the start of period t, the individual witnesses a distant telecom shock called Distant operators and decides whether to move to another region or not in response to the shock. All individuals living in the region j experience the same shock. We define the distant telecom shock algebraically as follows: Distant operators jt = ∑ d∈D Θ jd Local operators dt (1.2a) where, D is the set of all regions outside the region-of-residence j of individual i, Θ jd is the current population’s share of in-migrants in j who moved from a region d outside the region j before 1991, and Local operators dt is the number of telecom operators in region d at time t. We can interpret Distant operators jt as the migration share-weighted number of telecom operators 27 outside the region j. 21 The underlying hypothesis for the construction of Distant operators in equation (1.2a) is that telecom expansion in one region affects the migration decisions of individuals living in another region. The channel of effect is the pre-existing migration link between the two regions measured here as Θ jd , which also quantifies the strength of connection between the two regions. Even thoughΘ jd is based on a unidirectional movement of people from d to j, it measures how well the two regions are connected through historical migration channels. This is because the share of in-migrants living in j who moved from d would predict the share of in-migrants in d who moved from j. This idea is based on the fact that there is a lot of back-and-forth movement between pairs of regions in India (Deshingkar, 2008). 22 As an example, consider the states of Gujarat and Maharashtra. On the one hand, the second largest share, or about one-sixth of in-migrants living in Mumbai, the largest city in Maharashtra, came from Gujarat. On the other hand, the largest share, or about a quarter of in-migrants living in Gujarat, are from Maharashtra. 23 We see a similar pattern of migration linkages between state pairs such as Bihar and West Bengal, Punjab and Uttar Pradesh, Andhra Pradesh and Orissa, etc. 1.4.2.2 Rural-urban migration We estimate the impact of the Distant operators jt shock on the migration decisions of an individ- ual i living in rural areas of a region j using the following two-way fixed effects (TWFE) linear equation: Rural-urban migrant ijt = δ Distant operators jt +ϕ i +τ t +ν it (1.2b) where, Rural-urban migrant ijt is a dummy variable equal to one if individual i moved to an urban area from rural areas of j in the post-2000 period, and zero otherwise, Distant operators jt is as de- fined in equation ( 1.2a),ϕ i andτ t are the individual fixed effects and the time trend, respectively, 21 We know the district-of-residence of non-migrants in the year 2000. For each non-migrant’s district of residence, we calculate the shock based on the formula given in equation (1.2a). However, we only know the state of origin of a migrant in the year 2000. Hence, we calculate a state-level shock value for each migrant in two steps. First, we calculate the value of the shock for each district within a state using equation (1.2a). And then, we take an average of the shock values for all districts in the state. 22 For a historical narrative on pre-existing migration links between regions of India, see Tumbe (2018). 23 Since we do not have data on district-to-district migration flows, we compute state-wide averages to compare out-migration from one region to another. 28 and ν it is the error term. There are two periods in our analysis: the year 2000 and post-2000, which is 2001-2005. Our regression sample includes individuals who stayed back in rural areas in the post-2000 period and those who migrated from rural to urban areas across districts of India between 2001 and 2005. In other words, we only consider inter-district movements. We identify equation (1.2b) using temporal variation within each unit of observation, i.e., an individual. The results from the TWFE linear probability model (LPM) estimation of equation (1.2b) are presented in columns (1) and (3) of panel (a) in table 1.5. Column (1) of panel (a) provides estimates of equation (1.2b) with Rural-urban migrant ijt equal to one for any rural-urban migrant, whereas in column (3) of panel (a), Rural-urban migrant ijt is equal to one for rural-urban migrants who moved solely for employment purposes. The sample in column (3) excludes migrants who moved for purposes other than employment. Furthermore, the sample in column (3) is restricted to individuals between the ages of 18 and 52 years in the year 2000 or 25-59 years old at the time of the survey during 2007-2008. The age restriction is imposed to ensure that the sample consists of individuals who were of working age both at the time of shock in the year 2000 and during the survey in 2007-20008. 24 We subsample individuals who were of working age at the time of the shock in the year 2000 because we examine their labor mobility decisions. We select individuals who were of working age at the time of the survey because, in section 1.5.1, we use the same sample of individuals to calculate the wage premium obtained by migrants based on observed earnings in the survey. Column (1) of panel (a) in table 1.5 indicates that rural residents’ probability of any-purpose migration to urban areas went up by 0.6 percentage points with every additional migration share-weighted telecom operator outside their region-of-residence. Column (3) indicates that the probability of rural-urban migration for employment purposes among working-age individu- als increased by 0.4 percentage points with every additional migration share-weighted distant telecom operator. The average rural resident received a shock of 3 additional migration share- weighted telecom operators outside their region of residence in the post-2000 period. This implies that the telecom growth increased the probability of rural residents migrating to urban areas for any purpose by 1.8 percentage points, and the probability of working-age individuals’ migration 24 The upper bound of 59 is based on the fact that the retirement age in India was 60 during 2007-2008. See https://documents.doptcirculars.nic.in/D2/D02est/25012_8_98-Estt-A.pdf. 29 Table 1.5: Effect of telecom expansion on individual migration Panel (a): Regression estimates Dependent variable = Migration dummy Any migrant 18-52 yr. labor migrant Rural-urban Urban-rural Rural-urban Urban-rural (1) (2) (3) (4) Distant operators 0.006*** 0.004*** (0.000) (0.000) Local operators 0.002*** 0.001*** (0.000) (0.000) Individual fixed effects ✓ ✓ ✓ ✓ Time trend ✓ ✓ ✓ ✓ N 473,170 255,380 194,332 118,812 R 2 0.019 0.007 0.012 0.003 Panel (b): Gross estimates Any migrant 18-52 yr. labor migrant Rural-urban Urban-rural Rural-urban Urban-rural (1) (2) (3) (4) Telecom operator growth 3.02 3.22 3.01 3.22 Migration probability 0.018 0.006 0.012 0.003 Relevant population (millions) 743 286 341 151 Migration (millions) 13.37 1.72 4.09 0.45 Source: Authors’ calculations. Note: Panel (a) in table presents estimates from two-way fixed effects linear probability regressions at the individual level using employment and migration data from the NSS survey conducted during July 2007-June 2008. All individ- uals’ district or state of residence in the year 2000 and their subsequent residence, if they migrated to another district between 2001-2005, are known. Column (1) sample consists of 236,585 individuals who lived in rural areas in 2000; some of them moved to urban areas for any purpose between 2001 and 2005. Column (2) sample consists of 127,690 individuals who lived in urban areas in 2000; some of them moved to rural areas for any purpose between 2001 and 2005. Column (3) sample consists of 97,166 individuals who were between the ages of 18 and 52 years and lived in rural areas in 2000; some of them moved to urban areas for employment between 2001 and 2005. Column (4) sample consists of 59,406 individuals who were between the ages of 18 and 52 years and lived in urban areas in 2000; some of them moved to rural areas for employment between 2001 and 2005. Dummy variables for rural-urban and urban-rural migration are created based on this information. Local operators is the average number of telecom operators in the place of residence of an individual. Distant operators is the average number of migration share-weighted telecom operators corresponding to an individual’s region of residence in 2000. It is calculated by first taking the product of the share of in-migrants who moved from another region to the individual’s region-of-residence before 1991 and the number of telecom operators in the other region and then adding the products for all regions outside the region-of- residence of the individual. Standard errors are clustered at the household level and reported in parentheses. * p < 0.10, ** p< 0.05, *** p< 0.01. Panel (b) in table presents gross estimates of migration from extrapolation of regression estimates to the total pop- ulation. Distant operator growth given in columns (1) and (3), and local operator growth given in columns (2) and (4). Migration probability equals the regression estimate multiplied by the telecom operator growth in a column. Relevant population figures are at the national level from Census 2001. Columns (1) and (2) provide the total rural and urban populations of India in 2001, respectively. Columns (3) and (4) provide the rural and urban 18-52 year old populations, respectively. Migration in millions is telecom growth-induced migration; figures are obtained by multiplying the migration probability with the relevant population figures within a column. 30 for employment purposes increased by1.2 percentage points (see columns (1) and (3) of panel (b) in table 1.5). These findings broadly confirm our earlier results in section 1.4.1 that the telecom expansion of the early 2000s led to increased rural-urban migration in India. While the evidence presented thus far in tables 1.4 and 1.5 suggests that there was increased mobility for all purposes from rural to urban areas as a result of India’s 2000s telecom boom, the findings need not necessarily suggest that urban areas were growing as a result of the telecom growth-induced migration. There is a possibility that individuals were also moving from urban to rural areas as a result of the telecom expansion. Young (2013) shows that individuals move be- tween rural and urban areas at similar rates in developing countries. Therefore, we test whether India’s telecom expansion led to a reverse outward movement from urban to rural areas as well. 1.4.2.3 Urban-rural migration To test whether India’s early 2000s telecom expansion also led to urban-rural migration, we use a sample of individuals living in urban areas of a region j in the year 2000. We estimate the impact of local telecom expansion on urban-rural migration from j in the post-2000 (2001-2005) period with the following TWFE linear equation: Urban-rural migrant ijt = δ ′ Local operators jt +ϕ i +τ t +ν it (1.2c) where, Urban-rural migrant ijt is a dummy variable equal to one if an individual i moved from region j’s urban areas to another region’s rural areas in the post-2000 period, and zero otherwise, Local operators jt is the number of telecom operators in j at time t,ϕ i andτ t are the individual fixed effects and the time trend, respectively, andν it is the error term. As in the case of equation (1.2b), we only consider inter-district mobility here. Our regression sample includes individuals who stayed in urban areas in the post-2000 period and those who migrated from urban to rural areas across districts of India between 2001 and 2005. We identify equation (1.2c) the same way as in equation (1.2b). The results from the TWFE LPM estimation of equation (1.2c) are presented in columns (2) and (4) of panel (a) in table 1.5. Column (2) presents estimates with Urban-rural migrant ijt equal to one for any urban-rural migrant, whereas in column (4), Urban-rural migrant ijt is equal to one 31 for urban-rural migrants who moved solely for employment purposes. The sample in column (4) excludes migrants who moved for purposes other than employment, and the column (2) sample includes such migrants. As in column (3), we restrict the sample used in column (4) to those aged 25-59 years at the time of the survey, or 18-52 years in the year 2000 (see section 1.4.2.2 for details). Column (2) indicates that urban residents’ probability of any-purpose migration to rural areas went up by 0.2 percentage points with every additional local telecom operator. Column (4) indicates that working-age urban residents were 0.1 percentage points more likely to move to rural areas with every additional local telecom operator. The average urban resident received a shock of 3.2 additional local telecom operators. This implies that the telecom growth increased urban residents’ probability of migrating to rural areas for any purpose by 0.6 percentage points, and working-age individuals were 0.3 percentage points more likely to move to rural areas for employment (see columns (2) and (4) of panel (b) in table 1.5). The increase in urban-rural migration resulting from the telecom expansion is not surprising. Information enables labor market matching, leading to increased mobility in all directions as workers sort based on their skills and other preferences (Young, 2013). Information flows, which telecom infrastructure facilitate, increase workers’ understanding of the types of firms and job opportunities in distant labor markets, including rural labor markets. Average urban wages may be higher than average rural wages in India, but some types of workers may nevertheless receive a wage premium by moving from urban to rural areas or may derive other forms of non- pecuniary value. The increase in urban-rural migration may have been bolstered by the National Rural Employment Guarantee Scheme, a large-scale rural public works program implemented in India during the 2000s, which boosted rural wages (Imbert and Papp, 2015). 1.4.2.4 Net effects In panel (b) of table 1.5, we calculate gross telecom growth-induced migration figures by ex- trapolating our estimates to the national-level population of India in 2001. We estimate that the telecom expansion led to 14 million additional individuals moving from rural to urban areas across districts of India between 2001 and 2005, out of which more than 4 million working-age individuals moved for employment purposes. These figures are substantially higher than the 32 estimate of 0.5 million we obtained based on our aggregate analysis. The wide gap between the NSS and Census estimates of rural-urban migration is not entirely surprising. First, as we stated earlier, we construct aggregated migration variables using Census migration stock figures—individuals that moved between 1991 and 2005 and were staying at their destinations in 2011. It is likely that many migrants who moved from rural to urban areas between 2001 and 2005, moved back to rural areas between 2005 and 2011, thereby leading to underestimates of actual movement. Second, the Census enumerates migrants who move on a permanent basis in India. Other temporary forms of mobility, such as circular and seasonal, are not captured by the Census figures. Imbert and Papp ( 2020) suggests that every year more than eight million Indians migrate seasonally from rural to urban areas for one to six months in search of employment before moving back to their homes. Moreover, extrapolating sample estimates from the NSS employment and migration survey to the entire population of India suggests that each year about five million individuals move on a temporary but non-seasonal basis with the intention of returning to their homes about a year after migrating. Hence, the telecom-induced migration estimate of 14 million is plausible. We calculate gross telecom growth-induced urban-rural migration figures as well. In panel (b) of table 1.5, we see that the telecom expansion led to 1.7 million additional individuals moving from urban to rural areas across districts of India between 2001 and 2005, out of which less than half a million working-age individuals moved for employment. Subtracting the urban-rural migration figures from rural-urban migration, we find that the telecom shock led to a net inter-district rural-urban migration of about 12 million individuals. Out of these, almost 4 million working-age individuals moved for employment. These estimates suggest that net rural-urban migration for purposes other than employment was affected more than labor mobility by the telecom shock. This finding is not surprising given that the vast majority of internal migrants in India consists of women moving from their ancestral homes to their spouses’ homes after marriage (Rosenzweig and Stark, 1989). 25 The fact that the telecom shock affected non-labor mobility more than labor migration underscores the need to explore the heterogeneity of movement in the context of countries like India. 25 In2011, over45% of the stock of internal migrants in India consisted of women who moved at the time of marriage (Census of India, 2011b). 33 To put the net rural-urban migration estimates into context, we calculate the share of urban growth explained by telecom expansion-induced migration. The urban population of India grew by 91 million, from 286 million in 2001 to 377 million in 2011. Assuming a constant rate of growth, the urban population will have grown by 45 million between 2001 and 2005, which is our post-2000 period. Therefore, the telecom expansion-induced net rural-urban migration explains roughly 27% of the urban population growth of India during the early 2000s. 1.5 Channels of effects In section 1.4, we show that mobility increased as a result of the telecom expansion in India during the early 2000s. There was a net movement of about 12 million individuals from rural to urban areas across districts of India. Of these 12 million migrants, 4 million were working-age individuals who moved for employment. In this section, we explore the mechanisms through which the telecom shock could have led to increased migration. First, we examine the impact of telecom expansion on rural-urban migrants’ wage premia. Next, we investigate whether in- creased information flows could explain the higher proclivity of individuals to migrate out of rural areas. And finally, we conduct some robustness checks by exploring alternative channels which could act as confounders in our estimates. 1.5.1 Rural-urban migration premium We test whether telecom growth affected migration through rural-urban migrants’ higher wage premia earned after moving to cities. In other words, we examine whether the telecom shock increased the wage premium or the gap between rural-urban migrants’ and rural resident non- migrants’ wage earnings. The wage premium could have grown as a result of telecom expansion if improved communications infrastructure spurred firm growth, creating a labor demand shock, or if increased mobile phone access among rural-urban migrants enhanced their ability to bargain for higher wages due to greater access to information. First, we estimate the telecom growth- induced migration wage premium. Then, we estimate the wage premium earned by rural-urban migrants who moved before the telecom shock of 2001. Finally, we compare the two premia estimates. 34 1.5.1.1 Telecom shock-induced migration premium Consider a working-age individual i who lived in a rural area in the year 2000. We observe their daily wage earning at the time of the survey in 2007-2008. We also know whether the individual had migrated from a rural to an urban area across districts of India between 2001 and 2005 for employment reasons or whether they chose to stay put. We denote this information with a dummy variable Rural-urban labor migrant i , which equals one if the individual migrated for employment, and zero, if they did not move. We estimate the telecom shock-induced migration wage premium using the following equa- tion: Log wage i = π Rural-urban labor migrant i × ∆Distant operators d +ψz i +ω j +υ i (1.3a) where, ∆Distant operators d is the change in the number of migration share-weighted distant telecom operators (defined in section 1.4.2.1) between the pre- and post-2000 periods at individ- uals’ residence d in the year 2000, z i is a vector of individual characteristics that determine wage earnings such as gender, age, and education, ω j denotes fixed effects for individual i’s district- of-residence at the time of the survey in 2007-2008 (note that d = j for non-migrants), and υ i is the error term. The regression sample for estimating equation (1.3a) consists of rural resident non-migrants and rural-urban migrants who moved for employment across districts of India between 2001 and 2005. The interaction term coefficient π provides an estimate of the additional wage earned by labor migrants who moved because of the telecom shock compared to rural residents who did not migrate after 2001. In other words, π is the telecom shock-induced migration premium obtained by rural-urban migrants. In the regressions, we control for the number of years rural- urban migrants spent in urban areas after migrating (set to zero for rural resident non-migrants). We do this to account for the fact that individuals’ earnings steadily increase because of learning experiences and human capital accumulation after moving to urban areas (Lucas, 2004; Roca and Puga, 2017). The OLS estimates of equation (1.2a) are given in column (1) of table 1.6. Every additional 35 Table 1.6: Migration premia estimation Dependent variable = Log daily wage earning OLS Heckman-corrected (1) (2) (3) (4) Distant operators× Labor migrant 0.107*** 0.100*** (0.018) (0.018) Labor migrant pre-2000 0.297*** 0.274*** (0.041) (0.040) Female dummy -0.383*** -0.387*** -0.311*** -0.281*** (0.009) (0.009) (0.023) (0.027) Std. age 0.162*** 0.163*** 0.161*** 0.156*** (0.017) (0.017) (0.017) (0.017) Std. age squared -0.018 -0.016 -0.005 0.002 (0.013) (0.013) (0.013) (0.013) School graduate or above 0.618*** 0.633*** 0.628*** 0.645*** (0.012) (0.011) (0.013) (0.012) Years in urban 0.007 0.012*** 0.008 0.013*** (0.013) (0.002) (0.012) (0.002) District fixed effects ✓ ✓ ✓ ✓ N 30,398 31,317 97,166 98,828 R 2 0.467 0.487 Source: Authors’ calculations. Note: Table presents OLS estimates in columns (1) and (2) and Heckman-corrected estimates in columns (3) and (4) at the individual level using employment and migration data from the NSS survey conducted during July 2007-June 2008. Samples consist of individuals who were between the ages of 18 and 52 years in 2000. Column (3) sample consists of 97,166 individuals who were living in rural areas in the year 2000; some of them moved to urban areas for employment after 2000. Column (4) sample consists of 98,828 individuals who were living in rural areas in the year 2000; some of them moved to urban areas for employment before 2000. Subsamples of columns (3) and (4) used in columns (1) and (2), respectively. Subsamples consist of individuals who were participating in the labor force in 2007-2008. Labor migrant is a dummy equal to one if an individual migrated from a rural to an urban area after 2000, and zero otherwise. Labor migrant pre-2000 is a dummy equal to one if the individual migrated from a rural to an urban area before 2000, and zero otherwise. Distant operators is the average number of migration share-weighted telecom operators corresponding to an individual’s region of residence in 2000. It is calculated by first taking the product of the share of in-migrants who moved from another region to the individual’s region-of-residence before 1991 and the number of telecom operators in the other region and then adding the products for all regions outside the region-of-residence of the individual. Years spent in urban is equal to the number of years since a rural-urban migrant moved to an urban area. Variables used in the selection equation for the Heckman corrected estimates include the female dummy, marital status dummy, age, and categorical variables for education and relationship of an individual to the household head. Standard errors are clustered at the household level and reported in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. migration share-weighted distant telecom operator increased the wage gap between rural-urban migrants and their rural resident non-migrant counterparts by almost 11 percentage points. The average working-age rural resident received a shock of 3 migration share-weighted distant op- erators. Hence, the gross effect of the telecom shock led to a wage gap of 32 percentage points between rural-urban migrants and rural resident non-migrants. 36 Since the OLS regression sample for the wage equation only includes individuals participat- ing in the labor force, there is a selection bias in the estimates presented in column (1) of table 1.6 (Heckman, 1979). Therefore, we conduct a Heckman correction with a selection equation that accounts for individual characteristics predicting labor force participation, such as whether the individual is female, their marital status, age, education, and their relationship to their house- hold’s head. The Heckman-corrected estimates are given in column (3) of table 1.6. We see that the magnitude of the telecom shock-induced wage premium estimates is similar to that in column (1) of table 1.6. Thus, even with the Heckman correction, the wage gap between rural-to-urban migrants and rural residents is 30 percent. 1.5.1.2 Migration premium in the pre-telecom era Now, consider a working-age individual i who was either living in a rural area in the year 2001 or had migrated from a rural area to an urban area across districts of India for employment at any point in time before 2001. We denote the migration status of the individual with a dummy variable Labor migrant pre-2000 i , which equals one if the individual migrated from a rural area to an urban area before the year 2001, and zero, if they did not move out of rural areas. We also observe their daily wage earning at the time of the survey. We estimate the rural-urban migration wage premium obtained by labor migrants who moved before the year 2001 with the following equation: Log wage i = ρ Labor migrant pre-2000 i +ψz i +ω j +υ i (1.3b) where, z i is a vector of individual characteristics that determine wage earnings (as in equa- tion (1.3a)), ω j represents fixed effects for individual i’s district-of-residence at the time of the survey in 2007-2008, and υ i is the error term. The regression sample for estimating equation (1.3b) consists of rural resident non-migrants and rural-urban migrants who moved for employment across districts of India before 2001. The term ρ estimates the wage gap between rural-urban labor migrants and rural residents who did not migrate until 2001. As in the case of equation (1.3a), the vector z i includes the number of years rural-urban migrants spent in urban areas after migrating to control for additional earnings 37 arising from learning by working in cities. Note that the counterfactual groups in the regression samples used to estimate equations (1.3a) and (1.3b) are identical. Hence, the difference between the wage gap estimates of π and ρ reflects the divergence of rural-urban migrants’ wage premia resulting from the telecom shock. The OLS and Heckman-corrected estimates of equation (1.3b) are given in columns (2) and (4) of table 1.6, respectively. The OLS estimates suggest that the wage gap between rural-urban migrants who moved before2001 and rural resident non-migrants was about30%. The Heckman- corrected estimates suggest a lower gap of 27%. Note that while the years spent in urban areas have no impact on the earnings of migrants who moved after 2000, it is significant and positive for migrants who moved before 2001. This is because the time spent by post-2000 migrants in urban areas was insufficient to affect their earnings significantly. At the time of the survey, migrants who moved after 2000 spent, on average, four years in urban areas, whereas migrants who moved before 2001 spent 19 years in cities after migrating. There could also be a non-linear relationship between time spent in urban areas and earnings. 1.5.1.3 Comparing migration premia Comparing the migration wage premia estimates from sections 1.5.1.1 and 1.5.1.2, we find that rural-urban labor migrants who moved after2000 earned2.4-2.6 percentage points higher premia compared to those who moved before 2001. This translates to a widening of the wage gap between rural-urban migrants and rural resident non-migrants by 8-10%. 26 The higher wage premium partly explains the telecom shock-induced rural-urban migration. We note an important point regarding our estimates of the migration wage premia. Rural- urban labor migration is endogenous to both rural and urban wages. This is because migration leads to an adjustment in the labor market equilibria through demand and supply changes in both rural and urban areas. Increased rural-urban migration would lead to lower urban and higher rural wages. Therefore, in a general equilibrium framework, the gap between rural-urban migrants’ and rural residents’ earnings would be wider. In other words, our estimates of the migration premia are lower bounds for the true rural-urban migration wage premia. 26 We get the 8% figure by dividing 0.024 by the OLS estimate of 0.297 (column (2) of table 1.6). Similarly, we obtain the 10% figure by dividing 0.026 by the Heckman-corrected estimate of 0.274 (column (4) of table 1.6). 38 1.5.2 Improved information flows We estimate whether improved information flows resulting from the telecom shock could have contributed to increased migration. Better telecom infrastructure can enable rural residents to access more information about distant labor markets through mobile phones (Aker et al., 2012). We use IHDS data to find that rural Indian households with access to mobile phones were about three percentage points more likely to send a migrant to urban areas (see table A.5). Moreover, non-migrant rural residents gain knowledge about urban labor markets from former household members who migrated to cities in the past (Lu et al., 2016). This also explains why migrants move to places where they have existing social networks (Munshi, 2003). Based on Lu et al. (2016), we build an identification strategy that can provide suggestive evidence of increased migration as a result of telecom shock-induced information flows. We test whether a rural household’s former members living outside the household’s district of residence partly explain the impact of telecom growth on the out-migration of current rural households’ members. We use information on former household members who out-migrated in the past from the NSS employment and migration dataset to conduct our analysis. Unfortunately, the NSS data do not tell us whether a former member moved to an urban or a rural area. Hence, we examine the impact of telecom growth-induced information flows on out-migration from rural areas of a district to both rural and urban areas of another district. Consider an individual i from household k residing in rural areas of district j in the year 2000. The individual i experiences the distant telecom shock, defined in section 1.4.2.1, at the start of the pre-2000 period by virtue of living in district j. The shock is the number of migration share- weighted telecom operators or Distant operators. Suppose that the individual i’s household sent a member outside the household’s district of residence in the past. We denote this information with a dummy variable Pre-2000 migrant HH k , which equals one if household k sent at least one previous member to another district before the year 2001, and zero otherwise. We estimate the impact of the telecom shock on the probability of current rural residents’ mi- grating out to other districts that is incident through their households’ former members residing 39 in other districts using the following two-way fixed effects equation: Out-migrant ikjt = λ 0 Distant operators jt +λ 1 Distant operators jt × Pre-2000 migrant HH k +ϕ i +τ t +ν it (1.4) where, Out-migrant ikjt is a dummy variable equal to one if individual i chose to out-migrate to another district in the post-2000 period, and zero otherwise, ϕ i and τ t are individual fixed effects and the time trend, respectively, and ν it is the error term. There are two periods in the analysis—the year 2000 and the post-2000 period, which is 2001-2005. The interaction term’s coefficient λ 1 estimates the effect of telecom growth on migration prob- ability incident through the presence of a former household member in another district. In other words, λ 1 captures, to some extent, the impact of improved information flows on migration re- sulting from telecom growth, with former household members being the medium of information transmission. The estimates of equation (1.4) are given in table1.7. We present the estimates of equation (1.4) without the interaction term in columns (1) and (3) and with the interaction term in columns (2) and (4) of table 1.7. In columns (1) and (2), Out-migrant ikjt is equal to one for any migrant, and in columns (3) and (4), it is equal to one for migrants who moved solely for employment purposes. As in section 1.4.2, we restrict the sample for regressions in columns (3) and (4) to individuals who were between the ages of 18 and 52 years in the year 2000, or 25-59 years old at the time of survey during 2007-2008. We see that individuals who belonged to households with a former member living outside their district of residence were 0.6 percentage points more likely to move to another district with every additional migration share-weighted distant telecom operator compared to individuals belonging to households without an out-migrant from the past. Working-age individuals were 0.2 percentage points more likely to move out of their district of residence for employment if their households sent a labor migrant to another district in the past. Since the average rural resident experienced a telecom shock of three migration share-weighted operators, the overall impact of the shock was to increase the probability of out-migration by 1.8 percentage points for any purpose and 0.6 percentage points for employment for individuals with a household member 40 Table 1.7: Telecom expansion-induced follower migration Dependent variable = Post-2000 out-migrant dummy Any migrant 18-52 yr. labor migrant (1) (2) (3) (4) Distant operators 0.008*** 0.007*** -0.001 -0.001 (0.001) (0.001) (0.001) (0.001) Distant operators× 0.006*** 0.002*** Pre-2000 migrant in HH (0.000) (0.001) Individual fixed effects ✓ ✓ ✓ ✓ Time trend ✓ ✓ ✓ ✓ N 581,658 581,658 272,694 272,694 R 2 0.054 0.055 0.048 0.048 Source: Authors’ calculations. Note: Table presents estimates from two-way fixed effects linear probability regressions at the individual level using employment and migration data from the NSS survey conducted during July 2007-June 2008. All individuals’ district or state of residence in the year2000 and their subsequent residence, if they migrated to another district between2001- 2005, are known. Sample of 290,829 individuals, some of whom migrated to another district for any purpose between 2001 and 2005, used in columns (1) and (2). Sample of 136,347 individuals, some of whom migrated to another district for employment between 2001 and 2005, used in columns (3) and (4). Pre-2000 migrant in HH is a dummy equal to one if the individual’s household had sent an out-migrant outside the household’s district of residence before 2001, and zero otherwise. In columns (1) and (2), the pre-2000 migrant dummy is equal to one if the previous migrant moved for any purpose. In columns (3) and (4), the pre-2000 migrant dummy is equal to one if the previous migrant moved solely for employment. Distant operators is the average number of migration share-weighted telecom operators corresponding to an individual’s region of residence in 2000. It is calculated by first taking the product of the share of in-migrants who moved from another region to the individual’s region-of-residence before 1991 and the number of telecom operators in the other region and then adding the products for all regions outside the region-of-residence of the individual. Standard errors are clustered at the household level and reported in parentheses. * p < 0.10, ** p < 0.05, *** p< 0.01. living outside their district-of-residence. Another way of interpreting these results would be that current rural residents were 1.8 percentage points more likely to follow in the footsteps of former household members to other districts as a result of the telecom shock. The impact of the telecom shock on rural working-age individuals’ out-migration for em- ployment can be entirely explained by the presence of a former household member outside the district. In other words, improved information flows serve as the primary channel of effect of telecom growth on labor migration. Note that these estimates are not directly comparable to the estimates of telecom growth’s impact on individual-level rural-urban migration given in table 1.5 because our rural out-migration variable includes both rural-urban and rural-rural migrants. The coefficient of the distant operator variable in column ( 3) of table 1.7 is insignificant, even though the estimate of the same variable is significant in column ( 3) of table 1.5. We also note that improved information flows could enable labor migrants to bargain for higher wages after mi- 41 grating. Hence, the higher wage premium obtained by migrants as a result of the telecom shock (see section1.5.1) is consistent with telecom shock affecting migration through better information flows. 1.5.3 Robustness checks In the previous sections, we tested the impact of telecom growth on migration. We used the 2001 telecom spectrum auction as a pivot to conduct our analysis. However, several other events occurred during the same time, which could also explain the increased migration. Dutta et al. (2021a) finds that the National Highways Development Project Phase I, better known as the Golden Quadrilateral (GQ) highway upgrade project, increased internal migration in India be- tween 2001 and 2011. 27 Moreover, India became a major exporter of Information Technology (IT) services as a result of the trade liberalization during the 1990s combined with higher interna- tional demand for IT services. The resulting IT boom was a labor demand shock, which also increased internal migration in India (Ghose, 2021). In this section, we examine whether these events can partly explain the effect of telecom shock on migration. We examine the confounding effects of the GQ highway upgrade project and a labor demand shock resulting from export growth on telecom shock-induced migration using district-level ag- gregated data. Following Dutta et al. (2021a), we define a dummy variable GQ jt which is equal to one if district j received highway infrastructure investments under the GQ project in the post- 2000 period (2001-2005), and zero otherwise. In order to study the second possible confounding effect, trade expansion-induced labor demand, we construct a labor demand shock variable based on sectoral employment shares and international trade volumes (see next paragraph). We com- pare the impact of these events with the impact of the number of telecom operators on district j’s inter-district rural-urban migrant stock. The comparison enables us to ascertain the relative importance of telecom infrastructure growth and confounding factors in explaining the telecom shock-induced migration. Suppose that the share of employment of sector s in district j in the year 2001 was θ 2001 sj 27 The central government of India implemented the National Highways Development Project Phase I (NHDP I) as a highway upgrade program in 2001. It was meant to upgrade preexisting highways connecting the four largest metropolitan areas of India—Delhi, Mumbai, Kolkata, and Chennai—from two lanes to four lanes. See Dutta et al. (2021a) for details. 42 (based on Census data). We denote the national-level export volume of sector s as Export st (in US dollars). We define the Labor demand jt shock variable as follows: Labor demand jt = ∑ s∈S θ 2001 sj Log Export st (1.5a) where, S is a set of four sectors—agriculture, manufacturing, construction and mining, and services. Changes in the export demand act as international trade shocks, and hence, induce labor demand. Following Bartik (1991), we can interpret Labor demand jt as a shift-share shock. We estimate the impact of the confounding factors and the number of telecom operators on inter-district rural-urban migration using the difference-in-difference setup with interaction terms discussed in equation (1.1b). We estimate the following two-way fixed effects equation: Log Migration jt = µ 0 Auction jt +µ 1 Auction jt × Shock jt +σx jt +κ j +η t +ϵ jt (1.5b) where, Auction jt is the auction dummy which equals one if district j received a bid in India’s2001 telecom spectrum auction, and zero otherwise, Shock jt ={Operators jt , GQ jt , Labor demand jt }, is a vector of linearly varying treatments in district j, x jt is a vector of controls (as in equation (1.1b)), κ j and η t are district fixed effects and the time trend, respectively, and ϵ jt is the error term. We identify equation (1.5b) the same way we identify equation (1.1b). The estimates of equation (1.5b) are given in table 1.8. In column (1) of table 1.8, we present the estimates of equation (1.5b) without the interaction term, and in columns (2)-(4), we present the estimates of each confounding factor and the number of telecom operators interacted with the auction bid dummy. Column (2) of table 1.8 is identical to column (4) of table 1.4. In column (3) of table 1.8, we see that districts that received a bid in the telecom spectrum auction and were recipients of the GQ highway upgrade project witnessed a 4.3% growth in rural-urban migration. Even though this average treatment effect is similar to the estimate of the telecom operator growth impact on migration given in column (2) of table 1.8, the overall effect of the telecom shock on migration through operator growth is 15% or about four times the effect of the shock through the GQ project. We also note that a large portion of the impact of the telecom spectrum auction is still unexplained after removing GQ’s confounding effect, as can be 43 Table 1.8: Comparing coincident events’ impact on rural-urban migration Dependent variable = Log inter-district rural-urban migration (1) (2) (3) (4) Auction 0.073*** -0.124* 0.065*** -0.443 (0.018) (0.065) (0.019) (0.396) Auction× Operators 0.045*** (0.014) Auction× GQ 0.043* (0.023) Auction× Labor demand 0.050 (0.038) Log urban population 0.299*** 0.290*** 0.295*** 0.298*** (0.081) (0.076) (0.080) (0.080) Share literate urban -0.005 0.101 0.020 0.035 (0.531) (0.535) (0.530) (0.528) Share employed urban -0.102 0.039 -0.099 -0.122 (0.680) (0.672) (0.689) (0.680) Log consumption urban -0.032 -0.031 -0.035 -0.033 (0.025) (0.024) (0.024) (0.025) District fixed effects ✓ ✓ ✓ ✓ Time trend ✓ ✓ ✓ ✓ N 838 838 838 838 R 2 0.759 0.768 0.760 0.759 Source: Authors’ calculations. Note: Table presents two-way fixed effects regression estimates using a balanced panel of 419 districts observed during 1991-2000 or the pre-2000 period and during 2001-2005 or the post-2000 period. Dependent variable is the log of annualized stock of inter-district rural-urban migrants who were living in a district in 2011 and had moved during the pre- and post-2000 periods from rural areas of all other districts of India. Auction is a dummy equal to one if a district’s telecom circle received a bid in India’s 1800 MHz spectrum band auction held in 2001, and zero otherwise. Operators is the average number of local telecom operators in a district. GQ is a dummy equal to one if the district was a recipient of the Golden Quadrilateral highway upgrade project after 2001, and zero otherwise. Labor demand represents a labor demand shock variable calculated by first multiplying the 2001 share of employment in four sectors–—agriculture, manufacturing, mining and construction, and services—–with the log volume of the corresponding sector’s exports (in USD) in 2001 (for pre-2000) and 2005 (post-2000), and then adding the products over the four sectors for each period. Standard errors are clustered at the district level and reported in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. seen in the significant estimate of the auction bid dummy in column ( 3) of table 1.8. In addition, the estimate of the GQ dummy interaction term is only marginally significant at 90%. Hence, we conclude that the potentially confounding effect of the GQ project in our estimated effect of the telecom shock on rural-urban migration is not a major concern. Finally, we note that in column (4) of table 1.8, the interaction of the auction bid dummy and the labor demand shock variable is insignificant. In other words, the labor demand shock induced by trade expansion does not explain the impact of the telecom shock on rural-urban migration in India during the early 2000s. 44 1.6 Heterogeneity in migrants destinations In sections 1.4 and 1.5, we show that the telecom expansion in India during the early 2000s led to higher rural-urban migration. However, there is likely to be considerable heterogeneity in the nature of migration that ensued as a result of the telecom growth. Which cities received the most migrants? Did the income of urban areas matter in drawing migrants moving as a result of the telecom shock? In this section, we examine whether some cities received disproportionately more migrants than others and the role played by the wealth of cities in drawing migrants. This issue is worth exploring because it speaks to the question of whether technological change has differential impacts on wealthy and poor regions. We use district-level data to investigate whether rural-urban migrants were moving to richer or poorer cities as a result of the telecom expansion. We do this by first dividing districts into quintiles of income indicators and then estimating the impact of telecom expansion on rural- urban migration in each quintile. The two income indicators used here are the district-level mean monthly per capita consumption and the district-level mean weekly wage earnings. We compute the quintiles based on the distribution of these income indicators during 1999-2000. We use district-level data with two periods—1991-2000, or the pre-2000 period, and2001-2005, or the post-2000 period—to estimate the following TWFE equation: Log Migration jt = ∑ q∈Q γ q Operators jt × h q +σx jt +κ j +η t +ϵ jt (1.6) where, Migration jt is the stock of inter-district rural-urban migrants living in district j in period t, Q is the set of five quintiles based on the district-level consumption or wage distribution in the year 2000, h q is a dummy variable equal to one if district j belongs in quintile q, and zero otherwise, Operators jt is the number of telecom operators in district j at time t, x jt is a vector of controls (as in equation (1.1b)),κ j andη t are district fixed effects and the time trend, respectively, and ϵ jt is the error term. The coefficient γ q quantifies the marginal impact of an additional telecom operator on rural-urban migration to districts in quintile q. The TWFE estimates of equation (1.6) using monthly consumption and weekly wage quintiles are presented in figures 1.4a and 1.4b, respectively. We plot the estimates and the corresponding 45 Figure 1.4: Telecom growth impact on rural-urban migration by destination districts’ incomes 0.02 0.04 0.04 0.04 0.07 0.00 0.02 0.04 0.06 0.08 Local operator effect on rural−urban migration First Second Third Fourth Fifth Quintiles of district−level monthly mean consumption in 1999−2000 (a) Destination by household consumption 0.02 0.04 0.04 0.04 0.06 0.00 0.02 0.04 0.06 Local operator effect on rural−urban migration First Second Third Fourth Fifth Quintiles of district−level weekly average wages in 1999−2000 (b) Destination by weekly wage earning Source: Authors’ calculations. Note: Figures present estimates from two-way fixed effects regressions with a panel of 419 districts observed in the pre- and post-2000 periods. The log of annualized inter-district rural-urban migrant stock in a district is regressed on an interaction of the number of telecom operators in the district and a vector of dummies indicating the quintile of consumption or wage earning in which a district falls. The regressions in panel (a) include household consumption quintile dummies, and in panel (b), the individual weekly wage earning quintile dummies. The quintiles are based on the distribution of district-level means of consumption and wage during 1999-2000. All regressions include district fixed effects, the time trend, and a vector of controls that include the log of urban population, the shares of literate and employed in urban population, and the log of mean monthly per capita consumption in urban areas of the district. Regression estimates are rounded off to two decimal places. Vertical lines indicate the 95% confidence interval of regression estimates. 46 standard errors of γ for each quintile. Both plots indicate that the telecom expansion dispro- portionately increased rural-urban migration towards urban areas with the highest consumption and wage earnings in 1999-2000. Every additional telecom operator in districts belonging to the fifth (top) quintile of consumption and wage earnings led to 7% and 6% growth in inter-district rural-urban migration to those districts. In the fourth quintile, the effect of an additional tele- com operator on migration is 4% for both consumption and wage. The poorest cities in terms of consumption and wages saw the most minimal impact of telecom expansion on rural-urban migration. Our results indicate that telecom growth may not have led to an increase in rural-urban migrants’ expected earnings across all regions of India. Rural residents expected to earn more in cities that were already the wealthiest in India. 1.7 Discussion We show that the expansion of telecom technologies led to more rural-urban migration in India. Increased migration wage premia and improved information flows are some of the channels through which telecom growth increased migration. Our estimate that the telecom expansion led to a net increase of12 million rural-urban migrants demonstrates the importance of technological change as a driver of internal mobility. Since the existing literature has focused on low levels of permanent migration in India (Bhav- nani and Lacina, 2017; Kone et al., 2018), India’s interregional inequality, such as a large rural- urban wage gap, is attributed in part to low mobility (Dutta et al., 2022; Munshi and Rosenzweig, 2016). Underlying these claims is the idea that higher internal migration acts as a labor supply shock in migrants’ destinations and drives up labor demand in migrants’ origins, thereby acting as an equalizing force for wages across regions (Greenwood, 1975). Future research can examine whether increased mobility in countries like India actually leads to a lower income gap across regions and whether different drivers of migration lead to different impacts on interregional inequality. Our findings that telecommunication technologies’ growth led to more rural-urban migration in India also carry implications for policymakers aiming to increase rural dwellers’ access to opportunity in cities through migration. 47 What are the more general implications of increased internal migration for regional labor markets of India? Depending on the type of individuals who moved during the recent surge of migration in India, both urban and rural labor markets may have a skewed distribution of skill levels. Young (2013) posits that high-skilled workers are more likely to sort into urban areas because of a higher concentration of firms employing high-skilled labor in urban areas. However, Ghani et al. (2016) showed that firms in India have been moving away from large urban areas with better access to transportation infrastructure elsewhere. Therefore, understanding the skill distribution of migrants in India is, more generally, an important dimension for future research which can shed some light on how labor markets in different parts of the country respond to increased mobility. Our finding that India’s early 2000s telecom boom led to disproportionate migration towards the wealthiest urban areas raises other questions for future research. First of all, this finding highlights a difference between contemporary trends in developed countries like the US and developing countries like India. Many advanced economies have seen a slowdown in migra- tion to their most prosperous metropolitan areas, as labor and housing markets have become increasingly difficult to penetrate for all but the most educated workers (Ganong and Shoag, 2017; Moretti, 2012). Perhaps more significantly, our finding also raises the question of why telecom growth drew migrants to the wealthiest cities of India. Is it simply that migrants expected to earn more in cities that were already prosperous? Or was there a much larger labor demand shock in the wealthiest cities, suggesting that the firms in these cities were better positioned to make the most of technological change? Another explanation is that telecom expansion disproportionately supported more affluent or highly skilled migrants in making rural-urban moves, a population that may be more likely to find suitable employment in the wealthiest cities. Future research can address these questions while also investigating the role of other housing and labor market forces interacting with technological growth in drawing migrants to particular cities. Finally, our empirical findings suggesting that two-thirds of net rural-urban migration result- ing from telecom growth can be attributed to non-labor mobility underscores the need to study various forms of migration. A large share of Indians moving internally are women moving from their natal homes to their spouses’ homes at the time of marriage (Rosenzweig and Stark, 1989). 48 Moreover, the vast majority of migrants in India move on a temporary and seasonal basis rather than relocating permanently across regions (Imbert and Papp, 2020). Future research ought to explore these forms of mobility that have been less studied by academics researching developed countries. 1.8 Conclusion In this chapter, we use aggregated and microdata to show that India’s early 2000s telecom ex- pansion led to net rural-urban migration. This telecom expansion-induced migration explains more than a quarter of India’s urban population growth during the early 2000s. We attribute the increased mobility to increasing rural-urban migration premia and improved information flows. We conduct robustness checks and show that confounding factors like highway investments and labor demand shocks occurring at the same time as the telecom shock do not explain our results. Finally, we show that the wealthiest cities of India drew the most migrants moving as a result of the telecom shock. Our research contributes to the existing literature on improved communication and migration and, more generally, on the link between telecommunications infrastructure and urban growth. Moreover, we provide evidence linking technology, migration, and urban growth in developing countries. The empirical findings in this chapter raise several questions for future researchers to explore. Heterogeneity in the various forms of mobility undertaken for non-labor purposes, temporary migration, and the distribution of firms and workers’ skill levels across space are some dimensions of migration in developing countries like India that merit more investigation in future research. 49 Chapter 2 Distant Shocks, Migration, and Housing Supply Arnab Dutta Sahil Gandhi 1 Richard K. Green 2 2.1 Introduction Housing supply elasticity is an important and policy-relevant parameter of interest. It conveys critical information about a housing market, particularly one that is rapidly urbanizing. Yet, it is hard to precisely estimate the slope of the housing supply curve. Standard supply elasticity estimation techniques require demand shifters, or instruments, that satisfy the relevant exogene- ity assumptions. While geographical constraints, labor demand shocks, and migration have been argued to be exogenous to housing supply, caveats in the identification of the supply curve’s slope persist. In this chapter, we apply the Rosen-Roback spatial equilibrium framework (Roback, 1982; Rosen, 1979) to construct novel housing demand shifters. We demonstrate that, in some cases, migration-inducing shocks occurring in one region can affect the demand for housing in another without affecting the housing supply in the latter. Migration resulting from shocks, whose effects are concentrated within a region, leads to exogenous changes in the populations of regions that did not experience such shocks. Such an exogenous change in the population of a region, in turn, causes a shift in its demand for housing. We examine this hypothesis through a series of empirical tests and robustness checks. We demonstrate our estimation exercise with data on Indian districts. 3 India is a rapidly urbanizing country, with its current urban population estimated at around half a billion. 4 Fur- thermore, the case of India is particularly interesting with regard to the estimation of housing supply elasticity because of its variety of housing typologies. A formidable share of urban In- dia’s housing stock consists of informal houses made from non-durable substances, such as 1 Lecturer, University of Manchester. 2 Professor, University of Southern California. 3 A district in India is similar to a county in the United States. 4 For details, visit: https://data.worldbank.org/indicator/SP .URB.TOTL?locations=IN. 50 Figure 2.1: Shock-induced migration’s impact on housing demand Distant state B i h a r Local state Urban population growth Ma h a r a s h t r a Housing demand Out-migration Droughts Highway upgrade In-migration Source: Authors’ own. Note: Map presents a snapshot of the central part of India with the local state of Maharashtra and the distant state of Bihar highlighted. Negative rainfall shocks and highway upgradation occurs in Bihar. The resulting inter-state migration affects housing demand in Maharashtra through population growth. thatch, mud, plastic, etc. 5 Therefore, we estimate the formal housing supply elasticity, as well as the informal housing supply elasticity for urban India, between 2001 and 2011. In addition, we provide estimates for the supply elasticity of formal vacant houses since there was a large growth in the vacant housing stock of urban India during the 2000s. We use two migration-inducing events—droughts and highway infrastructure investments— occurring in one state as demand shifters for other states’ housing markets. To illustrate the spatial equilibrium mechanisms that explain how these shocks in one state may affect housing demand in another, we take the example of two Indian states—Maharashtra and Bihar. In spatial equilibrium, adjusting for location-specific amenities, the utility obtained by individuals living in Bihar and Maharashtra will be identical. Suppose that we want to estimate the housing sup- ply elasticity in urban areas of Maharashtra. We define Maharashtra as the local state. On the 5 Based on our definition of informal, in 2011, about 15% of the urban housing stock in India was informal (Cen- sus of India, 2011b). We define informality in section 2.2.1. 51 other hand, suppose that the state of Bihar experiences negative rainfall shocks (droughts) and a highway upgrade program. We call Bihar the distant state. The negative rainfall shocks and the highway upgrade program will affect wages and rents in Bihar, thereby, changing the utility obtained by individuals living in Bihar. The difference in util- ity obtained by those living in Bihar and Maharashtra will cause a state of spatial disequilibrium. In other words, there will be gains to movement between the two states. The resulting migration will affect the urban population growth of Maharashtra, and the change in urban population will, in turn, affect Maharashtra’s housing demand. Thus, negative rainfall shocks and a high- way upgrade program in the distant state of Bihar will act as demand shifters for urban housing markets in the local state of Maharashtra. Figure 2.1 provides a pictorial depiction of these spatial equilibrium mechanisms. Throughout the rest of the chapter, the terms distant and local are con- sistently used to refer to regions that experience shocks and regions where the housing markets are affected by the shocks, respectively. We use state-level data to show that negative rainfall shocks and a highway upgrade program increased inter-state migration in India between 2001 and 2011. 6 On the one hand, an additional rainfall deficit month increased decadal migration from affected states by 1.1%. On the other hand, a distant state’s inclusion in the Golden Quadrilateral (GQ) highway upgrade program increased migration both to and from such states. 7 This increased inter-state mobility led to urban population growth in the local state, or the migration-receiving regions. Next, using district-level data, we show that distant states’ shock-induced urban population growth increased the demand for housing in local districts. Furthermore, possibly because of higher floor space consumption by formal housing residents, the demand for formal housing units increased relatively more than the demand for informal units in response to the distant shock-induced urban population growth. Our findings indicate that both the negative rainfall shocks and the GQ highway upgrade program, occurring in distant states, are strong instruments for the number of formal, informal, and vacant houses in local urban markets. 6 Negative rainfall shocks are defined as decadal changes in the number of rainfall deficit months leading up to the two Census years, 2001 and 2011. For a given month, a state is rainfall deficit if the state’s total rainfall in the month is less than 80% of the long-term normal. See section 2.5.1 for details. 7 The Golden Quadrilateral (GQ), or the National Highways Development Project Phase I (NHDP I), was introduced as a highway upgrade program by the Central Government of India in2000, and it came into effect in2001. Its primary goal was to upgrade preexisting highways connecting India’s four largest metropolitan areas. 52 We conduct two robustness checks to alleviate endogeneity concerns in the identification strategy. The first robustness check addresses the possible confounding effects of spatial corre- lation of shocks by examining whether shocks in non-contiguous distant states have effects on migration that are similar to the effects of shocks in all distant states. The second robustness check investigates whether the distant state shocks affect the migration of individuals who move on a longer-term basis. Long-term movers to cities in India typically do not work in the construc- tion industry. Therefore, they would not affect the housing supply in cities through changes in construction costs. Both robustness checks show that the endogeneity concerns do not pose any threats to the identification strategy in the baseline results. With the same district-level dataset, using distant state shocks as instruments for housing demand, we estimate the national-level housing supply elasticities for urban India. Our hous- ing supply elasticity estimates can be summarized in three points. First, we estimate that the decadal supply elasticity of formal housing in urban India is 1.62. This estimate is very close to the supply elasticity of 1.75 obtained by Saiz (2010) for the average metropolitan area in the United States (US). Given that economists consider the US to be supply inelastic, India’s even lower elasticity figures indicate the existence of regulatory and institutional frictions. Second, we find that the supply elasticity of informal housing is - 0.49. The negative supply elasticity of informal housing may seem counterintuitive to the law of supply. But, a negative informal hous- ing supply elasticity is consistent with developing countries’ redevelopment processes studied in the literature (Henderson et al., 2021). More specifically, the negative informal supply elasticity indicates the existence of urban gentrification in Indian cities, which occurs in two ways. First, a simultaneous increase in rents paid by slum dwellers and land values around slums attract real estate developers. Slums are cleared for the construction of formal residential and commercial real estate space (Bhan, 2009). And second, slums are upgraded through various government and non-government programs that convert informal units to formal ones through in situ re- developments (Rains and Krishna, 2020; Rains et al., 2019). Finally, we estimate the elasticity of vacant formal residential housing units’ supply in urban India to be 2.62, which is larger than the supply elasticity of occupied formal housing units. The higher vacant elasticity reflects the fact that developers were engaged in speculative construction with the expectation of higher demand as market rents went up during the 2000s (Gandhi et al., 2022). 53 We use district-level variation to estimate formal housing supply elasticities for the 12 largest states of India. The state-level elasticities reveal substantial hetereogeneity across regions of India. The most supply-elastic state of Maharashtra is as elastic as cities in the relatively elastic state of Texas in the US. On the other hand, the least supply elastic state of West Bengal has substantially lower elasticity values than the least supply elastic metropolitan areas (Miami and Los Angeles) in the US. The heterogeneity in state-level elasticities underscores the role of land use regulations and institutional frictions in limiting the housing supply across different parts of India. Our contributions to the academic literature are three-fold. First, we exploit the Rosen-Roback framework to construct novel housing demand shifters. Prior research has used migration shocks such as international immigration (Saiz, 2010) and imputed migration based on historic flows (Paciorek, 2013) as housing demand shifters. Some papers have also used labor demand shocks in the form of shift-share instruments as housing demand shifters (Baum-Snow and Han, 2019; Paciorek, 2013; Saiz, 2010). A recent paper has used shift-share type instruments constructed from international capital flow shocks (Gorback and Keys, 2020). The problem with using migra- tion shocks as housing demand shifters is that migration decisions are endogenous to potential migrant destinations’ house prices and rents (Zabel, 2012). On the other hand, recent literature has shown that the identification of effects using shift-share instruments may not always be from exogenous shocks (Borusyak et al., 2022; Goldsmith-Pinkham et al., 2020). Therefore, shift-share instruments may not always satisfy the exogeneity assumptions required for their validity. We overcome the identification issues in the existing literature with better instruments. The strengths of our instruments are two-fold. First, we use migration-inducing shocks as demand shifters instead of migration. Shocks that affect rents and incomes only in the region of incidence are plausibly unrelated to non-migration factors that affect rents in another region. And second, by separating the shock-experiencing regions and the regions where we estimate the housing supply elasticities, we reduce the pathways through which omitted variable bias could occur as a result of the correlation between the shocks and the unobservables affecting housing supply. It may be hard, for instance, to argue that rainfall shocks, such as floods, are uncorrelated with housing supply within a given region. But, the geographical separation of housing markets and shock-experiencing regions alleviate such concerns. The spatial equilibrium framework provides the theoretical basis for identifying the indirect impact of such distant events on local housing 54 market outcomes, with migration being the channel of the impact. This idea is resonated in Boustan (2010), who used agricultural production shocks in the southern US as instruments for the Black Migration to the northern US during the post-war period. The second contribution in this chapter is providing a policy-relevant housing supply elastic- ity estimate for a large and urbanizing country like India. Prior academic literature has predomi- nantly focused on developed countries like the US (Baum-Snow and Han,2019; Green et al.,2005; Saiz, 2010). 8 Many studies have underscored the roles of regulations (Diamond, 2017; Glaeser et al., 2005a; Quigley and Raphael, 2005) and natural land constraints like hilly terrains (Saiz, 2010) in reducing the supply elasticity of housing in metropolitan areas of the US. Similar supply constraints also exist in developing countries like India. The land and housing markets in Indian cities are heavily regulated with floor-area-ratio (FAR) limits, ceilings on vacant land ownership, and rent control laws. 9 Studies have found that these regulations impose significant building costs on developers (Bertaud and Brueckner, 2005; Brueckner and Sridhar, 2012; Gandhi et al., 2021). Therefore, formal housing supply elasticity estimates almost surely reflect land-use policy decisions. Last but not least, we estimate the supply elasticity of informal housing, which is both an aca- demic contribution and a policy-relevant parameter for a developing country like India. Informal housing has been studied in the literature because its existence is associated with poverty (Marx et al., 2013) and institutional frictions, such as lack of property rights (Brueckner and Selod, 2009) and formal housing regulations (Henderson et al., 2021). Niu et al. (2021) underscored the critical role played by informal housing markets in reducing urbanization costs in Chinese cities by providing low-income migrants with cheaper housing. In similarly developing countries like India, informal housing accounts for a formidable share of the overall stock. Hence, without an informal housing supply elasticity estimate, our understanding of housing markets in the context of a developing country will be incomplete. To the best of our knowledge, this chapter provides the first informal housing supply elasticity estimate using direct observations on quantities and 8 Some studies have estimated housing supply elasticities in other countries such as Australia (McLaughlin, 2012), China (Wang et al., 2012), Italy (Accetturo et al., 2021), and the United Kingdom (Malpezzi and Maclennan, 2001). 9 The FAR of a building is equal to its total floor area divided by the area of the land parcel on which it is built. Lower FAR values indicate lower building height, and hence, stricter building regulations. Vacant land ownership was restricted in the largest urban areas of India under the urban land ceiling laws between the 1970s and the 2000s. These laws required firms and individuals to sell vacant land beyond a ceiling limit to the government at below-market prices (Sridhar, 2010). The urban land ceiling laws are studied in Chapter 3. 55 rents. The closest attempt at estimating an informal housing supply elasticity figure has been made by Niu et al. (2021) for Chinese cities. However, they calculate proxy informal housing supply elasticities using shares of village areas in the total urban built-up area on the edges of cities. Furthermore, we show that the informal housing supply elasticity in India is negative, which provides evidence in support of the existence of gentrification that occurs through informal- to-formal redevelopments in developing countries. The existing literature on informal housing has theorized that land parcels with informal housing units are redeveloped over time as rents in developing countries’ cities increase (Henderson et al., 2021). A recent study has also shown how the existence of informal housing can facilitate formal housing development through the easing of land and regulatory constraints (Guedes et al., 2021). However, to the best of our knowledge, empirical evidence on the conversion of informal units into formal houses at a large scale does not yet exist in the literature. Our evidence on gentrification fills this gap. The rest of this chapter is organized as follows. In section 2.2, we describe the data used for analysis and present some stylized facts about housing and migration in India in section 2.3. Sec- tion 2.4 provides a theoretical discussion of the Rosen-Roback spatial equilibrium setting applied in this chapter. Section 2.5 presents the empirical implementation of the theoretical mechanisms and the housing supply elasticity estimates. We present a discussion of the instruments and robustness checks in section 2.6. After a short discussion of policy implications in section 2.7, we provide concluding remarks in section 2.8. 2.2 Data For our analysis, we gather data from the National Sample Survey Organization (NSS), the Cen- sus of India, and the India Meteorological Department (IMD). We construct datasets at the state and district levels. We use the state-level datasets to study inter-state migration and population growth. Using district-level data, we study the impact of distant state shocks on local district- level outcomes. We estimate our elasticity figures using the district-level datasets. We construct a wide form panel for both datasets based on variable values from the Census years 2001 and 2011, which we then use to construct first-differenced variables for the analysis. In this section, we first 56 discuss the definition of informality of housing before briefly describing the datasets used in the analysis. 2.2.1 Informality of housing We use an approach to identify informal and formal housing similar to the UN-Habitat’s very general definition of slum and non-slum settlements. While the UN-Habitat classifies settlements as slums based on a range of living conditions, such as access to clean drinking water, sanitation, hygiene, good-quality housing, etc., we focus only on the quality of housing. We classify houses as informal if their roofs or walls are made of non-durable substances such as grass, thatch, bamboo, plastic, polythene, mud, unburnt brick, and wood. On the other hand, formal houses’ roofs and walls are made of durable substances like galvanized iron, metal, asbestos sheets, burnt bricks, stone, and concrete. Our definition of informality is closely linked to the Indian Census definitions of katcha and pucca (synonyms for non-durable and durable housing, respectively) and have been used by re- searchers in the past to identify informal houses in India (Gechter and Tsivanidis, 2017). Note that katcha or non-durable houses are not synchronous with slum housing in India. This is because there is a gap between the government’s definition of slums and the actual living con- ditions of those residing in slums and non-slum neighborhoods (Rains and Krishna, 2020; Rains et al., 2019). Therefore, using the durability of housing as an indicator for informal provides consistency in the measurement of the informal housing stock and may even be more accurate than the government-defined slum population figures in India. 2.2.2 State-level data A brief overview of the structure of aggregated in-migration data available from the Census of India is given in section 1.3.1 of chapter 1. To reiterate, the decennial data on in-migration from the Census provides the number of migrants by their time of movement (i.e., less than a year ago, 1-4 years ago, and 5-9 years ago), the distance migrants traveled from their last place of residence (inter-district, inter-state, etc.), the sector of origin (rural or urban), and their current place of residence (urban or rural). We use this information to construct decadal inter-state migration variables based on the number of individuals who moved into urban areas of a state from both 57 Table 2.1: Summary statistics of state-level variables 2001 2011 Mean SD Mean SD Variable (1) (2) (3) (4) Rainfall deficit months last decade 58 12 64 11 Inter-state migrants into urban last decade ('000) 319 543 452 691 Urban population (millions) 8 11 11 14 Mean monthly per capita consumption (INR) 890 274 1001 374 Urban surface area (sq. miles) 873 1,119 2,321 2,719 N 35 35 35 35 Source: Authors’ calculations based on Census of India, Labor Bureau of India, and National Sample Survey Organi- zation. Note: Table presents summary statistics of state-level variables. Migration given in thousands, population in mil- lions, and urban surface area in square miles. All values rounded off to the nearest integer. Real monthly per capita consumption in 2001 Indian National Rupees (INR) is calculated using the Consumer Price Index (CPI) data from the Labor Bureau of India. In PPP terms, $1 = 10 INR in 2001. For exchange rates, see: https://data.oecd.org/conversion/purchasing-power-parities-ppp.htm. rural and urban areas of another state in the decades leading up to the Census years, 2001 and 2011. The Census datasets also provide the urban population and the urban surface area for a given state. We obtain data on the mean monthly per capita consumption from the National Sample Sur- vey Organization and calculate real values based on the Consumer Price Index data provided by the Labor Bureau of India. We get our data on the expansion of National Highways Development Project Phase I, also known as the Golden Quadrilateral (GQ) highway upgrade program, from Ghani et al. (2016). And finally, we gather rainfall shock data from the Open Government Data (OGD) portal of the central government of India. 10 This dataset is sourced from the India Meteo- rological Department. It reports the percentage deviation of rainfall from the long-term average on a monthly basis between 1901 and 2015. We use this data to construct rainfall shock variables at the state level. 11 The summary statistics of all state-level variables are given in table 2.1. 10 The data can be downloaded from the following link: https://data.gov.in/. 11 The original data provides rainfall departure percentages for each of the 36 meteorological subdivisions in India. Meteorological subdivisions are roughly analogous to the state boundaries of India, with a few exceptions. Larger states consist of more than one subdivision, while some smaller states are clustered into one subdivision. We map these meteorological subdivisions to state boundaries and recalculate the rainfall departure values at the state level. 58 Table 2.2: Summary statistics of district-level variables 2001 2011 Mean SD Mean SD Variable (1) (2) (3) (4) Urban population ('000) 1,181 1,777 1,515 2,184 Informal housing units ('000) 46 42 46 39 Formal housing units ('000) 180 311 284 446 Vacant housing units ('000) 28 56 44 81 Mean informal rent (INR) 302 223 311 233 Mean formal rent (INR) 628 276 751 332 Mean rent for all residential houses (INR) 570 246 684 307 Mean monthly per capita consumption (INR) 1,051 247 1,110 338 Urban surface area (sq. miles) 104 126 134 142 Median no. of rooms per house 2 0 2 0 N 35 35 35 35 Source: Authors’ calculations based on Census of India, Labor Bureau of India, and National Sample Survey Organi- zation. Note: Table presents summary statistics of district-level variables. Population and housing units given in thou- sands, and urban surface area in square miles. All values rounded off to the nearest integer. Real monthly per capita consumption and rents in 2001 Indian National Rupees (INR) are calculated using the Consumer Price In- dex (CPI) data from the Labor Bureau of India. In PPP terms, $1 = 10 INR in 2001. For exchange rates, see: https://data.oecd.org/conversion/purchasing-power-parities-ppp.htm. 2.2.3 District-level data We obtain data on the number of various types of residential housing units—informal, formal occupied, and formal vacant—at the district level from the Census of India. In addition, we get data on district-level urban population and urban surface area from the Census. We also gather data on district-level mean per capita consumption and the mean housing rents for the various types of housing units in our analysis from the National Sample Survey Organization. These rent and consumption variables are inflation-adjusted to 2001 values based on the Consumer Price Index data provided by the Labor Bureau of India. Our final district-level data consists of 144 districts. The number of districts reduces in two ways. First, we recreate the actual administrative district boundaries to obtain time-consistent hypothetical boundaries because district boundaries are realigned in India very frequently. 12 This process is discussed briefly in section 1.3.1 of chapter 1. We lose some districts as a result of the redrawing of district boundaries. And second, we have data on mean informal housing 12 Between 2001 and 2011, the number of districts went up from 593 to 640. New districts are typically carved out of existing districts. 59 rent values for fewer districts. We estimate the housing supply elasticity of formal units using a larger sample of 339 districts and present the results in table A.8. The summary statistics of all district-level variables are given in table 2.2. 2.3 Stylized facts Housing markets in India have been understudied in the academic literature. Some quantitative papers on India have examined demand-side and affordability aspects of housing (Tiwari and Parikh, 1998; Tiwari et al., 1999), new housing construction (Dutta et al., 2021b; Gandhi et al., 2021), and rent control laws (Gandhi et al., 2022). A recent study has shown that rent control laws exacerbate rural-urban inequality in India by restricting rural-urban migration (Dutta et al., 2022). To the best of our knowledge, there does not exist other quantitative studies on housing and migration in India. Hence, the link between internal migration and housing in India is not well understood. In this section, we provide some stylized facts about the relationship between housing and migration in India. We argue that the growth in inter-state migration in India between the 1990s and the 2000s led to a shift in the demand for housing in India’s urban housing markets. Recall from figure 1a that inter-state movements in India constituted about 26% of overall decadal migration during the 2000s (16 million out of 62). Furthermore, table 2.1 indicates that state-level average inter-state migration to urban areas during the 2000s was about 5% of the average state’s urban population in 2011 (0.45 out of 11 million). Even though the volumes of inter-state migration are low, they are not insignificant, especially when we look at these numbers from a growth perspective. Figure 1a indicates that inter-state migration grew from 11 million to 16 million between the 1990s and the 2000s, a non-trivial growth of over 45%. To further alleviate concerns that inter-state migration’s impact on state-level urban popula- tion growth may be low, we plot the log of formal and informal housing units against the log values of inter-state migration at the state level. Figures 2.2a and 2.2b present the scatter plots. It is evidently clear that there is a strong correlation between the size of the housing stock and the number of inter-state migrants received by states. A relevant point is that the relationship between increased inter-state migration and housing 60 Figure 2.2: Inter-state migration and housing in urban India Jammu & Kashmir Himachal Pradesh Punjab Chandigarh Uttarakhand Haryana NCT of Delhi Rajasthan Uttar Pradesh Bihar Sikkim Arunachal Pradesh Nagaland Manipur Mizoram Tripura Meghalaya Assam West Bengal Jharkhand Odisha Chhattisgarh Madhya Pradesh Gujarat Daman & Diu Dadra & Nagar Haveli Maharashtra Andhra Pradesh Karnataka Goa Lakshadweep Kerala Tamil Nadu Puducherry Andaman & Nicobar Islands Jammu & Kashmir Himachal Pradesh Punjab Chandigarh Uttarakhand Haryana NCT of Delhi Rajasthan Uttar Pradesh Bihar Sikkim Arunachal Pradesh Nagaland Manipur Mizoram Tripura Meghalaya Assam West Bengal Jharkhand Odisha Chhattisgarh Madhya Pradesh Gujarat Daman & Diu Dadra & Nagar Haveli Maharashtra Andhra Pradesh Karnataka Goa Lakshadweep Kerala Tamil Nadu Puducherry Andaman & Nicobar Islands 7 10 13 16 19 Log urban informal housing units 7 10 13 16 19 Log inter−state in−migrants in urban areas 2001 Fitted line 2001 2011 Fitted line 2011 (a) Urban informal units and in-migrants Jammu & Kashmir Himachal Pradesh Punjab Chandigarh Uttarakhand Haryana NCT of Delhi Rajasthan Uttar Pradesh Bihar Sikkim Arunachal Pradesh Nagaland Manipur Mizoram Tripura Meghalaya Assam West Bengal Jharkhand Odisha Chhattisgarh Madhya Pradesh Gujarat Daman & Diu Dadra & Nagar Haveli Maharashtra Andhra Pradesh Karnataka Goa Lakshadweep Kerala Tamil Nadu Puducherry Andaman & Nicobar Islands Jammu & Kashmir Himachal Pradesh Punjab Chandigarh Uttarakhand Haryana NCT of Delhi Rajasthan Uttar Pradesh Bihar Sikkim Arunachal Pradesh Nagaland Manipur Mizoram Tripura Meghalaya Assam West Bengal Jharkhand Odisha Chhattisgarh Madhya Pradesh Gujarat Daman & Diu Dadra & Nagar Haveli Maharashtra Andhra Pradesh Karnataka Goa Lakshadweep Kerala Tamil Nadu Puducherry Andaman & Nicobar Islands 7 10 13 16 19 Log urban formal housing units 7 10 13 16 19 Log inter−state in−migrants in urban areas 2001 Fitted line 2001 2011 Fitted line 2011 (b) Urban formal units and in-migrants Source: Author’s calculations based on Census of India. Note: Figure in panel (a) presents a scatter plot of the log of state-level urban informal housing units and the log of inter-state migrants living in urban areas. The regression lines have slopes of 0.72 and 0.67, respectively, for 2001 and 2011, significant at 99%. Figure in panel (b) presents a scatter plot of the log of state-level urban formal housing units and the log of inter-state migrants living in urban areas. The regression lines have slopes of 0.80 and 0.71, respectively, for 2001 and 2011, significant at 99%. 61 Figure 2.3: Share of decadal migrants in urban population of India 13 16 4 4 9 12 0 5 10 15 20 Migrants in urban population (%) All internal Inter−state Intra−state 1991−2001 2001−2011 Source: Author’s calculations based on the Census of India. Note: Figure presents the urban population’s share of rural-urban and urban-urban migrants that moved during 1991- 2001 and 2001-2010 by migrants’ last residence (same or different state). Bars labeled with their corresponding values. demand could reflect structural transformation. However, despite its increasing levels, migration is still a small share of the urban population. Figure 2.3 shows that, at the national level, the urban population’s share of migrants increased very little during this time. Inter-state migrants as a share of India’s urban population remained flat at 4% between 2001 and 2011. 13 Therefore, the increasing volumes of migration do not reflect structural transformation. Another issue is that prior literature suggests that a major share of internal migrants to Indian cities move into slums and not formal housing (Mitra, 2010; Srivastava, 2011). These are usually poorer Indians from rural areas who move seasonally for one to six months to supplement farm incomes with urban informal earnings during lean agricultural seasons before moving back to their homes (Imbert and Papp, 2015, 2020; Rosenzweig and Udry, 2014). Hence, migration- inducing shocks may more likely capture informal rather than formal housing demand shifts. However, other studies have found that slums do not predominantly consist of migrants (Rains and Krishna, 2020). Besides, even if it is true that a larger share of relatively less wealthy 13 Not to be confused with the 5% share of state-level average inter-state migration in state-level average urban popu- lation. 62 Indians moves on a short-term basis, many affluent Indians also migrate and do so permanently rather than seasonally. For instance, the NSS data on employment and migration indicates that while about 12% of households had a seasonal migrant, about 27% of households reported that a former member had moved out permanently for employment or education. Moreover, the NSS data also indicates that more affluent and highly educated households are more likely to send a migrant out permanently and less likely to send migrants on a seasonal basis, which is consistent with past literature’s findings (Morten, 2019; Munshi and Rosenzweig, 2016). It is plausible that individuals from such affluent households who move permanently across regions choose formal housing over slums. 2.4 Theory We use the Rosen-Roback spatial equilibrium framework (Roback, 1982; Rosen, 1979) to analyze the effect of distant region shocks on inter-regional mobility and local housing demand. A shock that affects rents and incomes within a distant region induces spatial disequilibrium, spurring inter-regional mobility. Such mobility affects local housing demand if net inward mobility to the local region is non-zero. Therefore, distant region shocks that affect rents and incomes in the distant region act as demand shifters and can be used to estimate the local housing supply elasticity. In this section, we provide an analytical discussion of these effects. 2.4.1 Spatial equilibrium Consider an economy with a local region i where we are interested in estimating the housing supply elasticity and a distant region j that witnesses exogenous shocks to its economy. Consis- tently throughout the chapter, we refer to the distant region with the index j and the local region with the index i. The number of individuals occupying regions i and j are n i and n j , respectively. We assume that each individual is equivalent to a household in either region. 14 In both locations, individuals earn wage w and derive utility from housing services h, a numeraire consumption good c, and location-specific exogenously-given amenities a. The preference structure is assumed to be strictly quasi-concave. 14 As long as the number of households and the total population at i is monotonically related, relaxing this assump- tion will not alter the model mechanisms. 63 The market-clearing rent for housing services is given by r. The user-cost model relates r to the market-clearing house price p through the equation: r = p(K+ T+ D+ E). Here, K is the cost of capital, T is the property tax rate, D is the rate of depreciation, and E is the rate of expected appreciation (Poterba, 1984). The fact that market-clearing rent for housing services is an appropriate measure of market-clearing price for housing as a composite commodity is well established in the literature (Brueckner et al., 1987; Mills, 1967). Solving the consumer’s utility maximization problem, we get the demand for housing ser- vices, h d i (r i , w i ) and h d j (r j , w j ). The aggregate demand for housing, H D i and H D j , are products of total populations and consumers’ housing demand and can be expressed as follows: H D ι = n ι h d ι (r ι , w ι ) where ι={i, j} (2.1a) The utility-maximizing demand functions also provide the indirect utilities, V ι which, in turn, gives us the spatial equilibrium condition: V i (r i , w i , a i )= V j (r j , w j , a j )= ¯ V (2.1b) The values of r and w adjust, such that, conditional on amenities, the indirect utility is equal across both regions i and j. At this equilibrium, there are no gains to mobility between i and j. 2.4.2 Spatial disequilibrium Now consider a shock z j at the distant region j that does not affect amenities but changes the rent or income, or both, thus changing the utility of individuals at j. 15 The shock z could be a negative shock like a drought or a positive shock like a highway upgrade program. Because z j affects rent r j and income w j , it follows that V j (r j (z j ), w j (z j ), a j ) is an implicit function of z j . Hence, in response to z j , we have a state of spatial disequilibrium: ˜ V j = V j (z j )= V j (r j (z j ), w j (z j ), a j )̸= ¯ V = V i (2.1c) 15 The impact of z on mobility m between i and j will be determined by the implicit function m(V(r(z), w(z), a(z))). Hence, assuming that a remains exogenous to z does not alter the main mechanisms. Note, however, that while mobility will respond to r(z), w(z), and a(z), and r and w will, in turn, change in response to such mobility changes, a will not. In other words, a(z) changes only in response to z. 64 Since there are gains to mobility because of the difference in V i and V j , the shock z j will induce mobility between j and i until r and w adjust in both i and j, so that ˜ V j = ˜ V i = ˜ V. In other words, a shock affecting rents and incomes at a distant region j induces movement between the distant and the local regions so that rents and incomes change in both locations until spatial equilibrium is restored and there are no gains to moving. Boustan (2010) employed the same theoretical idea to estimate the impact of agricultural shocks in the southern US on the Black Migration in the post-war period. The empirical literature has also examined how labor and housing market shocks affect inter-regional mobility in the United States (Molloy et al., 2011; Saks and Wozniak, 2011). 2.4.3 Mobility We characterize mobility m between regions i and j as the vector (m ji , m ij ). m ji represents the number of individuals moving from j to i and m ij denotes the number of individuals moving from i to j. In other words, m ji represents in-migration from the distant region j to the local region i and m ij represents out-migration from local region i to the distant region j. This idea is consistent with the bi-directional movement of individuals across regions observed in the data. Note that the populations at i and j are functions of the migration vector (m ji , m ij ). On the one hand, in-migration from j to i increases the population at i and decreases the population at j. On the other hand, out-migration from i to j decreases the population at i and increases the population at j. Hence, for the local region i, we have: ∂n i ∂m ji (z j ) ≥ 0 and ∂n i ∂m ij (z j ) ≤ 0 (2.2a) The weak inequality follows from the fact that the natural rate of growth component in popu- lation changes is a major factor and can act as a countering force to both in- and out-migration effects on the local population. Since we do not explicitly model the natural rate of growth component in dn i , we allow for the possibility that population changes could be independent of migration. We put some more structure into the way we characterize mobility. We allow for migration to be an implicit function of the shock. To see this, suppose that m ij (.) and m ji (.) are two distinct 65 functions of the indirect utilities V i and V j . Spatial equilibrium implies that the net movement between two regions in equilibrium should be equal to zero. Therefore, we will have m ji ( ¯ V, ¯ V)= m ij ( ¯ V, ¯ V) in spatial equilibrium which implies net zero movement between i and j. Now, in response to the shock z j , the indirect utility at the distant state j changes from ¯ V to V j (z j ). The resulting migration functions can be written as follows: m ij ( ¯ V, V j (z j ))= m ij (z j ); m ji ( ¯ V, V j (z j ))= m ji (z j ) (2.2b) Equation (2.2b) implies that both in- and out-migration at i are implicit functions of the shock z j . 2.4.4 Housing demand Before we infer anything about the ways in which a distant shock may affect housing markets in the local region, we make the following assumptions: Assumption 2.1. In-migration from j to i and out-migration from i to j increase weakly with the distant shock z j . Assumption 2.2. If either in-migration from j to i or out-migration from i to j are unaffected by the shock z j , then mobility increases strictly with the distant shock z j in the direction in which the shock has a non-zero impact. Assumption 2.3. If the distant shock z j affects rents and wages at i, then such effects are only through migration. Assumption 2.2 implies that the shock z j should affect movement in at least one direction between i and j. Together, assumptions 2.1 and 2.2 restrict the universe of shocks to only those with a non-trivial effect on net mobility across regions. In terms of the empirical estimation, these two assumptions imply that in order for the shocks to be strong instruments for local housing demand, they need to have a strong impact on migration. Assumption2.3 is akin to the exclusion restriction in the empirical estimation. It implies that the distant shock cannot affect local rents and wages through any channel other than migration. 66 Proposition 2.1. Under the assumptions 2.1 to 2.3, a shock z j in the distant region j changes the demand for housing in the local region i if and only if the shock has a net non-zero impact on mobility between i and j. Proof: See appendix C. Proposition 2.1 means that the aggregate demand for local housing services responds to migration-inducing shocks at the distant region if the shocks produce non-zero mobility between the distant and the local regions. The intuition here is that the distant shocks affect mobility which, in turn, affects population changes in the local region. These population changes lead to a shift in the demand for housing. Proposition 2.2. Under the assumptions2.1 to2.3, a shock z j in the distant region j increases the demand for housing in i if and only if its magnitude of effect on in-migration from j to i outweighs its magnitude of effect on out-migration from i to j, and vice versa. Proof: See appendix C. Adding to the intuition of proposition 2.1, the idea in proposition 2.2 is that the demand for housing changes in the direction in which there is net mobility. In other words, if the shock leads to net non-zero mobility towards the local region, i.e. net in-migration, then housing demand in the local region will increase. The opposite will happen if there is net outward movement, i.e. net out-migration, from the local region. Note that propositions 2.1 and 2.2 rule out the special case where mobility of the same magnitude is induced in both directions by the distant shock. In such a case, net mobility will be zero despite the fact that the distant shock affects mobility in both directions. Propositions 2.1 and 2.2 imply that a distant shock-inducing net non-zero inter-regional mo- bility acts as a demand shifter in local housing markets. The driving mechanisms behind the distant region shock effect on local housing demand can be described as follows. First, the dis- tant shock affects rents and incomes in the distant region. This, in turn, changes the indirect utility in the distant region, thereby inducing a state of spatial disequilibrium in the economy. The resulting difference in utilities across the two regions implies gains to mobility. Individu- als move across regions. This movement causes a change in the local population growth, thus affecting local housing demand. 67 It is important to note here that distant state shocks, such as highway upgrades, can lead to increased trading of commodities across states which can, in turn, affect commodity prices. A change in commodity prices will lead to substitution between housing and non-housing expen- ditures, causing a second channel of impact on housing demand. This is particularly true if both the distant and the local states were recipients of the highway upgrade program. However, it is unlikely that the highway upgrade would have had a long-term impact on consumers’ expen- ditures on housing and non-housing goods since long-term non-housing commodity prices are affected by other factors such as productivity. We empirically test whether the Golden Quadrilateral (GQ) highway upgrade in India had any impact on recipient states’ commodity prices by regressing price changes of baskets of non- housing commodities consumed by urban households in a state on the state’s GQ recipient status. We find that a state’s inclusion in the GQ program did not impact its commodity prices in urban areas. The results from this regression are given in table A.6. We, therefore, ignore the trade channel of the impact of distant states’ inclusion in the GQ program on local housing demand. 2.4.5 Housing supply Suppose that the total housing stock H S i in the region i is supplied through a competitive market. In market equilibrium, we have H S i = H D i (r i , w i , n i ). Assuming that the supply function is log- linear, the reduced form for the inverse supply function at i can be written as follows: log(r i )= 1 η i log(H S i ) (2.3a) where, the housing supply elasticity at i isη i . Since the housing supply is never perfectly elastic, η i is a finite real number greater than zero. 16 Estimatingη i in equation (2.3a) presents a classic endogeneity problem since we only observe market equilibrium values of r i and H S i . Hence, we need exogenous demand shifters to trace the slope of the inverse supply curve. Propositions 2.1 and 2.2 indicate that exogenous shocks z j incident upon a distant region can act as a demand shifter at i if the shock z j induces net non-zero mobility between i and j. We can write the reduced form effect of z j on the aggregate demand 16 See Green et al. (2005) for a discussion on imperfect housing supply elasticities and the various reasons why that is the case in the context of a monocentric city model. 68 for housing services as follows: log(H D i )= βz j (2.3b) Proposition 2.2 implies that β could be either negative or positive, and its sign depends on the relative magnitude of the in- and out-migration effects of the shock z j . The predicted log(H D i ) ob- tained after estimating the parameter β is an exogenous demand shock which can be substituted in equation (2.3a) to estimate η i . The predicted log(H D i ) may not be exogenous if the shock-induced migration leads to labor market changes in the local region and, in turn, affects construction wages. In such a case, β may include supply-side factors as well, a concern we address with robustness checks in section 2.6.2. In our analysis, we hypothesize that z j is a demand shifter for all kinds of residential houses —formal occupied, informal, and formal vacant. However, their slopes will differ since these housing categories represent different markets. In other words, the coefficient β will be differ- ent for the three different types of residential housing used in our analysis. Therefore, in our empirical exercise, we will estimate three supply elasticity figures. 2.5 Empirical implementation In this section, we examine the hypothesis in propositions2.1 and2.2 through a series of empirical tests. First, we estimate the impact of two distant shocks, droughts and highway infrastructure investments, on inter-state mobility and the subsequent impact of such mobility on local urban population growth. Next, we use the same shocks as instruments for local urban population growth and estimate the urban population growth impact on local housing demand. Finally, we use the shocks as demand shifters and estimate the supply elasticities for formal occupied, informal, and formal vacant housing. 2.5.1 Defining instruments Based on our spatial equilibrium framework, discussed in section 2.4, we construct two instru- ments for all endogenous independent variables in the empirical estimation. The instruments are 69 two shocks occurring at the distant state j. The first shock is the number of drought spells, defined as the decadal change in the number of rainfall deficit months leading up to the Census years, 2001 and 2011. A state is rainfall deficit in a given month if the state’s rainfall levels in the month are less than 80% of the long-term normal. This definition of rainfall deficiency is based on the classifications used by the India Meteorological Department (IMD) to designate regions as rainfall deficit and has been used in the academic literature (see Bhavnani and Lacina (2017) and Jayachandran (2006)). The second shock is a dummy variable equal to one if the distant state was a recipient of the National Highways Development Project Phase I or the Golden Quadrilateral (GQ) highway upgrade program. The GQ program upgraded preexisting highways connecting the metropolitan cities of India—Chennai, Delhi, Kolkata, and Mumbai—from two to four lanes. The program was initiated in 2001, covering 14 states and union territories of India. In figure A. 2, we present a map of the GQ-recipient states and union territories. 2.5.2 Distant shocks, inter-state migration, and urban population growth We empirically estimate the effect of distant state shocks on inter-state mobility and the resulting impact of mobility on a local state’s urban population growth. For this estimation, we use a wide form state-level panel data with two periods—Census years 2001 and 2011. An observation is equivalent to a state-state pair. We construct first-differenced variables from the wide-form panel. First differences are denoted with∆, which represents changes in variable values between 2001 and 2011. 2.5.2.1 Empirical strategy Consider the local state i and the distant state j. Migration ji represents the number of individuals moving from j to i and Migration ij denotes the number of individuals moving from i to j. Our goal is to isolate the impact of the bi-directional migration flows, Migration ji and Migration ij , on the local state’s urban population growth between 2001 and 2011. We estimate the following first difference equation using data on i− j pairs of Indian states 70 between 2001 and 2011: ∆Log Urban population i = λ 1 ∆Log Migration ji +λ 2 ∆Log Migration ij +λ 3 ∆x i +υ ij (2.4a) where, x i is a vector of controls that include the log of per capita consumption (as proxy for income), urban surface area, and urban surface area squared at i and υ ij is the error term. All time-invariant unobservables will be eliminated in the first-difference set-up. However, inter-state migration flows are clearly endogenous to local urban population growth. Therefore, we instrument Migration ji and Migration ij with negative rainfall shocks and the GQ upgrade dummy occurring in distant states j, as defined in section 2.5.1. We estimate the impact of the two distant state shocks on migration flows using the following first-stage equation: ∆Log Migration k = µ 1 ∆Rain f all de f icit months j +µ 2 GQ dummy j +µ 3 ∆x i +ν k (2.4b) where, k = {ji, ij}, x i is as defined for equation ( 2.4a), and ν k is the error term. The identifica- tion of the coefficients µ 1 and µ 2 comes from the variation in rainfall shocks and the highway upgrade status between different states, respectively. As in equation (2.4a), all time-invariant unobservables are removed from equation (2.4b). 2.5.2.2 Covariates in estimation We discuss the inclusion of the covariates in equations (2.4a) and (2.4b). First, by including consumption as a proxy for income at i, we control for labor market equilibrium changes induced by urban population growth at i that is, in turn, caused by changes in migration flows between j and i. While we model the effect of changing population on the demand for housing at i in section 2.4.4, we do not say anything about the labor market effects of mobility at i. Since labor supply at i changes in response to the shock-induced mobility, we should expect the labor market equilibrium at i to reflect that. The resulting change in incomes will affect housing demand at i. Therefore, to control for such migration-induced income changes, we include consumption at i as a covariate in all regressions in our empirical analysis. Second, we include the urban surface area of i to account for changes in the total urban 71 surface area across states. Total urban surface areas of states and districts change over time because smaller urban-like settlements in India are reclassified and declassified as Census towns each Census year. 17 Since we have aggregated data for all urban areas, controlling for the urban surface area allows us to mitigate any effect on migration and urban population growth that can be attributed to the change in the urban surface area resulting from classification by Census authorities. We also control for urban surface area squared to account for the potentially non- linear relationship between urban surface area and the outcome variables. 2.5.2.3 Effect of inter-state migration on local urban population growth In the first stage, we estimate the impact of distant state shocks on inter-state migration. We then use the distant state shocks as instruments for migration and estimate migration’s impact on local urban population growth. The results are given in table 2.3. We observe a number of things in table 2.3. First, while the rainfall shock at j had a positive impact on migration from j to i, it did not affect migration from i to j. One additional rainfall deficit month at j increased migration from j to i by 1.1%. This is consistent with the literature that negative rainfall shocks spur outward mobility from affected regions in India (Bhavnani and Lacina, 2017; Bryan et al., 2014; Rosenzweig and Udry, 2014). Second, the highway upgrade at j had a positive and significant impact on migration from both j to i and i to j. A distant state’s inclusion into the GQ program increased migration from j to i by 27% and migration from i to j by 21%. On the one hand, the labor demand shock from firm relocation along the highway (Ghani et al., 2016) would have increased movement toward states included in the GQ program (Bartik, 1993). On the other hand, higher incomes, possibly resulting from the labor demand shock, would have spurred movement outward from those states because higher earners are better insured against uncertain migration outcomes (Bryan et al., 2014; Morten, 2019; Munshi and Rosenzweig, 2016). In the second stage, we see that an increase in migration from j to i caused the urban popu- lation at i to increase at the same rate as migration. But, migration from i to j had no impact on urban population growth at i, suggesting that the net effect of the distant shock-induced migra- tion was to cause an increase in urban population at i. One reason why out-migration from i to j 17 See appendix B.1 for details on urban administration in India and the definition of Census towns. 72 Table 2.3: Distant shock-induced migration and local urban population growth First-stage Second-stage Dependent variable =∆ Log indicators Migration 0-9 yrs. Urban population i← j i→ j at i ∆= 2011− 2001 (1) (2) (3) ∆ Log Migration 0-9 yrs. i← j 0.999*** (0.232) ∆ Log Migration 0-9 yrs. i→ j -0.365 (0.372) ∆ Rainfall deficit months at j 0.011*** -0.002 (0.003) (0.003) GQ highway dummy at j 0.269*** 0.211*** (0.043) (0.046) ∆ Log consumption at i -0.275* -0.069 0.451*** (0.159) (0.232) (0.166) ∆ Urban surface area at i 0.030 0.221*** 0.147* (0.028) (0.035) (0.081) ∆ Urban surface area at i squared 0.007*** -0.015*** -0.018** (0.002) (0.003) (0.007) F-stat on excluded instruments 29.9*** 10.4*** Anderson–Rubin Wald χ 2 (2) 254*** N 1,028 1,028 1,028 R 2 0.179 0.122 Source: Authors’ calculations. Note: Table presents two-stage least squares first-difference regression estimates using a sample of state-state pairs in India between 2001 and 2011. State i is the local region, and state j the distant. Excluding missing pairs, there are 1,028 i− j state pairs consisting of the 35 states and union territories in India. Migration i ← j is in-migration to i and i→ j is out-migration from i. Migrants are those who moved in the previous decade. Differences are calculated as first differences of values between the years 2001 and 2011. Log in- and out-migration are instrumented by the decadal change in the number of rainfall deficit months in state j and the GQ dummy, which equals one if state j was recipient of the Golden Quadrilateral highway upgrade program, and zero otherwise. Consumption, urban surface area, and urban surface area squared at i are additional covariates. Urban surface area unit in 1000 sq. miles. Robust standard errors presented in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. did not affect urban population growth at i is that our empirical estimation does not account for natural population growth, an empirical possibility we discussed in section 2.4. Studies have ar- gued that the natural growth component is a significant determinant of overall urban population growth in India (Mukhopadhyay et al., 2020). Therefore, it is quite likely that any out-migration effect on urban population growth is offset entirely by this natural growth component. We report the F-stat on excluded instruments and the Anderson-Rubin Wald test statistic. The F-stat values pass the threshold of 10, indicating that the instruments are not weak. The 73 Anderson-Rubin statistic indicates that the instruments are relevant. 2.5.3 Urban population growth and housing demand We estimate the impact of exogenous urban population growth on urban housing demand. We use a wide form district-level panel data with the two Census years, 2001 and 2011. Each obser- vation is a district-state pair. We use the wide-form panel to construct first-differenced variables. As in section 2.5.2,∆ denotes changes in variable values between 2001 and 2011. 2.5.3.1 Identification We estimate the impact of local urban population growth on the local demand for housing with the following equation: ∆Log Housing units i = γ 1 ∆Log Urban population i +γ 2 ∆ x ′ i +τ i (2.5a) where, x ′ i is a vector of covariates that consists of the median number of rooms, the log of district- level mean per capita consumption, the urban surface area, and the urban surface area squared and τ i is the error term. 18 A least square estimation of γ 1 in equation (2.5a) will yield inconsistent estimates because of the endogenous relationship between urban population growth and housing demand. Therefore, we instrument∆Log Housing units with negative rainfall shocks and the GQ highway upgrade dummy, defined in section 2.5.1. With data on i− j district-state pairs, we estimate the following first-stage equation: ∆Log Urban population i = π 1 ∆Rain f all de f icit months j +π 2 GQ dummy j +π 3 ∆x ′ i +ω ij (2.5b) where, x ′ i is as defined before and ω ij is the error term. The predicted exogenous population growth from equation (2.5b) is used to estimate equation (2.5a). 18 The type of housing can potentially determine the median number of rooms in a house, inducing a reverse causal effect of housing on median rooms. We address this endogeneity concern by running regressions without the median number of rooms as a covariate and find that the results are largely similar (see table A. 7). 74 2.5.3.2 Effect of local urban population growth on housing demand The results from estimating equations (2.5a) and (2.5b) with first difference regressions using data on i− j district-state pairs in India between 2001 and 2011 are presented in table 2.4. In the first-stage regressions, we find that both rainfall shocks and the highway upgrade program at the distant state j led to urban population growth at i. An additional rainfall deficit month at state j led to an increase in urban population by 1.1% at i. The inclusion of state j in the GQ program increased the urban population by 12% at i. This is consistent with the findings in table 2.3 that the shocks had a positive effect on migration from j to i and not on migration from i to j and that such in-migration from j to i led to urban population growth at i. In the second-stage regression results, we see that the distant state shock-induced urban pop- ulation growth had a positive impact on the growth of informal, formal occupied, and formal vacant housing units. A 1% increase in urban population led to a 0.24% increase in demand for informal houses, a 1.8% increase in demand for formal houses, and 2.4% increase in vacant houses. A higher impact of urban population growth on formal housing units’ growth compared to informal ones is consistent with the fact that individuals living in informal houses consume lower floor space than those living in formal houses. 19 The significant increase in vacant houses in response to urban population growth in India was partly due to developers engaging in spec- ulative construction with the expectation that future demand would be higher in response to rising urban population (Gandhi et al., 2022). We report three diagnostic tests—the F-stat on excluded instruments, the Anderson-Rubin Wald test statistic, and the Sargan Hansen J-statistic’s p-value. The F-stat is significantly higher than 10, indicating that the instruments are not weak. The significant Anderson-Rubin statistic supports the relevance condition. The J-statistic is reported because we have two instruments for one endogenous variable. The insignificance of the J-statistic suggests that the overidentification restrictions hold. 19 Data from the National Sample Survey Organization housing conditions survey conducted in 2012 across India indicates that while the average per capita floor space consumed by formal housing occupants was 77 square feet, floor area per person was 52 square feet among those living in informal houses. 75 Table 2.4: Distant shock-induced local urban population growth and housing demand First-stage Second-stage Dependent variable =∆ Log indicators Urban Housing units at i population at i Informal Formal Vacant ∆= 2011− 2001 (1) (2) (3) (4) ∆ Log urban population at i 0.237*** 1.85*** 2.42*** (0.024) (0.026) (0.035) ∆ Rainfall deficit months at j 0.011*** (0.000) GQ highway at j dummy 0.124*** (0.005) ∆ Log consumption at i 0.009 -0.210*** 0.040** -0.205*** (0.013) (0.013) (0.018) (0.020) ∆ Urban surface area at i 5.12*** 0.502** -1.46*** -2.89*** (0.202) (0.205) (0.225) (0.318) ∆ Urban surface area at i squared -5.51*** -1.38*** 1.13*** 1.13** (0.417) (0.324) (0.309) (0.464) ∆ Median no. rooms per unit at i -0.007 -0.142*** 0.035*** -0.030*** (0.008) (0.007) (0.007) (0.011) F-stat on excluded instruments 983*** Anderson–Rubin Wald χ 2 (2) 78.4*** 3281*** 2500*** Sargan-Hansen J-stat p-value 0.728 0.424 0.973 N 4,896 4,896 4,896 4,896 R 2 0.603 Source: Authors’ calculations. Note: Table presents two-stage least squares first-difference regression estimates using a sample of district-state pairs in India between 2001 and 2011. District i is the local region, and state j the distant. Excluding missing pairs, there are 4,896 i− j district-state pairs consisting of the 35 states and union territories and 144 districts of India in our sample. Differences are calculated as first differences of values between the years 2001 and 2011. Log urban population at i is instrumented by the decadal change in the number of rainfall deficit months in state j and the GQ dummy, which equals one if state j was recipient of the Golden Quadrilateral highway upgrade program, and zero otherwise. Consumption, urban surface area, urban surface area squared, and median number of rooms at i are additional covariates. Urban surface area unit in 1000 sq. miles. Robust standard errors presented in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. 2.5.4 Demand shifters and housing supply elasticity estimation We estimate the inverse supply elasticity of housing for urban India. We use the same wide form district-level panel data as in section 2.5.3 to construct first-differenced variables. Each observation is a district-state pair. As before,∆ denotes changes in variable values between 2001 and 2011. 76 2.5.4.1 Estimating equations To obtain the housing supply elasticity, ideally, we would like to estimate the following inverse supply equation: ∆Log Rent i = η∆Log Housing unit supply i +φ i (2.6a) where, η is the inverse supply elasticity and φ i is the error term. However, the quantity of housing units supplied is an unknown. We only observe the mar- ket equilibrium quantities of housing units. So, in reality, we can only estimate the following equation: ∆Log Rent i = α 1 ∆Log Housing units i +α 2 ∆x i +ε i (2.6b) where, x i is a covariate vector consisting of the log of district-level mean per capita consump- tion, urban surface area, and urban surface area squared and ε i is the error term. Consistently estimating the coefficient α 1 presents a classic endogeneity problem. To address the endogeneity problem, we need demand shifters. Propositions 2.1 and 2.2 provides the theoretical framework for using shocks occurring in a distant state j, that affect rents and incomes at j, as demand shifters for housing in district i. Sections 2.5.2 and 2.5.3 provides evidence in support of the channels through which we expect distant state shocks to act as demand shifters for local urban markets. Therefore, we use rainfall shocks and the GQ highway upgrade dummy, defined in sec- tion 2.5.1, as instruments for∆Log Housing units. We estimate the following first-stage equation using data on i− j district-state pairs in India between 2001 and 2011: ∆Log Housing units i = β 1 ∆Rain f all de f icit months j +β 2 GQ dummy j +β 3 ∆x i +ϵ ij (2.6c) where, x i is as defined before and ϵ ij is the error term. We estimate three sets of equations, one each for formal occupied, informal, and formal vacant housing units. There are two things to note here. First, contrary to the existing literature on housing supply 77 elasticity estimation, we do not control for construction cost in equations (2.6b) and (2.6c). This is because we do not have any data on construction costs at the district level in India. Second, a possible concern may arise due to the various rent control laws in Indian states that prohibit landlords from increasing rents (Harari, 2020). However, this is unlikely to be a cause for concern since our analysis uses a first difference estimation framework, and new rent control laws were not enacted in India after 2001. Amendments to the preexisting rent control laws did not have provisions that could affect rents paid by tenants (Gandhi et al.,2022). Hence, the first differences would mostly absorb the rent control law effects that were present from before 2001. 2.5.4.2 Housing supply elasticity estimates We estimate three supply elasticity figures, one each for the three different types of urban housing units in our analysis—informal, formal occupied, and formal vacant. We estimate the coefficients in equations (2.6b) and (2.6c) with first difference regressions using data on i− j district-state pairs in India between 2001 and 2011. The results are presented in table 2.5. In the first-stage regressions, we see that both rainfall shocks and the GQ highway upgrade program at a distant state j had a strong positive growth effect on all three types of housing units in district i. An additional rainfall deficit month in the distant state increased the demand for informal houses by 0.2%, formal occupied houses by 2.1%, and formal vacant houses by 2.7% in the local district. The inclusion of the distant state in the GQ program led to a 2.7% higher demand for informal houses, 23% higher demand for formal occupied houses, and 30% higher formal vacant houses in district i. These effects are consistent with the results from tables 2.3 and 2.4, which show that the distant shocks induced local urban population growth through increased inter-state migration and, therefore, led to higher housing demand in the local district. These results also empirically confirm propositions 2.1 and 2.2. The second-stage results provide the inverse supply elasticity estimates for urban India. First, we find that the inverse elasticity for informal houses is − 2.06 implying a decadal informal housing supply elasticity of− 0.49. A negative value for housing supply elasticity is contrary to the law of supply. However, a negative elasticity of supply for informal houses indicates that a process of urban gentrification or renewal is underway in Indian cities, where informal houses are converted to formal buildings. This sort of gentrification occurs in two ways. On the one 78 Table 2.5: Housing demand shifters and inverse supply elasticity estimation First-stage Second-stage Dependent variable =∆ Log indicators Housing units at i Rents at i Informal Formal Vacant Informal Formal Vacant ∆= 2011− 2001 (1) (2) (3) (4) (5) (6) ∆ Log informal units at i -2.06*** (0.564) ∆ Log formal units at i 0.617*** (0.037) ∆ Log vacant units at i 0.381*** (0.026) ∆ Rainfall deficit months at j 0.002*** 0.021*** 0.027*** (0.000) (0.001) (0.001) GQ highway at j dummy 0.027*** 0.226*** 0.300*** (0.007) (0.007) (0.011) ∆ Log consumption at i -0.195*** 0.054*** -0.181*** -0.515*** 0.453*** 0.397*** (0.015) (0.018) (0.026) (0.127) (0.041) (0.035) ∆ Urban surface area at i 1.88*** 7.96*** 9.53*** 6.67*** -2.18*** -0.011 (0.167) (0.322) (0.399) (1.408) (0.504) (0.424) ∆ Urban surface area at i -2.98*** -9.01*** -12.3*** -5.82*** 2.81*** 1.62** squared (0.368) (0.673) (0.812) (2.126) (0.759) (0.699) F-stat on excluded instruments 27.4*** 1699*** 1246*** Anderson–Rubin Wald χ 2 (2) 17.5*** 293*** 235*** Sargan-Hansen J-stat p-value 0.938 0.798 0.622 N 4,896 4,896 4,896 4,896 4,896 4,896 R 2 0.075 0.659 0.557 Source: Authors’ calculations. Note: Table presents two-stage least squares first-difference regression estimates using a sample of district-state pairs in India between 2001 and 2011. District i is the local region, and state j the distant. Excluding missing pairs, there are 4,896 i− j district-state pairs consisting of the 35 states and union territories and 144 districts of India in our sample. Differences are calculated as first differences of values between the years 2001 and 2011. Log housing units at i are instrumented by the decadal change in the number of rainfall deficit months in state j and the GQ dummy, which equals one if state j was recipient of the Golden Quadrilateral highway upgrade program, and zero otherwise. Consumption, urban surface area, and urban surface area squared at i are additional covariates. Urban surface area unit in 1000 sq. miles. Robust standard errors presented in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. hand, a simultaneous increase in rents paid by slum dwellers and land values around slums attract real estate developers. Slums are cleared for the construction of formal residential and commercial real estate space (Bhan, 2009). On the other hand, slums are upgraded through various government and non-government programs implementing in situ redevelopments (Rains and Krishna, 2020; Rains et al., 2019). The empirical evidence on gentrification is also consistent with existing economic theories about urbanization in developing countries. In a monocentric- 79 city model, Henderson et al. (2021) shows how informal houses are converted to formal houses when the rent gradient is driven up in a developing country’s city through productivity shocks. 20 Second, the inverse elasticity for formal occupied houses is 0.62 implying a decadal formal housing supply elasticity of 1.62. 21 This estimate is slightly lower than the estimate of 1.75 for the average metropolitan area in the United States (US) provided by Saiz (2010). Our estimate is also within the range of 1-3 for national housing supply elasticities proposed in the literature for the US (Gyourko et al., 2008). This indicates that housing markets in Indian cities during the 2000s responded at a pace similar to that in the US during 1970-2000. Considering that economists believe the US to be supply inelastic, however, India’s formal elasticity values indicate the existence of regulatory and institutional frictions that impede the new construction of houses. Finally, the decadal supply elasticity of vacant formal housing units is 2.62, which is substan- tially higher than that of occupied formal housing units. This is due to the fact that developers were engaged in speculative construction in India during the 2000s. Gandhi et al. (2022) ar- gues that a large number of investors were engaged in speculative home purchases during the 2000s. Developers responded to such speculation-driven higher market prices by building more units. Hence, as prices went up, the number of vacant houses in markets with speculative buyers increased. As in section 2.5.3.2, we report the F-stat on excluded instruments, the Anderson-Rubin Wald test statistic, and the Sargan Hansen J-statistic’s p-value in table 2.5. The F-stat is higher than 10 for all three first-stage regressions implying that the instruments are not weak. The Anderson- Rubin statistic’s significance suggests that the instruments are relevant. We report the J-statistics because we have two instruments for one endogenous variable in all three sets of regressions. The insignificance of the J-statistic means that the overidentification restrictions are true. 20 Note that a conversion of informal houses into formal buildings with rising rent gradient need not imply a negative informal supply elasticity. New informal houses are constructed even as redevelopments increase with the rent gradient. For a negative elasticity, the overall informal stock, that is, new informal houses net of redevelopments, will have to fall with rising rents. This is possible in two ways. On the one hand, if productivity shocks experienced by informal housing residents are lower in magnitude than those received by formal housing residents, the rent gradient will reflect that, and the new informal housing stock will grow at a slower pace than redevelopments. On the other hand, if informal houses are predominantly located in the outer city and transport costs increase at an increasing exponential rate, then the rent gradient near the city center will be steeper. Therefore, redevelopments will outpace new informal construction. 21 The regression sample used in table 2.5 is restricted to the 144 districts for which we have rent data for all types of houses. Using the full sample of 339 districts for which we have formal rents, we estimate that the supply elasticity for formal occupied units is 1.36, and the formal vacant housing supply elasticity is 1.78. We present these results in table A.8. 80 2.6 Discussion on instruments In this section, we discuss the identification issues concerning the two instruments, negative rainfall shocks and GQ highway upgrade program dummy, defined in section 2.5.1. First, we discuss the validity of the instruments. Then, we conduct some robustness checks to address endogeneity concerns. 2.6.1 Validity of instruments In any instrumental variable (IV) strategy, the instruments must satisfy the relevance condition and the exclusion restriction. We discuss how rainfall shocks and the GQ upgrade dummy satisfy these conditions. 2.6.1.1 Relevance For the relevance condition to hold, we need strong instruments, that is, the instruments will have to be highly correlated with the endogenous variables. The IV literature is extensive on how weak instruments can result in an inconsistent estimation of IV regressions. 22 We report the F- stat on excluded instruments and the Anderson-Rubin Wald chi-square test statistic in tables 2.3 to 2.5. All the reported diagnostics indicate that the instruments are not weak. Furthermore, prior literature’s findings on the economic impacts of negative rainfall shocks and GQ highway infrastructure in India are consistent with our results. Negative rainfall shocks act as negative income shocks in most parts of India due to largely rainfall-dependent agricultural practices. Hence, rainfall levels less than 80% of the long-term normal induce drought-like conditions and are unfavorable for agricultural output. A body of literature has examined the relationship between rainfall shocks and agricultural output (Jay- achandran, 2006), and the subsequent impact of rainfall shocks on migration (Morten, 2019; Rosenzweig and Udry, 2014). Rainfall shocks have also been used as an instrument to study civil conflict and dowry deaths in India (Sarsons, 2015; Sekhri and Storeygard, 2014). Bhav- nani and Lacina (2017) constructed an instrument from negative rainfall shocks to estimate the 22 The most recent study is by Mikusheva and Sun (2022) who propose a jackknifed version of the Anderson-Rubin weak IV test for the multiple-IV case. 81 effect of inter-state migration flows on fiscal federalism in India. Therefore, our empirical find- ing that negative rainfall shocks affect migration is not new to the literature. However, the fact that rainfall shock-induced migration can affect housing demand through population growth, and therefore, rainfall shocks in one region causing migration can be used as an instrument for population growth and housing demand in another, is novel. The GQ highway upgrade project has been documented in the literature as a positive eco- nomic shock since it affected firm relocation along the highway in states through which it passed (Abeberese and Chen, 2021; Ghani et al., 2016). Based on economic theory and evidence, we should expect two countervailing effects of the highway upgrade program in a state. First, due to firm relocation along the highways, we should expect to see a growth in employment in the recipient states, thereby, leading to migration towards such states (Molloy et al., 2011; Saks and Wozniak, 2011). And second, the firm and employment growth may also lead to a positive income shock in recipient states. 23 This would lead to more movement out of recipient states because higher incomes insure individuals against risky migration outcomes (Bryan et al., 2014; Morten, 2019; Munshi and Rosenzweig, 2016). Our empirical findings are consistent with these two countervailing effects of the GQ program on in- and out-migration. Therefore, both instruments used in the analysis are well grounded in theory and the existing literature. Given that the instruments also pass the diagnostics, we argue that the relevance condition holds. 2.6.1.2 Endogeneity concerns It is relatively harder to show that the instruments satisfy the exclusion restriction. For the most part, negative rainfall shocks and the GQ program implementation in one state are exogenous to population growth and housing markets in other states. There are some endogeneity issues that we discuss as well. The first endogeneity issue arises from the fact that both negative rainfall shocks and the GQ program implementation in one state can affect population growth and housing markets in 23 Whether the market clearing wage would change in GQ recipient states will depend on net migration resulting from the GQ program-induced labor demand shock. If net migration increased the labor supply more than the labor demand induced by the GQ program, then the market-clearing wage might as well have decreased. However, given that our estimates show out-migration resulting from the GQ program implementation, the likelihood of a wage increase is higher. 82 neighboring states. Such a spatial correlation of shocks would affect rents and incomes in the distant and local states. Therefore, the housing markets of local states will be affected through channels other than migration between the local and the distant states. Rainfall shocks may be spatially correlated because neighboring states are more likely to si- multaneously experience such shocks. For instance, monsoonal floods in the neighboring states of Bihar, West Bengal, and Orissa often occur simultaneously. Bhavnani and Lacina (2017) rec- ognizes this issue and controls for rainfall shocks in both migrants’ origin and destination states. Drought spells can also span multiple states at the same time. The inclusion of one state in the GQ program will not affect the inclusion of another because these highways were constructed on trade routes built during ancient and colonial times. For instance, one part of the GQ program, the National Highway II (NH2), was constructed on portions of the Grand Trunk Road that was first built by the emperor Chandragupta Maurya during the 3rd century BCE and later redeveloped under the rule of emperor Sher Shah Suri, the Mughals, and the British Raj (Elisseeff, 2000; Thapar, 2015). Contiguous states may still have a higher likelihood of being on the same trade route. Such contiguity effects will, however, be absorbed by the first-difference framework because there were no changes in state boundaries between 2001 and 2011. But, the GQ upgrade program implementation across contiguous states might have potentially affected housing supply if better contiguous-state road networks led to higher trading, and thus, reduced prices of construction material. 24 We address these endogeneity concerns by conducting robustness checks. In these robustness checks, we estimate equations (2.4a) and (2.4b) with negative rainfall shocks and the GQ program implementation occurring in only non-contiguous distant states as the instruments. The results from these regressions are discussed in section 2.6.2. Another endogeneity concern is that if the rainfall shocks and the GQ program implemen- tation in the distant state led to the migration of construction workers, there would have been housing supply effects in the local state. This is because a migration-induced labor supply shock in the construction sector would have led to changes in construction wages, thereby, affecting construction costs in the local region. This is particularly relevant in the Indian context since a 24 We expect construction material to be traded only across neighboring states. It is unlikely that such trading occurs across non-contiguous states since substances such as cement are heavy and difficult to transport. 83 large number of Indians migrate for construction work. However, such migrant workers are more likely to move seasonally for one to six months before moving back to their homes. The NSS survey on migration and employment (National Sample Survey Organization,2008) conducted in2007-08 indicates that36% of seasonal inter-state migrants move for construction work compared to only 1.5% of long-term inter-state migrants. Hence, if we eliminate short-run migrants from our analysis, we alleviate the endogeneity concern arising from the housing supply effects of distant state shocks. We conduct a second set of robustness checks by redefining the migration variables in equation ( 2.4a) to exclude short- run migrants who moved less than a year before the Census enumeration. In other words, the redefined migration variables exclude seasonal migrants. The results from this analysis are presented in section 2.6.2. The final concern with the identification is if negative rainfall shocks and the GQ program implementation led to the migration of firms as well as individuals across states. If firms moved too, then there may have been a restructuring of industries in regions that did not receive the shocks in ways that could have affected their population growth and housing supply. This, however, is unlikely to be a concern in our identification. First, negative rainfall shocks predominantly affect agricultural incomes in India (Jayachandran, 2006). Any sectoral realloca- tion of labor resulting from such shocks (Emerick, 2018) will not have likely led to the movement of firms. Second, past literature has shown that the GQ program led to firm concentration along the highways. Therefore, unless both the distant and the local states in our analysis were recip- ients of the GQ program, the endogeneity concern with firm relocation will not matter. In cases where both the distant and local states were recipients of the GQ program, only firm movement across non-contiguous states will be an untested identification concern. This is because our ro- bustness check with non-contiguous state shocks shows that contiguity effects do not drive our results. Any remaining concern with firms moving across non-contiguous states will not be a major one. This is because very few firms have the ability to relocate across non-contiguous states. There are considerable differences in property rights, laws, policies, regulations, culture, and languages spoken across states of India, and these differences widen with increasing distance between states. Therefore, the cost of firm relocation will be very high. Abeberese and Chen 84 (2021) found that only 12% of over 10,000 firms surveyed by the Annual Survey of Industries (ASI) between 1999 and 2007 relocated across districts of India, with the average firm moving only 42 miles (68 kilometers). In fact, the vast majority of firms in their sample moved across neighboring districts (mostly within states). Therefore, firm relocation as a result of rainfall shocks and the GQ program implementation is unlikely to be an endogeneity concern. 2.6.2 Robustness checks We conduct two robustness checks to address endogeneity concerns with the use of rainfall shocks and the GQ program implementation in the distant state as instruments for migration, urban population growth, and housing demand in the local state. In the first robustness check, we estimate the impact of non-contiguous distant state shocks on inter-state migration and the resulting impact of such migration on urban population growth in the local state. In the second, we estimate the impact of distant shocks on long-term migration. 2.6.2.1 Non-contiguous state shocks To address concerns that rainfall shocks and the GQ program implementation may have hous- ing supply effects across neighboring states, we conduct robustness checks by eliminating non- contiguous state pairs in our analysis. In the regressions using equations (2.4a) and (2.4b), we re- strict the universe of shocks to include only those that occurred in non-contiguous distant states. In other words, an observation in these regressions is equivalent to an i− j non-contiguous state pair. The coefficient estimates in table 2.6 are very similar in magnitude to those seen in table 2.3. Every additional rainfall deficit month in distant non-contiguous states increased migration from such states by 1%. A distant non-contiguous state’s inclusion in the GQ program increased migration from such states by 29% and to such states by 22%. Just as in table 2.3, local in- migration led to a positive significant impact on local urban population growth, whereas out- migration had no impact. In table A.9, we present the housing supply elasticity estimation results with non-contiguous state shocks and do not find any changes from the main estimates given in table 2.5. One contentious issue in the results given in table 2.6 is that the instruments for migration 85 Table 2.6: Distant non-contiguous states’ shock-induced migration and local urbanization First-stage Second-stage Dependent variable =∆ Log indicators Migration 0-9 yrs. Urban population i← j ′ i→ j ′ at i ∆= 2011− 2001 (1) (2) (3) ∆ Log migration 0-9 yrs. i← j ′ 1.03*** (0.261) ∆ Log migration 0-9 yrs. i→ j ′ -0.428 (0.424) ∆ Rainfall deficit months at j ′ 0.010*** -0.003 (0.004) (0.003) GQ highway at j ′ dummy 0.288*** 0.216*** (0.047) (0.051) ∆ Log consumption at i -0.276 -0.048 0.473** (0.168) (0.246) (0.189) ∆ Urban surface area at i 0.025 0.230*** 0.174* (0.031) (0.039) (0.096) ∆ Urban surface area at i squared 0.008*** -0.016*** -0.021** (0.003) (0.003) (0.009) F-stat on excluded instruments 26.2*** 9.01*** Anderson–Rubin Wald χ 2 (2) 241*** N 908 908 908 R 2 0.163 0.108 Source: Authors’ calculations. Note: Table presents two-stage least squares first-difference regression estimates using a sample of non-contiguous state-state pairs in India between 2001 and 2011. State i is the local region, and the non-contiguous state j ′ the distant. Excluding missing pairs, there are 908 i− j ′ non-contiguous state pairs consisting of the 35 states and union territories in India. Migration i← j is in-migration to i and i→ j is out-migration from i. Migrants are those who moved in the previous decade. Differences are calculated as first differences of values between the years 2001 and 2011. Log in- and out-migration are instrumented by the decadal change in the number of rainfall deficit months in state j ′ and the GQ dummy, which equals one if state j ′ was recipient of the Golden Quadrilateral highway upgrade program, and zero otherwise. Consumption, urban surface area, and urban surface area squared at i are additional covariates. Urban surface area unit in 1000 sq. miles. Robust standard errors presented in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. from i to j (column (2) of table 2.6) seem to be weak because the F-stat is less than 10. This indi- cates that the estimates presented in table 2.6 could be biased. This is not an unexpected finding, given that non-contiguous states’ shocks will produce lower volumes of migration compared to shocks that occur in all states. In other words, the strength of the instruments reduces by design. However, the F-stat in column (2) of table2.6 misses the strong-IV criterion marginally. Therefore, a potential bias will not significantly alter the estimates in table 2.6. Moreover, our results from previous sections and this one suggest that the local housing 86 demand shock from distant events (contiguous or not) is primarily through in-migration (i ← j) and not out-migration (i → j). This is because out-migration does not impact local urban population growth, whereas in-migration does. As long as the latter holds, we need not worry about the weak instrument problem in our supply estimation exercise. This is confirmed by the fact that both the negative rainfall shocks and the GQ upgrade program at distant non-contiguous states are strong instruments for the number of housing units in the local region (see table A.9). 2.6.2.2 Long-term migration We redefine the migration variables Migration ji and Migration ij in equations (2.4a) and (2.4b) to only include individuals who moved between 1-9 years before Census enumeration. In other words, we exclude migrants who moved less than a year before the Census enumerations or short-term and seasonal migrants. We rerun the regressions using equations (2.4a) and (2.4b). The results from these regressions are given in table 2.7. As in the case of overall migration (table 2.3), the distant state’s inclusion in the GQ highway upgrade program had a positive significant effect on long-term migration from j to i and from i to j, albeit with smaller magnitudes of effects. A distant state’s inclusion in the GQ program increased long-term migration from j to i by 22% and from i to j by 19%. However, contrary to table2.3, here we see that negative rainfall shocks at j do not affect long- term outward mobility. Instead, negative rainfall shocks in the distant state reduce long-term movement toward such regions. Even though the coefficient p-values are different in tables 2.3 and 2.7, the overall results still indicate that the distant state shock-induced long-term migration led to urban population growth in the local state. Similar to table2.3, here we see that a1% increase in long-term migration from j to i increased urban population growth at i by 1.44%, and out-migration from i had no impact on the local urban population. The higher impact of long-term (compared to short-term) migration on urban population growth is consistent with long-term movers settling down and starting families of their own at their destinations. 87 Table 2.7: Distant shock-induced long-term migration and local urbanization First-stage Second-stage Dependent variable =∆ Log indicators Migration 1-9 yrs. Urban population i← j i→ j at i ∆= 2011− 2001 (1) (2) (3) ∆ Log migration 1-9 yrs. i← j 1.44*** (0.390) ∆ Log migration 1-9 yrs. i→ j -0.630 (0.472) ∆ Rainfall deficit months at j 0.005 -0.007** (0.003) (0.003) GQ highway at j dummy 0.217*** 0.189*** (0.041) (0.045) ∆ Log consumption at i -0.198 -0.297 0.265 (0.148) (0.220) (0.272) ∆ Urban surface area at i -0.002 0.206*** 0.235** (0.028) (0.035) (0.104) ∆ Urban surface area at i squared 0.009*** -0.014*** -0.028*** (0.002) (0.003) (0.010) F-stat on excluded instruments 16.5*** 10.6*** Anderson–Rubin Wald χ 2 (2) 251*** N 1,013 1,013 1,013 R 2 0.126 0.083 Source: Authors’ calculations. Note: Table presents two-stage least squares first-difference regression estimates using a sample of state-state pairs in India between 2001 and 2011. State i is the local region, and state j the distant. Excluding missing pairs, there are 1,028 i− j state pairs consisting of the 35 states and union territories in India. Migration i← j is in-migration to i and i→ j is out-migration from i. Migrants are those who moved in the previous decade and were at their destinations for at least a year. Differences are calculated as first differences of values between the years 2001 and 2011. Log in- and out-migration are instrumented by the decadal change in the number of rainfall deficit months in state j and the GQ dummy, which equals one if state j was recipient of the Golden Quadrilateral highway upgrade program, and zero otherwise. Consumption, urban surface area, and urban surface area squared at i are additional covariates. Urban surface area unit in 1000 sq. miles. Robust standard errors presented in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. 2.7 Policy implications In this section, we discuss what the housing supply elasticity estimates from section 2.5.4 mean for India and the broader policy implications of the findings. We begin by presenting state-level supply elasticity figures. Since land administration is a state subject in India, state-level supply elasticity estimates would capture variation in land use regulations across India very well. We end this section with a discussion on the land use regulatory environment in India. 88 2.7.1 State-level Elasticities To get a sense of spatial heterogeneity in the housing supply elasticities across different parts of India, we provide some state-level elasticity figures for formal housing units. We do this by exploiting district-level variation within states for 12 of the largest states in India. 25 We estimate only the formal supply elasticities because not enough districts reported informal rents during the study period, making it impossible to estimate the informal elasticity even for larger states. 26 The formal housing supply elasticities are given in table 2.8. For comparison, column (3) of table 2.8 presents Metropolitan Statistical Areas (MSAs) in the US that are closest to these Indian states in terms of their supply elasticity estimates based on Saiz (2010). Table 2.8 shows that there is considerable variation in formal supply elasticities across states of India. On the higher end, Maharashtra has a supply elasticity of 3.06, comparable to Austin, Texas, in the US. Considering that cities in Texas are some of the most supply-elastic in the US, Maharashtra’s elasticity reflects a very responsive market. It is also not surprising that states like Orissa, Andhra Pradesh, and Gujarat have relatively higher supply elasticities. A 2009 World Bank report ranked cities, such as Bhubaneshwar (capital of Orissa), Hyderabad (capital of Andhra Pradesh), and Ahmedabad (capital of Gujarat), among the best for doing business in India. 27 It is also not surprising that Bihar and West Bengal have the lowest formal housing supply elasticities in India, with values of 0.49 and 0.38, respectively. These elasticity values are lower than the least supply elastic MSAs in the US: Miami and Los Angeles-Long Beach. The same World Bank report on ease of doing business in India estimated that the average time required to apply for a construction permit in Kolkata (capital of West Bengal) was 258 days, higher than any other major Indian city. Overall, these elasticity estimates indicate that while there are relatively responsive markets in India, such as Maharashtra and Orissa, the average urban area in India is perhaps as supply inelastic as the US. 25 We report the 12 states where at least 14 districts reported a formal rent figure for both 2001 and 2011. 26 For instance, Uttar Pradesh, the largest state in India, reported informal rents for only five districts in both 2001 and 2011. 27 Visit the link for the report: https://www.doingbusiness.org/content/dam/doingBusiness/media/Subnational- Reports/DB09-Sub-India.pdf. 89 Table 2.8: State-level formal housing supply elasticities Urban population Formal Comparable US MSAs (millions) elasticity (Saiz, 2010) State (1) (2) (3) Maharashtra 51 3.06 Austin, TX Odisha 7 2.05 Mobile, AL Tamil Nadu 35 1.92 Fresno, CA Andhra Pradesh 28 1.63 Phoenix, AZ Gujarat 26 1.31 Las Vegas, NV Madhya Pradesh 20 1.25 Detroit, MI Uttar Pradesh 44 1.17 Newark, NJ Rajasthan 17 1.06 Jacksonville, FL Karnataka 24 0.75 New York, NY Haryana 9 0.54 Miami, FL Bihar 12 0.49 West Bengal 29 0.38 Source: Authors’ calculations and Saiz (2010). Note: All reported states have observations on formal housing rents and quantities for at least 14 districts. States arranged in decreasing order of elasticity values. All elasticity values rounded off to two decimal places. Last column shows MSAs in the United States (US) that have comparable housing supply elasticities. No metropolitan areas in the US have elasticity figures comparable to Bihar and West Bengal. Miami is the least supply elastic MSA in the US, with an elasticity value of 0.6. 2.7.2 Land use regulations A discourse on housing supply is incomplete without a discussion of prevailing land use reg- ulations. A large body of literature has documented the supply restrictions imposed by land use regulations in the US (Diamond, 2017; Glaeser et al., 2005b; Quigley and Raphael, 2005). Similar regulations imposing supply-side constraints exist in India. These regulatory constraints increase the cost of construction (Brueckner and Sridhar, 2012; Gandhi et al., 2021). Three regu- lations merit some discussion in the context of housing supply in India. The first set of regulations comprises the various rent control laws in India. These laws restrict or prohibit landlords from increasing rents and evicting tenants. The resulting increase in the cost of renting reduces rural-urban migration and increases the rural-urban wage gap in India (Dutta et al., 2022). A recent study has shown that the rent control laws lead to more formal houses remaining vacant in India because of the higher costs borne by landlords in renting out their units to tenants (Gandhi et al., 2022). This may partly explain why the elasticity of vacant formal units is higher than occupied formal units (see section 2.5.4). Therefore, rent control laws are 90 clearly sources of frictions in housing markets of urban India and deserve further investigation in future research. The second set of regulations consists of the stringent building codes adopted by most urban areas of India. The most restrictive and widely studied of these building regulations is the floor area ratio (FAR) limit. All major cities in India, either through state- or municipal-level regulations, impose an FAR limit, usually around 1.5-2 (Sridhar, 2010). 28 The overly restrictive limits on FAR essentially reduce the floor space consumer per capita. Cities across the world that are similarly dense as Indian cities, such as Hong Kong and New York, have much higher FARs of around 10-15 (Sridhar, 2010). Needless to say, FAR restrictions may be severely limiting the housing supply in India. It is, however, difficult to tally the supply elasticity estimates with the existing FAR regula- tions. For instance, cities in Maharashtra impose some of India’s most restrictive FAR limits. Yet, we find that Maharashtra has the highest supply elasticity among major states. This is partly because of other factors that mitigate FAR restrictions, such as transferable development rights (TDRs). TDRs confer FAR credits, which can be resold in a market for FAR, to those surrendering parcels of land to the government for public purposes (Gupta, 2020). As a result, even though cities like Mumbai have very stringent freely available FAR limits of 1-1.3, the actual FAR may be much higher. 29 A thorough quantitative investigation of FAR limits and their effects on housing supply in Indian cities will be a policy and academically relevant question for future research to pursue. The third regulation concerns India’s urban land ceiling laws, which were in force across major urban areas between the 1970s and the 2000s. These regulations were part of the eminent domain restrictions imposed by the Indian government. The laws imposed ceiling limits on pri- vately owned vacant land. Land parcels above the limit were to be acquired and redistributed by governments to the urban poor. The urban land ceiling laws affected the way Indian cities de- veloped and expanded until the 1990s (Sridhar, 2010). Furthermore, the laws were implemented poorly, and most land parcels were neither acquired by the government nor used for any devel- 28 The FAR of a building is equal to its total floor area divided by the area of the land parcel on which it is built. 29 The housing conditions survey round from 2012, conducted by the National Sample Survey Organization, indi- cates that the average number of floors in a building in Mumbai is 4.23, much higher than the average of 1.23 in other parts of India. At the same time, the average lot size in Mumbai is not that different from other regions. 91 opment (National Institute of Urban Affairs,2002; Pethe and Nallathiga,2017). It is plausible that these regulations severely restricted the housing supply by freezing the development process on vacant land. The urban land ceiling laws were repealed during the2000s. Some studies have found positive impacts of this deregulatory reform on manufacturing growth (Duranton et al.,2015; Sood,2019). Therefore, an investigation into the effects of the reform on housing supply will provide critical insight into how housing supply can be improved in India. The third chapter studies how the repeal of the urban land ceiling laws in India affected housing supply between 2001 and 2011. 2.8 Conclusion Even though India is among the largest contributors to urban growth worldwide, the academic literature on housing in urban India is sparse. We fill this gap by estimating the supply elasticity of housing in urban India. We apply the Rosen-Roback spatial equilibrium framework to con- struct novel housing demand shifters. We show that shocks that induce inter-regional migration affect housing demand through population growth in regions that did not receive such shocks. We use negative rainfall shocks and a highway upgrade program implementation in one state as housing demand shifters for another. We begin by presenting some stylized facts on housing and migration in India. Next, we discuss the underlying theoretical framework, followed by a discussion of the empirical imple- mentation of the model mechanisms. We find that both negative rainfall shocks and the highway upgrade implementation induced inter-state migration in India during the 2000s. This increased migration led to urban population growth, increasing demand for housing in states that did not receive the shocks. We estimate national-level housing supply elasticity figures for urban India’s formal occupied, informal, and formal vacant residential housing units. While national and state-level housing supply elasticity estimates do not accurately depict metropolitan-level elasticities and the underlying heterogeneity across metropolitan areas of dif- ferent sizes and regulations, it is a relevant parametric estimate in the context of a large and urbanizing country. Further research with metropolitan-level price and new construction data would be required to provide granular elasticity estimates. 92 Chapter 3 Land Use Deregulation and Housing Supply Arnab Dutta 3.1 Introduction Regulations often have the equivalent effect of taxes in an economy. Within the context of an otherwise efficient market, regulations create the same kind of inefficiency wedges as taxes. Can deregulation reduce barriers to market efficiency? Institutional frictions, such as inadequate property rights, particularly in the context of developing countries, may inhibit or delay the benefits of deregulation. This chapter explores whether land use deregulation leads to housing supply growth in the presence of poorly maintained and updated land ownership and transaction records. Specifically, I conduct an ex post analysis of the repeal of urban land ceiling laws in India during the late 1990s and the 2000s. The urban land ceiling (ULC) laws, which came into force during the 1970s, put ceiling limits of 500-2000 square meters on privately owned vacant land in 70 of the largest cities in India. 1 Any excess vacant land identified under the ULC laws had to be surrendered to the government at fixed prices, which were often below market rates. The government’s role was to acquire and redistribute vacant land among the urban poor. The land acquisition process was, however, slow. By the 1990s, a consensus emerged that the laws had failed to achieve their objectives and were restricting housing supply in the affected cities. The ULC repeal act was promulgated by the central government of India in 1999 and was subsequently adopted by the different states between 1999 and 2008. I empirically estimate the impact of the ULC repeal act on formal and informal residential housing supply using a panel of202 cities in India between2001 and2011. 2 I employ a difference- in-difference estimation strategy by defining treatments based on the repeal of ULC. I classify cities in states that repealed the laws between 1999 and 2000 as early repeal and those which 1 The ceiling limits were inversely related to city population sizes. For more details, see section 3.2.1. 2 The definitions of formal and informal houses are given in Chapter 2. Formal houses are constructed using durable substances like concrete and metal. Informal houses are made of non-durable substances like grass, thatch, bamboo, etc. 93 repealed the laws between 2002 and 2008 as late repeal. I find that the formal residential housing supply was practically unaffected by the ULC’s repeal (early or late). Neither new formal housing construction nor the redevelopment of informal houses to formal buildings increased in ULC repeal cities relative to other cities. The ULC repeal act did not affect the supply of informal housing or the non-residential stock either. However, there was a decline in the new construction of informal houses. To investigate why the formal housing supply was unaffected by the ULC repeal, I dig into its institutional aspects. Under the ULC act, land acquisition by the government was slow, some- times spanning decades, primarily due to a slow process of ownership transfer and record upda- tion. State governments were in the process of acquiring a large number of land parcels when the ULC repeal act was passed in 1999. The repeal act failed to recognize this slow land acquisition process. It had no provision to decide the fate of land parcels in the process of acquisition. In other words, the repeal act did not specify whether land parcels in the process of acquisition were owned by governments or landowners. Hence, there was ambiguity about the ownership of such parcels after the repeal act was passed. Landowners and governments contested owner- ship rights over these parcels, which led to litigation. I provide anecdotal evidence of such legal proceedings. In the same sample of 202 cities, I examine whether land-related legal proceedings were higher in cities where ULC was enacted, after the repeal act was passed. I find that the number of land acquisition and land grabbing court cases per 100,000 people initiated between 2010 and 2016 was six times as high in cities that had enacted ULC compared to cities that never enacted ULC. These legal proceedings imposed costs on landowners. I then build a monocentric city model, following Henderson et al. (2021), to examine the mech- anisms through which the legal issues might explain how ULC repeal affected India’s housing markets. In this model, as a city radially expands, agricultural land on the city’s periphery is used for informal housing development. As rents are driven up over time through productivity shocks, informal houses are redeveloped into formal houses. I expand on Henderson et al.’s model to incorporate a heterogeneous cost of land development which must be incurred before housing construction. The development cost reflects both natural and institutional barriers to construction. A sufficiently high cost of development will imply that vacant land exists in the 94 city and vacant land is used only for formal housing construction. The ULC act upends the vacant land development process by imposing a regulatory tax on formal housing construction on vacant land. Because of its lax enforcement, the ULC-enforced tax did not apply to informal housing development. Therefore, under the ULC act, vacant land is used only for informal housing construction. When the ULC act is repealed, the regulatory tax disappears. But, shortly after the repeal, governments and landowners initiate legal proceedings to contest ownership rights over vacant parcels. Landowners then must incur a legal cost to settle disputes before using vacant parcels for construction. The legal cost applies to both formal and informal housing development because courts are more effective in enforcing rulings in India (Bhan, 2009; Mehta, 2007; Varshney, 1998). In this scenario, vacant land is used only for formal housing construction. However, such construction is pushed into the future until rents increase sufficiently to offset the development and legal costs. The model predictions are, for the most part, consistent with the empirical findings. More broadly, these findings indicate that the process of deregulatory reform matters, espe- cially in the context of developing countries where institutional frictions persist. It is important to note here that the research in this chapter does not suggest that deregulatory reform is bad. Neither does it suggest that deregulation is necessarily unsuccessful when institutional frictions exist. Deregulation is a slow process, and its benefits emerge in the long-run (Winston, 1998; Winston and Karpilow, 2016). Furthermore, the construction of houses, including the processes of getting the necessary permits, takes a long time, particularly in India (Gandhi et al., 2021). Therefore, it is likely that in the long run, the complete benefits of deregulation will be realized. In that case, institutional frictions will delay, rather than inhibit, the benefits of deregulation. In light of these arguments, it is best to think of the key empirical results presented in this chapter as the medium-run effects of the ULC repeal act rather than its long-run effects. This chapter connects and contributes to several bodies of academic literature. First, this chap- ter contributes to the voluminous literature on deregulation. Papers have shown that financial market reforms encourage innovation among private firms (Chava et al., 2013), enhance produc- tivity growth (Buera and Shin, 2013), and increase financial inclusivity by helping low-income households accumulate wealth (Célerier and Matray, 2019). But, there are general equilibrium ef- fects of deregulation which may undermine its benefits (Caselli and Gennaioli, 2008), and in turn, 95 have distributional consequences (Blanchard and Giavazzi, 2003; Marjit, 2003). This chapter con- tributes to this body of literature by showing that the institutional background and the process of deregulation determine, to some extent, the benefits that can be derived from deregulation. Second, this chapter connects with the empirical literature on the effects of property rights provision on economic outcomes in developing countries. 3 On the one hand, papers have shown that a poor provision of property rights can lead to low investment in public goods (Banerjee and Iyer, 2005; Banerjee et al., 2005) and slow economic growth (Collier and Gunning, 1999; Levine, 2005). On the other hand, many studies have shown the economic benefits of better property rights provision, such as higher incomes (Acemoglu et al., 2001; Besley and Burgess, 2000) and more efficient decision-making by households (Besley, 1995; Field, 2005, 2007; Galiani and Schar- grodsky, 2010). Better access to information on property ownership, through computerization, can also enhance credit supply (Deininger and Goyal, 2012). This chapter contributes to this literature by demonstrating how poor property rights provision can act as an institutional barrier to deregulatory reform. Finally, this chapter is directly related to the literature on land use regulation. Papers have shown that land use regulations have negative consequences for housing supply (Glaeser et al., 2005b; Glaeser and Ward,2009; Mayer and Somerville,2000; Quigley and Raphael,2005), through a reduction in the housing supply elasticity (Green et al., 2005; Saiz, 2010) and the imposition of a regulatory tax on builders (Glaeser et al., 2005a). Land use regulations also have negative macroeconomic effects on GDP (Herkenhoff et al., 2018; Hsieh and Moretti, 2019). In developing countries, land use regulations may enable the proliferation of slums by constraining formal housing supply (Cavalcanti et al., 2019; Henderson et al., 2021). However, the evidence on the effects of land use deregulation on housing markets is ambigu- ous. Some studies have found that land use deregulation can increase welfare gains (Herkenhoff et al., 2018), boost productivity growth (Parkhomenko, 2020), and enhance the affordability of housing (Kulka et al., 2022). On the other hand, housing market deregulation can lead to price appreciation (Autor et al., 2014) and gentrification (Davis, 2021; Rodríguez-Pose and Storper, 2020). Deregulation of financial markets in the US during the 1980s led to higher house prices because of more credit access combined with low housing supply elasticities (Favara and Imbs, 3 See Besley and Ghatak (2010), for a summary of the literature. 96 2015; Hoffmann and Stewen, 2020). This chapter shows that institutional frictions influence the way land use deregulation might affect housing supply, especially in the context of developing countries. The rest of this chapter is organized as follows. In section 3.2, I provide an overview of the legal and institutional aspects of the ULC act and its repeal. Section 3.3 discusses the data. In section 3.4, I present the empirical results from estimating the impact of ULC repeal on housing supply. In section 3.5, I empirically explore the relationship between land-related litigation and ULC, followed by a discussion of the mechanisms and policy implications in section3.6. I provide concluding remarks in section 3.7. 3.2 Background In this section, I provide information on India’s urban land ceiling (ULC) laws. I cover details on the enactment of the ULC laws, the land acquisition process under the ULC act, the repeal of the ULC act, and the legal issues that ensued after the ULC act was repealed. 3.2.1 The ULC act The Urban Land (Ceiling and Regulation) Act was enacted by the Parliament of India in 1976. This federal law applied to the64 largest urban agglomerations (UAs) or cities in India which had populations of 200,000 or more. Among the states which had cities with populations exceeding 200,000, Kerala and Jammu and Kashmir did not adopt the ULC act. 4 On the other hand, the state of Tamil Nadu enacted its own ULC law in 1978, which applied to its six largest cities with a population above 200,000. 5 The ULC laws had two major provisions. First, they put ceiling limits of 500-2000 square meters on privately owned vacant land. 6 Any excess vacant land beyond the ceiling limits had to 4 Article 252(1) of the constitution of India allowed the central government of India to create laws on state subjects such as land. The ULC act came into force in the 12 states of India, where legislatures had adopted article 252(1). The union territories, governed by the central government of India, also came under the purview of the ULC act. 5 The ULC act of Tamil Nadu had different ceiling limits, with a separate set of limits for individuals, family-owned properties, and industries. See Kimura (2011) for more details on how the Tamil Nadu act differed from the central government’s ULC act. I address the identification issues arising from the differences in the federal and Tamil Nadu laws in section 3.4.4. 6 The ceiling limit in the three largest cities of Delhi, Kolkata, and Mumbai was 500 square meters. In other cities with more than a million people, the limit was 1,000 square meters. In cities with populations between 300,000 and 97 be surrendered to the government at below-market rates (Sridhar, 2010). And second, the ULC act put restrictions on the transfer of property. Permissions were required from the government for land transactions. Competent Authorities (CAs), typically revenue collectors or other state government officials, were empowered by the government to oversee the ULC act’s enforcement. The government’s intention with the ULC laws was progressive. The idea was for the gov- ernment to acquire and redistribute excess vacant land among the urban poor and to prevent speculation in urban land markets. However, according to the government’s own assessment, the laws had failed to achieve their objectives (Parliamentary Standing Committee, 1999). By 1998, out of the roughly 850 square miles of vacant land identified by the central government’s ULC act in the 64 cities that it covered, only 74 square miles had actually been acquired. 7 For the most part, the remaining land parcels were either under litigations (Srinivas, 1991) or in the process of acquisition. In both cases, these land parcels remained vacant, and they were not used for any construction activities (National Institute of Urban Affairs, 2002; Pethe and Nallathiga, 2017). 8 3.2.2 Land acquisition under the ULC act The land acquisition process under the ULC act was slow, sometimes spanning over 20 years. There are at least two reasons for this. First, in a large number of acquisition cases, the CAs were not able to complete the various steps in the acquisition process in a timely manner. In a qualitative investigation, with detailed interviews of CAs from Mumbai, Siddiqi (2013) found that the power vested in CAs was dispro- portionately less than the degree of responsibilities they had to undertake. CAs had to coordinate the process of land acquisition between multiple government departments, which was, at times, extremely cumbersome and time-consuming. And second, institutional aspects played a role too. Land ownership in India is usually one million, the limit was 1,500 square meters. And finally, in cities with populations between 200,000 and 300,000, the limit was 2,000 square meters. 7 A quick back-of-the-envelope calculation shows that the ULC act-identified excess vacant land constituted 13% of land area in the covered cities. The average treated city had a total surface area of 99 square miles in 2001. Next, dividing the 850 square miles of total identified excess vacant land among the 64 cities covered by the federal law, we calculate that the average treated city had 13 square miles of excess vacant land. Dividing 13 by 99, we get the figure of 13%. 8 In some cases, state government officials engaged in corruption (Joshi and Little, 1996) and granted arbitrary exemptions using some provisions in the ULC act (Parliamentary Standing Committee, 1999; Siddiqi, 2013). 98 presumptive, and not even a complete set of transaction records, property tax documents, etc., are enough to ensure absolute ownership (Wadhwa, 2002). 9 Therefore, proving land ownership can be difficult. Furthermore, updating land mutation records can take several years, especially in places where records are not digitized, which was the case in most parts of India during the 2000s. 10 A complete land acquisition process under the ULC act by the government required that the government’s name be mentioned on mutation records. This could be done after concluding a prolonged process of back-and-forth notifications, approvals, and petitioning between landown- ers and the CAs, which could continue for years. Only after the completion of these processes could the CAs take physical possession of excess vacant land. At no step of the process could CAs delegate their responsibilities to subordinates. 3.2.3 Legal issues in the ULC repeal act With a consensus that the ULC act had failed to achieve its objectives, the Urban Land (Ceiling and Regulation) Repeal Act was promulgated in 1999. Several states adopted the repeal act between 1999 and 2000. The state of Tamil Nadu also repealed its ULC act in 1999. However, the states of Andhra Pradesh, Assam, Bihar, Maharashtra, and Orissa adopted the ULC repeal act between 2002 and 2008. The year of repeal for every state in the sample of analysis is given in table B.3. The ULC repeal act had two loopholes. First, it failed to recognize that the process of land acquisition under the ULC act was slow; the act was unclear about the fate of land parcels still in the process of acquisition. The repeal act mentioned that land acquired by the government would not be returned to landowners and that land not acquired would remain with landowners. This led to confusion about ownership rights of ULC act-identified excess vacant land parcels still in the process of acquisition. And second, the repeal act also allowed legal proceedings related to compensation received by landowners, for surrendering their parcels to the government, to 9 See this Ideas for India article on how several documents are required even to prove presumptive ownership: https://www.ideasforindia.in/topics/human-development/land-records-and-titles-in-india.html. 10 Digitization of land records in India began with the Computerization of Land Records scheme initiated by the central government of India in 1988-1989 (Habibullah and Ahuja, 2005). However, only a few states, like Gujarat and Karnataka, were able to digitize land records before 2011 (Behera, 2009). On a larger scale, digitization did not take off until the late 2000s, when the central government launched the National Land Record Modernization Programme. 99 continue. 11 These legal loopholes of the repeal law led to new legal proceedings initiated by various state governments and landowners against each other. Upon conducting a thorough investigation of numerous court cases, I did not find any compensation-related legal proceedings which continued after the repeal act was enacted in 1999. Plenty of such cases were resolved during the 1980s and the 1990s. However, there are sev- eral examples of legal proceedings initiated after 1999, where state governments and landowners disputed the ownership and possession of excess vacant parcels. The common theme in these cases is as follows. After the ULC repeal act was promulgated, governments assumed that they had de facto ownership over land parcels whose individual own- ers were already served notices by the CAs, that their excess vacant land would be acquired, a necessary step at the beginning of the process of land acquisition. On the other hand, individuals assailed this argument, saying that ownership had to be de jure, that is, governments could claim ownership only after the completion of all formalities required for the transfer of property. 3.2.4 Anecdotal evidence The most salient example of the legal proceedings which resulted from the disputed possession of ULC act-identified excess vacant land is that of State of Uttar Pradesh v. Hari Ram. 12 Briefly, the details of State of Uttar Pradesh v. Hari Ram are as follows. Hari Ram, a landowner from the state of Uttar Pradesh, initiated the process of handing over his excess vacant land to the government by providing the local CA with information about his landholdings in 1976. The office of the CA took 23 years to notify the regional land authorities that the excess vacant land owned by Hari Ram was to be acquired by the government. By then, the state of Uttar Pradesh had repealed the ULC act. Hari Ram filed a lawsuit, which ultimately went to the Supreme Court of India. In 2013, the Supreme Court judge noted that the CA had violated land acquisition rules, and mutation records were not updated with the government’s name. The court ruled that the acquisition order was invalid and Hari Ram was the legal owner. Many future cases cited the ruling of the Supreme Court in State of Uttar Pradesh v. Hari Ram and were resolved similarly. Examples of such cases include Raju v. State of Uttar Pradesh 11 These were legal proceedings under sections 11, 12, 13, and 14 of the ULC act. 12 Civil appeal no. 2326 of 2013. 100 and Mohammad Suaif and Another v. State of Uttar Pradesh at the high court of Uttar Pradesh (Allahabad) and Maan Singh v. State of Madhya Pradesh at the high court of Madhya Pradesh (Indore). 13 In a smaller number of cases, the Supreme Court ruled against landowners wherever owners violated some aspect of the ULC act. 14 3.3 Data and descriptives In this section, I briefly describe the data used for the analysis. First, I provide some key def- initions that are used throughout the chapter. Then, I discuss the data sources and descriptive statistics. More detailed information about data collection is provided in appendix B. 3.3.1 Definitions There are two key concepts used throughout the chapter that require some explanation. The first concept is that of the distinction between formal and informal housing. The second has to do with the definition of a city, its constituents, and its administration in India. The definitions of formal and informal housing used in this chapter are identical to the ones presented in section 2.2.1 of chapter 2. To reiterate, I classify houses as informal if their roofs or walls are made of non-durable substances such as grass, thatch, bamboo, plastic, polythene, mud, unburnt brick, and wood. On the other hand, formal houses’ roofs and walls are made of durable substances like galvanized iron, metal, asbestos sheets, burnt bricks, stone, and concrete. The definition of cities that I use is based on the Census of India’s definition of urban agglom- erations (UAs). UAs are conurbations of multiple municipalities and non-municipal areas. 15 In my analysis, I use data on 202 UAs consisting of about a thousand municipalities and non- municipal areas. 16 These cities are spread across 14 states of India. 17 I provide a detailed account of the urban administrative entities in UAs and the composition of UAs in my sample in ap- 13 Writ. petition no. 26861 of 2017, Writ. petition no. 12696 of 2009, and Writ. petition no. 4614 of 2022, respectively. 14 For example, State of Assam and Others v. Bhaskar Jyoti Sarma and Others, civil appeal no. 10565 of 2014. 15 Non-municipal areas within UAs have urban-like features, such as high population density and a large non- agricultural sector, but are governed by rural administrative bodies. 16 Figure A.3 presents a map of India with the 202 cities geocoded using Google Maps API. 17 Most of the excluded states did not have cities with 200,000 or more people in 1971. Among those that did have large cities, Jammu and Kashmir is dropped because of missing data in 1991. Jharkhand and West Bengal are dropped because these states still had not adopted the ULC repeal act during the study period of 2001-2011. The state of Orissa is also dropped from the sample because of several outliers in the data on legal proceedings. See appendix B.2 for more details. 101 pendix B.1. To avoid confusion in terminology, I consistently use the term cities instead of urban agglomerations throughout the chapter. 3.3.2 Data sources I use data primarily from the Census of India and the National Sample Survey Organization (NSS) to construct a panel of 202 cities for the two Census years—2001 and 2011. For some variables, I have city-level information, whereas, for others, I use either district-level data or repeated cross-section survey data to construct aggregated variables at the city level. I also use a household-level panel dataset from the India Human Development Survey (IHDS). I draw data at the district level from four sources. First, I use employment figures by different industrial sectors at the district level to construct an employment shift-share index. Second, I use land-related legal proceedings data from Ash et al. (2021). Third, I obtain data on daily rainfall and temperature at the 1°× 1° grid level from the India Meteorological Department. I assign temperature and rainfall values to districts based on the grid they belong to. And finally, I use household-level sample survey datasets from the NSS to construct aggregated variables at the district level, such as the share of households living in newly constructed units, using sampling weights provided by the NSS. These district-level variables are converted to the city-level with population weights. 18 I use data on a number of variables that are available at the city level from the Census of India. These variables include the total number of formal and informal residential houses, the total population, the total numbers of scheduled castes and tribes individuals, females, literate persons, and the total surface area. 19 I also use data on the number of non-residential buildings available only for municipalities within cities. Such non-residential buildings are used for non- residential purposes—schools, offices, shops, etc. And finally, I collect information about the ULC enactment and its repeal from various sources (legal proceedings, government documents, etc.). More details on data collection are given in appendix B.2, and the process of gathering ULC-related information is provided in appendix B.3. The summary statistics for the city-level 18 The weight for a given city-district pair equals the share of a city’s population living in the district in 2001. 19 The scheduled castes and tribes are among the most marginalized communities resulting from the hierarchical caste system in India. Individuals belonging to lower castes and indigenous groups are categorized as scheduled castes and scheduled tribes, respectively, by the Indian constitution. 102 Table 3.1: Summary statistics of city-level variables 2001 2011 Data Mean SD Mean SD source Variable (1) (2) (3) (4) (5) Population ('000) 624 1,443 804 1,759 Census Formal units ('000) 96 262 154 371 Census Informal units ('000) 19 29 18 25 Census Non-residential stock ('000) 38 113 57 176 Census Surface area (sq. miles) 37 57 50 81 Census Mean temperature (°F) 79 2 79 2 IMD Annual aggregate rainfall (inches) 43 25 41 23 IMD Share formal HH in new units 0.24 0.14 0.19 0.12 NSS Share informal HH in new units 0.19 0.22 0.13 0.24 NSS Share SC/ST 0.13 0.06 0.15 0.06 Census Share females 0.48 0.02 0.49 0.02 Census Share literate 0.70 0.08 0.74 0.07 Census Employment shift-share index 0.29 0.06 0.80 0.08 Census N 202 202 202 202 Source: Author’s calculations based on Census of India, India Meteorological Department, and National Sample Survey Organization. Note: Table presents summary statistics of city-level variables for 202 cities observed in 2001 and 2011. HH stands for household. Shares of formal and informal housing residents living in new units calculated from NSS survey data. New units are defined as those constructed in the previous decade. Shares of SC/ST, females, and literate calculated as shares of population. SC/ST stands for scheduled castes and tribes individuals. The employment shift-share index is a standard Bartik shifter calculated based on employment shares of 10 sectors in 1991. Figures for population and formal, informal, and non-residential stock are rounded off to thousands. Surface area, temperature, and rainfall are rounded off to integer values in square miles, Fahrenheit degrees, and inches, respectively. All share variables, including the shift-share index, are rounded off to two decimal points. variables are presented in table 3.1. Detailed summary statistics of the city-level variables for the treatment and control groups of cities are provided in tables A.10 and A.12. To identify in situ redevelopments of informal houses to formal houses, I turn to the India Human Development Survey (IHDS), conducted jointly by the National Council of Applied Eco- nomic Research and the University of Maryland (Desai et al., 2015). The project surveyed over 40,000 rural and urban households living in more than 350 districts of India. The first round of the survey was conducted during 2004-2005, and the second round was conducted during 2011-2012. 20 After accounting for missing data, I identify a subsample of 4,832 non-migrant households that lived in urban areas of districts where the 202 cities in my sample are located. In order to assign the ULC repeal treatments to the households, I conduct a matching of the districts 20 Data from the first round of survey has been used in chapter 1. 103 Table 3.2: Characteristics of households surveyed by IHDS 2004-2005 2011-2012 Mean SD Mean SD Variable (1) (2) (3) (4) HH lives in informal unit dummy 0.21 0.41 0.15 0.36 HH’s real annual income ('000 INR) 113 196 82 125 HH size 5.06 2.24 4.78 2.18 HH owns land dummy 0.08 0.26 0.08 0.27 N 4,832 4,832 4,832 4,832 Source: Author’s calculations based on Desai et al. and Labor Bureau of India. Note: Table presents summary statistics of a balanced panel of 4,832 non-migrant households surveyed once dur- ing 2004-2005 and a second time during 2011-2012. HH stands for household. Real income values rounded off to thousands integer INR values. Income is inflation-adjusted based on Consumer Price Index (CPI) values (base year = 2001) from the Labor Bureau of India. In PPP terms, $1 = 10 INR in 2001. For exchange rates, visit: https://data.oecd.org/conversion/purchasing-power-parities-ppp.htm. of residence of households to the cities in my sample of analysis. The data provides information on the type of households’ dwelling (formal or informal), annual household income, household size, and whether households owned land. Summary statistics for these variables are given in table 3.2. 3.3.3 Descriptive statistics The summary statistics of city-level variables presented in table 3.1 provide some insight into the characteristics of cities in the sample. The average city had 172,000 houses (informal and formal combined) for a population of about 800,000 in 2011, or a ratio of about 1 to 5, which is slightly higher than the ratio of 1 to 5.5 in 2001. The major portion of the housing stock growth (from 115,000 to 172,000 units) during the 2000s can be explained by the increase in the number of formal housing units since there was a decline in the informal housing stock. At the same time, there was a surge in the number of non-residential buildings by about 50%. 21 The increase in the non-residential stock is consistent with the large growth in the employment shift-share index during the 2000s. A breakdown of the growth rates of housing units and non-residential buildings by the treat- ment and control groups reveals a more nuanced story. Figure 3.1 plots the growth rates of informal and formal residential housing units and non-residential buildings, between 2001 and 21 Divide the growth of 19 million by the 2001 stock of 38 million. 104 Figure 3.1: City-level growth rates of houses and non-residential buildings (2001-2011) 58 57 63 4 2 −9 45 46 66 −30 0 30 60 90 Percentage growth rate in stock Formal residential Informal residential Non−residential No ULC Early repeal Late repeal 95% C.I. Source: Authors’ calculations based on Census of India. Note: Figure plots mean growth rates of residential formal and informal housing units and non-residential buildings between 2001 and 2011 by treatment and control groups of cities. Circles represent cities that never enacted ULC. Triangles represent early repeal cities that repealed ULC during 1999-2000. Squares represent late repeal cities that repealed ULC during 2002-2008. Vertical lines represent 95% confidence intervals. All points labeled with the corre- sponding mean percentage values. 2011, by the groups of no-ULC, early repeal, and late repeal cities. The formal housing stock grew almost at the same rate of roughly 60% in all three categories. But, while the informal housing unit growth was small and positive in no-ULC and early repeal cities, there was a de- cline by 9% in late repeal cities. On the other hand, the growth in the number of non-residential buildings was about 40% higher in the late repeal cities. The decline in the informal stock, along with an increase in the non-residential stock in late repeal cities, is consistent with the evidence of gentrification in urban India (Bhan, 2009; Dutta et al., 2021a), where slums are demolished to make space for commercial real estate. 22 To further understand how different the treatment and control cities were in the pre-treatment period, I present a balance test of all city-level variables with two-sample t-tests using data from 2001. The test statistics are given in table 3.3. Note that only the largest cities in India en- acted ULC. Therefore, we should expect the differences between treatment and control groups 22 See Chapter 2 for a discussion on gentrification in Indian cities. 105 Table 3.3: Balance test of city-level variables No Early Late Diff. Diff. ULC repeal repeal (2)-(1) (3)-(1) Variable (1) (2) (3) (4) (5) Log formal units 10.22 12.03 12.41 1.81 2.19 (0.05) (0.13) (0.32) [0.00] [0.00] Log informal units 8.90 10.27 10.93 1.37 2.03 (0.06) (0.14) (0.23) [0.00] [0.00] Log non-residential stock 9.34 11.02 11.36 1.68 2.02 (0.05) (0.14) (0.35) [0.00] [0.00] Share formal HH in new units 0.23 0.23 0.28 -0.00 0.05 (0.01) (0.02) (0.04) [0.83] [0.26] Share informal HH in new units 0.19 0.16 0.21 -0.03 0.02 (0.02) (0.04) (0.05) [0.48] [0.76] Log population 12.24 13.96 14.28 1.72 2.04 (0.04) (0.12) (0.31) [0.00] [0.00] Employment shift-share index 0.28 0.32 0.32 0.04 0.04 (0.00) (0.01) (0.02) [0.00] [0.03] Log surface area 2.56 4.26 4.42 1.70 1.86 (0.07) (0.12) (0.29) [0.00] [0.00] Share SC/ST 0.13 0.13 0.13 -0.00 -0.00 (0.00) (0.01) (0.02) [0.73] [0.64] Share females 0.48 0.47 0.48 -0.01 -0.00 (0.00) (0.00) (0.00) [0.03] [0.37] Share literate 0.70 0.70 0.73 0.00 0.03 (0.01) (0.01) (0.01) [0.51] [0.02] Mean temperature (°F) 78.73 78.26 79.68 -0.47 0.95 (0.18) (0.38) (0.77) [0.27] [0.26] Log annual aggregate rainfall 3.63 3.60 3.91 -0.03 0.28 (0.04) (0.07) (0.12) [0.70] [0.04] N 157 33 12 Source: Author’s calculations. Note: Table presents two-sample t-test statistics for 2001 values of city-level variables. No ULC represents cities that never enacted ULC. Early repeal represents cities that repealed ULC during 1999-2000. Late repeal represents cities that repealed ULC during 2002-2008. HH stands for household. Shares of formal and informal housing residents living in new units calculated from NSS survey data. New units are defined as those constructed in the previous decade. Variable means presented in columns (1)-(3), mean differences between early repeal and no-ULC cities in column (4), and mean differences between late repeal and no-ULC cities in column (5), all without parentheses or brackets. Standard errors are given in parentheses below variable means in columns (1)-(3). The p-values of two- sample t-tests are given in brackets below mean differences in columns (4) and (5). All values are rounded off to two decimal places. to be large in variables strongly correlated with population. This explains the significant differ- ences between the means of the number of housing units and surface area. On the other hand, with some exceptions, most variables that do not depend on city population, like the share of 106 households living in newly constructed units, climate, the share of SC/ST people, etc., are not significantly different between the treatment and control groups. The employment shift-share index is substantially higher in the treatment group, however, implying that the larger cities received higher labor demand shocks even before 2001. 3.4 Impact of ULC repeal on housing and non-residential supply In this section, I present empirical estimates for the impact of ULC repeal on residential housing and non-residential supply in 202 Indian cities between 2001 and 2011. First, I describe the em- pirical implementation and estimating equations. Next, I discuss the results from the estimation. Finally, I conduct robustness checks and falsification tests to validate the empirical results. 3.4.1 Empirical implementation I estimate the ULC repeal’s impact on the growth in formal and informal residential housing, non-residential stock, and the redevelopment of informal houses to formal houses. I exploit the repeal of ULC as a treatment variable in a difference-in-difference framework using a panel of 202 cities observed in the Census years 2001 and 2011. I construct two dummy variables—Early repeal and Late repeal—to capture refined treatment effects. Early repeal equals one if the state of administration of a given city repealed ULC between 1999 and 2000, and zero otherwise. Late repeal equals one if the state of administration of the city repealed ULC between 2002 and 2008, and zero otherwise. In the sample of analysis, there are 33 early repeal cities and 12 late repeal cities. All cities in the sample that had enacted ULC repealed it by 2008. Therefore, the control group consists of 157 cities that never enacted ULC. In order to write the estimating equation, I employ the conceptual framework from spatial equilibrium (Roback, 1982; Rosen, 1979). Within this framework, the equilibrium quantity of housing in a city will be affected by its population growth, land area, productivity, and city- specific amenities. 23 While population, productivity, and amenities control for demand-side fac- 23 A model can be used to derive the housing market equilibrium equation based on the exposition of spatial equilibrium in Glaeser and Gottlieb (2009). Such a model will consist of a tradable consumption good sector, a housing market, a labor market, and a market for non-traded capital (with price normalized to unity). Both the production of both the tradable good and housing require labor. Assuming that housing is built by non-resident landowners and the spatial utility equalizing criteria, the housing market equilibrium can be derived as a function of the city population, surface area, tradable good sector productivity, and city-specific amenities. A cost of regulation 107 tors, urban land area controls for natural supply constraints. Besides the log of population and the log of surface area, I include an employment shift-share index (Bartik, 1991) as an indicator for productivity growth and temperature and rainfall as indicators for city-based amenities in the estimating equation. 24 I estimate the following two-way fixed effects equation to identify the ULC repeal’s impact on housing and non-residential stock in city i: y it = β 1 Early repeal i × Post t +β 2 Late repeal i × Post t +β 3 X it +σ i + Post t +ϑ is × Post t +ϵ it (3.1) where, y is the outcome variable (housing and non-residential stock, etc.), Post is the time trend which is defined as a dummy variable equal to one for the year 2011, and zero otherwise, σ i represents city fixed effects, ϑ is represents state fixed effects, ϵ it is the error term, and X is a vector of controls that include the log of population, an employment shift-share index, the log of surface area, the population’s shares of scheduled castes and tribes individuals, females, and literate persons, mean temperature, and the log of annual aggregate rainfall. Since land administration is a state subject in India, the ULC repeal act was adopted by individual states at different points in time. Therefore, I cluster standard errors by the states of administration of cities. The administration of land by states is also the motivation for in- cluding the interaction of state fixed effects ϑ is and the Post dummy in equation (3.1). This interaction term captures, to a large extent, time-varying factors that may be affecting housing supply. The interaction term will also partly address identification issues resulting from selec- tion in treatment—the timing of the ULC repeal by different states may be affected by several simultaneously occurring political and institutional events specific to those states. The terms β 1 and β 2 are the two difference-in-difference coefficients of interest. They repre- on housing that is proportional to the quantity of housing produced will add a regulatory tax as an additional term in the equilibrium condition. 24 The shift-share index is calculated as a standard Bartik shifter with ten sectors and1991 employment shares of each sector in a city. The sectors are cultivation, non-cultivation agricultural activities, agriculture-allied industries, mining, household enterprises, manufacturing, construction, transport and communication, trade, and services. Suppose that employment in a sector j at time t in city i is E jit . The total employment in city i at time t is E it . Then, the shift-share index M it , for city i at time t, can be written as follows: M it =∑ 10 j=1 (E ji1991 /E i1991 )((E ji ′ t − E ji ′ 1991 )/E ji ′ 1991 ), where, i ′ denotes any urban area outside the city i. 108 sent the impacts of the ULC repeal on the outcome variables. The key assumption required to consistently estimate β 1 and β 2 is that the trajectory of housing and non-residential stock indi- cators would have remained the same in the treatment and control groups in a counterfactual world without the treatment. I test the validity of this parallel trends assumption in section 3.4.5. 3.4.2 Decomposing the housing stock Two of the outcome variables in equation (3.1) are the log of the number of formal and infor- mal residential housing units. Now, the growths in the number of formal and informal units are functions of the number of newly constructed units, the number of informal units that are redeveloped into formal units, and the number of demolished units. We can trace the changes in the housing stock over time with the following pair of equations: In f ormal stock t = In f ormal stock t− 1 + New in f ormal t − Redeveloped t − Demolished t (3.2a) Formal stock t = Formal stock t− 1 + New f ormal t + Redeveloped t − Demolished t (3.2b) To fully understand the impact of ULC repeal on residential housing supply, I run additional regressions with indicators for newly constructed units and redevelopment as outcome variables. This nuanced analysis allows me to trace the sources of the ULC repeal’s impact on housing stock growth. For newly constructed units, I create city-level variables that are sampling-weighted shares of households living in new units (defined as those constructed in the previous decade) based on information from the NSS data. The two city-level variables are the share of informal households living in new units and the share of formal households living in new units. 25 I use a panel of non-migrant households to identify redevelopments—changes in households’ dwelling status from informal to formal between survey rounds. A status change in the dwelling of non-migrant households indicates that there had been an in situ redevelopment between sur- 25 The share of households living in new units is a good proxy for the share of newly constructed units since the correlation between the number of households and the number of residential houses in the sample of analysis is 0.99. The impact of ULC repeal on the share of newly constructed units equals its impact on the log of newly constructed units minus its impact on the log of total number of units. Therefore, the impact of ULC repeal on the log of new units will have to be greater than its impact on the log of total number of units for the ULC repeal impact on the share of new units to be significant. In other words, the estimated impact of ULC repeal on the share of new units will be an underestimate of its impact on the number of new units. 109 vey rounds. In the absence of data, I cannot specifically test the impact of ULC repeal on demo- litions. 3.4.3 Results In this section, I present the difference-in-difference regression results from estimating equa- tion (3.1) with indicators for formal housing, informal housing, redevelopment, and non-residential stock as outcome variables. While the analysis of formal and informal housing and non-residential stock will be at the city level, I use household-level panel data to estimate the impact of ULC re- peal on redevelopments. 3.4.3.1 Formal housing The results from estimating equation (3.1) with city-level formal residential housing indicators as dependent variables are given in table 3.4. Columns (1)-(3) provides estimates for the impact of ULC repeal on the formal housing stock growth between2001 and2011. Columns (4)-(6) provides estimates for the impact of ULC repeal on changes in the share of formal housing residents living in newly constructed units between 2001 and 2011. The most striking feature in columns (1)-(3) of table 3.4 is that there is no significance on any of the coefficient estimates of the repeal dummies’ interaction terms. The lack of significance is also evident from the standard errors, which are, in most cases, larger than the coefficient estimates. The results remain unchanged across specifications when I include the controls and the interaction of state fixed effects and the post dummy in the regressions. In other words, the formal housing supply was unaffected by the ULC repeal. A similar picture unfolds when we look at the estimates in columns (4)-(6). In early repeal cities, new construction was unaffected. In column (4), the late repeal interaction term’s coef- ficient, which is positive and significant at 95%, indicates that the formal housing stock grew by 6% in late repeal cities. However, this effect subsides with the addition of controls and the additional fixed effects terms. 110 Table 3.4: ULC repeal impact on city-level formal residential housing Dependent variable = Formal housing indicators Log formal units Share HH in new formal (1) (2) (3) (4) (5) (6) Early repeal× Post -0.005 -0.028 -0.027 0.049 0.041 0.070 (0.034) (0.020) (0.017) (0.045) (0.032) (0.047) Late repeal× Post 0.034 0.019 0.001 0.063** 0.053 0.016 (0.070) (0.041) (0.022) (0.024) (0.034) (0.028) Log population 0.894*** 0.910*** 0.196 0.176 (0.090) (0.074) (0.182) (0.181) Shift-share index -0.051 0.146 -0.129 0.063 (0.138) (0.165) (0.411) (0.453) Log surface area 0.021 0.014 -0.015 -0.035 (0.035) (0.019) (0.052) (0.041) Share SC/ST -0.654 -0.566* -0.845 -1.331 (0.733) (0.307) (1.194) (1.248) Share females -0.816 0.159 -1.186 1.391 (1.680) (1.257) (3.326) (3.731) Share literate 0.245 1.062*** -0.183 0.012 (0.659) (0.340) (0.733) (1.100) Mean temperature -0.121*** -0.024 -0.003 0.110* (0.025) (0.041) (0.040) (0.056) Log rainfall 0.103 -0.036 0.265** 0.181 (0.108) (0.047) (0.119) (0.200) City fixed effects ✓ ✓ ✓ ✓ ✓ ✓ Time trend ✓ ✓ ✓ ✓ ✓ ✓ State fixed effects × Post ✓ ✓ Population-weighted ✓ ✓ ✓ Housing data source Census Census Census NSS NSS NSS N 404 404 404 404 404 404 R 2 0.864 0.949 0.966 0.027 0.106 0.198 Source: Author’s calculations. Note: Table presents two-way fixed effects regression estimates using a balanced panel of 202 cities observed in 2001 and 2011. Dependent variable in columns (1)-(3) is the log of formal housing stock. Dependent variable in columns (4)-(6) is the share of formal housing residents (HH) living in new units based on survey data. New formal units are defined as those constructed in the previous decade. Early repeal is a dummy equal to one if the city repealed ULC during1999-2000, and zero otherwise. Late repeal is a dummy equal to one if the city repealed ULC during2002-2008, and zero otherwise. Standard errors are clustered by the states of administration and reported in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. 3.4.3.2 Informal housing The results from estimating equation (3.1) with city-level informal residential housing indicators as dependent variables are given in table 3.5. Columns (1)-(3) provides estimates for the impact of ULC repeal on the informal housing stock growth between 2001 and 2011. Columns (4)-(6) 111 Table 3.5: ULC repeal impact on city-level informal residential housing Dependent variable = Informal housing indicators Log informal units Share HH in new informal (1) (2) (3) (4) (5) (6) Early repeal× Post 0.011 -0.009 -0.079 0.017 0.006 0.058 (0.066) (0.061) (0.047) (0.093) (0.087) (0.095) Late repeal× Post -0.138* -0.130** -0.046 -0.150*** -0.176** -0.211*** (0.066) (0.054) (0.070) (0.039) (0.061) (0.052) Log population 0.653*** 0.592*** 0.287 0.342 (0.106) (0.123) (0.202) (0.225) Shift-share index -0.297 -0.251 1.004 0.845 (0.398) (0.341) (0.988) (1.004) Log surface area 0.039 0.147** -0.018 0.010 (0.060) (0.061) (0.083) (0.106) Share SC/ST -0.442 0.412 -1.208 -1.937 (1.364) (0.790) (2.485) (2.216) Share females -1.282 -3.770 0.792 5.992 (4.917) (4.494) (7.372) (6.212) Share literate -1.261 -1.066 4.129 2.087 (1.586) (1.573) (2.748) (2.633) Mean temperature 0.202** 0.121 0.107 0.041 (0.072) (0.108) (0.124) (0.185) Log rainfall -0.164 -0.053 0.255 0.184 (0.153) (0.167) (0.193) (0.360) City fixed effects ✓ ✓ ✓ ✓ ✓ ✓ Time trend ✓ ✓ ✓ ✓ ✓ ✓ State fixed effects × Post ✓ ✓ Population-weighted ✓ ✓ ✓ Housing data source Census Census Census NSS NSS NSS N 404 404 404 404 404 404 R 2 0.008 0.178 0.434 0.086 0.134 0.272 Source: Author’s calculations. Note: Table presents two-way fixed effects regression estimates using a balanced panel of 202 cities observed in 2001 and 2011. Dependent variable in columns (1)-(3) is the log of informal housing stock. Dependent variable in columns (4)-(6) is the share of informal housing residents (HH) living in new units based on survey data. New informal units are defined as those constructed in the previous decade. Early repeal is a dummy equal to one if the city repealed ULC during1999-2000, and zero otherwise. Late repeal is a dummy equal to one if the city repealed ULC during2002-2008, and zero otherwise. Standard errors are clustered by the states of administration and reported in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. provides estimates for the impact of ULC repeal on changes in the share of informal housing residents living in newly constructed units between 2001 and 2011. All estimates of the early repeal and post dummy interaction term are insignificant. But, the estimates of late repeal interacted with the post dummy are negative and significant in most 112 columns. The negative coefficients of the late repeal interaction term in columns ( 1)-(3) indicate that the informal housing stock fell in late repeal cities, whereas the negative coefficients in columns (4)-(6) indicate that in late repeal cities, there was a decline in the share of informal households living in new units. The latter finding, in turn, implies that the number of newly constructed informal units also declined. The decline in the informal housing stock in late repeal cities was about 13% when we do not consider time-varying state-level unobservables (column (2) of table 3.5). However, with the inclusion of the state fixed effects and post dummy interaction term, the magnitude of the estimate falls, and the significance goes away. This reflects the fact that time-varying state-level unobservables, including policies during the 2000s, may have led to fewer slums in cities. This is consistent with the fact that several states of India undertook slum rehabilitation and housing redevelopment programs during the 2000s under various schemes, such as the Jawaharlal Nehru National Urban Renewal Mission (a point I revisit in section 3.4.4). The late repeal interaction term’s coefficient in column ( 6) is, however, substantially more negative than those in columns (4) and (5). These estimates imply two things. First, the late repeal of ULC led to a significant decline in the new construction of slums. The share of informal households living in newly constructed units fell by 21 percentage points. And second, time- varying state-level unobservables led to an increase in the new construction of informal houses. Together with the estimate in column (3), this implies that while state-level policies may have led to fewer slums overall through redevelopments, these policies may have also inadvertently led to new informal housing construction. This is because redevelopment programs would presumably lead to better quality housing, which would result in more migration from rural areas, thereby, driving up the demand for informal houses (Cavalcanti et al., 2019; Dutta et al., 2021a). 26 3.4.3.3 Non-residential stock The results from estimating equation (3.1) with the log of city-level non-residential stock as the outcome variable are given in table 3.6. The non-residential building stock equals the total num- ber of buildings that are used for any non-residential purpose. First, similar to its impact on informal and formal housing, the early repeal of ULC had no effect on non-residential stock 26 As discussed in Chapter 2’s section 2.3, a large share of migrants into Indian cities move into slums. 113 Table 3.6: ULC repeal impact on city-level non-residential stock Dependent variable = Log non-residential buildings (1) (2) (3) Early repeal× Post 0.019 0.009 0.018 (0.033) (0.040) (0.027) Late repeal× Post 0.152*** 0.133** 0.048 (0.042) (0.048) (0.029) Log population 0.479* 0.734*** (0.266) (0.179) Shift-share index -0.043 -0.013 (0.272) (0.336) Log surface area -0.016 -0.025 (0.042) (0.047) Share SC/ST -0.336 -0.988 (1.146) (1.027) Share females -0.160 -1.232 (2.688) (2.763) Share literate -0.155 -0.668 (0.995) (0.989) Mean temperature -0.003 0.029 (0.062) (0.056) Log rainfall 0.228 0.129 (0.154) (0.105) City fixed effects ✓ ✓ ✓ Time trend ✓ ✓ ✓ State fixed effects × Post ✓ N 404 404 404 R 2 0.786 0.811 0.867 Source: Author’s calculations. Note: Table presents two-way fixed effects regression estimates using a balanced panel of 202 cities observed in 2001 and 2011. Dependent variable is the log of non-residential or non-housing stock (buildings). Early repeal is a dummy equal to one if the city repealed ULC during1999-2000, and zero otherwise. Late repeal is a dummy equal to one if the city repealed ULC during 2002-2008, and zero otherwise. Standard errors are clustered by the states of administration and reported in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. growth. However, the estimates of the late repeal interaction term in columns (1) and (2) show that the late repeal cities witnessed a growth of about 13-15% in the non-residential stock. Al- though, column (3) shows that this effect goes away with the inclusion of the state fixed effects and post dummy interaction term. A comparison of the estimates of the late repeal and post dummy interaction term in columns (2) and (3) indicates that the growth in non-residential stock can be entirely explained by time-varying state-level unobservables. I explore this point again in section 3.4.4. 114 There is a measurement error in the non-residential stock data used in the regressions. The non-residential figures are based only on the municipalities in a city and do not include non- residential buildings located in non-municipal areas of cities. Such non-municipal areas are contiguous with the municipalities within cities and are urban enough to be considered part of the city (see appendix B.1). However, this measurement error is unlikely to cause a major problem in the analysis. The issue arises for less than half of the sample of cities where there exist non- municipal areas outside a central municipality. 27 Even in such cities where non-municipal areas exist, municipalities account for most of the population and surface area. 28 Therefore, the ULC repeal’s impact on non-municipal areas’ non-residential building stock growth will have to be substantial enough to cause the overall city-level numbers to be different from the presented estimates, which is implausible. 3.4.3.4 Redevelopment I use information on changes in dwelling status from informal to formal housing for a panel of 4,832 non-migrant households from the IHDS data to estimate the impact of ULC repeal on the probability of informal-to-formal redevelopments. Suppose that Tenure hdt represents a dummy variable which equals one if household h, residing in district d in state s, lives in an informal unit during survey round t. I estimate the following two-way fixed effects equation: Tenure hdt = γ 1 Early repeal d × Post t +γ 2 Late repeal d × Post t +γ 3 x hdt +ϱ h + Post t +ϑ hs × Post t +υ hdt (3.3) where, Post is the time trend which is defined as a dummy equal to one for the second round of survey, ϱ h represents household fixed effects, ϑ hs represents state fixed effects, υ hdt is the error term, and x includes controls at the household level, which are the log of real annual income, household size, and a dummy variable for whether the household owns land. The standard errors are clustered by the district of residence of households. The coefficients γ 1 andγ 2 in equation (3.3) are analogous to β 1 and β 2 in equation (3.1). They 27 In 2001 and 2011, 68% and 53% of cities consisted solely of municipalities, respectively. See table B.1 for details. 28 Municipalities accounted for 92% and 86% of cities’ populations in 2001 and 2011, respectively. The surface area covered by municipalities in cities was 83% in 2001 and 74% in 2011. 115 Table 3.7: ULC repeal impact on redevelopment from informal to formal housing Dependent variable = HH lives in informal unit dummy (1) (2) (3) Early repeal× Post 0.016 0.016 0.026 (0.038) (0.038) (0.034) Late repeal× Post -0.006 -0.006 0.083 (0.038) (0.038) (0.064) Log HH real income -0.009 -0.017** (0.009) (0.008) HH owns land dummy -0.010 -0.006 (0.024) (0.022) HH size -0.004 -0.004 (0.004) (0.004) Household fixed effects ✓ ✓ ✓ Time trend ✓ ✓ ✓ State fixed effects × Post ✓ N 9,664 9,664 9,664 R 2 0.016 0.016 0.045 Source: Author’s calculations. Note: Table presents two-way fixed effects regression estimates using a balanced panel of 4,832 non-migrant house- holds surveyed once during 2004-2005 and a second time during 2011-2012. HH stands for household. The ULC repeal treatments are assigned based on households’ district of residence. Early repeal is a dummy equal to one if the district-of-residence repealed ULC during 1999-2000, and zero otherwise. Late repeal is a dummy equal to one if the district of residence repealed ULC during2002-2008, and zero otherwise. Standard errors are clustered by households’ districts of residence and reported in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. provide estimates for the impact of the ULC early and late repeal treatments on the probability of households living in informal units. A higher or a lower probability of households living in informal units as a result of the ULC repeal would indicate a lower or higher probability of re- developments from informal to formal housing. Note that since the households in the regression samples are all non-migrant, the household fixed effects term captures time-invariant unobserv- ables at the city level (or, in this case, district level). Similar to equation (3.1), the interaction of state fixed effects and the post dummy controls for time-varying state-level unobservables. The results from estimating equation (3.3) are given in table 3.7. The estimates of all repeal dummies’ interaction terms are insignificant, with very large standard errors. The estimates do not change across specifications. These results mean that households living in the treatment groups of cities were not more or less likely to redevelop their dwelling units from informal to formal compared to those living in the control group of cities. In other words, the ULC repeal had no impact on redevelopments. 116 The results presented here have one caveat. The estimates capture the effect of ULC repeal on in situ redevelopments only. It is, however, possible that entire neighborhoods were converted from informal to formal houses with the relocation of previously residing informal households. In the absence of aggregated redevelopment data, it is hard to estimate such blanket forms of neighborhood-level redevelopments. 3.4.4 Robustness checks I attempt to address some identification issues by conducting robustness tests. The first issue I address is that of the departure of Tamil Nadu’s ULC act from the central government’s act, and therefore, the possibly different ULC repeal treatment received by cities of Tamil Nadu. I run regressions by excluding Tamil Nadu from the sample and do not find that the results are qualitatively different. The estimates for these regressions are given in table A.13. The second identification issue in the difference-in-difference specification in equation ( 3.1) is that the repeal of the ULC act happened in a phased manner. A recent body of literature has highlighted the problems with the two-way fixed effects estimation of difference-in-difference specifications when the treatment effect is not the same across groups (Athey and Imbens, 2022; Baker et al., 2022; Callaway and Sant’Anna, 2021; De Chaisemartin and d’Haultfoeuille, 2020; Goodman-Bacon, 2021). 29 This can happen when the treatment is phased. Unfortunately, the solutions offered to this problem in the literature require more than two time periods. Therefore, they are not applicable to my estimation strategy. I partially address this problem by running regressions with an alternative specification con- sisting of three treatment dummy variables—Late repeal decomposed into Late repeal I and Late repeal II, and Early repeal remaining unchanged. Late repeal I equals one if the given city repealed ULC between 2002 and 2006, and zero otherwise. Late repeal II equals one if the city repealed ULC between 2007 and 2008, and zero otherwise. The results for this alternative specification are given in table A.14. Qualitatively, these estimates are similar to the main regression results in tables 3.4 to 3.6. Note that the decomposition of the late repeal treatment effect does not re- solve the identification issue entirely because the repeal effect within the decomposed treatment groups is still likely to differ. The robustness check only indicates that the identification issue 29 For a review of the literature, see De Chaisemartin and d’Haultfoeuille (2022). 117 arising from the phased repeal may not be a major problem. The third identification issue arises due to the various confounding events that occurred dur- ing the 2000s, which may have also affected the housing supply in the sample cities. Although, controlling for the interaction of the state fixed effects and the post dummy in the regression results presented in tables 3.4 to 3.6 accounts for some of these confounding factors. It is still possible that within states there were time-varying unobservables that may have potentially af- fected land markets. The most important example of such a confounding event was the implementation of the Jawaharlal Nehru National Urban Renewal Mission (JNNURM). It was a project launched by the central government of India in 2004 to invest in urban public infrastructure. A condition for states to receive funds for the project was that they had to repeal the ULC act if they had not already done so. To test whether the JNNURM implementation could be a confounding event in the estimates of the ULC repeal impact presented in tables 3.4 to 3.6, I run triple difference regressions using the JNNURM implementation as an additional treatment. The results for these regressions are given in table 3.8. The implementation of JNNURM had no impact on formal and informal residential supply in early repeal cities and informal supply in late repeal cities (columns (1) and (2) of table 3.8). These effects are confirmed by the null effects of the interaction terms on both new formal and informal construction (columns (4) and (5) of table 3.8). However, column (1) shows that, in late repeal cities that implemented JNNURM, there was an increase in the formal residential housing stock by about 8%. Along with the null effect of the late repeal and JNNURM interaction term on new formal and informal residential construction, the growth in formal residential housing implies a redevelopment of informal to formal houses as a result of JNNURM. This is consistent with the fact that one of the goals of the JNNURM project was to rehabilitate slum dwellers through in situ redevelopments. The most important result, however, is that the early and late repeal dummies interacted with JNNURM and the post dummies have a positive and significant effect on the growth of non-residential stock (column (3) of table 3.8). The magnitude of this impact is around 13-14%. Recall from the estimates in table 3.6 that the impact of late ULC repeal on non-residential stock growth, without controlling for time-varying state-level unobservables, was of a similar 118 Table 3.8: Isolating JNNURM in the ULC repeal impact on city-level supply Dependent variable = Housing and non-residential indicators Log of stock Share HH in new units Formal residential Informal residential Non- residential Formal residential Informal residential (1) (2) (3) (4) (5) Early repeal× JNNURM 0.023 -0.056 0.139** -0.015 -0.122 × Post (0.043) (0.096) (0.055) (0.041) (0.289) Late repeal× JNNURM 0.079* -0.042 0.138* 0.010 -0.387 × Post (0.037) (0.118) (0.077) (0.057) (0.290) Early repeal× Post -0.010 0.008 0.003 0.005 -0.011 (0.019) (0.065) (0.040) (0.040) (0.091) Late repeal× Post -0.003 -0.121 0.126*** -0.010 0.053 (0.034) (0.087) (0.037) (0.062) (0.055) JNNURM× Post -0.056* 0.031 -0.140** 0.070** 0.166 (0.026) (0.080) (0.052) (0.031) (0.268) Log population 0.902*** 0.650*** 0.498* 0.174 0.270 (0.089) (0.105) (0.263) (0.179) (0.187) Shift-share index -0.016 -0.301 0.030 -0.205 0.948 (0.127) (0.384) (0.267) (0.404) (1.008) Log surface area 0.020 0.039 -0.018 -0.008 -0.026 (0.035) (0.060) (0.044) (0.050) (0.078) Share SC/ST -0.650 -0.495 -0.259 -0.800 -1.490 (0.705) (1.383) (1.228) (1.108) (2.647) Share females -0.814 -1.166 -0.310 -1.548 1.022 (1.710) (5.005) (2.810) (3.380) (7.419) Share literate 0.244 -1.255 -0.163 -0.137 4.112 (0.663) (1.579) (1.008) (0.712) (2.824) Mean temperature -0.123*** 0.203** -0.007 0.006 0.125 (0.026) (0.074) (0.060) (0.036) (0.119) Log rainfall 0.099 -0.156 0.208 0.246** 0.271 (0.111) (0.148) (0.147) (0.104) (0.195) City fixed effects ✓ ✓ ✓ ✓ ✓ Time trend ✓ ✓ ✓ ✓ ✓ Population-weighted ✓ ✓ Housing data source Census Census Census NSS NSS N 404 404 404 404 404 R 2 0.949 0.172 0.813 0.117 0.142 Source: Author’s calculations. Note: Table presents two-way fixed effects regression estimates using a balanced panel of 202 cities observed in 2001 and 2011. Dependent variables in columns (1)-(5) are the logs of formal housing, informal housing, and non- residential stock and the shares of formal and informal housing residents living in new units. New units are defined as those constructed in the previous decade. Early repeal is a dummy equal to one if the city repealed ULC during 1999-2000, and zero otherwise. Late repeal is a dummy equal to one if the city repealed ULC during 2002-2008, and zero otherwise. JNNURM is a dummy equal to one if the city implemented JNNURM, and zero otherwise. Standard errors are clustered by the states of administration and reported in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. 119 magnitude. This implies that the entirety of the non-residential stock growth in late repeal cities seen in columns (1) and (2) of table 3.6 can be explained by JNNURM. Column (3) of table 3.8 shows that JNNURM led to a growth in non-residential stock in early repeal cities as well. These estimates reflect the fact that the JNNURM heavily invested in public infrastructure, which may have led to a proliferation of industrial and business activities in the treated cities. The results presented here also contrast with existing evidence on the effect of ULC repeal on industrial growth in India. While prior literature has found that the repeal of ULC led to industrial growth (Sood, 2019) through a better allocation of factors of production (Duranton et al., 2015), these studies did not consider the confounding effect of JNNURM. The evidence presented in this chapter is more robust to the additional confounding events. 3.4.5 Falsification tests As with any difference-in-difference estimation strategy, I need to satisfy the parallel trends assumption. The literature on difference-in-difference has used a variety of techniques, including pre-treatment analysis and treatment within counterfactual observations. Unfortunately, I do not have data on any of the outcome variables at the city level prior to 2001. Therefore, I am unable to present any pre-treatment trends. Instead, I conduct falsification tests by estimating the impact of fake treatments generated in the counterfactual sample of 157 cities on the outcome variables—log of formal and informal housing units, log of non-residential stock, and shares of formal and informal households living in new units. I generate two sets of fake treatments based on pseudo-random variables following Bernoulli distributions with success probabilities reflecting the shares of treated and untreated cities in the full sample. In both sets of treatments, I first assign a false ULC act treatment to some of the 157 counterfactual cities. Then, in the first set of fake treatments, which I call treatment I, I first assign a False early repeal I treatment. The False late repeal I treatment is, by default, assigned to the remaining cities in the fake ULC act treatment sample since the False early repeal I and the False late repeal I treatment samples are mutually exclusive. In the second set of fake treatments that I call treatment II, I first assign a False late repeal II treatment, which also determines the False early repeal II treatment. The idea here is that the fake repeal treatments are assigned to counterfactual cities, conditional on them receiving the false ULC act treatment. 120 I run 1,000 regressions using the counterfactual sample with randomly-drawn fake repeal treatments at each iteration replacing the ULC repeal dummies in equation (3.1). The densities of p-values for the coefficients of fake treatments are plotted in figures A. 6 and A.7. For the most part, the jumps in the density of p-values do not occur below 0.1, suggesting that there is no particular pattern or significance of any of the fake treatment coefficient estimates. These results provide evidence in support of the parallel trends assumption. 3.4.6 Summary The results presented in sections 3.4.3 and 3.4.4 can be broadly summarized into three points. First, the ULC repeal (late or early) had no impact on formal housing supply. This finding is ex- plained by the fact that the repeal had no impact on either the new construction of formal housing or informal-to-formal redevelopments. Second, the informal housing stock was unaffected by the ULC repeal (late or early). But, there was a decline in the new construction of informal houses in late repeal cities. And finally, the ULC repeal had no impact on non-residential stock growth. The confounding effects of state-level unobservables and the JNNURM program implementation explain the increase in the non-residential stock in late repeal cities. The evidence presented in this section is consistent with some prior work showing that the property markets in ULC repeal cities did not show any signs of additional growth (Pethe and Nallathiga, 2017; Pethe et al., 2014) and property prices remained high in cities like Mumbai (Siddiqi, 2013). The null effect of ULC repeal on the non-residential stock is supported by the fact that, even though the ULC repeal may have led to better credit access among mid-sized firms because of their ability to assemble land parcels, the overall impact of the repeal on credit access was not positive (Kimura, 2013). On the other hand, the null effect on non-residential stock contrasts with the evidence provided by some papers (Duranton et al., 2015; Sood, 2019). 3.5 Legal proceedings in ULC enacting cities In this section, I empirically explore whether district courts of cities that had enacted ULC regis- tered higher numbers of land-related legal proceedings in the post-repeal era. 121 3.5.1 Legal challenges and housing markets The link between legal challenges and the construction of housing has been explored in the literature before. Gandhi et al. (2021) show that litigations against real estate developers slow and inhibit housing construction activities in India. This is because litigated land parcels cannot be used for construction until a court of law allows it. Furthermore, if legal proceedings concerning a land parcel arise due to disputed ownership rights, then a court of law has to settle such ownership disputes before any construction can take place. Even after courts pass their rulings, contesting parties may appeal against such decisions. Therefore, higher numbers of land-related legal proceedings may lead to frozen housing markets for a while. Section 3.4 show that the housing and non-residential supply were mostly unaffected by the repeal of ULC during the 2000s. There was a decline in the new construction of informal houses, but the construction and overall quantity of formal houses remained unaffected by the ULC repeal. Section 3.2 discussed the legal issues surrounding the ULC repeal act, which led to governments and landowners contesting ownership rights over ULC act-identified excess vacant land parcels in courts. It is possible that these legal proceedings that ensued after the ULC repeal act led to the freezing of several ULC act-identified vacant land parcels, thereby, slowing formal housing construction. It is hard to get data on the exact (or even an estimated) number of legal proceedings resulting from the ULC repeal’s legal loopholes. However, I can test whether district courts in cities where ULC was enacted witnessed higher numbers of land-related court cases in the post-repeal era. 30 In the following section, I present the empirical estimation of this hypothesis. 3.5.2 Estimating equation I use data on the number of land-related court cases registered in district courts of the 202 cities in the sample between 2010 and 2016 (Ash et al., 2021). The outcome variable is the number of land-acquisition and land-grabbing court cases per 100,000 people (based on 2011 population of a city). The treatment variable of interest is a ULC act dummy, which equals one if a city enacted ULC, and zero otherwise. 30 I use the terms legal proceedings and court cases interchangeably. 122 Figure 3.2: Land-related legal proceedings vs. 1971 city population quintiles 4.83 5.78 3.03 .72 2.39 0 4 8 12 No. of land−related court cases per 100,000 1 2 3 4 5 1971 city population quintiles Mean 95% C.I. Source: Author’s calculations based on Census of India and Ash et al. (2021). Note: Figure plots the mean number of land-related legal proceedings per 100,000 (based on 2011 population) reg- istered between 2010 and 2016 in each 1971 city population quintile. Points represent mean values (labeled) in a quintile. Vertical lines represent 95% confidence intervals. The 1971 population bounds for each quintile are as fol- lows. Quintile 1’s population ranges from 26,602 to 53,330. Quintile 2’s populations are between 53,331 and 80,101. Quintile 3 has populations between 80,102 and 125,183. Quintile 4’s populations are between 125,184 and 225,396. Finally, quintile 5’s population ranges from 225,397 to 6,596,370. The ULC act was implemented in cities with at least 200,000 people in 1971. The 1971 pop- ulation of a city, through its effect on the 2011 population, would also affect the number of land-related legal proceedings in the city. This is because the number of legal proceedings (per 100,000) may vary systematically across cities of different sizes. For instance, smaller cities may have greater infrastructure requirements as they become more urban. As a result, governments in smaller cities might be acquiring agricultural land in the cities’ peripheries for development projects, leading to more legal challenges against land acquisition. Hence, there could be an omitted variable bias in regressions that ignore the 1971 population distribution of the sample of cities. Figure 3.2 plots the average number of land-related legal proceedings per 100,000 registered in every 1971 city population quintile. The number of legal proceedings declines as one moves from the first to the fourth quintile. At the fifth quintile, there is a three-fold jump in the number 123 of legal proceedings from the fourth quintile. The fifth quintile’s lower bound is 225,397 people. Hence, most cities in the fifth quintile enacted ULC. This suggests that the ULC treatment may be driving up the number of legal proceedings in the fifth quintile. More broadly, figure 3.2 indicates that the relationship between legal proceedings and the 1971 population distribution may be non-linear. The regressions used to estimate the association between enacting ULC and the number of registered land-related legal proceedings are run at the city level. Consider city i in state s. Then, land-related legal proceedings can be regressed on the ULC act dummy with the following equation: Legal proceedings is = α 0 +α 1 ULC act is +α 2 X is + f(Z i )+ω s +ε is (3.4) where, Legal proceedings is represents the number of land-related court cases per 100,000 people (calculated based on 2011 population), ULC act is the treatment dummy, f(Z i ) is a non-linear function of the 1971 population (Z i ) of the city, ω s represents a vector of state fixed effects, ε is is the error term, and X is is a vector of controls that include the 2011 values of log population (not used in specifications where the log 1971 population is included, to avoid collinearity), an employment shift-share index, log surface area, and population’s shares of scheduled castes and tribes individuals, females, and literate persons, mean temperature, and the log of annual aggregate rainfall. The standard errors are clustered by the states of administration of cities. The term f(Z i ) captures any potential non-linear relationship between the number of land- related court cases and the 1971 city population. For this non-linear function, I choose three candidates. The first candidate is the log of 1971 population and its square. The second specifi- cation includes three dummy variables for three intervals of 1971 population. And for the third specification, I consider a vector of 1971 city population percentile fixed effects. Since many cities in the sample did not register any land-related legal proceedings between 2010 and 2016, there are a large number of zeros on the left-hand side of equation (3.4). 31 Therefore, instead of Ordinary Least Square (OLS) regressions, I employ the pseudo maximum- likelihood estimator of Poisson regression models discussed by Silva and Tenreyro (2006). This is 31 Out of the 202 cities in the sample, 43 registered at least one land-related court case. 124 to avoid inconsistency in the estimates ofα, which arises in OLS if the error term is heteroskedas- tic. 3.5.3 Empirical estimates The results from the estimation of equation (3.4) are presented in table 3.9. Column (1) provides the estimate without any non-linear function of 1971 city population as a control. Column (2) includes the log of 1971 population and its square. Column (3) includes three dummy variables for three 1971 population intervals: (1) 100,000 to 150,000, (2) 150,000 to 200,000, and (3) above 200,000. The omitted category is the population interval of 0 to 100,000. And finally, column ( 4) provides the results from including 1971 city population quintile fixed effects. The estimates given in column (1) of table 3.9 indicate that the relationship between ULC enactment and the number of land-related legal proceedings is positive but insignificant when we omit the non-linear function of 1971 population of the city. This is expected because figure 3.2 indicates that the raw correlation between the number of legal proceedings and ULC enactment should be either negative or weakly positive. When I include the non-linear indicators, the magnitude of the association between ULC enactment and the number of land-related legal proceedings increases, and the estimates turn significant. The column ( 2) estimate shows that ULC enactment is associated with five times (e 1.894 − 1) higher numbers of land-related legal proceedings. A similar magnitude of the esti- mate is realized in column (4), where the non-linear control is a vector of 1971 city population quintile fixed effects. Table A. 15 provides estimates with varying numbers of1971 city population percentile fixed effects. The estimates across these specifications are consistent. The estimate in column (3) is substantially higher in magnitude—ULC enactment is associ- ated with eight times (e 2.224 − 1) higher numbers of land-related legal proceedings. This higher magnitude could be the result of the interval bands, which are essentially interactions of the ULC act dummy and the interval band dummies. 32 In other words, the column (3) estimate presents a comparison of legal proceedings in ULC enacting and no-ULC cities within the highest popu- lation band (1971 population greater than 200,000). 32 ULC enacting cities exist only in the uppermost band with a 1971 population cutoff of 200,000. Therefore, the coefficient α 1 in equation (3.4) can be interpreted as an interaction between the ULC act dummy and 1{Z≥ 200000}. 125 Table 3.9: ULC act and land-related legal proceedings Dependent variable = # Land-related court cases, 2010-2016 (1) (2) (3) (4) ULC act 0.815 1.894*** 2.224** 1.758*** (0.798) (0.583) (0.933) (0.523) Log 1971 population 7.203 (10.232) Log 1971 population squared -0.365 (0.452) 1{100000≤ Z< 150000} -2.862* (1.481) 1{150000≤ Z< 200000} -3.973*** (0.875) 1{Z≥ 200000} -4.503 (3.372) Log population -1.199*** -0.101 -0.325 (0.345) (0.578) (0.445) Shift-share index 11.724*** 12.510*** 11.307*** 13.981*** (1.096) (1.540) (3.333) (1.541) Log surface area -0.165 -0.459*** 0.057 0.109 (0.196) (0.088) (0.447) (0.257) Share SC/ST 4.944*** 2.927*** 2.002*** -1.399 (0.303) (0.447) (0.758) (1.017) Share females -42.782 -60.310** -83.697 -80.864*** (28.772) (29.833) (65.700) (23.230) Share literate 2.363 2.654 0.387 1.676 (2.239) (2.072) (3.592) (3.180) Mean temperature 0.089 0.377* 0.265 0.532** (0.128) (0.228) (0.430) (0.212) Log rainfall -0.101 0.290 0.512 -0.599* (0.575) (0.609) (1.331) (0.343) Constant 19.921** -45.100 11.993 -7.066 (8.177) (64.501) (18.180) (7.932) (Z = 1971 population) State fixed effects ✓ ✓ ✓ ✓ Z quintile fixed effects ✓ N 202 202 202 202 R 2 0.693 0.707 0.746 0.743 Source: Author’s calculations. Note: Table presents pseudo maximum likelihood estimates of Poisson regressions using a cross-section of 202 cities. Dependent variable is the number of land-related legal proceedings per 100,000 (based on city population in 2011) registered between 2010 and 2016. ULC act is a dummy equal to one if the city had enacted the ULC act, and zero otherwise. Standard errors are clustered by the states of administration and reported in parentheses. * p< 0.10, ** p < 0.05, *** p< 0.01. 126 The results in table 3.9 indicate that cities that enacted ULC had substantially higher numbers of land-related legal proceedings compared to those that did not enact ULC. The most conserva- tive estimate is that the number of legal proceedings in ULC enacting cities was six times as high as in other cities (column (4) of table3.9). These results somewhat confirm the anecdotal evidence on legal issues surrounding the ULC repeal act discussed in section 3.2.3. The higher numbers of legal proceedings partly explain why the construction of formal housing was unaffected by the repeal of ULC. There are a few caveats in this analysis. First, the legal proceedings figures at the city level are calculated based on district-level figures, inducing noise in the data. However, despite the noise, the coefficient estimates and the standard errors seem relatively stable across specifications. Sec- ond, even after accounting for the non-linear function of the 1971 city population, the state fixed effects, and the vector of controls, there may be omitted variable bias. The ULC enactment may be associated with a variety of unobservable institutional factors that might also be correlated with the number of legal proceedings in ULC-enacting cities. And finally, the legal proceedings figures used in the outcome variable consist of all land-related acquisition cases. The number of land-related legal proceedings as a direct consequence of the ULC act or its repeal is lower than or equal to the total number of land-related legal proceedings. This suggests that the estimates presented in table 3.9 are upper bounds of the true impact of ULC on the number of land-related legal proceedings. 3.6 Discussion In this section, I integrate and explain the findings from sections 3.4.3 and 3.5 and discuss the policy implications of this research. 3.6.1 Mechanisms The empirical findings in sections 3.4.3 and 3.5 can be summarized into two salient points. First, the ULC repeal of India did not lead to any formal housing supply growth, and the late repeal cities witnessed a decline in the new construction of informal houses. Second, cities that enacted ULC registered six times as many land-related legal proceedings as cities that never enacted ULC. 127 I place these empirical findings in the context of a monocentric city model following Henderson et al. (2021). 3.6.1.1 Set-up Consider a monocentric city where the agents are resident consumers and non-resident landown- ers. Consumers consume housing and a numeraire tradable consumption good. Each landowner is born with a unit parcel of land in the city and constructs houses on them by hiring labor and capital from outside the city. Consumption and production are place- and time-specific. Place and time are represented with d and t, respectively. We can interpret d as the distance from the center of the city. I also incorporate a heterogeneous cost of development before any construction could occur on land parcels. The heterogeneity in development cost ensures that there will be vacant land parcels within the city. 3.6.1.1.1 Bid-rent Consumers derive utility from housing space v(d, t) and a consumption good m(d, t). They also obtain utility from amenities A that are fixed at the city level. The rental price per unit of housing space, or the bid-rent, is p(d, t). Consumers earn wage w(d, t), and their utility is given by a standard Cobb-Douglas function with parameter η. Since the city is open with costless migration, the utility obtained by consumers is fixed at U. Suppose that productivity λ is determined outside the city and it grows exponentially, at a constant rate, over time. 33 The transport cost-adjusted wage falls with distance d from the center at rate µ. Then, the bid-rent can be recovered as a function of productivity and transport cost, as follows: p(d, t)= pe λt− µx (3.5) The complete derivation of the bid-rent function is given in appendix C. Equation (3.5) implies that the bid-rent increases at a constant exponential rate of λ over time and falls at a constant 33 In a richer model, it is possible to incorporate a place-specific production of the tradeable sector commodity, allowing productivity to be determined endogenously. However, that is unlikely to affect the key intuitions in the model since I assume that landowners hire labor and capital from outside the city. 128 exponential rate of µ with distance from the center of the city. 3.6.1.1.2 Housing supply There are two housing sectors, informal and formal, that are different in their housing production technologies. I derive the optimal rents for the 2πd landowners owning unit parcels at each distance d from the city center. 3.6.1.1.2.1 Land values outside the city I deviate from the standard model where the land value outside the city is unique and is fixed by the exogenous agricultural rent. Instead, I incorporate a cost of development on land (Plantinga et al., 2002). Moreover, I introduce heterogeneity in land characteristics that induces differences in land values outside the city. Outside the city, at each distance d from the city center, θ∈ [0, 1] share of land parcels are unfit for cultivation or housing construction. Cultivation or housing construction on such parcels requires a fixed cost of development C. 34 The remaining (1− θ) share of landowners earn agricultural rent r from employing their parcels in cultivation. 3.6.1.1.2.2 Building technologies Landowners can choose between informal and formal housing supply in place d at time t. I use the indexes I and F to denote informal and formal housing, respectively. Housing supply is defined by the volume of space supplied on every parcel of land. Since I normalize the area of each parcel to unity, the volume of housing per parcel equals the height of the parcel’s building h T , where T = {I, F}. For both informal and formal housing, I model a convex iso-elastic cost function in the height of buildings, but with different parameters. 35 Furthermore, formal 34 The cost of development can be interpreted in two ways. First, there could be natural barriers to cultivation or housing construction on land parcels requiring development costs, such as forests, marshes, or steep slopes. In such cases, the cost would reflect the expenditure incurred by land owners in leveling land before development could commence. Second, there could be institutional barriers to development, such as ownership status. As discussed in section 3.2.2, land ownership is presumptive in India and can be challenged in the judicial system. In this case, the development cost represents the legal fees for settling ownership disputes before cultivation or housing development can commence. In all cases, the cost C is incurred by developers outside the city. Therefore, it does not affect the bid-rent in the city. 35 Owing to the differences in their height, Henderson et al. (2021) models different cost structures for informal and formal houses—informal houses increase in volume through cover, whereas, formal volume expands through height. This motivates the differences in cost functions between informal and formal housing production in their paper. However, in India, both informal and formal housing supply increases primarily through cover because Indian cities impose some of the world’s most restrictive floor area ratio (FAR) regulations (Brueckner and Sridhar, 2012). The average heights of informal and formal buildings are around 1-1.5 (National Sample Survey Organization, 2012). Therefore, the functional form of building technologies for informal and formal houses need not be different. Any 129 buildings, once constructed, remain for an infinite amount of time, whereas, informal houses undergo changes at every period. 36 Therefore, the construction costs are sunk in the case of formal houses, whereas, construction costs are incurred at every period for informal houses. 3.6.1.1.2.3 Informal housing production Suppose that the land rent from informal housing supply in d at time t is r I (d, t). The cost of construction of an informal building with height h I is iso-elastic in height, with elasticity parameter ψ> 1. Informal houses continuously evolve through time and grow taller. Therefore, the land rent that maximizes building height (derived in appendix C) will be: r I (d, t)= (ψ− 1) (ψ− 1) ψ ψ p(d, t) ψ (3.6) 3.6.1.1.2.4 Formal housing production Landowners constructing formal houses, choose the height of the building h F (d,τ), at the date of construction τ. Construction costs are sunk and iso-elastic in height with parameter φ > 1. Landowners’ rent from formal housing construction R F (d,τ), at time τ, that maximizes the building height, will be: R F (d,τ)= (φ− 1) (φ− 1) φ φ Φ(d,τ) φ (3.7a) where, Φ(d,τ) = R ∞ τ p(d, t)e − ρ(t− τ) dt is the accrued lifetime revenue per unit of space for a formal building at the time of construction. The formal land rent net of amortized construction costs, at any time t> τ, is 1 φ share of formal land revenue for the given period and can be written as follows: r F (d, t)= 1 φ p(d, t)h F (d,τ)= 1 φ p(d, t) φ− 1 φ Φ(d,τ) ! (φ− 1) (3.7b) Equations (3.7a) and (3.7b) are derived in appendix C. differences in technologies will be captured by the difference in parameters. 36 Unlike Henderson et al. (2021), I make this assumption because modeling formal housing redevelopment does not affect landowners’ decision to switch between informal and formal housing construction and construction on vacant plots. 130 3.6.1.1.3 Land development In the benchmark scenario without regulations, when there are no land development costs, hous- ing development occurs on agricultural land in the city peripheries as the city expands. First, informal housing development occurs when a parcel is on the city boundary at timeτ 0 , followed by formal housing construction at a future date τ 1 , when the parcel is inside the city. For this scenario, the present discounted value of a parcel of land can be written as follows: Π 0 (d)= Z τ 0 0 re − ρt dt+ Z τ 1 τ 0 r I (d, t)e − ρt dt+ R F (d,τ 1 )e − ρτ 1 (3.8) Landowners choose τ 0 andτ 1 to maximize the discounted land valueΠ 0 (d). The first-order con- ditions provide the optimal bid-rents p(d,τ 0 ) and p(d,τ 1 ) for which landowners initiate informal and formal housing construction, respectively. For informal housing construction to precede for- mal housing construction, the agricultural rent needs to be less than a constant term,Θ, defined in appendix C. 3.6.1.2 Analysis I derive the conditions under which vacant land exists in the city without regulations and the sequence of vacant land development. Then, I discuss how the ULC act perturbs the process of vacant land development. Finally, I explore the implications of the ULC repeal law for informal and formal housing construction and how the legal costs incurred by landowners with disputed ownership status affect housing supply. 3.6.1.2.1 Assumptions I begin by explicitly stating the model assumptions. Assumption 3.1. The cost functions for informal and formal housing are iso-elastic with parameters ψ and φ, respectively, such that, φ> ψ> 1. Assumption 3.2. The productivity parameter λ and the discount rate ρ are, such that, 1 > ρ > λφ > λψ> λ. Assumption 3.3. The strict upper bound of the agricultural rent r isΘ, such that,Θ> 1. 131 3.6.1.2.2 Land development without regulations In the scenario without regulations, first, I examine how the heterogeneous cost of development can lead to the existence of vacant land within the city. Next, I explore the circumstances under which only formal housing construction will take place on land parcels requiring development costs. Proposition 3.1. Under the assumptions 3.1 and 3.2, a necessary condition for theθ share of land parcels at distance d from the city center requiring development costs to stay vacant, once inside the city, is C> r ρ . Proof: See appendix C. To understand the intuition, note that proposition 3.1 implies that the development cost is higher than the agricultural rent (since ρ < 1). Therefore, when all land parcels at a distance d from the city center are rural (radius of the city is less than d), the θ share of land parcels requiring development costs are not used for agriculture, and they remain fallow. When the city expands to a radius of d, agricultural land is used for informal housing construction, but the fallow land parcels remain unused for a time. In other words, these land parcels exist as vacant lots inside the city. Once the bid-rent is high enough to offset the development cost, the vacant land parcels are used for housing construction. Proposition 3.2. Under the assumptions 3.1 and 3.2, a sufficient condition for the θ share of land parcels at distance d from the city center requiring development costs to stay vacant once inside the city, and a necessary condition for only formal housing construction to occur on vacant land, is C> Θ ρ Proof: See appendix C. Proposition 3.2 provides a more restrictive condition on the development cost than propo- sition 3.1 because Θ is a strict upper bound of r. The intuition in proposition 3.2 is that if the development cost is sufficiently high, there will be a jump in the development process, and vacant land parcels will be used only for formal housing construction. This is because the discounted values of rents earned by landowners from informal houses will not be enough to offset the development cost if proposition 3.2 holds. 3.6.1.2.3 Land development under regulations Under the given conditions, when landowners incur high land development costs, they keep land 132 vacant until it is optimal to initiate formal housing construction. However, land-use regulations may upend this process. In this section, I analyze the land development process under the ULC regulation. 3.6.1.2.3.1 Defining the ULC act To examine the impact of ULC on informal and formal housing supply, I introduce a ULC act- enforced fixed regulatory cost G on formal housing construction occurring on vacant land in the city. 37 Landowners incur the cost G only once, when they initiate formal housing construction on excess vacant plots. Anecdotal evidence suggests that informal housing development took place on excess vacant land identified by the ULC departments of cities like Mumbai (Pethe and Nallathiga, 2017). Therefore, because of the lax enforcement of the ULC act, the ULC regulatory cost does not apply to vacant land used for informal housing development. The city’s state government collects the regulatory revenue. The government expends such revenue on public goods outside the city. Therefore, the regulatory cost does not affect the bid-rent in the city. 3.6.1.2.3.2 Land development under the ULC act I examine the conditions under which development on vacant land occurs in a city regulated by the ULC act. Proposition 3.3. Under the assumptions 3.1 to 3.3, and a regulatory cost G imposed by the ULC act on formal housing construction on vacant land, informal housing development precedes formal housing construction on the θ share of land parcels at distance d from the city center requiring development costs, if the following condition holds: G> C n (ΘρC) ( ϕ− ψ ψ ) − 1 o Proof: See appendix C. The condition in proposition 3.3 implies that if the ULC act-enforced tax on formal housing construction on vacant land is greater than a given amount, the decision-making process is re- versed. Landowners find it optimal to develop informal housing on vacant land because of the 37 The ULC act applied to vacant land owned above ceiling limits of500-2000 square meters. But, in the model, every landowner owns a unit parcel of land. Therefore, for the sake of simplicity, I assume that the ULC act applies to all vacant land in the city. This assumption will not affect the key results because housing developments on vacant parcels are independent of developments on other land parcels in the city. A richer framework can relax this assumption and explore the implications of heterogeneous parcel size. 133 regulatory tax on formal construction. In the distant future, however, landowners will redevelop informal houses built on vacant land into formal houses when bid-rents increase sufficiently to offset the regulatory tax. 3.6.1.2.4 Land development under the ULC repeal act To see how the ULC repeal act may have affected the housing supply, suppose that the ULC act was in force until date τ L 0 . Before this date, only informal housing development was initiated on excess vacant land, and, at a later period, informal houses were redeveloped into formal houses. Under the ideal circumstances, after dateτ L 0 , landowners would initiate formal housing construc- tion on vacant land and redevelop informal houses, built on vacant land, into formal buildings. Therefore, the immediate impact of the repeal act would be a decline in the new construction of informal houses and an increment in the number of formal houses through redevelopment and new construction on vacant land. Now, consider the date τ L 1 > τ L 0 , when state governments and landowners initiate legal pro- ceedings against each other on land parcels with disputed possession and ownership. At this date, all informal and formal construction on ULC-regulated land stops. Landowners must incur a legal cost L before they can initiate new construction. The legal cost incurred by landowners is used by the judicial system outside the city. Therefore, the cost has no bearing on the bid-rent in the city. Unlike the ULC act-enforced regulatory tax, which applied only to formal housing construction on vacant land, the legal cost applies to both formal and informal housing develop- ment. This is because it is relatively difficult to bribe courts in India (Bhan, 2009; Mehta, 2007; Varshney, 1998), compared to government officials and bureaucrats. Therefore, courts are more effective at implementing their rulings. Since the legal cost applies to all forms of development on vacant land, landowners will build formal houses on vacant land if C+ L > Θ ρ , which is necessarily true under proposition 3.2. Therefore, in the post-repeal period, the pre-regulation status quo is restored. At the same time, since landowners incur the development cost and the legal cost on vacant land before initiating construction, the date for formal development is pushed into the future. In other words, the construction of formal houses on vacant land is delayed because of the legal cost. 134 3.6.1.3 Reconciling model implications with empirical estimates The model implies that immediately after the repeal of ULC, there will be a decline in the new construction of informal houses because vacant land is not used for informal housing develop- ment in the post-repeal era. Instead, landowners will initiate formal housing construction on vacant parcels and redevelop informal houses, built on vacant land, into formal buildings. The latter effect would be weaker because formal housing construction is prolonged. Therefore, in the short run, the ULC repeal should lead to (i) a decline in the new construction of informal houses, (ii) a marginal increase in new formal housing construction, (iii) a small increase in informal-to-formal redevelopment, and (iv) a decrease in the informal housing stock. There should be little to no effect on the formal housing stock in the short run. These implications are somewhat consistent with the empirical estimates of the late ULC re- peal treatment variable in tables 3.4 and 3.5. The estimates are null in the case of formal housing stock, as expected. And, informal housing construction fell in late repeal cities. Contrary to the model implications, however, there was no effect of the late repeal on new formal construction, redevelopment, and the informal stock. These empirical findings are not entirely inconsistent with the model since the model implies a weak effect on new formal construction and redevel- opment. And, in the case of the informal stock, even though insignificant, the sign of the late repeal coefficient is negative. As governments and landowners initiate legal proceedings on disputed parcels, all new in- formal and formal housing development on vacant land stops. There will be a marginal decline in the redevelopment of informal houses, built on vacant land, into formal houses. 38 As a result, the formal housing stock will fall by a small amount. The effect of the repeal on the informal housing stock could go either way. On the one hand, new informal construction declines. On the other hand, fewer informal houses are redeveloped into formal. In the medium run, the ULC repeal will (i) have no effect on the new construction of formal or informal houses, (ii) marginally reduce informal-to-formal redevelopment, (iii) marginally 38 Under the ULC act, informal houses are redeveloped into formal houses at a later date when the bid-rent increases sufficiently to offset the regulatory cost. With the imposition of the legal cost, such redevelopments will be pushed fur- ther into the future. This is consistent with anecdotal evidence that slums constructed on vacant land, identified by the ULC, exist to this day. For example, see this article in the Times of India, on the existence of slums on ULC act-identified vacant land, as of 2016, in Ahmedabad, Gujarat: https://timesofindia.indiatimes.com/city/ahmedabad/squatters-to- get-title-to-land-property-in-77-city-areas/articleshow/55882618.cms. 135 reduce the formal housing stock, and (iv) have an ambiguous effect on the informal housing stock. These implications are consistent with the estimates of the early ULC repeal treatment variable in tables 3.4 and 3.5. The early repeal of ULC had no impact on the new construction of formal and informal houses. The effects on redevelopment and the formal housing stock are null, which is not implausible considering that the model implies a weak negative effect on both. The informal housing stock is also unaffected in early repeal cities, which is explained by null effects on redevelopment and new informal construction. 3.6.2 Implications for policies and future research Since the research conducted in this chapter is an ex post analysis of a large-scale deregulatory reform, there are a couple of ways in which this work directly informs policymakers. First, this chapter shows that better and more efficient record-keeping of land ownership is essential in reducing the numerous litigations arising from property disputes in India. Second, this chapter shows that nuances in reforms and deeper institutional aspects require attention. A knowledge gap could arise if a top-down policymaking approach is followed. There- fore, the findings of this chapter underscore the need for decentralized policymaking. For in- stance, local municipal officials and competent authorities were aware of how land was acquired under the ULC act (Siddiqi, 2013). Therefore, rather than a centralized reform measure, if indi- vidual cities were allowed to implement their version of the ULC repeal act, the outcomes may have been different. In other words, decentralization may help eliminate the knowledge gap on institutional frictions that arise with centralized policymaking. The ambiguity in the effects of deregulation also merits investigation into other reforms. Future research can examine whether deregulatory measures, such as relaxing building height restrictions in Indian cities, can improve housing market outcomes by reducing the share of a city’s population living in slums. It is possible that relaxing building height restrictions may lead to a higher formal housing supply but also fewer redevelopments from informal to formal. In this case, the informal housing supply may not reduce. This would lead to lower informal rents, attracting rural-urban migrants to cities’ slums, resulting in a higher slum population (Cavalcanti et al., 2019). 136 3.7 Conclusion This chapter explores how land-use deregulation affects housing supply when there are insti- tutional frictions. Using data on over 200 cities in India, I show that the repeal of urban land ceiling (ULC) laws in India during the 1990s and the 2000s did not lead to formal housing sup- ply growth. This finding is explained by the ULC repeal act’s failure to recognize that a large number of vacant land parcels identified as excess under the ULC act were still in the process of acquisition, with unclear ownership, at the time of repeal. In the post-repeal period, ownership disputes between governments and landowners resulted in litigations. The number of legal pro- ceedings in the post-repeal period in cities where ULC was enforced was six times as high as in cities that never enacted ULC. The findings of this chapter indicate that the existing institutional frictions, in the form of inadequate property rights and inefficient record-keeping, can inhibit or delay the benefits of land use deregulation. More broadly, there could be other instances of institutional frictions impeding the gains from economic reforms, which merit investigation in future research. Lawmakers cannot ignore such frictions, especially in the context of developing countries, if they wish to experience the complete benefits of deregulatory reforms. 137 Conclusion This dissertation set out to study the housing markets of urban India. On the one hand, a critical aspect of housing demand, internal migration, is examined empirically. Three reasons behind the recent surge in India’s internal mobility are identified: technological growth, droughts, and highway infrastructure investments. This dissertation then estimates the slope of India’s hous- ing supply curves and provides important insight into the ways in which the housing supply is affected by existing regulatory frictions. Even though Indian cities are growing rapidly, in- creasingly through migration, the housing markets of urban India are not growing fast enough. The slow growth in housing markets can be attributed to a mix of regulatory and institutional frictions. While the primary goal of this dissertation is to make academic contributions and enhance our understanding of urban India’s housing markets, several solutions to India’s growing housing crisis emerge. The findings in chapters 2 and3 provide the insight behind the following solutions. First, as documented in a large body of academic literature, land use regulations impede housing supply. Indian cities are no strangers to such regulations. In fact, governments in India impose some of the strictest building regulations in the world. This dissertation’s key findings suggest that the existing land use regulations and eminent domain restrictions limit the supply of housing in India. Chapter 3 shows how the urban land ceiling laws, which were repealed 20 years ago, continue to impede formal housing supply in the largest Indian cities because of persisting institutional frictions. Therefore, the legacy of land use regulations may continue for a while even after they are eliminated. Chapter 3’s findings further indicate that a relatively seamless way of implementing land use deregulatory reforms can be achieved through better property rights provision. One pathway towards this goal would be to digitize ownership records. This will reduce land transaction costs and potentially lower the litigatory burden on the judicial system arising from poorly-defined ownership rights in India. While the government of India has been actively computerizing own- ership records for over two decades, a large share of the country’s land ownership records are yet to be digitized. The empirical findings in chapter 3 also suggest that a top-down approach to deregulatory 138 reform may not be successful. The federal and state governments are not always aware of the ways in which reforms may work on the ground. Therefore, decentralization of urban governance through the empowerment of urban local bodies and conferring the responsibility of reforms to municipalities can reduce this information gap in top-down administration. Despite the fact that the housing markets of urban India have been growing slowly, there is cause for optimism. In the past several years, various state governments have passed deregula- tory reforms in land use. The most important of these reforms have been initiatives in eliminating laws, by at least six states, that prohibited the purchase and conversion of agricultural land into non-agricultural use by non-farmers. 39 Several other states are in the process of undertaking land use reforms aimed at enhancing non-agricultural growth in India. A continuation of such reforms will surely improve the housing supply in India in the not-so-distant future. 39 These states are Haryana, Madhya Pradesh, Punjab, Rajasthan, and Tamil Nadu. 139 Bibliography Abeberese, A. B. and Chen, M. (2021). Intranational trade costs, product scope and productivity: Evidence from India’s Golden Quadrilateral project. Journal of Development Economics, page 102791. Accetturo, A., Lamorgese, A., Mocetti, S., and Pellegrino, D. (2021). Housing supply elasticity and growth: Evidence from Italian cities. Journal of Economic Geography, 21(3):367–396. Acemoglu, D., Johnson, S., and Robinson, J. A. (2001). The colonial origins of comparative development: An empirical investigation. American Economic Review, 91(5):1369–1401. Aker, J. C., Ksoll, C., and Lybbert, T. J. (2012). Can mobile phones improve learning? Evidence from a field experiment in Niger. American Economic Journal: Applied Economics, 4(4):94–120. Ash, E., Asher, S., Bhowmick, A., Bhupatiraju, S., Chen, D., Devi, T., Goessmann, C., Novosad, P ., and Siddiqi, B. (2021). In-group bias in the Indian judiciary: Evidence from 5 million criminal cases. Working Paper. Athey, S. and Imbens, G. W. (2022). Design-based analysis in difference-in-differences settings with staggered adoption. Journal of Econometrics, 226(1):62–79. Autor, D. H., Palmer, C. J., and Pathak, P . A. (2014). Housing market spillovers: Evidence from the end of rent control in Cambridge, Massachusetts. Journal of Political Economy, 122(3):661–717. Baker, A. C., Larcker, D. F., and Wang, C. C. (2022). How much should we trust staggered difference-in-differences estimates? Journal of Financial Economics, 144(2):370–395. Banerjee, A. and Iyer, L. (2005). History, institutions, and economic performance: The legacy of colonial land tenure systems in India. American Economic Review, 95(4):1190–1213. Banerjee, A., Iyer, L., and Somanathan, R. (2005). History, social divisions, and public goods in rural India. Journal of the European Economic Association, 3(2-3):639–647. Bartik, T. J. (1991). Who benefits from state and local economic development policies? 140 Bartik, T. J. (1993). Who benefits from local job growth: Migrants or the original residents? Regional studies, 27(4):297–311. Baum-Snow, N. and Han, L. (2019). The microgeography of housing supply. Work in progress, University of Toronto. Behera, H. C. (2009). Constraints in land record computerisation. Economic and Political Weekly, pages 21–24. Berry, C. R. and Glaeser, E. L. (2005). The divergence of human capital levels across cities. Papers in Regional Science, 84(3):407–444. Bertaud, A. and Brueckner, J. K. (2005). Analyzing building-height restrictions: Predicted impacts and welfare costs. Regional Science and Urban Economics, 35(2):109–125. Besley, T. (1995). Property rights and investment incentives: Theory and evidence from Ghana. Journal of Political Economy, 103(5):903–937. Besley, T. and Burgess, R. (2000). Land reform, poverty reduction, and growth: Evidence from India. The Quarterly Journal of Economics, 115(2):389–430. Besley, T. and Ghatak, M. (2010). Property rights and economic development. In Handbook of Development Economics, volume 5, pages 4525–4595. Elsevier. Bhan, G. (2009). “This is no longer the city I once knew”. Evictions, the urban poor and the right to the city in millennial Delhi. Environment and Urbanization, 21(1):127–142. Bhavnani, R. R. and Lacina, B. (2017). Fiscal federalism at work? Central responses to internal migration in India. World Development, 93:236–248. Blanchard, O. and Giavazzi, F. (2003). Macroeconomic effects of regulation and deregulation in goods and labor markets. The Quarterly Journal of Economics, 118(3):879–907. Boas, I. (2020). Social networking in a digital and mobile world: The case of environmentally- related migration in Bangladesh. Journal of Ethnic and Migration Studies, 46(7):1330–1347. Borusyak, K., Hull, P ., and Jaravel, X. (2022). Quasi-experimental shift-share research designs. The Review of Economic Studies, 89(1):181–213. 141 Boustan, L. P . (2010). Was postwar suburbanization “white flight”? Evidence from the black migration. The Quarterly Journal of Economics, 125(1):417–443. Brueckner, J. K. et al. (1987). The structure of urban equilibria: A unified treatment of the Muth- Mills model. Handbook of Regional and Urban Economics, 2(20):821–845. Brueckner, J. K. and Selod, H. (2009). A theory of urban squatting and land-tenure formalization in developing countries. American Economic Journal: Economic Policy, 1(1):28–51. Brueckner, J. K. and Sridhar, K. S. (2012). Measuring welfare gains from relaxation of land-use restrictions: The case of India’s building-height limits. Regional Science and Urban Economics, 42(6):1061–1067. Bryan, G., Chowdhury, S., and Mobarak, A. M. (2014). Underinvestment in a profitable technol- ogy: The case of seasonal migration in Bangladesh. Econometrica, 82(5):1671–1748. Buera, F. J. and Shin, Y. (2013). Financial frictions and the persistence of history: A quantitative exploration. Journal of Political Economy, 121(2):221–272. Cairncross, F. (1997). The death of distance: How the communications revolution will change our lives. Number C20-21. Harvard Business School. Callaway, B. and Sant’Anna, P . H. (2021). Difference-in-differences with multiple time periods. Journal of Econometrics, 225(2):200–230. Caselli, F. and Gennaioli, N. (2008). Economics and politics of alternative institutional reforms. The Quarterly Journal of Economics, 123(3):1197–1250. Cavalcanti, T., Da Mata, D., and Santos, M. (2019). On the determinants of slum formation. The Economic Journal, 129(621):1971–1991. Célerier, C. and Matray, A. (2019). Bank-branch supply, financial inclusion, and wealth accumu- lation. The Review of Financial Studies, 32(12):4767–4809. Census of India, P . (1971). Census of India 1971 Provisional Population Totals. Office of the Registrar General and Census Commissioner New Delhi. 142 Census of India, P . (2001a). Census of India 2001 Housing and Houselisting Tables. Office of the Registrar General and Census Commissioner New Delhi. Census of India, P . (2001b). Census of India2001 Provisional Population Totals. Office of the Registrar General and Census Commissioner New Delhi. Census of India, P . (2011a). Census of India 2011 Housing and Houselisting Tables. Office of the Registrar General and Census Commissioner New Delhi. Census of India, P . (2011b). Census of India2011 Provisional Population Totals. Office of the Registrar General and Census Commissioner New Delhi. Chattopadhyay, S. and Chatterjee, S. (2014). Telecom spectrum auctions in India: The theory and the practice. Available at SSRN 2400871. Chava, S., Oettl, A., Subramanian, A., and Subramanian, K. V . (2013). Banking deregulation and innovation. Journal of Financial Economics, 109(3):759–774. Collier, P . and Gunning, J. W. (1999). Explaining African economic performance. Journal of Economic Literature, 37(1):64–111. Davis, J. (2021). How do upzonings impact neighborhood demographic change? Examining the link between land use policy and gentrification in New York City. Land Use Policy, 103:105347. De Chaisemartin, C. and d’Haultfoeuille, X. (2020). Two-way fixed effects estimators with het- erogeneous treatment effects. American Economic Review, 110(9):2964–96. De Chaisemartin, C. and d’Haultfoeuille, X. (2022). Two-way fixed effects and differences-in- differences with heterogeneous treatment effects: A survey. Technical report, National Bureau of Economic Research. Deininger, K. and Goyal, A. (2012). Going digital: Credit effects of land registry computerization in India. Journal of Development Economics, 99(2):236–243. Desai, S., Dubey, A., Joshi, B., Sen, M., Shariff, A., Vanneman, R., et al. (2007-2015). Computer file, icpsr 37382. Ann Arbor: University of Maryland, National Council of Applied Economic Research, New Delhi and Inter-University Consortium for Political and Social Research. 143 Deshingkar, P . (2008). Circular internal migration and development in India. Migration and development within and across borders: Research and policy perspectives on internal and international migration. Geneva: International Organization for Migration/Social Science Research Council, pages 161–87. DGCIS, I. (2020). Foreign Trade Statistics. Directorate General of Commercial Intelligence and Statistics. Diamond, R. (2017). Housing supply elasticity and rent extraction by state and local governments. American Economic Journal: Economic Policy, 9(1):74–111. Duranton, G., Ghani, S. E., Goswami, A. G., Kerr, W., and Kerr, W. R. (2015). The misallocation of land and other factors of production in India. World Bank Policy Research Working Paper, (7221). Dutta, A., Gandhi, S., and Green, R. K. (2021a). Distant shocks, migration, and housing supply in India. SSRN working paper. Dutta, A., Gandhi, S., and Green, R. K. (2022). Do urban regulations exacerbate rural-urban inequality? Evidence from rent control in India. Evidence from rent control in India (April 25, 2022). Dutta, A., Green, R. K., Panchapagesan, V ., and Venkataraman, M. (2021b). Are move-in ready homes more expensive? Available at SSRN 3851745. Elisseeff, V . (2000). The silk roads: Highways of culture and commerce. Berghahn Books. Emerick, K. (2018). Agricultural productivity and the sectoral reallocation of labor in rural India. Journal of Development Economics, 135:488–503. Favara, G. and Imbs, J. (2015). Credit supply and the price of housing. American Economic Review, 105(3):958–92. Field, E. (2005). Property rights and investment in urban slums. Journal of the European Economic Association, 3(2-3):279–290. Field, E. (2007). Entitled to work: Urban property rights and labor supply in Peru. The Quarterly Journal of Economics, 122(4):1561–1602. 144 Galiani, S. and Schargrodsky, E. (2010). Property rights for the poor: Effects of land titling. Journal of Public Economics, 94(9-10):700–729. Gandhi, S., Green, R. K., and Patranabis, S. (2022). Insecure property rights and the housing market: Explaining India’s housing vacancy paradox. Journal of Urban Economics, page 103490. Gandhi, S., Tandel, V ., Tabarrok, A., and Ravi, S. (2021). Too slow for the urban march: Litigations and the real estate market in Mumbai, India. Journal of Urban Economics, 123:103330. Ganong, P . and Shoag, D. (2017). Why has regional income convergence in the us declined? Journal of Urban Economics, 102:76–90. Gaspar, J. and Glaeser, E. L. (1998). Information technology and the future of cities. Journal of Urban Economics, 43(1):136–156. Gechter, M. and Tsivanidis, N. (2017). Efficiency and equity of land policy in developing country cities: Evidence from the Mumbai mills redevelopment. Technical report, Working Paper. Ghani, E., Goswami, A. G., and Kerr, W. R. (2016). Highway to success: The impact of the Golden Quadrilateral project for the location and performance of Indian manufacturing. The Economic Journal, 126(591):317–357. Ghose, D. (2021). Trade, Internal Migration, and Human Capital. Glaeser, E. L. and Gottlieb, J. D. (2009). The wealth of cities: Agglomeration economies and spatial equilibrium in the United States. Journal of Economic Literature, 47(4):983–1028. Glaeser, E. L. and Gyourko, J. (2005). Urban decline and durable housing. Journal of Political Economy, 113(2):345–375. Glaeser, E. L., Gyourko, J., and Saks, R. (2005a). Why is Manhattan so expensive? Regulation and the rise in housing prices. The Journal of Law and Economics, 48(2):331–369. Glaeser, E. L., Gyourko, J., and Saks, R. E. (2005b). Why have housing prices gone up? American Economic Review, 95(2):329–333. Glaeser, E. L., Gyourko, J., and Saks, R. E. (2006). Urban growth and housing supply. Journal of Economic Geography, 6(1):71–89. 145 Glaeser, E. L. and Ward, B. A. (2009). The causes and consequences of land use regulation: Evidence from Greater Boston. Journal of Urban Economics, 65(3):265–278. Goldsmith-Pinkham, P ., Sorkin, I., and Swift, H. (2020). Bartik instruments: What, when, why, and how. American Economic Review, 110(8):2586–2624. Goodman-Bacon, A. (2021). Difference-in-differences with variation in treatment timing. Journal of Econometrics, 225(2):254–277. Gorback, C. S. and Keys, B. J. (2020). Global capital and local assets: House prices, quantities, and elasticities. Technical report, National Bureau of Economic Research. Green, R. K., Malpezzi, S., and Mayo, S. K. (2005). Metropolitan-specific estimates of the price elasticity of supply of housing, and their sources. American Economic Review, 95(2):334–339. Greenwood, M. J. (1975). A simultaneous-equations model of urban growth and migration. Journal of the American Statistical Association, 70(352):797–810. Guedes, R., Iachan, F. S., Sant’Anna, M., et al. (2021). Housing supply in the presence of infor- mality. SocArXiv. March, 23. Gupta, A. (2020). Transferable development rights. Gustafson, P . (2001). Roots and routes: Exploring the relationship between place attachment and mobility. Environment and Behavior, 33(5):667–686. Gustafson, P . (2014). Place attachment in an age of mobility. Place attachment: Advances in theory, methods and applications, pages 37–48. Gyourko, J., Saiz, A., and Summers, A. (2008). A new measure of the local regulatory envi- ronment for housing markets: The Wharton Residential Land Use Regulatory Index. Urban Studies, 45(3):693–729. Habibullah, W. and Ahuja, M. (2005). Land Reforms in India: Computerisation of Land Records, volume 10. SAGE Publications India. Harari, M. (2020). Cities in bad shape: Urban geometry in India. American Economic Review, 110(8):2377–2421. 146 Harris, J. R. and Todaro, M. P . (1970). Migration, unemployment and development: A two-sector analysis. The American Economic Review, pages 126–142. Heckman, J. J. (1979). Sample selection bias as a specification error. Econometrica: Journal of the Econometric Society, pages 153–161. Henderson, J. V ., Regan, T., and Venables, A. J. (2021). Building the city: From slums to a modern metropolis. The Review of Economic Studies, 88(3):1157–1192. Herkenhoff, K. F., Ohanian, L. E., and Prescott, E. C. (2018). Tarnishing the golden and empire states: Land-use restrictions and the US economic slowdown. Journal of Monetary Economics, 93:89–109. Hoffmann, M. and Stewen, I. (2020). Holes in the dike: The global savings glut, US house prices, and the long shadow of banking deregulation. Journal of the European Economic Association, 18(4):2013–2055. Hsieh, C.-T. and Moretti, E. (2019). Housing constraints and spatial misallocation. American Economic Journal: Macroeconomics, 11(2):1–39. Imbert, C. and Papp, J. (2015). Labor market effects of social programs: Evidence from india’s employment guarantee. American Economic Journal: Applied Economics, 7(2):233–63. Imbert, C. and Papp, J. (2020). Short-term migration, rural public works, and urban labor mar- kets: Evidence from india. Journal of the European Economic Association, 18(2):927–963. India Meteorological Department, P . (1901-2017). India Meteorological Department Datasets. http://www.imd.gov.in. Jain, M. (2018). Contemporary urbanization as unregulated growth in India: The story of census towns. Cities, 73:117–127. Jayachandran, S. (2006). Selling labor low: Wage responses to productivity shocks in developing countries. Journal of Political Economy, 114(3):538–575. Joshi, V . and Little, I. D. (1996). India’s economic reforms, 1991-2001. Clarendon Press. 147 Katz, E. and Stark, O. (1986). Labor migration and risk aversion in less developed countries. Journal of Labor Economics, 4(1):134–149. Kimura, Y. (2011). Property Rights Improvement and Credit Reallocation: Theory and Evidence from Urban India. Kimura, Y. (2013). Three essays on property rights reform and industrial development in urban india. PhD thesis, Boston University. Kone, Z. L., Liu, M. Y., Mattoo, A., Ozden, C., and Sharma, S. (2018). Internal borders and migration in India. Journal of Economic Geography, 18(4):729–759. Kulka, A., Sood, A., and Chiumenti, N. (2022). How to Increase Housing Affordability? Under- standing Local Deterrents to Building Multifamily Housing. Understanding Local Deterrents to Building Multifamily Housing (April 12, 2022). Labor Bureau of India, D. (2001-2012). Consumer Price Index Series. http://www.labourbureau.gov.in/. Lall, S. V . and Selod, H. (2006). Rural-urban migration in developing countries: A survey of theoretical predictions and empirical findings , volume 3915. World Bank Publications. Levine, R. (2005). Law, endowments and property rights. Journal of Economic Perspectives,19(3):61– 88. Lu, Y., Xie, H., and Xu, L. C. (2016). Telecommunication externality on migration: Evidence from Chinese villages. China Economic Review, 39:77–90. Lucas, Jr, R. E. (2004). Life earnings and rural-urban migration. Journal of Political Economy, 112(S1):S29–S59. Malpezzi, S. and Maclennan, D. (2001). The long-run price elasticity of supply of new residen- tial construction in the United States and the United Kingdom. Journal of Housing Economics, 10(3):278–306. Marjit, S. (2003). Economic reform and informal wage—a general equilibrium analysis. Journal of Development Economics, 72(1):371–378. 148 Marx, B., Stoker, T., and Suri, T. (2013). The economics of slums in the developing world. Journal of Economic perspectives, 27(4):187–210. Mayer, C. J. and Somerville, C. T. (2000). Land use regulation and new construction. Regional Science and Urban Economics, 30(6):639–662. McLaughlin, R. B. (2012). New housing supply elasticity in Australia: a comparison of dwelling types. The Annals of Regional Science, 48(2):595–618. Mehta, P . B. (2007). India’s unlikely democracy: The rise of judicial sovereignty. Journal of Democracy, 18(2):70–83. Mikusheva, A. and Sun, L. (2022). Inference with many weak instruments. The Review of Economic Studies, 89(5):2663–2686. Mills, E. S. (1967). An aggregative model of resource allocation in a metropolitan area. The American Economic Review, 57(2):197–210. Mitra, A. (2010). Migration, livelihood and well-being: Evidence from indian city slums. Urban Studies, 47(7):1371–1390. Molloy, R., Smith, C. L., and Wozniak, A. (2011). Internal migration in the United States. Journal of Economic Perspectives, 25(3):173–96. Moretti, E. (2012). The new geography of jobs. Houghton Mifflin Harcourt. Morten, M. (2019). Temporary migration and endogenous risk sharing in village India. Journal of Political Economy, 127(1):1–46. Mukhopadhyay, P ., Zérah, M.-H., and Denis, E. (2020). Subaltern urbanization: Indian insights for urban theory. International Journal of Urban and Regional Research, 44(4):582–598. Munshi, K. (2003). Networks in the modern economy: Mexican migrants in the US labor market. The Quarterly Journal of Economics, 118(2):549–599. Munshi, K. and Rosenzweig, M. (2016). Networks and misallocation: Insurance, migration, and the rural-urban wage gap. American Economic Review, 106(1):46–98. 149 Muto, M. (2009). The impacts of mobile phone coverage expansion and personal networks on migration: Evidence from Uganda. Technical report. National Institute of Urban Affairs, I. (2002). Assessment of the impact of the repeal of the Urban Land (Ceiling & Regulation) Act, 1976. New Delhi. National Sample Survey Organization, I. (1999-2012). Indian National Sample Surveys, Rounds 55, 58, 66, and 69. http://mospi.nic.in. National Sample Survey Organization, I. (2008). Indian National Sample Surveys, Round 64. http://mospi.nic.in. Niu, D., Sun, W., and Zheng, S. (2021). The role of informal housing in lowering China’s urban- ization costs. Regional Science and Urban Economics, 91:103638. Paciorek, A. (2013). Supply constraints and housing market dynamics. Journal of Urban Economics, 77:11–26. Parkhomenko, A. (2020). Local causes and aggregate implications of land use regulation. Available at SSRN 3739183. Parliamentary Standing Committee, U. R. D. (1998-1999). THE URBAN LAND (CEILING & REG- ULATION) REPEAL BILL. Department of Urban Development, Ministry of Urban Affairs & Employment. Pethe, A. and Nallathiga, R. (2017). Legislative Impact on Land Markets in Cities: The Case of ULCRA in Mumbai and Its Repeal. Environment and Urbanization ASIA, 8(2):201–213. Pethe, A., Tandel, V ., and Gandhi, S. (2014). The dynamics of urban governance in India. Plantinga, A. J., Lubowski, R. N., and Stavins, R. N. (2002). The effects of potential land develop- ment on agricultural land prices. Journal of Urban Economics, 52(3):561–581. Poterba, J. M. (1984). Tax subsidies to owner-occupied housing: an asset-market approach. The Quarterly Journal of Economics, 99(4):729–752. Quigley, J. M. and Raphael, S. (2005). Regulation and the high cost of housing in California. American Economic Review, 95(2):323–328. 150 Rains, E. and Krishna, A. (2020). Precarious gains: Social mobility and volatility in urban slums. World Development, 132:105001. Rains, E., Krishna, A., and Wibbels, E. (2019). Combining satellite and survey data to study Indian slums: Evidence on the range of conditions and implications for urban policy. Environment and Urbanization, 31(1):267–292. Rao, S. and Finnoff, K. (2015). Marriage migration and inequality in India, 1983–2008. Population and Development Review, 41(3):485–505. Roback, J. (1982). Wages, rents, and the quality of life. Journal of Political Economy,90(6):1257–1278. Roca, J. D. L. and Puga, D. (2017). Learning by working in big cities. The Review of Economic Studies, 84(1):106–142. Rodríguez-Pose, A. and Storper, M. (2020). Housing, urban growth and inequalities: The limits to deregulation and upzoning in reducing economic and spatial inequality. Urban Studies, 57(2):223–248. Rosen, S. (1979). Wage-based indexes of urban quality of life. Current Issues in Urban Economics, pages 74–104. Rosenzweig, M. R. and Stark, O. (1989). Consumption smoothing, migration, and marriage: Evidence from rural india. Journal of Political Economy, 97(4):905–926. Rosenzweig, M. R. and Udry, C. (2014). Rainfall forecasts, weather, and wages over the agricul- tural production cycle. American Economic Review, 104(5):278–83. Saiz, A. (2010). The geographic determinants of housing supply. The Quarterly Journal of Eco- nomics, 125(3):1253–1296. Saks, R. E. and Wozniak, A. (2011). Labor reallocation over the business cycle: New evidence from internal migration. Journal of Labor Economics, 29(4):697–739. Sarsons, H. (2015). Rainfall and conflict: A cautionary tale. Journal of Development Economics, 115:62–72. 151 Schultz, T. P . (1982). Lifetime migration within educational strata in Venezuela: Estimates of a logistic model. Economic Development and Cultural Change, 30(3):559–593. Sekhri, S. and Storeygard, A. (2014). Dowry deaths: Response to weather variability in India. Journal of Development Economics, 111:212–223. Siddiqi, F. J. (2013). Governing urban land: The political economy of the ULCRA in Mumbai. PhD thesis, Massachusetts Institute of Technology. Silva, J. S. and Tenreyro, S. (2006). The log of gravity. The Review of Economics and Statistics, 88(4):641–658. Skop, E. and Adams, P . C. (2009). Creating and inhabiting virtual places: Indian immigrants in cyberspace. National Identities, 11(2):127–147. Sood, A. (2019). Land Market Frictions and Manufacturing in India. Available at SSRN 3468519. Sridhar, K. S. (2010). Impact of land use regulations: Evidence from India’s cities. Urban Studies, 47(7):1541–1569. Sridhar, V . (2007). Growth of mobile services across regions in India. Journal of Scientific and Industrial Research, 66:281–289. Srinivas, L. (1991). Land and Politics in India: Working of Urban Land Ceiling Act,1976. Economic and Political Weekly, pages 2482–2484. Srivastava, R. (2011). Internal migration in india. Human Development in India. Storper, M. and Venables, A. J. (2004). Buzz: Face-to-face contact and the urban economy. Journal of Economic Geography, 4(4):351–370. Thapar, R. (2015). The Penguin history of early India: from the origins to AD 1300. Penguin UK. Tiwari, P . and Parikh, J. (1998). Affordability, housing demand and housing policy in urban India. Urban Studies, 35(11):2111–2129. Tiwari, P ., Parikh, K., and Parikh, J. (1999). Effective housing demand in Mumbai (Bombay) metropolitan region. Urban Studies, 36(10):1783–1809. 152 Todaro, M. P . (1969). A model of labor migration and urban unemployment in less developed countries. The American Economic Review, 59(1):138–148. Topel, R. H. (1986). Local labor markets. Journal of Political Economy, 94(3, Part 2):S111–S143. Tumbe, C. (2018). India moving: A history of migration. Penguin Random House India Private Limited. Varshney, A. (1998). India defies the odds: Why democracy survives. Journal of Democracy, 9(3):36–50. Wadhwa, D. (2002). Guaranteeing title to land. Economic and Political Weekly, pages 4699–4722. Wang, S., Chan, S. H., and Xu, B. (2012). The estimation and determinants of the price elasticity of housing supply: Evidence from China. Journal of Real Estate Research, 34(3):311–344. Winston, C. (1998). Us industry adjustment to economic deregulation. Journal of Economic Per- spectives, 12(3):89–110. Winston, C. and Karpilow, Q. (2016). Should the US eliminate entry barriers to the practice of law? Perspectives shaped by industry deregulation. American Economic Review, 106(5):171–76. World Trade Organization, W. (2016). Commercial services exports by sector – annual (1980-2013. World Trade Organization. Young, A. (2013). Inequality, the urban-rural gap, and migration. The Quarterly Journal of Eco- nomics, 128(4):1727–1785. Zabel, J. E. (2012). Migration, housing market, and labor market responses to employment shocks. Journal of Urban Economics, 72(2-3):267–284. 153 Appendix A Supplementary Figures and tables Figure A.1: Telecom circles in India (1996-2005) No data Jammu & Kashmir Himachal Pradesh Punjab Haryana Delhi Rajasthan Uttar Pradesh (W) Uttar Pradesh (E) Bihar North East Assam West Bengal Kolkata Orissa Madhya Pradesh Gujarat Maharashtra Mumbai Andhra Pradesh Karnataka Kerala Tamil Nadu Chennai Metro Category A Category B Category C No data Data source: Sridhar (2007). Note: Figure presents a map of India with the 23 telecom circles demarcated and labeled with their respective names. Each circle is categorized into one of four groups depending on the circle’s revenue-generating potential. In decreasing order of revenue-generating potential, these are called the metro, A, B, and C circles. The Chennai metro circle was merged with the Tamil Nadu circle at the end of 2005. 154 Figure A.2: Map of Golden Quadrilateral recipient states in India Jammu & Kashmir Himachal Pradesh Punjab Chandigarh Uttarakhand Haryana Delhi Rajasthan Uttar Pradesh Bihar Sikkim Arunachal Pradesh Nagaland Manipur Mizoram Tripura Meghalaya Assam West Bengal Jharkhand Odisha Chattisgarh Madhya Pradesh Gujarat Daman & Diu Dadra & Nagar Haveli Maharashtra Andhra Pradesh Karnataka Goa Lakshadweep Kerala Tamil Nadu Pondicherry Andaman & Nicobar Islands GQ non−recipient GQ recipient Data Source: Ghani et al. (2016). Note: Figure presents a map of India with the35 states and union territories demarcated. Light-colored states were not recipients of the National Highways Development Project Phase I or the Golden Quadrilateral (GQ) highway upgrade project. Dark-colored states were part of the GQ program. 155 Figure A.3: Map of India with geocoded cities Late repeal Early repeal No ULC Note: Figure presents a map of India with the 202 sample cities geocoded with markers. Each marker represents a city. Hollow circles represent late repeal cities that repealed ULC during 2002-2008. Grey-colored triangles represent early repeal cities that repealed ULC during 1999-2000. Black squares represent cities that never enacted ULC. The size of a marker is proportional to the population of the city it represents. Population figures are normalized to the range [0,1]. Geocodes for each city have been obtained using Google Maps API. 156 Figure A.4: Impact of false telecom shock on local telecom operators in counterfactual districts 0 2 4 Kernel density −.4 −.2 0 .2 .4 Coefficient (a) Density of coefficient estimates 0 .5 1 1.5 Kernel density 0 .2 .4 .6 .8 1 p−values (b) Density of p-values Source: Authors’ calculations. Note: Figures present the density of coefficient estimates and their p-values from two-way fixed effects regressions with a panel of 112 counterfactual districts observed in the pre- and post-2000 periods. The number of local telecom operators in a district is regressed on a false auction bid dummy, a vector of controls, district fixed effects, and the time trend. Panel (a) presents the density of coefficient estimates, and panel (b) presents the density of p-values. The kernel density estimates are based on the Gaussian kernel smoothing function. The half-width of the kernel, chosen from minimizing the mean integrated squared error around each point, is slightly wider. This explains the negative p-values in panel (b). The vertical line in panel (a) represents a coefficient estimate value of zero. The vertical line in panel (b) represents a p-value of 0.1. 157 Figure A.5: Impact of false telecom shock on rural-urban migration in counterfactual districts 0 4 8 12 Kernel density −.1 −.05 0 .05 .1 Coefficient (a) Density of coefficient estimates 0 .5 1 1.5 Kernel density 0 .2 .4 .6 .8 1 p−values (b) Density of p-values Source: Authors’ calculations. Note: Figures present the density of coefficient estimates and their p-values from two-way fixed effects regressions with a panel of 112 counterfactual districts observed in the pre- and post-2000 periods. The log of annualized inter- district rural-urban migrant stock in a district is regressed on a false auction bid dummy, a vector of controls, district fixed effects, and the time trend. Panel (a) presents the density of coefficient estimates, and panel (b) presents the density of p-values. The kernel density estimates are based on the Gaussian kernel smoothing function. The half- width of the kernel, chosen from minimizing the mean integrated squared error around each point, is slightly wider. This explains the negative p-values in panel (b). The vertical line in panel (a) represents a coefficient estimate value of zero. The vertical line in panel (b) represents a p-value of 0.1. 158 Figure A.6: Density of p-values from false treatment I falsification tests 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False early repeal I 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False late repeal I Panel (b): Log formal residential 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False early repeal I 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False late repeal I Panel (a): Log informal residential 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False early repeal I 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False late repeal I Panel (c): Log non−residential 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False early repeal I 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False late repeal I Panel (e): Share new formal 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False early repeal I 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False late repeal I Panel (d): Share new informal Source: Author’s calculations. Note: Figure presents Gaussian kernel density plots of p-values for the coefficients of false early repeal I and late repeal I dummies’ interacted with the time trend. The vertical bars indicate a p-value of 0.1. Each panel represents the density of p-values based on a set of 1,000 regressions estimating equation (3.1). The dependent variables in panels (a) through (e) are, respectively, log of formal and informal residential housing units, log of non-residential stock, and the shares of formal and informal households living in new units. For each iteration of the set of 1,000 regressions, the false early and late repeal I dummies are randomly drawn in two steps. First, in the full counterfactual sample of 157 cities, a fake ULC treatment is generated with a pseudo random variable following a Bernoulli distribution with success probability 0.22. Next, in the sample of observations where the generated fake ULC treatment is equal to one, False early repeal I is generated as another pseudo random number following a Bernoulli distribution with success probability 0.73. False late repeal I is equal to one for the observations where the generated fake ULC treatment is equal to one, and False early repeal I is equal to zero. All other values of False late repeal I and False early repeal I are zero. Other variables in the regressions include log population, an employment shift-share index, log surface area, population’s shares of scheduled castes and tribes individuals, females, and literate persons, mean temperature, and log annual aggregate rainfall. Standard errors are clustered by the states of administration. 159 Figure A.7: Density of p-values from false treatment II falsification tests 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False early repeal II 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False late repeal II Panel (b): Log formal residential 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False early repeal II 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False late repeal II Panel (a): Log informal residential 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False early repeal II 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False late repeal II Panel (c): Log non−residential 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False early repeal II 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False late repeal II Panel (e): Share new formal 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False early repeal II 0 .5 1 Density 0 .2 .4 .6 .8 1 p−values Year = 2011 x False late repeal II Panel (d): Share new informal Source: Author’s calculations. Note: Figure presents Gaussian kernel density plots of p-values for the coefficients of false early repeal II and late repeal II dummies’ interacted with the time trend. The vertical bars indicate a p-value of 0.1. Each panel represents the density of p-values based on a set of1,000 regressions estimating equation (3.1). The dependent variables in panels (a) through (e) are, respectively, log of formal and informal residential housing units, log of non-residential stock, and the shares of formal and informal households living in new units. For each iteration of the set of 1,000 regressions, the false early and late repeal II dummies are randomly drawn in two steps. First, in the full counterfactual sample, a fake ULC treatment is generated with a pseudo random variable following a Bernoulli distribution with success probability 0.22. Next, in the sample of observations where the generated fake ULC treatment is equal to one, False late repeal II is generated as another pseudo random number following a Bernoulli distribution with success probability 0.27. False early repeal II is equal to one for the observations where the generated fake ULC treatment is equal to one, and False late repeal II is equal to zero. All other values of False late repeal II and False early repeal II are zero. Other variables in the regressions include log population, an employment shift-share index, log surface area, population’s shares of scheduled castes and tribes individuals, females, and literate persons, mean temperature, and log annual aggregate rainfall. Standard errors are clustered by the states of administration. 160 Table A.1: Constituents and telecom operators by circles Telecom operators Pre-2000 Post-2000 Circle name Districts and states covered (1) (2) Andhra Pradesh Andhra Pradesh & Yanam 0.8 5.0 Assam Assam 0.3 1.8 Bihar Bihar & Jharkhand 0.5 3.0 Chennai Chennai, Kancheepuram & Thiruvallur 1.0 4.2 Delhi Delhi, Faridabad, Ghaziabad & Gurgaon 1.0 5.2 Gujarat DD, DNH & Gujarat 0.8 5.2 Haryana Haryana ∗ 0.8 4.4 Himachal Pradesh Himachal Pradesh 0.7 3.6 Jammu & Kashmir Jammu & Kashmir 0.0 0.8 Karnataka Karnataka 0.8 5.0 Kerala Mahe, Lakshadweep & Kerala 0.8 4.4 Kolkata Haora, Hugli, Kolkata, Nadia, & 24-Parganas 1.0 3.8 Madhya Pradesh Chhattisgarh & Madhya Pradesh 0.7 4.4 Maharashtra Maharashtra ⋆ 0.8 4.8 Mumbai Mumbai, Mumbai Suburban & Thane 1.0 4.6 North East AP , Manipur, Meghalaya, MZ, NL & Tripura 0.3 2.0 Orissa Orissa 0.4 3.4 Punjab Chandigarh & Punjab 0.4 4.8 Rajasthan Rajasthan 0.8 4.6 Tamil Nadu Karaikal, Pondicherry & TN † 1.0 5.0 Uttar Pradesh (E) 33 districts of eastern UP 0.8 3.2 Uttar Pradesh (W) 19 districts of western UP & Uttarakhand 0.8 3.8 West Bengal A & N Islands, Sikkim & WB ‡ 0.3 3.4 Mean All of India 0.7 3.9 Data source: Sridhar (2007). Note: Tables reports the constituent districts and states and the annual average number of telecom operators in each of the 23 telecom circles in India during the pre-2000 and the post-2000 periods. The pre-2000 period is between 1991-2000, and the post-2000 period is 2001-2005. All circles had zero operators before 1996. Telecom operator values are rounded off to one decimal point. Ghaziabad in Delhi includes the districts of Ghaziabad, Gautam Buddha Nagar, and Bulandshahr. Faridabad includes Faridabad and Palwal districts, and Gurgaon includes Gurgaon and Mewat districts. A & N Islands stand for the Andaman & Nicobar Islands. The abbreviations AP , DD, DNH, MZ, NL, TN, UP , and WB, respectively, stand for Arunachal Pradesh, Daman & Diu, Dadra & Nagar Haveli, Mizoram, Nagaland, Tamil Nadu, Uttar Pradesh, and West Bengal. 24-Parganas include North and South 24-Parganas. ∗ Excludes Gurgaon and Faridabad. ⋆ Excludes the Mumbai circle. † Excludes the Tamil Nadu circle. ‡ Excludes the Kolkata circle. 161 Table A.2: Summary statistics of aggregate variables by treatment and control districts Panel (a): Districts with an auction bid (treatment) Pre-2000 Post-2000 Mean SD Mean SD Variable (1) (2) (3) (4) Local telecom operators 0.78 0.16 4.37 0.78 Annualized inter-district rural-urban migrants ('000) 3.15 9.64 4.34 12.84 Urban population (millions) 0.82 1.29 1.07 1.62 Share literate urban 0.68 0.08 0.73 0.07 Share employed urban 0.31 0.04 0.34 0.04 Labor demand shock 9.41 0.23 10.35 0.24 Mean monthly per capita consumption (INR) 971 255 1,050 350 Mean weekly wages (INR) 181 98 189 96 GQ 0 0 0.20 0.40 N 307 307 307 307 Panel (b): Districts without an auction bid (control) Pre-2000 Post-2000 Mean SD Mean SD Variable (1) (2) (3) (4) Local telecom operators 0.39 0.09 2.79 0.62 Annualized inter-district rural-urban migrants ('000) 1.07 1.70 1.29 1.98 Urban population (millions) 0.29 0.39 0.39 0.50 Share literate urban 0.68 0.08 0.73 0.07 Share employed urban 0.30 0.05 0.33 0.04 Labor demand shock 9.35 0.21 10.32 0.24 Mean monthly per capita consumption (INR) 928 275 921 282 Mean weekly wages (INR) 195 150 178 105 GQ 0 0 0.17 0.38 N 112 112 112 112 Source: Authors’ calculations based on Census of India, Labor Bureau of India, National Sample Survey Organization, and Sridhar (2007). Note: Table presents summary statistics of district-level aggregate variables for a panel of 419 districts. Panel (a) and (b) presents the summary statistics for the 307 and 112 treatment and control districts, respectively. Treated districts received a bid in the 2001 auction of 1800 MHz spectrum band, and control districts did not. The pre- 2000 period is between 1991-2000, and the post-2000 period is 2001-2005. Migration figures are given in thousands. Population figures are given in millions. Literate and employed shares are the shares of urban residents in a district who are literate and employed, respectively. Consumption and wage earning figures based on urban households. Labor demand shock is calculated by first multiplying the 2001 share of employment in four sectors–—agriculture, manufacturing, mining and construction, and services—–with the log volume of the corresponding sector’s exports (in USD) in 2001 (for pre-2000) and 2005 (post-2000), and then adding the products over the four sectors for each period. GQ is a dummy equal to one if the district was a recipient of the Golden Quadrilateral highway upgrade project after 2001, and zero otherwise. Consumption and wage values rounded off to the nearest integers. All other values are rounded off to two decimal places. Real mean monthly per capita consumption and mean weekly wages in 2001 Indian National Rupees (INR) are calculated using the Consumer Price Index (CPI) data from the Labor Bureau of India. In PPP terms, $1 = 10 INR in 2001. For exchange rates, see: https://data.oecd.org/conversion/purchasing- power-parities-ppp.htm. 162 Table A.3: Summary statistics of IHDS variables Mean SD Variable (1) (2) HH sent out-migrant dummy variables: Any out-migrant 0.10 0.30 Out-migrated to urban areas only 0.08 0.26 Out-migrated to rural areas only 0.03 0.16 HH owns mobile phone 0.03 0.18 Log HH income 10.14 1.01 HH owns land 0.59 0.49 Non-agricultural HH 0.44 0.50 HH size 5.37 2.63 Highest educated school graduate 0.17 0.38 N 26,472 26,472 Source: Authors’ calculations based on Desai et al.. Note: Table presents summary statistics of household-level variables from the first wave of the India Human De- velopment Survey (IHDS I) conducted during November 2004-October 2005. The sample consists of 26,472 rural households surveyed by IHDS. HH stands for household. Any out-migrant is a dummy equal to one if the household sent a migrant to either rural or urban areas, and zero otherwise. Out-migrated to urban areas only is a dummy equal to one if the household sent a migrant to urban areas only, and zero otherwise. Out-migrated to rural areas only is a dummy equal to one if the household sent a migrant to rural areas only, and zero otherwise. HH owns mobile phone is a dummy equal to one if the household owns a mobile phone, and zero otherwise. HH owns land is a dummy equal to one if the household owns land, and zero otherwise. Non-agricultural HH is a dummy equal to one if the household’s principal occupation is in the non-agricultural sector, and zero otherwise. Highest educated school graduate is a dummy equal to one if the highest educated household member graduated high school. All values rounded off to two decimal places. 163 Table A.4: NSS microdata subsamples and their experienced telecom shock Sample Migration Telecom operators Type Size Reason Time Mean SD Type of individual (1) (2) (3) (4) (5) (6) Rural-urban migrant Full 4,160 Any 2001-2005 3.96 0.83 18-52 yr. 1,050 Labor 2001-2005 3.94 0.84 18-52 yr. 2,712 Labor Before 2001 - - Rural resident non-migrant Full 232,425 - - 3.65 0.98 18-52 yr. 96,116 - - 3.62 1.01 Urban-rural migrant Full 848 Any 2001-2005 4.24 0.89 18-52 yr. 159 Labor 2001-2005 4.34 0.93 Urban resident non-migrant Full 126,842 - - 3.90 1.16 18-52 yr. 59,247 - - 3.89 1.17 Out-migrant left rural areas Full 15,456 Any 2001-2005 3.84 0.90 Full 54,281 Any Before 2001 - - 18-52 yr. 6,588 Labor 2001-2005 3.72 0.96 Data sources: National Sample Survey Organization and Sridhar (2007). Note: Table reports the number of individuals and their experienced telecom shock for various subsamples of the microdata from the NSS survey on employment and migration conducted during July 2007-June 2008. For rural- urban migrants, rural resident non-migrants, and out-migrants who left rural areas, the telecom shock is the number of distant telecom operators. For urban-rural migrants and urban resident non-migrants, the shock is the number of local telecom operators. The time of migration and the reason stated for migration are given for migrants. The 18-52 yr. sample type reflects the sample of individuals who were between the ages of 18 and52 years in2000. Out-migrants who left rural areas went to either rural or urban areas. Their destinations are unknown. Distant telecom operators is the average number of migration share-weighted telecom operators corresponding to an individual’s region of residence in 2000. It is calculated by first taking the product of the share of in-migrants who moved from another region to the individual’s region-of-residence before 1991 and the number of telecom operators in the other region and then adding the products for all regions outside the region-of-residence of the individual. 164 Table A.5: Mobile phone ownership and migration among rural households Dependent variable = HH sent migrant to rural or urban areas urban areas only rural areas only (1) (2) (3) HH owns mobile phone 0.027** 0.025** 0.003 (0.012) (0.011) (0.005) Log HH income 0.019*** 0.017*** 0.001 (0.003) (0.002) (0.001) HH owns land 0.045*** 0.039*** 0.006** (0.005) (0.004) (0.003) Non-agricultural HH 0.018*** 0.018*** -0.000 (0.006) (0.005) (0.002) HH size -0.007*** -0.006*** -0.001*** (0.001) (0.001) (0.000) Highest educated school -0.004 -0.002 -0.003 graduate (0.005) (0.005) (0.003) District fixed effects ✓ ✓ ✓ N 26,472 26,472 26,472 R 2 0.009 0.010 0.000 Source: Authors’ calculations. Note: Table presents estimates from linear probability model regressions at the household level using data from the first wave of the India Human Development Survey (IHDS I) conducted during November 2004-October 2005. The samples used in all regressions include 26,472 rural households. Dependent variables in all columns are dummy variables indicating that a household sent a migrant. Column (1)’s migrant dummy equals one if the household sent a migrant to either rural or urban areas, and zero otherwise. Column (2)’s migrant dummy equals one if the household sent a migrant to urban areas only, and zero otherwise. Column (3)’s migrant dummy equals one if the household sent a migrant to rural areas only, and zero otherwise. Standard errors are clustered at the district level and reported in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. 165 Table A.6: Golden Quadrilateral and urban commodity prices Dependent variable =∆ Log price ∆= 2011− 2001 (1) GQ highway dummy 0.041 (0.040) ∆ Log weekly wage -0.015 (0.114) ∆ Log urban population -0.269*** (0.062) ∆ Urban surface area -0.025 (0.027) ∆ Urban surface area squared -0.000 (0.002) N 33 R 2 0.437 Source: Authors’ calculations. Note: Table presents first-difference regression estimates using a sample of 33 states in India between 2001 and 2011. Differences are calculated as first differences of values between the years 2001 and 2011. Commodity prices for a given state are calculated by first obtaining the prices of non-housing commodity baskets in the state’s urban areas, then multiplying the computed prices with the national average of each basket consumed by urban households between 2001 and 2011, and finally adding the products for all baskets. The GQ highway dummy equals one if the state was a recipient of the GQ program, and zero otherwise. Urban surface area unit in 1000 sq. miles. Robust standard errors presented in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. 166 Table A.7: Local urban population growth and housing demand without median rooms First-stage Second-stage Dependent variable =∆ Log indicators Urban Housing units at i population at i Informal Formal Vacant ∆= 2011− 2001 (1) (2) (3) (4) ∆ Log urban population at i 0.205*** 1.86*** 2.41*** (0.025) (0.026) (0.034) ∆ Rainfall deficit months at j 0.011*** (0.000) GQ highway at j dummy 0.124*** (0.005) ∆ Log consumption at i 0.009 -0.197*** 0.037** -0.203*** (0.013) (0.014) (0.018) (0.020) ∆ Urban surface area at i 5.12*** 0.824*** -1.54*** -2.82*** (0.202) (0.220) (0.226) (0.316) ∆ Urban surface area at i squared -5.53*** -1.84*** 1.24*** 1.03** (0.418) (0.348) (0.310) (0.463) F-stat on excluded instruments 1018*** Anderson–Rubin Wald χ 2 (2) 54.8*** 3402*** 2495*** Sargan-Hansen J-stat p-value 0.708 0.420 0.968 N 4,896 4,896 4,896 4,896 R 2 0.603 Source: Authors’ calculations. Note: Table presents two-stage least squares first-difference regression estimates using a sample of district-state pairs in India between 2001 and 2011. District i is the local region, and state j the distant. Excluding missing pairs, there are 4,896 i− j district-state pairs consisting of the 35 states and union territories and 144 districts of India in our sample. Differences are calculated as first differences of values between the years 2001 and 2011. Log urban population at i is instrumented by the decadal change in the number of rainfall deficit months in state j and the GQ dummy, which equals one if state j was recipient of the Golden Quadrilateral highway upgrade program, and zero otherwise. Consumption, urban surface area, and urban surface area squared at i are additional covariates. Urban surface area unit in 1000 sq. miles. Robust standard errors presented in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. 167 Table A.8: Formal and vacant housing supply elasticity estimation with full sample First-stage Second-stage Dependent variable =∆ Log indicators Housing units at i Rents at i Formal Vacant Formal Vacant ∆= 2011− 2001 (1) (2) (3) (4) ∆ Log formal units at i 0.738*** (0.029) ∆ Log vacant units at i 0.561*** (0.022) ∆ Rainfall deficit months at j 0.020*** 0.025*** (0.000) (0.001) GQ highway at j dummy 0.214*** 0.284*** (0.004) (0.006) ∆ Log consumption at i 0.044*** 0.006 0.175*** 0.128*** (0.010) (0.013) (0.024) (0.023) ∆ Urban surface area at i 9.42*** 11.4*** -6.88*** -5.65*** (0.217) (0.277) (0.485) (0.443) ∆ Urban surface area at i squared -10.9*** -15.0*** 8.02*** 7.95*** (0.500) (0.622) (0.767) (0.736) F-stat on excluded instruments 3696*** 3271*** Anderson–Rubin Wald χ 2 (2) 677*** 712*** Sargan-Hansen J-stat p-value 0.799 0.657 N 11,526 11,526 11,526 11,526 R 2 0.673 0.611 Source: Authors’ calculations. Note: Table presents two-stage least squares first-difference regression estimates using a sample of district-state pairs in India between 2001 and 2011. District i is the local region, and state j the distant. Excluding missing pairs, there are 11,526 i− j district-state pairs consisting of the 35 states and union territories and 339 districts of India in our complete sample on formal rents. Differences are calculated as first differences of values between the years 2001 and 2011. Log housing units at i are instrumented by the decadal change in the number of rainfall deficit months in state j and the GQ dummy, which equals one if state j was recipient of the Golden Quadrilateral highway upgrade program, and zero otherwise. Consumption, urban surface area, and urban surface area squared at i are additional covariates. Urban surface area unit in 1000 sq. miles. Robust standard errors presented in parentheses. * p < 0.10, ** p < 0.05, *** p< 0.01. 168 Table A.9: Housing supply elasticity estimation with non-contiguous state shocks First-stage Second-stage Dependent variable =∆ Log indicators Housing units at i Rents at i Informal Formal Vacant Informal Formal Vacant ∆= 2011− 2001 (1) (2) (3) (4) (5) (6) ∆ Log informal units at i -2.03*** (0.609) ∆ Log formal units at i 0.601*** (0.039) ∆ Log vacant units at i 0.377*** (0.027) ∆ Rainfall deficit months at j ′ 0.001*** 0.022*** 0.027*** (0.001) (0.001) (0.001) GQ highway at j ′ dummy 0.035*** 0.219*** 0.309*** (0.008) (0.008) (0.012) ∆ Log consumption at i -0.192*** 0.055*** -0.196*** -0.474*** 0.439*** 0.399*** (0.016) (0.019) (0.029) (0.135) (0.045) (0.038) ∆ Urban surface area at i 1.94*** 7.98*** 9.60*** 6.67*** -2.14*** -0.091 (0.181) (0.347) (0.429) (1.532) (0.532) (0.450) ∆ Urban surface area at i -3.09*** -9.17*** -12.5*** -5.91** 2.82*** 1.67** squared (0.397) (0.720) (0.870) (2.327) (0.807) (0.742) F-stat on excluded instruments 21.7*** 1491*** 1072*** Anderson–Rubin Wald χ 2 (2) 14.9*** 254*** 212*** Sargan-Hansen J-stat p-value 0.514 0.400 0.022 N 4,224 4,224 4,224 4,224 4,224 4,224 R 2 0.072 0.661 0.554 Source: Authors’ calculations. Note: Table presents two-stage least squares first-difference regression estimates using a sample of non-contiguous district-state pairs in India between 2001 and 2011. District i is the local region and a state j ′ that is non-contiguous with district i’s state the distant. Excluding missing pairs, there are 4,224 i− j ′ non-contiguous district-state pairs consisting of the 35 states and union territories and 144 districts of India in our sample. Differences are calculated as first differences of values between the years 2001 and 2011. Log housing units at i are instrumented by the decadal change in the number of rainfall deficit months in state j ′ and the GQ dummy, which equals one if state j ′ was recipient of the Golden Quadrilateral highway upgrade program, and zero otherwise. Consumption, urban surface area, and urban surface area squared at i are additional covariates. Urban surface area unit in 1000 sq. miles. Robust standard errors presented in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. 169 Table A.10: Summary statistics of city-level variables for early repeal cities 2001 2011 Data Mean SD Mean SD source Variable (1) (2) (3) (4) (5) Population ('000) 1,542 1,493 2,011 2,070 Census Formal units ('000) 236 258 387 472 Census Informal units ('000) 40 37 39 33 Census Non-residential stock ('000) 88 97 133 164 Census Surface area (sq. miles) 89 69 111 95 Census Mean temperature (°F) 78 2 79 2 IMD Annual aggregate rainfall (inches) 40 25 38 22 IMD Share formal HH in new units 0.23 0.12 0.22 0.13 NSS Share informal HH in new units 0.16 0.21 0.17 0.30 NSS Share SC/ST 0.13 0.04 0.14 0.05 Census Share females 0.47 0.02 0.48 0.02 Census Share literate 0.70 0.07 0.75 0.06 Census Employment shift-share index 0.32 0.05 0.83 0.08 Census N 33 33 33 33 Source: Author’s calculations based on Census of India, India Meteorological Department, and National Sample Survey Organization. Note: Table presents summary statistics of city-level variables for 33 early repeal cities observed in 2001 and 2011. HH stands for household. Shares of formal and informal housing residents living in new units calculated from NSS survey data. New units are defined as those constructed in the previous decade. Shares of SC/ST, females, and literate calculated as shares of population. SC/ST stands for scheduled castes and tribes individuals. The employment shift- share index is a standard Bartik shifter calculated based on employment shares of 10 sectors in 1991. Figures for population and formal, informal, and non-residential stock are rounded off to thousands. Surface area, temperature, and rainfall are rounded off to integer values in square miles, Fahrenheit degrees, and inches, respectively. All share variables, including the shift-share index, are rounded off to two decimal points. 170 Table A.11: Summary statistics of city-level variables for late repeal cities 2001 2011 Data Mean SD Mean SD source Variable (1) (2) (3) (4) (5) Population ('000) 2,979 4,514 3,616 5,102 Census Formal units ('000) 510 861 741 1,064 Census Informal units ('000) 75 63 62 47 Census Non-residential stock ('000) 207 394 329 602 Census Surface area (sq. miles) 129 130 160 152 Census Mean temperature (°F) 80 3 80 3 IMD Annual aggregate rainfall (inches) 54 26 55 23 IMD Share formal HH in new units 0.28 0.12 0.25 0.11 NSS Share informal HH in new units 0.21 0.17 0.04 0.09 NSS Share SC/ST 0.13 0.05 0.14 0.06 Census Share females 0.48 0.02 0.49 0.01 Census Share literate 0.73 0.04 0.77 0.04 Census Employment shift-share index 0.32 0.06 0.89 0.09 Census N 12 12 12 12 Source: Author’s calculations based on Census of India, India Meteorological Department, and National Sample Survey Organization. Note: Table presents summary statistics of city-level variables for 12 late repeal cities observed in 2001 and 2011. HH stands for household. Shares of formal and informal housing residents living in new units calculated from NSS survey data. New units are defined as those constructed in the previous decade. Shares of SC/ST, females, and literate calculated as shares of population. SC/ST stands for scheduled castes and tribes individuals. The employment shift- share index is a standard Bartik shifter calculated based on employment shares of 10 sectors in 1991. Figures for population and formal, informal, and non-residential stock are rounded off to thousands. Surface area, temperature, and rainfall are rounded off to integer values in square miles, Fahrenheit degrees, and inches, respectively. All share variables, including the shift-share index, are rounded off to two decimal points. 171 Table A.12: Summary statistics of city-level variables for no-ULC cities 2001 2011 Data Mean SD Mean SD source Variable (1) (2) (3) (4) (5) Population ('000) 250 205 335 364 Census Formal units ('000) 35 35 60 74 Census Informal units ('000) 10 9 10 10 Census Non-residential stock ('000) 14 12 21 24 Census Surface area (sq. miles) 20 24 29 53 Census Mean temperature (°F) 79 2 79 2 IMD Annual aggregate rainfall (inches) 42 24 41 23 IMD Share formal HH in new units 0.23 0.15 0.18 0.12 NSS Share informal HH in new units 0.19 0.23 0.13 0.24 NSS Share SC/ST 0.13 0.06 0.15 0.07 Census Share females 0.48 0.02 0.49 0.02 Census Share literate 0.69 0.08 0.74 0.07 Census Employment shift-share index 0.28 0.05 0.79 0.07 Census N 157 157 157 157 Source: Author’s calculations based on Census of India, India Meteorological Department, and National Sample Survey Organization. Note: Table presents summary statistics of city-level variables for 157 no-ULC cities observed in 2001 and 2011. HH stands for household. Shares of formal and informal housing residents living in new units calculated from NSS survey data. New units are defined as those constructed in the previous decade. Shares of SC/ST, females, and literate calculated as shares of population. SC/ST stands for scheduled castes and tribes individuals. The employment shift- share index is a standard Bartik shifter calculated based on employment shares of 10 sectors in 1991. Figures for population and formal, informal, and non-residential stock are rounded off to thousands. Surface area, temperature, and rainfall are rounded off to integer values in square miles, Fahrenheit degrees, and inches, respectively. All share variables, including the shift-share index, are rounded off to two decimal points. 172 Table A.13: ULC repeal impact on city-level supply excluding Tamil Nadu Dependent variable = Housing and non-residential indicators Log of stock Share HH in new units Formal residential Informal residential Non- residential Formal residential Informal residential (1) (2) (3) (4) (5) Early repeal× Post -0.034 -0.013 0.011 0.013 0.062 (0.025) (0.067) (0.044) (0.025) (0.072) Late repeal× Post 0.020 -0.128** 0.132** 0.056 -0.209*** (0.040) (0.058) (0.049) (0.037) (0.054) Log population 0.898*** 0.688*** 0.514* 0.186 0.342 (0.104) (0.115) (0.271) (0.166) (0.207) Shift-share index 0.006 -0.319 0.056 -0.392 1.722** (0.150) (0.423) (0.331) (0.365) (0.706) Log surface area 0.034 0.030 -0.016 -0.004 -0.050 (0.035) (0.064) (0.040) (0.045) (0.086) Share SC/ST -0.783 -0.470 -0.288 -1.681 1.138 (0.689) (1.387) (1.183) (1.092) (2.051) Share females -0.908 -0.890 -0.591 -0.967 2.266 (1.800) (5.132) (2.939) (3.026) (7.632) Share literate 0.647 -1.386 0.012 -0.181 4.533 (0.561) (1.624) (1.047) (0.876) (3.000) Mean temperature -0.096*** 0.193** 0.005 0.021 0.027 (0.014) (0.081) (0.072) (0.040) (0.120) Log rainfall 0.211** -0.257* 0.300 0.259* 0.365** (0.088) (0.139) (0.210) (0.120) (0.163) City fixed effects ✓ ✓ ✓ ✓ ✓ Time trend ✓ ✓ ✓ ✓ ✓ Population-weighted ✓ ✓ Housing data source Census Census Census NSS NSS N 366 366 366 366 366 R 2 0.949 0.182 0.817 0.131 0.170 Source: Author’s calculations. Note: Table presents two-way fixed effects regression estimates using a balanced panel of 183 cities, observed in 2001 and 2011. The sample excludes 19 cities in Tamil Nadu that are included in the full sample. Dependent variables in columns (1)-(5) are the logs of formal housing, informal housing, and non-residential stock and the shares of formal and informal housing residents living in new units. New units are defined as those constructed in the previous decade. Early repeal is a dummy equal to one if the city repealed ULC during 1999-2000, and zero otherwise. Late repeal is a dummy equal to one if the city repealed ULC during 2002-2008, and zero otherwise. Standard errors are clustered by the states of administration and reported in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. 173 Table A.14: ULC repeal impact on city-level supply with decomposed late repeal treatment Dependent variable = Housing and non-residential indicators Log of stock Share HH in new units Formal residential Informal residential Non- residential Formal residential Informal residential (1) (2) (3) (4) (5) Early repeal× Post -0.028 -0.008 0.009 0.041 0.007 (0.020) (0.061) (0.040) (0.033) (0.088) Late repeal I× Post 0.016 0.097 0.183 -0.005 -0.472** (0.093) (0.199) (0.126) (0.101) (0.197) Late repeal II× Post 0.019 -0.170** 0.124** 0.057 -0.157** (0.049) (0.075) (0.050) (0.034) (0.056) Log population 0.894*** 0.661*** 0.481* 0.194 0.274 (0.091) (0.108) (0.267) (0.184) (0.197) Shift-share index -0.050 -0.374 -0.060 -0.092 1.190 (0.125) (0.348) (0.287) (0.429) (0.994) Log surface area 0.021 0.038 -0.016 -0.015 -0.016 (0.035) (0.059) (0.042) (0.053) (0.081) Share SC/ST -0.656 -0.302 -0.305 -0.990 -1.940 (0.702) (1.269) (1.166) (1.200) (2.453) Share females -0.799 -2.628 -0.457 -0.725 3.122 (1.732) (5.279) (2.711) (3.522) (7.716) Share literate 0.246 -1.291 -0.161 -0.183 4.126 (0.662) (1.566) (1.001) (0.758) (2.667) Mean temperature -0.121*** 0.197** -0.004 0.000 0.122 (0.025) (0.073) (0.062) (0.041) (0.126) Log rainfall 0.103 -0.125 0.237 0.253* 0.196 (0.111) (0.159) (0.164) (0.129) (0.170) City fixed effects ✓ ✓ ✓ ✓ ✓ Time trend ✓ ✓ ✓ ✓ ✓ Population-weighted ✓ ✓ Housing data source Census Census Census NSS NSS N 404 404 404 404 404 R 2 0.949 0.182 0.811 0.107 0.146 Source: Author’s calculations. Note: Table presents two-way fixed effects regression estimates using a balanced panel of 202 cities observed in 2001 and 2011. Dependent variables in columns (1)-(5) are the logs of formal housing, informal housing, and non- residential stock and the shares of formal and informal housing residents living in new units. New units are defined as those constructed in the previous decade. Early repeal is a dummy equal to one if the city repealed ULC during 1999-2000, and zero otherwise. Late repeal I is a dummy equal to one if the city repealed ULC during 2002-2006, and zero otherwise. Late repeal II is a dummy equal to one if the city repealed ULC during 2007-2008, and zero otherwise. Standard errors are clustered by the states of administration and reported in parentheses. * p< 0.10, ** p< 0.05, *** p< 0.01. 174 Table A.15: ULC act and legal proceedings with 1971 population percentile fixed effects Dependent variable = # Land-related court cases, 2010-2016 (1) (2) (3) (4) (5) (6) ULC act 0.923 2.361*** 1.670* 2.805*** 2.562*** 2.239*** (1.421) (0.829) (0.889) (0.319) (0.502) (0.537) Log population -0.043 -0.637 -0.382 0.123 -0.318 -0.544 (0.615) (0.406) (0.656) (0.372) (0.367) (0.370) Shift-share index 11.927*** 12.108*** 11.477*** 16.594*** 13.247*** 17.730*** (2.856) (1.397) (2.778) (5.571) (3.087) (5.718) Log surface area 0.099 -0.105 0.251 0.147 -0.032 0.496 (0.384) (0.184) (0.540) (0.468) (0.338) (0.434) Share SC/ST 1.574** 2.561 1.970*** -5.413** 1.075 -4.910*** (0.669) (2.540) (0.720) (2.276) (1.069) (1.372) Share females -89.605 -57.840*** -85.857 -130.006** -69.806* -128.665** (62.799) (14.067) (63.743) (61.741) (36.562) (63.937) Share literate -0.074 0.911 0.317 0.604 4.723 -0.623 (3.114) (2.289) (3.243) (4.079) (3.637) (4.666) Mean temperature 0.303 0.293** 0.239 0.565 0.244 0.631 (0.419) (0.125) (0.369) (0.385) (0.327) (0.503) Log rainfall 0.604 0.102 0.411 -0.030 -0.670 -0.859 (1.300) (0.644) (1.141) (0.908) (0.474) (1.314) Constant 9.962 4.084 17.571 5.159 8.641 9.544 (18.658) (9.611) (22.545) (14.872) (17.757) (24.330) State fixed effects ✓ ✓ ✓ ✓ ✓ ✓ 1971 pctile fixed effects ✓ ✓ ✓ ✓ ✓ ✓ No. of percentiles 2 3 4 6 7 8 N 202 202 202 202 202 202 R 2 0.752 0.711 0.753 0.796 0.739 0.787 Source: Author’s calculations. Note: Table presents pseudo maximum likelihood estimates of Poisson regressions using a cross-section of 202 cities. Dependent variable is the number of land-related legal proceedings per 100,000 (based on city population in 2011) registered between 2010 and 2016. ULC act is a dummy equal to one if the city had enacted the ULC act, and zero otherwise. Each column includes 1971 population percentile fixed effects. The number of percentiles (given in table) varies for each column. Standard errors are clustered by the states of administration and reported in parentheses. * p < 0.10, ** p< 0.05, *** p< 0.01. 175 Appendix B Data In this section, I expand on the details about city-level data collection and processing provided in section 3.3 of Chapter 3. First, I present an overview of urban administration in India. Next, I describe the steps of data processing employed to obtain the sample of analysis. Finally, I discuss the various sources from which I gathered data on ULC enactment and repeal across cities of India. B.1 Urban administration in India Urban administration in India is a state subject. State governments are responsible for setting up municipalities in places that transform from rural to urban. However, governments are slow at setting up urban local bodies in fast-urbanizing settlements with rural administration. This has led to a form of in situ urbanization in India (Jain, 2018; Mukhopadhyay et al., 2020). The Census of India recognizes this gap in urban administration and urbanization and de- fines two types of urban areas. The first group of urban areas, termed “Statutory towns”, consists of settlements and cities where there exists some form of urban local bodies—municipal boards, committees, or corporations. The second group of urban areas, termed “Census towns”, consists of smaller settlements that have urban-like features—a population of at least 5,000, a population density of at least 400 persons per square kilometer, and at least 75% of the male workforce em- ployed in non-agricultural activities—but with non-urban administrative governance structures such as panchayats. In 2011, out of the roughly 8,000 towns enumerated by the Census, about half consisted of Census towns, accounting for 14% of India’s urban population (Census of India, 2011b). I use the definition of urban agglomerations (UAs) from the Census of India to identify the 202 cities in my sample. The Census of India defines a UA as “a continuous urban spread constituting a town and its adjoining outgrowths (OGs), or two or more physically contiguous towns together with or without outgrowths of such towns. An Urban Agglomeration must consist of at least a statutory town and its total population (i.e. all the constituents put together) should not be less than 20,000.” The 176 Table B.1: Composition of UAs Percentages 2001 2011 UA constituent indicators (1) (2) Municipalities exist in UAs (%) 100 100 Non-municipal areas exist in UAs (%) 32 47 Mean shares in UAs with non-municipal areas: (i) Municipalities’ share of UA population 92 86 (ii) Non-municipal share of UA population 8 14 (iii) Municipalities’ share of UA surface area 83 74 (iv) Non-municipal share of UA surface area 17 26 Mean shares in all UAs: (i) Municipalities’ share of UA population 97 94 (ii) Municipalities’ share of UA surface area 94 88 Data source: Census of India. Note: Table presents the composition of UAs in terms of the mean percentage shares of populations and surface areas of UAs’, accounted for by non-municipal areas and municipalities. Non-municipal areas are Census towns. Municipalities include smaller urban areas like Town Panchayats, Cantonment Boards, and Notified areas; mid-sized cities with Municipal Boards, Committees, etc.; and large cities with Municipal Corporations. word “town” refers to both Census towns and Statutory towns. The 202 UAs in the sample of analysis consisted of 594 towns in 2001 and 1,031 towns in 2011. The number of constituent towns was higher in 2011 because UAs expanded over time to include more Census towns and municipalities. In table B.1, I present the composition of UAs in terms of their population and surface area distribution among non-municipal areas or Census towns and statutory towns or municipalities. The Census also defines a third group of relatively smaller urban areas on the fringes of statutory towns that are, strictly speaking, not within the confines of urban local bodies as “out- growths”. Outgrowths are not the same as Census towns. More specifically, outgrowths are defined by the Census of India as “a viable unit such as a village or a hamlet or an enumeration block made up of such village or hamlet and clearly identifiable in terms of its boundaries and location.” In the various enumeration tables provided by the Census of India, outgrowths are included in the figures for statutory towns. Out of the 202 UAs in the sample of analysis, outgrowths exist in 85. 177 B.2 Data processing It is hard to gather data at the city level in India because most aggregated datasets are available at the state or district level. However, the larger cities in India usually account for the majority of the population of districts in which they are located. Since I obtain data at the district level for several variables, I choose the largest cities in India so that there is minimal noise when I convert data from the district level to the city level. I begin with the list of 388 UAs with at least 100,000 people in 2001. One reason for choosing the cutoff of 100,000 is that the Census provides many tables with information only on UAs with over 100,000 people. Moreover, the 388 UAs that had at least 100,000 people in 2001 also had more than 100,000 people in 2011. Hence, the cutoff choice of 100,000 for 2001 allows me to construct a consistent panel of UAs between 2001 and 2011. Next, I identify the constituents—municipalities and non-municipal areas—of the 388 UAs in both 2001 and 2011. This exercise is hard because not only does the number of constituents in UAs change over time, but the composition of municipalities within UAs also changes. 1 I use the Class 1 cities’ 2011 table (which contains most, but not all, information on constituent towns of UAs) provided by the Census as a reference to identify the constituent municipalities and non-municipal areas of all the 388 UAs. 2 I cross-verify the information in Class 1 cities table to that of other town- and UA-level datasets using a mixed approach—programmatic and manual. Ultimately, I identified 1,296 constituents in 2001 and 1,929 in 2011 for all the 388 UAs combined. I gather data at three levels—households, districts, and UA constituents (i.e., municipalities, Census towns). 3 Details on the data sources and the initial units of observation are given in table B.2. The conversion of data from the constituent level to the UA level is based on the matching of constituents to UAs. On the other hand, I aggregate household-level data from the National Sample Surveys to the district level. Then, I convert all district-level data (including 1 For instance, in the UA of Hyderabad, there were 19 municipalities and Census towns in 2001, of which the Hyderabad Municipal Corporation was the largest. In 2007, 12 municipalities and Census towns were merged with the Hyderabad Municipal Corporation to form the Greater Hyderabad Municipal Corporation. Hyderabad UA itself expanded, between 2001 and 2011, to include another 18 Census towns and municipalities. Similar compositional changes occurred in cities like Ahmedabad, Bangalore, Surat, and Vasai-Virar. 2 Appendix table A-04 (I), available online: https://censusindia.gov.in/census.website/data/census-tables. 3 Raw data on temperature and rainfall are at the 1°× 1° grid level. I assign grid-level values to districts based on a district-grid match. 178 Table B.2: UA-level data collection Sources of data Source Type Table/round Initial unit Class of variables (1) (2) (3) (4) Residential houses Census Enumeration H-4 Constituent Non-residential buildings ⋆ Census Enumeration H-1 Constituent Population Census Enumeration PCA Constituent Surface area Census Enumeration TD Constituent Sectoral employment ‡ Census Enumeration B-4 District SC/ST population Census Enumeration PCA Constituent Female population Census Enumeration PCA Constituent Literate population Census Enumeration PCA Constituent HH in new units NSS Survey 58 & 69 Household Climate IMD Estimation Grid Note: Table presents the sources of data for the city-level variables used in the analysis. Column (1) provides the source of data. The three major data sources are the Census, the National Sample Surveys (NSS), and the India Meteorological Department (IMD). Column (2) provides the type of data (population Census, survey, or estimated) used. Column (3) provides the Census table or survey round numbers. Column (4) provides the unit of observation in the original dataset. HH stands for household. SC/ST stand for scheduled castes and tribes. PCA stands for Primary Census Abstract. TD stands for town directory. B and H stand for the employment and housing table numbers. ⋆ Available only for 298 municipalities. ‡ Data for the base year (1991) obtained from the Primary Census Abstract. aggregated variables constructed from NSS surveys) to the UA level using population weights. The population weight for a given district-UA pair is the share of a UA’s population living in the district in 2001. In the final dataset, many UAs are dropped because of missing data. Furthermore, some states are dropped to account for other data issues. Notably, the states of Jharkhand and West Bengal are dropped because these states had ULC active until at least 2011. Orissa is dropped because of a large number of outliers in the legal proceedings data. 4 Finally, I dropped the state of Jammu and Kashmir because it did not have a Census in 1991. 4 For instance, Puri, a city of 200,000 people, registered 485 land-related court cases between 2010 and 2016. The average city in the data registered 84 cases. I investigated these numbers with newspaper reports and found no evidence of mass litigations that might explain the high numbers. 179 B.3 ULC data The Parliament of India passed the ULC act in 1976, and it came into force in the 64 largest UAs of India. 5 The law was adopted by most states of India, with the exceptions being Jammu and Kashmir and Kerala. The Tamil Nadu government enacted its own ULC laws, which applied to its six UAs. In my sample of analysis, there are 45 UAs where the ULC laws (Tamil Nadu and federal act combined) were enacted. 6 The ULC act was repealed in a phased manner. While nine states adopted the repeal act between 1999 and 2000, shortly after the central government of India promulgated the repeal bill, the states of Andhra Pradesh, Assam, Bihar, Jharkhand, Maharashtra, Orissa, and West Bengal did not repeal ULC immediately. Andhra Pradesh, Assam, Bihar, Maharashtra, and Orissa repealed ULC between 2002 and 2008. The state of Jharkhand eventually repealed ULC in 2011. However, as of 2023, ULC is still active in West Bengal. The collection of data on the year of ULC repeal by different states is less than simple. The information about the timing of ULC repeal by different states is inconsistent in prior literature (Duranton et al., 2015; Sridhar, 2010). For instance, while Sridhar (2010) mentions that ULC was active in Andhra Pradesh, Himachal Pradesh, Maharashtra, Uttarakhand, and West Bengal during the mid-2000s, Duranton et al. (2015) says that ULC was active in Andhra Pradesh, Assam, Bihar, Maharashtra, and West Bengal, around the same time. The discrepancy in information on states such as Assam, Bihar, Uttarakhand, etc., is unclear from the authors’ accounts of data collection. Therefore, I conduct a thorough search on the internet, using various combinations of key- words, such as “urban land ceiling India”, “urban land ceiling repeal India”, etc., along with state names to gather an extensive list of web links referring to various court case documents, gov- ernment gazette notifications, newspaper reports, etc. I cross-verify several documents for each state to ensure that the repeal date is consistent across sources. I present one source reference for each state in table B.3. 5 The number is 62 if we include Thane and Ulhasnagar as part of the Mumbai UA, which Census 1971 did not, but later Census years did. 6 The 23 UAs where the ULC act was enforced but are not in the sample are, in order of 1971 populations, Kolkata, Delhi, Jaipur, Dhanbad, Jamshedpur, Gwalior, Solapur, Rajkot, Jalandhar, Ranchi, Ajmer, Aligarh, Asansol, Chandi- garh, Gorakhpur, Cuttack, Jamnagar, Bhavnagar, Dehradun, Bikaner, Durgapur, Sangli, and Pondicherry. 180 Table B.3: State-level ULC repeal data collection ULC repeal data sources Repeal year Source State name (1) (2) Andhra Pradesh 2008 Surendra Raj Jaiswal & Anr. v. State of AP Assam 2003 The Assam Gazette (Extraordinary) Bihar 2006 Sheonath Prasad & Ors. v. State of Bihar Chhattisgarh 2000 Shyam Lal & Ors. v. State of Chhattisgarh Gujarat 1999 State of Gujarat v. Manibhai Kashibhai Patel Karnataka 1999 Kuratti Veerappa v. State of Karnataka Madhya Pradesh 2000 Lalaji Choubey v. State of Madhya Pradesh Mahrashtra 2007 Vithabai Bama Bhandari v. State of Maharashtra Punjab 1999 The Gazette of India (Extraordinary) Rajasthan 1999 RIICO v. Asia Printpack Limited & Anr. Tamil Nadu 1999 TN Urban Land (Ceiling & Regulation) Act Uttar Pradesh 1999 State of Uttar Pradesh v. Hari Ram Uttarakhand 1999 Anand Mani v. State of Uttarakhand Note: Table presents information about the repeal of ULC act by each state in the sample of analysis. Column (1) provides the year of repeal. Column (2) provides the primary source of ULC repeal information. The repeal year has been cross-verified with other court cases, newspaper reports, etc. AP stands for Andhra Pradesh. TN stands for Tamil Nadu. RIICO stands for Rajasthan State Industrial Development and Investment Corporation Limited. Anr. stands for another. Ors. stands for others. The states of Haryana, Jharkhand, Kerala, Orissa, and West Bengal and the union territories of Delhi and Pondicherry are not presented here because there are no UAs in the sample of analysis from these states which had enacted ULC. Note that the information provided on ULC repeal in table B.3 is not exhaustive in terms of the ULC’s coverage of cities. I do not include information on states, such as Jharkhand and West Bengal, because these are not part of the sample of analysis. The other states which are part of the analysis sample, but do not have any cities where ULC was enforced, are Haryana and Kerala. 181 Appendix C Derivations and proofs Proof of proposition 2.1: First, note that we can use the implicit function theorem to get the partial derivative of pop- ulation at i with respect to the distant shock, as follows: dn i dz j = ∂n i ∂m ji m ′ ji (z j )+ ∂n i ∂m ij m ′ ij (z j ) (C.1a) Next, differentiating the aggregate housing demand at i, H i , with respect to z j , and substituting equation (C.1a) in equation (2.1a), we get the following: dH D i dz j = ∂H D i ∂n i dn i dz j h d i (r i , w i )= ∂H D i ∂n i ∂n i ∂m ji m ′ ji (z j )+ ∂n i ∂m ij m ′ ij (z j ) h d i (r i , w i ) (C.1b) We know that h d i (r i , w i ) > 0 since utility is strictly quasiconcave. From equation (2.1a), we have ∂H D i ∂n i > 0, because the demand for housing will be strictly increasing in the total population. Now, since ∂n i ∂m ji (z j ) ≥ 0 and ∂n i ∂m ij (z j ) ≤ 0, a net non-zero sign on the sum of these two terms will imply that dH D i dz j ̸= 0. Now, on the other hand, if dH D i dz j ̸= 0, then it must be the case that ∂n i ∂m ji (z j ) + ∂n i ∂m ij (z j ) ̸= 0. Proof of proposition 2.2: The fact that dH D i dz j ⋚ 0 implies ∂n i ∂m ji m ′ ji (z j ) ⋚ ∂n i ∂m ij m ′ ij (z j ) directly follows from equation (C.1b) and the inequality ∂H D i ∂n i > 0. Now, to see the if condition, note first that ∂n i ∂m ji , m ′ ji (z j ) and m ′ ij (z j ) are all weakly positive and ∂n i ∂m ij is weakly negative. Hence, we have ∂n i ∂m ji m ′ ji (z j ) ≥ 0 and ∂n i ∂m ij m ′ ij (z j ) ≤ 0. If ∂n i ∂m ji m ′ ji (z j ) = ∂n i ∂m ij m ′ ij (z j ) , then dH D i dz j = 0 trivially follows from equation (C.1b). If ∂n i ∂m ji m ′ ji (z j ) ̸= ∂n i ∂m ij m ′ ij (z j ) , then there are three possibilities. First, we can have ∂n i ∂m ji m ′ ji (z j )> 0 and ∂n i ∂m ij m ′ ij (z j ) = 0, in which case equation (C.1b) implies dH D i dz j > 0. The second possibility is where ∂n i ∂m ji m ′ ji (z j ) = 0 and ∂n i ∂m ij m ′ ij (z j ) < 0, in which case dH D i dz j < 0 follows from equation (C.1b). And finally, we can have ∂n i ∂m ji m ′ ji (z j ) > 0 and ∂n i ∂m ij m ′ ij (z j ) < 0, in which case we have dH D i dz j > 0 if ∂n i ∂m ji m ′ ji (z j ) > ∂n i ∂m ij m ′ ij (z j ) . 182 Derivation of equation (3.5): Solving the consumer’s optimization problem, we get the following indirect utility: u(A, p(d, t), w(d, t))= Aw(d, t)p(d, t) − η = U (C.2a) The bid-rent p(d, t) is a function of city-wide amenities, the wage, and the outside indirect utility: p(d, t)= w(d, t) u ! 1 η where, u= U A (C.2b) Suppose the productivity and transport cost parameters are λ ′ and µ ′ , respectively. Then, equa- tion (C.2b) can be rewritten as a function of the initial central city wage value w, the amenity- adjusted utility u, and the parameters η, λ ′ , and µ ′ , as follows: p(d, t)= pe λt− µx where, p= w u ! 1 η , λ= λ ′ η , µ= µ ′ η (C.2c) Derivation of equation (3.6): The informal land rent is a function of the bid-rent p(d, t) and building height: r I (d, t)= p(d, t)h I (d, t)− h I (d, t) ( ψ ψ− 1 ) (C.3a) Maximizing r I (d, t) with respect to building height h I (d, t), we get the following: h I (d, t)= ψ− 1 ψ p(d, t) ! (ψ− 1) (C.3b) Substituting equation (C.3b) in equation (C.3a), we get equation (3.6). Derivation of equation (3.7a): Landowners’ rent from formal housing construction is: R F (d,τ)= h F (d,τ)Φ(d,τ)− h F (d,τ) ( φ φ− 1 ) (C.4a) 183 Maximizing R F (d,τ) with respect to h F (d,τ), we get the following: h F (d,τ)= φ− 1 φ Φ(d,τ) ! (φ− 1) (C.4b) Substituting equation (C.4b) in equation (C.4a), we get equation (3.7a). Derivation of equation (3.7b): The shareΥ of amortized cost of construction in lifetime formal land revenue can be solved fromΥh F (d,τ)Φ(d,τ) = h F (d,τ) ( φ φ− 1 ) to obtainΥ = 1− 1 φ . This means that amortized construc- tion costs represent 1− 1 φ fraction of formal lifetime land revenue. Therefore, formal land rent net of amortized construction costs is 1 φ share of formal land revenue for the given period. Benchmark land development and derivation ofΘ: The present discounted value of a parcel of land without development cost or regulation, where informal development precedes formal development, is: Π 0 (d)= Z τ 0 0 re − ρt dt+ Z τ 1 τ 0 r I (d, t)e − ρt dt+ R F (d,τ 1 )e − ρτ 1 (C.5a) The first-order condition from maximizing Π 0 (d) with respect to τ 0 is given as follows: r = r I (d,τ 0 ) (C.5b) =⇒ p(d,τ 0 )= r 1 ψ ψ (ψ− 1) ψ− 1 ψ (C.5c) The first-order condition from maximizing Π 0 (d) with respect to τ 1 is given as follows: r I (d,τ 1 )= r F (d,τ 1 ) ρ− φλ ρ− λ ! (C.5d) =⇒ p(d,τ 1 )= {φ(ρ− λ)} φ (φ− 1) (φ− 1) (ψ− 1) (ψ− 1) ψ ψ 1 ρ− φλ ! 1 φ− ψ (C.5e) 184 The present discounted value of a parcel of land, where only formal development occurs, is: Π 2 (d)= Z τ 2 0 re − ρt dt+ R F (d,τ 2 )e − ρτ 2 (C.5f) The first-order condition from maximizing Π 2 (d), with respect to τ 2 , yields the following: p(d,τ 2 )= r 1 φ φ (φ− 1) φ− 1 φ ρ− λ (ρ− φλ) 1 φ (C.5g) If informal development occurs first on agricultural land followed by formal construction, then we have τ 2 > τ 1 . This implies that p(d,τ 2 ) > p(d,τ 1 ) since the bid-rent p(d, t) increases at a constant exponential rate with time t. Therefore, for informal development to occur first on agricultural land, we need the following condition to hold: r< ψ (ψ− 1) ψ− 1 ψ (φ− 1) φ− 1 φ φ (ρ− φλ) 1 φ ρ− λ ! ϕψ ϕ− ψ =Θ (C.5h) In other words, informal development exists if r<Θ. Proof of proposition 3.1: The present discounted value of a land parcel that requires development costs can be written as follows: Π c (d)= Z τ c 1 τ c 0 r I (d, t)e − ρt dt− Ce − ρτ c 0 + R F (d,τ c 1 )e − ρτ c 1 (C.6a) The first-order condition from maximizing Π c (d) with respect to τ c 0 is given as: ρC = r I (d,τ c 0 ) (C.6b) Now, the if condition in proposition3.1 and equation (C.5b) imply that r I (d,τ c 0 )> r I (d,τ 0 ), which in turn means that p(d,τ c 0 ) > p(d,τ 0 ), since r I (d, t) is monotonically increasing in the bid-rent p(d, t). The bid-rent at a given distance d from the city center increases exponentially with time at a constant rate. Therefore, we have τ c 0 > τ 0 . Land parcels, at distance d from the city center, 185 that require development costs, become part of the city at time τ c 0 , defined by equation (C. 6b). But, the other parcels that do not require any development cost enter the city at date τ 0 , which is given by the first-order condition in equation (C. 5b). The parcels converting from agricultural land to informal houses at τ 0 also define the boundary of the city, d b . Therefore, the 2θπd land parcels requiring development costs remain vacant within the city for time interval [τ 0 ,τ c 0 ). Proof of proposition 3.2: First, the sufficient condition for the existence of vacant land follows from the fact that C > Θ ρ > r ρ . Now, for the necessary condition, consider the scenario where informal development precedes formal construction. In this case, the first order condition for informal development to begin solves for τ c 0 , defined in equation (C. 6b). The resulting expression for the bid-rent function is, as follows: p(d,τ c 0 )= C 1 ψ ρ 1 ψ ψ (ψ− 1) ψ− 1 ψ (C.7a) Now, consider the situation where vacant land is only used for formal housing construction. The present discounted value of land rent, in this case, is: Π f (d)= R F (d,τ f 2 )e − ρτ f 2 − Ce − ρτ f 2 (C.7b) In this scenario, land remains vacant during the time interval [τ 0 ,τ f 2 ). Landowners choose τ f 2 to maximize equation (C.7b). The first-order condition from maximizing Π f (d) with respect to τ f 2 can be written, as follows: ρC = r F (d,τ f 1 ) ρ− φλ ρ− λ ! (C.7c) =⇒ p(d,τ f 2 )= C 1 φ ρ 1 φ φ (φ− 1) φ− 1 φ ρ− λ (ρ− φλ) 1 φ (C.7d) Substituting the expression forΘ, the necessary condition in proposition 3.2 can be rewritten, as 186 follows: C 1 ψ ρ 1 ψ ψ (ψ− 1) ψ− 1 ψ > C 1 φ ρ 1 φ φ (φ− 1) φ− 1 φ ρ− λ (ρ− φλ) 1 φ (C.7e) Equations (C.7a), (C.7d) and (C.7e) imply that p(d,τ c 0 ) > p(d,τ f 2 ). Since the bid-rent p(d, t) increases exponentially at a constant rate with time, we have τ c 0 > τ f 2 . Therefore, formal housing construction occurs at an earlier date τ f 2 . Proof of proposition 3.3: Suppose that vacant land under the ULC act is used for informal housing development first, followed by formal housing construction. The date of informal development on vacant land is τ c 0 which is given by equation (C.6b). The corresponding bid-rent function is given in equa- tion (C.7a). Since, from proposition 3.2, we have τ c 0 > τ f 2 , under this scenario, land requiring development costs remain vacant for the time interval [τ 0 ,τ c 0 ). Now, consider the case where vacant land under the ULC act is directly developed into formal housing. Landowners incur an additional fixed cost G at the beginning of formal construction as a result of the ULC act. The present discounted value of land rent for this case is: Π g 2 (d)= R F (d,τ g 2 )e − ρτ g 2 − (C+ G)e − ρτ g 2 (C.8a) Here, land remains vacant during the time interval [τ 0 ,τ g 2 ). The first-order condition from maxi- mizingΠ g 2 (d) with respect to τ g 2 can be derived, as follows: ρ(C+ G)= r F (d,τ g 1 ) ρ− φλ ρ− λ ! (C.8b) =⇒ p(d,τ g 2 )=(C+ G) 1 φ ρ 1 φ φ (φ− 1) φ− 1 φ ρ− λ (ρ− φλ) 1 φ (C.8c) The if condition in proposition 3.3 and equations (C.6b) and (C.8c) imply that p(d,τ g 2 )> p(d,τ c 0 ). This implies thatτ g 2 > τ c 0 since the bid-rent p(d, t) increases exponentially at a constant rate with time. Therefore, informal development precedes formal construction on vacant land under the ULC act. 187

Asset Metadata

Creator Dutta, Arnab (author)

Core Title Essays on internal migration and housing markets of urban India

Contributor Electronically uploaded by the author (provenance)

School School of Policy, Planning and Development

Degree Doctor of Philosophy

Degree Program Urban Planning and Development

Degree Conferral Date 2023-05

Publication Date 04/07/2025

Defense Date 02/09/2023

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Housing,India,migration,OAI-PMH Harvest

Format theses (aat)

Language English

Advisor (Boarnet, Marlon G.), Green, Richard K. (committee chair), De La Roca, Jorge (committee member), Kahn, Matthew E. (committee member)

Creator Email arnab.dutta822@gmail.com,arnabdut@usc.edu

Unique identifier UC112956282

Identifier etd-DuttaArnab-11585.pdf (filename)

Legacy Identifier etd-DuttaArnab-11585

Document Type Dissertation

Format theses (aat)

Rights Dutta, Arnab

Internet Media Type application/pdf

Type texts

Source 20230410-usctheses-batch-1018 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email uscdl@usc.edu

Abstract (if available)

Abstract This dissertation explores how Indian cities’ housing supply responds to urban growth, which is increasingly driven by migration. The first chapter shows that technological growth-induced rural-urban migration accounted for a quarter of India’s urban population growth during the early 2000s. Rural residents moved to urban areas in response to India’s mobile phone service expansion because of higher urban wages and relatively seamless transmission of distant labor market information. The effects were, however, primarily concentrated in the wealthiest urban areas that drew the most migrants. The second chapter shows that droughts and highway infrastructure investments induced interstate migration in India and increased housing demand in Indian cities during the 2000s. This chapter then estimates the slopes of urban India’s formal, informal, and vacant housing supply curves. The estimates indicate that while the formal housing supply in India is inelastic, developers engage in speculative construction as well. A negative informal housing supply elasticity estimate confirms the existence of gentrification. The third chapter shows that the repeal of India’s urban land ceiling laws, which allowed governments to acquire privately owned vacant land through ceiling limits, did not lead to an expected growth in formal housing supply during the 2000s. Disputes in the ownership of vacant land parcels led to legal battles between governments and landowners after the reform, which delayed the new construction of formal houses. The dissertation’s findings indicate that while Indian cities are expanding, their housing markets are not growing fast enough because of land use regulations and institutional frictions.