Inequality, Demography and Fiscal Policy 9819905176, 9789819905171

Citation preview

Applied Economics and Policy Studies

Weijie Luo

Inequality, Demography and Fiscal Policy

Applied Economics and Policy Studies Series Editors Xuezheng Qin , School of Economics, Peking University, Beijing, China Chunhui Yuan, School of Economics and Management, Beijing University of Posts and Telecommunications, Beijing, China Xiaolong Li, Department of Postal Management, Beijing University of Posts and Telecommunications, Beijing, China

The Applied Economics and Policy Studies present latest theoretical and methodological discussions to bear on the scholarly works covering economic theories, econometric analyses, as well as multifaceted issues arising out of emerging concerns from different industries and debates surrounding latest policies. Situated at the forefront of the interdisciplinary fields of applied economics and policy studies, this book series seeks to bring together the scholarly insights centering on economic development, infrastructure development, macroeconomic policy, governance of welfare policy, policies and governance of emerging markets, and relevant subfields that trace to the discipline of applied economics, public policy, policy studies, and combined fields of the aforementioned. The book series of Applied Economics and Policy Studies is dedicated to the gathering of intellectual views by scholars and policymakers. The publications included are relevant for scholars, policymakers, and students of economics, policy studies, and otherwise interdisciplinary programs.

Weijie Luo

Inequality, Demography and Fiscal Policy

Weijie Luo School of Economics Beijing International Studies University Beijing, China Department of Economics and Related Studies University of York York, UK

ISSN 2731-4006 ISSN 2731-4014 (electronic) Applied Economics and Policy Studies ISBN 978-981-99-0517-1 ISBN 978-981-99-0518-8 (eBook) https://doi.org/10.1007/978-981-99-0518-8 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

This book is dedicated to my son, Yichen Luo.

Foreword

In an article published in the Journal of Political Economy in 1981, Meltzer and Richard (1981) presented what has become the seminal theoretical contribution to the political economy literature on the size of government. The theory is simple and elegant. It combines the logic of the median voter theorem with a fiscal system that can tax labor income proportionally to fund a universal public good that benefits everyone. The theory makes two predictions which can best be thought of as representing two sides of the same coin. The first prediction is that an increase in (labor income) inequality, which makes the median voter poorer relative to the average voter, leads to an increase in the size of government under universal suffrage. The second prediction applies to a semi-democratic society with restrictions on who can vote. It says that an extension of the voting franchise that gives the right to vote to individuals with lower incomes will, for a given income distribution, lead to an increase in the size of the government. In both cases, the powerful logic is that the median voter demands more redistribution—a higher tax rate and more public goods—when they become poorer (as inequality increase) or when the median voter among the enfranchised citizens become poorer. These predictions have become known as the Meltzer– Richard hypothesis, and they feature prominently both in Acemoglu and Robinson (2000)’s theory of franchise extension and in Lindert (2004a, b)’s historical analysis of public sector growth. The hypothesis has been subject to very extensive empirical investigation, and despite the appealing logic, the evidence that higher levels of inequality or franchise extension is associated with more redistribution and larger government is, at best, mixed. Some studies find evidence that is consistent with the logic (e.g., Borge and Rattsø 2004; Aidt and Jensen 2013) while many others do not (e.g., Benabou 1996; Aidt et al. 2022). Inequality, Demography and Fiscal Policy takes the theoretical and empirical analysis of the political economy of the fiscal state one important step forward. It demonstrates how one can resolve many of the ambiguities of past empirical research on the Meltzer–Richard hypothesis. The principal argument of the book is that the theory needs to be enriched and that empirically relevant extensions must be integrated. Once these additional features are added, new testable implications

vii

viii

Foreword

emerge and when they are confronted with data, they fare a lot better than the original Meltzer–Richard hypothesis. This is quite an achievement! The book highlights three important features that have been overlooked or downplayed in the existing literature. The first is that the type of inequality matters. Meltzer and Richard (1981) focus on labor income inequality. While important, in recent decades there has been a rise in capital income inequality. With inequality along both dimensions factored into the analysis, it becomes clear that they affect the median voter’s desire for state-sponsored redistribution in opposite ways. The second feature is that the set of policy instruments available to governments to fund their activities is much richer than the simply proportional tax on labor income considered by Meltzer and Richard (1981). In particular, governments can also use indirect commodity taxes, VAT, or sales taxes or they can issue debt to postpone increasing taxes to the future. Again, this makes a big difference to the demand for redistribution. The third feature is demographic changes related to population aging. Since distribution not only takes place within a generation but also across generations, the demographic structure makes a key difference to the political economy of redistribution. The book, step-by-step, builds these features into the theoretical structure and derives predictions which are, then, tested against real-world observational data. It is a marked strength of the book that it combines formal theoretical modeling with econometric testing. This demonstrates the relevance of the theoretical arguments, and it helps us understand how the big trends—increasing inequality (along different dimensions) and population ageing—affect the policy choices made by (democratic) governments and how those, in turn, affect the size of government. On top of that, the analysis in the book goes one step further. It investigates how these trends, via their impact on fiscal policy, affect economic growth. The existing literature, starting with the classical article by Persson and Tabellini (1994), have not been able to find a robust link between inequality and growth. The book shows that this may very well be due to the fact that the distinction between labor and capital income inequality has been ignored. This important book will appeal to anyone with an interest in the political economy of inequality, redistribution, population aging, democracy, and economic growth. Toke S. Aidt Faculty of Economics University of Cambridge Cambridge, UK [email protected] CESifo Munich, Germany

Foreword

ix

References Acemoglu D, Robinson J (2000) Why did the west extend the franchise? Democracy, inequality, and growth in historical perspective. Q J Econ 115(4):1167–1199 Aidt TS, Jensen PS (2013) Democratization and the size of government: evidence from the long 19th century. Public Choice 157:511–542 Aidt TS, Winer SL, Zhang P (2022) Franchise extension and fiscal structure in the UK 1820-1913: a new test of the redistribution hypothesis. Cliometrica 16:547–574 Benabou R (1996) Inequality and growth. National Bureau of Economic Research Macroeconomics Annual 11:11–74 Borge L-E, Rattsø J (2004) Income distribution and tax structure: empirical test of the MeltzerRichard hypothesis. Eur Econ Rev 48(4):805–826 Lindert PH (2004a) Growing public. Social spending and economic growth since the eighteenth century. The story, vol I. Cambridge University Press, Cambridge Lindert PH (2004b) Growing public. Social spending and economic growth since the eighteenth century. Further evidence, vol II. Cambridge University Press, Cambridge Meltzer A, Richard S (1981) A rational theory of the size of government. J Polit Econ 89(5):914– 927 Persson T, Tabellini G (1994) Is inequality harmful for growth? Am Econ Rev 83(3):600–621

Preface

The rising inequality in many developed countries has sparked renewed interest among policymakers, academics, and the general public over the last decade, as evidenced by the attention Piketty’s (2014) academic book has generated. Following up on Kuznets’ (1953) pioneering study, a number of authors (Alvaredo et al. 2011– 2022) have constructed long-run series of top income shares to measure income inequality. For example, estimates from the World Wealth and Income Database show that income concentration is high and growing in the United States: The richest 1% of households earned 20% of total income in 2015, up from 11% in 1978, while the bottom 50% experienced a complete collapse, seeing a total income decline from 20% to 12% (Piketty et al. 2018). In contrast and in spite of a similar trend, in 2015, the top 1% in China remained proportionally smaller than the bottom 50% though it began approaching the levels seen in the United States (Piketty et al. 2019); the same was observed to an even lesser extent in France as a typical representative of the West European pattern (Garbinti et al. 2017). Despite the global resurgence of income inequality, there is still much doubt and debate as to whether such inequality is desirable. This debate is not new as the phenomenon is often highly contested, but in recent years it has fallen under the category of the “redistribution and growth” debate. However, Meltzer and Richard’s (1981) seminal paper, building on Romer (1975) and Roberts’ earlier research (1977), offers a sanguine prediction: Greater before-tax income inequality implies divergence between mean and median income, and under universal suffrage, leads to increased redistribution. Democracy thus provides a corrective to increased inequality. However, evidence supporting the Meltzer and Richard (1981) hypothesis is generally weak. Perotti (1996), Benabou (1996), Persson and Tabellini (2003), and Shelton (2007) have all found an insignificant or even negative relationship between government size and the degree of inequality. Chapter 1 of this book begins by testing this hypothesis in the context of the United States. However, the prediction was empirically rejected empirically using American state-level data. It was found that inequality induced by differences in capital income, derived from the Panel Study of Income Dynamics data, was negatively associated with state government size. Moreover, this chapter shows that capital xi

xii

Preface

income inequality plays a key role in poor states, while labor income inequality is key in rich places regarding the explanation of the change in state government. Hence, this book seeks to contribute to both the theoretical and empirical literature on the sources of income inequality. In the original mechanism (Meltzer and Richard 1981; Persson and Tabellini 1994), labor is the sole income channel, and the rich have higher income by means of higher productivity (individual-specific skills, in other words). Nevertheless, in reality, labor is not the sole channel of income for the rich, as has been widely observed (see Piketty et al. 2019), and moreover, the labor share of income has consistently declined in recent years (see Azmat et al. 2012; Karabarbounis and Neiman 2014). Indeed, Piketty (2014) has linked increasing inequality to the declining labor share: If the rate of return on capital is greater than the rate of economic growth, then the capital share increases, and if ownership is highly concentrated within a small number of groups, then inequality inexorably rises. Furthermore, capital income has recently become more unequal as well as more important. Kaymak and Poschke (2016) and Saez and Zucman (2016) documented considerable increases in the concentration of wealth in the United States over the past 50 years. Hence, the current inequality redistribution and inequality–growth literature suffers from one crucial omission: By focusing on the impacts of inequality and using the often assumed (aggregate) inequality induced solely by differences in labor productivity, the analysis fails to comprehensively consider capital income inequality. Chapter 2 attempts to address this omission by constructing a median voter model with the distinction that income inequality is engendered by differences in capital income as well as differences in labor productivity. The chapter, therefore, asks how inequality stemming from capital income affects government size. The key issue is that labor income is taxable, while income from capital is harder to tax. Evidence of tax evasion or avoidance in the case of the latter (perhaps also due to capital mobility and international tax competition) abounds. It is harder to escape from PAYE. Like labor income, individuals differ in their endowment of capital, with there being a right-skewed distribution of capital income. The majority of individuals, endowed with limited (or zero) assets or wealth, are compelled to supply labor for their income, which is taxed. On the other hand, those paid in capital income are, to a meaningful extent, able to avoid the same tax obligation, resulting in relatively lower taxation exposure for the capital-rich. When income differences are induced by capital income, the median voter’s capacity to redistribute through the tax system is restricted as the capital-rich supply less (taxable) labor. If capital income inequality rises such that the capital-rich supply less labor, then the demand for tax on labor declines as the capital-poor (median voter) cannot expropriate the rich. The original Meltzer and Richard (1981) hypothesis is thus reversed: Increased inequality in capital income leads to smaller government. Chapter 2 reports on an empirical investigation of the relationship between government size and inequality using a panel of Organisation for Economic Co-operation and Development (OECD) countries and including a measure of capital income inequality as an additional explanatory variable. Due to limited availability of direct measures of capital income inequality, it was proxied in the empirical analysis by the

Preface

xiii

top 1% total income share, obtained from the World Wealth and Income Database. Piketty (2014) provided a theoretical justification for this proxy as in that study, capital is disproportionately owned by a small number of groups. In this analysis, rising capital income with fixed capital ownership results in a larger top income share. Certainly, as in Piketty et al. (2019), capital income is an important component of the income of the top 1%. The empirical work also separately employs specific measures of productivity-induced labor income inequality as distinct from capital income inequality. These two measures are empirically and conceptually distinct from one another. Consistent with the theory, government size is negatively associated with capital income inequality. The negative relationship held when the lagged dependent variable was controlled for and also when capital income inequality was instrumented with measures of technological progress and capital market access. Moreover, controlling for capital income inequality yielded a positive and significant relationship between government size and labor income inequality, consistent with Meltzer and Richard (1981) and in contrast to the voluminous empirical work testing their hypothesis. Public debt continues to rise in advanced countries. Japan’s, the United Kingdom’s, and the United States’ combined quantitative easing programs and the European Union’s (EU) liquidity injections have been of an unprecedented scale. Chapter 3 aims to understand how government debt can be caused by changes in the income distribution, utilizing the Luxembourg Income Survey. When income differences are generated by capital income, the median voter’s ability to redistribute through taxation is constrained, affecting the public debt accumulation. If capital income inequality increases (and it is the rich who enjoy capital income) such that redistributive policies are reduced, then public debt falls because governments’ borrowing incentives decline. In 2012, the European Commission presented an assessment of the economic effect of population aging in the EU and supported the idea that fiscal challenges increasingly resultant from both a larger proportion of an older population and a fall in the proportion of the population that is economically active. Hence, a separate strand of recent research focuses on the effect of population aging on fiscal policy, again related to Meltzer and Richard (1981). Razin et al. (2002) theorized that increases in the dependency ratio lead to lower labor tax rates and a fall in the generosity of social transfers in democracies. The government redistributes funds emanating from labor income taxes to both workers and retirees, and under democracy, the equilibrium tax rate is that which the median voter prefers. An increasing dependency ratio implies a decline in the population growth rate and lowers income taxes and transfers because the median voter is a worker who increasingly resents redistribution as the retired population increases. Empirical evidence generally has not supported this hypothesis that focuses exclusively on income taxes. Disney (2007), Sanz and Velazquez (2007), and Shelton (2008) have all found a positive relationship between population aging and the size of the welfare state. Hence, this book also seeks to contribute to both the theoretical and empirical political economy literature on the composition of taxes. This idea was motivated by recent rises in taxes on goods and services and concurrent population aging, whereas

xiv

Preface

income taxes are the only revenue source in the original Razin et al. (2002) hypothesis. Revenue sources outside of income taxes are thus becoming increasingly important empirical components of total revenue (e.g., expenditure taxes have accounted for approximately 30% of the total tax revenue in the United Kingdom in recent decades). This is potentially a paradox as it might have been expected that countries with larger numbers of retirees would have higher income taxes and lower expenditure taxes, reflecting the increased political clout of the retired population, who, presumably, would prefer income taxes. Chapter 4 of this book analyzes the effect of population aging on the composition of taxes in a political economy model. In an overlapping generations model, taxes are levied on both income and expenditure, financing redistribution to both the working and retired populations, with a balanced budget period by period. Only workers pay income taxes, but both generations pay expenditure taxes. As in Razin et al. (2002), the median voter’s choice is the unique Condorcet winner. When the median voter is of working age (as widely observed), then population aging increases the demand for expenditure rather than income taxes in order to increase the tax burden on the retired population. An unambiguous finding reported in this chapter is that the composition of taxes, defined as the extent to which taxes are levied on income relative to expenditure, theoretically always declines with the share of retirees. Chapter 5 reports an empirical investigation of the relationship between population aging and the extent of taxes on income relative to taxes on expenditure using international panel data. Following Pickering and Rajput (2018), the dependent variable was constructed based on the ratio of taxes on income, profits, and capital gains to taxes on goods and services, and the key demographic measure was the percentage of the population over the age of 65 years, with reference to the World Development Indicators database. Consistent with the theory, the extent of taxes on income relative to expenditure was found to be negatively associated with the fraction of the retired population. This relationship holds more strongly in stronger democracies. Apart from a higher share of the total older-age population, the decline in the share of the population that is youthful deserves our attention. Chapter 6 investigates how the younger demographic structure affects the composition of taxes. We introduce a model featuring unbalanced fertility in the income distribution. The model theorizes that a higher rate of population growth increases income inequality, leading to more demand for income taxes rather than expenditure taxes, as the median voter is relatively poor and thus resents being double-taxed on the consumption side. International panel evidence supports this hypothesis, and the statistical relationship holds most significantly in stronger democracies. Chapter 7 turns to the influence of population aging on government debt. The chapter argues that government debt is not induced by large spending, particularly in an era of a steady government expenditure. Instead, population aging leads to more demand on expenditure rather than income taxes, which causes the government to issue more debt as the rise in expenditure taxes can hardly mitigate the negative consequences of the fall in income taxes. The next part of the book takes a step further to study how income inequality and demographic change affect subsequent economic growth globally in the short and

Preface

xv

medium term. This section of the book comprises four chapters, from Chaps. 8 to 11. Chapter 8 is a detailed review and summary of the existing literature; it systematically reviews the theories and empirical studies related to the relationship between income inequality and economic growth. Much of the following literature focuses on the impact of income inequality on economic growth, typically based on Meltzer and Richard (1981). One important benchmark Persson and Tabellini’s (1994) endogenous growth model; those scholars argued that, if in a society, the political decisions regarding redistribution generate economic policies that tax investment, then inequality should harm growth as it increases redistributive tax pressures. Empirical support for this hypothesis is generally weak. For example, Forbes (2000) found a positive and significant relationship between an increase in the level of income inequality in a given country with subsequent growth rates in the short and medium term, by controlling for country-specific effects and period effects. Chapter 9, building on Chap. 2, discusses how inequality affects economic growth in an endogenous growth model. The benchmark is Persson and Tabellini (1994), who argued that productivity-induced income inequality leads to lower growth as distortionary taxes increase and harm capital accumulation. When income differences are generated by capital income, the median voter’s ability to redistribute through taxation is constrained, while such redistributive policies are financed by distortionary taxes, in principle, affecting capital accumulation and growth-promoting activities. If capital income inequality increases (and it is the rich who enjoy capital income) such that labor tax rates fall, then the subsequent rate of economic growth increases because distortionary taxes fall and investment is facilitated. In direct contrast to Persson and Tabellini (1994), increased inequality in capital income leads to higher economic growth. Chapter 9 reports the results of testing the relationship between inequality and growth in a panel of OECD countries. Augmenting Forbes’ (2000) work, capital income inequality was included as an additional explanatory variable. Consistent with the theory, growth was found to be positively associated with capital income inequality in the short and medium term. The positive relationship survived under a variety of econometric specifications, including when difference and system generalized method of moments techniques were used to address potential endogeneity problems. Moreover, once capital income inequality was controlled for, the impact of labor income inequality became negative, consistent with Persson and Tabellini (1994) and in contrast to Forbes’ (2000) empirical work, for instance. Despite the impact of inequality on growth, Chap. 10 includes two sections exploring how demographic change affects economic growth, especially via the influence of tax composition. First, population aging increases the demand for expenditure taxes rather than income taxes, leading to higher growth since the extent of distortionary relative to non-distortionary taxes falls and investment is facilitated. Second, an increase in the youthful fraction of the population increases the demand for income taxes rather than expenditure taxes, as expenditure taxes are shared among three generations whereas income taxes are mainly afforded by the working-age population that increasingly dislikes being double-taxed on consumption while raising their

xvi

Preface

dependent children. This reduces growth since the extent of distortionary relative to non-distortionary taxes rises and investment is constrained. Chapter 11 further discusses the impact of the relative use of distortionary and non-distortionary taxes on economic growth using a panel of OECD countries and a sample period spanning 1980 to 2015. In particular, our research examined whether the impact of non-distortionary taxes on economic growth increases with aging, which has increased in Western countries over the past 30 years. We did so by extending the dataset to include the dramatic change in the demographic structure seen in OECD countries since 1990. Using this extended data range allowed us to find a negative effect of distortionary taxation as expected. However, our results also show that an increase in non-distortionary taxation is negatively associated with growth. Hence, in this chapter, we argue that recent distortions from non-distortionary taxation can be accounted for by rapid population aging. Economists often formulate ideal tax systems, but politicians cannot always adopt the ideal tax design because of the risk of losing popularity with the electorate. Our analytical results could be applied to design tax systems that can be implemented in the real world. The last part of this book analyzes the relationship between demographics and income inequality. It builds upon the Malthusian trap that occurs when population growth outpaces agricultural production, causing famine or war and resulting in poverty and depopulation. As argued in Chap. 12, demographic changes increase the demand for expenditure rather than income taxes, leading to a rise in income inequality since expenditure is widely distributed in poor households, and increased expenditure relative to income taxes intensifies inequality. In summary, this book aims to empirically and theoretically study how income inequality and demographic change have affected fiscal policy and subsequent economic growth globally in the past decades. The features of this book can be summarized as follows. Firstly, it briefly reviews the dynamic changes in income sources that contribute to inequality. Secondly, it distinguishes between income inequality induced by differences in labor productivity and that induced by differences in capital income. Thirdly, it briefly reviews the dynamic changes in tax composition in the age of demographic change. Lastly, it discusses the impacts of changes in age structure on the extent of taxes on income relative to expenditure. This book offers a comprehensive discussion toward understanding and analyzing the reasoning, performance, and challenges of fiscal policy and economic growth from the perspective of inequality and demographics. In addition to students, teachers and researchers in the areas of equity, demography, public finance, political economy, and economic policy, this book is also of great interest to policymakers, planners, and non-government agencies concerned with understanding and addressing poverty- and aging-related issues in developed and developing countries. Beijing, China

Weijie Luo

Preface

xvii

References Alvaredo F, Atkinson A, Piketty T, Saez E, Zucman G (2011–present) The World Inequality Database. online at http://www.wid.world Azmat G, Manning A, Reenen JV (2012) Privatization and the decline of labour’s share: international evidence from network industries. Economica, 79(315):470–492. Benabou R (1996) Inequality and growth. Nat Bur Econ Res Macroecon Ann 11:11–74 Disney R (2007) Population ageing and the size of the welfare state: Is there a puzzle to explain?. Eur J Political Econ 23(2):542–553 Forbes KJ (2000) A reassessment of the relationship between inequality and growth. Am Econ Rev 90(4):869–887 Garbinti B, Goupille J, Piketty T (2017) Inequality dynamics in France, 1900–2014: Evidence from distributional national accounts (DINA). WID.world Working Paper Karabarbounis L, Neiman B (2014) The global decline of the labor share. Q Econ 129(1):61–103 Kaymak B, Poschke M (2016) The evolution of wealth inequality over half a century: The role of taxes, transfers and technology. J Monetary Econ 77:1–25 Kuznets S, Jenks E (1953) Shares of upper income groups in income and savings. In: Shares of upper income groups in income and savings. Natl Bur Econ Res 171–218 Meltzer AH, Richard SF (1981) A rational theory of the size of government. J Polit Econ 89(5):914– 927 Perotti R (1996) Growth, income distribution, and democracy: what the data say. J Econom Growth 1(2):149–187 Persson T, Tabellini G (1994) Is inequality harmful for growth? Am Econom Rev 84(3):600–621 Persson T, Tabellini G (2003) The Economic Effects of Constitutions. Massachusetts Institute of Technology Press, Cambridge MA Pickering A, Rajput S (2018) Inequality and the composition of taxes. Int Tax Public Finance, 25(4):1001–1028 Piketty T (2014) Capital in the Twenty-First Century. Harvard University Press, Cambridge MA Piketty T, Saez E, Zucman G (2018) Distributional national accounts: methods and estimates for the United States. Q J Econ 133(2):553–609 Piketty T, Yang L, Zucman G (2019) Capital accumulation, private property, and rising inequality in China, 1978–2015. Am Econ Rev 109(7):2469–96 Razin A, Sadka E, Swagel P (2002) The aging population and the size of the welfare state. J Political Econ 110(4):900–918 Roberts KW (1977) Voting over income tax schedules. J Publ Econ 8(3):329–340 Romer T (1975) Individual welfare, majority voting, and the properties of a linear income tax. J Publ Econ 4(2):163–185 Saez E, Zucman G (2016) Wealth inequality in the United States since 1913: evidence from capitalized income tax data. Q J Econ 131(2):519–578 Sanz I, Velázquez FJ (2007) The role of ageing in the growth of government and social welfare spending in the OECD. Eur J Polit Econ 23(4):917–931 Shelton CA (2007) The size and composition of government expenditure. J Publ Econ 91(11– 12):2230–2260 Shelton CA (2008) The aging population and the size of the welfare state: is there a puzzle? J Public Econ 92(3–4):647–651

Acknowledgements

The vague idea for this book can be traced back to my postgraduate study at the University of Birmingham during the period 2012–2013. It has been almost ten years since then. Ten years ago, I discussed the idea of economic development and income inequality with my classmate Liaoqi Yu. I then met with Prof. Peter Sinclair to discuss this childish idea. During our conversation he appreciated my rigorous academic attitude and encouraged me to continue my inequality research after graduation. He believed that the economic development and poverty reduction path deserved in-depth study. Thanks to his encouragement, I was brave enough to pursue deeper studies in economics. The contents of this book originated in my Ph.D. thesis compiled at the University of York during the period 2014–2018. Compared with the manuscript from four years ago, I have made a lot of revisions and updates to produce this version. During the writing process, I received help from many people, and I would like to express my gratitude to them. I want to thank my doctoral supervisors, Dr. Andrew Pickering and Dr. Paulo Santos Monteiro, for their comments. You are much appreciated for your continuous support of my Ph.D. study. All staff and students at the Department of Economics and Related Studies at the University of York contributed to my fantastic experience and great learning environment. I owe many thanks to Prof. Frank Cowell and Prof. Gulcin Ozkan, members of my examination committee, for their valuable comments and stimulating discussions. Part of this work and the idea for it came into being while I was studying at the University of Cambridge, and I am grateful for their generous support and hospitality, as well as the insightful advice from my mentor, Dr. Toke Aidt. Whenever I pick up this book, my mind will flash back to my Birmingham, York, and Cambridge days. My last remark is to express my sincere gratitude to all my co-authors who contributed to each chapter of the book, including Dr. Andrew Pickering, Dr. Paulo Santos Monteiro, Dr. Xiaoge Zhang, and Jingci Zhu. I extend the same appreciation to my former students at the University of York and the Central University of Finance and Economics. Without their warm support, the fruition of this volume would not have possible. I am also grateful to my friends Difu Liang, Zimin (Michael) Sun,

xix

xx

Acknowledgements

and Yanhao Wang for their insightful comments on economics and for their encouragement. I owe Emily Zhang, Shalini Selvam, and their colleagues at Springer my thanks as well because they encouraged me to write this book. I also want to thank the press’ anonymous reviewers for their comments. Finally, I want to express my deepest appreciation to my whole family, particularly my parents, who supported me unconditionally, as usual. Without their endorsement and support, my academic advancement would not have been possible.

Contents

Part I

Inequality and Fiscal Policy

1

Inequality and the Size of US State Government . . . . . . . . . . . . . . . . . 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1 Taxation Encouraging Effects of Inequality . . . . . . . . . . . 1.2.2 Taxation Damaging Effects of Inequality . . . . . . . . . . . . . 1.2.3 Recent Development in the Study of Inequality . . . . . . . 1.3 Data and Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.1 Baseline Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.2 Instrumental Variables Estimation . . . . . . . . . . . . . . . . . . . 1.4.3 Robustness and Heterogeneity . . . . . . . . . . . . . . . . . . . . . . 1.4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3 3 6 6 7 8 9 11 11 13 13 15 16 16

2

Inequality and Government Size: A Political Economy Theory and OECD Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Economic Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.2 The Median Voter’s Choice of Tax Policy . . . . . . . . . . . . 2.2.3 Capital Income Inequality and Redistribution . . . . . . . . . 2.3 Data and Econometric Specification . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.1 Panel Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.2 Instrumental Variables Estimation . . . . . . . . . . . . . . . . . . . 2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

19 19 21 22 25 26 29 35 35 37 40 42 44

xxi

xxii

3

Contents

Inequality and Government Debt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Data and Econometric Specification . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1 Baseline Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.2 Instrumental Variables Estimation . . . . . . . . . . . . . . . . . . . 3.3.3 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Part II 4

5

6

47 47 49 51 51 51 53 54 55

Demography and Fiscal Policy

Population Aging and the Composition of Taxes: A Political Economy Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Background and Related Literature . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 The Economic Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Political-Economy Equilibrium: The Choice of Tax Policy . . . . . 4.4.1 Income Taxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.2 Expenditure Taxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.3 The Composition of Taxes . . . . . . . . . . . . . . . . . . . . . . . . . 4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5.2 Further Research Directions . . . . . . . . . . . . . . . . . . . . . . . . Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

59 59 61 64 66 66 68 70 72 72 73 73 77

Population Aging and the Composition of Taxes: Evidence from International Panel Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Data and Econometric Specification . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.1 Baseline Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.2 Further Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

79 79 81 83 83 83 87 87

Youthful Dependents and the Composition of Taxes . . . . . . . . . . . . . . 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.1 Theoretical Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.2 Demographic Structure and Tax Preference . . . . . . . . . . . 6.3 Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.1 Data and Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.2 Baseline Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.3 Heterogeneous Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . .

89 89 91 91 93 95 95 96 98

Contents

7

xxiii

6.3.4 Income Taxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.5 Expenditure Taxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.6 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

98 100 103 104 104 106

Demography and Government Debt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.1 Effects of Population Aging . . . . . . . . . . . . . . . . . . . . . . . . 7.2.2 Public Debt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3 Data and Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.1 Baseline Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . 7.4.2 Further Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . . 7.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

109 109 111 111 112 113 115 115 116 119 119

Part III Short- and Medium-Term Perspective 8

Inequality and Economic Growth: A Literature Review . . . . . . . . . . . 8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2.1 Effects of Economic Growth on Inequality . . . . . . . . . . . 8.2.2 Effects of Inequality on Growth . . . . . . . . . . . . . . . . . . . . . 8.2.3 Effects of Inequality on Growth . . . . . . . . . . . . . . . . . . . . . 8.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

123 123 124 125 126 128 130 130

9

Inequality and Economic Growth in the Twenty-First Century . . . . 9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2.1 Economic Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2.2 Capital Income Inequality and Growth . . . . . . . . . . . . . . . 9.3 Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.3.1 Data and Econometric Specification . . . . . . . . . . . . . . . . . 9.3.2 Panel Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.3.3 Robustness and Further Estimation . . . . . . . . . . . . . . . . . . 9.3.4 Generalized Method of Moments Estimation . . . . . . . . . . 9.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

133 133 136 137 140 143 143 147 151 153 154 155 158

xxiv

Contents

10 Demography and Economic Growth: The Effect of Tax Composition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.1 Population Aging and Economic Growth . . . . . . . . . . . . . . . . . . . . 10.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.1.2 Data and Econometric Specification . . . . . . . . . . . . . . . . . 10.1.3 Baseline Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . 10.1.4 Further Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . . 10.1.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.2 Youthful Dependents and Economic Growth . . . . . . . . . . . . . . . . . 10.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.2.2 Data and Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.2.3 Baseline Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . 10.2.4 Instrumental Variables Estimation . . . . . . . . . . . . . . . . . . . 10.2.5 Further Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . . 10.2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

161 161 161 163 164 166 168 168 168 170 171 172 172 176 176

11 Tax Composition and Economic Growth in the Age of Demographic Change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.3 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.4.1 Contemporary Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.4.2 Dynamic Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.4.3 System Generalized Method of Moments Estimation . . . 11.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

179 179 182 183 187 187 188 190 192 195 197

Part IV Re-thinking of the Malthusian Trap 12 Demography and Income Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.2 Data and Econometric Specification . . . . . . . . . . . . . . . . . . . . . . . . . 12.3 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.3.1 Baseline Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.3.2 Further Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.3.3 Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

201 201 203 204 204 205 208 208 208

Contents

xxv

Appendix A: Labor Income Inequality, Taxation and Growth: A Political Economy Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 Appendix B: Capital Income Inequality, Taxation and Growth: A Political Economy Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 217 Appendix C: Alternative Measure of Inequality and Growth: Evidence from OECD Countries . . . . . . . . . . . . . . . . . . . . . . . . 227 Appendix D: Demography and Taxation: Further Global Evidence . . . . . 239

List of Figures

Fig. 1.1

Fig. 2.1 Fig. 2.2 Fig. 2.3 Fig. 2.4 Fig. 2.5 Fig. 2.6 Fig. 2.7 Fig. 3.1

Fig. 4.1 Fig. 4.2 Fig. 5.1

Fig. 7.1 Fig. 9.1

Scatter plot of average US state government tax revenue as a percentage of income and average income inequality by state. Notes Income inequality is defined as the ratio of mean to median household labor income . . . . . . . . . . . . . . . . . Capital income inequality and top 1% income share . . . . . . . . . . Capital income and productivity joint distribution (Assumption 2.2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Increase in capital income inequality . . . . . . . . . . . . . . . . . . . . . . . The size of government, 1960–2007 . . . . . . . . . . . . . . . . . . . . . . . Capital income inequality, 1960–2007 . . . . . . . . . . . . . . . . . . . . . Labor income inequality, 1960–2007 . . . . . . . . . . . . . . . . . . . . . . Labor income inequality and capital income inequality, 1960–2007 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Scatter plot of average central government debt as a percentage of GDP and average income inequality by country (using top 1 percent income share data) . . . . . . . . . . . Correlation between aging and personal income taxes (cross-country data, 1990–2020) . . . . . . . . . . . . . . . . . . . . . . . . . . Correlation between personal income taxes and expenditure taxes, 1990–2020 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Correlation between aging and income taxes (cross-country data over 1990–2014). Aging is defined as the ratio of the population above 65 years old to the population between 15 and 64 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Government expenditure from 1970 to 2018 . . . . . . . . . . . . . . . . . Scatter plot of income inequality derived from the University of Texas Inequality Project and the share of income received by the top one percent of the population derived from the World Inequality Database . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4 22 27 28 30 31 33 34

48 62 64

80 110

145

xxvii

xxviii

Fig. 9.2 Fig. 10.1

Fig. 10.2

Fig. 11.1 Fig. 11.2 Fig. 11.3 Fig. 12.1

Fig. C.1 Fig. C.2 Fig. D.1 Fig. D.2 Fig. D.3

List of Figures

Scatter plot of labor income inequality and capital income inequality derived from the Luxembourg income study . . . . . . . . Correlation between aging and growth (cross-country data over 1985–2014). Aging is defined as the ratio of the population above 65 years old to the population between 15 and 64 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Correlation between economic growth and the ratio of children to workers (cross-country five-year panel data over 1980–2014) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Population aging from 1980 to 2015 . . . . . . . . . . . . . . . . . . . . . . . Per capita gross domestic product and non-distortionary taxation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Marginal effect of non-distortionary taxes on economic growth for different levels of the aging measure . . . . . . . . . . . . . . Correlation between income inequality and aging (cross-country data over 1990–2014). Aging is defined as the ratio of the population above 65 years old to the population between 15 and 64 . . . . . . . . . . . . . . . . . . . . . . . Growth and capital income inequality, 1960–2010 . . . . . . . . . . . . Growth and labor income inequality, 1960–2010 . . . . . . . . . . . . . Correlation between aging and the composition of taxes, 1990–2014 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Correlation between income taxes and expenditure taxes, 1990–2014 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Income taxes in OECD and non-OECD countries, 1990–2014 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

146

162

169 182 185 192

202 228 229 240 242 242

List of Tables

Table 1.1 Table 1.2 Table 1.3 Table 1.4 Table 2.1 Table 2.2 Table 2.3 Table 2.4 Table 3.1 Table 3.2 Table 3.3 Table 3.4 Table 5.1 Table 5.2 Table 5.3 Table 6.1 Table 6.2 Table 6.3 Table 6.4 Table 6.5 Table 6.6 Table 7.1 Table 7.2 Table 7.3 Table 9.1 Table 9.2 Table 9.3 Table 9.4 Table 9.5

Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Basic estimation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Instrumental variables estimation results . . . . . . . . . . . . . . . . . . Robustness and extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Panel estimation results with fixed effects . . . . . . . . . . . . . . . . . Instrumental variables estimation results . . . . . . . . . . . . . . . . . . Instrumental variables estimation results . . . . . . . . . . . . . . . . . . Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Panel regressions of inequality on debt . . . . . . . . . . . . . . . . . . . . Instrumental variables estimation results . . . . . . . . . . . . . . . . . . Robustness and extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Basic estimation results—the composition of taxes . . . . . . . . . . Estimation results—the composition of taxes . . . . . . . . . . . . . . . Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Basic estimation results—the composition of taxes . . . . . . . . . . Panel estimation results with fixed effects—the composition of taxes . . . . . . . . . . . . . . . . . . . . . . . . Panel estimation results with fixed effects—income taxes . . . . Panel estimation results with fixed effects—expenditure taxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Robustness—the composition of taxes . . . . . . . . . . . . . . . . . . . . Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Basic estimation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Further estimation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Panel regressions of inequality on economic growth . . . . . . . . . Panel regressions of inequality on economic growth . . . . . . . . . Robustness and extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Difference generalized method of moments regressions . . . . . .

10 12 13 14 32 36 40 41 49 52 53 54 82 84 86 96 97 99 101 102 103 114 116 118 144 149 150 152 153 xxix

xxx

Table 9.6 Table 10.1 Table 10.2 Table 10.3 Table 10.4 Table 10.5 Table 10.6 Table 10.7 Table 11.1 Table 11.2 Table 11.3 Table 11.4

Table 11.5 Table 11.6 Table 11.7 Table 12.1 Table 12.2 Table 12.3 Table 12.4 Table C.1 Table C.2 Table C.3 Table C.4 Table D.1 Table D.2 Table D.3 Table D.4 Table D.5 Table D.6 Table D.7 Table D.8 Table D.9

List of Tables

Appendix table: descriptive statistics for labor income and capital income . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Basic estimation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Further estimation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Basic estimation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Instrumental variables estimation results . . . . . . . . . . . . . . . . . . Further estimation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Panel regressions of per capita gross domestic product growth on the composition of taxes . . . . . . . . . . . . . . . . . . . . . . . Panel regressions of per capita gross domestic product growth on the composition of taxes . . . . . . . . . . . . . . . . . . . . . . . System generalized method of moments regressions of per capita gross domestic product growth on the composition of taxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . List of countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Theoretical aggregation of the functional classifications . . . . . . Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Basic estimation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Further estimation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mechanism check . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Panel estimation results with fixed effects . . . . . . . . . . . . . . . . . Difference and system generalized method of moments regressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sensitivity analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Basic estimation results—the composition of taxes (annual data) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Estimation results—the composition of taxes (annual data) . . . Estimation results—the composition of taxes (annual data) . . . Estimation results—income taxes (annual data) . . . . . . . . . . . . . Estimation results—expenditure taxes (annual data) . . . . . . . . . Estimation results—the composition of taxes (cross-country data) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Estimation results—income taxes (cross-country data) . . . . . . . Estimation results—expenditure taxes (cross-country data) . . .

158 163 165 167 170 173 174 175 186 189 191

193 194 195 196 204 205 206 207 228 232 234 235 241 244 246 248 250 252 253 254 255

Part I

Inequality and Fiscal Policy

Part I of the text covers the analysis of the relationship between income inequality and fiscal policy. It briefly reviews the dynamic changes of income sources that contribute to inequality and builds upon a distinction between income inequality induced by differences in labor productivity and income inequality induced by differences in capital income.

Chapter 1

Inequality and the Size of US State Government

1.1 Introduction Does income inequality necessarily lead to a rise in the size of government? One seminal paper is a political economy model designed by Meltzer and Richard (1981), who argue that in a median voter framework increased income inequality indicates larger distance between mean and median income and therefore, leads to greater demand for redistribution. Empirical evidence generally has not supported the Meltzer and Richard (1981) hypothesis. For example, Fig. 1.1 depicts the raw correlation between state government tax revenue as a share of state GDP over 1976–1997 in the US and the ratio of mean to median household labor income, indicating no positive association between state government size and income inequality.1 Rising inequality in many developed countries has received renewed interest among policy-makers, academics, and the general public over the last decade, as shown by the attention generated by an academic book by Piketty (2014). Following up on Kuznets and Jenks (1953) pioneering study, a number of authors (Alvaredo et al. 2011) have constructed long-run series of top income shares to measure income inequality. For example, estimates from the World Inequality Database find that income concentration is high and growing in the United States: the richest 1% of households earned 20% of total income in 2015, up from 11% in 1978, while the bottom 50% experienced a complete collapse, from 20% to 12% of total income (Piketty et al. 2018). In contrast, and in spite of a similar trend, the top 1% share remains smaller than the bottom 50% in China but approaching the US level in 2015 (Piketty et al. 2019), and even less so in France as a typical representative of the West European pattern (Garbinti et al. 2017). 1

Perotti (1996), Persson and Tabellini (2005), Shelton (2007), and Luo et al. (2017) all find an insignificant or even negative relationship between the size of government and the degree of inequality at country-level. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 W. Luo, Inequality, Demography and Fiscal Policy, Applied Economics and Policy Studies, https://doi.org/10.1007/978-981-99-0518-8_1

3

1 Inequality and the Size of US State Government

State government tax revenue as a share of income 4 6 8 2

4

State Government Size Versus Inequality Minnesota New Mexico Mississippi South Carolina New York Massachusetts North Carolina Arkansas Kentucky California Arizona Iowa Georgia Indiana MarylandWashington Pennsylvania Jersey Ohio Virginia Illinois Oregon New Tennessee Missouri

Michig Alabama Louisiana

Colorado Florida

Texas

1

1.2 1.4 1.6 Ratio of mean to median labor income (mean) tratio

1.8

Fitted values

Fig. 1.1 Scatter plot of average US state government tax revenue as a percentage of income and average income inequality by state. Notes Income inequality is defined as the ratio of mean to median household labor income

Despite the global resurgence of income inequality, there is still much doubt and debate as to whether such inequality is desirable. This debate is not a new phenomenon, often highly contested, and in recent years has fallen under the category of ‘redistribution and growth’ debate. However, the seminal paper by Meltzer and Richard (1981), building on earlier research by Romer (1975) and Roberts (1977), offers a sanguine prediction: greater before-tax income inequality implies divergence between mean and median income and so, under universal suffrage, leads to increased redistribution. Democracy thus provides a corrective to increased inequality. However, evidence supporting the Meltzer and Richard (1981) hypothesis is generally weak. Perotti (1996), Benabou (1996), Persson and Tabellini (2005), and Shelton (2007) all find an insignificant or even negative relationship between the size of government and the degree of inequality. Much of the following literature has focused on the impact of income inequality on economic growth, typically based on Meltzer and Richard (1981). One important benchmark is an endogenous growth model by Persson and Tabellini (1994), who argue that if in a society the political decisions regarding redistribution generate economic policies that tax investment, then inequality should harm growth as it increases redistributive tax pressures. Empirical support for this hypothesis is generally weak. For example, Forbes (2000) finds that an increase in the level of income inequality

1.1 Introduction

5

in a country has a positive and significant relationship with subsequent growth rates in the short and medium term, by controlling for country-specific effects and period effects. However, in the original mechanism (Meltzer and Richard 1981; Persson and Tabellini 1994), labor is the sole channel of income and the rich have higher income by means of higher productivity (individual-specific skills, in other words). Nevertheless, in reality, labor is not the sole channel of income for the rich, as widely observed (see Piketty et al. 2018), and moreover, the labor share of income has consistently declined in recent years (see Azmat et al. 2012; Karabarbounis and Neiman 2014). Indeed, Piketty (2014) links increasing inequality to the declining labor share. Furthermore, capital income has recently become more unequal as well as more important. Kaymak and Poschke (2016), and Saez and Zucman (2016) document considerable rises in the concentration of wealth in the US over the past 50 years. Hence the current inequality-redistribution literature has one crucial omission: through focusing on the impacts of inequality and using the often assumed (aggregate) inequality solely induced by differences in labor productivity, the analysis is lacking a comprehensive consideration of capital income inequality. Hence, following Luo (2020) and Luo (2022), with the twist that income inequality is engendered from differences in capital income as well as differences in labor productivity, the purpose of this paper is to analyze how the sources of income inequality affect the size of US state government. The relationship between inequality and government size is investigated empirically using a panel of US states, including new measures of both capital and labor income inequality as additional explanatory variables. This paper constructs the measures of inequality making use of householdlevel income data from the Panel Study of Income Dynamics (PSID). Both meanlevel income and median-level income for these two types (labor income and capital income) can be constructed, which allows more direct measurement of these two kinds of inequality. As in Meltzer and Richard (1981), labor income inequality is equal to the ratio of mean to median labor income. This paper thus constructs this, and also a comparable measure for capital income inequality following Luo (2022). Our analysis makes use of the US state data and adopts the panel regression with fixed effects to document three main findings. First, an increase in capital income inequality is found to be negatively associated with government size. A one standard deviation rise in capital income inequality is statistically associated with a decline in state tax revenue of 0.14% of state GDP. Second, the results hold across various econometric specifications employed, such as utilizing five-year average data, instrumental variables estimation, and alternative measures of ideology. Third, controlling for capital income inequality yields a positive relationship between state government size and labor income inequality in states with lower income levels. Moreover, our results also show that capital income inequality plays a key role in poor states, whilst labor income inequality does in rich places regarding the explanation of the change in state government. We believe that this paper makes two crucial contributions to the current literature. First, this paper seeks to contribute to the empirical literature on the sources of income inequality. Our analysis examines the sources of income inequality theory within a

6

1 Inequality and the Size of US State Government

developed country, under the premise that the existing literature mainly focuses on the analysis of cross-country. Second, we show capital income inequality as one new explanation of the slowdown in the size of government in US, which sheds light on further research direction. In addition, the analysis of this paper can be applied in the design of tax system, given that government did raise revenue through labor income taxes 30 years ago and it might have reached the peak of Laffer curve. The rest of the paper is organized as follows. The next section introduces the existing literature. Section 1.3 describes the data and the methodology. Section 1.4 contains the estimation results, and Sect. 1.5 concludes.

1.2 Literature Review The debate of the relationship between inequality and the redistributive taxes appeared from 1980s. There are two main groups of this analysis could be identified: one group argues that greater inequality would generate greater redistribution and more distortionary taxation, while the other group claims that more equal nation could lead to higher redistribution or taxes.

1.2.1 Taxation Encouraging Effects of Inequality The most significant and oldest notion of taxation promoting effects of inequality could trace back to the study of Meltzer and Richard mentioned previously. On the theoretical side, the positive relationship between inequality and taxation has received attention of literatures. An interesting idea of this is what Persson and Tabellini (1994) have clarified in the democratic regimes. If some countries are initially high inequality, the median voter would be relatively poor in those economies and thus, this relatively less-income median voter would like to vote the higher redistributive taxes to be the optimal fiscal policy. This favor policy of relatively higher taxes could be ordinarily wasteful if they are not constructed in a suitable way, which would finally hurt growth in the future. Another key research of this argument is the work of Perotti (1996). In Perotti’s theoretical part, the component of political mechanism shows that higher income inequality would generate rising pressures for redistributive taxes. This will in turn discourage the motivations in the physical and human capital accumulation due to higher distortionary taxation. This is consistent with the implication pointed out by Bertola (1993). Based on the majority rule, the level of taxes and redistribution is the consequence of process of democratic voting. The pre-tax individual income mainly decides the preference of voter. Thus, a highly unequal democratic economy would clam more redistribution that is financed by the unattractive distortionary taxes. In terms of empirical side, it is meaningful to point out there are also plenty of studies trying to evaluate the positive relationship between income inequality and

1.2 Literature Review

7

redistribution analyzed in the theoretical way mentioned previously. As Milanovic (2000) has demonstrated, one proper test of the hypothesis of median voter taking redistribution into account would be proposed. Having included the required data of 24 democracies, the results of this research support the relation that economies with higher factor income inequality would have more redistribution, which is in line with implication of theoretical models discussed above. Additionally, Alesina and Rodrik (1994) have tried to endogenize public policy in an endogenous growth model in order to study the link between economic growth and political conflict. Both the theoretical and empirical part of their study presents that in a democratic economy the higher inequality the wealth distribution becomes, the higher the taxation rate is.

1.2.2 Taxation Damaging Effects of Inequality In contrast, it is valuable to point out that there also exist evidences, both in theoretical model and empirical analysis, though rare, supporting the idea that income inequality in an economy would have a negative effect on redistributive fiscal policy, in particular taxes. In other words, if an economy is more equal in income distribution, the fiscal redistribution or distortionary taxation would be greater. This runs in an opposite way to those predicted by early political economic models which indicate that more unequal democratic economies tend to distribute more, as having discussed above. It is interesting to note that the theoretical part of the work provided by Perotti (1996) aims to illustrating the argument that initially greater inequality in democratic economies tends to distribute more in the political mechanism, while its empirical section does not seem fairly supportive of its theoretical explanation by using the cross-nation data. As Perotti’s empirical contribution has shown, those distributive policies are normally correlated with inequality in such an opposite sign to its implication in theoretical part, which presents that if economies are initially more unequal, the fiscal redistribution would tend to be less, rather than greater. Further, one crucial and rare theoretical model is the work of Benabou (2000), which discusses the imperfect insurance and capital markets and analyzes how different agents vote for the distributive policies. Based on the framework of Benabou, the first relationship between distributive policy and inequality is U-shaped. This means that if the income inequality is relatively less, there is nearly tending to the efficient distributive policy, and as income inequality rises it would reduce the level of distributive policy firstly due to an increase in the proportion of rich agents, and then raise it again because of a great number of poor. The second relationship implied by Benabou is the advance of human capital accumulation. Having invested greater in human capital, the poor individuals would increase their relative income. This shows that inequality is a diminishing function of distributive rate. Taking these two into account, it can easily find that the distribution is a decreasing function of inequality, and two stable equilibriums could be identified. The first one is the thought of Welfare State, showing that low level of inequality while high level of government transfers; another one is the idea of Laissez-Faire, clarifying that more unequal links

8

1 Inequality and the Size of US State Government

with less level of distributive expenditure, which are the main ideas of the framework of Benabou. Moreover, there also exists empirical contribution supporting the implications of Benabou. As Muinelo and Roca (2013) have discussed the relationship between inequality and growth by the way of fiscal policy. By constructing a panel data of 21 upper income OECD countries during the year 1972–2006, their test demonstrates that the level of gross income inequality plays a crucial role in determining the outcomes of fiscal policy, and fiscal policy would also affect the growth and inequality in turn. It is meaningful to point out that there is a negative relationship between the initial gross income inequality and the fiscal policy, which means that higher inequality would connect with lower level of distribution, as the empirical results imply.

1.2.3 Recent Development in the Study of Inequality Given that capital income has become more unequal and more important in recent years (see Piketty 2014; Kaymak and Poschke 2016; Saez and Zucman 2016), a series of recent articles (i.e., Luo et al. 2017; Luo 2018, 2020, 2022) question and discuss the sources of income inequality in particular in an era of declining share of labor income. Following the series of literature working on the sources of income inequality, the argument proposed in this paper invokes a median voter model with the twist that income inequality is engendered from differences in capital income as well as differences in labor productivity.2 The key issue is that capital income is difficult to tax, whilst income from labor is taxable. When income differences are induced by capital income, the capacity of the median voter to redistribute through the tax system is restricted as the capital-rich supply less (taxable) labor. If capital income inequality rises such that the capital-rich supply less labor, then the demand for tax on labor declines as the capital-poor (median voter) cannot expropriate the rich. Thus, increased inequality in capital income leads to smaller government. What is the recent development in the study of inequality? And how does capital contribute to the dynamic of inequality? A recent series of literature argues that inequality has rised less than previously thought, particularly owing to a more modest rist of wealth and capital income at the top. Smith et al. (2021) hold that the increase of the top 0.1% wealth share is half as large as previous estimates. Smith et al. (2019) classify three-quarters of pass-through profit as human capital income, and thus indicate that labor income is more prevalent at the top income compared with the estimates in Piketty et al. (2018). Auten and Splinter (2018) state that the top 1% share of income has not rised at all on a post-tax basis from the 1960s onwards.

2

Please find a simple theoretical model in Chap. 2 in Luo (2018).

1.3 Data and Methodology

9

1.3 Data and Methodology The source of data for the dependent and control variables is Pickering and Rockey (2013), who provide comprehensive state-level policy and institutional data for the period over 1960–1997.3 Following their analysis the dependent variable is measured by total state taxes (i.e., the sum of sales, income and corporate taxes) per capita divided by state income per capita. The argument invoked in this paper emphasizes the median voter framework hence states in the US are the appropriate sample. Table 1.1 contains descriptive statistics of all the variables. The measure of income inequality used in traditional empirical analysis is an aggregate level of inequality. One common measure of inequality widely employed is the share of income received by the top 1% of the population, taken from the World Inequality Database. If the argument invoked in this paper is crucial, which implies that capital income inequality and labor income inequality may capture different influences on government size, especially at different levels of income, then arguably previous analysis has suffered from an omitted variable bias. The separate measures of inequality are preferable to the top income share and other frequently used aggregate measures of inequality, as they have incorporated all types of income within an index while capital income inequality has become more unequal as well as more important. The ideal measure of inequality, based on the logic in Meltzer and Richard (1981), is the ratio of mean to median income. As this paper argues, in the case of capital income, the greater this ratio the lower demands for redistribution and therefore, smaller government size. For most of the US states, it is possible to access household level micro-data from the Panel Study of Income Dynamics (PSID) that would allow more direct measurement of these two kinds of inequality (the advantage of split between labor and capital income), whilst the number of states and subsequent observations falls due to data availability.4 Following Meltzer and Richard (1981), labor income inequality is equal to the ratio of mean to median labor income (LabI neq). Table 1.1 contains statistics for this measure showing a mean value of 1.26, hence (as expected) mean labor income is generally greater than the median in the PSID data. The PSID data also give the opportunity to construct a measure of capital income inequality. Since the data of median capital income contains many relatively small values (mostly zero), I use the natural logarithm of the difference between mean and median capital income to measure capital income inequality (Cap I neq). As a result, the construction of capital income inequality differs slightly from that of 3

Note that Pickering and Rockey (2013) obtain the data from Besley and Case (2003). Note that, when I have less than 100 individuals matching the criteria for the construction of inequality measures in the PSID, I report the index as missing. Current data availability for labor and capital income precludes using the data before 1976.

4

10

1 Inequality and the Size of US State Government

Table 1.1 Descriptive statistics Obs t/y g/y CapIneq LabIneq Income FDLH FDUH DemGov DemBoth PolComp TranShare I deoc I deog,AD A/C O P E I deog,N O M

1056 1104 782 782 1104 1174 1173 1152 1450 1173 1450 1450 1450 1450

Mean

Std. Dev.

Min

Max

5.62 11.90 6.89 1.26 12.31 0.59 0.59 0.54 0.65 −0.04 −0.01 0.46 0.50 0.54

1.14 2.49 0.84 0.17 2.34 0.18 0.19 0.50 0.48 0.05 0.09 0.15 0.19 0.10

1.96 6.72 3.54 0.99 7.70 0.16 0.11 0 0 −0.25 −0.24 0.15 0.05 0.26

9.88 23.19 9.47 2.35 22.95 1 1 1 1 0 0.25 0.86 0.94 0.76

Notes The table gives descriptive statistics for the variables. t/y is the ratio of total state taxes to state income. g/y is the ratio of total state expenditure to state income. LabI neq is the ratio of mean to median household labor income, and Cap I neq is equal to the natural logarithm of the difference between mean and median household capital income—both data are constructed based on the Panel Study of Income Dynamics database. F DL H and F DU H are respectively the fraction of Democrats in the lower and upper houses. DemGov is an indicator variable set equal to one when the governor is a Democrat. Dem Both is an indicator variable set equal to one when the Democrats control both houses. PolComp is a measure of political competition that takes higher values when seat shares are more equal. T ranShar e is the total net value of transfers received as a share of state income. I deoc , I deog,AD A/C O P E and I deog,N O M are different measures of ideology, with higher numbers denoting greater ‘liberalism’ (i.e., increasingly left-wing ideology). All variables, except LabI neq and Cap I neq, are provided by Pickering and Rockey (2013)

labor income inequality. Table 1.1 show that the mean value this measure takes is 6.89, which serves to highlight the fact that most households do not receive any capital income, and shows that inequality in capital income is severe. Thus, the empirical analysis uses the inequality measure as the key explanatory variable. Following Pickering and Rockey (2013), the regression analysis also includes their control variables in the analysis of state government size. In particular I control for state income per capita in real terms, demographic composition, a set of political representation variables, percentage net value of transfers received, and alternative measures of ideology. The benchmark empirical specification is thus   t  = β1 I nequalit yi,t + xi,t  + αi + ηt + u i,t y i,t

(1.1)

1.4 Empirical Results

11

where i represents each state and t represents each time period, and u i,t is the error term. The left-hand-side variable, yt , is a measure of government size in state i in year t. The variable of interest is Inequality, including Cap I neq and LabI neq, as described above. The coefficient, β, hence indicates the impact of inequality on the size of state government. A positive and significant β suggests that income inequality exerts a positive effect on the state government size, whilst a negative and significant β implies that the inequality pushes the level of state redistribution lower. Control variables analyzed above are included in the vector xi,t . We also include yearspecific dummy variables, ηt , to control for shocks and trends that shape government size over time, and state-specific dummy variables, αi , to control for time-invariant, unobserved state characteristics that shape redistribution across states.

1.4 Empirical Results 1.4.1 Baseline Estimation This section is to test whether and how total state government tax revenue as a share of state GDP changes with income inequality in the presence of fixed state and year effects. Column 1 of Table 1.2 is a simple specification with just the measure of labor income inequality (LabI neq) as the only regressor using annual data OLS regression, with robust standard errors clustered by state. As mentioned above, column 2 then extends the regression to include two types of inequality, LabI neq and Cap I neq, to separate out labor and capital income. In this specification the sign of the coefficient estimate relating to capital income inequality is negative, and statistically significant at the 1% level. This is consistent with the argument - an increase in capital income inequality leads to smaller size of state government. Following Besley and Case (2003) the rest columns further augment state income per capita, demographic composition, and a set of political representation variables (i.e. the fraction of Democrats in the lower and upper houses, indicator if the governor is a Democrat, indicator if the Democrats control both houses, and a measure of political competition) as further controls. Moreover, in their analysis, one potential bias source may come from the fiscal transfers between states through the Federal government. Therefore, the total net value of transfers received is included as a further control. Following Pickering and Rockey (2013) I deoi is defined as the moving average of different ideology measures over the past ten years of data (where i = c, g AD A/C O P E, g N O M). As mentioned in Pickering and Rockey (2011), the effect of ideology on government size is conditional on income. Thus, column 3 further includes an ideology measure (I deoc ) and the interaction term (I ncome ∗ I deoc ) as well as the controls mentioned above. The results similarly demonstrate an increased tendency to reduce states’ taxes as capital income inequality rises, as predicted in this paper. The results for the specifications using I deog,AD A/C O P E or I deog,N O M as alternative measures of ideology are similar, with the estimated

12

1 Inequality and the Size of US State Government

Table 1.2 Basic estimation results (1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

−0.271*** (0.0642)

−0.168*** (0.0564)

−0.200*** (0.0583)

−0.189*** (0.0534)

−0.265* (0.139)

−0.334** (0.147)

−0.317** (0.138)

−0.325* (0.179)

−0.239 (0.194)

−0.402** (0.192)

−0.391** (0.187)

0.227 (0.446)

−0.310 (0.405)

−0.270 (0.420)

Income

−0.0830 (0.109)

−0.159 (0.132)

−0.0890 (0.187)

−0.0728 (0.143)

−0.107 (0.206)

−0.0192 (0.283)

FDLH

0.0417 (0.668)

−0.0758 (0.705)

−0.0668 (0.686)

−0.589 (1.210)

−0.927 (1.310)

−0.940 (1.288)

FDUH

−0.732 (0.571)

−0.534 (0.628)

−0.572 (0.622)

−0.776 (1.063)

−0.0565 (1.031)

−0.135 (1.041)

DemGov

0.0468 (0.0609)

0.0508 (0.0648)

0.0500 (0.0630)

−0.00245 (0.103)

−0.00626 (0.105)

−0.00189 (0.0994)

DemBoth

−0.0205 (0.0802)

−0.0482 (0.0834)

−0.0431 (0.0849)

0.0405 (0.223)

−0.0951 (0.221)

−0.0778 (0.228)

PolComp

0.471 (1.335)

−0.927 (1.691)

−1.093 (1.328)

−0.485 (1.777)

−3.078 (2.405)

−3.062 (1.828)

TranShare

−0.760 (0.623)

−0.828 (0.574)

−0.803 (0.603)

0.226 (1.072)

0.356 (0.969)

0.388 (1.093)

I deoi

−3.244 (2.052)

−0.670 (1.894)

0.392 (3.226)

−4.719* (2.672)

−0.0160 (2.653)

1.573 (4.825)

I ncome ∗ I deoi

−0.0609 (0.123)

0.0457 (0.132)

−0.0522 (0.247)

−0.0310 (0.169)

0.0172 (0.195)

−0.112 (0.377)

CapIneq LabIneq

−0.349* (0.185)

Observations 579

579

548

548

548

108

108

108

States

30

30

30

30

30

30

30

30

Data

Annual

Annual

Annual

Annual

Annual

5-year average

5-year average

5-year average

Estimation method

FE

FE

FE

FE

FE

FE

FE

FE

Year dummies R2

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

0.182

0.229

0.323

0.285

0.285

0.419

0.358

0.359

Notes Dependent variable is total state taxes measured as a percentage of state income. Estimations use panel regression with state fixed effects and robust standard errors clustered by state in parentheses. Year dummies are included in all regressions. Columns (3)–(5) extend column (2) to include I ncome, F DL H , F DU H , DemGov, Dem Both, PolComp, state population, and the proportions of children and senior citizens, T ranShar e, lagged ideology measures, I deoi , and their interaction with state income, I ncome ∗ I deoi , as additional control variables (where i = c, g AD A/C O P E, g N O M corresponding to each column). Columns (6)–(8) again test columns (3)–(5) using five-year average of the data instead of annual data. *, **, and *** respectively denote significance levels at 10, 5 and 1%

statistical significance of capital income inequality unaffected—remaining at the 1% level. Using the estimate from column 3 of Table 1.2, the estimated coefficient for capital income inequality is negative and the estimated relationship is sizable: a one standard deviation increase in capital income inequality is statistically associated with a fall in state tax revenue of 0.14% of state GDP. Results are also presented using five-year averages of the data, and the results essentially duplicate those found, establishing that the observed correlation is not driven by the cyclical features in the data.

1.4 Empirical Results

13

Table 1.3 Instrumental variables estimation results (1) (2) −0.406*** (0.131) LabIneq −0.332** (0.136) Observations 518 States 30 Data Annual Estimation method IV Year dummies Yes Ideology measured by I deoc L .Cap I neq Instruments 0.354*** (0.0420) Weak instruments F = 71.061 R2 0.893 CapIneq

−0.506*** (0.142) −0.473*** (0.141) 518 30 Annual IV Yes I deog,AD A/C O P E L .Cap I neq 0.345*** (0.0423) F = 66.466 0.883

(3) −0.464*** (0.135) −0.471*** (0.140) 518 30 Annual IV Yes I deog,N O M L .Cap I neq 0.357*** (0.0422) F = 71.383 0.884

Notes Instrumental variables regression of the ratio of total state taxes to state income on capital income inequality using L .Cap I neq as an instrument. All control variables, state fixed effects and year dummies are included in all regressions. Columns (1)–(3) respectively correspond to ideology measured by I deoc , I deog,AD A/C O P E and I deog,N O M . *, **, and *** respectively denote significance levels at 10, 5 and 1%

1.4.2 Instrumental Variables Estimation The empirical analysis shown above establishes a robust negative statistical association between state government size and capital income inequality in the presence of a substantial set of controls. However, the results presented do not establish causality, insofar that the movements in capital income inequality may be endogenous to the policy variable, or alternatively both variables co-move in response to an unseen third variable. To address this, in Table 1.3 I instrument for capital income inequality by employing the lag of capital income inequality. Analogous to columns 3–5 of Table 1.2, columns 1–3 of Table 1.3 respectively correspond to ideology measured by I deoc , I deog,AD A/C O P E and I deog,N O M and subsequent interaction effects. The hypothesis that these instruments are weak can be rejected given that the F-statistic of the first stage regression in all cases exceeds 65. Importantly the estimated coefficient for Cap I neq is still found to be negative and statistically significant.

1.4.3 Robustness and Heterogeneity Table 1.4 investigates the robustness and contains estimation results from fixed state and year effects. Since the size of government is moving slowly and the lack of

14 Table 1.4 Robustness and extensions Dep. Var. (1) (2) t/y t/y

1 Inequality and the Size of US State Government

(3) t/y

Panel A. Ideology measured by I deoc L .(t/y) 0.665*** 0.593*** (0.0406) (0.0407) CapIneq −0.101*** −0.112*** −0.125 (0.0347) (0.0284) (0.0949) LabIneq −0.214** −0.173* −0.549** (0.0988) (0.0976) (0.230) Observations 526 467 290 States 30 29 28 R2 0.625 0.392 Panel B. Ideology measured by I deog,AD A/C O P E L .(t/y) 0.685*** 0.619*** (0.0402) (0.0445) CapIneq −0.118*** −0.133*** −0.142 (0.0355) (0.0317) (0.0853) LabIneq −0.242** −0.186* −0.619*** (0.106) (0.0975) (0.212) Observations 526 467 290 States 30 29 28 R2 0.624 0.375 Panel C. Ideology measured by I deog,N O M L .(t/y) 0.686*** 0.622*** (0.0423) (0.0464) CapIneq −0.112*** −0.125*** −0.155* (0.0360) (0.0323) (0.0829) LabIneq −0.243** −0.196** −0.618*** (0.104) (0.0978) (0.214) Observations 526 467 290 States 30 29 28 R2 0.623 0.373 Data Full Full Higher income Estimation FE ArellanoFE method Bond Year dummies Yes Yes Yes

(4) t/y

(5) g/y

−0.159*** (0.0385) 0.289 (0.350) 258 26 0.500

−0.212 (0.178) −0.242 (0.263) 577 30 0.576

−0.165*** (0.0449) 0.157 (0.365) 258 26 0.519

−0.257 (0.166) −0.315 (0.255) 577 30 0.576

−0.138*** (0.0403) 0.147 (0.371) 258 26 0.493 Lower income

−0.236 (0.167) −0.312 (0.267) 577 30 0.572 Full

FE

FE

Yes

Yes

Notes Column (1) includes lagged dependent variable, L .(t/y). Column (2) contains Arellano-Bond estimation with lagged values of both the predetermined and endogenous variables as instruments. Columns (3) and (4) respectively correspond to higher and lower state income levels. Column (5) instead uses g/y as an alternative dependent variable. Panels A-C respectively correspond to ideology measured by I deoc , I deog,AD A/C O P E and I deog,N O M . *, **, and *** respectively denote significance levels at 10, 5 and 1%

1.4 Empirical Results

15

dynamics may result in biased estimates of standard errors and autocorrelated residuals, column 1 in panel A uses the same specification as column 3 of Table 1.2 but including the lagged dependent variable, confirming that state government size is highly persistent. Column 2 contains Arellano-Bond dynamic panel estimation results especially in the presence of possible endogeneity problems. Statistical significance in columns 1 and 2 implies that the estimates are stable across these specifications, which in turn supports the argument proposed. It is natural to see whether the results vary with the level of development. Columns 3 and 4 split the sample by economic development according to the median value of state income per capita. In column 3 the (relatively) high income sample again returns a negative coefficient for capital income inequality but with reduced statistical significance, whilst labor income inequality plays a significant role in this case. In column 4 the (relatively) low income sample also returns the same sign of the coefficient for Cap I neq. It is also noteworthy that the coefficient estimate for labor income inequality is now positive, though is not statistically significant. This is in line with the findings in Luo et al. (2017) who focus on the sample of OECD countries, which implies that the US is in general richer than the average level of OECD and as a consequent, the group of poor states may be comparable to them. The results reported so far are if anything strengthened when state government expenditure is employed instead as the measure of state government size. The last column presents the relevant results and shows that using the expenditure data the estimated coefficient for capital income inequality keeps negative, although the significance level falls. Panels B and C further use alternative measures of ideology and mirror the results found in panel A.

1.4.4 Summary In summary, it is worth enriching our results with a discussion on the research results in the context of the added value made by this paper in two aspects. First, in contrast to Meltzer and Richard (1981), Inequality induced by differences in capital income is found to be negatively associated with government size, when the US state data is utilized. Second, as for the sources of income inequality theory, the existing literature mainly focuses on the analysis of cross-country (i.e., Luo et al. 2017), while this paper conducts the analysis within a developed country and obtains different results. Luo et al. (2017) use OECD data and argue that government size and capital income inequality are found to be negatively related. Moreover, controlling for capital income inequality yields a positive and significant relationship between government size and labor income inequality. However, this paper shows the heterogeneous effects of income inequality in the US. Capital income inequality plays a significant role in poor states, whilst labor income inequality instead does in rich regions with respect to the explanation of the change in state government size.

16

1 Inequality and the Size of US State Government

1.5 Conclusion Utilizing the data on the US household income from the Panel Study of Income Dynamics database and the panel estimation with state fixed effects, this paper empirically examines how income inequality affects the size of US state government. We find three main findings. First, an increase in capital income inequality is found to be negatively associated with government size. Second, the results hold across various econometric specifications employed, such as instrumental variables estimation. Third, controlling for capital income inequality yields a positive relationship between state government size and labor income inequality in states with lower income levels. We believe our results contribute to the literature in two ways. First, our analysis examines the sources of income inequality theory within a developed country, under the premise that the existing literature mainly focuses on the analysis of cross-country. Second, we show capital income inequality as one new explanation of the slowdown in the size of government in US, which sheds light on further research direction. One implication for practice from our results is that the local government is advised to pay attention to the sources of income inequality (i.e., labor income inequality and capital income inequality) and their impact on the size of local government. In addition, the analysis of this paper can be applied in the design of tax system as well as the income taxation literature. The composition of taxes is always neglected by the government and researchers. Tax system is set ideally by economists, but politician cannot follow this ideal tax design. This is due to the risk of losing votes in politician competition. Government did raise revenue through labor income taxes 30 years ago. While it might probably have reached the peak of Laffer curve. This indicates that we need to look elsewhere, such as capital income taxes.

References Alesina A, Rodrik D (1994) Distributive politics and economic growth. Q J Econ 109(2):465–490 Alvaredo F, Atkinson A, Piketty T, Saez E, Zucman G (2011-present) World inequality database. Online at http://www.wid.world Auten G, Splinter D (2018) Income inequality in the United States: using tax data to measure long-term trends. Joint Committee on Taxation, Washington, DC Azmat G, Manning A, Reenen JV (2012) Privatization and the decline of labour’s share: international evidence from network industries. Economica 79(315):470–492 Benabou R (1996) Inequality and growth. Nat Bur Econ Res Macroecon Ann 11:11–74 Benabou R (2000) Unequal societies: income distribution and the social contract. Am Econ Rev 90(1):96–129 Bertola G (1993) Factor shares and savings in endogenous growth. Am Econ Rev 83(5):1184–1198 Besley T, Case A (2003) Political institutions and policy choices: evidence from the United States. J Econ Lit 41(1):7–73 Forbes KJ (2000) A reassessment of the relationship between inequality and growth. Am Econ Rev 90(4):869–887

References

17

Garbinti B, Goupille-Lebret J, Piketty T (2017) Inequality dynamics in France, 1900–2014: evidence from distributional national accounts (DINA). WID.world Working Paper Karabarbounis L, Neiman B (2014) The global decline of the labor share. Q Econ 129(1):61–103 Kaymak B, Poschke M (2016) The evolution of wealth inequality over half a century: the role of taxes, transfers and technology. J Monetary Econ 77:1–25 Kuznets S, Jenks E (1953) Shares of upper income groups in income and savings. In: Shares of upper income groups in income and savings. National Bureau of Economic Research, pp 171–218 Luo W (2018) Essays on inequality and fiscal policy. PhD thesis, University of York Luo W (2020) Inequality and government debt: evidence from OECD panel data. Econ Lett 186:108869 Luo W (2022) Inequality and growth in the twenty-first century. Scott J Polit Econ 69(4):345–366 Luo W, Pickering A, Monteiro PS (2017) Inequality and the size of government. Discussion Papers 17/02, Department of Economics, University of York Meltzer AH, Richard SF (1981) A rational theory of the size of government. J Polit Econ 89(5):914– 927 Milanovic B (2000) The median-voter hypothesis, income inequality, and income redistribution: an empirical test with the required data. Eur J Polit Econ 16(3):367–410 Muinelo-Gallo L, Roca-Sagalés O (2013) Joint determinants of fiscal policy, income inequality and economic growth. Econ Modell 30:814–824 Perotti R (1996) Growth, income distribution, and democracy: what the data say. J Econom Growth 1(2):149–187 Persson T, Tabellini G (1994) Is inequality harmful for growth? Am Econom Rev 84(3):600–621 Persson T, Tabellini G (2005) The economic effects of constitutions. MIT Press, Cambridge MA Pickering A, Rockey J (2011) Ideology and the growth of government. Rev of Econ Stat 93(3):907– 919 Pickering A, Rockey J (2013) Ideology and the size of US state government. Publ Choice 156(3– 4):443–465 Piketty T (2014) Capital in the Twenty-First Century. Harvard University Press, Cambridge MA Piketty T, Saez E, Zucman G (2018) Distributional national accounts: methods and estimates for the United States. Q J Econ 133(2):553–609 Piketty T, Yang L, Zucman G (2019) Capital accumulation, private property, and rising inequality in China, 1978–2015. Am Econ Rev 109(7):2469–96 Roberts KW (1977) Voting over income tax schedules. J Publ Econ 8(3):329–340 Romer T (1975) Individual welfare, majority voting, and the properties of a linear income tax. J Publ Econ 4(2):163–185 Saez E, Zucman G (2016) Wealth inequality in the United States since 1913: evidence from capitalized income tax data. Q J Econ 131(2):519–578 Shelton CA (2007) The size and composition of government expenditure. J Publ Econ 91(11– 12):2230–2260 Smith M, Yagan D, Zidar O, Zwick E (2019) Capitalists in the twenty-first century. Q J Econ 134(4):1675–1745 Smith M, Zidar OM, Zwick E (2021) Top wealth in America: new estimates and implications for taxing the rich (No. w29374). National Bureau of Economic Research

Chapter 2

Inequality and Government Size: A Political Economy Theory and OECD Evidence

This chapter is coauthored with Andrew Pickering and Paulo Santos Monteiro at Department of Economics and Related Studies, University of York.

2.1 Introduction Neoclassical models of democracy, as articulated by Meltzer and Richard (1981),1 offer a sanguine prediction: greater before-tax income inequality implies divergence between mean and median income and so, under universal suffrage, raises redistribution. Democracy, in principle, thus provides a corrective to increased inequality, and we should expect increased ex-ante inequality to lead to an increase in redistribution. However, evidence supporting the Meltzer and Richard (1981) hypothesis is generally weak. For example, the United States and other Anglo-Saxon countries have greater income inequality but lower public sector spending as a share of total GDP, while Scandinavian countries have relatively equal income distributions and a larger government spending share. Perotti (1996), Benabou (1996), Bassett et al. (1999) and Persson and Tabellini (2003) all find an insignificant or even negative link between the size of government and the degree of inequality.2 In response to this puzzle, new theoretical work has proposed mechanisms through which greater inequality levels can coexist with smaller government under democracy. For instance, Benabou (2000) identifies a functional role for the government to provide insurance (which implies redistribution) under capital market imperfections. The capacity for society to reach consensus on this role increases as the income distribution becomes more equal and risks become aligned and so government grows with equality. However, this type of mechanism also implies that government size should be positively correlated with economic growth and the evidence relating to 1

Romer (1975) and Roberts (1977) are important antecedents. More recent empirical literature (De Mello and Tiongson 2006; Shelton 2007; Muinelo and Roca 2013) is also unsupportive.

2

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 W. Luo, Inequality, Demography and Fiscal Policy, Applied Economics and Policy Studies, https://doi.org/10.1007/978-981-99-0518-8_2

19

20

2 Inequality and Government Size: A Political …

the so-called ‘Armey curve’ surveyed by Bergh and Henrekson (2011) if anything points to a negative relationship, at least for high income countries.3 The approach taken in this chapter instead revisits Meltzer and Richard (1981) more closely. In the original mechanism, labor is the only source of income and the rich earn more by dint of higher productivity. However, labor is not the only source of income for the rich and, moreover, the labor share of income has declined in recent years (see Azmat et al. 2012; Karabarbounis and Neiman 2014). Indeed, Piketty (2014) links rising inequality to the declining labor share: if the return to capital exceeds the rate of economic growth, then the capital share grows, and if ownership is concentrated within a small number of dynasties, then inequality inexorably increases. Furthermore, capital income has become more unequal as well as more important. Kaymak and Poschke (2016) document considerable increases in the concentration of wealth in the US over the past 50 years. Hence we instead ask how inequality stemming from capital income affects government size. Individuals differ in their capital endowment, with a right skewed capital income distribution. The majority of individuals are endowed with limited (or zero) assets or wealth and so are compelled to supply labor for their income, which is taxed. In contrast, if capital-income is not taxed then the capital-rich are relatively less exposed to taxation. In direct contrast to Meltzer and Richard (1981), the key result is that increased inequality in capital income leads to smaller government. When income differences are driven by capital income, the capacity of the median voter to redistribute through the tax system is reduced because the capitalrich supply less (taxable) labor. If capital income inequality increases such that the capital-rich supply less labor, then the preferred labor income tax rate falls because the (capital-poor) median voter cannot effectuate redistribution. Our work is related to Krusell and Rios (1999), who study a version of Meltzer and Richard’s model that includes inequality not only in labor income but also in wealth. However, we differ from Krusell and Rios (1999) as we assume capital income cannot be taxed, for the reasons explained below. The relationship between the size of government and inequality is investigated empirically using a panel of fifteen OECD countries, including a measure of capital income inequality as an additional explanatory variable. Direct measures of capital income inequality are not widely available.4 In the empirical work this is proxied by the top 1% total income share, taken from the World Wealth and Income Database 3

Other mechanisms are proposed by Persson (1995) and Rodriguez (2004). In the former, utility depends on relative consumption. In this model there is increasingly a problem of excessive labor supply in more equal societies and taxes work to increase utility by reducing labor. As in Benabou (2000), greater equality increases the capacity for agreement to tax, which again solves a market failure. Taxes work to eliminate the negative externalities associated with individual labor supply. Rodriguez (2004) instead models the political power of the rich as increasing with inequality, thereby reducing their obligation to pay tax. The democratic constraint is therefore undermined. 4 Limited capital income data indicates that the rich do hide their income from capital, and in other words, it is difficult to tax their capital income. This in turn implies the lack of capital income tax (and data) that the median voter can effectuate, consistent with OECD evidence (extremely small size of capital income taxation), as well as the lack of capital income inequality data.

2.2 The Model

21

(WID).5 A theoretical justification for this approach is Piketty (2014), wherein capital is disproportionately owned by a small number of dynasties. In this analysis the larger top income share stems from increasing capital income with fixed capital ownership. Certainly capital income represents an important component of the income of the top 1%. Frydman and Saks (2010) document the increasing importance of stock options and long-term bonuses (also in the form of capital payments) in the remuneration of executives in large publicly traded corporations in the US. Examination of disaggregated capital income data for a subset of countries provides empirical justification for this proxy. The WID contains non-wage (i.e. capital) income data for the top 1% and the top 10% for Australia, Canada, France and the United States. We posit that the higher the ratio of the share of non-wage income going to the top 1% relative to the top 10% the more unequal the capital income distribution. Ideally given our theory we would require that the numerator and denominator would respectively be the mean and 50th percentile non-wage income, but such data are not available. Nonetheless it seems plausible that inequality between the top 1% and the top 10% would be correlated with the theoretical ideal. Figure 2.1 plots this measure of capital income inequality together with the top 1% income share for these countries. In all four cases there is a strong correspondence between the direct measure of capital income inequality and the top income share, giving some credence to using the latter to proxy for the former for the wider sample of countries. The empirical analysis below also separately employs specific measures of productivity-induced labor income inequality as distinct from capital income inequality. As we discuss below the two measures are empirically as well as conceptually distinct from one another. Consistent with our theory, the size of government is negatively associated with capital income inequality. A one standard deviation increase in capital income inequality leads to a reduction in the size of government of around 2.6% of GDP. The negative relationship holds up when the lagged dependent variable is controlled for, and also when capital income inequality is instrumented with measures of technological progress and capital market access. We also find that once capital income inequality is controlled for, then the impact of labor income inequality becomes positive, consistent with Meltzer and Richard (1981) and in contrast to the voluminous empirical work testing their hypothesis. The next section theoretically analyzes how the size of government changes with capital income inequality. Section 2.3 contains the empirical work, and Sect. 2.4 concludes.

2.2 The Model As in Meltzer and Richard (1981), individuals, indexed by i, have preferences defined over consumption ci and leisure li , represented by a strictly concave, continuous and 5

The 0.1% income share could alternatively be used, though the results are very similiar because the correlation between the 0.1 and 1% income shares is around 0.98.

2 Inequality and Government Size: A Political …

Canada

France

United States

5 20 15 10 5

Top 1% income share

10

15

.1 .15 .2 .25 .3

20

Australia

.1 .15 .2 .25 .3

capital income inequality

22

2020

2000

1980

1960

1940

2020

2000

1980

1960

1940

year capital income inequality

Top 1% income share

Graphs by Country

Fig. 2.1 Capital income inequality and top 1% income share

twice-differentiable utility function, u i (ci , li ). Consumption and leisure are both normal goods. Following the original, we first analyze the equilibrium behavior conditional on a given tax policy and then address the tax policy choice itself.6

2.2.1 Economic Environment Income may be derived from both labor and capital. All individuals possess a unit of time to allocate to labor n i , or leisure li = 1 − n i . Individual labor income yi = xi n i depends on productivity, xi , as well as hours worked, and is taxed at a linear rate t. Capital income varies exogenously across individuals and is denoted by Ri .7

6

In order to compare with the Meltzer and Richard (1981) model, we start with a static model and focus on the tax policy choice generated by different sources of income inequality, rather than over-generation pension wealth decision. 7 Capital income analyzed throughout this chapter is the income with zero opportunity cost, such as rental income. Housing price in large cities has consistently increased in recent years and has been accumulated as high levels of housing wealth. Higher levels of housing wealth do lead to larger inequality in wealth, while it cannot be taxed until it is sold (in the case of rental income, it can be claimed as smaller size or other items to avoid tax).

2.2 The Model

23

Following Meltzer and Richard (1981), consumption is also financed by lump-sum redistribution, r , common to all individuals, hence: ci = (1 − t) xi n i + Ri + r.

(2.1)

To clarify the argument, capital income is assumed to be untaxed.8 In practice it is often more difficult to raise taxes on capital than on labor. Capital is often highly mobile internationally, whilst labor is not, and given this Diamond and Mirrlees (1971) show that small open economies should not tax capital income. Whilst in practice capital income taxation rates are positive, Gordon et al. (2004) observe they are lower than average labor income taxes in most countries. Moreover, capital income taxation liability can be reduced or even avoided altogether due to various loopholes including differential rates for different types of capital income (thereby enabling arbitrage opportunities) and the fact that interest payments are often tax-deductible. Indeed Gordon and Slemrod (1988), using US tax return data from 1983, estimated that the tax revenue loss from eliminating capital income taxation completely would be zero, hence that the tax burden on capital was effectively non-existent. Conceivably the perceived deadweight and/or capital flight losses from increasing capital income taxation nullify it as an instrument.9 Thus we focus on the choice of the labor income tax. Each individual chooses labor supply so as to maximize:   u i (ci , li ) = u i (1 − t) xi n i + Ri + r, 1 − n i .

(2.2)

The first-order condition is: (1 − t) xu c − u l = 0,

(2.3)

which determines the labor supply, n [(1 − t) x, R, r ], for those who wish to work.10 Since leisure is a normal good, we have that: ∂n (1 − t) xu cc − u cl =− < 0, ∂R D 8

(2.4)

The results would all still stand if we instead modeled capital income taxation as fixed (and unresponsive to inequality), as observed from OECD data. The difficulty to collect capital income tax also underpins this argument. The rich are anti-tax: it could be easily observed that large companies always try to reclassify their labor-capital income in order to find tax haven. In rich economies with low self-employment, tax evasion is small on aggregate but high at the top, strong gradient within top 1% (Alstadsæter et al. 2018). 9 Deadweight loss as well as capital flight loss leads to a loss of function of capital income taxation, which constraines the ability of median voter to influence over capital income tax rates. 10 For simplicity (but without loss of generality) we henceforth assume that the joint distribution of x and R is such that n i > 0 for all i, so that everyone supplies a strictly positive amount of market work.

24

2 Inequality and Government Size: A Political …

with D = [(1 − t) x]2 u cc − 2 (1 − t) xu cl + u ll < 0, given the assumption that u is strictly concave. Similarly, since consumption is a normal good we have that: ∂n ∂c u cl x (1 − t) − u ll =1+ > 0, (1 − t) x = − ∂R ∂R D

(2.5)

a condition which imposes additional restrictions on u cl . Hence, all else equal, people who are relatively capital-rich supply less labor and enjoy higher consumption. There are two sources of heterogeneity that determine differences in before-tax labor income. Firstly productivity, as analyzed by Meltzer and Richard (1981), and secondly capital income endowments. At the individual level increases in productivity will all else equal increase labor income.11 On the other hand increases in capital income will all else equal reduce the labor supply and, therefore, labor income. This underpins their proclivity towards taxation of labor income. Average labor income can thus be written by integrating: ∞ ∞ y¯ = 0

  xn R, r, (1 − t) x f (x, R) dxd R.

(2.8)

0

where f (x, R) is the joint density function of x and R. Individual productivity and capital endowments conceivably are correlated with each other to some extent: if, for example, high productivity individuals simultaneously enjoy high capital income. Finally, the government’s balanced budget requirement (in per capita terms) is given by: t y¯ = r. (2.9) Note that analogous to (2.4), we have: ∂n (1 − t) xu cc − u cl =− < 0. ∂r D

11

(2.10)

Notice that, as in Meltzer and Richard (1981), the sign of ∂n (1 − t) u c + (1 − t)2 xnu cc − (1 − t) nu cl =− ∂x D

(2.6)

is indeterminate. Hence, the labor supply could be backward bending as productivity increases. Still, pre-tax labor income may never decline following an increase in productivity. To see this notice that, for any individual earning positive labor income, we have ∂y ∂n =n+x ∂x ∂x (1 − t) xu c + n [u cl (1 − t) x − u ll ] =− > 0, D which must be positive given condition (2.5).

(2.7)

2.2 The Model

25

Hence for given productivity and capital income endowment, individual labor supply falls with increased redistribution. Therefore: ∂ y¯ = ∂r

∞ ∞ x 0

0

∂n f (x, R) dxd R < 0. ∂r

(2.11)

This establishes that the left-hand side of (2.9) is strictly decreasing with r . Moreover, t y¯ is non-negative and bounded above by t x, ¯ where x¯ is average productivity. In turn, the right-hand side of (2.9) is strictly increasing with r . Thus, there is a unique value of r to satisfy (2.9) for any t.

2.2.2 The Median Voter’s Choice of Tax Policy We now turn to the policy-setting decision. Crucially, the median voter is still a Condorcet winner even though the electorate is heterogeneous on two dimensions. The logic of this is that the preferred tax rate remains a monotonic function of the labor income alone, regardless of the underlying determinants of that labor income. Hence high labor income (whether induced by either high productivity or low capital income) will engender aversion to taxes, whilst low labor income (whether induced by low productivity or a generous capital income inheritance) will engender support for tax-financed redistribution. Formally, the median voter, m, is denoted as the owner of the median labor income. She sets taxes to maximize utility subject to the budget constraint (2.2), the government budget constraint (2.9), and a rational anticipation of how taxation will affect the incentives to supply labor in the economy. The first-order condition for the median voter with respect to the tax rate is:  y¯ − y m + t

d y¯ dt

 = 0,

(2.12)

where y m is the labor income of the median voter. Condition (2.12) yields the following solution for the tax rate chosen by the median voter t=

m − 1 + ηr , m − 1 + ηr + mητ

(2.13)

with ηr < 0 and ητ > 0 the partial elasticities of average income (assumed constant, as in Meltzer and Richard (1981)), and m = y¯ /y m .12 The key insight of Meltzer and Richard (1981) is that an increase in labor income inequality raises taxation, since an increase in income inequality raises m and from (2.13) we have that 12

Details are available in the Appendix.

26

2 Inequality and Government Size: A Political …

dt > 0. dm

(2.14)

Finally, although we impose almost no restrictions on the joint distribution f (x, R), we wish to guarantee that: (i) the chosen tax rate is positive; and that (ii) the individuals that are in the top of the capital income distribution are never the decisive voter. Thus, in the sequel we make the following two (empirically supported) assumptions: Assumption 2.1 The joint distribution f (x, R) is such that the labor income distribution is right-skewed. Thus, y m < y¯ and the chosen tax rate is positive. From (2.12) we see that Assumption 2.1 guarantees that the chosen tax rate is positive. Assumption 2.2 The joint distribution f (x, R) is such that the set of individuals i ∈ K with capital income Ri (the top K% of the capital income distribution) has productivity xi which is sufficiently high so that yi = xi n i > y m for all i ∈ K. Figure 2.2 illustrates the condition imposed by Assumption 2.2. The locus denoted y = y m represents productivity and capital income pairs, (x, R), for which labor income y is equal to the median voter’s labor income, y m . To the right of this locus, y > y m , since ∂∂ xy > 0 and ∂∂ Ry < 0. The dashed line denoted Q(1−K) % represents the (1 − K) %–quantile of the capital income marginal density function.13 Assumption 2.2 is a condition requiring that the set K of all individuals with capital income above Q(1−K) % is located to the right of the locus y = y m , as shown in Fig. 2.2.14

2.2.3 Capital Income Inequality and Redistribution We are interested in the consequences of higher capital income inequality. To study this issue we consider an increase in the capital income earned by the individuals in the set K of all individuals with capital income above Q(1−K) % . This is represented in Fig. 2.3: the individuals in the set K that correspond to the original individuals in the top K% of the capital income distribution receive an exogenous increase in capital income; thus, the set K shifts upwards in the space (x, R), but still satisfying the restriction imposed by Assumption 2.2, that guarantees that the median voter does not belong to the members of the set K (the new set is represented by the triangle 13

The size of the shaded area depends on the size of the set K (if we choose a larger set K, then the shaded area will be larger). The position of this shaded area indicates initial levels of capital income that individuals in the set K have (and we will discuss below regarding to the consequence of increased capital income inequality). We focus on the 99% percentile because in the empirical section that follows we use the income share of the top 1% as our measure of capital income inequality. 14 Consistent with Assumption 2.2, Atkinson and Lakner (2013) found that in the United States the tax units at the top of the labor income distribution are more likely to also be at the top of the capital income distribution.

2.2 The Model

27

Fig. 2.2 Capital income and productivity joint distribution (Assumption 2.2)

above, in Fig. 2.3). Notice that this experiment constitutes an increase in capital income inequality, since we maintain the capital income of all the other individuals unchanged and, hence, the capital income share of the top K% is increased.15 Under a right-skewed labor income distribution y m < y¯ , and given (2.13) above then t > 0. As with Meltzer and Richard (1981) demand for redistribution stems from changes in the labor income distribution. However, the labor income distribution may now change depending on the distribution of capital income as well as the productivity distribution. To see the consequences of higher capital income inequality, notice that all the individuals in the set K will choose to work less, because they enjoy an increase in their capital income and leisure is a normal good. This will tend to lower the average labor income y¯ , since we have that y¯ = p (K) y¯ (K) + (1 − p (K)) y¯ (∼ K) ,

(2.15)

where y¯ (K) denotes the average labor income of the individuals in the set K, y¯ (∼ K) denotes the average labor income of the individuals not in the set K, and p (K) is the probability measure of the set of individuals K. Notice that Assumption 2.2 guarantees that y¯ (K) > y m . On the other hand, the reduction in y¯ implies that the individuals not in the set K will receive fewer transfers and, therefore, work more. From Assumption 2.2, the individual earning the median labor income is not in the set K and, thus, y m will 15

It is not, however, a mean preserving spread in capital income. But lowering the capital income of the bottom Q(1−K) % capital income earners in order to preserve the mean capital income would only reinforce our results.

28

2 Inequality and Government Size: A Political …

Fig. 2.3 Increase in capital income inequality

increase. The upshot is that m = y¯ /y m is decreased. Hence, the effect of the increase in capital income going to the top capital-income recipients is to reduce the gap between taxable mean and median labor income. Hence an increase in overall income dt > 0, it inequality can coexist with a reduction in labor income inequality. Since dm follows that an increase in capital income inequality unambiguously lowers the tax rate chosen. Proposition 2.1 Suppose the top capital-income recipients are sufficiently productive that they also earn labor income above the median labor income (Assumption 2.2), and consider an increase in capital-income inequality represented by an increase in the capital income earned by the top capital-income recipients. Then the labor income tax rate t falls as capital income inequality rises. The proof of Proposition 2.1 is in the Appendix. In direct contrast to Meltzer and Richard (1981) government size diminishes with increased capital income inequality. If inequality increases such that the share of capital income going to the top income recipients increases, then the preferred tax rate falls because the (capital) rich are supplying less taxable labor income and hence the capacity of the median voter to redistribute is reduced. The intuition here is that if capital income is highly concentrated within a small group of top (capital) rich, then these top rich individuals who are also with high productivity work less (and avoid being taxed their capital income through various ways). The aggregate tax revenue generated from income from labor (which is hard to escape) will be smaller, and therefore the aggregate level of redistribution is reduced. The key issue is the extent to which the median voter can effectively redistribute through the tax system. As discussed above there are good reasons to believe that taxation of relatively mobile capital is considerably more difficult than taxation of

2.3 Data and Econometric Specification

29

labor income. If the rich are rich primarily due to capital income, perhaps because of the rising capital share, and perhaps due to successful reclassification of their income streams, then the capacity of the median voter to redistribute is curtailed. Moreover if rising inequality translates into further reductions in the supply of taxable labor then it follows that the demand for redistribution will fall.

2.3 Data and Econometric Specification The empirical analysis examines a panel of fifteen OECD countries over the period 1960–2007.16 Following Pickering and Rockey (2011) and Facchini et al. (2017), the dependent variable is total government outlays as a percentage share of GDP, extracted from the OECD Economic Outlook database. Figure 2.4 depicts these data, showing all countries experienced an upward trend in the earlier years followed by a period of stasis or even slight decline since around 1990. Table 2.1 contains descriptive statistics of all the variables used in the analysis. Figure 2.5 depicts the top income share data for all 15 countries. Note that the increases in the top income share to some extent coincides with the reversal of the growth of government noted above. Clearly there are interesting differences across the countries, for instance stronger recent increases in the English-speaking countries as discussed by Piketty and Saez (2006). The argument advanced in this chapter is the following: as the top income share increases, the supply of taxable labor of the rich falls, and hence support for taxation of labor income falls. As noted above previous empirical literature has generally been unsupportive of the original Meltzer and Richard (1981) hypothesis. If the mechanism put forward in the present chapter is important, and capital-income inequality and productivity differences are correlated with each other, then arguably previous analyses have suffered from an omitted variable bias. A measure of productivity heterogeneity is thus also included in the empirical analysis. This measure is taken from the University of Texas Inequality Project’s Estimated Household Income Inequality data.17 These data (denoted by U T I P) use Theil’s T statistic—measured across sectors within each country—to estimate wage inequality. Assuming competitive labor markets, then wage inequality should be capturing underlying heterogeneity in productivity.18 16

Specifically the countries included are Australia, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Japan, the Netherlands, Norway, Spain, Sweden, the United Kingdom, and the United States. Current data availability for the top income share precludes using other countries. The sample ends in 2007 due to the substantial toll on government outlays in many countries following the global financial crisis. 17 See Galbraith and Kum (2005). 18 To utilize Theil’s T statistic—measured across sectors within each country—shows the evolution of economic inequality. We can do this with many different data sets, including at the regional or provincial level. With the U T I P data, we can review changes in global inequality both across countries and through time. Nothing comparable can be done with previous data set (i.e. Deininger and Squire), for the measurements are too sparse and too inconsistent.

30

Fig. 2.4 The size of government, 1960–2007

2 Inequality and Government Size: A Political …

2.3 Data and Econometric Specification

Fig. 2.5 Capital income inequality, 1960–2007

31

32

2 Inequality and Government Size: A Political …

Table 2.1 Descriptive statistics Obs Mean OUTLAYS TOPINC UTIP SHARE FP ln(y) IDEO PROP1564 PROP65 TRADE YGAP OIL_EX OIL_IM INTERNET KAOPEN

720 617 625 548 622 720 683 720 720 710 720 720 720 720 564

41.50 7.89 34.61 67.97 57.73 2.94 0.043 65.20 12.79 53.58 0.026 4.60 15.06 10.91 1.41

Std. dev.

Min

Max

10.14 2.44 3.25 6.03 12.94 0.439 0.117 2.62 2.84 28.40 1.34 12.05 15.86 22.50 1.22

12.8 3.49 27.42 44.74 27.96 1.57 −0.266 57.63 5.73 8.93 −4.75 0 0 0 −1.89

72.4 18.33 43.16 82.10 82.46 3.93 0.337 69.89 21.02 178.25 5.83 72.36 72.36 87.76 2.39

Notes OU T L AY S denotes total government outlays as a percentage of GDP—taken from the OECD Economic Outlook database. T O P I N C is the top 1% income share—taken from the WID. U T I P is the University of Texas Inequality Project’s Estimated Household Income Inequality. S H A R E is the business sector labor share—taken from the OECD database. F P is the female labor force as a percentage of the female population between 15 and 64—also taken from the OECD database. y is real GDP per capita in $000s of 2005 prices—taken from the Penn World Tables. I D E O is ideology used in Pickering and Rockey (2011). P R O P1564 and P R O P65 are respectively the proportion of the population aged between 15 and 64, and 65 and above—taken from WDI database. T R AD E is the sum of exports and imports as a percentage of GDP. Y G A P is the difference between the actual output and its trend value in percentage—also taken from WDI database. O I L_E X and O I L_I M are respectively the oil price times a dummy variable equal to 1 if net exports of oil are positive; and the oil price times a dummy variable equal to 1 if net exports of oil are negative—taken from US Energy Information Administration. I N T E R N E T is the number of internet users per 100 people—also taken from WDI database. K AO P E N is the Chinn and Ito (2006) index for financial openness

Figure 2.6 depicts these data, which also exhibit increases in recent years, varying across countries. This measure is thus close to Meltzer and Richard (1981) original conception of the driver of the demand for redistribution—productivity-based inequality. A natural objection here is that the top income share will also be picking up productivity-induced inequality. Inevitably there is a correlation between productivity inequality as measured by U T I P and the income share of the top 1%, but this is somewhat weaker than might be expected. Figure 2.7 depicts a scatter plot of the two series, exhibiting a correlation coefficient of around 0.53. Hence there is meaningful separate information in the two series. Our argument is that the top income share is especially informative about capital income inequality rather than productivity-

2.3 Data and Econometric Specification

Fig. 2.6 Labor income inequality, 1960–2007

33

34

2 Inequality and Government Size: A Political …

Fig. 2.7 Labor income inequality and capital income inequality, 1960–2007

induced labor income inequality. The small sample of countries depicted in Fig. 2.1 discussed in the introduction lends some credence to this argument. The analysis includes control variables following Facchini et al. (2017). Controls include the natural logarithm of GDP per capita in constant chained PPP US$ (ln(y)), taken from the Penn World Tables (e.g. see Ram (1987)). Ideology (denoted I D E O) and its interaction with income (denoted I N T E R AC T ) as used in Pickering and Rockey (2011), are also included as standard. Following Facchini et al. (2017) the labor share of income (denoted S H A R E) from the OECD database is also included to capture (falling) cost-push effects. Following Kau and Rubin (2002) and Winer et al. (2008) female participation (F P) in the labor force is also included. Further controls follow Persson and Tabellini (2003). Demographic effects are encapsulated in the percentage of the population between 15 and 64 years of age and the percentage over the age of 65 (denoted P R O P1564 and P R O P65), taken from the World Development Indicators (WDI) database. Following Rodrik (1998) the trade share (the sum of exports and imports as a percentage of GDP, denoted T R AD E) is also employed in the regression analysis.19 Total government outlays in OECD countries vary counter-cyclically. There may also be cyclical movements in inequality. To address this potential problem the regres19

The results would all still stand if we ideally incorporate polity variables as one further control variable. The reason why we omit here is to easily compare with the template work by Facchini et al. (2017), and that the sample countries we have are all with high democracy scores.

2.4 Evidence

35

sion analysis employs the Persson and Tabellini (2003) cyclical control variables— the output gap (denoted Y G A P) and oil price effects (depending on whether or not the country is a net oil-exporter or importer, denoted O I L_E X and O I L_I M) are also included in the analysis when annual data are used. To summarize, the frist approach to estimate the effect of inequality on total government outlays is to consider the following econometric model:   + αi + ηt + u i,t OU T L AY Si,t = β1 T O P I N Ci,t + β2 U T I Pi,t + xi,t

(2.16)

where i represents each country and t represents each time period, all control variables analyzed above are included in the vector xi,t , αi are country dummies, ηt are period dummies, and u i,t is the error term.

2.4 Evidence 2.4.1 Panel Estimation Table 2.2 contains estimation results from fixed-effects panel regressions with total outlays as a percentage of GDP as the dependent variable. Column 1a represents the current consensus, augmenting the benchmark specification in Facchini et al. (2017) with productivity-induced inequality (U T I P), and finding it to be highly insignificant. This insignificance coheres with the findings in Perotti (1996), Persson and Tabellini (2003), De Mello and Tiongson (2006), and Shelton (2007). Column 1b further augments this specification with capital income inequality. The estimated coefficient for capital income inequality is negative, with a p-value of 1.7% and the estimated relationship is sizable: A one standard deviation increase in capital income inequality is statistically associated with government size which is smaller by 2.63% of GDP, consistent with the theoretical reasoning given here. It is also noteworthy that the coefficient estimate for productivity-induced labor income inequality increases substantially, though is still not statistically significant. Following Facchini et al. (2017) results are also presented (in columns 2a and 2b) using five-year averages of the data, and the results essentially duplicate those in column 1, establishing that the observed correlation is not caused by the cyclical features in the data. Column 3 of Table 2.2 contains Arellano-Bond dynamic panel estimation results extending the specification used in column 2 to include the lagged dependent variable (L .OU T L AY S). Here the negative relationship between government size and capital income inequality holds up, and indeed the coefficient estimate pertaining to labor income inequality is now positive, consistent with the Meltzer and Richard (1981) hypothesis, and significantly different from zero at the 5% level. This evidence suggests that previous tests of the Meltzer and Richard (1981) hypothesis were hampered by the conflation of capital and labor income inequality.

36

2 Inequality and Government Size: A Political …

Table 2.2 Panel estimation results with fixed effects (1a) (1b) (2a)

(2b)

(3)

0.132 (0.496) 0.695*** (0.200) −0.055 (0.221) −6.217 (4.501) −58.428** (23.033) 1.448* (0.805) 0.885 (0.579) 1.926*** (0.604) −0.031 (0.054)

−1.134*** (0.367) 0.497 (0.408) 0.522*** (0.163) −0.096 (0.257) 2.342 (4.222) −34.991 (24.511) 0.388 (0.863) 0.549 (0.591) 1.318** (0.572) −0.087 (0.053)

0.423*** (0.096) −0.632* (0.361) 0.932** (0.375) 0.696*** (0.164) 0.141 (0.112) −0.641 (4.419) −7.101 (19.769) 0.145 (0.723) 0.168 (0.358) 0.160 (0.388) −0.057 (0.065)

113 15 5-year averages 0.41

113 15 5-year averages 0.47

L.OUTLAYS

0.139 (0.422) SHARE 0.473*** (0.155) FP −0.064 (0.188) ln(y) −7.484* (3.771) IDEO −53.942** (22.948) INTERACT 1.434 (0.877) PROP1564 0.593 (0.507) PROP65 1.967*** (0.620) TRADE −0.026 (0.047) YGAP −0.682*** (0.158) OIL_EX 0.031 (0.049) OIL_IM 0.052 (0.030) Obs 506 No. Countries 15 Data Annual

−1.079** (0.401) 0.730 (0.460) 0.364** (0.127) −0.019 (0.192) −1.139 (3.786) −38.339 (23.633) 0.516 (0.918) 0.239 (0.503) 1.102* (0.609) −0.046 (0.046) −0.562*** (0.180) 0.003 (0.036) 0.045* (0.024) 462 15 Annual

R 2 (within)

0.48

TOPINC UTIP

0.42

98 15 5-year averages

Notes Panel regressions of government outlays as a percentage share of GDP including fixed effects, S H A R E, F P, ln(y), I D E O, I N T E R AC T , P R O P1564, P R O P65, T R AD E, Y G A P, O I L_E X , O I L_I M as control variables. Column (3) contains Arellano-Bond estimation with lagged values of both the predetermined and endogenous variables as instruments. Robust standard errors are shown in parentheses. Standard errors are clustered by country. *, **, and *** respectively denote significance levels at 10, 5 and 1%

2.4 Evidence

37

2.4.2 Instrumental Variables Estimation The empirical analysis presented above establishes a robust negative statistical association between government size and capital income inequality in the presence of a substantial set of controls. However, these results do not establish causality, insofar that the movements in capital income inequality may be endogenous to the size of government, or alternatively both variables may co-move in response to an unobserved driver not accounted for in the controls. What is required for identification is a source of exogenous variation in capital income inequality. In this section we describe and deploy two potential instruments. An advantage of using two independent instruments is that it enables an overidentification test of the exclusion restriction that the instruments are not correlated with the error term in the second stage regression. The first instrument is the number of internet users in percentage of the total population (I N T E R N E T ), encapsulating technological change.20 Skill-biased technological change has been advanced as a (if not the) principle driver of rising inequality in general terms (for example in Goldin and Katz 2009). Conceivably this process has especially underpinned increasing capital income inequality.21 Atkinson et al. (2011) indeed document that a large part of the top income share derives from capital income.22 There are a number of channels through which advancing information technology could increase capital income inequality. One, as noted above is simply the mechanism advanced in Piketty (2014): if capital income rises with fixed ownership concentration, then capital inequality rises. Another stems from the observation that information technology is ‘weightless’ and in such circumstances the distinction between labor and capital income becomes somewhat arbitrary. Thus one can equally describe Mark Zuckerberg as being an extremely productive worker, or as having created a company with enormous capital value. Relatedly, information technology plausibly has allowed many diverse activities to upscale their operations, resulting in significant increases in profitability which has in no small part been manifest in increased capital income for share owners or business partners. What is relevant for the theory above is liability for labor, as distinct from capital, income taxation. In particular in the case of new information technology, the new high earners face an interesting problem of how to classify their income. Plausibly, and indeed empirically as observed above in related situations, they (or their accountants) will classify and organize their income so as to minimize taxation obligations. Given that it is almost universally the case that top marginal labor income 20

Taken from the WDI database. Note that any effect of technological change through labor income inequality, or the labor share, is closed off due to these variables separately being included as controls in the analysis. It is still nonetheless possible that technology is correlated with the error term in the second-stage regression (i.e. violating the exclusion restriction), though the mechanism is not easy to see given the extensive set of controls. Moreover the exclusion restriction is tested below using the Hausman over-identification test. 22 For instance in their Fig. 2.3 capital gains, capital income and business income represent well over half of the income of the top 0.1% in the US. 21

38

2 Inequality and Government Size: A Political …

taxes are higher than the (effective) top marginal capital income taxes, then income will likely be declared as capital income. To summarize, new technology has resulted in enormous rewards for a small number of people who have substantially registered these rewards in the form of capital income. Our second instrument encapsulates exogenous variation in what we term as financial inclusiveness. By definition capital income requires capital ownership, and historically such ownership has not been widespread, even in the OECD. A necessary condition for mass ownership of capital assets and equity in particular is an established level of financial inclusion. A well developed financial system is one where it is easy, for all members of the population, to acquire (and sell) different types of capital assets. When financial inclusion is low, then conceivably at least some forms of asset ownership are not feasible for much of the population, and likely those with low income. Following this line of reasoning we conjecture that capital income inequality falls, conditionally, with financial inclusion. The standard measure of financial inclusion is the ratio of stock market capitalization to GDP. However there are two problems with using this measure as an instrument in the context of our research objective. Firstly stock market capitalization is unlikely to be exogenous: a large public sector by construction implies a small private sector, hence lower stock market capitalization all else equal. Secondly, and more prosaically, the standard source for these data (the World Bank Global Financial Development Database) provides data only from 1989. To uncover exogenous variation in financial inclusion we use the Chinn-Ito index for financial openness (K AO P E N ), an institutional measure that Chinn and Ito (2006) establish leads to changes in financial development, and therefore financial inclusion once legal systems and institutions are sufficiently developed (conditions which apply in the OECD). Notably these authors rule out reverse causality from financial inclusion to financial openness hence the Chinn-Ito index more plausibly satisfies the exogeneity requirement. To summarize the argument: The Chinn and Ito (2006) index exogenously drives financial inclusion. Exogenous increases in financial inclusion permit wider asset ownership thereby causing capital income inequality to fall. Hence we posit that capital income inequality exogenously falls with increases in the Chinn-Ito index.23 Table 2.3 contains the results of the IV estimation. Column 1 contains results using only the I N T E R N E T instrument, and column 2 contains results using only the K AO P E N instrument. The first-stage coefficient estimates for both instruments exhibit signs as hypothesized. Capital income inequality is estimated to (conditionally) increase with internet coverage, and the hypothesis that this particular instrument is weak can be rejected given that the F-statistic of the first stage regression 23

Dabla-Norris et al. (2015) find that overall inequality actually increases with financial openness. The mechanism discussed therein is skills-bias—financial openness productively adds especially to the highly-skilled, thus increasing wage-inequality. It should be clear that this is a distinct hypothesis from ours, which emphasizes access to capital markets. Note again that labor income inequality is controlled for in both the first and second stages of the IV estimation. Hence the estimated effect of the Chinn-Ito index on capital income inequality is already conditional on any effect it has on labor income inequality.

2.4 Evidence

39

exceeds 14. On the other hand capital income inequality is estimated to conditionally fall with capital market openness. The F-stat in this instance does not quite reach the threshold value of 10, but is not far off. Column 3 employs both instruments, with the advantage that this enables application of the overidentification test. The null hypothesis here is that the exclusion restriction is violated, and clearly the test statistic does not indicate rejection of this hypothesis. This test result thus supports the exclusion restriction that the instruments are not correlated with the second-stage error term. Using the results from column 3, the coefficient estimate for T O P I N C in the second stage indicates that a one standard deviation increase in this variable all else equal causes a fall in the size of government of about 6% (i.e. using the data in Table 2.1 around 60% of a standard deviation). Importantly the assumption that all else is equal here is strong: we have already documented the positive correlation between T O P I N C and labor income inequality (U T I P), and indeed the coefficient estimate for the latter variable suggests an offsetting effect if both types of inequality simulataneously increase. What is clear from these results is that the effects of inequality in general terms are more complex than implied in the original Meltzer and Richard (1981) model. Labor income inequality now positively affects government size—consistent with Meltzer and Richard (1981). The top income share—which we interpret as a proxy especially for capital income inequality—negatively affects the size of government. This is consistent with the theoretical reasoning in this chapter. When it is difficult to tax capital income, then those who rely on labor income become averse to labor income taxation. Columns 4 and 5 contain estimation results using 5-year averages of the data. For these regressions the lag of the top income share is used as an instrument, because I N T E R N E T and K AO P E N are not sufficiently strong in this setting, where much of the time variation is averaged out. In column 4 T O P I N C is again estimated to have a significantly negative impact on government size, whilst labor income inequality (U T I P) remains positive and statistically significant. The negative impact of T O P I N C survives the addition of the lagged dependent variable in column 5, though the impact of productivity-induced labor income inequality is here reduced. The concerns motivating IV estimation for capital income inequality (T O P I N C) should also apply to labor income inequality (U T I P). To address this, in Table 2.4 we instrument for both T O P I N C and U T I P by employing I N T E R N E T , K AO P E N , the fifth lag of the top income share, and the fifth lage of labor income inequality. In this specification the estimated coefficient for T O P I N C is again negative. The hypothesis that these instruments are weak can be rejected given that the F-statistic of the first stage regression exceeds 23, and the result of the overidentification test supports the exclusion restriction that the instruments are not correlated with the second-stage error term.

40

2 Inequality and Government Size: A Political …

Table 2.3 Instrumental variables estimation results (1) (2) (3)

(4)

L .OU T L AY S −4.105∗∗∗ T O P I NC (1.186) 1.932∗∗∗ UT I P (0.485) Obs 462 No. Countries 15 Method IV Data Annual

−2.462∗∗ (1.213) 1.420∗∗∗ (0.460) 457 15 IV Annual

Instruments

I N T E R N E T K AO P E N 0.017∗∗∗ −0.246∗∗∗ (0.004) (0.082)

F pχ 2

14.78

9.038

−3.404∗∗∗ (0.903) 1.753∗∗∗ (0.370) 457 15 IV Annual

−1.754∗∗∗ (0.392) 0.836∗∗ (0.350) 112 15 IV 5-year averages I N T E R N E T L .T O P I N C 0.015∗∗∗ 0.801∗∗∗ (0.004) (0.056) K AO P E N −0.208∗∗ (0.082) 10.64 205.3 0.359

(5) 0.523∗∗∗ (0.067) −0.653∗∗ (0.326) 0.269 (0.280) 112 15 IV 5-year averages L .T O P I N C 0.825∗∗∗ (0.062)

177.9

Notes IV is estimated by two-stage-least squares. First stage coefficients are reported below the named instruments in the Instruments row. F is an F-statistic for the statistical significance of the instruments in the first stage regression. pχ 2 is the p-value for the Chi-squared test of overidentifying restrictions. See also notes for Table 2.2 for other details

2.5 Conclusion This chapter analyzes how inequality in the capital income distribution affects the size of government. Capital income is quite distinct from labor income. We define it as rental income, and also model it as untaxed, hence redistribution is financed solely by taxation applied to labor income, and voters have preferences over the tax rate based on their position in the capital income distribution. Despite the fact that there are two underlying sources of heterogeneity in the populations, the median voter is still the unique Condorcet winner because tax preferences are monotonic in labor income. The result relating taxation levels to capital income inequality is novel. In contrast to Meltzer and Richard (1981) increased capital-income inequality now leads to smaller government. Agents who are endowed with capital income are less averse to labor-income taxation. The choice of labor income tax depends on the distribution of capital income: if the share of capital income of the rich increases, then their taxable labor supply falls and the preferred tax rate falls because the median voter has a reduced capacity to redistribute through taxation.

2.5 Conclusion

41

Table 2.4 Instrumental variables estimation results (1) TOPINC UTIP SHARE Obs No. Countries Method Data Instrument for L5.UTIP L5.TOPINC INTERNET KAOPEN F pχ 2

−1.787*** (0.307) 0.628 (0.409) 0.292*** (0.077) 406 15 IV Annual TOPINC 0.348*** (0.040) 0.586*** (0.042) 0.002 (0.004) −0.065 (0.067) 23.74 0.475

UTIP 0.573*** (0.043) −0.035 (0.045) 0.018*** (0.004) −0.083 (0.073)

Notes IV is estimated by two-stage-least squares. First stage coefficients are reported below the named instruments in the Instruments row. F is an F-statistic for the statistical significance of the instruments in the first stage regression. pχ 2 is the p-value for the Chi-squared test of overidentifying restrictions. In this method we instrument for both T O P I N C and U T I P using L5.T O P I N C, L5.U T I P, I N T E R N E T , and K AO P E N . See also notes for Table 2.2 for other details

The relationship between the size of government and inequality is tested in a panel of OECD countries, augmenting the analysis of Pickering and Rockey (2011) and Facchini et al. (2017) to include capital income inequality as an additional explanatory variable. The measure of capital income inequality in the analysis is the top 1% income share. Consistent with the theory, government size is found to be negatively associated with capital income inequality. Moreover controlling for the top income share renders a consistently positive estimate for the impact of labor income inequality on government size, in line with the original Meltzer and Richard (1981) hypothesis. The negative impact of capital income inequality on government size survives a variety of econometric specifications, including when capital income inequality is instrumented with variables encapsulating technology and access to the capital market.

42

2 Inequality and Government Size: A Political …

Appendix Derivation of Equations (2.12) and (2.13) The problem of the median voter m is to choose the tax rate so as to maximize   u m (cm , l m ) = u m (1 − t) x m n m + R m + t y¯ , 1 − n m ,

(2.17)

and the first-order condition for the median voter with respect to the tax rate is  y¯ − y m + t

d y¯ dt



  dn m   u c + (1 − t) x m u c − u l = 0. dt

(2.18)

Thus, making use of equation (2.3), the tax rate chosen by the median voter must satisfy   d y¯ m = 0. (2.19) y¯ − y + t dt Changes in the tax rate t affect average income via two channels: its effect on the opportunity cost of leisure, and its effect on transfers (from the government’s budget constraint r = t y¯ ). In particular, we have that ∂ y¯ dr ∂ y¯ d y¯ = − , dt ∂r  dt ∂τ  ∂ y¯ d y¯ ∂ y¯ = y¯ + t − . ∂r dt ∂τ

(2.20)

with τ = 1 − t. Thus, the total derivative of average income with respect to changes in the tax rate is given by d y¯ y¯r y¯ − y¯τ = < 0, (2.21) dt 1 − t y¯r with y¯r = ∂∂ry¯ and y¯τ = ∂∂τy¯ . Finally, making use of (2.21) to substitute in (2.19), we obtain  y¯r y¯ − y¯τ , 1 − t y¯r

  ηr y¯ (1 − t) − ητ y¯ t , = y¯ − y m (1 − t) + 1 − ηr 

0 = y¯ − y m + t

(2.22)

where ηr = y¯r (r/ y¯ ) and ητ = y¯τ (τ/ y¯ ) are the partial elasticities of average income. Solving the above equation for t, yields

2.5 Conclusion

43

t=

m − 1 + ηr , m − 1 + ηr + mητ

(2.23)

with m = y¯ /y m .

Proof of Proposition 2.1 We begin with the following decomposition of average income y¯ = p (K) y¯ (K) + (1 − p (K)) y¯ (∼ K) ,

(2.24)

where y¯ (K) is the average income of the individuals in set K and y¯ (∼ K) is the average income of the individuals not in set K. From Assumption 2.2 we have that y¯ K > y m . Taking the total derivative of y¯ with respect to R (K), the capital income of the individuals in set K in Eq. (2.24) we obtain     d y¯ ∂ y¯ (K) d y¯ ∂ y¯ (K) ∂ y¯ (∼ K) d y¯ = p (K ) + t + (1 − p (K)) t , d R (K ) ∂ R (K ) ∂r d R (K) ∂r d R (K ) ∂ y¯ (K) ∂ y¯ d y¯ + t, = p (K) ∂ R (K ) ∂r d R (K) ∂ y¯ (K) d y¯ = p (K) + ηr , ∂ R (K ) d R (K )

(2.25) d y¯ , where we used the fact that ηr = ∂∂ry¯ ry¯ = ∂∂ry¯ ty¯y¯ = ∂∂ry¯ t. Using (2.25) to solve for d R(K) we obtain d y¯ p (K) ∂ y¯ (K) = < 0, (2.26) d R (K) 1 − ηr ∂ R (K) since leisure is a normal good. Thus, average income y¯ must fall. In turn, we have that dy m ∂ y m ∂ y¯ = t > 0. d R (K) ∂r ∂ R (K)

(2.27)

Thus, we have established that y¯ must fall and y m must increase following an increase in the capital-income going to the top capital-income recipients. Therefore, m = y¯ /y m falls and the increase in capital income inequality lowers labor income inequality. The upshot is that the increase in the capital income going to the top capital-income recipients results in a lower t, the labor income tax chosen by the median voter.

44

2 Inequality and Government Size: A Political …

References Alstadsæter A, Johannesen N, Zucman G (2018) Who owns the wealth in tax havens? Macro evidence and implications for global inequality. J Public Econ 162:89–100 Atkinson AB, Lakner C (2013) Wages, capital and top incomes: The factor income composition of top incomes in the USA, 1960–2005. The Society for the Study of Economic Inequality Atkinson AB, Piketty T, Saez E (2011) Top incomes in the long run of history. J Econom Lit 49(1):3–71 Azmat G, Manning A, Reenen JV (2012) Privatization and the decline of labour’s share: international evidence from network industries. Economica 79(315):470–492 Bassett WF, Burkett JP, Putterman L (1999) Income distribution, government transfers, and the problem of unequal influence. Eur J Polit Econ 15(2):207–228 Benabou R (1996) Inequality and growth. National Bureau Econom Res Macroecon Annu 11:11–74 Benabou R (2000) Unequal societies: income distribution and the social contract. Am Econom Rev 90(1):96–129 Bergh A, Henrekson M (2011) Government size and growth: a survey and interpretation of the evidence. J Econom Surv 25(5):872–897 Chinn MD, Ito H (2006) What matters for financial development? Capital controls, institutions, and interactions. J Dev Econ 81(1):163–192 De Mello L, Tiongson ER (2006) Income inequality and redistributive government spending. Publ Finance Rev 34(3):282–305 Dabla-Norris ME, Kochhar MK, Suphaphiphat MN, Ricka MF, Tsounta ME (2015) Causes and consequences of income inequality: a global perspective. Int Monetary Fund Diamond PA, Mirrlees JA (1971) Optimal taxation and public production I: production efficiency. Am Econom Rev 61(1):8–27 Facchini F, Melki M, Pickering A (2017) Labour costs and the size of government. Oxford Bull Econ Stat 79(2):251–275 Frydman C, Saks RE (2010) Executive compensation: a new view from a long-term perspective, 1936–2005. Rev Financial Stud 23(5):2099–2138 Galbraith JK, Kum H (2005) Estimating the inequality of household incomes: a statistical approach to the creation of a dense and consistent global data set. Rev Income Wealth 51(1):115–143 Goldin C, Katz LF (2009) The race between education and technology. Harvard University Press, Cambridge MA Gordon R, Kalambokidis L, Slemrod J (2004) Do we now collect any revenue from taxing capital income? J Publ Econom 88(5):981–1009 Gordon RH, Slemrod J (1988) Do we collect any revenue from taxing capital income? Tax Policy and the Economy 2:89–130 Karabarbounis L, Neiman B (2014) The global decline of the labor share. Q J Econ 129(1):61–103 Kau JB, Rubin PH (2002) The growth of government: sources and limits. Public Choice 113(3):389– 402 Kaymak B, Poschke M (2016) The evolution of wealth inequality over half a century: the role of taxes, transfers and technology. J Monetary Econ 77:1–25 Krusell P, Rios-Rull JV (1999) On the size of US government: political economy in the neoclassical growth model. Am Econom Rev 89(5):1156–1181 Meltzer AH, Richard SF (1981) A rational theory of the size of government. J Polit Econ 89(5):914– 927 Muinelo-Gallo L, Roca-Sagalés O (2013) Joint determinants of fiscal policy, income inequality and economic growth. Econ Model 30:814–824 Perotti R (1996) Growth, income distribution, and democracy: what the data say. J Econ Growth 1(2):149–187 Persson M (1995) Why are taxes so high in egalitarian societies? Scandinavian J Econom 569–580 Persson T, Tabellini G (2003) The economic effects of constitutions. MIT Press, Cambridge MA Pickering A, Rockey J (2011) Ideology and the growth of government. Rev Econ Stat 93(3):907–919

References

45

Piketty T (2014) Capital in the twenty-first century. Harvard University Press, Cambridge MA Piketty T, Saez E (2006) The evolution of top incomes: a historical and international perspective. Am Econ Rev 96(2):200–205 Ram R (1987) Wagner’s hypothesis in time-series and cross-section perspectives: Evidence from “real” data for 115 countries. Rev Econ Stat 69(2):194–204 Roberts KW (1977) Voting over income tax schedules. J Publ Econ 8(3):329–340 Rodriguez F (2004) Inequality, redistribution, and rent-seeking. Econ Politics 16(3):287–320 Rodrik D (1998) Why do more open economies have bigger governments? J Polit Econ 106(5):997– 1032 Romer T (1975) Individual welfare, majority voting, and the properties of a linear income tax. Journal of Public Economics 4(2):163–185 Shelton CA (2007) The size and composition of government expenditure. J Public Econ 91(11– 12):2230–2260 Winer SL, Tofias MW, Grofman B, Aldrich JH (2008) Trending economic factors and the structure of Congress in the growth of government, 1930–2002. Public Choice 135(3):415–448

Chapter 3

Inequality and Government Debt

This chapter is originally published in Economics Letters, 2020, 186, p. 108869.

3.1 Introduction Does income inequality necessarily lead to a rise in government debt? One important work is a multicountry politico-economic model designed by Azzimonti et al. (2014), who argue that the incentives of governments to borrow rise both as financial markets become internationally integrated and as income inequality increases if this is associated with higher income risk. Empirical evidence generally has not supported the Azzimonti et al. (2014) hypothesis. For example, Fig. 3.1 depicts the raw correlation between central government debt as a share of GDP over 1970–2010 and the share of income earned by the top 1 percent of the population, indicating no positive association between debt and income inequality. In most theoretical literature, labor is the only source of income and the rich have higher income by dint of higher productivity. However, labor is not the only source of income for the rich and moreover, the labor share of income has declined in recent years (see Azmat et al. 2012; Karabarbounis and Neiman 2014). Indeed, Piketty (2014) relates rising inequality to the falling labor share: if the rate of return on capital is greater than the rate of economic growth, then the share of capital rises, and if ownership is concentrated within a small number of groups, then inequality inexorably increases. Further, capital income has recently become more unequal as well as more important. Kaymak and Poschke (2016) document considerable increases in the concentration of wealth in the U.S. over the past 50 years.

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 W. Luo, Inequality, Demography and Fiscal Policy, Applied Economics and Policy Studies, https://doi.org/10.1007/978-981-99-0518-8_3

47

48

3 Inequality and Government Debt

100

Debt Versus Inequality Hungary

Average Debt by country 40 60 80

Italy Belgium

Greece Ireland

Japan

Netherlands Poland

Canada

United States

Sweden Denmark New Great Zealand France Austria Britain Iceland Turkey

20

Portugal Spain Finland Norway Germany Korea Switzerland Australia

5

10

15 Average Inequality by country

(mean) cgd_gdp

20

25

Fitted values

Fig. 3.1 Scatter plot of average central government debt as a percentage of GDP and average income inequality by country (using top 1 percent income share data)

As in Luo et al. (2017), labor income is taxable, whilst income from capital is harder to tax. Evidence abounds of tax evasion or avoidance in the case of the latter.1 It is harder to escape from PAYE. Like labor income, the capital income distribution is again right-skewed with the majority of individuals endowed with limited (or zero) assets or wealth being compelled to supply labor for their income, which is taxed. On the other hand, those paid in capital income are to a meaningful extent able to avoid the same tax obligation, then the capital-rich are relatively less exposed to taxation. When income differences are generated by capital income, the ability of the median voter to redistribute through taxation is constrained, affecting the public debt accumulation. If capital income inequality increases (and it is the rich who enjoy capital income) such that redistributive policies reduce, then the public debt falls because the incentives of governments to borrow decline.

1

Some may argue that dividends are often taxed at source. Note that only cash dividends can be taxed while the rich can reclassify cash dividends through different ways, or simply transform to stock dividends to avoid being taxed.

3.2 Data and Econometric Specification

49

3.2 Data and Econometric Specification Following Reinhart and Rogoff (2011), the dependent variable is total (domestic plus external) gross central government debt measured as a percentage of GDP. The sample covers the period 1970–2010 in countries which have been OECD members since 1975. The argument invoked in this paper emphasizes the median voter framework hence established democracies are the appropriate sample. Table 3.1 contains descriptive statistics of all the variables. The measure of income inequality usd in most empirical literature is an aggregate level of inequality. A common measure of aggregate inequality, taken from the University of Texas Inequality Project’s Estimated Household Income Inequality data (see Galbraith and Kum 2005), is therefore tested in the empirical analysis. These data (U T I P) use Theil’s T statistic, measured across sectors within each country, to estimate inequality. Another measure of inequality frequently used is the share of income earned by the top 1 percent of the population (T O P I N C), taken from the World Inequality Database. Thus, the empirical analysis further uses this alternative measure to test the hypothesis.

Table 3.1 Descriptive statistics Obs Mean DEBT UTIP TOPINC CAPINEQ LABINEQ ln(y) TRADE PROP1564 PROP65 INTERNET KAOPEN

1030 1172 984 687 687 1244 1186 1320 1320 1360 1134

45.78 36.67 8.25 0.94 1.27 2.96 70.42 65.47 12.57 16.17 1.06

Std. dev.

Min

Max

30.22 4.97 3.60 0.09 0.23 0.51 42.57 3.35 3.46 26.67 1.43

3.30 20.58 2.04 0.33 0.98 0.78 9.10 49.76 3.33 0.00 −1.89

189.10 50.76 28.26 1.00 2.97 4.42 349.85 72.68 22.96 93.39 2.39

Notes The table gives descriptive statistics for the variables. D E BT is total (domestic plus external) gross central government debt measured as a percentage of GDP—taken from Reinhart and Rogoff (2011). U T I P is the University of Texas Inequality Project’s Estimated Household Income Inequality. T O P I N C is the income share of top 1&—taken from The World Inequality Database. L AB I N E Q is the ratio of mean to median household labor income, and C A P I N E Q is equal to one minus the ratio of median to mean household capital income—both data are taken from the Luxembourg Income Study Database. Income y is real GDP per capita in $000s of 2005 prices—taken from the Penn World Tables. T R AD E is the sum of exports and imports as a percentage of GDP—taken from the WDI database. P R O P1564 and P R O P65 are respectively the proportion of the population aged between 15 and 64, and 65 and above—taken from the WDI database. I N T E R N E T is the number of internet users per 100 people—taken from WDI database. K AO P E N is the Chinn and Ito (2006) index for financial openness

50

3 Inequality and Government Debt

If the argument invoked in this paper is important, which means that labor income inequality and capital income inequality have opposite effects on government debt, then arguably previous analyses have suffered from an omitted variable bias. The ideal measure of inequality, given the logic in Meltzer and Richard (1981), is the ratio of mean to median income. As this paper argues, in the case of labor income, the greater this ratio the higher demands for debt. Conversely, greater ratio of mean to median capital income results in less debt because incentives to borrow are constrained. For most of the OECD countries, it is possible to access household level micro-data from the Luxembourg Income Study (LIS) that would permit more direct measurement of these two types of inequality (the advantage of separating out labor and capital income), whilst the number of observations falls due to data availability. Following Meltzer and Richard (1981), labor income inequality is constructed by the ratio of mean to median labor income (L AB I N E Q). Table 3.1 contains statistics for this measure showing a mean value of 1.27, hence (as expected) mean income is generally greater than the median in the LIS data. The LIS data also give the opportunity to construct a measure of capital income inequality. Due to the relatively small value (mostly zero) in the data of median capital income, I use one minus the ratio of median to mean capital income to measure capital income inequality (C A P I N E Q). For this reason the construction of capital income inequality differs slightly from that of labor income inequality. Table 3.1 shows that the maximum value this measure takes is 1, which is also very common. This serves to highlight the fact that most households are not recipient to any capital income. While there are instances of capital income being generally more widespread, for instance in the case of the minimum value of 0.333 (Germany, 1983). Following Persson and Tabellini (2005), the regression analysis also includes standard control variables in the analysis of central government primary budget surplus data. In particular I control for the natural log of real GDP per capita in constant dollars, the degree of trade openness, the proportion of the population aged between 15 and 64, and 65 and above. The benchmark empirical specification is thus   + αi + ηt + u i,t Debti,t = β1 I nequalit yi,t + xi,t

(3.1)

where i represents each country and t represents each time period, control variables analyzed above are included in the vector xi,t , αi are country dummies, ηt are year dummies, and u i,t is the error term.

3.3 Evidence

51

3.3 Evidence 3.3.1 Baseline Estimation This section is to test whether and how central government debt as a percentage of GDP changes with income inequality in the presence of fixed country and year effects. Column 1 of Table 3.2 is a simple specification with just inequality measure (U T I P) and the lagged dependent variable using annual data OLS regression, with robust standard errors clustered by country. Column 2 further uses an alternative measure of inequality (T O P I N C) instead. The insignificance findings together with opposite estimated coefficients for two inequality measures challenge the argument by Azzimonti et al. (2014). As mentioned above, column 3 then extends the regression to include two types of inequality, L AB I N E Q and C A P I N E Q, to separate out labor and capital income. In this specification the sign of the coefficient estimate relating to capital income inequality (C A P I N E Q) is negative, and statistically significant at the 5% level. It is also noteworthy that the coefficient estimate for labor income inequality (L AB I N E Q) is positive, and statistically significant at the 10% level. This is consistent with the argument—an increase in labor income inequality causes governments to have more incentives to borrow whilst capital income inequality relieves the stress. Columns 4–6 repeat the analysis of columns 1–3 including full controls instead. The results in the presence of fixed country and year effects, as well as the control variables, support those already found. Using the estimate from column 6 of Table 3.2, the estimated coefficient for the measure of capital income inequality is negative, with a p-value of 7.5% and the estimated relationship is sizable: a one standard deviation increase in capital income inequality is statistically associated with a fall in central government debt of 0.67% of GDP. On the other hand, the estimated coefficient for the measure of labor income inequality is positive, with a p-value of 2.8%: a one standard deviation increase in labor income inequality is associated with an increase in debt of 1.02% of GDP.

3.3.2 Instrumental Variables Estimation The empirical analysis presented above establishes a robust negative statistical association between central government debt and capital income inequality in the presence of a substantial set of controls. However, these results do not establish causality, insofar that the movements in capital income inequality may be endogenous to debt, or alternatively both variables co-move in response to an unseen third variable. What is required for identification is a source of exogenous variation in capital income inequality. In an attempt to identify such movements I employ two instrumental variables following Luo et al. (2017). The first instrument is technological change, measured as the number of internet users in percentage of the total popula-

52

3 Inequality and Government Debt

Table 3.2 Panel regressions of inequality on debt (1)

(2)

(3)

(4)

(5)

(6)

L.DEBT

0.970∗∗∗ (0.0178)

0.975∗∗∗ (0.0206)

0.917∗∗∗ (0.0241)

0.940∗∗∗ (0.0220)

0.942∗∗∗ (0.0232)

0.902∗∗∗ (0.0372)

UTIP

0.454 (0.285)

0.396 (0.241) −0.192 (0.197)

TOPINC

−0.0661 (0.238)

CAPINEQ

−7.560∗∗ (3.582)

−7.521∗ (4.027)

LABINEQ

3.307∗ (1.622)

4.449∗∗ (1.899)

ln(y)

1.492 (3.048)

−5.905 (4.027)

−5.856 (4.975)

TRADE

−0.0395 (0.0320)

−0.0352 (0.0407)

−0.0138 (0.0479)

PROP1564

−0.0334 (0.281)

0.408 (0.331)

−0.165 (0.469)

PROP65

0.811∗∗∗ (0.284)

0.895∗ (0.512)

−0.0522 (0.480)

790

568

902

790

568

No. Countries 27

Observations

902

26

24

27

26

24

Estimation method

FE

FE

FE

FE

FE

FE

Year dummies

Yes

Yes

Yes

Yes

Yes

Yes

R2

0.951

0.949

0.906

0.952

0.951

0.907

Notes Dependent variable is gross central government debt measured as a percentage of GDP. Estimations use panel regression with country fixed effects and robust standard errors clustered by country in parentheses. Year dummies are included in all regressions. Columns (4)–(6) again test columns (1)–(3) including ln(y), T R AD E, P R O P1564, and P R O P65 as additional control variables. ∗ , ∗∗ , and ∗∗∗ respectively denote significance levels at 10%, 5% and 1%

tion (I N T E R N E T ).2 Skill-biased technological change has been advanced as a (if not the) principle driver of rising inequality in general terms (for example in Goldin and Katz (2009). Conceivably this process has especially underpinned increasing capital income inequality. The second instrument is financial inclusion, measured as the Chinn-Ito index (K AO P E N ).3 A well developed financial system is one where it is easy, for all members of the population, to acquire (and sell) different types of capital assets. When financial inclusion is low, then conceivably at least some forms of asset ownership are not feasible for much of the population, and likely those with low income. The presence of a second instrument allows overidentification tests to investigate the exclusion restrictions.

2 3

Taken from the WDI database. Taken from Chinn and Ito (2006).

3.3 Evidence

53

Table 3.3 Instrumental variables estimation results (1) L.DEBT CAPINEQ LABINEQ Observations No. Countries Data Estimation method Year dummies Overid Weak Inst. R2

0.885∗∗∗ (0.017) −10.660∗ (5.597) 4.576∗∗ (2.018) 514 24 Annual IV Yes Sargan χ 2 = 2.877 (p = 0.237) Basmann χ 2 = 2.522 (p = 0.283) F=70.404 0.972

Notes: Instrumental variables regression of central government debt on capital income inequality and labor income inequality using technological progress, capital market access, L2.CAPINEQ, and L2.LABINEQ as instruments. ∗ , ∗∗ , and ∗∗∗ respectively denote significance levels at 10%, 5% and 1%

The concerns motivating instrumental variables estimation for capital income inequality (C A P I N E Q) should also apply to labor income inequality (L AB I N E Q). To address this, in Table 3.3 I instrument for both inequality measures by employing I N T E R N E T , K AO P E N , and the second lags of two inequality. The hypothesis that these instruments are weak can be rejected given that the F-statistic of the first stage regression exceeds 70. Both the Sargan and Basman overidentification tests are not rejected, which supports the assumptions of instrument exogeneity, and the associated exclusion restrictions. Importantly the estimated coefficient for C A P I N E Q is still found to be negative and that for L AB I N E Q keeps positive, and statistically significant.

3.3.3 Robustness Table 3.4 investigates the robustness and contains estimation results from fixed country and year effects. Column 1 uses the same specification as column 6 of Table 3.2 but excluding Asian countries (e.g. Japan and Korea) to examine whether the regional coverage of the sample affects the results. Column 2 further excludes non-EU countries. As can be seen the results are essentially unaltered given their exclusion. Columns 3 and 4 split the sample by economic development according to the median value of GDP per capita. In column 3 the (relatively) high income sample again

54

3 Inequality and Government Debt

Table 3.4 Robustness and extensions (1) (2) L.DEBT

0.903∗∗∗

(0.0373) CAPINEQ −7.589∗ (4.038) 4.302∗∗ LABINEQ (1.941) ln(y) −5.792 (5.012) TRADE −0.00848 (0.0483) PROP1564 −0.157 (0.471) PROP65 −0.0551 (0.481) Observations 562 No. Countries 22 Data Excluding Asia FE Estimation method Year dummies Yes 0.907 R2

(3)

(4)

(5)

0.946∗∗∗

0.978∗∗∗

0.827∗∗∗

(0.0211) −8.078∗ (4.311) 7.289∗∗∗ (2.109) −18.44∗∗∗ (6.107) −0.0196 (0.0497) 0.686 (0.440) −0.322 (0.368) 392 16 Excluding non-EU FE

(0.0332) −8.202∗ (4.419) 5.143∗∗∗ (0.746) −20.29∗∗ (8.746) 0.0262 (0.0779) −0.624 (1.033) −2.710∗∗∗ (0.879) 284 21 Higher income FE

(0.0604) −1.112 (5.189) 2.614 (3.046) −7.548 (9.839) −0.0872 (0.0936) −1.217 (0.849) 1.668 (1.013) 284 20 Lower income

0.902∗∗∗ (0.0350) −7.460∗∗ (3.802) 4.462∗∗ (1.787) −5.783 (4.690) −0.0142 (0.0453) −0.163 (0.439) −0.0466 (0.455) 543 23 Full

Yes 0.942

Yes 0.905

Yes 0.885

FE

ArellanoBond Yes

Notes As for Table 3.2. Column 1 excludes Japan and Korea observations. Column 2 excludes non-EU observations. Columns 3 and 4 respectively correspond to higher and lower income levels. Column 5 contains Arellano-Bond estimation with lagged values of both the predetermined and endogenous variables as instruments

returns a negative coefficient for capital income inequality and a positive coefficient for labor income inequality of very similar magnitudes to that found for the full sample. In column 4 the (relatively) low income sample also returns same signs of coefficients but with reduced statistical significance. The debt-inequality relationship is somewhat looser under lower economic development. The last column contains Arellano-Bond dynamic panel estimation results. The negative relationship between government debt and capital income inequality holds up, and indeed the coefficient estimate pertaining to labor income inequality is still positive, and significantly different from zero at the 5% level.

3.4 Conclusion This paper analyzes how income inequality affects central government debt. In contrast to Azzimonti et al. (2014), an increase in labor income inequality is found to be

References

55

positively associated with debt level, whilst capital income inequality leads to lower debt. The empirical results hold across various econometric specifications employed.

References Azmat G, Manning A, Reenen JV (2012) Privatization and the decline of labour’s share: international evidence from network industries. Economica 79(315):470–492 Azzimonti M, De Francisco E, Quadrini V (2014) Financial globalization, inequality, and the rising public debt. Am Econ Rev 104(8):2267–2302 Chinn MD, Ito H (2006) What matters for financial development? Capital controls, institutions, and interactions. J Dev Econ 81(1):163–192 Galbraith JK, Kum H (2005) Estimating the inequality of household incomes: a statistical approach to the creation of a dense and consistent global data set. Rev Income Wealth 51(1):115–143 Goldin CD, Katz LF (2009) The race between education and technology. Harvard University Press, Cambridge, MA Karabarbounis L, Neiman B (2014) The global decline of the labor share. Quarterly J Econ 129(1):61–103 Kaymak B, Poschke M (2016) The evolution of wealth inequality over half a century: the role of taxes, transfers and technology. J Monetary Econ 77:1–25 Luo W, Pickering A, Monteiro PS (2017) Inequality and the size of government. Discussion Papers 17/02, Department of Economics, University of York Meltzer AH, Richard SF (1981) A rational theory of the size of government. J Political Econ 89(5):914–927 Persson T, Tabellini GE (2005) The economic effects of constitutions. MIT Press, Cambridge, MA Piketty T (2014) Capital in the twenty-first century. Harvard University Press, Cambridge, MA Reinhart CM, Rogoff KS (2011) From financial crash to debt crisis. Am Econ Rev 101(5):1676– 1706

Part II

Demography and Fiscal Policy

Part II of the text covers the analysis of the relationship between demography and fiscal policy. It briefly reviews the dynamic changes of tax composition in the age of demographic change and builds upon the impacts of changes in age structure on the extent of taxes on income relative to expenditure.

Chapter 4

Population Aging and the Composition of Taxes: A Political Economy Theory

This chapter is originally published in Policy Studies, 2023. https:// doi.org/ 10.1080/ 01442872.2023.2202905.

4.1 Introduction As observed in many developed countries, the influences of rapid demographic change on various elements have attracted rising attention among academics and policymakers alike. One crucial aspect of concern is the effect of population aging on government budgets and on the sustainability of public finances. Although the existing literature has widely and comprehensively covered aging-induced changes on government expenditure (i.e., Balassone et al. 2011; European Commission 2018), the effect of population aging on government revenue remains understudied. Moreover, in the cross-country context, there are many potential effects of demographic change. The literature on population aging specifically examines tax shocks (Ferraro and Fiori 2020), college attainment (Conesa et al. 2020), demand for public services (Guest and McDonald 2000), fiscal sustainability (Andersen 2012; Lee et al. 2017; Van Sonsbeek 2010), and fiscal stress (Guest 2006; Kudrna et al. 2019). How does population aging affect the political economy of tax policy? The political economy model developed by Razin et al. (2002), building on Meltzer and Richard (1981) and Saint (1994), argues that in democracies, an increase in the dependency ratio leads to lower tax rates and a fall in the size of social transfers. In Razin et al. (2002), the role of government is to redistribute funds emanating solely from income taxes to both workers and retirees, and under democracy, the equilibrium tax rate is that preferred by the median voter. In the case of positive population growth, the median voter is a worker. An increase in the dependency ratio, consistent with a fall in the population growth rate, lowers income taxes and transfers. Admittedly, people generally receive less income and consume differently when retired (Aigner and Doring 2012); however, on average they are more wealthy. Under © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 W. Luo, Inequality, Demography and Fiscal Policy, Applied Economics and Policy Studies, https://doi.org/10.1007/978-981-99-0518-8_4

59

60

4 Population Aging and the Composition of Taxes …

the premise that different consumption goods and different kinds of income are differentially taxed around the world, rapid demographic change has fiscal implications as well as influences on government revenue. Given the increased empirical importance of expenditure taxes, Luo (2019) focused on a panel of international data to examine the relationship between the extent of taxes on income relative to expenditure and the fraction of the retired population. Despite this, relatively few studies have provided a theoretical analysis of the political-economic determination of the tax composition.1 For example, one related work by Pickering and Rajput (2018) links the composition of taxes to inequality. Related studies by Aidt and Jensen (2009) and Keen and Lockwood (2010) have analyzed the adoption of tax instruments with the development process. Thus, a key aim of this paper is to provide a political economy theory to scrutinize how population aging affects the composition of taxes. Moreover, it provides a possible explanation for the puzzling situation where countries with a somewhat larger proportion of retirees have relatively higher labor income taxes and lower expenditure taxes to reflect the increased political clout of the retired population, whose members presumably prefer labor income taxes over expenditure taxes.2 However, evidence supporting this hypothesis is generally weak. For example, Luo (2019) found a negative and significant relationship between the fraction of the retired population and the extent of taxes on income relative to expenditure. This paper involved the development of an overlapping generations model with two types of agents: workers and retirees. They vote simultaneously on income tax and expenditure tax, whose tax proceeds are redistributed in a lump sum to both generations, with a balanced budget across consecutive periods.3 As in Razin et al. (2002) the choice of the median voter is the unique Condorcet winner.4 In the case of a positive population growth rate, the median voter is a young individual (because the younger generation is larger than the older generation), a factor that determines political equilibrium tax rates. This paper contributes to the existing literature in the theoretical effect of population aging on tax revenue, as follows. The theoretical findings are in line with Razin et al. (2002) in positing that an increased fraction of retirees in the population always leads to lower labor income taxes as long as the decisive voter is young. The outcomes relating to expenditure taxes are novel, implying that this depends on initial levels of income taxes and expenditure taxes. When income taxes are fairly large, expenditure tax rates decline monotonically with population growth. When income taxes are relatively small, a threshold relationship exists. At lower initial 1

Luo (2019) only developed a verbal theory. One alternative interpretation is that, arguably, countries with a higher dependency ratio tend to be fairly richer, whereas richer countries tend to collect a greater share of income taxes. 3 The author of this paper is aware of the high debt-to-GDP ratio, especially in member states of the Organisation for Economic Co-operation and Development (OECD). Note that this paper primarily focuses on the composition of taxes rather than debt decisions. 4 According to Persson and Tabellini (2002), if all voters have single-peaked policy preferences over a given ordering of policy alternatives, a Condorcet winner always exists and coincides with the median-ranked bliss point. 2

4.2 Background and Related Literature

61

levels of expenditure tax rates, an increase in the share of retirees in the population also leads to lower expenditure taxes. However, if the initial levels of expenditure taxes pass a certain threshold level, then the increased size of the retired population results in higher expenditure taxes, even if this comes at the price of a greater size of deadweight expenditure tax losses, because the decisive voter wants to shift the tax burden onto the retired population.5 Nonetheless, the main proposition is that the composition of taxes–defined as the extent to which taxes are levied on income relative to expenditure—falls unambiguously along with the share of retirees. The logic is similar to Razin et al. (2002). If the median voter is a worker, then increasing the size of the retired population compels a shift in tax composition toward expenditure taxes, as only workers pay income taxes, whereas both generations pay expenditure taxes.6 The next section introduces related literature. Section 4.3 covers the economic environment. Section 4.4 theoretically analyzes how the composition of taxes shifts with population aging, and Sect. 4.5 provides the conclusion.

4.2 Background and Related Literature A chief argument of concerns over population aging is that countries undergoing swifter demographic change tend to lower income taxes and transfers (see Razin et al. 2002). Several scholars have attempted to study this phenomenon. For example, Creedy et al. (2010), Felix and Watkins (2013), and Yashio and Hachisuka (2014) respectively explored the influence of population aging on taxation for New Zealand, the United States, and Japan. Based on a counterfactual analysis, they found that per capita income tax revenue declined considerably with aging, whereas its effect on consumption or expenditure taxes remains unclear to some extent. However, a series of empirical studies has generally not supported the hypothesis regarding the negative effect of aging on income taxes (see Disney 2007; Sanz and Velazquez 2007; Shelton 2008). For example, Fig. 4.1 depicts the raw correlation between personal income taxes as a share of total taxation between 1990 and 2020 and the proportion of the population over the age of 65, indicating no negative relationship between aging and the size of personal income taxes. The correlation between these two series is 0.314. In response to this conundrum, new theoretical literature has appeared asserting that a larger proportion of retirees can lead to higher income taxes and more generous social transfers. Galasso and Profeta (2007) proposed that population aging has a political effect: the median voter becomes older and hence more willing to support a larger system. Further theoretical analyses of welfare states 5

In practice, Japan, as a country with an aging population, provides interesting anecdotes. In April 2014, Japan’s government increased its expenditure tax rate from 5% to 8%, and again raised this rate to 10% in October 2019. 6 The author of this paper is also aware that income taxes are primarily levied on pensions and labor/self-employment income. Owing to the lower social security contributions and lower pension incomes, the effective tax rate on the retired population is lower overall.

62

4 Population Aging and the Composition of Taxes …

Personal income taxes as a % of taxation 0 20 40 60

Income Taxes Versus Population Aging

0

5

10 Proportion over 65 incometax

15

20

Fitted values

Fig. 4.1 Correlation between aging and personal income taxes (cross-country data, 1990–2020)

and aging, as argued by Simonovits (2007), have expanded beyond the mechanism investigated in Razin et al. (2002) in attempting to link increasing income taxes to aging. Notes: Aging is defined as population aged 65 and above as a percentage of the total population. Population is based on the de facto definition of population, which counts all residents regardless of legal status or citizenship. Tax on personal income is defined as the taxes levied on the net income (gross income minus allowable tax reliefs) and capital gains of individuals. This indicator relates to government as a whole (all government levels) and is measured in percentage of total taxation. Sources: The share of the population 65 and older was obtained from the WDI database. Tax on personal income was obtained from the OECD Data. Prammer (2019) built on the work of Creedy et al. (2010), Felix and Watkins (2013), and Yashio and Hachisuka (2014) to explore the effect of population aging on income tax revenue and social security contributions in a dynamic analysis. In this context, a dynamic analysis was performed linking population scenarios to long-term scenarios that assume rising real wages and pension benefits, causing an increase in income tax revenue per capita and social security contributions. The theory proposed in this paper was inspired by recent increases in the sizes of taxes on goods and services, in particular in today’s era of population aging (Luo 2022). As a consequence, expenditure taxes, as one source of revenue outside of income taxes, are empirically becoming an increasingly important component of total revenue. Cross-country evidence shows that over the period of 1990–2020, taxes on personal income, on average, came to represent a smaller fraction of total revenue than taxes on goods and services (i.e., personal income taxes are, on average, approximately 23% of total taxation, whereas taxes on goods and services account

4.2 Background and Related Literature

63

for over 33% of total taxation on average). Figure 4.2 portrays a scatter plot of the two series. In addition, this paper observed that the capacity to raise revenue through income taxes normally might entail fewer limitations in countries with fairly high incomes. For instance, among OECD members, taxes on income are on average 25% of total revenue, whereas the average global level of income taxes accounts for just 22% of total revenue. In contrast, Luo (2018) indicated that the capacity to raise revenue through income taxes is normally limited in countries with comparatively low incomes, whereby OECD countries collect more income taxes (as a share of total revenue) relative to non-OECD countries. This is also well documented in Besley and Persson (2014)).7 One underlying reason is that levying income taxes requires better tax administration capacity than levying consumption taxes, for example. Besley and Persson (2014) deduced that political factors (i.e., weak institutions, fragmented polities, and a lack of transparency) and socio-cultural factors (i.e., a weak sense of national identity and poor norms for compliance) may stifle the collection of tax revenue in developing countries. However, this paper focuses on the relationship between demographic structure and the structure of taxation since the central argument involves the median voter framework, given that the median voter more plausibly drives policy under a stronger democracy (Luo 2019). Countries undergoing swifter demographic change are more likely to be comparatively richer (Acemoglu and Restrepo 2017), whereas richer countries tend to be endowed with a higher level of democracy (Acemoglu et al. 2008). Arguably, the expenditure taxes used in Fig. 4.2 should ideally only include consumption taxes, rather than all forms of spending under the premise of the model’s focus on voters. Systematic cross-country data isolating taxation revenue derived from private consumption, as opposed to other forms of expenditure, are not available. However, because aggregate consumption typically represents around 60% of GDP in most countries, it seems likely that the data are reflecting the underlying variable of interest as stated in Pickering and Rajput (2018). Notes: Tax on personal income is defined as the taxes levied on the net income (gross income minus allowable tax reliefs) and capital gains of individuals. This indicator relates to the government as a whole (all government levels), and is measured in percentage of total taxation. Tax on goods and services is defined as all taxes levied on the production, extraction, sale, transfer, leasing or delivery of goods, and the rendering of services, or on the use of primarily or permission to use goods or to perform activities. They consist mainly of value added and sales taxes. This covers: multi-stage cumulative taxes; general sales taxes - whether levied at manufacture/production, wholesale or retail level; value-added taxes; excises; taxes levied on the import and export of goods; taxes levied in respect of the use of goods and taxes on permission to use goods, or perform certain activities; taxes on the extraction, processing or production of minerals and other products. This indicator relates to government as a whole (all government levels) and is measured in percentage of total taxation. 7

Besley and Persson (2014) found that low-income nations typically collect taxes of between 10 and 20% of gross domestic product (GDP), whereas the average for high-income states is approximately 40%.

64

4 Population Aging and the Composition of Taxes …

Fig. 4.2 Correlation between personal income taxes and expenditure taxes, 1990–2020

Sources: Both tax on personal income as a percentage of taxation and tax on goods and services as a percentage of taxation were obtained from the OECD Data. Recent literature has generally focused on the composition of taxes instead of income taxes only because of the declining status of tax revenue from income. For instance, Pickering and Rajput (2018) analyzed how income inequality affects the extent to which taxes are levied on income relative to expenditure. Luo (2018, 2019) then extended the Pickering and Rajput (2018) framework to explore the relationship between demography and the composition of taxes. Luo (2020) further considered the effect of demographic structure on economic growth through the mechanism of tax composition. Luo and Zhu (2021), based on Luo (2020), expanded to the threegeneration framework and established how the young population affects economic growth. With the increased importance of taxes on goods and services as a source of government revenue, this paper analyzes the theoretical effect of population aging on tax revenue from a more profound perspective of tax structure to fill existing research gaps.

4.3 The Economic Environment This section entails the use of a theoretical framework to examine the effect of population aging on the composition of taxes below. The model extends Razin et al. (2002) to include expenditure taxes as well as income taxes, consisting of an overlapping generations model with population growth rate (n) where individuals live for two periods: a working (young) period and a retirement (old) period. The utility u of an individual born in period t depends on their consumption in the two periods (c1,t and c2,t+1 ):

4.3 The Economic Environment

65

u t = u(c1,t , c2,t+1 )

(4.1)

where u is a strictly concave and a twice-differentiable utility function. Owing to the presence of expenditure (or consumption) taxes τc,t and τc,t+1 , consumption is less than expenditure in period t (x1,t ) and in period t + 1 (x2,t+1 ): c1,t = (1 − τc,t )x1,t

(4.2)

c2,t+1 = (1 − τc,t+1 )x2,t+1 .

(4.3)

First-period expenditure, x1,t , and second-period expenditure, x2,t+1 , are determined by budget constraints (4.4) x1,t = (1 − τ y,t )yt + rt x2,t+1 = rt+1

(4.5)

where labor income yt is taxed at a linear income tax rate τ y,t , and rt and rt+1 are lump-sum redistributions in period t and in period t + 1.8 The assumption that individuals work only in the first period is implicit in Eq. (4.5). This also indicates that retirees do not pay any income taxes. The government budget is assumed to be balanced period by period. Redistribution (rt ) is financed by consumption and income tax revenue, and is paid to both working and retired people as in Razin et al. (2002) (rt is assumed to be equal for young and old, as in Razin et al. (2002)). Hence,     rt N0 (1 + n)t−1 + (1 + n)t = τc,t x¯t N0 (1 + n)t−1 + (1 + n)t + τ y,t y¯t N0 (1 + n)t (4.6) where N0 is the initial quantity of young individuals, and x¯t and y¯t are the average levels of expenditure and income in period t. Further, expenditure equals income at the aggregate level in period t,   x¯t N0 (1 + n)t−1 + (1 + n)t = y¯t N0 (1 + n)t .

(4.7)

Combining (4.6) and (4.7), the lump-sum redistribution equals rt = (τc,t + τ y,t )

8

1+n y¯t . 2+n

(4.8)

In the model, expenditure taxes seem to be broadly associated with consumption taxes. As indicated above, we can also rename expenditure taxes as consumption taxes. Moreover, the income tax defined in the model appears to be consistent with the definition of personal income taxes, which are levied on physical persons as opposed to corporate entities. For brevity and following Pickering and Rajput (2018), this paper will describe these as “income taxes” in the text.

66

4 Population Aging and the Composition of Taxes …

Since the government budget is balanced period by period, it follows that redistribution in period t + 1, rt+1 , is independent of tax rates, and that τ y,t and τc,t in period t are analogous to Razin et al. (2002). In voting on tax rates τ y,t and τc,t , individuals living in period t thus take rt+1 as exogenous because there is no serial dependence between rt and rt+1 . The political-economic equilibrium for tax rates, τ y,t and τc,t , is then determined by the majority of voting individuals alive in period t, without being influenced by preceding or future generations. A final component is that mean income is modeled to be declining in taxes to capture the perspective of Meltzer and Richard (1981) as in Pickering and Rajput (2018)), (4.9) y¯t = yt∗ e−δ y τ y,t −δc τc,t where yt∗ is potential income, and 0 < δ y < 1 and 0 < δc < 1 capture the sensitivity of income in relation to income and expenditure taxes, respectively. The parameters δ y and δc reflect deadweight losses due to the costs of tax collection and/or their influences on economic activity. High values of δ y and δc indicate high costs for tax collection or alternatively low levels of tax base. The properties of (4.9) indicate that d y¯t = −δ y y¯t and dτd y¯c,tt = −δc y¯t , and hence that the proportionate deadweight losses, dτ y,t d y¯t /dτ y,t y¯t

and

d y¯t /dτc,t y¯t

, are constant, which rules out scale effects.

4.4 Political-Economy Equilibrium: The Choice of Tax Policy 4.4.1 Income Taxes This section addresses policy-setting. The pivotal voter alive in period t chooses the vector of policies q = {τ y,t , τc,t , rt } to maximize their utility. Substituting (4.4) and (4.5) into (4.2) and (4.3) and then substituting into (4.1) and using (4.8) gives    1+n  y¯t , (1 − τc,t+1 )rt+1 (4.10) u t = u (1 − τc,t ) (1 − τ y,t )yt + (τc,t + τ y,t ) 2+n which indicates that the policy problem is multidimensional (τc and τ y ).9 The important point of departure from Razin et al. (2002) is that there are now two tax instruments being set. In general, the Condorcet winner does not exist when the policy problem has two (or more) dimensions. However, following Persson and Tabellini (2002), a simple sufficient condition can be found that again ensures the existence of a Condorcet winner: Voter heterogeneity is limited in that voters’ preferences for a multidimensional policy can be projected onto a unidimensional space in which different voters can be ordered by type. Thus, the equation shown above can be 9

In voting on tax rates in period t individuals take τc,t+1 as exogenous.

4.4 Political-Economy Equilibrium: The Choice of Tax Policy

67

re-written in terms of (unidimensional) intermediate preferences, indicating that the choice of the median voter is pivotal. As in Grandmont (1978), if voters only differ along one dimension (here, income), and the indirect utility function (W (q;y i )) can be expressed as W (q; y i ) = J (q) + K (y i )H (q) then the choice of the median voter is a Condorcet winner. It is clear that Eq. (4.10) satisfies this requirement. As long as the rate of population growth is positive, n > 0, the condition where there are more young versus older individuals (or more working individuals than retired ones) always holds. As in Razin et al. (2002), since individuals work only in the first period, the ratio of retirees to workers is given by 1/(1 + n), and the overall dependency ratio–retired as a share of the total population—is given by 1/(2 + n). This implies that the median voter, in determining equilibrium tax rates, is still among the working-age population. As such, the preferred policy of the median voter is the unique Condorcet winner, even though the policy problem is two-dimensional. For a given n, the political equilibrium (τ y or τc ) is constant over time. This ensures that the time subscript t is suppressed henceforth. Maximization of (4.10) with respect to τ y , given (4.9), yields: ∂u = V (τ y , n) ∂τ y 1+n 1+n y¯ − (τc + τ y ) δ y y¯ = 0 = −yd + 2+n 2+n

(4.11)

where yd is the income of the decisive voter. The mathematical derivations are found in the appendix. Equation (4.11) is analogous to the well-known condition (4.13) in Meltzer and Richard (1981) and Eq. (4.11) in Razin et al. (2002), which determines the preferred income tax rate of the decisive voter. Equation (4.11) then delivers the outcome in Razin et al. (2002) where an increase in n raises the political equilibrium income tax rate as discussed below. According to Eq. (4.11), τ y is defined by the first-order condition. Given the assumption that u is strictly concave, I have the second-order condition ∂ V (τ y , n) ∂ 2u = ≤ 0. ∂τ y2 ∂τ y

(4.12)

Here, I examined the effect of changes in the population growth rate and thus population aging on the equilibrium income tax rate. Total differentiation of (4.11) with respect to n implies ∂ V (τ y ,n) dτ y /∂n = − ∂ V (τ y ,n) . (4.13) dn /∂τ y

68

4 Population Aging and the Composition of Taxes …

Since

∂ V (τ y ,n) ∂τ y

≤ 0 (see Eq. 4.12), it follows that the direction of the effect of changes

in n on the equilibrium income tax rate, τ y , is determined by the sign of differentiating equation (4.11) with respect to n, I conclude that ∂ V (τ y , n) 1 1 y¯ − (τc + τ y ) δ y y¯ . = ∂n (2 + n)2 (2 + n)2

∂ V (τ y ,n) . ∂n

By

(4.14)

If the sign of ∂ V (τ y , n)/∂n is positive, then an increase in the rate of population growth, n, increases the political economy equilibrium income tax rate, τ y . Inspecting y¯ the right-hand side of (4.14) shows that it contains one positive term, (2+n) 2 , whereas the other term is negative. Consider, for concreteness, the case in which the decisive voter is young and the population growth rate rises (the fraction of retirees falls). In this case, there is a decline in the amount of tax revenue collected from the median voter that “leaks” to retirees, who become a smaller share of the population with higher n. This is a pro-tax factor. However, the per capita marginal efficiency cost 1 ¯ , rises, which is an anti-tax factor. In of distortionary taxation, −(τc + τ y ) (2+n) 2 δy y this instance, there is no ambiguity: The pro-tax factor always dominates the anti-tax factor and hence ∂ V (τ y , n)/∂n is positive. The proof of this is in the appendix. Lemma 4.1 The equilibrium income tax rate rises as the rate of population growth rises, that is, dτ y > 0. dn In this case, an increase in n (a smaller fraction of retirees) increases the political economy equilibrium income tax rate with exactly the same underpinning as that provided in Razin et al. (2002) (who only considered income taxes). The aging of the population affects the equilibrium income tax rate in two directions: the increased number of retirees raises the demand for benefits while simultaneously restraining the willingness of the working-age population to accede to higher income taxes, as current workers are net losers from the welfare state. If the decisive voter is not among the retirees, as is still the case in all Western countries, then the increased size of the retired population leads to lower income taxes since the decisive voter is adversely affected because they are a net contributor to the welfare system.

4.4.2 Expenditure Taxes Now let us turn to the political equilibrium choice of the expenditure tax rate. Maximization of (4.10) with respect to τc , given (4.9), yields ∂u = W (τc , n) ∂τc 1 + n   1+n 1+n  = (1 − τc ) y¯ − (τc + τ y ) δc y¯ − (1 − τ y )yd + (τc + τ y ) y¯ = 0. 2+n 2+n 2+n

(4.15)

4.4 Political-Economy Equilibrium: The Choice of Tax Policy

69

The mathematical derivations are again found in the appendix. According to (4.15), τc is defined by the first-order condition. Given the assumption that u is strictly concave, I have the second-order condition ∂ W (τc , n) ∂ 2u = ≤ 0. ∂τc2 ∂τc

(4.16)

I next explored the effect of changes in the population growth rate and thus population aging on the equilibrium expenditure tax rate. Total differentiation of (4.15) with respect to n implies ∂ W (τc ,n) dτc /∂n = − ∂ W (τ ,n) . (4.17) c dn /∂τc (τc ,n) Since ∂ W∂τ ≤ 0 (see Eq. 4.16), it follows that the direction of the effect of changes c (τc ,n) . in n on the equilibrium expenditure tax rate, τc , is determined by the sign of ∂ W ∂n By differentiating equation (4.15) with respect to n, I concluded that

  ∂ W (τc , n) 1 1 1 = (1 − τc ) y¯ − (τc + τ y ) δc y¯ − (τc + τ y ) y¯ . 2 2 ∂n (2 + n) (2 + n) (2 + n)2 (4.18) If the sign of ∂ W (τc , n)/∂n is negative, then an increase in the rate of population growth, n, decreases the political economy equilibrium expenditure tax rate, τc . The y¯ right-hand side of (4.18) contains one positive term, (1 − τc ) (2+n) 2 , whereas the other terms are negative. Furthermore, consider for concreteness the case in which the decisive voter is young and the population growth rate rises (the fraction of retirees falls). As with income taxes, within the bracket, the pro-tax factor does not always 1 ¯, dominate the distortionary element (which is an anti-tax factor), −(τc + τ y ) (2+n) 2 δc y given that 0 < τc < 1, 0 < τ y < 1, and 0 < δc < 1, so the whole term (1 − τc )



 1 1 y ¯ − (τ + τ ) δ y ¯ c y c (2 + n)2 (2 + n)2

is indeterminate. Note that the last term on the right-hand side of (4.18) is also negative, which indicates that the sign of dτc /dn depends on the values of τ y , τc , and δc . This is evidenced in the appendix. 1 Lemma 4.2 If τ y > 1+δ , then the equilibrium expenditure tax rate declines as the c rate of population growth rises, that is,

dτc < 0; dn 1 If τ y < 1+δ , then the effect of changes in the population growth rate on the c equilibrium expenditure tax rate is given by:

70

4 Population Aging and the Composition of Taxes …

dτc dn



>0 τc∗

When τ y is relatively large, this lemma indicates that expenditure tax rates fall monotonically along with population growth. However, the income tax rate in this case must be greater than 50% (in case of δc = 1). Assuming a more realistic or empirically found deadweight loss δc of anything between 0 and 0.5, then the income tax rate should be above 66%, which is not or hardly ever observed in reality. Hence, economically, this unambiguous result has little relevance. When τ y is fairly small, this lemma describes a threshold relationship. At lower initial levels of τc , an increase in the population growth rate also leads to higher expenditure tax rates as with income tax rates because at low tax levels, the size of deadweight losses are somewhat small. However, if τc is greater than a certain threshold level, the greater size of deadweight expenditure tax losses drive the decisive voter to opt for lower tax rates. At higher initial levels of τc , an increase in n (a smaller fraction of retirees) decreases the political economy equilibrium expenditure tax rate. The aging of the population affects the equilibrium expenditure tax rate in two directions: a greater number of retirees raises the demand for benefits; meanwhile, the working-age and retired populations are less willing to progress to higher expenditure taxes since both groups must contribute to the welfare state. If the decisive voter is among the working population, then the increased size of the retired population leads to higher expenditure taxes. This situation occurs even if the shift in taxes comes at the price of the greater size of deadweight expenditure tax losses. This is because the decisive voter wants to raise the tax burden on the retired population, and a greater fraction of retirees indicates a larger taxation base arising from spending by the elderly.

4.4.3 The Composition of Taxes Now let us examine the composition of taxes. The ratio of income to expenditure taxes is given by τy (4.19) T ≡ . τc Combining equations (4.11) and (4.15) yields proposition 4.1. The proof of this is in the appendix. Proposition 4.1 The ratio of income to expenditure taxes rises as the rate of population growth rises, and the ratio falls as n falls, that is, dT > 0. dn

4.4 Political-Economy Equilibrium: The Choice of Tax Policy

71

c The mechanism under which the ambiguous sign of dτ is not translated into an dn dT ambiguous sign of dn can be summarized into two aspects. First, Lemma 4.1 shows that the political economy equilibrium income tax rate rises monotonically along with population growth, and the first half of Lemma 4.2 reveals that the equilibrium expenditure tax rate declines monotonically along with population growth if the income tax rate is relatively large. Combining both arguments, the ratio of income to expenditure taxes increases as the rate of population growth rises if the income tax rate is relatively large. Second, if the income tax rate is fairly small, then our analysis can be divided into two sides. As mentioned in the second half of Lemma 4.2, at higher initial levels of τc , a rise in n reduces the political economy equilibrium expenditure tax rate (τc ). Since a rise in n monotonically increases the political economy equilibrium income tax rate (τ y ) as argued in Lemma 4.1, the ratio of τ income to expenditure taxes ( τcy ) increases along with population growth n at higher initial levels of τc . In contrast, as mentioned above, at lower initial levels of τc , an increase in the population growth rate leads to increases in both the political economy equilibrium income tax rate (τ y ) and the expenditure tax rate (τc ). Under the premise that the model invokes the median voter theorem, this framework provides a better tax administration capacity to levy income taxes.10 Hence, the ratio of income to expenditure taxes rises along with population growth in this context. Proposition 4.1 is straightforward but novel. Increases in the fraction of the retired population (or decreases in the population growth rate) lead to increases in consumption taxes relative to income taxes. If the fraction of the retired population expands, then income taxes decline because of the smaller share of the working-age population whose members are net contributing to income taxes with all else being equal. To guarantee the degree of redistribution, there will be a tradeoff between income and expenditure taxes preferred by the decisive voter. Although the political clout of the retired population (whose members presumably prefer labor income taxes) is rising, the working-age population (including the decisive voter) wants to shift the tax-burden onto the retired population, rather than taxed labor income being only from younger individuals, who are acting as net contributors to the welfare system. In sum, the extent to which taxes are levied on income relative to expenditure falls unambiguously along with the share of retirees. For completeness, this paper shall also briefly consider the case in which the median voter is among the retired population. This may occur if population growth happens through longevity rather than fertility, or if older adults become more likely to vote than the young. In this case, the political economy equilibrium income tax rate maximizes the transfer r . By contrast, when the median voter is a member of the working-age population, the political economy equilibrium income tax rate maximizes r plus another term, (1 − τ y )y that is decreasing in τ y .11 Thus, the political economy equilibrium income tax rate “jumps” upward when older adults become the majority; that is, as the young/old balance switches from being younger to being

10

In practice, taxes on income are more progressive in nature than taxes on expenditure. The median voter is not among the retirees, as is probably still the case in all developed countries as well as developing nations.

11

72

4 Population Aging and the Composition of Taxes …

older while the political economy equilibrium expenditure tax rate is independent of the young/old balance if the median voter is among the retired population. In this case, the expenditure tax rate depends on the value of the income tax rate (this means that the expenditure tax rate could be increasing in τ y ).

4.5 Conclusion 4.5.1 Summary This paper theorizes how population aging affects the composition of taxes. The theoretical model extends Razin et al. (2002) to include expenditure taxes in an overlapping generations model, in addition to income taxes. Taxes are levied on labor income and on expenditure, and they finance redistribution policies that benefit two generations (workers and retirees). A key outcome is that a smaller fraction of retirees (an increase in the population growth rate) increases the equilibrium income tax rate, as in Razin et al. (2002). In other words, a greater share of retirees in the population leads to lower income taxes. Conversely, the effect of population aging on expenditure taxes is ambiguous. The outcomes relating to the composition of taxes, defined as the extent of taxes on income relative to expenditure, are novel. Increases in the share of the retired population conceivably lead to reductions in income taxes relative to expenditure taxes. This is consistent with the working-age generation–including the median voter—wanting to rebalance the tax-burden toward retirees—who only pay expenditure taxes in the model. Hence, the degree to which taxes are levied on income relative to expenditure taxes falls unambiguously along with the share of retirees. As such, this paper seeks to contribute to the theoretical political economy literature on the composition of taxes. This idea is motivated by recent rises in taxes on goods and services and concurrent population aging, whereas income taxes are the only source of revenue in the original Razin et al. (2002) hypothesis. Revenue sources outside of income taxes are thus empirically becoming increasingly important components of total revenue (i.e., expenditure taxes account for approximately 30% as a share of total tax revenue in the United Kingdom in recent decades). This is potentially a paradox as it might have been expected that countries with larger numbers of retirees would have higher income taxes and lower expenditure taxes, reflecting the increased political clout of the retired population, whose members presumably prefer income taxes.

4.5 Conclusion

73

4.5.2 Further Research Directions One limitation of this paper is that this study has yet to elaborate on how the results change with Simonovits (2007) critique of Razin et al. (2002). However, this paper still points out further research directions to generalize the theoretical model in three ways: (1) The benefits of the retirees and younger individuals are proportional. (2) Consider a correlation between current contributions and future retired-age benefits. (3) The lifetime total income of any worker is a decreasing function of the tax rate. The analysis of this paper can be applied in the design of the tax system as well as the income taxation literature. Although this paper already quotes several works that discuss the composition of taxes and aging (Pickering and Rajput 2018; Prammer 2019), governments and researchers seldom examine dynamic changes in the tax composition in relation to the age of demographic shifts.

Appendix Derivation of equations (4.11) and (4.15) The problem of the decisive voter is to maximize:    1+n  y¯t , (1 − τc,t+1 )rt+1 . u (1 − τc,t ) (1 − τ y,t )yd + (τc,t + τ y,t ) 2+n The properties of (4.9) indicate that: d y¯t = −δ y y¯t dτ y,t d y¯t = −δc y¯t . dτc,t The first-order condition for the pivotal voter with respect to the labor income tax rate is:  ∂u  1+n 1+n y¯t − (τc,t + τ y,t ) δ y y¯t (1 − τc,t ) − yd + = 0. 2+n 2+n ∂c1,t and the first-order condition for the pivotal voter with respect to the expenditure tax rate is:    1 + n 1+n 1 + n  ∂u (1 − τc,t ) y¯t − (τc,t + τ y,t ) δc y¯t − (1 − τ y,t )yd + (τc,t + τ y,t ) y¯t = 0. 2+n 2+n 2+n ∂c1,t

74

4 Population Aging and the Composition of Taxes …

For a given n, the political equilibrium τ y is constant over time, such that the time subscript t is suppressed henceforth. Thus, the labor income tax rate chosen by the decisive voter yields Eq. (4.11) in the text −yd +

1+n 1+n y¯ − (τc + τ y ) δ y y¯ = 0. 2+n 2+n

and the expenditure tax rate chosen by the decisive voter yields equation (4.15) in the text (1 − τc )

1 + n 2+n

y¯ − (τc + τ y )

 1+n 1+n δc y¯ − (1 − τ y )y + (τc + τ y ) y¯ = 0. 2+n 2+n

Proof of Lemma 4.1 The direction of the effect of changes in n on the equilibrium income tax rate, τ y , is determined by the sign of Eq. (4.14): ∂ V (τ y , n) 1 1 = y¯ − (τc + τ y ) δ y y¯ . 2 ∂n (2 + n) (2 + n)2 1 Given that (2+n) ¯ is positive and exists in each term, the sign of the above equation 2 y depends on 1 − (τc + τ y )δ y .

Utilizing proof by contradiction, if inequation 1 − (τc + τ y )δ y < 0 holds, then after a few mathematical derivations, we can see that inequation τc + τ y >

1 δy

holds. However, considering a more realistic or empirically found deadweight loss δ y of anything between 0 and 0.5, then the sum τc + τ y should be above 2 (in other words, 200%), which is not or hardly ever observed in reality. Therefore, we can obtain Lemma 4.1.

4.5 Conclusion

75

Proof of Lemma 4.2 The direction of the effect of changes in n on the equilibrium expenditure tax rate, τc , is determined by the sign of equation (4.18):   ∂ W (τc , n) 1 1 1 y¯ − (τc + τ y ) δc y¯ − (τc + τ y ) y¯ . = (1 − τc ) 2 2 ∂n (2 + n) (2 + n) (2 + n)2 1 Given that (2+n) ¯ is positive and exists in each term, the sign of the above equation 2 y depends on   (1 − τc ) 1 − (τc + τ y )δc − (τc + τ y ).

After a few mathematical derivations, we can see that the sign of the above equation depends on δc τc2 + (δc τ y − δc − 2)τc + 1 − δc τ y − τ y . This shows a quadratic function with respect to δc , and the parabola opens upwards (given that δc > 0). First, we can check the axis of symmetry: −

δc τ y − δc − 2 b =− 2a 2δc 1 1 τy >1 = + − δc 2 2

given that τ y < 1 and δc < 1. Now, let us consider an extreme case in which τc = 1 (in other words, expenditure is taxed at a rate of 100%). The sign of equation (4.18) depends on: 1 y¯ . −(1 + τ y ) (2 + n)2 c This indicates that if τc = 1, then dτ < 0. dn Second, we can check the intercept of this parabola. Thus, let us now consider another extreme case in which τc = 0 (in other words, expenditure is taxed at a rate of 0%). The sign of equation (4.18) depends on:

(1 − τ y δc − τ y )

1 y¯ . (2 + n)2

1 This indicates that if 1 − τ y δc − τ y < 0 (in other words, τ y > 1+δ ), then we have a c dτc negative intercept. In this case, dn < 0 always holds when the value of τc is between 0 and 1.

76

4 Population Aging and the Composition of Taxes …

1 If 1 − τ y δc − τ y > 0 (in other words, τ y < 1+δ ), then we have a positive intercept. c ∗ c > 0 when the value of τc is This means that τc has a threshold level τc in which dτ dn dτc ∗ between 0 and τc , and dn < 0 when the value of τc is between τc∗ and 1.

Proof of Proposition 4.1 From (4.15), I have (1 − τc )

1 + n 2+n

y¯ − (τc + τ y )

 1+n 1+n δc y¯ = (1 − τ y )y + (τc + τ y ) y¯ . 2+n 2+n

Given that y = y¯ and dividing by y yields (1 − τc )

1 + n 2+n

− (τc + τ y )

1+n  1+n δc = (1 − τ y ) + (τc + τ y ) . 2+n 2+n

From (4.11), substituting for (τc + τ y ) and τ y using τc + τ y = (1 − τc )

1 + n 2+n

1+n 2+n

−

−1  δc = 1 + τc −

δy

1+n −1 2+n 1+n δ 2+n y

1+n 2+n −1 1+n 2+n δ y

+

1+n 2+n

implies −1

δy

.

Solving for τc yields   1+n 1+n 1+n ( 2+n − 1) 2+n (δ y − δc ) − ( 2+n − 1)   . τc = 1+n 1+n 1+n ( 2+n + 1)δ y − ( 2+n − 1)δc 2+n Substituting into τ y =

1+n 2+n −1 1+n 2+n δ y

− τc yields

    1+n 1+n 1+n 1+n 1+n 1+n ( 2+n − 1) ( 2+n + 1)δ y − ( 2+n − 1)δc − δ y ( 2+n − 1) 2+n (δ y − δc ) − ( 2+n − 1)   . τy = 1+n 1+n 1+n ( 2+n δ y 2+n + 1)δ y − ( 2+n − 1)δc

Simplifying the above two equations yields T ≡

τy (3 + 2n)δ y + δc  −1 =  τc δ y (1 + n)(δ y − δc ) + 1

References

77

The first-order condition with respect to n yields:     2 2δ (1 + n)(δ δ y (δ y − δc ) − δ ) + 1 − (3 + 2n)δ + δ y c y c y dT =   2 dn δ y (1 + n)(δ y − δc ) + 1 δ 2y (2 − δ y ) + δ y δc2 =  2 > 0 δ y (1 + n)(δ y − δc ) + 1 given that 0 < δ y < 1 and 0 < δc < 1.

References Acemoglu D, Johnson S, Robinson JA, Yared P (2008) Income and democracy. Am Econ Rev 98(3):808–42 Acemoglu D, Restrepo P (2017) Secular stagnation? The effect of aging on economic growth in the age of automation. Am Econ Rev Papers Proc 107(5):174–179 Aidt TS, Jensen PS (2009) The taxman tools up: an event history study of the introduction of the personal income tax. J Public Econ 93(1–2):160–175 Aigner-Walder B, Döring T (2012) The effects of population ageing on private consumption—a simulation for Austria based on household data up to 2050. Eurasian Econ Rev 2(1):63–80 Andersen TM (2012) Fiscal sustainability and demographics—should we save or work more? J Macroeconomics 34(2):264–280 Balassone F, Cunha J, Langenus G, Manzke B, Pavot J, Prammer D, Tommasino P (2011) Fiscal sustainability and policy implications: a post-crisis analysis for the euro area. Int J Sustain Econ 3(2):210–234 Besley T, Persson T (2014) Why do developing countries tax so little? J Econ Perspect 28(4):99–120 Conesa JC, Kehoe TJ, Nygaard VM, Raveendranathan G (2020) Implications of increasing college attainment for aging in general equilibrium. Eur Econ Rev 122:103363 Creedy J, Enright J, Gemmell N, Mellish A (2010) Population ageing and taxation in New Zealand. NZ Econ Papers 44(2):137–158 Disney R (2007) Population ageing and the size of the welfare state: is there a puzzle to explain? Eur J Political Econ 23(2):542–553 European Commission (2018) The 2018 ageing report: economic and budgetary projections for the EU Member States (2016–2070). Institutional paper 079 Felix A, Watkins K (2013) The impact of an aging US population on state tax revenues. Econ Rev 4:95–127 Ferraro D, Fiori G (2020) The aging of the baby boomers: demographics and propagation of tax shocks. Am Econ J: Macroeconomics 12(2):167–93 Galasso V, Profeta P (2007) How does ageing affect the welfare state? Eur J Political Econ 23(2):554–563 Grandmont JM (1978) Intermediate preferences and the majority rule. Econometrica: J Econometric Soc 317–330 Guest R (2006) Population ageing, fiscal pressure and tax smoothing: a CGE application to Australia. Fiscal Stud 27(2):183–203 Guest RS, McDonald IM (2000) Population ageing and projections of government social outlays in Australia. Austr Econ Rev 33(1):49–64

78

4 Population Aging and the Composition of Taxes …

Keen M, Lockwood B (2010) The value added tax: its causes and consequences. J Dev Econ 92(2):138–151 Kudrna G, Tran C, Woodland A (2019) Facing demographic challenges: pension cuts or tax hikes? Macroeconomic Dyn 23(2):625–673 Lee SH, Kim J, Park D (2017) Demographic change and fiscal sustainability in Asia. Soc Indicators Res 134(1):287–322 Luo W (2018) Essays on inequality and fiscal policy. PhD thesis, University of York Luo W (2019) Demography and the composition of taxes: evidence from international panel data. Econ Lett 183:108518 Luo W (2020) Demography and economic growth: the effect of tax composition. Appl Econ Lett 27(20):1629–1634 Luo W, Zhu J (2021) Youthful dependents and economic growth: the effect of tax composition. Appl Econ Lett 28(8):675–680 Luo W (2022) Tax composition and economic growth in the age of demographic change. Policy Stud 1–20 Meltzer AH, Richard SF (1981) A rational theory of the size of government. J Political Econ 89(5):914–927 Persson T, Tabellini G (2002) Political economics: explaining economic policy. MIT press Pickering A, Rajput S (2018) Inequality and the composition of taxes. Int Tax Public Finance 25(4):1001–1028 Prammer D (2019) How does population ageing impact on personal income taxes and social security contributions? J Econ Ageing 14:100186 Razin A, Sadka E, Swagel P (2002) The aging population and the size of the welfare state. J Political Econ 110(4):900–918 Saint-Paul G (1994) Unemployment, wage rigidity, and the returns to education. Eur Econ Rev 38(3–4):535–543 Sanz I, Velázquez FJ (2007) The role of ageing in the growth of government and social welfare spending in the OECD. Eur J Political Econ 23(4):917–931 Shelton CA (2008) The aging population and the size of the welfare state: is there a puzzle? J Public Econ 92(3–4):647–651 Simonovits A (2007) Can population ageing imply a smaller welfare state? Eur J Political Econ 23(2):534–541 Van Sonsbeek JM (2010) Micro simulations on the effects of ageing-related policy measures. Econ Model 27(5):968–979 Yashio H, Hachisuka K (2014) Impact of population aging on the personal income tax base in Japan: simulation analysis of taxation on pension benefits using micro data. Public Policy Rev 10(3):519–542

Chapter 5

Population Aging and the Composition of Taxes: Evidence from International Panel Data

This chapter is originally published in Economics Letters, 2019, 183, p. 108518.

5.1 Introduction How does population aging affect fiscal policy? Razin et al. (2002), building on Meltzer and Richard (1981), theorize that increases in the dependency ratio lead to lower labor tax rates and a reduction in the generosity of social transfers in democracies. The logic is that the median voter is a worker, who increasingly dislikes redistributing as the retired population increases. This paper develops the Razin et al. (2002) hypothesis to consider the composition of taxes, in particular the setting of income versus expenditure taxes. The main theoretical prediction is that the extent of taxes on income relative to taxes on expenditure falls with population aging. The logic is similar to Razin et al. (2002). Income taxes are paid solely by workers, whilst expenditure taxes are paid by both generations. If the median voter is a worker, then increasing the size of the retired population compels a shift in tax composition towards expenditure taxes.1 International panel evidence supports this hypothesis. Empirical evidence generally has not supported the Razin et al. (2002) hypothesis, which focuses only on income taxes. For example, Fig. 5.1 depicts the raw correlation between income taxes as a portion of total revenues over 1990–2014 and the ratio of the population above 65 to the population between 15 and 64, indicating no negative association between aging and the level of income taxes.2 The argument proposed in this paper is motivated by recent rises in taxes on goods and services, and concurrent population aging. Consequently, revenue sources outside of income taxes (i.e. expenditure taxes) are empirically becoming increasingly 1

Please find a simple theoretical model in Chap. 4 in Luo (2018). On the other hand, Disney (2007), Sanz and Velazquez (2007), and Shelton (2008) all find a positive relationship between population aging and the size of the welfare state in regression analyses.

2

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 W. Luo, Inequality, Demography and Fiscal Policy, Applied Economics and Policy Studies, https://doi.org/10.1007/978-981-99-0518-8_5

79

80

5 Population Aging and the Composition of Taxes: Evidence …

Income taxes as a portion of total revenues 60 20 40 0

Income Taxes Versus Population Aging

0

10

20

30

Ratio of old to young taxes_ipcg_revenue

Fitted values

Fig. 5.1 Correlation between aging and income taxes (cross-country data over 1990–2014). Aging is defined as the ratio of the population above 65 years old to the population between 15 and 64

important components of total revenue.3 This is potentially a paradox, as it might have been expected that countries with a relatively large retired population would have higher labor income taxes and lower expenditure taxes, reflecting the increased political clout of the retired population who presumably would prefer labor income taxes.4 In an overlapping generations model, taxes are levied on both income and expenditure, which finance redistribution to both the working and retired population, with a balanced budget period by period (see Luo 2018). In a median voter framework, the observed tax rate is the median voter’s ideal income-to-expenditure tax ratio. Although population aging leads to an increase in the number of voters who prefer a higher ratio, the preference of the median voter shifts the other way as the median voter wants to increase the tax burden on retirees, instead of being taxed income from the young who solely contribute to the welfare system. 3

For example, it accounts for approximately 30% as a share of the total tax revenue in the United Kingdom in recent years. 4 Given the increased empirical importance of expenditure taxes this paper examines how population aging affects the composition of taxes. Despite the fact that there is an enormous literature focusing on optimal taxation, starting with Diamond and Mirrlees (1971), relatively little of this literature provides a positive analysis of the political-economic determination of tax composition. One related work by Pickering and Rajput (2018) links the composition of taxes to inequality. Other related work analyzes the adoption of tax instruments with the development process (see Aidt and Jensen 2009; Keen and Lockwood 2010), but this literature neglects the effect of demography on the composition of taxes.

5.2 Data and Econometric Specification

81

5.2 Data and Econometric Specification This empirical analysis focuses on a panel of international data. Cross-country annual data on income and expenditure tax revenue over the period 1990–2014 are available from the World Development Indicators (hereafter WDI). In spite of over 20 years, there is much more variation in both demography and the policy data across rather than within countries.5 Following Pickering and Rajput (2018), the main dependent variable is the ratio τ of income to expenditure taxes, T = τcy , constructed by the ratio of taxes on income, profits and capital gains (as a share of total tax revenue) to taxes on goods and services (as a share of total tax revenue). Both are extracted from the WDI database.6 In practice (and also within countries) rates of tax vary with different types of income and goods, but the measure of ratio proposed is a way to capture the extent to which taxes are levied on income relative to expenditure. Due to the relatively small value in the data of taxes on goods and services in the case of some countries, following Pickering and Rajput (2018) I use the natural logarithm of T , ln(T ), in the below regression analysis. The argument proposed predicts that the extent of taxes on income relative to expenditure declines with an increased fraction of the retired population. The measure of the proportion of retirees in the population used is the percentage of the population over the age of 65 (PROP65). This measure of the retired fraction is preferable to the dependency ratio used by Razin et al. (2002). The dependency ratio includes children as well as retirees which would have different impacts on taxes, as shown by Shelton (2008). One important determinant of the composition of taxes is the level of development. To capture tax base effects as argued by Kenny and Winer (2006), I include the natural logarithm of GDP per capita in constant chained PPP US$ (ln(y)) as a first control in the regression analysis. To fully capture demographic effects, the econometric analysis includes the percentage of the population between 15 and 64 years of age (PROP1564) as an additional control. Pickering and Rajput (2018) find that the extent to which taxes are levied on income relative to expenditure rises with inequality. Therefore, the inequality measure (UTIP), obtained from the University of Texas Inequality Project’s Estimated Household Income Inequality data of Galbraith and Kum (2005), is also included as a control. Governments collect tax revenue through means beyond taxation on income and consumption. One important source is the revenue from import duties and tariffs due to openness (Rodrik 1998). Thus the trade share (TRADE) is also employed in the 5

For example, the standard deviation of cross-country retired fraction is 4.62, whilst the withincountry standard deviation is 0.956. For the main policy variable (ln(T )—described below), the standard deviation of the cross-country is 1.04, whilst the within-country standard deviation is 0.383. 6 According to the International Monetary Fund, for most countries central government finance data have been consolidated into one account, but for others only budgetary central government accounts are available. The aggregation method (to address the progressivity of the tax rate) is calculated at the median household.

82

5 Population Aging and the Composition of Taxes: Evidence …

Table 5.1 Descriptive statistics Obs

No. of countries

Mean

Std. dev. (overall)

Std. dev. (between)

Std. dev. (within)

Min

Max

ln(T )

2091

157

−0.261

1.00

1.04

0.383

−4.00

5.09

τy

2142

158

22.45

12.75

12.12

5.06

0.349

75.24

τc

2141

160

29.06

13.76

13.47

5.48

0.024

89.22

PROP65

4633

194

7.01

4.72

4.62

0.956

0.335

25.08

PROP1564 4633

194

61.20

7.01

6.57

2.48

45.29

85.81

UTIP

1556

129

42.79

6.71

6.49

2.20

22.75

59.96

ln( y )

3652

166

8.58

1.29

1.27

0.287

5.03

11.73

TRADE

4263

191

86.94

51.96

47.42

ln(POP)

5302

213

15.07

2.35

2.36

20.30 0.142

0.309

531.7

9.11

21.04

POLITY 2

3790

165

3.04

6.69

6.20

2.46

−10

10

YGAP

4668

198

0

0.034

0.002

0.034

−0.609

0.505

Notes τ y denotes taxes on income, profits and capital gains as a percentage of revenue—taken from the World Development Indicators (WDI). τc denotes taxes on goods and services as a percentage τ of revenue—taken from the WDI. T = τcy . PROP1564 and PROP65 are respectively the proportion of the population aged between 15 and 64, and 65 and above—taken from the WDI. UTIP is the University of Texas Inequality Project’s Estimated Household Income Inequality. y is real GDP at chained PPPs in millions of 2005 US dollars per capita—taken from the Penn World Tables. TRADE is the sum of exports and imports as a percentage of GDP—taken from the WDI. POP is the size of country population—taken from the WDI. POLITY 2 is a measure of democracy provided by the Polity IV project, with −10 denoting the highest level of autocracy, and 10 denoting the highest level of democracy. YGAP is the difference between the actual output and its trend value in percentage

regression analysis. In addition to these control variables the natural logarithm of the total population size (ln(POP)) is also included, thus capturing to some extent any scale (dis-)economies related to particular forms of tax collection. There may also be cyclical movements in policy variables. To address this potential problem the regression analysis includes the output gap (Y G A P) as a further control. The policy variables may also be affected by the degree of democracy either directly through channels other than those mentioned above, or indirectly serving as a proxy for tax capacity. In democracies with low quality institutions the link between the median voter and policy is blurred, while in countries with stronger institutions the median’s influence is stronger. Thus the democracy score provided by the Polity IV project is included as a final control (POLITY 2). Table 5.1 contains descriptive statistics of the variables used in the regression analysis.7

7

Note that there is considerable dispersion in how countries raise their tax revenue. Over the sample period taxes on income on average represent a smaller fraction of total revenue than taxes on goods and services. This indicates that the capacity to raise revenue through income taxes is normally limited in those countries with relatively low income. For instance among OECD members, taxes on income are on average 32% of total revenue, whilst outside the OECD income taxes account for just 20% of total revenue on average.

5.3 Evidence

83

5.3 Evidence 5.3.1 Baseline Estimation This section is to test whether and how the ratio of income to expenditure taxes across different countries systematically changes with the fraction of the population that is retired.8 Column 1 of Table 5.2 is a simple specification with just the fraction of the retired population (PROP65) and GDP per capita as regressors using annual data OLS regression, with robust standard errors clustered by country. Column 2 extends the regression of column 1 to include country fixed effects. In these specifications the sign of the coefficient estimate relating to the fraction of the retired population is negative in all cases, and all are statistically significant at the 1% level. This is consistent with the argument—an increase in the retired fraction increases expenditure taxes relative to income taxes. Columns 3 and 4 repeat the analysis of columns 1 and 2 using full controls instead. The results using panel estimation with full control variables support those already found. The estimated statistical significance of the fraction of the retired population is unaffected and even remains at the 5% level in column 4. Using the estimate from column 4 of Table 5.2, a one standard deviation increase in the fraction of the retired population is statistically associated with a fall of 0.63 in the ratio of income to expenditure taxes, holding all else equal. The magnitude of this estimated correlation is sizable—implying more than a half of the raw standard deviation in the policy variables. Further, columns 5 and 6 instead use the ratio of the population above 65 against those between the ages of 15 and 64, RATIO, to measure population aging, and mimic columns 3 and 4. The results similarly demonstrate an increased tendency to use expenditure taxes rather than income taxes as population aging increases.9

5.3.2 Further Estimation It is natural to investigate whether or not the reported results change with the degree of democracy, given that the argument invokes on the median voter framework. Columns 1 and 2 of Table 5.3 thus extend the regression results by splitting the sam8

This paper also separately examines how different categories of tax measures respectively comove with the fraction of retirees. I also report results from cross-country regressions with data measured by within-country averages. Both results emphasize the need and motivation to examine the relationship between population aging and the extent to which taxes are levied on income relative to expenditure, instead of income taxes only. 9 I also include central government debt levels as a further control and obtain similar results. In Table 5.2, the results relating to the control variables are of some interest. There is a positive relationship with income per capita, which likely indicates greater potential to increase income taxes in richer countries. In addition, more populous countries are found to have greater reliance on income taxes (when their estimated coefficients are significant), which reflects more labor force and thus more labor income to be taxed.

1849 146 Panel No 0.135

1849 146 Panel Yes 0.070

0.584*** (0.218)

−0.0899*** (0.0423)

0.399*** (0.120)

(2)

(1)

−0.102*** (0.0267)

−0.00553 (0.0144) 0.795*** (0.147) 0.001000 (0.00138) 0.170*** (0.0557) −0.0130 (0.0213) −0.00379 (0.941) 796 87 Panel No 0.414

−0.0891*** (0.0213) −0.0810*** (0.0182)

(3)

−0.0126 (0.0184) 1.072*** (0.231) 0.000615 (0.00210) −0.000519 (0.608) −0.0136 (0.0123) −0.932 (0.734) 796 87 Panel Yes 0.172

−0.134** (0.0510) −0.0525** (0.0245)

(4)

−0.0752*** (0.0186) 0.00258 (0.0190) 0.623*** (0.172) −0.00142 (0.00154) 0.109* (0.0574) −0.0272 (0.0218) −0.739 (0.939) 796 87 Panel No 0.314

(5)

−0.0656* (0.0335) −0.0162 (0.0222) 0.754*** (0.189) 0.000791 (0.00201) −0.517 (0.608) −0.0205* (0.0108) −1.002 (0.817) 796 87 Panel Yes 0.130

(6)

Notes Table 5.2 contains results using OLS regressions of the composition of taxes, ln(T ), including PROP1564, UTIP, ln(y), TRADE, ln(POP), POLITY 2, and YGAP as control variables described in the text. Columns (5) and (6) instead use the ratio of the population above 65 to those between the ages of 15 and 64, RATIO, to measure population aging, and mimic columns (3) and (4). Robust standard errors are shown in parentheses. Standard errors are clustered by country. *, **, and *** respectively denote significant levels at 10%, 5% and 1%

Observations Countries Data Fixed effects? R2

YGAP

POLITY 2

ln(POP)

TRADE

ln(y)

UTIP

RATIO

PROP1564

PROP65

Table 5.2 Basic estimation results—the composition of taxes

84 5 Population Aging and the Composition of Taxes: Evidence …

5.3 Evidence

85

ple by levels of democracy (depending on the median value of democracy score). Column 1 (column 2) contains results for countries with stronger (weaker) democratic credentials. When the sample is separated it becomes clear that the negative relationship between the fraction of the retired population and the ratio of income to expenditure taxes holds only in the subsample of democratic regimes.10 The use of an interaction term provides an alternative way to examine how the results change with the extent of the franchise. In column 3, POLITY 2 is then multiplied by the aging measure, thereby generating an interaction term. The hypothesis here is that the relationship between the tax composition measure and aging will be increasingly negative under democracies, hence that the coefficient estimate for the interaction term is negative. The estimation results confirm this, although the significance level declines.11 The political equilibrium is preferred as a balance between those who receive and those who contribute from a tax-transfer policy. While there is a rise in the number of the retired population who prefers a higher ratio, the preference of the median voter shifts the other way, thus ensuring that the ratio falls, as the median voter wants to shift the tax-burden onto the retired population, rather than being taxed labor income only from the young who are acting as a net contributor to the welfare system. Moreover, the median voter more plausibly drives policy under stronger democracy, consistent with the empirical results. In contrast, policy is less likely to respond to changes in the preference of the median voter the less democratic is the country. Note that it is also of interest to ask whether there are other stories, such as the Laffer Curve, which might explain the recent tendency to increase expenditure taxes. The Laffer Curve suggests that when income tax rates increase from low levels, the tax revenue collected by government also increases. If tax rates keep increasing after a certain point, then it would cause people not to work as hard as before, thereby reducing tax revenue. Columns 4 and 5 of Table 5.3 split the sample with stronger democratic credentials by τ y (determined by the median value of τ y ). If the story of the Laffer Curve could explain, when τ y increases from lower to higher levels, then the tax revenue generated by taxes on income relative to expenditure firstly increases and then declines. This indicates that the sign of the coefficient on PROP65 should be reversed at lower and higher levels of τ y , holding all else equal. Statistical significance in columns 4 and 5 implies that the estimates are stable across these subsamples, which in turn supports the argument proposed. It is also natural to see whether the results vary with the level of development. Columns 6 and 7 of Table 5.3 split the sample by levels of GDP per capita (determined by the median value of GDP per capita). As can be seen in all cases, the ratio of income to expenditure taxes is negatively correlated with the fraction of the retired population. However, this negative relationship only holds to a significant degree in 10

In column 1 the p-value for the estimated coefficient for the fraction of the retired population is 1.1%, and the estimated effect is sizable: A one standard deviation increase in the fraction of the retired population is statistically associated with the policy variable ln(T ), which is smaller by 0.78, holding all else equal. 11 I obtain essentially identical results if I entered a dummy variable that takes the value of 1 when POLITY 2 is positive and the value of 0 when the index is negative in the interaction term.

0.000446 (0.0211)

−0.0231 (0.0206)

1.627*** (0.464)

0.00132 (0.00258)

1.824* (1.077)

UTIP

ln(y)

TRADE

ln(POP)

POLITY 2

0.00664 (0.0214)

−0.0111 (0.0207)

−0.154 (1.012)

0.0141 (0.157)

−1.751 (1.400)

496

Panel

Yes

High POLITY 2

0.278

YGAP

Observations

Data

Fixed effects?

Sample

R2 0.177

Full

Yes

Panel

796

0.248

High POLITY 2 and high τ y

Yes

Panel

273

−0.451 (1.129)

0.397 (0.365)

0.715 (1.340)

0.00486 (0.00449)

0.892* (0.447)

−0.0231 (0.0169)

−0.106** (0.0496)

−0.111* (0.0580)

(4)

0.347

High POLITY 2 and low τ y

Yes

Panel

223

−1.662 (1.412)

−0.154** (0.0717)

0.214

High income

Yes

Panel

559

−2.434* (1.239)

−0.0249* (0.0142)

−0.955 (1.228)

0.000949 (0.00219)

−0.000184 (0.00258) 3.177** (1.347)

1.171*** (0.412)

−0.0437* (0.0252)

−0.0253 (0.0287)

−0.129** (0.0620)

(6)

1.596*** (0.293)

−0.00984 (0.0237)

−0.0778* (0.0421)

−0.124*** (0.0405)

(5)

0.215

Low income

Yes

Panel

237

−0.873 (0.850)

−0.00594 (0.0131)

0.747 (0.811)

0.00140 (0.00354)

0.967*** (0.328)

0.0199 (0.0189)

−0.100** (0.0432)

−0.128 (0.248)

(7)

Notes Regression specification is the same as column 4 of Table 5.2. Columns (1) and (2) respectively correspond to higher and lower democracy levels. Column (3) includes an interaction term described in the text. Columns (4) and (5) respectively correspond to higher and lower levels of τ y under the sample with higher democracy level. Columns (6) and (7) respectively correspond to higher and lower levels of income

0.201

Low POLITY 2

Yes

Panel

300

−0.0619 (0.613)

−0.985* (0.512)

−0.896 (0.735)

0.000473 (0.00215)

1.080*** (0.236)

−0.0117 (0.0185)

−0.0566** (0.0262)

−0.00328 (0.00207)

−0.0971* (0.0528)

(3)

0.000853 (0.00278)

0.876*** (0.298)

−0.0115 (0.0166)

−0.117 (0.0955)

−0.119** (0.0506)

(2)

(1)

−0.164** (0.0628)

PROP1564

PROP65*POLITY 2

PROP65

Table 5.3 Estimation results—the composition of taxes

86 5 Population Aging and the Composition of Taxes: Evidence …

References

87

the group of countries with higher income level. Rich countries have a proportionally larger retired population, and therefore tax revenue collected by taxes on income is decreased relative to expenditure.12

5.4 Conclusion This paper analyzes how population aging affects the composition of taxes. The extent of taxes on income relative to expenditure is found to be negatively associated with the fraction of the retired population. The empirical results hold across various econometric specifications employed. In particular, the results hold significantly in countries with strong democratic credentials.

References Aidt TS, Jensen PS (2009) The taxman tools up: an event history study of the introduction of the personal income tax. J Public Econ 93(1–2):160–175 Diamond PA, Mirrlees JA (1971) Optimal taxation and public production I: production efficiency, and II: tax rules. Am Econ Rev 61(1):8–27; (3):261–278 Disney R (2007) Population ageing and the size of the welfare state: is there a puzzle to explain? Eur J Polit Econ 23(2):542–553 Galbraith JK, Kum H (2005) Estimating the inequality of household incomes: a statistical approach to the creation of a dense and consistent global data set. Rev Income Wealth 51(1):115–143 Keen M, Lockwood B (2010) The value added tax: its causes and consequences. J Devel Econ 92(2):138–151 Kenny LW, Winer SL (2006) Tax systems in the world: an empirical investigation into the importance of tax bases, administration costs, scale and political regime. Int Tax Public Finan 13(2):181–215 Luo W (2018) Essays on inequality and fiscal policy. PhD thesis, University of York Meltzer AH, Richard SF (1981) A rational theory of the size of government. J Polit Econ 89(5):914– 927 Pickering A, Rajput S (2018) Inequality and the composition of taxes. Int Tax Public Finan 25(4):1001–1028 Razin A, Sadka E, Swagel P (2002) The aging population and the size of the welfare state. J Polit Econ 110(4):900–918 Rodrik D (1998) Why do more open economies have bigger governments? J Polit Econ 106(5):997– 1032 Sanz I, Velázquez FJ (2007) The role of ageing in the growth of government and social welfare spending in the OECD. Eur J Polit Econ 23(4):917–931 Shelton CA (2008) The aging population and the size of the welfare state: is there a puzzle? J Public Econ 92(3–4):647–651 12

In column 6 the p-value for the estimated coefficient for the fraction of the retired population is 4.1%, and the estimated effect is sizable: a one standard deviation increase in the fraction of the retired population is statistically associated with a reduction of 0.61 in the policy variable ln(T ). As a robustness check, panel regressions using RATIO as the key independent variable, testing whether the results change with the degree of democracy and the level of development, produce essentially identical results.

Chapter 6

Youthful Dependents and the Composition of Taxes

This chapter is coauthored with Xiaoge Zhang at School of Finance, Southwestern University of Finance and Economics.

6.1 Introduction How does age structure affect fiscal policy? One crucial work is a politico-economic model designed by Razin et al. (2002), who argue that greater dependency ratio results in lower income tax rates and a fall in the generosity of social transfers in democracies, since the median voter is among the working-age population who increasingly feels distaste for redistribution when the retired population rises. Empirical evidence generally has not supported the Razin et al. (2002) hypothesis, which focuses only on income taxes. For example, Disney (2007), Sanz and VealAzquez (2007), and Shelton (2008) all find a positive relationship between population aging and the size of the welfare state in regression analyses. In response to this, a set of theoretical literature has appeared through which a larger proportion of retirees can lead to higher income taxes and more generous social transfers. Galasso and Profeta (2007) propose that population aging has a political effect: the median voter becomes older and hence more willing to support a larger system. Further theoretical analyses studying welfare states and aging, as also argued by Simonovits (2007), extend beyond the mechanism analyzed in Razin et al. (2002), trying to link increasing income taxes to aging. Unlike both empirical and theoretical literature mentioned which concerns an upward trend of the fraction of retired population, this paper instead pays attention to youthful dependents as argued by Shelton (2008), since the current fraction of youthful dependents also plays a significant role in the future demographic structure. Luo and Zhu (2021), based on Shelton (2008), extend the three-generation framework to consider how the young population affects economic growth.

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 W. Luo, Inequality, Demography and Fiscal Policy, Applied Economics and Policy Studies, https://doi.org/10.1007/978-981-99-0518-8_6

89

90

6 Youthful Dependents and the Composition of Taxes

The theory put forward in this paper is also motivated by recent increases in taxes on goods and services. Revenue sources outside of income taxes are thus empirically becoming increasingly important components of total revenue. Moreover, a series of recent literature generally focuses on the composition of taxes instead of income taxes only, given a declining status of tax revenue coming from income. For instance, Pickering and Rajput (2018) analyze how income inequality affects the extent to which taxes are levied on income relative to expenditure. Luo (2018, 2019) then extend the Pickering and Rajput (2018) framework to explore the relationship between demography and the composition of taxes. Luo (2020) further considers the impact of demographic structure on economic growth through the mechanism of tax composition. Hence this paper develops the Pickering and Rajput (2018) framework to consider how the younger population affects the composition of taxes, in particular the setting of income versus expenditure taxes. The main theoretical prediction is that the extent of taxes on income relative to expenditure rises with youthful dependents. The logic is that unbalanced fertility in the demographic structure leads to greater income inequality (as argued by Dahan and Tsiddon 1998), bringing about more demand for income taxes rather than expenditure taxes, as the median voter is relatively poor and s/he dislikes being taxed (again) at the consumption-side. Theoretically, we extend the median-voter theory to a dynamic framework. The model features unbalanced fertility across the distribution of income level. We assume the richer is less willingly to produce many offspring. With this assumption, the youthful dependents inevitably become larger over time since fertility is unevenly distributed. In the meantime, the position of the median voter becomes increasingly poor when the inequality is raising. As usual, he/she separately weights income and expenditure tax between the utility undermined by the taxation and utility gained from redistribution. However, in general the model predicts that the median voter dynamically becomes more and more in favour of income tax rather than expenditure tax which corresponds to the trend of larger proportion of youth dependents. The relationship between youthful dependents and the extent of taxes on income relative to taxes on expenditure is investigated empirically using a panel of over 100 countries over the period 1990–2014. Following Pickering and Rajput (2018) and Luo (2019), the dependent variable is constructed by the ratio of taxes on income, profits and capital gains (as a share of total tax revenue) to taxes on goods and services (as a share of total tax revenue), and the key explanatory variable is the ratio of the population either below 14 or above 65 to the population between the ages of 15 and 64. These data are all taken from the World Development Indicators database. Consistent with the theory, data measuring the extent to which taxes are levied on earnings relative to expenditure are consistently positively correlated with the measure of youthful dependents. This relationship is robust across different econometric specifications employed. In the panel estimation with fixed effects, a one-standarddeviation-rise in the ratio of children to workers is statistically associated with a rise of 0.62 in the ratio of income to expenditure taxes, holding all else equal. The magnitude of this estimated correlation is sizable—showing more than a half of the raw standard deviation in the policy variable. The statistical relationship holds

6.2 The Model

91

most significantly in countries with higher levels of democracy, in support of the mechanism proposed in this paper. Moreover we also separately examine how the different tax instruments co-move with the proportion of youth. The data indicate a positive correlation between income taxes and the fraction of the youth population, and a negative correlation between expenditure taxes and the fraction of the youth population. The rest of the paper is organized as follows. The next section provides a theoretical analysis of the political economics of demography and tax composition. Section 6.3 contains the empirical analysis, and Sect. 6.4 concludes.

6.2 The Model 6.2.1 Theoretical Framework In this section, we firstly develop a theoretical framework of the measure of inequality under our uneven fertility assumption. Then we combine the our framework with the median voter theorem to establish the relation between demographic structure and tax preference. We set an economy with continuum of individuals i spreading on an income distribution between (0, 1). i represents the quantile of the individual’s income in the population. For instance, i = 0.99 represents the individual who is richer than 99% of the population. Time t is discrete and starts from 0. At time 0, the economy is endowed with population N0 . Each individual is endowed with income W0 . We assume the richer (poorer) is of a higher income growth rate and a lower (higher) fertility rate. Across countries, there is a strong negative correlation between GDP and fertility. Among many others, Croix and Matthias (2003) discuss this relation. The intuition is that high birth rates tend to place a greater burden of child rearing and education on populations already struggling with poverty. Moav (2004) develops a model to show that the poor will choose high fertility rates with low investment in child human capital. The intuition is that educated individuals (the rich) with relatively high human capital have a comparative advantage in raising educated children. Hence they place higher priority in child quality than quantity. Henceforce, for i, we directly set the offspring reproducing rate as e1−i . The population is ⎛ N (t) = N0 ⎝

1

⎞t e1−i di ⎠ = N0 (e − 1)t

(6.1)

0

At the beginning of each time period, all descendants share wealth from their ancestors. Then during the period, their wealth will grow at a rate of gi,t . Let us start with the simplistic case of gi,t = 1 for all t and i, which indicates that income per capita will not grow. Later, we will discuss the relaxation of this assumption. Under this condition, the average income is W0 / (e − 1)t .

92

6 Youthful Dependents and the Composition of Taxes

The dynamics of the median income is slightly more difficult to obtain. Since the reproducing is uneven, an individual’s quantile position q is changing over time. In the basic case without income growth, the dynamics of an individual’s quantile position q is captured by ⎛ 1 ⎞ qt−1  e1−i di = qt ⎝ e1−i di ⎠ 0



qt−1

0

e = ln qt (1 − e) + e

 (6.2)

For example, in the last period, an individual is at the median position (qt−1 = 50%). Due to the heavier “wealth dilution” on the left hand-side, his offspring will be at the position of qt = 62.2% in this period. One period later, the next generation will be at the position of qt+1 = 73.3%. Further, with Eq. (6.2), we have the following lemma to describe the dynamics of the median income. Lemma 6.1 The income of the median (M E D t ) individual in time t in this economy follows M E Dt =

W0 t−1 t− τ =0 qτ e

(6.3)

where the series of {qt }∞ t=0 follows that  qt = 50%; qt−1 = ln

e qt (1 − e) + e



Additionally, M E D t is decreasing over time t. Proof For a specific time period t, one can set qt = 0.5 and obtain {qt }tt=0 by backward calculation through Eq. (6.2). Then the income of the median (M E D t ) individual in time t follows, M E Dt =

W0 1−q 1−q 0 1 e e e1−q2

. . . e1−qt−1

=

W0 t−1 t− τ =0 qτ e

(6.4)

Obviously, with the increasing of time t, there are more terms in the denominator of Eq. (6.4). For larger t, the corresponding q0 will be closer and closer to 0. Therefore, M E D t is decreasing with time t.  With Lemma 6.1, we can establish the measure of inequality m as the ratio between average income and median income in the following proposition. Proposition 6.1 The measure of inequality m, defined as the ratio of average income to median income, is

6.2 The Model

93

mt =

W0 / (e − 1)t W0 t−1

et−

τ =0

(6.5)

qτ

where the series of {qt }∞ t=0 follows that  qt = 50%; qt−1 = ln

e qt (1 − e) + e



m t is increasing with time t. Proof Since 1 − e < 0, we know that qt−1 and qt have a positive relation within the range of qt ∈ (0, 1). The series of {qt }tt=0 is monotonically increasing with t. Therefore, with qt = 0.5, a larger t means a smaller q0 . Now we consider m t , mt =

W0 / (e − 1)t e1−qt−1 e1−qt−2 · · · e1−q0 = W0 (e − 1)t

(6.6)

t−1 et− τ =0 qτ

According to Eq. (6.2), with qt = 0.5, e1−qt−1 ≈ 1.86 > e − 1. Hence we conclude that the ratio of average to median income, m t , will increase with time t.  Additionally, we discuss relaxing the assumption of growth rate of income per capita gi,t . Firstly, it is clear that if the growth rate is even (i.e., gi,t = gt for all i), the conclusion of Proposition 6.1 will not be penetrated since the average income and median income grow at the same rate. Secondly, if the per capita income growth rate, gi,t , is uneven across i, the existence of Proposition 6.1 depends on the distribution of gi,t across i. However, as long as the income per capita growth rate, gi,t , dose not smooth out the inequality brought by the uneven fertility, we believe that Proposition 6.1 is still reasonable. Here, to keep the model parsimonious, we want to avoid exploring more depth of the relation between growth rate of income and inequality. Since we do not see the significant amelioration in the inequality in recent decades, we adopt Proposition 6.1 and continue our analysis.1

6.2.2 Demographic Structure and Tax Preference Further, in any moment of t, the median voter theory applies. Based on Meltzer and Richard (1981), Pickering and Rajput (2018) develop a model that utilizes the median-voter framework to explain the relationship between inequality, income taxes and expenditure taxes. Intuitively, for any given level of total taxation, richer 1

For detailed data of the trend of the world-wide inequality measure, one can visit https:// ourworldindata.org/income-inequality.

94

6 Youthful Dependents and the Composition of Taxes

(poorer) individuals prefer increasing amounts of expenditure (income) taxes relative to income (expenditure) taxes, and the median voter decides. According to their model, for given moment of t, the median voter faces the consumption maximization problem of

 Max cm = (1 − τc ) 1 − τ y ym + r τc , τ y

(6.7)

subject to the constraints of r = τc x¯ + τ y y¯ x¯ = y¯ y¯ = y ∗ e−δ y τ y −δc τc

(6.8) (6.9) (6.10)

where cm is the median-voter’s consumption, ym is his/her income. τc and τ y are consumption and income tax rate respectively. r is the redistribution. x¯ and y¯ are average income and average expenditure in the aggregate level, hence we have x¯ = y¯ . y ∗ is the potential income and Eq. (6.10) represents the ‘deadweight’ cost of taxation. δ y and δc capture the sensitivity of the cost to income tax and expenditure tax respectively. The first order conditions of this static problem give:

 (m − 1) m δ y − δc − (m − 1)

τc = m δ y (m + 1) − δc (m − 1) (m − 1) − τc τy = δy m τy δ y (m + 1) − δc (m − 1)

−1 =  τc δ y δ y − δc m − (m − 1)

(6.11) (6.12) (6.13)

where m = y¯ /ym represents the ratio between average income and median income, which is consistent with the measure m defined in Eq. (6.5). Pickering and Rajput (2018) offer the proposition confirming that, with certain conditions,2 the ratio between income tax and consumption tax in Eq. (6.13) is unambiguously increasing with the inequality m. We can combine the median-voter framework with the dynamics generated in Proposition 6.1. We find that, over time, the wealth in the modelled economy will become increasingly unevenly distributed due to the assumption that the richer is of low fertility. The median voter becomes increasingly poorer and more in favor of income tax related to consumption tax. In the meantime, due to the uneven demographic growth, the economy becomes relatively young. Since the median voter becomes relatively poorer over time, the median is relatively in favour of income tax to consumption tax since they benefit a relatively large share in their disposable income from redistribution which income tax will not be levied on. Additionally, the 2

The condition required is 0 < δc < δ y < 1 < m, see Pickering and Rajput (2018) for details.

6.3 Evidence

95

optimal level of consumption tax and income tax are determined by 3 contradicting forces. Firstly, higher tax rate will of course hurt taxpayer’s income and utility. However, higher tax means higher redistribution which increases utility. Thirdly, higher tax means higher tax collection cost which undermines the aggregate output overall. This heuristic model shows that as the demographic structure in a society becomes younger, which is a natural consequence of unbalanced fertility, we may discover it choose income tax relative to consumption tax.

6.3 Evidence 6.3.1 Data and Methodology The empirical analysis examines the proposition using a panel of international data over the period 1990–2014. Following Pickering and Rajput (2018) and Luo (2019), the dependent variable is the natural logarithm of the ratio of income taxes (i.e., taxes on income, profits and capital gains) to expenditure taxes (i.e., taxes on goods τ and services), ln( τcy ). Data on both taxes are taken from the World Development Indicators (WDI) database. Moreover we separately examine how different categories of tax measures respectively co-move with the fraction of children. Table 6.1 contains descriptive statistics of the variables used in the analysis. The measure of the dependency ratio widely used in the literature is constructed at an aggregate level as in Razin et al. (2002). Thus, the ratio of the population either below 14 or above 65 to the population between the ages of 15 and 64 (Dependency ratio) is tested in the empirical analysis. Following Luo and Zhu (2021), this paper afterwards replaces this ratio with two alternative dependency measures: the ratio of retirees to workers (Retirees) and the ratio of children to workers (Youth). If our theory is significant, then retirees and children have opposite effects on tax policy and consequently, the analysis then employs both alternative measures to examine the hypothesis. The analysis also augments control variables following Luo (2019) which mainly focuses on the upward trend of the fraction of retired population. In particular we control for the natural logarithm of real GDP per capita in constant dollars, the measure of income inequality, the degree of trade openness, the natural logarithm of the total population size, the output gap, and the democracy score.3 Moreover, as demonstrated by Luo (2019) and under the premise that the theory proposed above is based on a median voter framework, in democracies with low quality institutions the link between the median voter and policy is blurred, whilst in countries with stronger institutions the median’s influence is stronger. For this reason, the sample is further separated into countries that score highly on this measure and those that do not. The

3

Table 6.1 contains the measure and the data source of the variables.

96

6 Youthful Dependents and the Composition of Taxes

Table 6.1 Descriptive statistics Obs Mean τy τc Dependency ratio Retirees Youth UTIP ln(y) TRADE ln(POP) POLITY 2 YGAP

Std. dev.

Min

Max

2142 2141 4633

22.45 29.06 65.62

12.75 13.76 19.55

0.349 0.024 16.54

75.24 89.22 120.8

4633 4633 1556 3652 4263 5302 3790 4668

11.02 54.60 42.79 8.58 86.94 15.07 3.04 0

6.65 23.98 6.71 1.29 51.96 2.35 6.69 0.034

0.390 15.28 22.75 5.03 0.309 9.11 −10 −0.609

40.53 115.0 59.96 11.73 531.7 21.04 10 0.505

Notes The table gives descriptive statistics for the variables. τ y denotes taxes on income, profits and capital gains as a percentage of revenue, τc denotes taxes on goods and services as a percentage of revenue—both taken from the World Development Indicators (WDI) database. Dependency ratio is equal to the ratio of the sum of the population above 65 and the population below 14 to the population between the ages of 15 and 64, Retirees is the ratio of the population above 65 to the population between the ages of 15 and 64, and Youth is the ratio of the population below 14 to the population between the ages of 15 and 64—data on demography are taken from the WDI database. UTIP is the University of Texas Inequality Project’s Estimated Household Income Inequality. y is real GDP at chained PPPs in millions of 2005 US dollars per capita—taken from the Penn World Tables. TRADE is the sum of exports and imports as a percentage of GDP—taken from the WDI database. POP is the size of count

expectation is that the demographic variable(s) will be more strongly related to the policy variable in the more democratic subsample.

6.3.2 Baseline Estimation The next step is to explore the hypothesis proposed above—whether and how the ratio of income to expenditure taxes across different countries and time periods systematically changes with the fraction of youthful dependents. Column 1 of Table 6.2 is a simple specification with just the aggregate dependency ratio and income level as regressors using annual data OLS regression, with robust standard errors clustered by country. Column 2 splits this aggregate measure into Retirees and Youth, whilst column 3 only uses Youth as the key explanatory variable. Columns 4–6 then extend columns 1–3 to include full control variables as well as country fixed effects on the right-hand side. In these specifications it becomes distinct that aging is found to be negatively related to ratio of income to expenditure taxes as in Luo (2019), whilst the impact of Youth shifts the other way, resulting in a positive coefficient on Dependency ratio. Note that the sign of the coefficient estimate regarding Youth

6.3 Evidence

97

Table 6.2 Basic estimation results—the composition of taxes (1) (2) (3) (4) Dependency 0.0270*** ratio (0.00755) Retirees Youth

−0.0341* (0.0181) 0.0233*** (0.00707)

0.0301*** (0.00691)

0.399*** (0.107)

0.604*** (0.138)

0.566*** (0.131)

1849 146 Panel No

1849 146 Panel No

1849 146 Panel No

−0.0211 (0.0237) 0.836*** (0.195) −0.00150 (0.00189) 0.269 (0.685) −0.0209* (0.0113) −1.027 (0.787) 796 87 Panel Yes

0.106

0.191

0.168

0.139

TRADE ln(POP) POLITY 2 YGAP Observations Countries Data Fixed effects? R2

(6)

−0.0672** (0.0287) 0.0264** (0.0106) −0.0107 (0.0186) 1.087*** (0.224) 0.0000768 (0.00209) 0.231 (0.640) −0.0121 (0.0118) −0.952 (0.734) 796 87 Panel Yes

0.0261** (0.0115) −0.0178 (0.0219) 0.952*** (0.214) −0.00127 (0.00195) 0.395 (0.700) −0.0174 (0.0118) −1.001 (0.767) 796 87 Panel Yes

0.179

0.157

0.0199** (0.00961)

UTIP ln(y)

(5)

τ

Notes Table 6.2 contains results using OLS regressions of the composition of taxes, ln( τcy ). Column (1) is with just Dependency ratio and ln(y) as regressors using annual data OLS regression over the period 1990–2014. Column (2) instead includes both Retirees and Youth as regressors. Column (3) only uses Youth as the key explanatory variable. Columns (4)–(6) again test columns (1)–(3) but including UTIP, TRADE, ln(POP), POLITY 2, and YGAP as additional control variables, with country fixed effects. Robust standard errors are shown in parentheses. Standard errors are clustered by country. *, **, and *** respectively denote significance levels at 10%, 5% and 1%

keeps positive in all cases, and all are statistically significant. This is consistent with our theory—an increase in youthful dependents increases income rather than expenditure taxes. Using the estimate from column 6 of Table 6.2, the estimated coefficient for Youth is positive, with a p-value of 2.5%: a one-standard-deviation-rise in the ratio of children to workers is statistically associated with a rise of 0.62 in the ratio of income to expenditure taxes, holding all else equal. The magnitude of this estimated correlation is sizable—showing more than a half of the raw standard deviation in the policy variable.

98

6 Youthful Dependents and the Composition of Taxes

6.3.3 Heterogeneous Analysis It is meaningful to test whether or not the reported results vary with the degree of democracy, given that our theory is based on a median voter framework. Columns 1 and 2 of Table 6.3 therefore present the results by separating the sample according to the median value of democracy score—column 1 shows those with stronger democratic credentials whilst otherwise in column 2. When the sample is distinguished it becomes obvious that the positive relationship between the ratio of children to workers and the ratio of income to expenditure taxes holds solely in the subsample of democratic regimes. The use of an interaction term is another method to investigate the change of the results given the extent of the franchise. In column 3, the democracy score is multiplied by Youth, hence reaching an interaction term. The hypothesis in this case is that the relationship between Youth and the tax composition measure will be increasingly positive under democracies and as a result, the coefficient estimate for the interaction term is positive. The estimation results confirm this, and statistically significant at the 5% level. The median voter has more capacity to affect policy under stronger democracy, consistent with the theory proposed, whilst the influence of the median voter in less democratic countries declines. Note that it is also of interest to ask whether the results change with the level of economic development. Columns 4 and 5 of Table 6.3 separate the sample depending on the median value of income level. As can be observed, the ratio of income to expenditure taxes is positively correlated with the ratio of children to workers. However, this positive relationship only holds to a significant degree in the subsample of countries with lower income level. Poor countries have a proportionally more younger population, and therefore tax revenue collected by taxes on income is increased relative to expenditure. The last two columns further split the sample into OECD members and the rest. The policy-youth relationship is to some extent looser under the OECD subsample. Statistical significance in the regression without the OECD membership, which is normally regarded as poor countries, in turn supports the argument proposed.

6.3.4 Income Taxes In Tables 6.4 and 6.5, results are presented respectively for income taxes, τ y , and expenditure taxes, τc , the numerator and denominator in the main dependent variable, using Panel regressions as above. In Table 6.4 the findings for income taxes, τ y , τ are quite similar to the results found for ln( τcy ) though with lower significance levels. If all control variables are excluded except ln(y), then the p-value of the coefficient estimate pertaining to Youth in column 1 improves to p = 0.005. Increases in the ratio of children to workers are generally found to be positively correlated with the extent to which taxes are levied on income, but more so in the stronger democracies. In countries where a full sample with a set of control variables is included,

0.270

R2

0.195

0.0000442 (0.0206) 0.822*** (0.266) −0.000640 (0.00297) −0.803 (0.497) −0.0117 (0.0194) −0.148 (1.030) 300 Panel Yes Low POLITY 2

0.00988 (0.00833)

(2)

0.175

0.0255** (0.00971) 0.000984** (0.000442) −0.0166 (0.0208) 0.998*** (0.218) −0.000809 (0.00214) 0.308 (0.696) −0.0720*** (0.0243) −0.864 (0.721) 796 Panel Yes Full

(3)

0.225

−0.0628** (0.0238) 0.717** (0.272) −0.000844 (0.00169) 0.0698 (1.114) −0.0314*** (0.0104) −0.971 (1.056) 503 Panel Yes High income

0.0249 (0.0164)

(4)

0.250

0.0287* (0.0162) 1.203*** (0.262) 0.00338 (0.00261) 0.676 (0.776) −0.00246 (0.0116) 0.110 (1.097) 293 Panel Yes Low income

0.0461*** (0.0127)

(5)

0.169

−0.0168 (0.0239) 1.030*** (0.244) −0.00110 (0.00269) 0.416 (0.754) −0.0172 (0.0126) −1.187 (0.846) 542 Panel Yes Non-OECD countries 0.167

0.0283** (0.0125)

−0.0228 (0.0296)

−0.0138 (0.0159) 0.198 (0.189) −0.00190 (0.00146) 0.988 (1.297) −0.0427* (0.0221) 0.790 (1.216) 254 Panel Yes OECD Members

(7)

(6)

Notes Regression specification is the same as column (6) of Table 6.2. Columns (1) and (2) respectively correspond to higher and lower democracy levels. Column (3) includes an interaction term described in the text. Columns (4) and (5) respectively correspond to higher and lower levels of income. Columns (6) and (7) respectively correspond to OECD members and non-OECD countries. Robust standard errors are shown in parentheses. Standard errors are clustered by country. *, **, and *** respectively denote significance levels at 10%, 5% and 1%

YGAP

POLITY 2

ln(POP)

TRADE

ln(y)

Observations Data Fixed effects? Sample

0.0593*** (0.0216)

−0.0321 (0.0227) 1.478*** (0.437) 0.000102 (0.00219) 2.498** (1.072) 0.0315 (0.164) −1.670 (1.391) 496 Panel Yes High POLITY 2

Youth × POLITY 2 UTIP

Youth

(1)

Table 6.3 Panel estimation results with fixed effects—the composition of taxes

6.3 Evidence 99

100

6 Youthful Dependents and the Composition of Taxes

the estimated effect remains positive, though is not statistically significant. When the stronger democratic criterion (depending on the median value of democracy score) is employed, the estimated effect increases and is statistically significant at the 5%. When the sample is refined further to those countries with higher income level throughout the same period, the positive coefficient estimate is sustained, although statistical significance is in this instance low. Using the estimate of column 3, a one standard deviation increase in the ratio of children to workers is statistically associated with a rise of 15.36% in τ y , holding all else equal. Given that this is nearly half of a standard deviation in the policy variable, the magnitude of the estimated correlation is sizable. This weak estimated relationship indicates that there is a very slight variation in income taxes within countries over the sample period. This in turn emphasizes the need and motivation to examine the relationship between the youth fraction and the extent to which taxes are levied on income relative to expenditure, instead of income taxes only. In Table 6.4, the results relating to the control variables are of some interest. One regularity is that consistent with Besley and Persson (2014), there is a positive relationship with income per capita, which likely indicates greater potential to increase tax in richer countries. In addition, trade is found to be positively associated with income taxes, as in Rodrik (1998), which shows a greater potential to increase taxes in countries with higher level of openness.

6.3.5 Expenditure Taxes Table 6.5 contains estimation results relating to τc , the extent to which taxes are raised through expenditure on goods and services. In contrast to income taxes, increases in the ratio of children to workers are generally found to be negatively related to the extent to which expenditure taxes are used, and again this result is particularly strong in the stronger democracies. In countries with lower democratic level, the estimated relationship is negative, though it is not statistically significant, whilst in countries where the stronger democratic requirement is applied, the estimated effect is statistically significant at the 1% level. Analogous to Table 6.3, when the sample is separated according to income level and OECD membership, the coefficient estimate is also negative and statistically significant when the sample is refined to those relatively poor countries. Using the estimate of column 3, a one standard deviation increase in the ratio of children to workers is statistically associated with τ a fall of 17.76% in τc , holding all else equal. As with ln( τcy ), this again represents more than a raw standard deviation in τc , so this is still a sizable effect. There are some differences between the results relating to the controls for income taxes and expenditure taxes. For instance there is a clear positive relationship between expenditure taxes and income inequality, which runs opposite to the findings on income taxes and implies that if the median voter becomes relatively poor, then he/she is likely to tax more on expenditure instead of income. Further in contrast to

0.104

R2

0.085

0.120 (0.123) −0.158 (0.266) 6.916* (3.829) 0.0346 (0.0214) −1.596 (10.34) −0.408* (0.237) −13.10 (13.09) 826 Panel Yes Full 0.262

0.641** (0.288) −0.370 (0.254) 15.47** (6.093) 0.0626** (0.0264) 36.07*** (11.83) 0.561 (1.761) −15.85 (19.75) 515 Panel Yes High POLITY 2 0.111

0.269 (0.209) −0.459 (0.334) 0.0512 (4.521) 0.0538** (0.0237) 8.186 (9.147) −0.144* (0.0853) 23.57* (13.98) 533 Panel Yes High income

−0.0692 (0.144) 0.113 (0.275) 3.035 (4.552) 0.0342 (0.0375) −18.15* (9.502) −0.332 (0.252) −14.70 (17.91) 311 Panel Yes Low POLITY 2 0.086

(5)

(4)

0.171

−0.0385 (0.276) 0.174 (0.262) 9.657 (6.605) 0.0444 (0.0521) −8.935 (14.88) −0.535* (0.266) −30.18* (17.63) 293 Panel Yes Low income

(6) −0.194 (0.289) 0.116 (0.219) 3.038 (2.658) 0.00174 (0.0225) −3.290 (16.21) 0.195 (0.401) 28.65* (15.65) 268 Panel Yes OECD Members 0.154

(7)

0.132 (0.139) −0.158 (0.292) 7.704* (4.424) 0.0425 (0.0283) −1.957 (11.20) −0.419* (0.245) −18.03 (14.16) 558 Panel Yes Non-OECD countries 0.094

(8)

Notes Table 6.4 contains results using OLS regressions of income taxes, τ y . Column (1) only uses Youth as the key explanatory variable. Column (2) again tests column (1) but including UTIP, TRADE, ln(POP), POLITY 2, and YGAP as additional control variables, with country fixed effects. Columns (3) and (4) respectively correspond to higher and lower democracy levels. Columns (5) and (6) respectively correspond to higher and lower levels of income. Columns (7) and (8) respectively correspond to OECD members and non-OECD countries. Robust standard errors are shown in parentheses. Standard errors are clustered by country. *, **, and *** respectively denote significance levels at 10%, 5% and 1%

1899 Panel Yes Full

5.719*** (1.487)

0.209*** (0.0734)

Observations Data Fixed effects? Sample

YGAP

POLITY 2

ln(POP)

TRADE

ln(y)

UTIP

Youth

Table 6.4 Panel estimation results with fixed effects—income taxes (1) (2) (3)

6.3 Evidence 101

0.108

R2 0.237

−0.742*** (0.263) 0.563** (0.249) −15.91*** (4.563) 0.0448 (0.0289) −23.57 (15.52) 0.457 (1.875) 25.50** (11.78) 496 Panel Yes High POLITY 2

(3)

0.098

−0.325 (0.220) 0.0846 (0.316) −7.129** (3.498) 0.0313 (0.0449) 1.325 (16.31) −0.0982 (0.236) −3.740 (8.853) 301 Panel Yes Low POLITY 2

(4)

0.267

−0.439* (0.237) 0.953*** (0.180) −10.98*** (3.938) 0.0574** (0.0233) −4.585 (16.74) 0.348*** (0.0785) 26.33** (10.41) 503 Panel Yes High income

(5)

0.230

−0.725*** (0.126) −0.361 (0.290) −8.047*** (2.495) −0.0273 (0.0349) −11.28 (10.77) −0.311* (0.154) −10.92 (8.150) 294 Panel Yes Low income

(6) 0.297 (0.394) 0.175 (0.261) −0.341 (4.166) 0.0420* (0.0223) −18.06 (20.82) 1.443 (0.909) 2.576 (19.33) 254 Panel Yes OECD Members 0.136

(7)

−0.453** (0.188) 0.329 (0.292) −9.354*** (2.952) 0.0476 (0.0346) −8.279 (13.14) −0.0615 (0.190) 9.291 (9.941) 543 Panel Yes Non-OECD countries 0.126

(8)

Notes Table 6.5 contains results using OLS regressions of expenditure taxes, τc . Column (1) only uses Youth as the key explanatory variable. Column (2) again tests column (1) but including UTIP, TRADE, ln(POP), POLITY 2, and YGAP as additional control variables, with country fixed effects. Columns (3) and (4) respectively correspond to higher and lower democracy levels. Columns (5) and (6) respectively correspond to higher and lower levels of income. Columns (7) and (8) respectively correspond to OECD members and non-OECD countries. Robust standard errors are shown in parentheses. Standard errors are clustered by country. *, **, and *** respectively denote significance levels at 10%, 5% and 1%

1885 Panel Yes Full

0.120

−0.438** (0.177) 0.307 (0.269) −9.249*** (2.633) 0.0444 (0.0277) −7.682 (12.47) −0.0307 (0.185) 9.222 (8.975) 797 Panel Yes Full

−5.132*** (1.654)

(2)

(1)

−0.335*** (0.0813)

Observations Data Fixed effects? Sample

YGAP

POLITY 2

ln(POP)

TRADE

ln(y)

UTIP

Youth

Table 6.5 Panel estimation results with fixed effects—expenditure taxes

102 6 Youthful Dependents and the Composition of Taxes

6.3 Evidence

103

Table 6.6 Robustness—the composition of taxes (1) (2)

(3)

(4)

−0.0107 (0.0185) 0.968*** (0.227) −0.000103 (0.00252) 1.264 (0.918) −0.0126 (0.0122) −0.533 (0.704)

−0.0606* (0.0347) 0.0230** (0.00962) −0.00842 (0.0178) 1.050*** (0.224) −0.0000706 (0.00237) 0.656 (0.964) −0.00950 (0.0117) −0.561 (0.697)

0.0182* (0.00916) −0.0103 (0.0183) 1.005*** (0.231) −0.000213 (0.00252) 1.172 (0.892) −0.0116 (0.0122) −0.540 (0.699)

796 87 Panel Fixed effects

796 87 Panel Fixed effects

796 87 Panel Fixed effects

0.0155*** (0.00540) 0.00118 (0.00799) 0.551*** (0.122) 0.00250** (0.00116) 0.139 (0.359) −0.0321*** (0.00757) −0.814* (0.448) 0.514*** (0.0525) 597 79 Panel Arellano-Bond

Yes 0.187

Yes 0.201

Yes 0.192

Dependency ratio 0.0142* (0.00808) Retirees Youth UTIP ln(y) TRADE ln(POP) POLITY 2 YGAP L.(DEP VAR) Observations Countries Data Estimation method Year dummies? R2

Notes Columns (1)–(3) extend columns (4)–(6) of Table 6.2 to include year dummies on the righthand side. Column (4) contains the Arellano-Bond dynamic panel data estimation results. *, **, and *** respectively denote significance levels at 10%, 5% and 1%

τ y there is a negative relationship between τc and income per capita, which reflects the ability to collect revenue through taxes on income in particular.

6.3.6 Robustness Table 6.6 presents a robustness check. We would expect that both the demographic variable(s) and the tax policy (ratio) to be highly persistent. However, some may worry that treating observations as being independent over time as we did above seems likely to cause problems (i.e., not least as demographic data are often I(1) or I(2)). This table thus provides us with some evidence that different specifications

104

6 Youthful Dependents and the Composition of Taxes

would not make a large difference to the existing findings. First, we have yet to treat observations as being independent over time here, and extend the baseline estimation (i.e., columns 4–6 of Table 6.2) to include year dummies on the right-hand side. In this specification the sign of the coefficient estimate relating to the demographic variables retains, and the significance level survives. Second, the last column displays the Arellano-Bond dynamic panel data estimation results. The positive relationship between the ratio of children to workers and the ratio of income to expenditure taxes holds up, and significantly different from zero at the 1% level.

6.4 Conclusion This paper analyzes how the younger population affects the composition of taxes. The paper, based on the Pickering and Rajput (2018) framework, theorizes that a rise in the youthful proportion of the population leads to more demand for income rather than expenditure taxes. We establish a model which combines median-voter theory with dynamic evolution in the demographic structure. With unbalanced fertility, the position of median-voter naturally becomes poorer over time while the demographic structure becomes younger. The main proposition is that the extent of taxes on income relative to expenditure rises with youthful dependents. The logic is that unbalanced fertility in the demographic structure leads to greater income inequality, bringing about more demand for income taxes rather than expenditure taxes, as the median voter is relatively poor and s/he dislikes being taxed (again) at the consumption-side. The relationship between younger population and the composition of taxes is tested in an international panel data, including the ratio of children to workers as an explanatory variable. Data for tax composition and the ratio of children to workers are all taken from the WDI database. Consistent with the theory, the extent of taxes on income relative to taxes on expenditure is found to be positively associated with the ratio of children to workers. Moreover, income taxes as a proportion of total revenues rise with the youth fraction, whilst expenditure taxes as a proportion of total revenues fall with the youth fraction. The empirical results hold across the various econometric specifications employed. In particular, the fact that the results hold significantly in countries with strong democratic credentials is supportive of the mechanism proposed in this paper.

Appendix Proof of Proposition 6.1 Proof. Equation (6.11) is nothing but substituting Eq. (6.5) into Eq. (6.10).

6.4 Conclusion

105 τ

Firstly, we re-exam the relation between τcy and m in Eq. (6.10) shown in Pickering and Rajput (2018). The function can be rearranged into  δ y + δc + m δ y − δc τy −1  = τc δ y + mδ y δ y − δc − 1

(6.14)

Immediately, we have   d τ y /τc δ2 − δy − 2 δy =  c 2 dm δ y m δc − δ y + 1 − 1

(6.15)

To generate a meaningful policy, we assume that the ‘deadweight cost’ of tax will not exceed the tax itself, hence we apply that 0 < δ y < 1 and 0 < δc < 1. Therefore τ y /τc is unambiguously increasing with m.4 Secondly, we show the inequality is increasing with time t in this economy. m (t) =

et − 1 t

te 2

(6.16)

We have m  (t) =

et (t − 2) + t + 2 t

2t 2 e 2

(6.17)

Define f (t) ≡ et (t − 2) + t + 2

(6.18)

f  (t) = et (t − 1) + 1 f  (t) = et t

(6.19) (6.20)

We have

One will see that, when t > 0, f  (t) > 0 and f  (0) = 0. Hence we have f  (t) > 0. With the condition f (0) = 0, we conclude f (t) > 0 and m  (t) > 0. Again, with the condition of limt→0 m  (t) = 0,5 we conclude that m (t) is monotonically increasing when t > 0. Finally, we show the population growth rate N g is increasing with time t. N g (t) =

1 et − t e −1 t

(6.21)

This proof shows that the condition suggested in Pickering and Rajput (2018), 0 < δc < δ y < 1 < m, is not necessary. We only require δ y and δc are between 0 and 1. 5 This can be shown by L’Hospital’s rule. 4

106

6 Youthful Dependents and the Composition of Taxes

We have N g  (t) =

 et et − t 2 − 2 + 1 (et − 1)2 t 2

(6.22)

Define g (t) ≡ et − t 2

(6.23)

g  (t) = et − 2t g  (t) = et − 2

(6.24) (6.25)

We have

When t > 0, g  (t) reaches its minimum 2 − 2 log  2 ≈ 0.61 > 0 at t = log 2. Therefore g  (t) > 0. Hence, g (t) and et et − t 2 − 2 + 1 is increasing with t when t > 0. Again, when t = 0, the numerator of N g  (t) is 0. Consequently, the numerator of N g  (t) and N g  (t) itself is positive. Hence, N g (t) is monotonically increasing with t.

References Besley T, Persson T (2014) Why do developing countries tax so little? J Econ Perspect 28(4):99–120 Croix D, Matthias D (2003) Inequality and growth: why differential fertility matters. Am Econ Rev 93(4):1091–1113 Dahan M, Tsiddon D (1998) Demographic transition, income distribution, and economic growth. J Econ Growth 3(1):29–52 Disney R (2007) Population ageing and the size of the welfare state: is there a puzzle to explain? Eur J Polit Econ 23(2):542–553 Galasso V, Profeta P (2007) How does ageing affect the welfare state? Eur J Polit Econ 23(2):554– 563 Luo W (2018) Essays on inequality and fiscal policy. PhD thesis, University of York Luo W (2019) Demography and the composition of taxes: evidence from international panel data. Econ Lett 183:108518 Luo W (2020) Demography and economic growth: the effect of tax composition. Appl Econ Lett 27(20):1629–1634 Luo W, Zhu J (2021) Youthful dependents and economic growth: the effect of tax composition. Appl Econ Lett 28(8):675–680 Meltzer AH, Richard SF (1981) A rational theory of the size of government. J Polit Econ 89(5):914– 927 Moav O (2004) Cheap children and the persistence of poverty. Econ J 115(500):88–110 Pickering A, Rajput S (2018) Inequality and the composition of taxes. Int Tax Public Finan 25(4):1001–1028 Razin A, Sadka E, Swagel P (2002) The aging population and the size of the welfare state. J Polit Econ 110(4):900–918 Rodrik D (1998) Why do more open economies have bigger governments? J Polit Econ 106(5):997– 1032

References

107

Sanz I, Velazquez FJ (2007) The role of ageing in the growth of government and social welfare spending in the OECD. Eur J Polit Econ 23(4):917–931 Shelton CA (2008) The aging population and the size of the welfare state: is there a puzzle? J Public Econ 92(3–4):647–651 Simonovits A (2007) Can population ageing imply a smaller welfare state? Eur J Polit Econ 23(2):534–541

Chapter 7

Demography and Government Debt

7.1 Introduction How does population aging affect government debt? One important work is an overlapping generations growth model designed by Ono (2003), building upon Pecchenino and Pollard (1997), who theorizes that as population aging leads to a heavy burden of social security expenditures and social security tax alone is hardly sufficient to cover such expenditures, the government has to issue more public debt in order to finance the payment. Empirical evidence generally has not supported the mechanism put forward by Ono (2003). For example, public debt is not induced by a greater size of government spending, in particular in an era of a steady size of government expenditure. Figure 7.1 depicts these data, showing that both OECD members and all the world experienced a period of stasis in the earlier years followed by a consistent fall since the 2008/9 Global Financial Crisis. Given a upward trend in public debt, this indicates no positive association between debt and government spending. Luo (2019), according to model suggested by Razin et al. (2002), develops that population aging raises the demand for expenditure taxes instead of income taxes for the sake of shifting the tax burden onto the retired population if the median voter lies within working age. This paper develops the Luo (2019) hypothesis to consider how population aging affects government debt, especially via the effect of tax composition. The main theoretical prediction is that debt rises with population aging as the extent of taxes on income relative to expenditure declines. The logic is analogous to Luo (2019). Income taxes are uniquely borne by workers, whilst expenditure taxes are shared by both workers and retirees. If the median voter is a worker, then a rising size of the retired population compels a transfer in tax composition towards expenditure taxes, affecting the public debt accumulation. While the rise in expenditure taxes can hardly mitigate the negative consequences of the fall in income taxes, then the incentives of governments to borrow increase. Unlike the argument invoked in this paper, current empirical literature generally has concentrated on the relationship between the fraction of retired population and © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 W. Luo, Inequality, Demography and Fiscal Policy, Applied Economics and Policy Studies, https://doi.org/10.1007/978-981-99-0518-8_7

109

110

7 Demography and Government Debt

Fig. 7.1 Government expenditure from 1970 to 2018

economic growth, especially in the past few decades when the change in GDP per capita is found to be positively associated with the dramatic change the structure of demography has undergone (see Acemoglu and Restrepo 2017; Luo 2020a; Luo and Zhu 2021). In an overlapping generations model, taxes are collected from both expenditure and income, which transfer to both generations, with a balanced budget period by period. In a median voter framework, as argued by Luo (2019) the observed tax policy is the ideal income-to-expenditure tax ratio of the median voter. Countries experiencing more rapid aging have issued more debt, since the median voter is inclined to increase the tax burden on the retired population rather than being uniquely taxed income and is fond of a lower tax ratio, which reduces income taxes and causes the government to issue more debt. The relationship between population aging and government debt is investigated empirically using a panel of over 100 countries over the period 1990–2014. Following Reinhart and Rogoff (2011) and Luo (2020b), the dependent variable is total (domestic plus foreign) gross central government debt measured as a percentage of GDP. The key explanatory variable is the ratio of the population either below 14 or above 65 to the population between the ages of 15 and 64. These data are all taken from the World Development Indicators database. Consistent with the theory invoked, data measuring public debt as a share of GDP are consistently positively correlated with the measure of the fraction of the retired population. This relationship is robust across different econometric specifications employed. In the panel estimation with fixed effects, a one standard deviation rise in the ratio of retirees to workers is statistically associated with an increase in central government debt by 0.32% of GDP, holding all else equal. The magnitude of this

7.2 Literature Review

111

estimated correlation is sizable—showing nearly a half of the raw standard deviation in the policy variable. The statistical relationship holds most significantly in countries with higher levels of democracy, with OECD membership, or with higher levels of income, in support of the mechanism proposed in this paper. Moreover, we also examine how the results change with the tax policy, given that public debt declines as the ratio of income to expenditure taxes in the argument proposed in this paper. The data indicate that population aging leads to a rise in the extent of expenditure relative to income taxes, and thus public debt is accumulated. The rest of the paper is organized as follows. The next section introduces the existing literature. Section 7.3 presents the data and identification strategy. Section 7.4 illustrates the empirical results, and the last section concludes.

7.2 Literature Review The issue of population aging observed in many advanced economies has been widely debated in recent years, and thus a growing literature in selected aspects regarding this topic occurs. In particular, the existing literature mainly focuses on the impacts of population aging on health expenditure, tax policy, and business cycle, while in the field of public debt, research on the direct influencing mechanism between aging and debt is relatively scarce.

7.2.1 Effects of Population Aging The first and immediate effects of population aging are observed on health expenditure through an increase in the healthcare demand. As highlighted by the World Health Organization (2015), the use of healthcare services rises with age and per capita expenditures on medical care are relatively higher among the elderly. Further studies put forward that given a longer life expectancy aging induces a relatively strong reaction from health spending (Lopreite and Zhu 2020). Indeed, increased longevity without an improvement in health conditions leads to more demand for health services over a longer period of lifetime, increasing total healthcare spending (Breyer et al. 2010; Zweifel et al. 2005). Impacts on the tax policy of population aging emerge as well. The increase in the old-age dependency ratio also raises the expenditure for pensions (Komisyonu 2018; Verbiˇc and Spruk 2014), hampering the compliance with the fiscal sustainability rules (Beetsma and Oksanen 2007). Based on the median voter framework (Luo 2019) discovers that population aging increases the demand for expenditure relative to income taxes so as to increase the tax burden on the retired population. Day and Day (2021) also provide evidence of lower income tax rates while based on the lowerability median voter. Felix and Watkins (2013), Yashio and Hachisuka (2014), and Creedy et al. (2010) investigate the impact of population aging on taxation for the US,

112

7 Demography and Government Debt

Japan, and New Zealand, respectively. They find that income tax revenue decreases significantly with population aging, whilst the effect on consumption/sales taxes is not clear-cut. However, a dynamic analysis combining the population scenarios with long-run scenarios under growing real wages and pension benefits, leads to increasing income tax revenue and social security contributions (Prammer 2019). Except for income taxes and expenditure taxes which account for a major portion, environmental tax reform arises under a dynamic CGE model when negative impacts of aging are reinforced by automation (Costantini and Sforna 2020). Seen from the dynamic cycle of macroeconomic society, population aging weakens the effectiveness of fiscal policy in boosting an economy without considering the state of business cycle (Yoshino and Miyamoto 2017; Basso and Rachedi 2021; Miyamoto and Yoshino 2020), while aging would call for a larger fiscal stimulus to support an aggregate demand during recession along with a large fiscal swing space (Honda and Miyamoto 2021). As consumers and entrepreneurs become more risk-averse with longer life expectancy, countries with an aging population show a lower cyclical volatility of consumption and investment but higher output volatility (Guimarães and Tiryaki 2020).

7.2.2 Public Debt Public debt is another topic receiving increased attention. Existing literature mainly focuses on the effects of government debt on other fields, especially on inequality, rather than treating government debt as an endogenous variable. Röhrs and Winter (2017) find that quantitatively sizable welfare gains can be achieved by reducing government debt when comparing stationary equilibra through raising the amount of capital available for production, while these welfare gains disappear once the transition is incorporated. Similar points of view are provided in the Borissov and Kalk (2020) research. They propose that a lower debt-to-GDP ratio results in a convergence of the economy on a unique egalitarian equilibrium with perfect equality and a higher growth rate. In addition to inequality, public debt also links to economic growth. Countries with higher level of government debt have a faster growth in industries with greater liquidity, and the positive liquidity effect of public debt mainly stems from domestic debt, rather than external debt (Grobéty 2018). Another strand of literature argues that government debt can foster growth by enhancing the supply of liquid assets or collateral (Woodford 1990). When taking government debt as the outcome variable, Gnangnon (2021) provides evidence that a great deal of tax reform reduces the instability of public debt especially in developing countries, and this relevance could be through the trade openness channel, given the relationship between tax reform and trade (Gnangnon 2019). Therefore, tax is an important channel when exploring government debt. A few scholars have studied the relationship between aging and government debt theoretically. A heavy burden of social security payments brought by aging oblige governments to finance through issuing bonds under a overlapping gener-

7.3 Data and Methodology

113

ations model, and economies would achieve two dynamically inefficient equilibra (Ono 2003). The growth and welfare effects of debt-financed public investment under the golden rule of public finance using the Yaari Blanchard model show that both tax rates and public debt positively depend on longevity. The rising tendency of public debt under population aging can be observed in countries such as the United Kingdom, Germany, and Japan (Kamiguchi and Tamai 2019). However, as mentioned above, such articles are only carried out from theoretical models rather than using empirical methods in combination with related data. In the empirical literature, Pan and Wang (2012) provide evidence that an increase in the old-age dependency ratio causes an increase in the ratio of public debt to GDP in the Euro area. ChecheritaWestphal and Rother (2012), however, do not include a single statement on the effects of population aging on debt. The dependency ratio is included together with debt as an explanatory variable to explain developments in the saving rate. While such empirical articles are still limited in regions and only consider the case of European countries, which lacks general significance. More importantly, they are lack of an essential rational mechanism behind the causality neither. In order to make up for the absence of academic research in this area especially empirical articles, one key contribution of this paper is that we utilize international panel data, and use the tax structure as the channel to study the influence of population aging. To our knowledge, this is the first empirical paper that attempts to analyze the effects of aging on government debt through increasing the expenditure taxes to compensate for the decrease in income taxes.

7.3 Data and Methodology Following Reinhart and Rogoff (2011) and Luo (2020b), the dependent variable is total (domestic plus foreign) gross central government debt measured as a percentage of GDP. The sample covers a panel of international data over the period from 1990 to 2014. The argument invoked in this paper emphasizes the median voter framework and thus, countries with various quality institutions are the appropriate sample. Table 7.1 contains descriptive statistics of all the variables. The key explanatory variable in the analysis is the fraction of the retired population. The measure of it frequently employed in current literature is the proportion of the population above the age of 65 (P R O P65). Another common measure of the dependency ratio, computed by the ratio of the population above 65 to the population between the ages of 15 and 64 (R AT I O), is also tested in the empirical analysis for the purpose of robustness check. These measures are preferable to an aggregate level of the dependency ratio used in Razin et al. (2002), as they exclude the impact from children on policy. Following Luo (2020b), which focuses solely on OECD evidence, the empirical analysis also includes standard control variables in the investigation of central government primary budget surplus data. In particular I control for the natural logarithm

114

7 Demography and Government Debt

Table 7.1 Descriptive statistics Obs Mean DEBT PROP65 PROP1564 UTIP ln(y) TRADE POLITY 2 OECD ty tc

1722 4633 4633 1556 3652 4263 3790 5325 2142 2141

56.11 7.01 61.20 42.79 8.58 86.94 3.04 0.138 22.45 29.06

Std. Dev.

Min

Max

37.75 4.72 7.01 6.71 1.29 51.96 6.69 0.345 12.75 13.76

0.214 0.335 45.29 22.75 5.03 0.309 − 10 0 0.349 0.024

289.8 25.08 85.81 59.96 11.73 531.7 10 1 75.24 89.22

Notes The table gives descriptive statistics for the variables. DEBT is total (domestic plus foreign) gross central government debt measured as a percentage of GDP—taken from the World Development Indicators (WDI) database. P R O P1564 and P R O P65 are respectively the proportion of the population aged between 15 and 64, and 65 and above—taken from the WDI database. U T I P is the University of Texas Inequality Project’s Estimated Household Income Inequality. y is real GDP at chained PPPs in millions of 2005 US dollars per capita—taken from the Penn World Tables. T R AD E is the sum of exports and imports as a percentage of GDP—taken from the WDI database. P O L I T Y 2 is a measure of democracy provided by the Polity IV project, with −10 denoting the highest level of autocracy, and 10 denoting the highest level of democracy. τ y denotes taxes on income, profits and capital gains as a percentage of revenue, τc denotes taxes on goods and services as a percentage of revenue—both taken from the WDI database

of real GDP per capita in constant dollars, the measure of income inequality, the degree of trade openness, and the demographic effect. Apart from these control variables, this paper also includes a final control variable to examine the argument invoked. Luo (2019) finds that democracy does matter for tax composition. In nations with low quality institutions the association between the median voter and policy is to some extent constrained, whilst in those with stronger institutions the impact of the median voter is stronger. If the effect of policy variables is considered, as advanced in this paper, then the accumulation of public debt will be influenced. Beyond that, government debt may also be influenced by the democracy level indirectly regarding as a proxy for tax capacity. Thus, the democracy score provided by the Polity IV is employed in the analysis. The benchmark empirical specification is thus   + αi + u i,t Debti,t = α0 + β Agingi,t + xi,t

(7.1)

where i represents each country and t represents each time period, and u i,t is the error term. The left-hand-side variable, Debti,t , is a measure of government debt in country i in year t. The variable of interest is Agingi,t , measured by the fraction of the retired population. The coefficient, β, hence indicates the impact of population aging on public debt. A positive and significant β suggests that population aging exerts a

7.4 Empirical Results

115

positive effect on the decision of public debt, whilst a negative and significant β implies that aging pushes the level of debt lower. Control variables analyzed above are included in the vector xi,t . We also include country-specific dummy variables, αi , to control for time-invariant, unobserved country characteristics that shape public debt across countries. Standard errors are clustered at the country level due to the potential correlation of error term, u, within a country. As this paper argues that population aging affects government debt through the effect of tax composition, both taxes on income, profits and capital gains as a share of total tax revenue and taxes on goods and services as a share of total tax revenue are included as key regressors to explore the growth effect of tax policy. Following Luo (2019), I also instead use the income-to-expenditure tax ratio as the key explanatory variable. The argument proposed predicts that government debt falls with an increase in the extent of taxes on income relative to expenditure.

7.4 Empirical Results 7.4.1 Baseline Estimation Results This section is to test whether and how central government debt as a share of GDP changes with the fraction of the population that is retired in the presence of fixed country effects. Column 1 of Table 7.2 is a simple specification with just the fraction of the retired population (P R O P65) as the only regressor using annual data OLS regression, with robust standard errors clustered by country. Column 2 further uses an alternative measure of aging, the ratio of retirees to workers (R AT I O), instead. In these specifications the sign of the coefficient estimate relating to the aging measure is positive in all cases, and all are statistically significant. This is consistent with the argument—an increase in the retired fraction causes governments to have more incentives to borrow. Columns 3 and 4 repeat the analysis of columns 1 and 2 using full controls instead. The results using panel estimation with full control variables support those already found. The estimated statistical significance of the alternative aging measures is unaffected and even remains at the 1% level in both columns. Using the estimate from column 4 of Table 7.2, the estimated coefficient for the measure of population aging is positive, with a p-value of 0.4% and the estimated relationship is sizable: a one standard deviation rise in the ratio of retirees to workers is statistically associated with an increase in central government debt by 0.32% of GDP, holding all else equal. The presence of fixed country effects goes some means towards controlling for unobserved determinants of government debt. Nonetheless, it is possible that unobserved country-specific effects may be time-varying. To further control for unobserved country-specific effects the regression analysis is next augmented to employ the lagged dependent variable. The last two columns of Table 7.2 contain the results.

116

7 Demography and Government Debt

Table 7.2 Basic estimation results (1) (2) PROP65

4.637** (1.913)

RATIO

(3)

(4)

7.556*** (2.831) 3.439*** (1.172)

UTIP

1.844** (0.733) −27.41** (11.59) 0.0219 (0.124) −0.860 (1.366)

ln(y) TRADE PROP1564

1618 118 Panel Yes 0.048

1618 118 Panel Yes 0.060

798 85 Panel Yes 0.165

(6)

1.708* (0.896) 4.861*** (1.649) 1.795** (0.747) −26.94** (12.70) 0.0224 (0.122)

L.(D E P V A R) Observations Countries Data Fixed Effects? R2

(5)

798 85 Panel Yes 0.159

0.413 (0.354) −6.014* (3.476) −0.0291 (0.0318) −0.416 (0.394) 0.765*** (0.0644) 701 79 Panel Yes 0.793

1.167** (0.557) 0.378 (0.352) −6.941** (3.435) −0.0295 (0.0326)

0.766*** (0.0641) 701 79 Panel Yes 0.792

Notes Table 7.2 contains results using OLS regressions of total central government debt as a share of GDP. Estimations use panel regression with country fixed effects, and robust standard errors clustered by country in parentheses. Columns (3) and (4) again test columns (1) and (2) including full control variables. Columns (5) and (6) extend columns (3) and (4) to include the lagged dependent variable. *, **, and *** respectively denote significance levels at 10, 5 and 1%

The point estimate of β1 for both alternative measures of aging remains positive and is still statistically significant even in this demanding econometric specification. In Table 7.2, the results relating to the control variables are also of some interest. There is a negative relationship with income per capita, which likely indicates greater potential for richer countries to increase income taxes, which relieves their stress to borrow. In addition, countries with greater income inequality levels are found to have greater reliance on public debt, which reflects more redistributive policies produced.

7.4.2 Further Estimation Results It is natural to explore whether or not the reported results change with the level of democracy, given the argument invoked on a median voter framework. Column 1 of Table 7.3 thus firstly follows the specification of column 4 of Table 7.2 whilst also adds the degree of democracy on the right-hand side. The estimated statistical significance of the aging ratio is unaffected and even remains at the 5% level. The

7.4 Empirical Results

117

result similarly demonstrates an increased tendency for governments to issue more debt as population aging increases. Columns 2 and 3 then extend the estimated results by splitting the sample by the degree of democracy (based on the median value of democracy score). Column 2 (column 3) contains the result for countries with stronger (weaker) democratic credentials. When the sample is divided it becomes distinct that the positive relationship between the ratio of retirees to workers and public debt holds only in the subsample of democratic regimes. In column 2 the p-value for the estimated coefficient for the aging ratio is 1.6%: A one standard deviation increase in the ratio of retirees to workers is statistically associated with central government debt, which is higher by 0.26% of GDP, holding all else equal. Though here is an increase in the number of the retired population who is in favor of income rather than expenditure taxes, leading to a lower level of debt, the median voter influences towards the opposite way, for s/he aims to share the tax-burden with retirees, instead of being taxed income solely from the working population. Moreover, under strong quality institutions the median voter more plausibly affects policy and therefore driving public debt. The use of a dummy variable of OECD membership provides an alternative way to examine how the results change with the extent of the franchise. As the argument advanced in this paper emphasizes the median voter framework, established democracies are the appropriate sample. Columns 4 and 5 hence respectively present the result for countries with/without OECD membership, and again confirm the story proposed in this paper. Similarly, columns 6 and 7 split the sample by economic development depending on the median value of income level. In column 6 the (relatively) high income sample again returns a positive coefficient for the aging ratio of very slightly smaller magnitude to that found for the full sample. In column 7 the (relatively) low income sample also returns same sign of coefficient whilst with declined statistical significance. The debt-aging relationship is somewhat looser under lower economic development, as the degree of democracy falls. Note that it is also of interest to examine how the results change with the tax policy, under the premise that public debt falls as the ratio of income to expenditure taxes in the argument proposed in this paper. Therefore, column 8 includes both income taxes and expenditure taxes as key regressors. The sign of coefficients on two separate policy variables confirms that increased tax revenue can to some extent relieve the stress for financing by debt. As mentioned previously, the observed tax rate is the ideal tax ratio of the median voter. Column 9 then uses the income-to-expenditure tax ratio instead of two separate policy variables as the key explanatory variable. Statistical significance in the regression with tax ratio indicates that while the extent of expenditure relative to income taxes rises due to aging, the rise in expenditure taxes is far away to cover the negative outcomes of the fall in income taxes given its small initial size, and thus public debt is accumulated.

−51.47*** (16.80) 0.301*** (0.104)

−9.271 (14.21)

−0.0511 (0.115)

9.320 (14.94)

−33.09*** (11.35)

0.120 (0.0855)

−1.206* (0.684)

ln(y)

TRADE

POLITY 2

−1.074 (0.751)

0.227* (0.131)

−48.59*** (17.06)

0.981 (0.793)

1.075 (3.990)

(7)

(8)

326

High POLITY2

Full

0.168

Sample

R2

451

0.253

Low POLITY2

Yes

0.275

OECD Countries

Yes

0.167

Non-OECD countries

Yes

Panel

397

0.186

High income

Yes

Panel

380

0.220

Low Income

Yes

Panel

0.035

Full

Yes

Panel

1186

Panel

0.004

Full

Yes

Notes Column (1) contains results using OLS regressions of total central government debt as a share of GDP, including UTIP, ln(y), TRADE, POLITY 2, and country fixed effects, with annual panel data over the period 1990–2014. Columns (2) and (3) respectively correspond to higher and lower democracy levels. Columns (4) and (5) respectively correspond to OECD and non-OECD countries. Columns (6) and (7) respectively correspond to higher and lower income levels. Column (8) includes both t y and tc as key regressors. Column (9) instead includes t y /tc as the key explanatory variable. Robust standard errors are shown in parentheses. Standard errors are clustered by country. *, **, and *** respectively denote significance levels at 10, 5 and 1%

0.170

Yes

Fixed Effects? Yes

Panel

1186

Panel

Data

302

777

Panel

Observations

Panel

−0.0107* (0.00600)

(9)

t y /tc

475

−0.142 (0.369)

−0.689 (4.208)

−0.0537 (0.116)

−1.308 (17.88)

1.301 (1.273)

3.689** (1.662)

(6)

tc

−1.065 (0.746)

0.269** (0.114)

−37.29** (14.55)

1.251* (0.734)

1.102 (3.319)

(5)

−0.695** (0.272)

−1.437 (2.244)

−0.0674 (0.100)

−8.840 (15.47)

1.439 (1.295)

4.022** (1.564)

(4)

ty

−1.208 (1.021)

0.747 (0.834)

1.243 (1.216)

1.343** (0.636)

3.245 (3.473)

UTIP

3.976** (1.581)

(3)

3.629** (1.728)

(2)

RATIO

(1)

Table 7.3 Further estimation results

118 7 Demography and Government Debt

References

119

7.5 Conclusion This paper analyzes how population aging affects government debt. The paper, based on the Luo (2019) framework, theorizes that a rise in the fraction of the retired population leads to more demand for public debt. The logic is that a frequent change in demography leads to a transfer in tax composition towards expenditure taxes, affecting the public debt accumulation. The relationship between the fraction of the retired population and government debt is itested in an international panel data, including the ratio of retirees to workers as an explanatory variable. Data for public debt and the ratio of retirees to workers are all taken from the WDI database. Consistent with the theory, the debt level is found to be positively associated with the fraction of the retired population, as a result of a falling ratio of income to expenditure taxes. The empirical results hold firmly in countries with strong democratic credentials, with OECD membership, or with higher levels of economic development, which is supportive of the mechanism proposed in this paper. Consequently, it is recommended that the authorities should have an eye on the relative change between income and expenditure taxes in an era of aging, and its effect on public debt through the mechanism of tax composition.

References Acemoglu D, Restrepo P (2017) Secular stagnation? The effect of aging on economic growth in the age of automation. Am Econ Rev 107(5):174–79 Basso HS, Rachedi O (2021) The young, the old, and the government: demographics and fiscal multipliers. Am Econ J Macroecon 13(4):110–41 Beetsma RM, Oksanen H (2007) Pension systems, ageing and the stability and growth pact. CESifo Working Paper No. 2141, Available at SSRN: https://ssrn.com/abstract=1029588 Borissov K, Kalk A (2020) Public debt, positional concerns, and wealth inequality. J Econ Behav Organ 170:96–111 Breyer F, Costa-Font J, Felder S (2010) Ageing, health, and health care. Oxford Rev Econ Pol 26(4):674–690 Checherita-Westphal C, Rother P (2012) The impact of high government debt on economic growth and its channels: an empirical investigation for the euro area. Eur Econ Rev 56(7):1392–1405 Costantini V, Sforna G (2020) A dynamic CGE model for jointly accounting ageing population, automation and environmental tax reform. European Union as a case study. Econ Modell 87:280– 306 Creedy J, Enright J, Gemmell N, Mellish A (2010) Population ageing and taxation in New Zealand. New Zealand Econ Papers 44(2):137–158 Day C, Day G (2021) Aging, voters and lower income tax: a role for pension design. Econ Model 94:560–569 Felix A, Watkins K (2013) The impact of an aging US population on state tax revenues. Econ Rev 4:95–127 Gnangnon SK (2019) Tax reform and trade openness in developing countries. J Econ Integr 34(3):498–519 Gnangnon SK (2021) Tax reform and public debt instability in developing countries: the trade openness and public revenue instability channels. Econ Anal Pol 69:54–67 Grobéty M (2018) Government debt and growth: the role of liquidity. J Int Money Finance 83:1–22

120

7 Demography and Government Debt

Guimarães SD, Tiryaki GF (2020) The impact of population aging on business cycles volatility: international evidence. J Econ Ageing 17:100285 Honda J, Miyamoto H (2021) How does population aging affect the effectiveness of fiscal stimulus over the business cycle? J Macroecon 68:103288 Kamiguchi A, Tamai T (2019) Public investment, public debt, and population aging under the golden rule of public finance. J Macroecon 60:110–122 Komisyonu A (2018) The 2018 ageing report: economic and budgetary projections for the 28 EU member states (2016–2070). European Economy Institutional Paper, European Commission, p 079 Lopreite M, Zhu Z (2020) The effects of ageing population on health expenditure and economic growth in China: a Bayesian-VAR approach. Soc Sci Med 265:113513 Luo W (2019) Demography and the composition of taxes: evidence from international panel data. Econ Lett 183:108518 Luo W (2020) Demography and economic growth: the effect of tax composition. Appl Econ Lett 27(20):1629–1634 Luo W (2020) Inequality and government debt: evidence from OECD panel data. Econ Lett 186:108869 Luo W, Zhu J (2021) Youthful dependents and economic growth: the effect of tax composition. Appl Econ Lett 28(8):675–680 Miyamoto H, Yoshino N (2020) A note on population aging and effectiveness of fiscal policy. Macroecon Dynam pp 1–11. https://doi.org/10.1017/S1365100520000607 Ono T (2003) Social security policy with public debt in an aging economy. J Popul Econ 16(2):363– 387 Pan H, Wang C (2012) Government debt in the euro area evidence from dynamic factor analysis. Econ Lett 115(2):272–275 Pecchenino RA, Pollard PS (1997) The effects of annuities, bequests, and aging in an overlapping generations model of endogenous growth. Econ J 107(440):26–46 Prammer D (2019) How does population ageing impact on personal income taxes and social security contributions? J Econ Ageing 14:100186 Razin A, Sadka E, Swagel P (2002) The aging population and the size of the welfare state. J Polit Econ 110(4):900–918 Reinhart CM, Rogoff KS (2011) From financial crash to debt crisis. Am Econ Rev 101(5):1676– 1706 Röhrs S, Winter C (2017) Reducing government debt in the presence of inequality. J Econ Dyn Cont 82:1–20 Verbiˇc M, Spruk R (2014) Aging population and public pensions: theory and macroeconometric evidence. Panoeconomicus 61(3):289–316 Woodford M (1990) Public debt as private liquidity. Am Econ Rev 80(2):382–388 World Health Organization (2015) World report on ageing and health. World Health Organization Yashio H, Hachisuka K (2014) Impact of population aging on the personal income tax base in Japan: simulation analysis of taxation on pension benefits using micro data. Publ Pol Rev 10(3):519–542 Yoshino N, Miyamoto H (2017) Declined effectiveness of fiscal and monetary policies faced with aging population in Japan. Jap World Econ 42:32–44 Zweifel P, Steinmann L, Eugster P (2005) The sisyphus syndrome in health revisited. Int J Health Care Financ Econ 5(2):127–145

Part III

Short- and Medium-Term Perspective

We now turn to some more specialized topics that are not usually covered in a oneterm, contemporary analysis. Part III of the text studies how income inequality and demographic change affect subsequent economic growth globally in the short and medium terms. Fiscal policy plays a crucial role in this process.

Chapter 8

Inequality and Economic Growth: A Literature Review

8.1 Introduction The relationship between income inequality and economic growth, which is one of significant features of this day and age, has attracted much attention from scholars. Different countries with different growth experience have generated an increasing strand of literature both in theoretical and empirical trying to analyze various behaviors of inequality. The growth rate and income distribution of a country are both endogenous results of this economic system. Thus they are actually subject to normal influence, both with respect to macroeconomic policies for example. As Musgrave (1959) has pointed out, it is extensively acknowledged that fiscal policy is an effective instrument and a redistributive tool to enhance economic growth for one country. As one of the key mechanisms, the composition and combination of these policies are selected to achieve a goal of trade-off between equity and efficiency. It is believed that the effect of fiscal policy on growth and inequality comes to mind again during recent crises. It is found that financial crises broke out frequently during these years and it also brought some losses to countries. It is necessary to point out that the income distribution is disproportionately affected in different scopes not only during the crisis but also after it. It is argued that fiscal policy is a significant way to affect the income distribution. The change of government spending that is normally issued as the reaction of crises deeply hurt poor households. Roine et al. (2009) explore the determinants of income inequality by constructing a panel data analysis of 16 nations over the whole twentieth century, showing that the public spending (normally government expenditure) is positive for families with low-income and negative relationship with the upper middle group. Thus, the changes of government expenditure between different fields, further leading to the decreasing in labor demand, would affect income distribution among families and the provision of social services. These types of shifts eventually hurt those people who highly depend on those public services, particularly the poor. Consequently, based on the discussion

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 W. Luo, Inequality, Demography and Fiscal Policy, Applied Economics and Policy Studies, https://doi.org/10.1007/978-981-99-0518-8_8

123

124

8 Inequality and Economic Growth: A Literature Review

presented above, it is concluded that the fiscal policy constructed by the government plays an increasingly indispensable role in economic growth and income distribution. It is found that some developing countries (such as China, Brazil, India, Indonesia, South Africa, etc.) experience a path of high economic growth together with high inequality these years. For instance in 2008, the Chinese government issued a combination of moderately loose monetary policy and relatively positive fiscal policy when facing great recession due to the global financial crisis, such as the huge government expenditure of four trillion yuan. This stimulus package of four trillion yuan aims at keeping high economic growth in China but might lead to further income inequality. As the China Household Finance Survey (2012) has indicated in the report, the Gini coefficient in China in 2010 reached 0.61, which means that income inequality in China is increasingly severe. Nevertheless, nations facing different situations would considerably construct different fiscal policies. Some countries with high inequality experience the path of lower growth since the median voter in those nations will be relatively poor, therefore the relatively poor median voter would choose relatively higher redistributive taxes as the favor fiscal policy. This fiscal policy of higher taxes chosen by the median voter would be purely wasteful if they could not be used for a suitable way, and then this policy would dampen economic growth as Persson and Tabellini (1994) clarify in democratic regimes. While there may well be other ways to redistribute which do not lower growth but might actually raise growth. In many countries the government is responsible for affording the cost of essential public services including social protection, health care, housing and education. A significant example of this situation would be government expenditures on education which would increase physical and human capital, hence increase the growth rate (through a Lucas-model type mechanism). Importantly, if these education expenditures were constructed in a right way, it would expand human capital and hence, lower inequality (at least in the mid-run to long-run).

8.2 Literature Review In this section, this research first presents previous literatures about the relationship between growth, inequality and fiscal policy, in particular those associated with the effect of economic growth on inequality; secondly, this research introduces those effects of inequality on growth; and lastly, this research explain macroeconomic effects of fiscal policy in previous papers.

8.2 Literature Review

125

8.2.1 Effects of Economic Growth on Inequality 8.2.1.1

The Traditional Approach

As Lewis (1954), Kuznets (1955) and Kaldor (1955) demonstrate in their researches that, economic development largely determines the level of income inequality. They analyze how income distribution is affected by economic development in the long run, pointing out that the effect of growth on inequality is increasing firstly but decreasing later during the stages of economic development (‘inverted-U hypothesis’). The neoclassical growth model proposed by Solow (1956) shows a theoretical model to illustrate the relationship between capital accumulation and inequality, under the assumptions of perfect credit markets and linear saving functions. In the beginning during the path of capital accumulation, the wealth and income distribution is going to be more and more unequal, but when wealth has become sufficiently accumulated, the distribution of income and wealth would become equal since investment returns drop and wages increase enough, as suggestions from Stiglitz (1969) and Tamura (1991). Further, some previous papers tried to use the neoclassical underpinning to investigate the influences of fiscal policy on economic activities. For instance, the effects of taxes on growth are studied in the works of Sato (1967) and Krzyzaniak (1967), while the dynamic effects of fiscal policy are raised by Summers (1981) and Auerbach and Kotlikoff (1987) accepting overlapping generations in the model. Endogenous saving ratios come to mind when Judd (1985) and Chamley (1986) study the impacts of fiscal policy by using the model presented by Koopmans (1965) and Cass (1965). In short, models presented above stress the temporary effects of different fiscal policies, which means policies of taxes and expenditures could be one important factor to determine the level of output, but their effects on economic growth are not significantly permanent. By contrast, the neoclassical growth models of public-policy propose implications of the endogenous growth models in another way, showing that investment in physical and human capital could affect the equilibrium rate of growth. Therefore, fiscal policies like government expenditure and tax play an increasing role in the process of growth. As Barro (1990) and Barro and Sala-i-Martin (1992) describe, the endogenous models become permanent impacts of fiscal policy on growth, no longer be temporary effects. This means that fiscal policy could affect both the short run output level and the long run growth rate. This literature of endogenous growth points new fields that fiscal policy could permanently affect economic growth. In turn, as below, it also lights new avenues that growth and inequality might affect each other, and fiscal policy might affect aggregate demand for the purpose of redistribution.

8.2.1.2

The New Growth Theories

In the eighties and nineties century, the OECD countries experienced an increasing path of inequality, which is not consistent with the prediction of Kuznets (1955). The

126

8 Inequality and Economic Growth: A Literature Review

new theories of growth put forward three contributions with the development of new endogenous growth models: the improvement of new technologies, the increasing of international trade, and the appearance of new organizations. Hence, some papers have paid attention to skill-biased technological progress (as the major force of economic growth) and normally conclude that it might lead to higher inequality since it affects labor productivity in different ways. As Krusell et al. (2000) have argued, higher productivity leads to dropping prices of equipment, which would induce an increasing input of capital in production, therefore wage inequality would become higher across different educational levels. The second one concentrates on the influences of fast growth in international trade. According to Wood (1994) and Wood and Ridao-Cano (1999), the profits from international trade would mainly affect the production factor in the most abundant one. It is believed that developing countries are rich in unskilled labor while developed countries are rich in skilled labor. Thus, higher international trade would lead to more exports of skilled labor-intensive goods but more imports of unskilled labor-intensive goods, which could raise the demand for skilled workers and therefore enlarge the wage differential compared to unskilled workers. Finally, the third contribution turns to the effect of new organizations. The advance of technology would change the inside organization of firms, horizontal or direct communication and diversified tasks become more important. As Lindbeck and Snower (1996) mention, skilled workers have the ability to finish different tasks and learn from them. They conclude that if skilled workers could show their educational comparative edge, they would achieve wage premiums, and hence income inequality among skilled and unskilled workers would increase. These new growth theories describe that economic growth induced by the change of technology and the appearance of human capital would influence income distribution. Additionally, fiscal policy plays an increasingly indispensable role in affecting growth as an instrument and influencing redistribution as a tool. It is important to note that growth-promoting fiscal policy (like government expenditure on education or human capital investment) could have various effects on income inequality. On the one hand, those policies, which provide more opportunities for individuals to approach general knowledge, would reduce income inequality. At the same time, those policies, which encourage progress of skill-biased technology, would finally enlarge wage premiums and therefore, higher income inequality in the long term.

8.2.2 Effects of Inequality on Growth In contrast, during the 1990s s the literature concentrates on the effects of wealth and income inequality on the proceeding of economic growth, and the causation between growth and inequality goes in the opposite direction as a result. Two groups of researchers can be divided when considering this issue: one group argues that greater initial inequality would enhance economic growth through various transfer channels; while the other group argues that initial inequality might dampen growth

8.2 Literature Review

127

through some economic and political-economy channels. Therefore, the effects of redistributive policies on growth would be contrary in turn.

8.2.2.1

The Growth-Enhancing Effects of Inequality

The literature suggesting the grow-enhancing influences of inequality virtually concentrates on different preferences of saving behavior among economic agents, incentive considerations and investment indivisibilities. The growth models proposed by Lewis (1954) and Kaldor (1957) indicate that the rich shows more preference to save than the poor. If the growth rate of one country depends on the percentage of its national saving level, the majority countries unequal in distribution would see a faster growth rate than economies stressing egalitarianism. More modernly, Galor and Moav (2004) show that the aggregate output relates to the initial distribution of income under a saving function. They conclude that the output level is higher due to allowing higher propensity to save of the rich, which means that the economy is more unequal. The second theory pays attention to indivisibilities of investment, especially, it is pointed out that the implemented technical innovations and new industries would bring large sunk costs together. Aghion and Howitt (1998) discuss the imperfect credit market, suggesting that the wealth needs to be collected enough in order to cover such costs, and as a result, a new industry forms. Therefore, a more unequal country constructing more investment projects would show a faster growth trend than a country with equal wealth distribution. Eventually, the third idea could trace back to Mirrlees (1971) and is based on incentive considerations which means there is a trade-off between efficiency and equity. They accept that there is a moral risk if the reward relates to output performance: if the reward is constant and independent from output performance, the investment efforts of employees would be discouraged. However, if the reward is too sensitive it might lead to risk aversion.

8.2.2.2

The Growth-Dampening Effects of Inequality

On the other hand, the second group of researchers tries to illustrate how initial wealth and income inequality lowers growth in the long run. These theoretical literatures argue that through some economic ways the initial inequality might dampen growth. Consequently, those redistributive fiscal policies, aiming at reducing initial inequality, would enhance growth in the medium and long run at least. The economic channels mainly discuss imperfect capital market, the size of domestic market and endogenous fertility. In a neoclassical model, Stiglitz (1969) considers the aggregate capital stock as an important input of production. He argues that when capital markets are imperfect, the level of individual wealth would not be convergent and this redistribution might affect the aggregate level of output. Due to the imperfections of capital market, the poor cannot invest in indivisible production and therefore, the negative effect of inequality on growth will be severer with more unequal distribution of initial wealth. The second argument introduces the domestic

128

8 Inequality and Economic Growth: A Literature Review

market size and the impact of income distribution on various goods. In this case, the initial distribution of income could impact the growth rate in the long run by changing the composition and size of domestic market. In this sense, low-income households are not considered as participants in some consumption activities since the imperfect markets give a chance for producers to continue high prices and consequently, the long run growth would be encouraged. Finally, the third argument concerns the endogenous fertility. The initial income inequality could decrease the long run growth rate due to its positive impact on average rate of fertility. Poor parents usually prefer not to invest in the education of their children, while rich parents normally prefer to do it. In this sense, a country with more equal distribution usually experiences lower average fertility rate due to an advanced transfer of human capital assets or income. Becker and Barro (1988) explain that if the drop in average fertility rate comes with an increasing investment of human capital, the economic growth would see a rising trend in the long run. Overall, it could conclude from the above economic arguments that higher inequality might have a negative effect on growth. However, redistributive policies, aiming at alleviating income inequality, could promote economic growth. Further, these policies could improve growth rate not only in developing countries but also in numerous developed countries.

8.2.3 Effects of Inequality on Growth 8.2.3.1

The Linkage Between Fiscal Policy and Economic Development

Previous papers have widely examined the empirical impacts of fiscal policy and have various views. The majority of macroeconomic researches try to use time series models to examine the impacts of different fiscal policies on economic activities, while the sign and magnitude of those effects are various among countries. Other researches try to use a cross-country method to estimate the aggregate effect of fiscal policies on growth with a sample of countries. Nevertheless, these outcomes are not very robust, indicating that the significance and effect of these fiscal variables are based on both the chosen control variables and the conditions of the country. Besides, in order to analyze the redistributive effect of fiscal policy, some empirical papers put their variables into an inequality equation. Precisely, there are two groups of contributions can be distinguished. The first group of studies explains the effect of government spending on income distribution in OECD countries and as Galli and van der Hoeven (2001) suggest, the impact of total public spending on income inequality is significantly negative. In the case of the United States, Wolff and Zacharias (2007) use a panel data of states and conclude that government spending rather than taxes plays an important role in reducing inequality. The second group of papers discusses cases in developing countries and show that the distributive effects of fiscal policies in those countries are very weak in general. Therefore, in the case of Latin American countries, Goñ et al. (2011) show that the poor performance of fiscal policies is due

8.2 Literature Review

129

to the small size of collecting and transferring resources. Further, Lopez et al. (2010) investigate a sample of 40 developing countries, pointing that transfers government expenditure from private goods to public goods are related to poverty reduction, but the redistributive impact is neutral as targeting poor. To sum up, the results of the above empirical researches clarify that the redistributive effect of fiscal policies might be significantly associated with the development and government capacity of the country targeted.

8.2.3.2

A Better Choice

Moreover, both García et al. (2007) and Chaterjee and Turnovsky (2010) try to present theoretical models to discuss the impacts of various types of fiscal policy on growth and inequality. Their studies finally concern that some role of government spending (like public investment) could both stimulate economic growth and reduce inequality. Consequently, García et al. (2007) analyze an endogenous growth model including an elastic supply of labor and evaluate the distributive effect of ways of financing a subsidy. Their conclusions show that the pro-growth policies would lead to an increasing inequality of pre-tax income. These policies raise the return of capital and therefore, higher income inequality because capital is distributed in a more unequal way compared to labor. Nevertheless, their results also introduce that some policies could reduce inequality of post-tax, and stress that gross income inequality is an ineffective indicator for the assessment of the impacts of different policies on the welfare distribution. It is believed that the effects of some policies on the pre-tax and post-tax income distribution could be opposite, hence it is possible for a more equal post-tax income distribution to promote a faster growth. More recently, Chaterjee and Turnovsky (2010) study the impacts of growth-enhancing policies on the inequality of wealth and income and how the type of financing influences these relationships. They include heterogeneous agents in an equilibrium endogenous growth model, but the heterogeneity results from the initial private wealth endowment. Their analysis shows that government expenditure in infrastructure would raise wealth inequality no matter how this investment is financed. In contrast, they indicate that the results of income inequality are related to the way of public investment financed and might be under a inter-temporal trade-offs. This means that government spending which is collected by consumption tax results in a short-term decrease in income inequality but a long-term increase in inequality. Finally, from an empirical point of view, there are few papers that focus on the possible influences of fiscal policy on economic growth and inequality. As Muinelo Gallo and Roca Sagalés (2011) have presented, the effects of various types of fiscal policy on growth and inequality by constructing a panel method of 43 high-income countries. They argue that a rise in the government size by the ways of expenditures and taxes could lower growth but reduce inequality. It seems that the public investment would be the only one policy that could break the goal of trade-off between equity and efficiency because the rise in this policy would lessen inequality without damaging growth.

130

8 Inequality and Economic Growth: A Literature Review

Overall, the above theoretical and empirical literatures suggest that the unattractive trade-off between equity and efficiency could be broken under specific circumstances of fiscal policy. Precisely, an increasing of public investment could enhance growth while reduce income inequality in the short term.

8.3 Conclusion Regarding the above reviews of previous researches, it is obvious that previous papers and studies presented above essentially focus on the effects of economic growth on inequality, effects of inequality on growth, and macroeconomic effects of fiscal policy. However, there are still not enough studies and researchers that mainly pay attention to the possible impacts of fiscal policy on income inequality and economic growth. Especially, it seems that none of these papers attempts to concentrate on that different fiscal policies have different influences on the relationship between growth and inequality.

References Aghion P, Howitt P (1998) Endogenous growth theory. MIT press Auerbach AJ, Kotlikoff LJ (1987) Dynamic fiscal policy. Cambridge University Press Barro RJ (1990) Government spending in a simple model of endogeneous growth. J Political Econ 98(5, Part 2):S103–S125 Barro RJ, Sala-i-Martin X (1992) Public finance in models of economic growth. Rev Econ Stud 59(4):645–661 Becker GS, Barro RJ (1988) A reformulation of the economic theory of fertility. Q J Econ 103(1):1– 25 Cass D (1965) Optimum growth in an aggregative model of capital accumulation. Rev Econ Stud 32(3):233–240 Chamley C (1986) Optimal taxation of capital income in general equilibrium with infinite lives. Econometrica: J Econ Soc 607–622 Chatterjee S, Turnovsky SJ (2010) The distributional consequences of government spending. Available at SSRN 1100163 China Household Finance Survey (2012) Report of China Household Income Inequality [online]. Available from: http://chfs.swufe.edu.cn/upload/shourubupingdeng.pdf [Accessed 24 Aug 2013] Galli R, van der Hoeven R (2001) Is inflation bad for income inequality: the importance of the initial rate of inflation. In International Labor Organization Employment Paper, 2001/29 Galor O, Moav O (2004) From physical to human capital accumulation: inequality and the process of development. Rev Econ Stud 71(4):1001–1026 García Peñalosa C, Turnovsky SJ (2007) Growth, income inequality, and fiscal policy: what are the relevant tradeoffs? J Money, Credit and Banking 39(2–3):369–394 Goñi E, López JH, Servén L (2011) Fiscal redistribution and income inequality in Latin America. World Dev 39(9):1558–1569 Judd KL (1985) Redistributive taxation in a simple perfect foresight model. J Public Econ 28(1):59– 83 Kaldor N (1955) Alternative theories of distribution. Rev Econmic Stud 23(2):83–100

References

131

Kaldor N (1957) A model of economic growth. Econ J 67(268):591–624 Koopmans TC (1965) On the concept of optimal economic growth, in (Study Week on the) Econometric Approach to Development Planning, chap. 4 Krusell P, Ohanian LE, Ríos Rull JV, Violante GL (2000) Capital skill complementarity and inequality: a macroeconomic analysis. Econometrica 68(5):1029–1053 Krzyzaniak M (1967) Long-run burden of a general tax on profits in a neoclassical world. Publ Finance-Finances Publ 22(4):472–491 Kuznets S (1955) Economic growth and income inequality. Am Econ Rev 45(1):1–28 Lewis W (1954) Economic development with unlimited supplies of labor. Manchester School 22(2):139–191 Lopez RE, Thomas V, Wang Y (2010) The effect of fiscal policies on the quality of growth, No. 53855. The World Bank, pp 1–60 Lindbeck A, Snower DJ (1996) Reorganization of firms and labor-market inequality. Am Econ Rev 86(2):315–321 Mirrlees JA (1971) An exploration in the theory of optimum income taxation. Rev Econom Stud 38(2):175–208 Muinelo Gallo L, Roca Sagalés O (2011) Economic growth and inequality: the role of fiscal policies. Australian Economic Papers 50(2–3):74–97 Musgrave R (1959) The theory of public finance. McGraw-Hill, New York Roine J, Vlachos J, Waldenström D (2009) The long-run determinants of inequality: what can we learn from top income data? J Publ Econom 93(7–8):974–988 Sato K (1967) Taxation and neo-classical growth. PublFinance-Finances Publ 22(3):346–373 Solow RM (1956) A contribution to the theory of economic growth. Q J Econ 70(1):65–94 Stiglitz JE (1969) Distribution of income and wealth among individuals. Econometrica: J Econ Soc pp 382–397 Summers LH (1981) Taxation and capital accumulation in a life cycle growth model. Am Econ Rev 71(4):547–60 Tabellini G, Persson T (1994) Is inequality harmful for growth? Am Econ Rev 84(3):600–621 Tamura R (1991) Income convergence in an endogeneous growth model. J Polit Econ 99(3):522–540 Wolff EN, Zacharias A (2007) The distributional consequences of government spending and taxation in the US, 1989 and 2000. Rev Income Wealth 53(4):692–715 Wood A (1994) North-South trade, employment and inequality: changing fortunes in a skill-driven world. Clarendon Press Wood A, Ridao-Cano C (1999) Skill, trade, and international inequality. Oxford Econ Papers 51(1):89–119

Chapter 9

Inequality and Economic Growth in the Twenty-First Century

This chapter is originally published in Scottish Journal of Political Economy, 2022, 69(4), pp.345–366.

9.1 Introduction Is inequality necessarily harmful for growth? One important benchmark is an endogenous growth model designed by Persson and Tabellini (1994),1 who embed the Meltzer and Richard (1981) argument and propose that if the political decisions in a society regarding redistribution produce economic policies that tax investment and constrain growth-promoting activities, then inequality should harm growth because it increases redistributive tax pressures. A substantial amount of evidence has attempted to test the impact of inequality on growth, but the literature has not provided a satisfactory conclusion so far. For example, earlier cross-country OLS studies (e.g. see Alesina and Rodrik 1994; Persson and Tabellini 1994; Perotti 1996; Deininger and Squire 1998) all find negative consequence of higher inequality for economic performance. However, estimation using panel data, for example that assembled by Deininger and Squire (1996), generally challenges the negative effect of inequality on growth found in cross-country regressions. Barro (2000) finds little overall link between income inequality and economic growth in a panel of countries, reporting a negative effect in poor countries and a positive effect in rich countries. Perhaps the most surprising result is Forbes (2000). By controlling for country-specific effects and period effects, she finds that

1

Alesina and Rodrik (1994) and Bertola (1993) also provide similar anecdotes.

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 W. Luo, Inequality, Demography and Fiscal Policy, Applied Economics and Policy Studies, https://doi.org/10.1007/978-981-99-0518-8_9

133

134

9 Inequality and Economic Growth in the Twenty-First Century

in the short- and medium-term, an increase in the level of income inequality in a country has a positive and significant relationship with subsequent growth rates.2 In response to this puzzle, new theoretical literature has proposed mechanisms through which greater levels of income inequality can promote economic growth. For instance, Galor and Moav (2004) study the effect of inequality on growth along the process of development. In the early stages of development, when physical capital accumulation is the prime engine of growth, inequality stimulates growth as it channels resources towards individuals with more incentive to save. The positive effect of inequality on growth is reversed when human capital accumulation instead of physical capital is the primary engine for growth, where equality alleviates constraints on human capital accumulation and therefore stimulates growth.3 The theory proposed in this paper is also motivated by recent falls in the size of redistribution in particular in democracies. For example, as argued by Bonica et al. (2013), redistribution is to some extent limited through which higher rates of taxation reduce labor supply, and the model developed by Bolton and Roland (1997) shows that redistribution is limited by the consequence of deadweight loss in taxation. Furthermore, there is also a considerable literature in political science discussing the role of the media, money in politics, and non-fiscal issues (Frank 2007). The mechanism analyzed in this paper instead revisits Persson and Tabellini (1994) more closely. In their model, productivity is the only source of heterogeneity and directly affects labor income and thus capital accumulation, and only capital is actually taxed. However, labor (productivity) is not the unique source of income especially for the rich. Moreover, as Azmat et al. (2012) and Karabarbounis and Neiman (2014) have found, there has been a considerable decline in the share of labor income in recent years. Piketty (2014) further links rising inequality to the falling labor share. As we can observe, capital income has currently become more unequal as well as increasingly significant. For example, Kaymak and Poschke (2016) and Saez and Zucman (2016) show considerable rises in the wealth concentration in the U.S. over the past few decades. Luo et al. (2017), building upon Meltzer and Richard (1981), link rising capital income inequality to declining redistribution: if inequality increases such that the share of capital income going to the top capitalincome recipients increases, then the preferred tax rate falls because the (capital) rich

2

Li and Zou (1998) also find the positive link by using an improved data set on income inequality again compiled in Deininger and Squire (1996). One following empirical work is that of Frank (2009), who, estimating a dynamic panel data model but using regional data from different U.S. states, provides evidence that the long-run relationship between inequality and growth in the United States is positive and in principle driven by the upper end of the income distribution. More recently, a series of literature focusing on the relationship between inequality and growth (i.e. Breunig and Majeed 2020; Islam and McGillivray 2020; Saha and Mishra 2020) produces mixed results. 3 Moreover, Foellmi and Zweimüller (2006) study an innovation-based growth model and identify that an increasing unequal distribution of income affects the incentive to innovate through a price effect, where greater inequality allows innovators to charge higher prices, and a market-size effect, with an opposite direction. It turns out that the price effect always dominates the market-size effect, and thus increased inequality simulates growth.

9.1 Introduction

135

are supplying less taxable labor income and hence the capacity of the median voter to redistribute is reduced.4 Hence this paper instead asks how inequality stemming from capital income affects economic growth in an endogenous growth model. Individuals differ in the endowment of capital income, and the capital income distribution is right-skewed where the majority of individuals own limited (or zero) assets or wealth. However, evidence abounds of tax evasion or avoidance in capital income (see Alstadsæter et al. 2019). Beyond that, OECD data indicates that capital income taxation is unresponsive to inequality (induced by the capital endowment). Persson and Tabellini (1994) argue that productivity-induced income inequality leads to lower growth as distortionary taxes increase and harm capital accumulation. When income differences are generated by capital income, the ability of the median voter to redistribute through taxation is constrained, while such redistributive policies are financed by distortionary taxes, in principle, affecting capital accumulation and growth-promoting activities. If capital income inequality increases (and it is the rich who enjoy capital income) such that labor tax rates fall, then the subsequent rate of economic growth increases because distortionary taxes fall and investment is facilitated.5 In direct contrast to Persson and Tabellini (1994), increased inequality in capital income leads to higher growth. The relationship between inequality and growth is investigated empirically using a panel of OECD countries over the period 1975–2015, including new measures of both capital and labor income inequality as additional explanatory variables. This paper constructs the measures of inequality making use of household-level income data from the Luxembourg Income Study (LIS). The data permit more direct measurement of these two types of inequality (the advantage of separating out labor and capital income), and therefore both mean and median level income for these two types can be constructed (given several thousand households within countries in the LIS). The ratio of mean to median labor income theoretically leads to lower economic growth as in Persson and Tabellini (1994). This paper thus constructs this, and also a comparable measure for capital income inequality to test the proposed hypothesis that increased capital income inequality drives higher growth. The empirical analysis below separately includes specific measures of labor income inequality as distinct from capital income inequality. As discussed below, the two measures are empirically as well as conceptually distinct from one another. Consistent with the theory proposed, an increase in capital income inequality has a positive and significant relationship with subsequent economic growth. A one stan4

Luo (2020) extends the Luo et al. (2017) argument to analyze how government debt can be caused by changes in the income distribution. 5 If we consider that the government does engage in public investment targeted for inequality, then higher distortionary tax generated revenue could finance public investment to foster growth. Note that this paper follows Persson and Tabellini (1994) through which the incentives for productive accumulation (that determines growth) depend on the capacity of individuals to appropriate privately the fruits of their efforts, which in turn depends on the adoption of tax and regulatory policies. When political decisions tend to produce policies that allow more private appropriation, it generates more accumulation and higher growth.

136

9 Inequality and Economic Growth in the Twenty-First Century

dard deviation increase in capital income inequality is statistically correlated with a 0.42% increase in average annual growth over the next five years. The positive relationship holds up across various econometric specifications employed, including when different sample sets are considered, and also when the difference generalized method of moments technique is used to deal with the potential endogenous problem. I also find that once capital income inequality is controlled for, then the impact of labor income inequality becomes negative, consistent with Persson and Tabellini (1994) and in contrast to the empirical work using aggregative measures of inequality. A one standard deviation rise in labor income inequality is associated with a fall of 0.41% in per capita GDP growth.6 This paper is part of a small literature that attempts to obtain a better grasp of the empirical picture with respect to the growth-inequality relationship. Earlier empirical contributions include Voitchovsky (2005), Castelló-Climent (2010), and Halter et al. (2014). The first-mentioned paper questions previous empirical literature that used aggregate indicators of inequality (e.g. Gini coefficient) which may mask different impacts of the upper and bottom part of the income distribution on growth. CastellóCliment (2010), consistent with Barro (2000), states that the results of inequality are different for rich and poor countries, finding a positive effect in the group of rich countries but a negative effect in the poor group.7 Finally, Halter et al. (2014), by contrast, examine this relationship in the time dimension rather than the regional dimension (among rich and poor countries), and indeed find a positive effect in the short term but a negative effect further into the future. None of these papers, however, links inequality in the distribution of capital income to economic growth. The next section theoretically analyzes how the rate of growth changes with capital income inequality. Section 9.3 contains the empirical work, and Sect. 9.4 concludes.

9.2 The Model This model revisits Persson and Tabellini (1994) to include labor income taxation as well as wealth taxation.8 I study an overlapping generations model with constant population, where individuals live within two distinct periods. Individuals born in period t, indexed by i, have preferences defined over consumption when young 6

Some may concern the significance of this study if capital income accounts for a minor proportion in total income (as shown in Appendix Table 9.6). Note that this paper does not only focus on capital income to measure income inequality. Instead, I separate the aggregate measure of income inequality into labor income inequality and capital income inequality, and explore how both of inequality affect subsequent economic growth. The results show that the estimated relationship is sizable. 7 Banerjee and Duflo (2003) revisit both Perotti (1996) and Barro (2000) specifications, and argue that the growth rate is an inverted U-shaped function of changes in inequality. They believe that this non-linearity can account for different estimates of the relationship between inequality and growth within previous research. 8 Note that in Persson and Tabellini (1994) labor income is assumed to be untaxed.

9.2 The Model

137

ci , leisure when young l i , and consumption when old d i , represented by a strictly i ). Conconcave, continuous, twice-differentiable utility function vti = U (cti , lti , dt+1 sumption and leisure are both normal goods. Following the original, I first analyze the equilibrium behavior conditional on a given tax policy and then address the tax policy choice itself.

9.2.1 Economic Environment Income may be derived from both labor and capital, and the stock of asset, k, accumulated on average by the previous generation has a positive externality on the income of the newborn generation, as in Persson and Tabellini (1994). All individuals possess a unit of time to allocate to labor n i , or leisure l i = 1 − n i . Individual labor income yti = n i ei kt depends on productivity, ei , as well as hours worked, and is taxed at a linear rate τ . Capital income varies exogenously across individuals and is denoted by R i kt , taxed at a linear rate ϑ. Following Meltzer and Richard (1981), consumption is also financed by lump-sum redistribution, r , common to all individuals, hence the budget constraints are: i = (1 − τt )n i ei kt + rt + (1 − ϑt )R i kt cti + kt+1

(9.1)

i i = γ kt+1 dt+1

(9.2)

where k i is the individual accumulation of asset, and γ is the exogenous rate of return on asset.9 Individuals make decisions between consumption and investment when young, financed by labor and capital income as well as lump-sum transfers, and benefit from the return on that investment when old. Note that the stock of aggregate capital is accumulated as average productivity of all individuals increases. With homothetic preferences, the ratio of consumption in the two periods is independent di = D. Equivalently, every individual has of wealth and labor income taxation, ct+1 i t the same “saving rate”. In practice it is accepted that raising taxes on capital is more difficult than that on labor. Given that capital, relative to labor, is often highly mobile internationally, Diamond and Mirrlees (1971) argue that small open economies should not tax capital income. Indeed, international tax competition limits the capacity of countries to tax capital income. Although the rates of capital income taxation are positive, Gordon et al. (2004) observe lower average rates than for labor income in most countries. Further, the argument that capital income taxation is difficult to be collected also echoes with a series of academic literature (i.e. Landier and Plantin 2017;

9

Throughout the paper I use superscripts to denote individual-specific variables and no superscripts to denote average variables.

138

9 Inequality and Economic Growth in the Twenty-First Century

Alstadsæter et al. 2019).10 Over recent decades, for example, capital income taxes (as a share of GDP) have tended to be small and relatively constant (at around 2–3%) in OECD countries. Evidence from OECD data shows that they have not changed much despite other changes in the size of government, in support of that capital income taxation is comparatively fixed and unresponsive to inequality. It is an open question why the median voter would tolerate this, while conceivably the perceived deadweight and/or capital flight losses from rising capital income tax rates to some extent nullifies it as an effective instrument. Thus this paper focuses on the choice of the labor income tax. Each individual chooses labor supply so as to maximize the indirect utility function:   γ  (1 − τt )n i ei kt + rt + (1 − ϑt )R i kt , 1 − n i , vti = U γ+D (9.3)  γD  (1 − τt )n i ei kt + rt + (1 − ϑt )R i kt . γ+D The first-order condition with respect to labor supply is: γD γ (1 − τt )ei kt Uc − Ul + (1 − τt )ei kt Ud = 0 γ+D γ+D

(9.4)

which determines the labor supply, n[(1 − τt )ei , rt , R i ], for those who wish to work.11 Since leisure is a normal good, I have that ∂n i =− ∂ Ri

∂ 2 vti ∂n i ∂ R i ∂ ∂vti ( ∂n i ∂n i

)

0 for all i, so that everyone supplies a strictly positive amount of market work. 12 In detail, using (9.4), I have that ∂n i = ∂ Ri

∂ 2 vti ∂n i ∂ R i ∂v i − ∂ i ( ti ) ∂n ∂n

γ γD γD γ γ2D ( γ +D )2 (1 − τt )ei kt Ucc + ( γ +D )2 (1 − τt )ei kt Udd − γ +D Ucl + 2 (1 − τt )ei kt Ucd − γ +D Udl (γ +D)2 = (1 − ϑt )kt − < 0,

9.2 The Model

139

∂cti γ kt ∂n i [1 − ϑ = + (1 − τt )ei ], t ∂ Ri γ+D ∂ Ri γ (1 γ +D

(1 − ϑt )γ kt = γ+D

− τt )ei kt Ucl +

γD (1 γ +D

− τt )ei kt Udl − Ull

−

(9.6) > 0,

a condition which imposes additional restrictions on both Ucl and Udl . Hence, all else equal, individuals who are relatively capital-rich supply less labor and enjoy higher consumption. There are two sources of heterogeneity that determine differences in before-tax labor income. Firstly productivity, as analyzed by Meltzer and Richard (1981) and Persson and Tabellini (1994), and secondly capital income endowments, as proposed by this paper. At the individual level increases in productivity will lead to a rise in labor income all else being equal.13 On the other hand, increases in capital income will result in a fall in the labor supply all else being equal and, consequently, labor income. This then underpins their proclivity towards taxation of labor income. Average labor income can thus be written by integrating: ∞ ∞ y¯t = kt

ei n[(1 − τt )ei , rt , R i ] f (ei , R i )dei d R i 0

(9.8)

0

where f (ei , R i ) is joint distribution function of ei and R i . Individual capital and productivity endowments are conceivably correlated with each other to some extent: if, for example, highly productive individuals are simultaneously in possession of high capital income. Finally, the balanced budget requirement of government (in per capita terms) is given by: ¯ t + τt y¯t = rt (9.9) ϑt Rk

2 2  γD γ γ i i γ +D (1 − τt )e kt Ucc + Ull + γ +D (1 − τt )e kt Udd − 2 γ +D (1 −  2 γ D τt )ei kt Ucl + 2( γ +D (1 − τt )ei kt DUcd − 2 γγ+D (1 − τt )ei kt Udl < 0. 13 Note that, as in Meltzer and Richard (1981), the sign of ∂n i is indeterminate, but for any individual ∂ei with

∂ ∂n i

∂v i

( ∂nti ) ≡  =



with positive labor income I have ∂ yti ∂ei

= kt (n i + ei = kt

∂n i ) ∂ei

 γ   γ  γD γD ei γ +D (1 − τt )kt Uc + γ +D (1 − τt )kt Ud + n i γ +D (1 − τt )ei kt Ucl + γ +D (1 − τt )ei kt Udl − Ull −

> 0,

(9.7) must be positive given condition (9.6).

140

9 Inequality and Economic Growth in the Twenty-First Century

where R¯ is average capital income. For the average individual, by use of (2) and (8) I can thus solve the growth rate of k 14 ∞ ∞

¯ ei n[(1 − τt )ei , rt , R i ] f (ei , R i )dei d R i + R) − 1. γ+D (9.10) Note that analogous to (9.5), I have: D( kt+1 − kt gt = = kt

0

0

∂n i =− ∂rt

∂ 2 vti ∂n i ∂rt ∂ ∂vti ( ∂n i ∂n i

)

0,16 as in Meltzer and Richard (1981). Note that this paper is raises taxation, dm instead interested in the consequences of higher capital income inequality. If income inequality stems from a rise in capital income going to the top capital-income recipients, then the gap between taxable mean and median labor income is shortened as the capital rich supply less taxable labor income. Under the premise of the key insight of Meltzer and Richard (1981), it follows that an increase in capital income inequality lowers the tax rate chosen. Thus, I can have Lemma 9.1.17 Lemma 9.1 Consider an increase in capital-income inequality represented by a rise in the capital income earned by the top capital-income recipients. Then the labor income tax rate τ falls as capital income inequality rises. This implies that government size falls with rising capital income inequality.18 If inequality rises such that the proportion of capital income flowing to the top income recipients rises, then the preferred tax rate declines since the (capital) rich are supplying less taxable labor income and therefore, the capacity of the median voter to redistribute is constrained. The next step turns to the effect of capital income inequality on economic growth via the channel of redistribution. Combining (10) and the total derivative of y¯ , I have Lemma 9.2. Lemma 9.2 The growth rate falls as the labor income tax rate τ rises, e.g., D d( dg = dτ γ+D

∞ ∞ 0

0

¯ ei n[(1 − τ )ei , r, R i ] f (ei , R i )dei d R i + R) < 0. (9.13) dτ

Thus all else being equal, the higher the labor income taxation, the lower the growth rate. Appendix contains more mathematical details. In addition, from the properties of the g and τ functions derived above, I can obtain Lemma 9.3. Lemma 9.3 A more unequal distribution of labor income decreases economic growth, e.g., dg dτ dg = < 0. (9.14) ddm dτ dm This indicates that productivity-induced income inequality leads to lower growth as distortionary taxes increase and harm capital accumulation, which is identical in spirit to Persson and Tabellini (1994). Now consider the consequences of higher capital income inequality and the mechanism analyzed above. This paper thus establishes the following Proposition: The ratio of mean to median labor income, m = y¯ /y m , denotes labor income inequality. Please find a simple proof for Lemma 1 in Chap. 3 in Luo (2018). 18 The lemma is identical to proposition 1 in Luo et al. (2017), who assume that capital income is not taxed. 16 17

142

9 Inequality and Economic Growth in the Twenty-First Century

Proposition 9.1 Greater inequality in capital income leads to reduced demand for redistribution and thus higher growth, if the top capital-income recipients are sufficiently productive (that they also earn labor income above the median labor income). In direct contrast to Persson and Tabellini (1994), economic growth increases with increased capital income inequality. When income differences are driven by capital income, the capacity of the median voter to redistribute through taxation is reduced since the capital-rich supply less (taxable) labor.19 Such redistributive policies, financed by distortionary taxes, in principle, affect capital accumulation and growth-promoting activities, which in turn is actually detrimental to growth. If capital income inequality increases (and it is the rich who enjoy capital income) such that the labor income tax rate chosen by the median voter falls as the (capitalpoor) median voter cannot effectuate redistribution, then this generates a smaller size of redistributive policies that are financed by less distortionary taxes. If declining distortionary taxes translate into further reduced restriction on aggregate capital accumulation (and lower redistributive tax pressures), then the original Persson and Tabellini (1994) hypothesis gets reversed and it follows that subsequent economic growth will increase. For completeness, this paper shall also consider briefly what the equilibrium would look like if capital taxation was not assumed to be fixed. In this case, the decision mechanism for capital taxation is the same as for labor taxation. Governments and Parliaments, appointed on the basis of democratic elections, decide on both of them.20 If capital income inequality increases (and it is the rich who enjoy capital income, as proposed in this model), then capital income taxation rises whilst labor income taxation falls (given that the rich with extra capital income provide less labor supply). The change in total tax revenue depends on the taxation efficiency on both labor and capital, since the fall in labor taxation to some extent offsets the rise in capital taxation. As we can see, it is more difficult to escape from PSYE (labor income), whilst income from capital is harder to tax (this is also why there are several tax heavens and some practices of capital taxation have been deprecated also within the EMU, even if the introduction of such taxes has been proposed by prominent politicians and major political parties in various countries). If the taxation on capital is not fully efficient (like 100%, and this is obviously true in practice), then the fall in labor taxation can counterbalance the rise in capital taxation, leading to higher growth. These opposing effects on the subsequent economic growth rate is next examined empirically.

19

The condition that the top capital-income recipients are sufficiently productive indicates that the individuals that are in the top of the capital income distribution are never the decisive voter. 20 Within the median voter framework, it requires assumptions to get a Condorcet winner with two policy instruments (i.e. labor tax rate and capital tax rate). Following the strategy by Shepsle (1979), this paper assumes that the capital tax rate is voted on first and then for a given capital tax rate, the labor tax rate is voted on.

9.3 Evidence

143

9.3 Evidence 9.3.1 Data and Econometric Specification This section is to test whether and how economic growth changes with income inequality. The sample covers the period 1975–2015 in countries which have been OECD members since 1975.21 The mechanism invoked in this paper emphasizes the median voter framework hence established democracies are the appropriate sample. The dependent variable is the average rate of growth of income per capita over a fiveyear period as yearly growth rates incorporate short-run disturbances. For instance, this means that growth rate in period 2 is averaged over 1981–1985 and is regressed on explanatory variables measured during period 1 (1976–1980).22 This reduces yearly serial correlation from business cycles. The innovation upon their model is to include capital income inequality and labor income inequality instead of aggregate inequality. The final data set, with means, and standard deviations is contained in Table 9.1. The measure of income inequality used in traditional empirical studies is an aggregate level of inequality. One common measure of inequality frequently employed is derived from the University of Texas Inequality Project’s Estimated Household Income Inequality data.23 These data (denoted U T I P) that use the Theil’s T statistic—measured across sectors within each country—to estimate inequality, is thus tested in the empirical analysis. Assuming competitive labor markets, then wage inequality should be capturing underlying heterogeneity in productivity.24 Another measure of inequality widely accepted is the share of income received by the top one percent of the population (denoted T O P I N C), obtained from the World Inequality Database. As a result, the empirical analysis then uses this alternative measure to test the hypothesis. The top income share data show that OECD countries experienced a downward trend in the earlier years followed by a period of stasis or even slight increase since around 1990.25 A natural challenge here is that the top income share will also be picking up productivity-induced inequality. Inevitably there is a correlation between productivity inequality as measured by U T I P and the income share of the top 1%, though 21

Specifically, the countries included are Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Italy, Japan, Luxembourg, the Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, Turkey, the United Kingdom, and the United States. 22 In practice, each explanatory variable is measured in 1980, except the inequality data, which are sometimes not available in a specific year and is taken from the year closest to 1980. 23 See Galbraith and Kum (2005), and Chen and Galbraith (2012). 24 The U T I P data exhibit at increased trend in recent years, varying across countries, which indicates that this measure is close to the original conception of Persson and Tabellini (1994) of the driver of the policy-productivity-based inequality. 25 Obviously there are interesting differences across the countries, for example greater recent increases in the English-speaking countries, as discussed by Piketty and Saez (2006).

144

9 Inequality and Economic Growth in the Twenty-First Century

Table 9.1 Descriptive statistics Obs Mean UTIP TOPINC CAPINEQ LABINEQ ln(y) MEDU FEDU PPPI SHARE

182 165 116 120 216 192 192 192 156

36.92 8.67 7.49 1.24 3.32 3.59 3.16 92.26 65.01

Std. dev.

Min

Max

4.18 3.82 0.69 0.25 0.45 1.28 1.32 23.16 9.23

27.68 3.65 5.85 0.98 2.03 0.63 0.25 42.94 21.45

50.39 28.26 9.52 2.97 4.58 7.25 6.57 179.06 82.10

Notes The table presents descriptive statistics for the variables. U T I P is the University of Texas Inequality Project’s Estimated Household Income Inequality. T O P I N C is the income share of top 1%—taken from the World Inequality Database. C A P I N E Q is equal to the natural logarithm of the difference between mean and median household capital income in constant chained PPP US, and L AB I N E Q is the ratio of mean to median household labor income - both capital and labor income data are taken from the Luxembourg Income Study database. Income y is real GDP per capita in $000s of 2011 prices—taken from the Penn World Tables. M E DU and F E DU are respectively the average years of secondary schooling in the male and female population aged over 25—taken from Barro and Lee (2013). P P P I is the price level of investment measured as the PPP of investment over exchange rate relative to the United States—taken from the Penn World Tables. S H A R E is the business sector labor share—taken from the OECD database

this is somewhat weaker than might be expected. Figure 9.1 depicts a scatter plot of the two series, exhibiting a correlation coefficient of around 0.585. As noted above, previous empirical literature including both country dummies and period dummies has generally been unsupportive of the original Persson and Tabellini (1994) hypothesis. If the mechanism proposed in this paper is important, which suggests that labor income inequality and capital income inequality have opposite effects on economic growth, then arguably previous analyses have suffered from an omitted variable bias.26 Separate measures of capital income inequality and labor income inequality are therefore employed in the empirical analysis. The ideal, given the logic in Meltzer and Richard (1981), is the ratio of mean to median income. As argued in Persson and Tabellini (1994), in the case of labor income, the greater this ratio the less the stimulus for economic growth. Conversely, as this paper argues above, a greater ratio of mean to median capital income results in increased growth because distortionary taxes are constrained. For most of the OECD countries, it is applicable to access household level microdata from the Luxembourg Income Study (LIS) which would provide more direct measurement of these two types of inequality (the advantage of distinction between labor and capital income), whilst the number of observations declines due to data

26

In addition, this is also contributed by that capital-income inequality and productivity-based inequality are correlated with each other.

9.3 Evidence

145

University of Texas Inequality Project (UTIP) 30 40 45 25 35 50

UTIP Versus Top one percent income share

5 10 15 20 25 30 The share of income earned by the top one percent of the population (TOPINC) utip

Fitted values

Fig. 9.1 Scatter plot of income inequality derived from the University of Texas Inequality Project and the share of income received by the top one percent of the population derived from the World Inequality Database

availability (mostly from 1990 onwards).27 Following Meltzer and Richard (1981), labor income inequality is constructed by the ratio of mean to median household labor income in constant chained PPP US (denoted L AB I N E Q). Table 9.1 contains statistics for this measure showing a mean value of 1.24, hence (as expected) mean labor income is generally greater than the median in the LIS data. In addition to the measure of labor income inequality, the LIS data also give the chance to construct a measure of capital income inequality.28 Due to the relatively small value (mostly zero) in the data of median household capital income, this paper uses the natural logarithm of the difference between mean and median household capital income in constant chained PPP US to measure capital income inequality (denoted C A P I N E Q). For this reason the construction of capital income inequality differs slightly from that of labor income inequality. Table 9.1 shows a mean value of 7.49, indicating that capital income is concentrated within a small number of 27

The LIS contains harmonised income microdata, spanning decades, from countries mainly in Europe, North America, Latin America, Asia, and Australasia. Some may concern that the (harmonized) survey data will suffer from the shortcoming in capturing the top-end of the income/wealth distributions. Note that the LIS’ data experts harmonise the microdata into a common, cross-national template, and create comprehensive documentation, and ensure the reliability of the harmonised data. http://www.lisdatacenter.org/wp-content/uploads/brochure.pdf (accessed August 6, 2020). 28 According to the LIS, capital income is defined as cash payments from property and capital (including financial and non-financial assets), including interest and dividends, rental income and royalties, and other capital income from investment in self-employment activity. https://www. lisdatacenter.org/wp-content/uploads/files/data-lis-guide.pdf (accessed August 6, 2020).

146

9 Inequality and Economic Growth in the Twenty-First Century

1

Labor income inequality (LABINEQ) 2 3 1.5 2.5

Labor income inequality Versus Capital income inequality

6

7 8 9 Capital income inequality (CAPINEQ) labineq

10

Fitted values

Fig. 9.2 Scatter plot of labor income inequality and capital income inequality derived from the Luxembourg income study

people. This serves to highlight the fact that most households are not recipient to any capital income. While there are instances of capital income being generally more widespread, for example in the case of the minimum value of 5.85 (Spain, 1980). The theoretical argument in this paper has a direct testable prediction—labor income inequality falls with an increase in capital income inequality. Figure 9.2 confirms this with a negative relationship between these two types of inequality. The negative correlation, with an coefficient of around −0.015, also implies that these two measures are empirically and conceptually distinct from one another.29 The argument put forward in this paper is the following: as capital income inequality increases, the supply of taxable labor of the rich falls, which is likely to result in policies that allow less labor income taxation and therefore more accumulation and higher growth. The analysis includes control variables following Forbes (2000). These include the natural logarithm of real GDP per capita in constant chained PPP US$ (denoted y). Per capita GDP y and the resultant growth rates are taken from the Penn World Tables (e.g. Ram 1987). Following most empirical studies of income distribution and growth (e.g. Alesina and Rodrik 1994; Persson and Tabellini 1994) human capital effects are also included, and are represented by average years of secondary schooling in the 29

Note that the relatively higher correlation between U T I P and T O P I N C indicates that these two data to some extent capture similar information, whilst the relatively lower correlation between C A P I N E Q and L AB I N E Q again confirms that these two measures are respectively constructed by different income sources (i.e. labor and capital income) and contains conceptually different information.

9.3 Evidence

147

male and female population aged over 25 (denoted M E DU and F E DU ), drawn from the data set compiled in Barro and Lee (2013). These two schooling variables proxy for the stock of human capital at the beginning of each of the estimation periods.30 The price level of investment (the PPP of investment over exchange rate relative to the United States, denoted P P P I ) as used in Perotti (1996) are also employed in the regression analysis to capture market distortions that affect the cost of investment, also taken from the Penn World Tables. Additionally, as argued above, inequality stemming from capital income is likely to be correlated with the declined share of labor income and therefore, the labor share of income (denoted S H A R E) from the OECD database is also employed to capture whether the association between capital income inequality and growth holds strongly in smaller size of labor share.31 Finally, the country dummies are employed to control for time-invariant omitted-variable bias, and the period dummies are employed to control for global shocks that may affect aggregate growth in any periods but are not captured by other explanatory variables. To summarize, the growth model central to this section is   + αi + ηt + u i,t Gr owth i,t = β1 I nequalit yi,t−1 + xi,t−1

(9.15)

where i represents each country, t represents each time period, and u i,t is the error term. The left-hand-side variable, Gr owth, is a measure of economic growth in country i in period t. Control variables analyzed above are included in the vector xi,t . The analysis also includes country dummies, αi , and period dummies, ηt .32

9.3.2 Panel Estimation Table 9.2 contains estimation results regarding the effect of inequality on economic growth in the presence of fixed country and period effects. In this table the average rate of growth of income per capita is the dependent variable. Column 1 is a specification with an inequality measure (U T I P) and the lagged income level using five-year periods data OLS regression, with robust standard errors clustered by country. Column 2 further employs an alternative measure of inequality (T O P I N C) instead. 30 Note that M E DU and F E DU are very correlated and there is no reason to include both measures in the regression. Persson and Tabellini (1994) used historical data going back to mid-1800 whilst OECD countries after 1975 have very similar education rates across genders (as clearly emerges from the descriptive statistics), so in the below analysis this paper only includes one measure of education level (i.e. M E DU ) and leaves the other (i.e. F E DU ) as robustness check. 31 As in Facchini et al. (2017), a recent declining labor share has played a part in explaining the slowdown in the growth of government size and hence, less distortions and higher growth. 32 This paper analyzes the effect of capital income inequality on subsequent economic growth through the channel of redistribution. As presented above, only the reduced form of the model is estimated, which calls for some mechanism checks. Using OECD data, as shown in Luo et al. (2017) redistribution is found to be negatively related to capital income inequality, which reinforces the argument proposed in this paper.

148

9 Inequality and Economic Growth in the Twenty-First Century

These two columns examine the original Persson and Tabellini (1994) hypothesis, finding its coefficient to be insignificant but negative. This negative sign coheres with the argument by Persson and Tabellini (1994) but contradicts the results in Forbes (2000). As mentioned above, Column 3 utilizes the LIS data and incorporates two types of inequality, C A P I N E Q and L AB I N E Q, to separate out capital and labor income. The insignificance findings together with opposite estimated coefficients compared to previous two inequality measures challenge the argument by Persson and Tabellini (1994) as well as that by Forbes (2000). Columns 4–6 repeat the analysis of columns 1–3 including full controls instead. In the specification with two separate measures of inequality as well as the control variables, the sign of the coefficient estimate relating to capital income inequality (C A P I N E Q) is positive, and statistically significant at the 10% level. It is also noteworthy that the coefficient estimate for labor income inequality (L AB I N E Q) is negative, and statistically significant at the 10% level. This is consistent with the argument—an increase in labor income inequality is harmful for economic growth, whilst capital income inequality leads to a better growth performance. The last column contains the estimation result when the education level is measured as female average schooling years rather than that of male. The negative relationship between economic growth and labor income inequality holds up, and indeed the coefficient estimate pertaining to capital income inequality is still positive, and significantly different from zero at the 10% level. Using the estimate from column 6 of Table 9.2, the estimated coefficient for the measure of capital income inequality is positive, with a p-value of 5.9% and the estimated relationship is sizable: A one standard deviation rise in capital income inequality is statistically correlated with a 0.42% increase in average annual growth over the next five years, holding all else equal. On the other hand, the estimated coefficient for the measure of labor income inequality is negative, with a p-value of 6.6%: a one standard deviation rise in labor income inequality is associated with a fall of 0.41% in per capita GDP growth, holding all else equal.33 Most of the coefficient estimates of control variables agree with those traditionally reported in typical literature. As indicated by models considering conditional convergence, the coefficient on the initial income level is negative and statistically significant. Note that the coefficient on female education is positive.34 A rise in initial female attainment leads to more backwardness and thus faster subsequent growth since the economy converges toward a steady state. The growth promoted effect of female is meanwhile underpinned by their participation in the analysis. In addition, the coefficient on the share of labor income is negative. A reduced labor share means that capital income plays a more significant role in particular in an era of unequal capital income distribution arised, and even dominates growth activities, in support of our argument. 33

Appendix presents an approach developed by Oster (2019) to examine the sensitivity of the estimates to omitted variables. 34 As mentioned above, this paper only includes one measure of education level due to similar education rates across genders, which is different from the previous design (i.e. Perotti 1996; Forbes 2000; Barro and Sala-i-Martin 2003), which incorporates both male and female education levels in the regression.

(2)

(3)

182 Full Fixed effects Yes 0.439

165 Full Fixed effects Yes 0.453

116 Full Fixed effects Yes 0.558

Average growth rate of GDP per capita −0.0307 (0.146) −0.179 (0.196) 0.438 (0.272) 1.595 (1.545) ∗∗ ∗∗ −5.556 −5.933 −9.911∗∗∗ (2.407) (2.601) (0.996)

(1)

0.00295 (0.00885) −0.0592 (0.0481) 152 Full Fixed effects Yes 0.633

−11.10∗∗∗ (1.573) 0.186 (0.159)

−0.0279 (0.171)

(4)

−0.00388 (0.00961) −0.0563 (0.0398) 140 Full Fixed effects Yes 0.622

−9.995∗∗∗ (1.428) 0.111 (0.146)

−0.152 (0.164)

(5)

0.0121 (0.0144) −0.112∗ (0.0613) 100 Full Fixed effects Yes 0.701

0.603∗ (0.298) −1.638∗ (0.834) −11.10∗∗∗ (1.999) 0.231 (0.147)

(6)

0.323∗ (0.157) 0.0119 (0.0144) −0.107∗ (0.0603) 100 Full Fixed effects Yes 0.704

0.544∗ (0.304) −1.573∗ (0.856) −11.09∗∗∗ (2.045)

(7)

Notes Dependent variable is the average growth rate of GDP per capita. Estimations use panel regression with country fixed effects and robust standard errors clustered by country in parentheses. Year dummies are included in all regressions. Columns (4)–(6) extend columns (1)–(3) to include the lagged values of M E DU , P P P I , and SHARE as additional control variables. Column (7) again tests column (6) using the lagged F E DU as an alternative measure of human capital instead of M E DU . ∗ , ∗∗ , and ∗∗∗ respectively denote significance levels at 10%, 5%, and 1%

Observations Data Estimation Period dummies R2

L.SHARE

L.PPPI

L.FEDU

L.MEDU

L .ln(y)

L.LABINEQ

L.CAPINEQ

L.TOPINC

DEP VAR = L.UTIP

Table 9.2 Panel regressions of inequality on economic growth

9.3 Evidence 149

(2)

182 Full Fixed effects Yes 0.942

0.720∗∗∗ (0.112)

165 Full Fixed effects Yes 0.933

0.702∗∗∗ (0.120)

Log of GDP per capita −0.00125 (0.00720) −0.00881 (0.00955)

(1)

116 Full Fixed effects Yes 0.931

0.0223 (0.0132) 0.0597 (0.0713) 0.515∗∗∗ (0.0506)

(3)

0.000118 (0.000434) −0.00288 (0.00232) 152 Full Fixed effects Yes 0.951

0.462∗∗∗ (0.0788) 0.00898 (0.00767)

−0.00130 (0.00829)

(4)

−0.000226 (0.000466) −0.00300 (0.00186) 140 Full Fixed effects Yes 0.946

0.517∗∗∗ (0.0717) 0.00543 (0.00698)

−0.00768 (0.00790)

(5)

0.000553 (0.000708) −0.00546∗ (0.00297) 100 Full Fixed effects Yes 0.954

0.0298∗ (0.0150) −0.0815∗ (0.0438) 0.462∗∗∗ (0.0989) 0.0112 (0.00708)

(6)

0.0156∗ (0.00754) 0.000544 (0.000707) −0.00520∗ (0.00292) 100 Full Fixed effects Yes 0.954

0.0269∗ (0.0153) −0.0783∗ (0.0450) 0.462∗∗∗ (0.101)

(7)

Notes Dependent variable is the natural logarithm of GDP per capita. Estimations use panel regression with country fixed effects and robust standard errors clustered by country in parentheses. Year dummies are included in all regressions. Columns (4)–(6) extend columns (1)–(3) to include the lagged values of M E DU , P P P I , and SHARE as additional control variables. Column (7) again tests column (6) using the lagged F E DU as an alternative measure of human capital instead of M E DU . ∗ , ∗∗ , and ∗∗∗ respectively denote significance levels at 10%, 5%, and 1%

Observations Data Estimation Period dummies R2

L.SHARE

L.PPPI

L.FEDU

L.MEDU

L .ln(y)

L.LABINEQ

L.CAPINEQ

L.TOPINC

DEP VAR = L.UTIP

Table 9.3 Panel regressions of inequality on economic growth

150 9 Inequality and Economic Growth in the Twenty-First Century

9.3 Evidence

151

Table 9.3 reports that the broad picture is also similar when we alternatively use the level of the natural logarithm of real GDP per capita on the left-hand-side.35 The statistical significance of C A P I N E Q in the rest of the table implies that the estimates are stable when more direct measurement of two types of income inequality is applied. This evidence suggests that previous tests of the Persson and Tabellini (1994) hypothesis were hampered by the conflation of capital and labor income inequality, which in turn supports the theory proposed.

9.3.3 Robustness and Further Estimation Table 9.4 tests the robustness and contains estimation results in the presence of fixed country and period effects. Given the substantial toll on government outlays in many countries following the 2008/9 global financial crisis, column 1 shows that if the broad picture is also similar when we focus on the pre-crisis sample. Column 1 therefore repeats the analysis of column 6 of Table 9.2 using alternative period coverage, and the results support those already found. Column 2 uses the same specification but excluding Asian and Oceanian countries (such as Japan, Australia, New Zealand, and etc.) to examine whether the regional coverage of the sample affects the results. Column 3 further adds full observations available in the LIS data (rather than a restriction on the countries that have been OECD members since 1975). As can be seen the results are essentially unaltered given their exclusion and inclusion. Columns 4 and 5 split the sample by economic development according to the median value of GDP per capita. In column 4 the (relatively) high income sample again returns a positive coefficient for capital income inequality and a negative coefficient for labor income inequality of slightly larger magnitudes to that found for the full sample. In column 5 the (relatively) low income sample captures an opposite sign of the coefficient on labor income inequality, together with reduced statistical significance on the coefficient for capital income inequality. The growth-inequality relationship is somewhat looser under lower economic development. Using the estimate from column 4 of Table 9.4, the result similarly demonstrates an increased tendency to boost the economic development as capital income inequality arises. In this column, the p-value for the estimated coefficient for the measure of capital income inequality is 0.8%, and the estimated effect remains sizeable: a one standard deviation increase in capital income inequality is statistically associated with the growth in GDP per capita, which is higher by 0.62%, holding all else equal. On the other hand, the estimated coefficient for labor income inequality is still negative, with a p-value of 6.2%: an increase in labor income inequality by one standard deviation is associated with a drop in average rate of growth of GDP per capita by around 0.40%.

35

As robustness check, this paper also obtains similar patterns with a different measure of economic growth—change in the natural logarithm of real GDP per capita.

(4)

(5)

Fixed effects Yes

127 0.942 Included Full LIS sample

Fixed effects Yes

50 0.952 Included Higher income

Fixed effects Yes

50 0.933 Included Lower income

127 50 50 0.663 0.881 0.741 DEPENDENT VARIABLE = Log of GDP per capita 0.0268∗ 0.0447∗∗∗ 0.0353 (0.0223) (0.0153) (0.0148) −0.107∗∗ −0.0796∗ 0.360∗∗∗ (0.0426) (0.0396) (0.112)

Fixed effects Yes

0.175 (0.167) −0.0778 (0.0490) −0.00217 (0.00255) 100 0.954 Included Full

3.768 (3.453) −1.557 (0.939) −0.0472 (0.0527) 100 0.706

(6)

Notes: As for Tables 9.2 and 9.3. Column (1) only includes the pre-financial-crisis sample. Column (2) excludes the Asian and Oceanian observations. Column (3) includes all observations given the Luxembourg Income Study database. Columns (4) and (5) respectively correspond to higher and lower income levels. Column (6) includes an interaction term described in the text

Fixed effects Yes

Estimation Period dummies

(3) DEPENDENT VARIABLE = Average growth rate of GDP per capita 0.553∗ 0.934∗∗∗ 0.715 (0.443) (0.305) (0.297) −2.142∗∗ −1.477∗ 7.303∗∗∗ (0.817) (0.791) (2.265)

94 0.953 Included Excluding Asia and Oceania Fixed effects Yes

0.0338∗∗ (0.0154) −0.0911∗ (0.0431)

0.0286∗ (0.0155) −0.0948∗∗ (0.0404)

90 0.947 Included Pre-crisis

94 0.710

0.681∗∗ (0.308) −1.831∗∗ (0.824)

(2)

90 0.698

0.579∗ (0.306) −1.914∗∗ (0.757)

L.(C A P I N E Q *S H A R E) Observations R2 Control variables Data

L.LABINEQ

L.(C A P I N E Q *S H A R E) Observations R2 Panel B L.CAPINEQ

L.LABINEQ

Panel A L.CAPINEQ

Table 9.4 Robustness and extensions (1)

152 9 Inequality and Economic Growth in the Twenty-First Century

9.3 Evidence

153

It is natural to investigate whether or not the reported results change with the level of labor share, given that the argument invokes on the distribution of capital income. The use of an interaction term provides a way to examine how the results change with the extent of the share of labor. In column 6 of Table 9.3, S H A R E is then multiplied by the measure of capital income inequality, thereby generating an interaction term. The hypothesis here is that the relationship between the measure of economic growth and capital income inequality will be increasingly negative under larger share of labor, hence that the coefficient estimate for the interaction term is negative. The estimation results confirm this, although the significance level declines. Given a fall in the share of labor, capital income becomes more crucial and more unequal, leading to a higher rate of growth. Moreover, as for Tables 9.2 and 9.3, Panel B of Table 9.4 also reports similar patterns with an alternative measure of economic performance.

9.3.4 Generalized Method of Moments Estimation Previous literature on the effect of income inequality on economic growth discusses the necessity to deal with potential endogeneity. Following the specification by Forbes (2000), Table 9.5 applies difference generalized method of moments (GMM) by Arellano and Bond (1991) to a panel of OECD countries over 1975–2015, in the presence Table 9.5 Difference generalized method of moments regressions (1) (2) (3) DEP VAR = Average growth Log of GDP p.c. Average growth rate of GDP p.c. rate of GDP p.c. L.CAPINEQ L.LABINEQ Observations Controls Human capital measured as Data Estimation Period dummies Hansen test AR(2) p-value

(4) Log of GDP p.c.

0.603∗∗ (0.270) −1.638∗∗ (0.756) 82 Included MEDU

0.0298∗∗ (0.0136) −0.0815∗∗ (0.0396) 82 Included MEDU

0.544∗∗ (0.275) −1.573∗∗ (0.775) 82 Included FEDU

0.0269∗ (0.0139) −0.0783∗ (0.0407) 82 Included FEDU

Full Difference GMM Yes 4.81 0.936

Full Difference GMM Yes 4.80 0.926

Full Difference GMM Yes 6.44 0.910

Full Difference GMM Yes 6.47 0.900

Notes Estimations use the difference GMM of Arellano and Bond (1991), with robust standard errors. Full control variables and period dummies are included in all regressions. Endogenous variables in the regression are used as instruments. In column (1), dependent variable is the average growth rate of GDP per capita. In column (2), dependent variable is the natural logarithm of GDP per capita. Columns (3) and (4) again test columns (1) and (2) using an alternative measure of human capital, F E DU , instead of M E DU . ∗ , ∗∗ , and ∗∗∗ respectively denote significance levels at 10%, 5%, and 1%

154

9 Inequality and Economic Growth in the Twenty-First Century

of period effects and a set of control variables. In column 1, dependent variable is the average growth rate of GDP per capita. The basic difference GMM regression, eliminating the fixed effects and using lags of the endogenous variables as instruments, produces similar results as shown above. In particular, the sign of the coefficient estimate relating to capital income inequality (C A P I N E Q) is positive, and that for labor income inequality (L AB I N E Q) is negative, and both statistically significant at the 5% level. Columns 2 further uses an alternative measure of economic development (the natural log of per capita GDP). Columns 3 and 4 instead use female rather than male education attainment to measure human capital, and repeat the analysis of columns 1 and 2. It should also be noted that the AR(2) test and the Hansen J test show that there is no further serial correlation, and the overidentifying restrictions are not rejected. While heightening the concern is the problem of weak instruments in difference GMM, which led to the development of system GMM by Arellano and Bover (1995) and Blundell and Bond (1998), and could reinforce endogeneity bias. The perfect p-value of 1.00 for the Hansen test is a classic sign of instrumental proliferation.36 More importantly, throughout Table 9.5 the positive and significant coefficients on capital income inequality combined with the negative and significant coefficients on labor income inequality ultimately underpin the proposed theory.

9.4 Conclusion This paper analyzes how inequality in the capital income distribution affects growth. Capital income is quite distinct from labor income. I define it as rental income, and voters have preferences over the tax rate based on their position in the capital income distribution. Despite the fact that there are two underlying sources of heterogeneity in the populations, the median voter is still the unique Condorcet winner because tax preferences are monotonic in labor income. The result, relating growth to capital income inequality, is novel. In contrast to Persson and Tabellini (1994) increased capital-income inequality now leads to higher growth. Agents who are endowed with capital income are less averse to labor-income taxation. If the share of capital income of the rich increases such that their taxable labor supply falls and the preferred tax rate falls, then the subsequent rate of economic

36

This paper is aware that difference GMM can suffer from the problem of weak instruments, and it may be better to utilize the benefit of system GMM, which augments the equation estimated by difference GMM, simultaneously estimating an equation in levels with suitable lagged differences of endogenous variables as instruments. Note that the availability of data on inequality from the LIS (mostly starting from the year 1990) rules out this attempt. Additionally, more recent literature on weak instruments (in system GMM estimation in particular) has indicated that if instruments are weak, then inferences based on conventional Wald statistics can be misleading. However, this is still an open question, and we should not simply conclude that the system GMM estimator is not a useful tool for conducting cross-country growth empirics (see Bazzi and Clemens 2013; Kraay 2015).

9.4 Conclusion

155

growth increases because distortionary taxes fall and capital accumulation is less constrained. The relationship between inequality and growth is tested in a panel of OECD countries, including separate measures of capital and labor income inequality as additional explanatory variables. The measures are directly utilizing the survey data from the Luxembourg Income Study. Consistent with the theory, subsequent growth rate is found to be positively associated with capital income inequality. Moreover controlling for the top income share renders a consistently negative estimate for the impact of labor income inequality on growth, in line with the original Persson and Tabellini (1994) hypothesis. The positive impact of capital income inequality on growth survives in a variety of econometric specifications.

Appendix Derivation of Equation (9.13) The problem of the median voter m is to choose the tax rate so as to maximize  γ  ¯ t + (1 − ϑt )R m kt , 1 − n m , (1 − τt )n m em kt + τt y¯t + ϑt Rk γ+D  γD  ¯ t + (1 − ϑt )R m kt , (1 − τt )n m em kt + τt y¯t + ϑt Rk γ+D (9.16) and the first-order condition for the median voter with respect to the tax rate is vtm = U



 d y¯t  γ γD Uc + Ud dτt γ + D γ+D  dn m  γ γD (1 − τt )em kt Uc − Ul + (1 − τt )em kt Ud + = 0. γ+D γ+D dτt



y¯t − ytm + τt

(9.17)

Thus, making use of Eq. (9.4), the tax rate chosen by the median voter must satisfy y¯t − ytm + τt

d y¯t = 0. dτt

(9.18)

For a given labor income inequality, the political equilibrium τ is constant over time, so that the time subscript t is suppressed henceforth. Changes in the tax rate τ affect average income via two channels: its effect on the opportunity cost of leisure, and its effect on transfers (from the government’s budget constraint). In particular, I have that

156

9 Inequality and Economic Growth in the Twenty-First Century

∂ y¯ dr ∂ y¯ dθ d y¯ = + , dτ ∂r dτ ∂θ dτ ∂ y¯ d y¯ ∂ y¯ = ( y¯ + τ ) − ∂r dτ ∂θ

(9.19)

with θ = 1 − τ . Thus, the total derivative of average labor income with respect to changes in the tax rate is given by y¯r y¯ − y¯θ d y¯ = < 0, dτ 1 − τ y¯r

(9.20)

with y¯r = ∂∂ry¯ and y¯θ = ∂∂θy¯ . Moreover, for the average individual in (9.1) and (9.2), I have ¯ t − ct , kt+1 = y¯t + Rk ¯ t − dt+1 , = y¯t + Rk D γ ¯ t − kt+1 . = y¯t + Rk D

(9.21)

Solving the above equation for kt+1 , yields kt+1 =

¯ t) D( y¯t + Rk . γ+D

(9.22)

Combining the above equation and (9.8), the growth rate of k can be obtained kt+1 − kt , kt   ∞ ∞ D 0 0 ei n[(1 − τt )ei , rt , R i ] f (ei , R i )dei d R i + R¯ − 1. = γ+D

gt =

(9.23)

Again for a given labor income inequality, the political equilibrium τ and g are constant over time, so that the time subscript t is suppressed henceforth. Thus, the effect of taxation on growth, making use of (9.20), yields dg D d = dτ γ+D

 ∞ ∞ 0

0

D k1 d y¯ < 0. = γ + D dτ

 ei n[(1 − τ )ei , r, R i ] f (ei , R i )dei d R i + R¯ , dτ

(9.24)

9.4 Conclusion

157

Sensitivity to Omitted Variables Notwithstanding the data at hand can at best establish robust correlations that are consistent with the theory proposed, one may still be concerned that these results are an artifact of selection. Building upon Altonji et al. (2005), Oster (2019) develops an approach to examine the sensitivity of the estimates to omitted variables. In the Oster (2019) setting, if the relationship between the variable of interest (treatment) and the observed control variables (observables) is proportional to the relationship between the variable of interest and the omitted variables (unobservables), then the magnitudes of movements in the coefficient of interest and the R-squared value after inclusion of control variables are informative about the size of the omitted variable bias. Therefore, I examine the potential role of unobserved variables using the Oster (2019) Stata program psacalc that explore the stability of the coefficient of interest, CAPINEQ, in the face of increasing the set of control variables. This paper then calculates the estimates of the parameter δ, developed in Oster (2019), that can be interpreted as the level of selection on unobserved variables, as a proportion of the level of selection on observed variables, required to drive the estimated treatment effect to zero. A higher (absolute) value of δ implies that a higher level of selection on unobserved variables would be required for the results to be completely explained by omitted variables bias. The results of the δ test for the specifications used in the basic estimation (columns 6 and 7 of Table 9.2, with the value of −1.42 and −1.33) do not indicate omitted variables are driving the main results. Assessing the change in the δ parameter, this figure typically declines in absolute value moving from the specification of column 6 to that of column 7, potentially suggesting that we have worsened the selection problem by controlling for these variables. As a result, I select the specification of column 6 (of Table 9.2) as the preferred specification for the remaining analyses. A negative estimate of δ can be generated if the observables are positively correlated with the treatment, and the unobservables are negatively correlated with the treatment (Oster (2019)). To elaborate, negative values of δ indicate that the coefficient increases in magnitude when control variables are added. For example, the δ of −1.42 for capital income inequality in the growth equation indicates that the coefficient on CAPINEQ is larger in magnitude when the full set of controls is included. Although negative δs make this method uninformative about the size of the potential omitted variable bias, Graham et al. (2017) argue that negative values of δ indicate that results are unlikely to be driven by omitted variables. This is due to the fact that moving from zero controls to the full set of controls strengthens the coefficient of interest when δ is negative, making it unlikely that including additional unobservables would drive the estimated coefficient to 0. Overall, these results suggest that selection on unobservables is unlikely to be driving the main results.

158

9 Inequality and Economic Growth in the Twenty-First Century

Labor Income and Capital Income Table 9.6 Appendix table: descriptive statistics for labor income and capital income Variable Definition Year Mean Std. dev. Min Max Labor income

Mean household labor income

Capital income

Mean household capital income

1975 1980 1985 1990 1995 2000 2005 2010 1975 1980 1985 1990 1995 2000 2005 2010

36899.39 33499.27 30847.8 31250.48 28306.47 29660.05 33529.43 33427.67 2216.188 2389.911 1885.264 2042.408 1675.947 1648.585 2304.83 1755.311

9212.113 9591.762 10975.99 10370.58 12040.1 14827.55 15998.13 15641.51 861.2387 2390.266 1810.433 1906.911 1475.713 1439.129 2725.219 1516.935

27637.21 19142.15 11291.03 8593.753 8186.642 8592.507 10777.65 9128.012 1246.134 348.8937 411.8062 242.6643 35.32222 42.08769 57.56039 58.23211

49651.44 49342.52 49735.34 48968.01 52991.73 63228.51 61861.96 60546.98 2960.982 7557.011 6362.933 8389.775 7282.951 5283.755 13688.75 5172.714

Source The measures are utilizing the survey data from the Luxembourg Income Study, in constant chained PPP US

References Alesina A, Rodrik D (1994) Distributive politics and economic growth. Quarterly J Econ 109(2):465–490 Alstadsæter A, Johannesen N, Zucman G (2019) Tax evasion and inequality. Am Econ Rev 109(6):2073–2103 Altonji JG, Elder TE, Taber CR (2005) Selection on observed and unobserved variables: assessing the effectiveness of Catholic schools. J Political Econ 113(1):151–184 Arellano M, Bond S (1991) Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Rev Econ Stud 58(2):277–297 Arellano M, Bover O (1995) Another look at the instrumental variable estimation of errorcomponents models. J Econometrics 68(1):29–51 Azmat G, Manning A, Reenen JV (2012) Privatization and the decline of labour’s share: international evidence from network industries. Economica 79(315):470–492 Banerjee AV, Duflo E (2003) Inequality and growth: what can the data say? J Econ Growth 8(3):267– 299 Barro RJ (2000) Inequality and growth in a panel of countries. J Econ Growth 5(1):5–32 Barro RJ, Lee JW (2013) A new data set of educational attainment in the world, 1950–2010. J Dev Econ 104:184–198 Barro RJ, Sala-i-Martin X (2003) Economic growth. McGraw-Hill, New York

References

159

Bazzi S, Clemens MA (2013) Blunt instruments: avoiding common pitfalls in identifying the causes of economic growth. Am Econ J: Macroeconomics 5(2):152–86 Bertola G (1993) Factor shares and savings in endogenous growth. Am Econ Rev 83:1184–1199 Blundell R, Bond S (1998) Initial conditions and moment restrictions in dynamic panel data models. J Econometrics 87(1):115–143 Bolton P, Roland G (1997) The breakup of nations: a political economy analysis. Quarterly J Econ 112(4):1057–1090 Bonica A, McCarty N, Poole KT, Rosenthal H (2013) Why hasn’t democracy slowed rising inequality? J Econ Perspect 27(3):103–24 Breunig R, Majeed O (2020) Inequality, poverty and economic growth. Int Econ 161:83–99 Castelló-Climent A (2010) Inequality and growth in advanced economies: an empirical investigation. J Econ Inequality 8(3):293–321 Chen J, Galbraith J (2012) Austerity and fraud under different structures of technology and resource abundance. Cambridge J Econ 36(1):335–343 Deininger K, Squire L (1996) A new data set measuring income inequality. World Bank Econ Rev 10(3):565–591 Deininger K, Squire L (1998) New ways of looking at old issues: inequality and growth. J Dev Econ 57(2):259–287 Diamond PA, Mirrlees JA (1971) Optimal taxation and public production I: Production efficiency. Am Econ Rev 61(1):8–27 Facchini F, Melki M, Pickering A (2017) Labour costs and the size of government. Oxford Bull Econ Statistics 79(2):251–275 Foellmi R, Zweimüller J (2006) Income distribution and demand-induced innovations. Rev Econ Stud 73(4):941–960 Forbes KJ (2000) A reassessment of the relationship between inequality and growth. Am Econ Rev 90(4):869–887 Frank MW (2009) Inequality and growth in the United States: evidence from a new state-level panel of income inequality measures. Econ Inquiry 47(1):55–68 Frank T (2007) What’s the matter with Kansas?: how conservatives won the heart of America, Picador Galbraith JK, Kum H (2005) Estimating the inequality of household incomes: a statistical approach to the creation of a dense and consistent global data set. Rev Income Wealth 51(1):115–143 Galor O, Moav O (2004) From physical to human capital accumulation: inequality and the process of development. Rev Econ Stud 71(4):1001–1026 Gordon RH, Slemrod J (1988) Do we collect any revenue from taxing capital income?. Tax Policy Econ 2:89–130 Gordon R, Kalambokidis L, Slemrod J (2004) Do we now collect any revenue from taxing capital income?. J Public Econ 88(5):981–1009 Graham BA, Miller MK, Strøm KW (2017) Safeguarding democracy: powersharing and democratic survival. Am Political Sci Rev 111(4):686–704 Halter D, Oechslin M, Zweimüller J (2014) Inequality and growth: the neglected time dimension. J Econ Growth 19(1):81–104 Islam MR, McGillivray M (2020) Wealth inequality, governance and economic growth. Econ Modelling 88:1–13 Karabarbounis L, Neiman B (2014) The global decline of the labor share. Quarterly J Econ 129(1):61–103 Kaymak B, Poschke M (2016) The evolution of wealth inequality over half a century: the role of taxes, transfers and technology. J Monetary Econ 77:1–25 Kraay A (2015) Weak instruments in growth regressions: implications for recent cross-country evidence on inequality and growth. World Bank Policy Research Working Paper (7494) Landier A, Plantin G (2017) Taxing the rich. Rev Econ Stud 84(3):1186–1209 Li H, Zou HF (1998) Income inequality is not harmful for growth: theory and evidence. Rev Dev Econ 2(3):318–334

160

9 Inequality and Economic Growth in the Twenty-First Century

Luo W (2018) Essays on inequality and fiscal policy. PhD thesis, University of York Luo W (2020) Inequality and government debt: evidence from OECD panel data. Econ Lett 186:108869 Luo W, Pickering A, Monterio PS (2017) Inequality and the size of government. Discussion Papers 17/02, Department of Economics, University of York Meltzer AH, Richard SF (1981) A rational theory of the size of government. J Political Econ 89(5):914–927 Oster E (2019) Unobservable selection and coefficient stability: theory and evidence. J Bus Econ Statistics 37(2):187–204 Perotti R (1996) Growth, income distribution, and democracy: what the data say. J Econ Growth 1(2):149–187 Persson T, Tabellini GE (1994) Is inequality harmful for growth? Am Econ Rev 84(3):600–621 Piketty T (2014) Capital in the twenty-first century. Harvard University Press, Cambridge MA Piketty T, Saez E (2006) The evolution of top incomes: a historical and international perspective. Am Econ Rev 96(2):200–205 Ram R (1987) Wagner’s hypothesis in time-series and cross-section perspectives: evidence from “real⣞ data for 115 countries. Rev Econ Statistics 69(2):194–204 Saez E, Zucman G (2016) Wealth inequality in the United States since 1913: evidence from capitalized income tax data. Quarterly J Econ 131(2):519–578 Saha AK, Mishra V (2020) Genetic distance, economic growth and top income shares: evidence from OECD countries. Econ Model 92:37–47 Shepsle KA (1979) Institutional arrangements and equilibrium in multidimensional voting models. Am J Political Sci 27–59 Voitchovsky S (2005) Does the profile of income inequality matter for economic growth? J Econ Growth 10(3):273–296

Chapter 10

Demography and Economic Growth: The Effect of Tax Composition

Section 10.1 is originally published in Applied Economics Letters, 27(20), pp. 1629–1634. Section 10.2 is coauthored with Jingci Zhu at School of Foreign Studies, Central University of Finance and Economics, and is originally published in Applied Economics Letters, 28(8), pp. 675–680.

10.1 Population Aging and Economic Growth 10.1.1 Introduction Is population aging necessarily harmful for economic growth? Several theories document the negative effect of population aging on economic growth, either due to the smaller size of the labour force and consequent reduced productivity (see Gordon 2017), or due to an excess of savings over desired investment (see Baldwin and Teulings 2014). Luo (2019), building upon Razin et al. (2002), argues that population aging increases the demand for expenditure rather than income taxes in order to increase the tax burden on the retired population when the median voter is of working age. This paper develops the Luo (2019) hypothesis to consider how population aging affects economic growth, in particular the effect of tax composition. The main theoretical prediction is that growth increases with population aging as the extent of taxes on income relative to taxes on expenditure falls. The logic is similar to Luo (2019). Income taxes are paid solely by workers, whilst expenditure taxes are paid by both generations.1 If the median voter is a worker, then increasing the size of the retired population compels a shift in tax composition towards expenditure taxes, and therefore distortionary taxes fall and investment is facilitated, leading to higher growth. International panel evidence supports this hypothesis. 1

Admittedly, some countries include the pensions as income within the personal income tax. For simplicity, this paper follows Luo (2019) and assumes that income taxes are mainly contributed by workers. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 W. Luo, Inequality, Demography and Fiscal Policy, Applied Economics and Policy Studies, https://doi.org/10.1007/978-981-99-0518-8_10

161

162

10 Demography and Economic Growth: The Effect of Tax Composition

4

Economic Growth Versus Population Aging

Growth in GDP per capita 2 0 1 3

Equatorial Guinea

-1

El Salvador Macao SAR China China Egypt Mongolia Korea India Iran Botswana Qatar Singapore Lao PDR Vietnam Maldives Argentina Sri Lanka Cape Verde Poland Thailand Indonesia Chile Cambodia Ireland Bhutan Grenada Luxembourg Panama Hong Malaysia Malta Romania Portugal Mauritius Peru Kong SAR China Bahrain Oman Uruguay Dominican Republic Spain Brazil Iraq Bolivia Albania Antigua and Barbuda Nepal Sudan Austria Philippines Paraguay S uriname Lesotho Hungary Bangladesh Swaziland Seychelles Germany Norway Syrian Arab Republic Uganda Jordan Angola Lebanon Namibia Turkey Jamaica Belgium Morocco Tunisia Netherlands St. Vincent and the Grenadines Ecuador Costa Rica Pakistan Saudi Arabia Guatemala Mozambique Finland Belize Colombia United Kingdom Ghana Mali Trinidad Denmark and Tobago Japan Sweden Zambia New Zealand France Greece Israel Cyprus Kuwait Australia Switzerland Italy Burkina Faso St. Lucia Honduras Bulgaria Venezuela United States Fiji Mauritania Canada Iceland Ethiopia Chad Mexico South Africa Gabon Kenya Rwanda The Bahamas Cote d Congo Tanzania Guinea-Bissau Benin Nigeria Barbados Senegal Sierra Leone Togo Madagascar Liberia Niger Burundi Cameroon Djibouti The Gambia Brunei Guinea Malawi Comoros Dem. Rep. Congo Zimbabwe Central African Republic

0

10

20

30

Ratio of old to young gr

Fitted values

Fig. 10.1 Correlation between aging and growth (cross-country data over 1985–2014). Aging is defined as the ratio of the population above 65 years old to the population between 15 and 64

Empirical evidence generally has not supported the hypothesis regarding the negative effect of aging on growth (see Acemoglu et al. 2019). Figure 10.1 depicts the raw correlation between the change in GDP per capita between 1985 and 2015 and the initial ratio of the population above 65 to the population between 15 and 64, indicating no negative association between aging and the growth of GDP. In response to this puzzle, new theoretical work has proposed mechanisms through which greater fraction of the retired population can coexist with higher growth. For instance, Acemoglu and Restrepo (2017) argue that countries undergoing more pronounced demographic changes have experienced an era of the more rapid adoption of automation technologies. In an overlapping generations model, taxes are levied on both expenditure and income, financing transfer to both workers and retirees, with a balanced budget period by period (see Luo 2018). In a median voter framework, the observed tax rate is the ideal income-to-expenditure tax ratio of the median voter. Countries experiencing more rapid aging have grown more, as the median voter wants to increase the tax burden on retirees instead of being solely taxed income and prefers a lower tax ratio, which increases non-distortionary rather than distortionary taxes and facilitates investment.

10.1 Population Aging and Economic Growth Table 10.1 Descriptive statistics Obs Mean PROP65 RATIO ln(y) ln(POP) POLITY 2 ty tc

6553 6553 5606 5606 5143 2142 2141

6.65 10.60 8.77 1.78 1.94 22.45 29.06

163

Std. dev.

Min

Max

4.48 6.31 1.24 1.97 7.19 12.75 13.76

0.33 0.39 4.96 −3.20 −10 0.35 0.02

25.08 40.53 11.99 7.22 10 75.24 89.22

Notes The table gives descriptive statistics for the variables. PROP65 is the proportion of the population aged 65 and above—taken from the World Development Indicators (WDI) database. P65 RATIO = PPRROOP1564 , in which PROP1564 is the proportion of the population aged between 15 and 64. Income y is real GDP per capita in $000s of 2011 prices—taken from the Penn World Tables. POP is the size of country population—taken from the WDI. POLITY 2 is a measure of democracy provided by the Polity IV project, with −10 denoting the highest level of autocracy, and 10 denoting the highest level of democracy. t y denotes taxes on income, profits and capital gains as a percentage of revenue, and tc denotes taxes on goods and services as a percentage of revenue—both taken from the WDI

10.1.2 Data and Econometric Specification The empirical analysis examines a panel of international data over the period 1985– 2014. The dependent variable is the change in (log) GDP per capita as yearly growth rates incorporate short-run disturbances. For example, this means that the growth rate in period 2 is regressed on explanatory variables measured during period 1. This reduces yearly serial correlation from business cycles. Table 10.1 contains descriptive statistics of the variables used in the regression analysis. The key explanatory variable is the proportion of retirees in the population. The measure of it used is the percentage of the population over the age of 65 (PROP65) as in Luo (2019). This measure is preferable to the dependency ratio used by Razin et al. (2002), as the ratio includes both children and retirees which would have different impacts on taxes, as indicated by Shelton (2008). The analysis includes control variables following Acemoglu and Restrepo (2017), which focuses only on cross-country evidence. Controls include per capita GDP in constant chained PPP US$ (denoted y). Per capita GDP y and the resultant growth rates are taken from the Penn World Tables. To fully capture to some extent any scale (dis-)economies related to particular forms of tax collection that affect growth, the natural logarithm of the total population size (ln(POP)) is also included as a control. In addition to these control variables, this paper also includes some other controls for purpose of robustness check. Acemoglu et al. (2019) find that democracy does cause growth. Beyond that, economic growth rates may also be affected by the degree of democracy indirectly serving as a proxy for tax capacity. As argued in Luo (2019), in democracies with low quality institutions the link between the median voter and policy is blurred, while in countries with stronger institutions the median’s influence

164

10 Demography and Economic Growth: The Effect of Tax Composition

is stronger. If the effect of policy variables is taken into account, as proposed in this paper, then the subsequent growth will be affected. Therefore, the democracy score provided by the Polity IV project is included as a further control (POLITY 2). As this paper argues that population aging affects economic growth through the effect of tax composition, both income and expenditure tax revenue (i.e. taxes on income, profits and capital gains as a share of total tax revenue (t y ), and taxes on goods and services as a share of total tax revenue (tc )) are included to examine the growth effect of policy variables. In practice rates of tax vary with different types of income and goods, but the measure of income-to-expenditure tax ratio is a way to capture the extent to which taxes are levied on income relative to expenditure. Therefore, in the specification the tax ratio is employed instead of two separate policy variables. Due to the relatively small value in the data of taxes on goods and services in the case of some countries, following Pickering and Rajput (2018) and t Luo (2019), I use the natural logarithm of the ratio, ln( tcy ), in the below regression analysis. The argument proposed predicts that economic growth declines with an increase in the extent of taxes on income relative to expenditure.

10.1.3 Baseline Estimation Results This section is to test whether and how the change in (log) GDP per capita across different countries systematically changes with the fraction of the population that is retired. Column 1 of Table 10.2 is a simple specification with just the fraction of the retired population (PROP65) as the only regressor using five-year period data OLS regression, with robust standard errors clustered by country. Column 2 extends the regression of column 1 to include initial log GDP per capita and initial log population on the right-hand side, while column 3 also adds country fixed effects. In these specifications the sign of the coefficient estimate relating to the fraction of the retired population is positive in all cases, and all are statistically significant at the 1% level. This is consistent with the argument—an increase in the retired fraction fosters economic growth. Further, column 4 instead uses the ratio of the population above 65 against those between the ages of 15 and 64, RATIO, to measure population aging, and mimic column 3. The result similarly demonstrates an increased tendency to boost the economic development as population aging increases. Columns 5 and 6 show that if the broad picture is also similar when we focus on the post-1995 sample (1995–2014), where concerns about aging have become more prominent. Columns 5 and 6 therefore repeat the analysis of columns 3 and 4 using alternative period coverage, and the results support those already found. Columns 7 and 8 present similar patterns with a greater size of interval—regressions of the change in GDP per capita on the measure of aging using ten-year period data. Using the estimate from column 7 of Table 10.2, a one standard deviation increase in the fraction of the retired population is statistically associated with an increase of 0.31 changes in GDP per capita over ten year period, holding all else equal.

1100 164 5-year panel 1985–2014 No 0.004

0.00387*** (0.00126)

−0.0268*** (0.00979) 0.00225 (0.00433) 1100 164 5-year panel 1985–2014 No 0.014

0.00853*** (0.00221)

−0.352*** (0.0525) 0.501*** (0.0562) 1100 164 5-year panel 1985–2014 Yes 0.262

0.0599*** (0.0169)

(3)

0.0318*** (0.0104) −0.323*** (0.0453) 0.523*** (0.0562) 1100 164 5-year panel 1985–2014 Yes 0.240

(4)

−0.459*** (0.0603) 0.671*** (0.0895) 820 164 5-year panel 1995–2014 Yes 0.300

0.0825*** (0.0183)

(5)

0.0460*** (0.0122) −0.426*** (0.0520) 0.694*** (0.0922) 820 164 5-year panel 1995–2014 Yes 0.269

(6)

−0.581*** (0.0742) 0.670*** (0.0824) 608 164 10-year panel 1985–2014 Yes 0.368

0.0695*** (0.0222)

(7)

0.0323** (0.0136) −0.536*** (0.0677) 0.696*** (0.0809) 608 164 10-year panel 1985–2014 Yes 0.351

(8)

Notes Table 10.2 contains results using OLS regressions of the change of logarithm of GDP per capita. Column (1) is with just L.PROP65 as a regressor using 5-year period data OLS regression over the period 1985–2014. Column (2) includes L.ln(y) and L.ln(POP) as further control variables based on column (1). Column (3) extends column (2) to include fixed effects. Column (4) uses L.RATIO as the key explanatory variable instead of L.PROP65. Columns (5) and (6) again test columns (3) and (4) over the period 1995–2014. Columns (7) and (8) again test columns (3) and (4) using 10-year period data. Robust standard errors are shown in parentheses. Standard errors are clustered by country. *, **, and *** respectively denote significance levels at 10%, 5% and 1%

Observations Countries Data Period Fixed effects? R2

L.ln(POP)

L.ln(y)

L.RATIO

L.PROP65

Table 10.2 Basic estimation results (1) (2)

10.1 Population Aging and Economic Growth 165

166

10 Demography and Economic Growth: The Effect of Tax Composition

10.1.4 Further Estimation Results It is natural to investigate whether or not the reported results change with the degree of democracy, given that the argument invokes on the median voter framework. Column 1 of Table 10.3 thus firstly follows the specification of column 7 of Table 10.2 whilst also adds the initial degree of democracy. The estimated statistical significance of the fraction of the retired population is unaffected and even remains at the 1% level. Columns 2 and 3 then extend the regression results by splitting the sample by levels of democracy (depending on the median value of democracy score). Column 2 (column 3) contains results for countries with stronger (weaker) democratic credentials. When the sample is separated it becomes clear that the positive relationship between the fraction of the retired population and the change in GDP per capita holds only in the subsample of democratic regimes. In column 2 the p-value for the estimated coefficient for the fraction of the retired population is 4%, and the estimated effect is sizable: A one standard deviation increase in the fraction of the retired population is statistically associated with the change in GDP per capita, which is higher by 0.17, holding all else equal. The use of an interaction term provides an alternative way to examine how the results change with the extent of the franchise. In column 4, POLITY 2 is then multiplied by the aging measure, thereby generating an interaction term. The hypothesis here is that the relationship between the measure of economic growth and aging will be increasingly positive under democracies, hence that the coefficient estimate for the interaction term is positive. The estimation results confirm this, whilst the significance level falls.2 While there is a rise in the number of the retired population who prefers a higher tax ratio, leading to a lower rate of growth, the preference of the median voter shifts the other way, as she wants to shift the tax-burden onto the retired population, rather than being taxed income only from the young. Moreover, the median voter more plausibly drives policy and hence affecting economic growth under stronger democracy, consistent with the empirical results. In contrast, policy and subsequent growth is less likely to respond to changes in the preference of the median voter the less democratic is the country. Note that it is also of interest to ask how the results change with income and expenditure taxes, under the premise that economic growth rises as income taxes decline whilst expenditure taxes rise in the argument proposed in this paper. Therefore, column 5 thus include both income taxes (t y ) and expenditure taxes (tc ) as further control variables. As mentioned above, the observed tax rate is the ideal tax ratio of the median voter in this paper. Column 6 then use the income-to-expenditure tax ratio rather than two separate policy variables to examine the growth effects in the econometric analysis. If the story of the tax composition effect could explain, then income taxes will be distortionary and have negative impact on growth whilst expenditure taxes will be non-distortionary and do not reduce growth. In the case of 2

This paper also obtains essentially identical results if the author entered a dummy variable that takes the value of 1 when POLITY 2 is positive and the value of 0 when the index is negative in the interaction term.

t

532 146 10-year panel 1985–2014 Yes Full 0.378

−0.599*** (0.0840) 0.613*** (0.100) 0.00938** (0.00405)

0.0714*** (0.0244)

299 164 10-year panel 1985–2014 Yes High POLITY 2 0.341

−0.486*** (0.0978) 0.759*** (0.136) 0.0101** (0.00454)

0.0386** (0.0186)

(2)

233 164 10-year panel 1985–2014 Yes Low POLITY 2 0.478

−1.018*** (0.0884) 0.381* (0.194) 0.00517 (0.00748)

0.195 (0.166)

(3)

532 164 10-year panel 1985–2014 Yes Full 0.378

0.0687* (0.0398) 0.000215 (0.00160) −0.598*** (0.0842) 0.617*** (0.119) 0.00829 (0.0101)

(4)

155 164 10-year panel 1985–2014 Yes Full 0.712

−0.996*** (0.105) 1.806*** (0.261) 0.00401 (0.00900) −0.00711 (0.00704) 0.0119*** (0.00449)

0.0894*** (0.0291)

(5)

−0.227*** (0.0589) 155 164 10-year panel 1985–2014 Yes Full 0.724

−0.869*** (0.113) 1.806*** (0.251) 0.00527 (0.00858)

0.0766** (0.0296)

(6)

Notes Table 10.2 contains results using OLS regressions of the change of logarithm of GDP per capita, including L.ln(y), L.ln(POP), L.POLITY 2, and country fixed effects, with 10-year panel data over the period 1985–2014. Columns (2) and (3) respectively correspond to higher and lower democracy levels. Column t (4) includes an interaction term described in the text. Column (5) includes L .t y and L .tc as further control variables, and column (6) instead include L.ln( tcy ). Robust standard errors are shown in parentheses. Standard errors are clustered by country. *, **, and *** respectively denote significance levels at 10%, 5% and 1%

Observations Countries Data Period Fixed effects? Sample R2

L.ln( tcy )

L .tc

L .t y

L.POLITY 2

L.ln(POP)

L.ln(y)

L.(PROP65*POLITY 2)

L.PROP65

Table 10.3 Further estimation results (1)

10.1 Population Aging and Economic Growth 167

168

10 Demography and Economic Growth: The Effect of Tax Composition

tax ratio, economic growth will be negatively associated with the ratio. The results show that the sign of coefficients on two separate policy variables confirms this story, although the significance level on that of income taxes is weak. Moreover, statistical significance in the regression with tax ratio implies that if the extent of income relative to expenditure taxes rises, then investment is dampened and growth is promoted, which in turn supports the argument proposed.

10.1.5 Conclusion This paper analyzes how population aging affects economic growth. The change in the logarithm of GDP per capita is found to be positively associated with the fraction of the retired population. The empirical results hold across various econometric specifications employed. In particular, the results hold significantly in countries with strong democratic credentials. As a result, it is advised that governments should pay attention to the relative change between income and expenditure taxes especially in an era of aging, and the effect of aging on economic growth through the mechanism of tax composition.

10.2 Youthful Dependents and Economic Growth 10.2.1 Introduction How does age structure affect economic growth? Several theories document the positive effect of younger population on growth due to the larger size of the labor force. Luo (2019), building on Razin et al. (2002), argues that population aging leads to more demand for expenditure instead of income taxes so as to raise the tax burden on retirees when the median voter is among the working-age population. This paper develops Luo (2019) hypothesis to further consider how the youthful fraction of the population affects growth, in particular the effect of three-generation. The main theoretical prediction is that growth falls with more youthful dependents as the extent of taxes on income relative to expenditure rises. The logic is similar to Luo (2019). Income taxes are afforded solely by workers, whilst expenditure taxes are shared by three generations. As in Shelton (2008), the most likely way by which children affect the extent of tax system is through the preferences of their parents because children do not vote. Parents generally prefer a higher income-to-expenditure tax ratio rather than their child-less counterparts since they need to pay for their dependent children and at that moment, their after-tax (or disposable) income is (again) taxed at consumption. If the median voter is among the workers, then increasing the size of the youth population compels a shift in tax composition towards income taxes, and

10.2 Youthful Dependents and Economic Growth

169

-2

Growth in GDP per capita -1 0 1

2

Economic Growth Versus Youthful Dependents

20

40

60 80 Ratio of children to workers growth

100

120

Fitted values

Fig. 10.2 Correlation between economic growth and the ratio of children to workers (cross-country five-year panel data over 1980–2014)

therefore distortionary taxes rise and investment is not facilitated, leading to lower growth. International panel evidence supports this hypothesis. Empirical evidence generally has not supported the hypothesis regarding the positive effect of population rejuvenation on growth. For instance, Fig. 10.2 provides a glimpse of the raw correlation between the change in GDP per capita between 1980 and 2014 and the initial ratio of the population below 14 to the population between the ages of 15 and 64, indicating no positive association between the ratio of children to workers and the growth of GDP. Unlike the argument invoked in this paper, current empirical literature generally has focused on the upward trend of the fraction of retired population, especially in the past few decades when the structure of demography has been experiencing a dramatic change (see Acemoglu and Restrepo 2017; Luo 2019, 2020). In a three-generation model, taxes are financed by both income and expenditure, redistributing to children, workers and retirees, with a balanced budget period by period. In a median voter framework, the observed tax rate is the ideal income-toexpenditure tax ratio of the median voter as in Luo (2019). Countries experiencing more rapid growth of youth population have grown less, as the median voter wants to increase the tax burden on her/his income instead of being taxed again when pays for her/his dependent children, and prefers a higher tax ratio, which increases distortionary rather than non-distortionary taxes and investment is not facilitated.3 3

Admittedly, if the younger generation are given proper education, then the human capital they possess can to some extent mitigate the negative consequences of distortions of taxes on investment.

170

10 Demography and Economic Growth: The Effect of Tax Composition

Table 10.4 Descriptive statistics Obs Mean

Std. dev.

Min

Max

Dependency ratio Retirees Youth ln(y) ln(POP) POLITY 2 t ln( tcy )

8374

68.25

19.55

15.74

117.89

8374 8374 5606 5606 5143 2091

10.58 57.54 8.77 1.78 1.94 −0.26

6.22 23.81 1.24 1.97 7.19 1.00

0.80 14.87 4.96 −3.20 −10 −4.00

41.17 112.37 11.99 7.22 10 5.09

Median age POP growth

1608 1608

25.30 17.44

7.89 11.05

14.40 −20.98

46.35 42.29

Notes The table gives descriptive statistics for the variables. Dependency ratio is equal to the ratio of the sum of the population above 65 and the population below 14 to the population between the ages of 15 and 64, Retirees is the ratio of the population above 65 to the population between the ages of 15 and 64, and Youth is the ratio of the population below 14 to the population between the ages of 15 and 64—data on demography are taken from the World Development Indicators (WDI) database. Income y is real GDP per capita in $000s of 2011 prices—taken from the Penn World Tables. POP is the size of country population—taken from the WDI database. POLITY 2 is a measure of democracy provided by the Polity IV project, with −10 denoting the highest level of autocracy, and 10 denoting the highest level of democracy. t y denotes taxes on income, profits and capital gains as a percentage of revenue, and tc denotes taxes on goods and services as a percentage of revenue—both taken from the WDI database. Median age is the value of age in the 50th percentile of the population, and POP growth is the natural growth rate of the population, measured by the difference between the crude birth rate and the crude death rate (as a thousandth of the population)—both taken from the 2019 Revision of World Population Prospects, United Nations

10.2.2 Data and Methodology This empirical analysis focuses on a panel of international data over the period from 1980 to 2014. Following Luo (2020), the dependent variable is the change in the natural logarithm of GDP per capita since yearly growth rates employ short-run disturbances.4 This indicates that the measure of growth rate in period 2 is regressed on explanatory variables incorporated in period 1, in order to lessen yearly serial correlation caused by business cycles. Table 10.4 presents descriptive statistics of the variables used in the analysis. The measure of the dependency ratio frequently used in the empirical literature is of aggregate level. For example, Razin et al. (2002) measure this ratio with one minus the fraction of the population in the labor force. A common measure of the dependency ratio, computed by the ratio of the population either below 14 or above 65 to the population between the ages of 15 and 64 (Dependency ratio), is thus tested in the empirical analysis. Note that, however, as argued by Shelton (2008), the labor force excludes children as well as retirees. This paper then replaces the dependency 4

Data on GDP per capita are taken from the Penn World Tables.

10.2 Youthful Dependents and Economic Growth

171

ratio with two alternative measures: the ratio of children to workers (Youth) and the ratio of retirees to workers (Retirees), as in Shelton (2008).5 If the argument invoked in this paper is crucial, then children and retirees have opposite impacts on tax policy, leading to different impacts on economic growth. Therefore, the empirical analysis further uses the alternative measures to test the hypothesis. The analysis also augments control variables following Acemoglu and Restrepo (2017) and Luo (2020). In particular we control for the natural logarithm of per capita GDP in constant dollars (ln(y)) and the natural logarithm of the total population size (ln(POP)) to sufficiently capture to some extent any scale (dis-)economies related to particular manners of tax collection that affect subsequent growth. Besides, according to Acemoglu et al. (2019) democracy does cause growth. More importantly, given our argument that children affect the tax policy is through the preferences of their parents who vote in a median voter framework, growth rates may be affected by the level of democracy indirectly performing as a proxy for tax capacity. As a result, the democracy score (POLITY 2) provided by the Polity IV project is included as a further control. As in Luo (2019), in democracies with stronger institutions the median voter more plausibly drives policy, whilst in countries with low quality institutions the median’s influence is weaker. For this reason, the expectation is that the ratio of children to working-age will be more firmly related to growth rates in stronger democracies. As this paper argues that the rejuvenation of population affects economic growth t through the effect of tax composition, the tax policy (ln( tcy )), defined as the natural logarithm of the ratio of taxes on income, profits and capital gains as a share of total tax revenue (t y ) to taxes on goods and services as a share of total tax revenue (tc ), is included in the regression to test the growth effect of policy variables.6 The argument proposed predicts that economic growth falls with a rise in the extent of taxes on income relative to expenditure.

10.2.3 Baseline Estimation Results This section starts by showing that the relationship depicted in Fig. 10.2 is robust. Column 1 of Table 10.5, in the presence of fixed effects, is a simple specification with just the dependency ratio as the only regressor using five-year period data OLS regression, with robust standard errors clustered by country. Column 2 splits the dependency ratio into Retirees and Youth, whilst columns 3 and 4 respectively use the alternative measures as a regressor. Columns 5–8 then extend columns 1–4 to 5

Data on population by age are extracted from the World Development Indicators database. As a robustness check, we also examine how youthful dependents affect the composition of taxes t (ln( tcy )), utilizing the dataset used in Luo (2019). The replication of Table 10.2 in Luo (2019) instead using the youthful fraction of the population as the key explanatory variable yields a significant positive relationship between the measure of youth and the tax ratio, as conjectured by the argument invoked in this paper.

6

172

10 Demography and Economic Growth: The Effect of Tax Composition

include log GDP per capita and initial log population on the right-hand side.7 In these specifications it becomes clear that aging is found to be positively related to growth as in Acemoglu and Restrepo (2017) and Luo (2020) whilst the influence of Youth shifts the other way, leading to a negative coefficient on the dependency ratio. Note that the sign of the coefficient estimate relating to Youth is negative in all cases, and all are statistically significant at the 1% level. This is consistent with the argument—an increase in the ratio of children to workers reduces economic growth. Using the estimate from column 8 of Table 10.5, a one-standard-deviation-increase in the ratio of children to workers is statistically associated with a fall of 0.27 changes in GDP per capita, holding all else equal.

10.2.4 Instrumental Variables Estimation The empirical analysis contained above establishes a robust negative statistical association between economic growth and the ratio of children to workers in the presence of a set of controls. However, these results do not establish causality, insofar that the movements in the fraction of children to workers may be endogenous to growth, or alternatively both variables co-move in response to an unobserved third variable. What is required for identification is a source of exogenous variation in Youth. In an attempt to identify such movements we employ two instrumental variables. The first instrument is Median age, measured as the value of age in the 50th percentile of the population. Conceivably the higher the value of median age, the smaller the fraction of children to workers. The second instrument is the natural growth rate of the population (POP growth).8 When the rate of population growth is high, then conceivably this process will underpin increased youthful dependents. The presence of a second instrument allows overidentification tests to examine the exclusion restrictions. As observed in Table 10.6, the hypothesis that these instruments are weak can be rejected given that the F-statistic of the first stage regression exceeds 20. The Hansen overidentification test is not rejected, which supports the assumptions of instrument exogeneity, and the associated exclusion restrictions. Importantly the coefficient for Youth is still found to be negative.

10.2.5 Further Estimation Results It is natural to examine whether or not the results reported change with the level of democracy, in consideration of that the argument invokes on the median voter framework. Column 1 of Table 10.7 thus at first follows the specification of column 8 of Table 10.5 whilst also adds the initial level of democracy. The estimated statistical significance of the ratio of children to workers is unaffected and even remains at 7

As a robustness check, we also obtain similar patterns with a greater size of interval—regressions of the change in GDP per capita on different measures of demography using ten-year period data. 8 Both data are derived from the 2019 Revision of World Population Prospects, United Nations.

0.0020 (0.00696) −0.0060*** (0.00118)

0.0123* (0.00700)

(3)

(4) −0.0099*** (0.00225)

(5)

(6)

0.0241*** (0.00848) −0.0061*** −0.0098*** (0.00112) (0.00243) −0.3510*** −0.3996*** (0.0489) (0.0644) 0.3133*** 0.3357*** (0.0739) (0.0748) 1079 1079 1079 1079 1079 1079 161 161 161 161 161 161 Five-year panel Five-year panel Five-year panel Five-year panel Five-year panel Five-year panel 1980–2014 1980–2014 1980–2014 1980–2014 1980–2014 1980–2014 Yes Yes Yes Yes Yes Yes 0.042 0.044 0.006 0.044 0.261 0.299

−0.0062*** (0.00116)

(8)

−0.0112*** (0.00256) −0.3253*** −0.3771*** (0.0445) (0.0564) 0.5301*** 0.2943*** (0.0565) (0.0785) 1079 1079 161 161 Five-year panel Five-year panel 1980–2014 1980–2014 Yes Yes 0.244 0.281

0.0340*** (0.00995)

(7)

Notes Table 10.5 contains results using OLS regressions of the change of logarithm of GDP per capita. Estimations use panel regression with country fixed effects, and robust standard errors clustered by country in parentheses. Column (1) is with just Dependency ratio as a regressor using five-year panel data OLS regression over the period 1980–2014. Column (2) instead includes both Retirees and Youth as regressors. Columns (3) and (4) respectively use Retirees and Youth as the key explanatory variable. Columns (5)–(8) again test columns (1)–(4) but including ln(y) and ln(POP) as additional control variables. *, **, and *** respectively denote significance levels at 10%, 5% and 1%

Observations Countries Data Period Fixed effects? R2

ln(POP)

ln(y)

L.Youth

L.Dependency ratio L.Retirees

Table 10.5 Basic estimation results (1) (2)

10.2 Youthful Dependents and Economic Growth 173

174

10 Demography and Economic Growth: The Effect of Tax Composition

Table 10.6 Instrumental variables estimation results (1) L.Youth Observations No. countries Data Period Estimation method Instruments

Overidentification Weak instruments R2

−0.00630*** (0.000865) 1079 161 Five-year panel 1980–2014 Instrumental variables estimation Median age −2.231*** (0.234) POP growth 1.029*** (0.147) Hansen J = 1.793 ( p = 0.1806) F = 23.32 0.096

Notes Instrumental variables regression of the change of logarithm of GDP per capita on the ratio of children to workers using the value of age in the 50th percentile of the population and the natural growth rate of the population as instruments. *, **, and *** respectively denote significance levels at 10%, 5% and 1%

the 1% level.9 Columns 2 and 3 further extend the regression results by splitting the sample through levels of democracy (based on the median value of democracy score). Column 2 (column 3) presents the results for countries with stronger (weaker) democratic credentials. When the sample is splitted it becomes obvious that the negative relationship between the ratio of children to workers and the change in GDP per capita holds in both subsamples. Given stable results shown, the use of an interaction term offers a chance to examine if Y outh is more strongly related to growth rates in more democratic regimes. In column 4, POLITY 2 is thus multiplied by the youth measure, thereby generating an interaction term. The hypothesis in this case is that the relationship between economic growth and Y outh will be increasingly negative under democracies, hence the coefficient estimate for the interaction term is negative. The estimation result confirms this, though the significance level drops.10 While there is a rise in the number of youthful dependents who cannot vote, the median voter prefers a higher tax ratio, as s/he increasingly dislikes being doubletaxed at consumption when raises her/his dependent children, leading to a lower rate of growth. 9

The estimated effect is sizable: A one-standard-deviation-increase in the ratio of children to workers is statistically associated with the change in GDP per capita, which is lower by 0.21, holding all else equal. 10 This paper also obtains essentially identical results if we, in the interaction term, entered a dummy variable that takes the value of 1 when POLITY 2 is positive and the value of 0 when the index is negative.

934 Five-year panel 1980–2014 FE Full 0.251

−0.310*** (0.0593) 0.280*** (0.0798) 0.00436* (0.00257)

−0.00896*** (0.00264)

499 Five-year panel 1980–2014 FE High POLITY 2 0.293

−0.337*** (0.0436) 0.0849 (0.0771) 0.00540* (0.00288)

−0.0139*** (0.00261)

(2)

435 Five-year panel 1980–2014 FE Low POLITY 2 0.278

−0.330*** (0.0887) 0.395*** (0.142) 0.00211 (0.00566)

−0.0115*** (0.00397)

(3)

934 Five-year panel 1980–2014 FE Full 0.252

−0.00881*** (0.00266) −0.0000669 (0.0000797) −0.314*** (0.0591) 0.295*** (0.0831) 0.00879* (0.00490)

(4)

355 Five-year panel 1980–2014 FE Full 0.36

−0.479*** (0.0716) 0.507*** (0.183) 0.00265 (0.00497) −0.00157 (0.0399)

−0.0132*** (0.00442)

(5)

0.188*** (0.0249) 784 Five-year panel 1980–2014 Arellano-Bond Full

−0.796*** (0.0262) 0.191*** (0.0523) 0.00786*** (0.00240)

−0.0173*** (0.00145)

(6)

Notes Column (1) contains results using OLS estimation of the change of logarithm of GDP per capita, including L.ln(y), L.ln(POP), L.POLITY 2, and country fixed effects, with five-year panel data over the period 1980–2014. Columns (2) and (3) respectively correspond to higher and lower democracy levels. Column t (4) includes an interaction term described in the text. Column (5) includes L.ln( tcy ) as a further control variable. Column (6) contains the Arellano-Bond dynamic panel data estimation results. *, **, and *** respectively denote significance levels at 10%, 5% and 1%

Observations Data Period Estimation method Sample R2

L.(DEP VAR)

L.ln( tcy )

t

L.POLITY 2

L.ln(POP)

L.(Youth × POLITY 2) L.ln(y)

L.Youth

Table 10.7 Further estimation results (1)

10.2 Youthful Dependents and Economic Growth 175

176

10 Demography and Economic Growth: The Effect of Tax Composition

Note that it is also of interest to ask how the results change with the composition of taxes, under the premise that economic growth falls as income taxes rise whilst expenditure taxes fall in the argument invoked in this paper. Therefore, column 5 t thus includes the income-to-expenditure tax ratio, ln( tcy ), as a further control variable to examine the growth effects in the econometric analysis. If the story of the effect of tax composition could explain, then income taxes will be distortionary whilst expenditure taxes will be non-distortionary and consequently, economic growth will be negatively related to the tax ratio. The result indicates that the sign of coefficient on the policy variable confirms this story, though the significance level on that is weak. This shows that if the extent of income relative to expenditure taxes increases, then investment is constrained and growth is depressed, in support of our argument. The last column presents the Arellano-Bond dynamic panel data estimation results. The negative relationship between subsequent growth and Youth holds up, and significantly different from zero at the 1% level.11

10.2.6 Conclusion This paper analyzes how younger population affects economic growth. The change in the natural logarithm of GDP per capita is found to be negatively associated with the ratio of children to workers. The empirical results hold across various econometric specifications. Thus, it is suggested that governments should pay more attention to the relative change between income and expenditure taxes in particular in an era of falling fertility rates, and the effect of less children on economic growth through the mechanism of tax composition. One limitation of this paper is that our literature review is relatively limited, and we have yet to anticipate any economy-wide substitution and income impacts of taxes through changes in relative prices.

References Acemoglu D, Restrepo P (2017) Secular stagnation? The effect of aging on economic growth in the age of automation. Am Econ Rev 107(5):174–79 Acemoglu D, Naidu S, Restrepo P, Robinson JA (2019) Democracy does cause growth. J Polit Econ 127(1):47–100 Baldwin R, Teulings C (2014) Secular stagnation: facts, causes and cures. Centre for Economic Policy Research-CEPR, London Gordon RJ (2017) The rise and fall of American growth: the US standard of living since the civil war, vol 70. Princeton University Press Luo W (2018) Essays on inequality and fiscal policy. PhD thesis, University of York 11

In Table 10.7, the results relating to the control variables are of some interest. As documented by models considering conditional convergence, the coefficients on initial income is negative and statistically significant. In addition, countries with stronger democratic credentials are found to have a higher rate of growth, in line with Acemoglu et al. (2019).

References

177

Luo W (2019) Demography and the composition of taxes: evidence from international panel data. Econ Lett 183:108518 Luo W (2020) Demography and economic growth: the effect of tax composition. Appl Econ Lett 27(20):1629–1634 Pickering A, Rajput S (2018) Inequality and the composition of taxes. Int Tax Public Finan 25(4):1001–1028 Razin A, Sadka E, Swagel P (2002) The aging population and the size of the welfare state. J Polit Econ 110(4):900–918 Shelton CA (2008) The aging population and the size of the welfare state: is there a puzzle? J Public Econ 92(3–4):647–651

Chapter 11

Tax Composition and Economic Growth in the Age of Demographic Change

This chapter is originally published in Policy Studies, 2022. https:// doi.org/ 10.1080/ 01442872.2022.2096212.

11.1 Introduction Does the composition of taxes affect the long-run growth rate? There are two schools of thought depending on the type of public policy growth model adopted for the analysis. On the one hand, public policy neoclassical growth models regard fiscal policy as one of the determinants of output. Hence, the steady-state growth rate is determined by exogenous factors such as technological progress and population growth, while fiscal policy only plays a role in the transition towards this steady state. On the other hand, public policy endogenous growth models (i.e. Barro 1990; Barro and Sala-i-Martin 1992; Mendoza et al. 1997) demonstrate the mechanisms by which fiscal policy can affect both output and the steady-state growth rate. Under endogenous growth models, the elements of government taxation can be classified into distortionary and non-distortionary taxation. Distortionary and nondistortionary taxes are distinguished according to whether they affect the decision to invest in physical and/or human capital, leading to tax wedges and thus distorting the steady-state growth rate. These results can be extended by allowing different forms of taxation to be distortionary to different degrees (Devarajan et al. 1996; Mendoza et al. 1997). For example, Mendoza et al. (1997) argue that if leisure enters the utility function, then consumption taxation becomes distortionary (while it is nondistortionary in Barro 1990 model) by affecting education/labour-leisure and further capital/labour ratios in production. Tax decisions are also affected by the aging of the population. The proportion of the population aged 65 years or older increased to around 17% in Organisation for Economic Co-operation and Development (OECD) countries in 2015. This rapid aging of the population in developed countries is seen as one of the most dangerous © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 W. Luo, Inequality, Demography and Fiscal Policy, Applied Economics and Policy Studies, https://doi.org/10.1007/978-981-99-0518-8_11

179

180

11 Tax Composition and Economic Growth in the Age of Demographic Change

threats to their economies. Luo (2018) theorizes that population aging affects the tax structure, thereby increasing demand for expenditure rather than income taxes. Hence, to analyse the taxation–growth nexus empirically, it is crucial to include the aging measure in the vector of the non-fiscal variables as well as an interaction term of the aging measure and expenditure taxes (non-distortionary taxes in this context). Building on the prediction of endogenous growth models that shifting government revenue away from distortionary taxes towards non-distortionary taxes can promote growth, in this study, we hypothesize that the impact of non-distortionary taxes on economic growth increases with aging. In other words, the impact of nondistortionary taxes on growth is context dependent. The decline in economic growth due to levying non-distortionary taxes expands as the proportion of older people in the population grows. Thus, the declines in the growth effect of non-distortionary taxes for lower proportions of older people tend to be small, in line with Barro (1990) argument. However, as the proportion of older people rises, the government’s stance on revenue shifts and declines in growth manifest more concretely. The theory proposed in this paper is also motivated by recent increases in taxes on goods and services in the present era of rapid demographic change. Revenue sources outside income taxes such as taxes on goods and services are thus becoming increasingly important components of revenue for OECD countries. For example, they have risen to approximately 30% of total taxation in the United Kingdom in recent decades. Moreover, recent demographic studies have tended to focus on the importance of taxes on goods and services, given the declining proportion of tax revenue from income. For instance, Luo (2018, 2019) explores the relationship between demographic characteristics and the extent to which taxes are levied on income relative to expenditure. Luo (2020) further considers the impact of the demographic structure on economic growth through the mechanism that population aging increases demand for expenditure taxes rather than income taxes; meanwhile, Luo and Zhu (2021) extend this argument to the three-generation framework. However, no studies have thus far examined how the tax composition (i.e. the balance between distortionary and non-distortionary taxes) affects economic growth, particularly in those countries undergoing the most pronounced demographic changes. To bridge this gap in the body of knowledge, the relationship between the composition of taxes and growth is investigated empirically in this study using an updated dataset, namely, a panel of OECD countries over 1980–2015 in five-year intervals. We select this extended data range to include the recent dramatic change in the demographic structure in these nations. The dependent variable is the growth rate of real per capita gross domestic product (GDP) taken from the Penn World Tables. The fiscal data used are collected from the Government Finance Statistics Yearbook of the International Monetary Fund (IMF). Following Kneller et al. (1999), this study treats income and profit taxes, social security contributions, and payroll and property taxes as ‘distortionary’ and consumption (expenditure-based) taxes as ‘non-distortionary’. By analysing data that include these recent demographic changes, an increase in distortionary taxation is found to be negatively associated with per capita GDP growth, consistent with the original prediction and most of the empirical literature.

11.1 Introduction

181

However, a negative effect of non-distortionary taxation on growth is also found during this period. This negative relationship is robust to different econometric specifications (e.g. when the budget constraint is mis-specified). Indeed, in the presented panel estimation with country and year dummies, a one standard deviation increase in non-distortionary taxation is significantly associated with a fall of 0.54% in average annual growth over the five-year period, holding all else equal. This raises an interesting question: why have expenditure taxes, previously described and examined as ‘non-distortionary’, now become so ‘distortionary’? In contrast to Barro (1990) model, expenditure taxes (whether constant or timedependent) become distortionary and have a negative effect on growth when leisure enters the utility function (in other words, labour supply is elastic), as in Mendoza et al. (1997) model. In this case, expenditure taxes indirectly distort the decision to invest by affecting the choice between labour/education and leisure, which in turn affects the capital/labour ratio in production. The results of this study thus reflect a recent decline in the effective opportunity cost of leisure by depressing consumption forgone by working less; in other words, labour supply has become more elastic. However, this study instead argues that the common practice of setting expenditure taxes at a variety of rates for different goods and services leads to distortions in an era of rapid population aging. Figure 11.1 depicts the proportion of the population aged above 65 during 1980–2015, with data taken from the World Development Indicators (WDI). It shows that OECD countries first experienced a period of stability, followed by an upward trend in the later years. Luo (2019), based on Razin et al. (2002), proposes that population aging thus raises demand for expenditure taxes rather than income taxes to shift the tax burden towards older people. The main theoretical prediction in this study is that growth falls with a rise in taxes on goods and services induced by population aging. The logic is that if the government levies a higher tax on goods as a result of aging, then the quantity of outputs traded declines, leading to a larger deadweight loss and slower growth. Evidence from the OECD panel data supports this hypothesis. We find that the relationship between economic growth and taxes on goods and services is increasingly negative under the aging measure. Further, this finding remains robust when system generalized method of moments (GMM) estimations are used to mitigate potential endogeneity. This study makes two crucial contributions to the literature. First, it enriches existing knowledge on the relationship between the tax structure and economic growth. In particular, focusing on OECD countries and adopting an empirical approach, our findings confirm that income and profit taxes, social security contributions, and payroll and property taxes (i.e. distortionary taxes) are negatively related to economic growth, as already found in existing studies. However, we also find that consumption taxes are negatively related to economic growth, in contrast to previous studies that have not shown any significant relationship. This negative relationship between consumption taxes and economic growth is interpreted to be related to the recent rapid aging of the population in OECD countries, which is captured by our investigated time period. Second, the negative relationship between consumption taxes and economic growth relating to the aging of the population could provide practical suggestions for countries suffering rapid demographic change (i.e. advanced countries) and for

182

11 Tax Composition and Economic Growth in the Age of Demographic Change

Fig. 11.1 Population aging from 1980 to 2015

countries likely to fight the problem of population aging in the future (i.e. developing countries). Our results suggest that developing countries should pay attention to the shift from direct to indirect taxes during the period of population aging, while advanced countries undergoing rapid demographic change should adjust their tax systems given that expenditure taxes have become distortionary to a certain extent. The remainder of the paper is organized as follows. Section 11.2 reviews the related literature on taxation and economic growth. Section 11.3 describes the model and data. Section 11.4 presents the estimation results, and Sect. 11.5 concludes.

11.2 Literature Review Neoclassical growth models (e.g. Solow 1956; Swan 1956) indicate that tax and expenditure measures do not affect the steady-state growth rate; however, Jones et al. (1993), Stokey and Rebelo (1995), building on the work of Barro (1990), King and Rebelo (1990), Lucas (1990), use endogenous growth models to extend the analysis and demonstrate the conditions under which fiscal variables can affect growth. The more recent theoretical literature (Park and Philippopoulos 2003; Peretto 2003, 2007) is also informative. However, empirical evidence derived from examining the predictions of endogenous growth models emphasizes the importance of a complete structure of both expenditure and taxation. In contrast to focusing only on the expenditure side, as in Devarajan et al. (1996), Mendoza et al. (1997) consider only the taxation side and

11.3 Materials and Methods

183

argue that the tax composition has no significant effect on growth—even if it produces significant private investment effects. This finding is borne out by the seminal empirical work of Kneller et al. (1999), which includes a full specification of the government budget constraint. They test the growth effects of fiscal policy for a panel of 22 OECD countries over 1970–1995 using the criteria put forward by Barro (1990) to classify relevant fiscal data and find strong support for Barro (1990) model. More importantly, they show that distortionary taxation (e.g. taxation on income and profit) reduces growth, while non-distortionary taxation (taxation on goods and services) does not. Further, they show that productive government expenditure (e.g. transport and communication expenditure) raises growth, while non-productive expenditure (e.g. social security and welfare expenditure) does not. More recent empirical studies have continued to investigate the impact of the tax structure on economic growth. Bleaney et al. (2001), Gemmell et al. (2011) show that only distortionary taxes lead to lower growth rates. A series of research, including Widmalm (2001), Angelopoulos et al. (2007), Romero-Avila and Strauch (2008), Ojede and Yamarik (2012), argue that tax categories have different effects on growth. Lee and Gordon (2005), Acosta-Ormaechea et al. (2019), Yanikkaya and Turan (2020) employ data on a larger group of countries and reconfirm this finding. Moreover, the results of Acosta-Ormaechea et al. (2019) on tax and growth ranking have attracted significant scholarly attention. Baiardi et al. (2019), using data on OECD countries, show that the relationship between taxes and growth that exists to some extent becomes less significant when the number of countries and time period are extended. Beyond that, Xing (2012) finds no robust growth ranking among taxes on corporate income, personal income, and consumption. The main arguments for the relationship between taxation and growth in OECD countries and the European Union are formulated in several ways, including by examining the effect of growth on marginal tax rates (Padovano and Galli 2001), the effect of growth on tax components (Tosun and Abizadeh 2005), a tax design for inclusive growth (Brys et al. 2016), the effect of the tax burden on growth (Blanco and Delgado 2019), and the relationship between the tax structure and growth (Arnold 2008; Johansson et al. 2008; Stoilova 2017).

11.3 Materials and Methods This section introduces our proposed model, which is similar to that used in most empirical work on fiscal policy and growth. The aim of this model is to estimate growth as a function of a bundle of fiscal variables, initial income, the investment ratio, and the labour force growth rate. Apart from a set of control variables, the estimations in this study use panel regression with country dummies to control for time-invariant omitted variable bias and period dummies to control for global shocks. In this sense, our model is almost identical to that used by Kneller et al. (1999) find a negative effect of distortionary taxation on growth, but no effect of nondistortionary taxation. However, one key change from their work is to extend the fiscal

184

11 Tax Composition and Economic Growth in the Age of Demographic Change

dataset to 2015 to incorporate the period in which OECD countries experienced rapid population aging. While this change might affect aggregate growth in any period, it is not adequately captured by existing work. This study starts by analysing those countries that have been OECD members since 1970 and then moves onto examining a sample of current members.1 In accordance with usual practice, the dependent variable is the log difference in annual per capita GDP, as yearly growth rates incorporate short-run disturbances. The analysis also follows the standard practice of using five-year periods to eliminate yearly serial correlation from business cycles. Hence, we estimate six five-year periods of per capita GDP growth for each OECD country over 1980–2015, including only those countries with observations for at least two consecutive periods. Applying these criteria to the dataset results in a sample of 30 countries and 130 observations. Following Kneller et al. (1999), we classify the fiscal variables into four categories: distortionary/non-distortionary taxes and productive/non-productive expenditure. We also add the budget surplus of the government as well as revenue and expenditure for which the classification is unclear (labelled ‘other revenue’ and ‘other expenditure’, respectively). All these fiscal data come from the Government Finance Statistics Yearbook of the IMF, and they are aggregated into six main categories in this study, as described in Appendix Table 11.6.2 In this context, one crucial issue is the allocation of taxes to the categories of distortionary and non-distortionary taxes. Given that most major taxes used in OECD countries are distortionary in some respect, it is accepted that the relevant distortion is that to the incentive to invest (i.e. in physical and/or human capital) when we test endogenous growth models. Following Kneller et al. (1999), this study thus treats income and property taxes as ‘distortionary’ and consumption (expenditurebased) taxes as ‘non-distortionary’. This is because expenditure-based taxes do not lower returns on investment despite being likely to affect the education/labour–leisure decision. As one important determinant of growth is the initial level of development, we include per capita GDP in constant chained PPP US$, taken from the Penn World Tables, as the first control in the regression analysis. As in the classical Barro-type regression, the investment ratio (e.g. gross fixed capital formation as a share of GDP) and the labour force growth rate are also included, taken from the WDI database.3 Further controls are employed in the regression since both the demographic structure and openness affect economic growth (Acemoglu and Restrepo 2017; Grossman and Helpman 1990) as well as tax policy (Persson and Tabellini 2003). Demographic 1

Appendix Table 11.5 lists the countries included in the analysis. Appendix Table 11.7 presents the correlation coefficients of the variables. This study also tests the variance inflation factors, which range from 1.74 to 5.52 due to the mis-specification of the budget constraint. Both results indicate that multicollinearity is not an issue. 3 The Durbin–Wu–Hausman test statistics indicate that most of the variables on the right-hand side are endogenously determined. This implies that the ordinary least squares (OLS) results are inconsistent due to endogeneity issues. The procedure used to carry out the Durbin–Wu–Hausman test in STATA is explained at https://www.stata.com/support/faqs/statistics/durbin-wu-hausmantest/ (accessed May 16, 2020). 2

11.3 Materials and Methods

185

0

per capita real GDP 40 60 20

80

real GDP p.c. versus non-distortionary taxation

5

10 15 non-distortionary taxation (as % of GDP) yp_pwt

20

Fitted values

Fig. 11.2 Per capita gross domestic product and non-distortionary taxation

effects are encapsulated by the percentage of the population aged between 15 and 64 years and the percentage aged over 65 (denoted P R O P1564 and P R O P65, respectively), also taken from the WDI database. Openness is measured by the trade share (the sum of exports and imports as a percentage of GDP, denoted T rade), again taken from the WDI database. Table 11.1 presents the descriptive statistics of the variables used in the analysis. The sample countries grew, on average, by 1.95% per capita per annum, with investment ratios in excess of 23% and labour force growth below 0.9% per annum. Among the taxation variables, the distortionary tax category yields about twice as much revenue (around 22.5% of GDP on average) as non-distortionary taxes. Figure 11.2 depicts a scatterplot of per capita GDP and non-distortionary taxes, showing a correlation of around −0.20. As this indicates a negative relationship between income and non-distortionary taxes over 1980–2015, it is meaningful to examine whether non-distortionary taxes have reduced growth in recent years compared with earlier. In addition, this study has four other novelties compared with Kneller et al. (1999). First, the estimations switch from the contemporary estimation (as in Kneller et al. 1999) to a panel data fixed effects regression with lagged values of the tax (and other control) variables, since fiscal policy as well as investment and labour take time and ignoring this leads to low levels of significance.4 Applying lags can also make endogeneity less likely to some extent. Second, in accordance with the standard growth model combined with initial income value, economic development is 4

For example, this means that the growth rate in period 2 is regressed on the explanatory variables measured in period 1.

186

11 Tax Composition and Economic Growth in the Age of Demographic Change

Table 11.1 Descriptive statistics Obs Mean GDP p.c. Investment Labour force growth Net lending Distortionary taxation Nondistortionary taxation Other revenue Productive expenditure Nonproductive expenditure Other expenditure Budget surplus Trade PROP1564 PROP65

Std. dev.

Min

Max

1025 1034 875

29.95 23.04 0.88

13.39 4.05 1.56

6.85 11.55 −4.71

84.42 39.40 10.85

831 603

0.98 22.45

4.01 6.22

−29.28 4.49

20.67 36.66

662

11.15

2.78

3.83

17.98

598 580

7.73 25.18

3.59 4.21

2.07 15.27

26.22 45.34

580

16.82

4.74

6.10

28.28

580

2.33

0.99

−8.91

5.52

638

−2.27

4.54

−32.12

18.70

1034 1085 1085

80.47 66.54 13.68

48.95 2.60 3.72

16.01 53.27 3.92

419.53 73.02 26.34

Notes The table provides the descriptive statistics of the variables. Income and resultant growth are taken from the Penn World Tables. The investment ratios and labour force growth rates are taken from the World Development Indicators (WDI) database. The fiscal data are collected from the Government Finance Statistics Yearbook of the International Monetary Fund. The data are consolidated and cover all levels of government. All the fiscal variables are expressed as percentages of gross domestic product (GDP). P R O P1564 and P R O P65 are the proportion of the population aged between 15 and 64 and aged 65 and above, respectively, taken from the WDI database. T rade is the sum of exports and imports as a percentage of GDP

measured by the natural logarithm of per capita GDP rather than per capita GDP, as used by Kneller et al. (1999). Third, the empirical analysis then incorporates an interaction term between non-distortionary taxation and the aging measure to capture whether the novel results are due to population aging. The expectation is that non-distortionary taxation is negatively and more strongly related to growth in the subsample with the more severe aging problem. Fourth, this study further employs the system GMM technique of Blundell and Bond (1998) with a level equation, a standard model in growth regressions, to control for potential endogeneity bias and allow us to compare the fixed effect results.

11.4 Results

187

11.4 Results 11.4.1 Contemporary Effects This section presents the results of testing whether and how the change in (log) per capita GDP across OECD countries systematically changes with the composition of taxes. Panel A of Table 11.2 shows the estimation results from the OLS regressions of per capita GDP growth on the tax composition, using the set of control variables and country and period dummies (with standard errors clustered at the country level). The data for this analysis are the five-year data from between 1980 and 2015.5 Column (1) presents the current consensus, augmenting the benchmark controls used in Kneller et al. (1999) with non-productive expenditure as the implicit financing element. Here, the sample is those OECD members that have joined since 1970. In line with most of the empirical literature on taxes and growth, the estimated coefficient of distortionary taxation in column (1) of Panel A becomes negative, with a p-value of 1.9%. Further, the estimated relationship is sizeable: A one standard deviation increase in distortionary taxation per percent of GDP is significantly associated with a 0.35% fall in average annual growth. By contrast, non-distortionary taxation is also found to be negatively associated with per capita GDP growth, significant at the 10% level, in contrast to our expectations. Moreover, to interpret the fiscal parameters, it is important to fully specify the government budget constraint, as in Kneller et al. (1999). They demonstrate that failure to do so (e.g. omitting or mis-specifying the budget constraint) will lead to serious errors and incorrect conclusions. To examine whether the results, particularly for non-distortionary taxation, change in this setting, columns (2)–(5) of Table 11.2 show the results with the mis-specified budget constraint. In column (2), the three expenditure variables are omitted from the regression, while only distortionary and non-distortionary taxation are included in column (3). Further, columns (4) and (5) only include one tax variable each. In these specifications, the effect of non-distortionary taxation on growth remains negative.6 Columns (1)–(5) are based on the five-year averages of the years with the final digits 1–5 and 6–10 to use the full dataset and follow convention. We also explore the consequences of changing the time periods to the years with the final digits 2–6 and 7–11, which employ a similar number of observations, and find similar results. Using the same set of countries, these results indicate that the relationship between non-distortionary taxes and growth is different from that found in the traditional literature. Hence, we investigate whether the results change with the entry or exit of countries from the OECD. Columns (6)–(10) of Table 11.2 (Panel A) include the sample of countries that joined the OECD before 2015, using the same specification 5

The use of clustered standard errors prevents us from inflating the accuracy of the estimates due to the potential within-group correlation of the error terms. 6 Indeed, initial per capita GDP enters the regressions with a significant negative coefficient, implying the conditional convergence of growth rates over this period.

188

11 Tax Composition and Economic Growth in the Age of Demographic Change

as in columns (1)–(5). For this analysis, the number of observations rises from 92– 128 and the number of countries rises from 21 to 30 (see column (6)). We find that the results for the tax variables persist following the entry of new countries and that the significance levels of the two tax variables even increase slightly (which is expected given the larger sample). More importantly, the negative relationship between non-distortionary taxation and growth persists. We also explore the consequences of changing the time periods to those years with the final digits 2–6 and 7–11, which employs countries with OECD membership before 2015 over the full sample period of 1980–2015. The results are broadly similar. Although the significance levels of the two tax variables tend to be marginally lower, they are still significantly negative. The point estimates of the coefficients are around −0.22% for distortionary taxation and −0.72% for non-distortionary taxation on average.7 Given that the results in Panel A of Table 11.2 are based on the sample period of 1980–2015, we finally assess whether the negative relationship between non-distortionary taxation and growth changes with the dramatic change in the demographic structure between 1990 and 2015 (i.e. the period of rapid population aging). As shown in Panel B, the broad picture is similar for the post-1990 sample period.8

11.4.2 Dynamic Effects To overcome the problems in Kneller et al. (1999) work, as mentioned above, we run a series of additional analyses. Table 11.3 presents the estimation results from the fixed effects panel regressions of growth on the (lagged) tax composition, with a battery of (lagged) control variables and period dummies (with standard errors clustered at the country level), using the same sample of countries as in columns (6)–(10) of Table 11.2 over 1980–2015. In addition to time dummies, time trends are employed to be certain we are using detrended regressors (Wooldridge 2013). Again, we obtain essentially identical results. To investigate whether the results change with the measure of aging, column (1) of Panel A adds initial demographic variables and openness on the right-hand side. In this specification, the coefficient of the proportion of older people (P R O P65) is positive and significant at the 5% level, in line with the findings of Luo (2020), which are derived using international panel data. 7

Productive expenditure has a significant and positive coefficient, and the point estimate suggests that a one percentage point increase in GDP raises the growth rate by 0.35% points on average. 8 Easterly and Rebelo (1993) argue that the significance of the fiscal variables in econometric regressions is sensitive to the inclusion of an initial income term. The removal of this term collapses the basic regression to a simple form of a growth accounting equation. As the term of initial per capita GDP is a significant regressor in the tables presented thus far, it would not be surprising if the results are sensitive to its exclusion. However, in an unreported analysis, we find that the coefficients of the two tax variables are close to those presented previously; this implies that the significance of the tax variables in the growth regression is not sensitive to this change in the specification based on the data used.

−0.278** (0.109)

0.798

R 2 (within)

0.786

93

−0.313 (0.238)

−0.164* (0.0920)

Fixed effects

Yes

(5)

0.754

93

−0.518** (0.233)

−0.185** (0.0813)

0.717

93

−0.0862 (0.0788)

0.754

106

−0.540* (0.269)

0.737

128

−0.812*** (0.198)

−0.415** (0.158)

Yes

Fixed effects

All fiscal variables

0.810

78

−0.587** (0.219)

−0.152 (0.121)

Yes

Fixed effects

All fiscal variables

0.771

78

−0.0372 (0.120)

Yes

Fixed effects

All fiscal variables

0.802

88

−0.671** (0.322)

Yes

Fixed effects

None

0.714

109

−0.767*** (0.226)

−0.347** (0.161)

Yes

Fixed effects

All expenditure

0.703

115

−0.515*** (0.148)

−0.177** (0.0798)

0.731

134

−0.708*** (0.166)

−0.339*** (0.0808)

(7)

Yes

Fixed effects

All fiscal variables

0.643

116

−0.618*** (0.174)

−0.242** (0.0949)

0.685

138

−0.825*** (0.200)

−0.305*** (0.0966)

(8)

Yes

Fixed effects

All fiscal variables

0.598

116

−0.139 (0.103)

0.633

138

−0.249* (0.125)

(9)

Including countries joining the OECD before 2015 (6)

Yes

Fixed effects

All fiscal variables

0.661

126

−0.578*** (0.210)

0.652

151

−0.821*** (0.213)

(10)

Notes Columns (1) and (6) include Initial GDP p.c., Investment, Labour force growth, Net lending, Other revenue, Other expenditure, Budget surplus, and Productive expenditure as the control variables. Columns (2) and (7) omit Other expenditure, Productive expenditure, and Non-productive expenditure. Columns (3)–(5) and (8)–(10) omit Net lending, Other revenue, Other expenditure, Budget surplus, Productive expenditure, and Non-productive expenditure. Standard errors clustered at the country level are shown in parentheses. *, **, and *** denote the significance levels of 10, 5, and 1%, respectively. OECD: Organisation for Economic Co-operation and Development; GDP: gross domestic product

Yes

Fixed effects

Estimation

Time dummies

0.836

All expenditure

0.848

None

R 2 (within)

Omitted control variables

78

No. of observa- 77 tions

−0.123 (0.141)

−0.426** (0.196)

−0.262* (0.149)

Nondistortionary −0.667*** taxation (0.227)

Distortionary taxation

(4)

Estimates of the impact of the tax composition on economic growth from 1990 to 2015

92

No. of observations

Panel B.

(3)

Estimates of the impact of the tax composition on economic growth from 1980 to 2015

(2)

Including countries joining the OECD before 1970

(1)

Non-distortionary −0.472* taxation (0.266)

Distortionary taxation

Panel A.

Coverage

Table 11.2 Panel regressions of per capita gross domestic product growth on the composition of taxes

11.4 Results 189

190

11 Tax Composition and Economic Growth in the Age of Demographic Change

Using an interaction term can allow researchers to examine how the results change with the degree of population aging. Therefore, following Brambor et al. (2006), we use an interaction model to test the extent to which population aging increases the negative relationship between non-distortionary taxation and economic growth. In column (2) of Table 11.3 (Panel A), P R O P65 is multiplied by the measure of nondistortionary taxation to generate an interaction term. The hypothesis here is that the relationship between economic growth and non-distortionary taxation is increasingly negative as population aging worsens (i.e. the coefficient of the interaction term is negative). The estimation results, which are significant at the 5% level, support this hypothesis. Population aging increases demand for expenditure taxes relative to income taxes, and a higher tax on goods thus reduces the quantity of outputs traded, leading to a larger deadweight loss and lower growth rates. Figure 11.3 shows the marginal effect of non-distortionary taxes on economic growth for different levels of the aging measure. This relationship is robust when the budget constraint is misspecified (see columns (3) and (4) of Panel A) as well as when aging is alternatively measured by the ratio of the population above 65 to the population between 15 and 64 (see Panel B).9

11.4.3 System Generalized Method of Moments Estimation The regression assumes that all the variables on the right-hand side are exogenously determined. According to Easterly and Rebelo (1993), the most likely sources of simultaneity in the regression are the effect of the business cycle and Wagner’s law, which states that higher per capita GDP tends to be associated with higher government expenditure. Although we use five-year periods to control for the potential effect of the business cycle, some endogeneity may remain. Conversely, Wagner’s law is less of a concern because it indicates a correlation between GDP growth and the growth rate of government expenditure and taxation, whereas we focus on the levels of the fiscal variables. Nonetheless, to address any remaining concerns about endogeneity, we adopt the system GMM method of Blundell and Bond (1998), as the difference GMM approach proposed by Arellano and Bond (1991) can suffer from the weak instrument problem (see Roodman 2009). The results are shown in columns (1)–(4) of Table 11.4.10 To augment the difference GMM method, we simultaneously estimate an equation in levels using suitable lagged differences of the endogenous variables as instruments to examine the sensitivity of the results to reducing the num9

Although we acknowledge that the coefficient of distortionary taxation is negative but not significant in some of the columns, all the t-statistics of the coefficients of distortionary taxation are above one, suggesting weak significance. This is acceptable given that the number of observations is below 100. 10 We also apply the difference GMM approach, which eliminates the fixed effects and uses lags of the endogenous variables as instruments, to the sample of OECD countries for 1980–2015 using the data over five-year periods. We find that this approach produces similar results to those in Table 11.4, particularly regarding the significant and negative coefficients of distortionary and non-distortionary taxation, reconfirming the main results.

(2)

All expenditure Fixed effects Yes

Notes Columns (1) and (2) include the initial levels of log of GDP p.c., Investment, Labour force growth, Net lending, Other revenue, Other expenditure, Budget surplus, Productive expenditure, T rade, and P R O P1564 as the control variables. Column (3) omits the initial levels of Other expenditure, Productive expenditure, and Non-productive expenditure. Column (4) omits the initial levels of Net lending, Other revenue, Other expenditure, Budget surplus, Productive expenditure, and Non-productive expenditure. Standard errors clustered at the country level are shown in parentheses. *, **, and *** denote the significance levels of 10%, 5%, and 1%, respectively. GDP: gross domestic product

None Fixed effects Yes

All fiscal variables Fixed effects Yes

None Fixed effects Yes

Omitted control variables Estimation Time dummies

−0.00571** (0.00255) 0.0582*** (0.0201) 0.0807*** (0.0236) −0.00409** (0.00151) 107 31 0.769

(4)

−0.0109*** (0.00390) 0.0745*** (0.0220) 3.895*** (1.197) −0.375*** (0.104) 107 31 0.692

−0.00703** (0.00334) 0.0627** (0.0286) 0.0916*** (0.0277) −0.00440** (0.00183) 103 31 0.786

(3)

Panel B. Aging measured by the ratio of the population above 65 to the population between 15 and 64 −0.00737 −0.00809 −0.0116** L. (Distortionary taxation) (0.00640) (0.00557) (0.00462) L.(Non-distortionary taxation) −0.00551 0.0700** 0.0573* (0.0101) (0.0276) (0.0284) L.(PROP65/PROP1564) −0.518 3.498** 3.044** (0.500) (1.438) (1.354) L. (Non-distortionary taxation −0.354*** −0.303** × P R O P65P R O P1564) (0.122) (0.124) No. of observations 98 98 103 30 30 31 No. of countries 0.787 0.808 0.716 R 2 (within)

Panel A. Aging measured by the ratio of the population above 65 to the total population −0.00878 −0.00909 L. (Distortionary taxation) (0.00677) (0.00607) L. (Non-distortionary taxation) −0.00542 0.0634** (0.00954) (0.0287) L.(P R O P65) 0.0229** 0.0779*** (0.0112) (0.0268) L. (Non-distortionary taxation −0.00482** × P R O P65) (0.00190) No. of observations 98 98 30 30 No. of countries 0.804 0.821 R 2 (within)

(1)

Table 11.3 Panel regressions of per capita gross domestic product growth on the composition of taxes 11.4 Results 191

192

11 Tax Composition and Economic Growth in the Age of Demographic Change

-.1

Effects on linear prediction -.05 0 .05

.1

Average marginal effects of L.(Non-distortionary taxation) with 95% CIs

4

6

8

10

12

14 16 L.(PROP65)

18

20

22

24

Fig. 11.3 Marginal effect of non-distortionary taxes on economic growth for different levels of the aging measure

ber of instruments (see columns (5)–(8)). Further, the results of an AR(2) test and a Sargan test show that there is no further serial correlation and that the overidentifying restrictions are not rejected. In an additional analysis, we test the validity of the subsets of the instruments (levels, differenced, and standard instrumental variable (IV) instruments), as the Sargan test evaluates the entire set of overidentifying restrictions/instruments. Specifically, we use a difference-in-Sargan/Hansen test, known as the C-test (see Baum 2006). The null hypothesis of the C-test is that the specified variables are valid instruments. As shown in the table, the null hypothesis of the exogeneity of any of the GMM instruments used, or of the validity of the standard IV instruments, cannot be rejected.

11.5 Discussion This study analyses how the tax composition (i.e. the balance between nondistortionary and distortionary taxation) affects per capita GDP growth. We categorize income and profit taxes, social security contributions, and payroll and property taxes as ‘distortionary’ and consumption taxes as ‘non-distortionary’. Using a sample period of 1980–2015 to account for the period of dramatic change in the demographic structure in OECD member states, we find that distortionary taxation does reduce economic growth. However, we also show that non-distortionary taxation has consistent negative effects on growth throughout the sample period. This study thus argues that distortions from expenditure taxes are caused by the recent rapid rise in the proportion of older people in these countries. This relationship is

System GMM Yes 0.346 0.291 7.93 98

Estimation Time dummies AR(2) p-value Overidentification Difference-in-Hansen test No. of instruments Method to reduce instruments System GMM Yes 0.350 0.268 7.74 96 Lags 1–1

98 30 None

−0.00211*** (0.000810)

(5) −0.00167 (0.00120)

(7)

103 107 31 31 All expenditure All fiscal variables System GMM System GMM Yes Yes 0.325 0.326 0.138 0.011 9.95 16.46 84 65 Lags 1–1 Lags 1–1

−0.00198* (0.00120)

(6)

System GMM Yes 0.278 0.194 9.11 89 Lags 1–1

98 30 None

−0.128* (0.0693)

(8)

Notes Columns (1) and (5) include the initial levels of Distortionary taxation, Non-distortionary taxation, log of GDP p.c., Investment, Labour force growth, Net lending, Other revenue, Other expenditure, Budget surplus, Productive expenditure, T rade, P R O P65, and P R O P1564 as the control variables. Column (2) omits the initial levels of Other expenditure, Productive expenditure, and Non-productive expenditure. Column (3) omits the initial levels of Net lending, Other revenue, Other expenditure, Budget surplus, Productive expenditure, and Non-productive expenditure. Columns (4) and (8) include the initial levels of Distortionary taxation, Non-distortionary taxation, log of GDP p.c., Investment, Labour force growth, Net lending, Other revenue, Other expenditure, Budget surplus, Productive expenditure, T rade, and P R O P65/P R O P1564 as the control variables. *, **, and *** denote the significance levels of 10%, 5%, and 1%, respectively. GDP: gross domestic product; GMM: generalized method of moments

System GMM Yes 0.276 0.300 12.16 98

98 30 None

103 107 31 31 All expenditure All fiscal variables System GMM System GMM Yes Yes 0.342 0.176 0.372 0.229 12.38 14.95 103 107

98 30 None

No. of observations No. of countries Omitted control variables

(4)

−0.106* (0.0569)

−0.00149* (0.000863)

(3)

−0.00114 (0.000851)

(2)

L.(Non-distortionary taxation −0.00184** × P R O P65) (0.000757) L.(Non-distortionary taxation × P R O P65/P R O P1564)

(1)

Table 11.4 System generalized method of moments regressions of per capita gross domestic product growth on the composition of taxes

11.5 Discussion 193

194

11 Tax Composition and Economic Growth in the Age of Demographic Change

Table 11.5 List of countries Country AUSTRALIA AUSTRIA BELGIUM CANADA CHILE CZECH REPUBLIC DENMARK ESTONIA FINLAND FRANCE GERMANY GREECE HUNGARY ICELAND IRELAND ISRAEL ITALY JAPAN LUXEMBOURG MEXICO NETHERLANDS NEW ZEALAND NORWAY POLAND PORTUGAL SLOVAK REPUBLIC SLOVENIA SOUTH KOREA SPAIN SWEDEN SWITZERLAND TURKEY UK USA

Membership

Geographic location

7 June 1971 29 September 1961 13 September 1961 10 April 1961 7 May 2010 21 December 1995 30 May 1961 9 December 2010 28 January 1969 7 August 1961 27 September 1961 27 September 1961 7 May 1996 5 June 1961 17 August 1961 7 September 2010 29 March 1962 28 April 1964 7 December 1961 18 May 1994 13 November 1961 29 May 1973 4 July 1961 22 November 1996 4 August 1961 14 December 2000 21 July 2010 12 December 1996 3 August 1961 28 September 1961 28 September 1961 2 August 1961 2 May 1961 12 April 1961

Oceania Europe Europe North America South America Europe Europe Europe Europe Europe Europe Europe Europe Europe Europe West Asia Europe East Asia Europe North America Europe Oceania Europe Europe Europe Europe Europe East Asia Europe Europe Europe West Asia/Europe Europe North America

Notes This table shows the list of countries joining the Organisation for Economic Co-operation and Development before 2015

11.5 Discussion

195

Table 11.6 Theoretical aggregation of the functional classifications Theoretical classification Functional classification Distortionary taxation

Non-distortionary taxation Other revenue

Productive expenditure

Unproductive expenditure Other expenditure

Taxes on income, profits, and capital gains Social security contributions Taxes on payroll and workforce Taxes on property Taxes on goods and services Taxes on international trade Other tax revenue Non-tax revenue Expenditure on general public services Expenditure on defence Expenditure on education Expenditure on health Expenditure on housing and community amenities Expenditure on transport Expenditure on communication Expenditure on social protection Expenditure on economic affairs Other expenditure

Notes Functional classifications refer to the classifications given in the data source, as in Kneller et al. (1999)

robust to different econometric specifications (e.g. using interaction effects) as well as when system GMM estimations are used to mitigate potential endogeneity. Ideal tax systems are often formulated by economists, but politicians cannot always adopt such an ideal tax design because of the risk of losing popularity with the electorate. One practical implication of our results is that our analysis could be applied to design tax systems that can be implemented in the real world. Further, in this era of population aging, advanced economies should focus on whether expenditure taxes hamper growth, while developing countries should pay attention to the relative change between income and expenditure taxes. One limitation of this study is its failure to provide a theoretical basis for the empirical analysis. Moreover, the presented empirical analysis may not be a satisfactory way to identify the causal relationship between taxes and economic growth. Hence, a further research direction could be to derive a theoretical model to analyse how the tax structure affects economic growth during rapid demographic change. Another direction would be to find a better identification method to examine the causal impact of the tax structure on economic growth.

Appendix

ln(GDP p.c.)

−0.1840 0.0006 0.1789 −0.1185 0.4496

−0.0668

0.0837

−0.3454

−0.2364

−0.3962

1.0000

0.3516

0.1501 0.1123 −0.0497 0.3804 −0.1305 Non-productive expenditure

−0.3331 0.3483 0.1900 0.2711 0.5663 Productive expenditure 1.0000

1.0000 −0.2214 −0.1357 0.1111 0.0288

−0.2085 0.2941 0.0616 −0.1142 −0.4646 Other expenditure

−0.2343

−0.3175

0.1021

−0.1126 −0.0889

−0.0401

−0.3042

0.2125

1.0000 0.3145 −0.1356 −0.0849

Labour force growth

1.0000 0.0937 −0.0639 −0.2277 −0.2076

Investment

−0.0212

1.0000 −0.1106 0.1697 0.3321 0.4319 −0.2119

1.0000 0.1910 −0.0337 −0.1377

1.0000 0.2622 0.1699

1.0000 0.3161

0.1315 0.3257 −0.1278 0.0781 PROP1564

−0.1118

0.0487 −0.1590 0.5414 Trade

−0.0384

−0.1433

0.5601

0.2814 0.3879

1.0000

0.7931 0.0078 −0.1311 −0.1300 Budget surplus

0.7464

−0.0883 0.4739

1.0000 0.2206

Distortionary taxation Non-distortionary taxation

−0.2554

0.0161

0.5032 0.0016

1.0000 0.1178 0.1710

Net lending

Notes This table shows the correlations of the variables. GDP: gross domestic product

Productive expenditure Non-productive expenditure Other expenditure Budget surplus Trade PROP1564 PROP65

ln(GDP p.c.) Investment Labour force growth Net lending Distortionary taxation Non-distortionary taxation Other revenue Productive expenditure Non-productive expenditure Other expenditure Budget surplus Trade PROP1564 PROP65

Table 11.7 Correlation

1.0000

−0.1992 0.0819 PROP65

−0.0480

0.4573

−0.0331

0.1798

1.0000 0.2161

Other revenue

196 11 Tax Composition and Economic Growth in the Age of Demographic Change

References

197

References Acemoglu D, Restrepo P (2017) Secular stagnation? The effect of aging on economic growth in the age of automation. Am Econ Rev 107(5):174–79 Acosta-Ormaechea S, Sola S, Yoo J (2019) Tax composition and growth: a broad cross-country perspective. German Econ Rev 20(4):e70–e106 Angelopoulos K, Economides G, Kammas P (2007) Tax-spending policies and economic growth: theoretical predictions and evidence from the OECD. Eur J Polit Econ 23(4):885–902 Arellano M, Bond S (1991) Some tests of specification for panel data: monte Carlo evidence and an application to employment equations. Rev Econ Stud 58(2):277–297 Arnold JM (2008) Do tax structures affect aggregate economic growth?: empirical evidence from a panel of OECD countries. OECD Working Papers, OECD Arnold JM, Brys B, Heady C, Johansson A, Schwellnus C, Vartia L (2011) Tax policy for economic recovery and growth. Econ J 121(550):F59–F80 Baiardi D, Profeta P, Puglisi R, Scabrosetti S (2019) Tax policy and economic growth: does it really matter? Int Tax Publ Finance 26(2):282–316 Barro RJ (1990) Government spending in a simple model of endogeneous growth. J Polit Econ 98(5, Part 2):S103–S125 Barro RJ, Sala-i-Martin X (1992) Public finance in models of economic growth. Rev Econom Stud 59(4):645–661 Baum CF (2006) An introduction to modern econometrics using Stata. Stata Press Blanco F, Delgado FJ (2019) Taxation and economic growth in the European Union: a quantile approach. Revista de economía mundial 51:79–100 Bleaney M, Gemmell N, Kneller R (2001) Testing the endogenous growth model: public expenditure, taxation, and growth over the long run. Can J Econ 34(1):36–57 Blundell R, Bond S (1998) Initial conditions and moment restrictions in dynamic panel data models. J Econ 87(1):115–143 Brambor T, Clark WR, Golder M (2006) Understanding interaction models: improving empirical analyses. Polit Anal 14(1):63–82 Brys B, Perret S, Thomas A, O’Reilly P (2016) Tax design for inclusive economic growth. OECD Working Papers, OECD Devarajan S, Swaroop V, Zou HF (1996) The composition of public expenditure and economic growth. J Monetary Econ 37(2):313–344 Easterly W, Rebelo S (1993) Fiscal policy and economic growth. J Monetary Econ 32(3):417–458 Gemmell N, Kneller R, Sanz I (2011) The timing and persistence of fiscal policy impacts on growth: evidence from OECD countries. Econ J 121(550):F33–F58 Grossman GM, Helpman E (1990) Trade, innovation, and growth. Am Econ Rev 80(2):86–91 Johansson A, Heady C, Arnold JM, Brys B, Vartia L (2008) Taxation and economic growth. OECD Working Papers, OECD Jones LE, Manuelli RE, Rossi PE (1993) Optimal taxation in models of endogenous growth. J Polit Econ 101(3):485–517 Khattry B, Rao JM (2002) Fiscal faux pas?: an analysis of the revenue implications of trade liberalization. World Dev 30(8):1431–1444 King RG, Rebelo S (1990) Public policy and economic growth: developing neoclassical implications. J Polit Econ 98(5, Part 2):S126–S150 Kneller R, Bleaney MF, Gemmell N (1999) Fiscal policy and growth: evidence from OECD countries. J Public Econ 74(2):171–190 Lee Y, Gordon RH (2005) Tax structure and economic growth. J Public Econ 89(5–6):1027–1043 Lucas RE Jr (1990) Supply-side economics: an analytical review. Oxford Econ Papers 42(2):293– 316 Luo W (2018) Essays on inequality and fiscal policy. PhD thesis, University of York Luo W (2019) Demography and the composition of taxes: evidence from international panel data. Econ Lett 183:108518

198

11 Tax Composition and Economic Growth in the Age of Demographic Change

Luo W (2020) Demography and economic growth: the effect of tax composition. Appl Econ Lett 27(20):1629–1634 Luo W, Zhu J (2021) Youthful dependents and economic growth: the effect of tax composition. Appl Econ Lett 28(8):675–680 Mendoza EG, Milesi-Ferretti GM, Asea P (1997) On the ineffectiveness of tax policy in altering long-run growth: harberger’s superneutrality conjecture. J Publ Econ 66(1):99–126 Ojede A, Yamarik S (2012) Tax policy and state economic growth: the long-run and short-run of it. Econ Lett 116(2):161–165 Padovano F, Galli E (2001) Tax rates and economic growth in the OECD countries. Econ Inquiry 39(1):44–57 Park H, Philippopoulos A (2003) On the dynamics of growth and fiscal policy with redistributive transfers. J Publ Econom 87(3):515–538 Peretto PF (2003) Fiscal policy and long-run growth in R&D-based models with endogenous market structure. J Econ Growth 8(3):325–347 Peretto PF (2007) Corporate taxes, growth and welfare in a Schumpeterian economy. J Econ Theory 137(1):353–382 Persson T, Tabellini GE (2003) The economic effects of constitutions. MIT press Razin A, Sadka E, Swagel P (2002) The aging population and the size of the welfare state. J Polit Econ 110(4):900–918 Reinhart CM, Rogoff KS (2011) From financial crash to debt crisis. Am Econ Rev 101(5):1676– 1706 Rodrik D (1998) Why do more open economies have bigger governments? J Polit Econ 106(5):997– 1032 Romero-Avila D, Strauch R (2008) Public finances and long-term growth in Europe: evidence from a panel data analysis. Eur J Polit Econ 24(1):172–191 Roodman D (2009) How to do xtabond2: an introduction to difference and system GMM in Stata. Stata J 9(1):86–136 Solow RM (1956) A contribution to the theory of economic growth. Q J Econ 70(1):65–94 Stoilova D (2017) Tax structure and economic growth: evidence from the European Union. Contaduría y Administración 62(3):1041–1057 Stokey NL, Rebelo S (1995) Growth effects of flat-rate taxes. J Polit Econ 103(3):519–550 Swan TW (1956) Economic growth and capital accumulation. Econ Record 32(2):334–361 Tosun MS, Abizadeh S (2005) Economic growth and tax components: an analysis of tax changes in OECD. Appl Econ 37(19):2251–2263 Widmalm F (2001) Tax structure and growth: are some taxes better than others? Publ Choice 107(3–4):199–219 Wooldridge JM (2013) Introductory econometrics: a modern approach. Nelson Education Xing J (2012) Tax structure and growth: how robust is the empirical evidence? Econ Lett 117(1):379– 382 Yanikkaya H, Turan T (2020) Tax structure and economic growth: do differences in income level and government effectiveness matter? Singap Econ Rev 65(01):217–237

Part IV

Re-thinking of the Malthusian Trap

Part IV of the text covers the analysis of the relationship between demography and income inequality. It builds upon the Malthusian trap which occurs when population growth outpaces agricultural production, causing famine or war, resulting in poverty and depopulation.

Chapter 12

Demography and Income Inequality

12.1 Introduction How does population aging affect income inequality? Related studies can be traced back to Paglin (1975), who proposes a decomposition method to analyze the impact of demographic change on inequality. Both Lindert (1978) and Repetto (1978) then show that high proportion of the older population is associated with a high aggregate income inequality of an economy. Empirical evidence generally has not supported their hypothesis. For example, Fig. 12.1 depicts the raw correlation between income inequality over 1990–2014 and the ratio of the population above 65 to the population between 15 and 64, indicating no positive association between aging and the degree of inequality. On the other hand, Morley (1981) extends the method by Paglin (1975) and find that a high proportion of the young population intensifies income inequality and therefore, countries with demographic change in the older age group have a lower level of inequality. Through a cohort analysis method, subsequent studies focusing on the within-economy relationship, such as the United States, the United Kingdom, and Taiwan (Deaton and Paxson 1994), Japan (Ohtake and Saito 1998), Sri Lanka (Karunaratne 2000), South Korea (Jin 2009), and China (Dong et al. 2018), provide mixed results. Therefore, additional empirical studies, in particular cross-country analysis, are needed.1 Luo (2019) demonstrates that population aging raises the demand for expenditure rather than income taxes so as to increase the tax burden on the retired population when the median voter is of working age. This paper develops the Luo (2019) hypothesis to consider how population aging affects income inequality, in particular via the the effect of tax composition. The main theoretical prediction is that inequality increases with population aging as the extent of taxes on income relative to taxes on expenditure declines. The logic is similar to Luo (2019). Income taxes are paid solely by workers, whilst expenditure taxes are paid by both generations. If the 1

Although Luo et al. (2018) confirm this positive association, their results indicate that this relationship holds more firmly in Asia and Latin America regions. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 W. Luo, Inequality, Demography and Fiscal Policy, Applied Economics and Policy Studies, https://doi.org/10.1007/978-981-99-0518-8_12

201

202

12 Demography and Income Inequality

60

Income inequality Versus Population Aging Angola

30

40

UTIP

50

Cameroon Kuwait Malawi Guatemala Qatar Armenia Peru Gabon Burundi Tunisia Cambodia Oman Mozambique Trinidad and Tobago Lesotho Central Republic India The African Bahamas Tanzania Pakistan Bolivia Zambia Morocco Cote d Jamaica Azerbaijan Mongolia Bangladesh Swaziland Nepal Egypt Nigeria Ghana Senegal Brazil Yemen Uganda Georgia Botswana Turkey Puerto Rico El Salvador Panama Sudan Syrian Arab Republic Philippines Ecuador Chile Eritrea Indonesia Jordan Iraq Honduras Tonga Belize Kenya Zimbabwe Argentina Ethiopia Colombia Suriname Kazakhstan Myanmar Venezuela South Africa Barbados Mexico Albania Uruguay Madagascar Sri Lanka Paraguay Fiji Iran Kyrgyz Republic Greece Israel Thailand Moldova Costa Rica Macedonia Japan Russia Ukraine Hong Kong SAR China Algeria New Zealand Spain Malaysia China Hungary Lithuania United States Portugal Cyprus Canada Bulgaria Belgium Croatia Korea Latvia Mauritius Poland Afghanistan France Italy Australia Bosnia and Herzegovina United Kingdom Slovak Singapore Ireland Netherlands Romania Austria MaltaRepublic Norway Luxembourg Estonia Germany Iceland SloveniaFinland Switzerland Denmark Macao SAR China Sweden Czech Republic

0

10

20

30

Ratio of old to young (mean) texas

Fitted values

Fig. 12.1 Correlation between income inequality and aging (cross-country data over 1990–2014). Aging is defined as the ratio of the population above 65 years old to the population between 15 and 64

median voter is a worker, then increasing the size of the retired population compels a shift in tax composition towards expenditure taxes, and thus poor households are disproportionately damaged, leading to higher inequality. International panel evidence supports this hypothesis. The argument proposed in this paper is also motivated by recent rises in taxes on goods and services. Revenue sources outside of income taxes are thus empirically becoming increasingly important components of total revenue. A series of recent literature generally focuses on the tax structure instead of income taxes only, given a declining status of tax revenue coming from income. For instance, Pickering and Rajput (2018) analyze how income inequality affects the extent to which taxes are levied on income relative to expenditure. Luo (2019) then extends their framework to explore the relationship between demography and the composition of taxes. Luo (2020) and Luo and Zhu (2021) further consider the impact of demographic structure on economic growth through the mechanism of tax composition. In an overlapping generations model, taxes are levied on both income and expenditure, transferring to both retirees and workers, with a balanced budget period by period (Luo 2018). In a median voter framework, the observed tax rate is the median voter’s ideal income-to-expenditure tax ratio. Countries experiencing more rapid aging have delivered more inequality, as the median voter wants to improve the tax burden on retirees and prefers a lower tax ratio, which disproportionately raises the tax burden on lower-income households given their wide distribution of expenditure.

12.2 Data and Econometric Specification

203

12.2 Data and Econometric Specification This empirical analysis concentrates on a panel of international data over the period 1990–2014. The dependent variable is a measure of income inequality, obtained from the University of Texas Inequality Project’s Estimated Household Income Inequality data of Galbraith and Kum (2005). These data (denoted by UTIP) use Theil’s T statistic—measured across sectors within each country—to estimate wage inequality. Assuming competitive labor markets, then wage inequality should be capturing underlying heterogeneity in productivity. This measure is thus close to Meltzer and Richard (1981) original conception of the driver of the demand for redistribution— productivity-based inequality. As for the explanatory variable, the measure of the proportion of retirees in the population used is the percentage of the population over the age of 65 (PROP65) as in Luo (2019). This measure of the retired fraction is preferable to the dependency ratio used in the existing literature. The dependency ratio includes retirees as well as children which would have different influences on inequality. One important determinant of income inequality is the level of development. Thus, I include the natural logarithm of GDP per capita in constant chained PPP US$ (ln(y)) as a first control in the regression analysis. To fully capture the effect of international trade on inequality, the econometric analysis includes the trade share (TRADE) as an additional control. In addition, the natural logarithm of the total population size (ln(POP)) is also included, thus capturing to some extent fertility effects. Beyond that, this paper also controls the degree of democracy for purpose of robustness check. As proposed by Luo (2019), in democracies with low quality institutions the link between the median voter and policy is blurred, whilst in countries with stronger institutions the median’s influence is stronger. If the mechanism put forward in this paper is important, then inequality will be affected by policy variables. Therefore, the democracy score provided by the Polity IV project is employed as a final control (POLITY 2). Table 12.1 contains descriptive statistics of the variables used in the regression analysis. Since this paper implies that population aging affects inequality through the impact of tax composition, both income tax revenue (taxes on income, profits and capital gains as a share of total tax revenue (τ y )) and expenditure tax revenue (taxes on goods and services as a share of total tax revenue (τc )) are included to examine the distributive effects of taxation. For robustness, I also utilize the income-to-expenditure tax ratio instead of two separate policy variables. The argument proposed predicts that income inequality declines with a rise in the extent of taxes on income relative to expenditure.

204

12 Demography and Income Inequality

Table 12.1 Descriptive statistics Obs Mean UTIP PROP65 RATIO ln(y) TRADE ln(POP) POLITY 2 τy τc

1556 4633 4633 3652 4263 5302 3790 2142 2141

42.788 7.006 11.018 8.577 86.936 15.073 3.037 22.446 29.056

Std. dev.

Min

Max

6.708 4.715 6.655 1.294 51.957 2.351 6.685 12.746 13.760

22.752 0.335 0.390 5.031 0.309 9.105 −10 0.349 0.024

59.957 25.078 40.532 11.734 531.737 21.038 10 75.238 89.224

Notes The table gives descriptive statistics for the variables. UTIP is the University of Texas Inequality Project’s Estimated Household Income Inequality. PROP65 is the proportion of the population aged 65 and above—taken from the World Development Indicators (WDI) database. RATIO = P R O P65 P R O P1564 , in which PROP1564 is the proportion of the population aged between 15 and 64. Income y is real GDP at chained PPPs in millions of 2005 US dollars per capita—taken from the Penn World Tables. TRADE is the sum of exports and imports as a percentage of GDP—taken from the WDI. POP is the size of country population—taken from the WDI. POLITY 2 is a measure of democracy provided by the Polity IV project, with −10 denoting the highest level of autocracy, and 10 denoting the highest level of democracy. τ y denotes taxes on income, profits and capital gains as a percentage of revenue, and τc denotes taxes on goods and services as a percentage of revenue—both taken from the WDI

12.3 Empirical Results 12.3.1 Baseline Estimation This section is to test whether and how income inequality across different countries systematically changes with the fraction of the population that is retired, in the presence of country fixed effects. Column 1 of Table 12.2 is a simple specification with just the fraction of the retired population (PROP65) and GDP per capita as regressors using annual data OLS regression, with robust standard errors clustered by country. Column 2 instead uses the ratio of the population above 65 against those between the ages of 15 and 64, RATIO, to measure population aging, and mimic column 1. In these specifications the sign of the coefficient estimate relating to the fraction of the retired population is negative in all cases, and all are statistically significant at the 1% level. This is consistent with the argument—an increase in the retired fraction intensifies inequality. Columns 3 and 4 repeat the analysis of columns 1 and 2 using full controls instead. The results using panel estimation with full control variables support those already found. The estimated statistical significance of the fraction of the retired population is unaffected and even remains at the 1% level. Using the estimate from column 4 of Table 12.2, a one standard deviation increase in the fraction of the retired population is statistically associated with a rise of 4.02 in the measure of inequality, holding all else equal. The magnitude of this estimated

12.3 Empirical Results Table 12.2 Basic estimation results (1) PROP65

(2)

1.275*** (0.232)

RATIO ln(y)

205

(4)

0.991*** (0.244)

−1.781** (0.785)

0.784*** (0.158) −1.085 (0.789)

1493 123 Panel Yes 0.188

1493 123 Panel Yes 0.156

TRADE ln(POP) POLITY 2 Observations Countries Data Fixed effects? R2

(3)

−2.383*** (0.805) 0.00920 (0.00738) 4.414** (1.860) 0.0418 (0.0605) 1395 114 Panel Yes 0.161

0.604*** (0.152) −1.953** (0.812) 0.0120* (0.00717) 4.717** (1.910) 0.0533 (0.0616) 1395 114 Panel Yes 0.145

Notes Table 12.2 contains results using OLS regressions of income inequality, UTIP. Columns (2) and (4) instead use the ratio of the population above 65 to those between the ages of 15 and 64, RATIO, to measure population aging, and mimic columns (1) and (3). Robust standard errors are shown in parentheses. Standard errors are clustered by country. *, **, and *** respectively denote significant levels at 10%, 5% and 1%

correlation is sizable—implying more than a half of the raw standard deviation in the inequality measure. In Table 12.2, the results relating to the control variables are also of some interest. There is a negative relationship with income per capita, which likely indicates greater potential to reduce inequality in richer countries. In addition, more populous countries are found to suffer from more unequal distribution, which reflects the Malthusian trap.

12.3.2 Further Estimation It is natural to investigate whether or not the reported results change with the degree of democracy, given that the argument invokes on the median voter framework. Columns 1 and 2 of Table 12.3 thus extend the regression results by splitting the sample by levels of democracy (depending on the median value of democracy score (POLITY 2 = 7)). Column 1 (column 2) contains results for countries with stronger (weaker) democratic credentials. The positive relationship between the fraction of the retired population and income inequality holds in both columns. The use of an interaction term provides an alternative way to examine how the results change with the extent of the franchise. In column 3, I make use of a democracy indicator variable (DEMOCRACY ), defined as 0 or 1 depending on whether POLITY 2 > 7.

−1.073 (1.085) 0.00884 (0.0101) 2.849 (2.386) 0.370 (0.438) 886 Panel Yes Higher POLITY 2 0.115

0.581** (0.238)

−3.559*** (0.953) −0.0111 (0.0131) 8.495*** (1.762) −0.0301 (0.0744) 509 Panel Yes Lower POLITY 2 0.207

1.668** (0.685)

(2)

−2.410*** (0.793) 0.00889 (0.00751) 4.977*** (1.788) −0.0200 (0.0618) 1395 Panel Yes Full 0.175

0.774*** (0.255) 0.166* (0.0909)

(3)

−2.500*** (0.822) 0.0198*** (0.00748) 2.298 (2.317) 0.260* (0.133) 911 Panel Yes Higher income 0.273

0.909*** (0.221)

(4)

−1.789 (1.425) −0.0177 (0.0116) 7.369*** (2.186) −0.0460 (0.0609) 484 Panel Yes Lower income 0.076

0.850 (0.791)

(5)

−1.247 (1.553) 0.00628 (0.00910) 6.430 (4.443) 0.237* (0.131) 407 Panel Yes OECD countries 0.240

0.480* (0.238)

(6)

Notes Regression specification is the same as column (3) of Table 12.2. Columns (1) and (2) respectively correspond to higher and lower democracy levels. Column (3) includes an interaction term described in the text. Columns (4) and (5) respectively correspond to higher and lower levels of income. Column (6) only includes countries with OECD membership

Observations Data Fixed effects? Sample R2

POLITY 2

ln(POP)

TRADE

PROP65 × DEMOCRACY ln(y)

PROP65

Table 12.3 Further estimation results (1)

206 12 Demography and Income Inequality

12.3 Empirical Results Table 12.4 Mechanism check (1) τ DEP. VAR. = ln( τcy ) PROP65

−0.0899** (0.0423)

207

(2) τy

(3) τc

−0.500 (0.475)

1.258* (0.663)

τ

(4) UTIP

−0.689 (0.528)

ln( τcy ) τy τc ln(y)

0.584*** (0.218) Observations 1849 Countries 146 Data Panel Fixed effects? Yes R2 0.070

(5) UTIP

5.256** (2.327) 1899 147 Panel Yes 0.039

−2.766 (2.852) 1885 147 Panel Yes 0.021

1.760** (0.793) 837 93 Panel Yes 0.045

−0.0165 (0.0276) 0.0653** (0.0313) 1.619** (0.723) 837 93 Panel Yes 0.064

Notes The starting point—column (1)—is a replication of column (2) of Table 12.2 in Luo (2019). Robust standard errors are shown in parentheses. Standard errors are clustered by country. *, **, and *** respectively denote significant levels at 10%, 5% and 1%

This indicator variable is then multiplied by the aging measure, thereby generating an interaction term. The hypothesis here is that the relationship between the tax composition measure and inequality will be increasingly positive under democracies, hence that the coefficient estimate for the interaction term is positive. The estimation results confirm this, and statistically significant at the 10% level. While there is a rise in the number of the retired population who prefers a higher ratio, leading to a higher level of inequality, the preference of the median voter shifts the other way, as s/he wants to shift the tax-burden onto the retired population, rather than being taxed labor income only from the young. Moreover, the median voter more plausibly drives policy and hence affecting inequality under stronger democracy, consistent with the empirical results. In contrast, policy and inequality are less likely to respond to changes in the preference of the median voter the less democratic is the country. It is of interest to see whether the results vary with the level of development. Columns 4 and 5 of Table 12.3 split the sample by levels of GDP per capita (determined by the average value of GDP per capita). As can be seen in all cases, income inequality is positively correlated with the fraction of the retired population. However, this positive relationship only holds to a significant degree in the group of countries with higher income level. Rich countries have a proportionally larger retired population, and therefore the distribution of income becomes more unequal. Similarly, when the sample is restricted to countries with OECD membership (in column 6), the results are statistically significant.

208

12 Demography and Income Inequality

12.3.3 Mechanism Note that it is also natural to ask how the results change with income and expenditure taxes, under the premise that income inequality rises as income taxes decline whilst expenditure taxes rise in the argument proposed in this paper. Therefore, Table 12.4 examines the effect of aging on the income-to-expenditure tax ratio (in column 1), income taxes (in column 2), and expenditure taxes (in column 3). The results show that demographic changes increase the demand for expenditure rather than income taxes. Columns 4 and 5 then respectively use the tax ratio and two separate policy variables to investigate the distributive effects in the econometric analysis. If the story of the tax composition effect could explain, then income taxes will have a negative impact on inequality whilst expenditure taxes do not reduce inequality. The sign of coefficients on policy variables confirms this story, although the significance level on that of income taxes falls.

12.4 Conclusion This paper analyzes how population aging affects the distribution of income. Income inequality is found to be positively associated with the fraction of the retired population. The empirical results hold across various econometric specifications employed. In particular, the results hold significantly in countries with strong democratic credentials.

References Deaton A, Paxson C (1994) Intertemporal choice and inequality. J Polit Econ 102(3):437–467 Dong Z, Tang C, Wei X (2018) Does population aging intensify income inequality? Evidence from China. J Asia Pac Econ 23(1):66–77 Galbraith JK, Kum H (2005) Estimating the inequality of household incomes: a statistical approach to the creation of a dense and consistent global data set. Rev Income Wealth 51(1):115–143 Jin KS (2009) Aging and inequality of income and consumption in Korea. J Int Econ Stud 53(23):59– 72 Karunaratne HD (2000) Age as a factor determining income inequality in Sri Lanka. Developing Economies 38(2):211–242 Lindert PH (1978) Fertility and scarcity in America. Princeton University Press, Princeton, NJ Luo W (2018) Essays on inequality and fiscal policy. PhD thesis, University of York Luo W (2019) Demography and the composition of taxes: evidence from international panel data. Econ Lett 183:108518 Luo W (2020) Demography and economic growth: the effect of tax composition. Appl Econ Lett 27(20):1629–1634 Luo W, Zhu J (2021) Youthful dependents and economic growth: the effect of tax composition. Appl Econ Lett 28(8):675–680

References

209

Luo Z, Wan G, Wang C, Zhang X (2018) Aging and inequality: the link and transmission mechanisms. Rev Dev Econ 22(3):885–903 Meltzer AH, Richard SF (1981) A rational theory of the size of government. J Polit Econ 89(5):914– 927 Morley SA (1981) The effect of changes in the population on several measures of income distribution. Am Econ Rev 71(3):285–294 Ohtake F, Saito M (1998) Population aging and consumption inequality in Japan. Rev Income Wealth 44(3):361–381 Paglin M (1975) The measurement and trend of inequality: a basic revision. Am Econ Rev 65(4):598–609 Pickering A, Rajput S (2018) Inequality and the composition of taxes. Int Tax Public Finan 25(4):1001–1028 Repetto R (1978) The interaction of fertility and the size distribution of income. J Dev Stud 14(4):22– 39

Appendix A

Labor Income Inequality, Taxation and Growth: A Political Economy Theory

A.1 Economic Environment Different individuals have different incomes. The budget constraints, common to all individuals, are i = (1 − τt )yti + rt (A.1) cti + kt+1 i i = γ kt+1 dt+1

(A.2)

where y i is the individual’s labor income when young, and is taxed at a linear rate τ , k i is the individual accumulation of asset, r is lump-sum redistribution, and γ is the exogenous rate of return on asset. Individuals make decision between consumption and investment when young, financed by disposable labor income and lump-sum redistribution, and benefit from the return on that investment when old. The labor income when young is defined as yti = n i ei kt

(A.3)

where ei is productivity, and the stock of k accumulated on average by the previous generation has a positive externality on the income of the newborn generation as in Persson and Tabellini (1994). Note that the stock of aggregate capital is accumulated as average productivity of all individuals increases. With homothetic preferences, the ratio of consumption in the two periods is independent of wealth and labor income di = D. Equivalently, every individual has the same “saving rate”. taxation, ct+1 i t Each individual chooses labor supply so as to maximize vti = U



  γ  γD  (1 − τt )n i ei kt + rt , 1 − n i , (1 − τt )n i ei kt + rt . γ+D γ+D (A.4)

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 W. Luo, Inequality, Demography and Fiscal Policy, Applied Economics and Policy Studies, https://doi.org/10.1007/978-981-99-0518-8

211

212

Appendix A: Labor Income Inequality, Taxation …

The first-order condition is γD γ (1 − τt )ei kt Uc − Ul + (1 − τt )ei kt Ud = 0 γ+D γ+D

(A.5)

which determines the labor supply, n[(1 − τ )ei , r ], for those who wish to work. Note that k is given due to accumulation by the previous generation. The choice depends only on the size of redistribution, r , and after-tax wage, (1 − τ )ei . Let F denote the distribution function for individual productivity, so that F(ei ) is the fraction of the population with productivity less than ei . Average labor income is obtained by integrating ∞ y¯t = kt

  ei n (1 − τt )ei , rt d F(ei ).

(A.6)

0

Finally, the budget of government is balanced and all government spending is for redistribution of income. If per capita income is y¯ , then τt y¯t = rt .

(A.7)

For the average individual, kt+1 = y¯t − ct . By use of (A.2) and (A.6) I can therefore solve for the growth rate of k D kt+1 − kt = gt = kt

∞ 0

ei n[(1 − τt )ei , rt ]d F(ei ) − 1. γ+D

(A.8)

Since leisure is a normal good, I have ∂n i = ∂rt γ D 2 )2 (1 − τt )ei kt Ucc + ( γγ+D ) (1 − τt )ei kt Udd − ( γ +D

γ γ +D Ucl

D i + 2 (γγ+D) 2 (1 − τt )e k t Ucd − 2

γD γ +D Udl

(A.9)

− < 0,

with

∂2v ∂n 2 i

≡=

2 2  D γ γ (1 − τt )ei kt Ucc + Ull + γγ+D (1 − τt )ei kt Udd − 2 γ +D γ +D 2 γ D 2( γ +D (1 − τt )ei kt DUcd − 2 γγ+D (1 − τt )ei kt Udl < 0, given the 

(1 − τt )e kt Ucl + assumption that v is strictly concave. Hence for given productivity endowment, individual labor supply falls with increased redistribution. Therefore ∂ y¯t = kt ∂rt

∞ ei 0

∂n i d F(ei ) < 0. ∂rt

(A.10)

Appendix A: Labor Income Inequality, Taxation …

213

This establishes that the left-hand side of (A.7) is strictly decreasing with r . Moreover, τ y¯ is non-negative and bounded above by τ e, where e is average productivity. In turn, the right-hand side of (A.7) is strictly increasing with r . Thus, there is a unique value of r to satisfy (A.7) for any τ .

A.2 Political-Economic Equilibrium In order to characterize the political economic equilibrium, the median voter m sets taxes to maximize utility subject to the budget constraints (A.1) and (A.2), and the government budget constraint (A.7): vtm = U



  γ  γD  (1 − τt )n m em kt + τt y¯t , 1 − n m , (1 − τt )n m em kt + τt y¯t , γ+D γ+D

(A.11)

and the first-order condition for the median voter with respect to the tax rate is 

 d y¯t  γ γD Uc + Ud dτt γ + D γ+D (A.12)  dn m  γ γD (1 − τt )em kt Uc − Ul + (1 − τt )em kt Ud + = 0. γ+D γ+D dτt

y¯t − ytm + τt

Thus, making use of Eq. (A.5), the tax rate chosen by the median voter must satisfy d y¯t y¯t − ytm + τt = 0. (A.13) dτt For a given labor income inequality, the political equilibrium τ is constant over time, so that the time subscript t is suppressed henceforth. Let θ = 1 − τ be the fraction of earned income retained. From (A.6), y¯ depends on r and θ . The total derivative of average labor income, I have ∂ y¯ dr ∂ y¯ dθ d y¯ = + , dτ ∂r dτ ∂θ dτ   ∂ y¯ d y¯ ∂ y¯ = y¯ + τ − . ∂r dτ ∂θ

(A.14)

Thus, the total derivative of average labor income with respect to changes in the tax rate is given by y¯r y¯ − y¯θ d y¯ = < 0, (A.15) dτ 1 − τ y¯r with y¯r =

∂ y¯ ∂r

and y¯θ =

∂ y¯ . ∂θ

Finally, substituting (A.15) into (A.13) I have

214

Appendix A: Labor Income Inequality, Taxation …

y¯r y¯ − y¯θ , 1 − τ y¯r ηr y¯ (1 − τ ) − ηθ y¯ τ , = ( y¯ − y m )(1 − τ ) + 1 − ηr

0 = y¯ − y m + τ

(A.16)

where ηr = y¯r ry¯ and ηθ = y¯θ θy¯ are the partial elasticities of average income. Solving the above equation for τ , yields τ=

m − 1 + ηr m − 1 + ηr + mηθ

(A.17)

with m = yy¯m . Identical to the spirit of Meltzer and Richard (1981), Eq. (A.17) yields that an increase in labor income inequality raises taxation dτ > 0. dm

(A.18)

A.3 Labor Income Inequality and Growth For the average individual in (A.1) and (A.2), I have kt+1 = y¯t − ct , dt+1 , = y¯t − D γ kt+1 . = y¯t − D

(A.19)

Solving the above equation for kt+1 , yields kt+1 =

D y¯t . γ+D

(A.20)

Combining the above equation and (A.6), the growth rate of k can be obtained kt+1 − kt , k  ∞t i D 0 e n[(1 − τt )ei , rt ]d F(ei ) = − 1. γ+D

gt =

(A.21)

Again for a given labor income inequality, the political equilibrium τ and g are constant over time, so that the time subscript t is suppressed henceforth. Thus, the effect of taxation on growth, Combining (A.8) and making use of the total derivative of y¯ (A.15), yields

Appendix A: Labor Income Inequality, Taxation …

D d dg = dτ γ+D



∞ i 0 e n[(1

D k1 d y¯ < 0. = γ + D dτ

215

− τ )ei , r ]d F(ei ) dτ

 , (A.22)

Thus all else equal, the higher is the labor income taxation, the lower is the growth rate. Combining (A.18), therefore, the effect of labor income inequality on growth yields dg dτ dg = < 0. (A.23) dm dτ dm If labor income inequality increases such that divergence between mean and median labor income increases, then the preferred labor income tax rate (or redistribution) rises, and hence less growth because redistributive policies are coming from distortionary taxes that affect capital accumulation. This indicates that labor income inequality is harmful for growth which is identical in spirit to Persson and Tabellini (1994). References Meltzer AH, Richard SF (1981) A rational theory of the size of government. J Polit Econ 89(5):914–927 Persson T, Tabellini G (1994) Is inequality harmful for growth? Am Econ Rev 84(3):600–621

Appendix B

Capital Income Inequality, Taxation and Growth: A Political Economy Theory

This model revisits Persson and Tabellini (1994) to include labor income taxation instead of wealth taxation. I study an overlapping generations model with constant population, where individuals live for two periods. Individuals born in period t, indexed by i, have preferences defined over consumption when young ci , leisure when young l i , and consumption when old d i , represented by a strictly concave, continuous, i ). Consumption and leisure twice-differentiable utility function vti = U (cti , lti , dt+1 are both normal goods. Following the original, I first analyze the equilibrium behavior conditional on a given tax policy and then address the tax policy choice itself.

B.1 Economic Environment Income may be derived from both labor and capital, and the stock of asset, k, accumulated on average by the previous generation has a positive externality on the income of the newborn generation as in Persson and Tabellini (1994). All individuals possess a unit of time to allocate to labor n i , or leisure l i = 1 − n i . Individual labor income yti = n i ei kt depends on productivity, ei , as well as hours worked, and is taxed at a linear rate τ . Capital income varies exogenously across individuals and is denoted by R i kt . Following Meltzer and Richard (1981), consumption is also financed by lump-sum redistribution, r , common to all individuals, hence the budget constraints are: i = (1 − τt )n i ei kt + rt + R i kt (B.1) cti + kt+1 i i = γ kt+1 dt+1

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 W. Luo, Inequality, Demography and Fiscal Policy, Applied Economics and Policy Studies, https://doi.org/10.1007/978-981-99-0518-8

(B.2)

217

218

Appendix B: Capital Income Inequality, Taxation …

where k i is the individual accumulation of asset, and γ is the exogenous rate of return on asset.1 Individuals make decision between consumption and investment when young, financed by labor and capital income as well as lump-sum transfers, and benefit from the return on that investment when old. Note that the stock of aggregate capital is accumulated as average productivity of all individuals increases. With homothetic preferences, the ratio of consumption in the two periods is independent di = D. Equivalently, every individual has of wealth and labor income taxation, ct+1 i t the same “saving rate”. To clarify the argument, capital income is assumed to be untaxed. In practice it is often more difficult to raise taxes on capital than on labor. Capital is often highly mobile internationally, whilst labor is not, and given this Diamond and Mirrlees (1971) show that small open economies should not tax capital income. Indeed, international tax competition limits the democratic control over capital income taxation. Whilst in practice capital income taxation rates are positive, Gordon et al. (2004) observe lower average rates than for labor income in most countries. Moreover, the academic literature documents considerable difficulties with the collection of capital income taxation, primarily due to different types of capital income being taxed differentially (thereby, enabling arbitrage opportunities), and the fact that interest payments are tax-deductible. Indeed Gordon and Slemrod (1988), using US tax return data from 1983, estimated that the tax revenue loss from eliminating capital income taxation completely would be zero, hence that the tax burden on capital was effectively non-existent. It is an open question quite why the median voter would tolerate such a state of affairs, but conceivably the perceived deadweight and/or capital flight losses from increasing capital income taxation to some extent nullifies it as an instrument. Thus I focus on the choice of the labor income tax.2 Each individual chooses labor supply so as to maximize: γ  (1 − τt )n i ei kt + rt + R i kt , vti = U [ γ +D D  1 − n i , γγ+D (1 − τt )n i ei kt + rt + R i kt ].

(B.3)

The first-order condition is: γD γ (1 − τt )ei kt Uc − Ul + (1 − τt )ei kt Ud = 0 γ+D γ+D

(B.4)

which determines the labor supply, n[(1 − τt )ei , rt , R i ], for those who wish to work.3 Since leisure is a normal good, I have that 1

Throughout the paper I use superscripts to denote individual-specific variables and no superscripts to denote average variables. 2 The results would all still stand if I instead modeled capital income taxation as fixed (and unresponsive to inequality), as observed from OECD data. 3 Note again that k is given due to accumulation by the previous generation. Further, for simplicity t (but without loss of generality) I henceforth assume that the joint distribution of ei and R i is such that n i > 0 for all i, so that everyone supplies a strictly positive amount of market work.

Appendix B: Capital Income Inequality, Taxation …

∂n i =− ∂ Ri

∂ 2 vti ∂n i ∂ R i ∂ ∂vti ( ∂n i ∂n i

219

)

0,

a condition which imposes additional restrictions on Ucl and Udl . Hence, all else equal, people who are relatively capital-rich supply less labor and enjoy higher consumption. There are two sources of heterogeneity that determine differences in before-tax labor income. Firstly productivity, as analyzed by Meltzer and Richard (1981), and secondly capital income endowments. At the individual level increases in productivity will all else equal increase labor income.5 On the other hand increases in capital income will all else equal reduce the labor supply and, therefore, labor income. This underpins their proclivity towards taxation of labor income. Average labor income can thus be written by integrating: ∞ ∞ y¯t = kt

ei n[(1 − τt )ei , rt , R i ] f (ei , R i )dei d R i 0

4

(B.8)

0

In detail, using (B.4), I have that ∂n i = ∂ Ri

∂ 2 vti ∂n i ∂ R i ∂v i − ∂ i ( ti ) ∂n ∂n

γD γ γD γ γ2D (1 − τt )ei kt Ucd − γ +D Udl ( γ +D )2 (1 − τt )ei kt Ucc + ( γ +D )2 (1 − τt )ei kt Udd − γ +D Ucl + 2 (γ +D)2 < 0, = kt −

 γD 2 2 γ γ i i γ +D (1 − τt )e kt Ucc + Ull + γ +D (1 − τt )e kt Udd − 2 γ +D (1 − 2 γ γ D τt )ei kt Ucl + 2( γ +D (1 − τt )ei kt DUcd − 2 γ +D (1 − τt )ei kt Udl < 0. 5 Note that, as in Meltzer and Richard (1981), the sign of ∂n i is indeterminate, but for any individual ∂ei with

∂ ∂n i

∂v i

( ∂nti ) ≡  =



with positive labor income I have

∂ yti ∂n i = kt (n i + ei ) ∂ei ∂ei (B.7)   γ γ γD γD ei γ +D (1 − τt )kt Uc + γ +D (1 − τt )kt Ud + n i γ +D (1 − τt )ei kt Ucl + γ +D (1 − τt )ei kt Udl − Ull > 0, = kt −

must be positive given condition (B.6).

220

Appendix B: Capital Income Inequality, Taxation …

where f (ei , R i ) is joint distribution function of ei and R i . Individual productivity and capital endowments conceivably are correlated with each other to some extent: if, for example, high productivity individuals simultaneously enjoy high capital income. Finally, the government’s balanced budget requirement (in per capita terms) is given by: (B.9) τt y¯t = rt . For the average individual, by use of (B.2) and (B.8) I can thus solve for the growth rate of k ∞∞ D( 0 0 ei n[(1 − τt )ei , rt , R i ] f (ei , R i )dei d R i + R) kt+1 − kt −1 gt = = kt γ+D (B.10) where R is average capital income. Note that analogous to (B.5), I have: ∂n i =− ∂rt

∂ 2 vti ∂n i ∂rt ∂ ∂vti ( ∂n i ∂n i

)

0 the partial elasticities of average income (assumed constant, as in Meltzer and Richard 1981), and labor income inequality m = y¯ /y m . Proof The problem of the median voter m is to choose the tax rate so as to maximize vtm

γ  (1 − τt )n m em kt + τt y¯t + R m kt , = U [ γ +D D  1 − n m , γγ+D (1 − τt )n m em kt + τt y¯t + R m kt ],

and the first-order condition for the median voter with respect to the tax rate is

 γ γD y¯t − Uc + Ud γ+D γ+D  m

dn γ γD m m + = 0. (1 − τt )e kt Uc − Ul + (1 − τt )e kt Ud γ+D γ+D dτt ytm

d y¯t + τt dτt



Thus, making use of Eq. (B.4), the tax rate chosen by the median voter must satisfy y¯t − ytm + τt

d y¯t = 0. dτt

For a given labor income inequality, the political equilibrium τ is constant over time, so that the time subscript t is suppressed henceforth. Changes in the tax rate τ affect average income via two channels: its effect on the opportunity cost of leisure, and its

222

Appendix B: Capital Income Inequality, Taxation …

effect on transfers (from the government’s budget constraint r = τ y¯ ). In particular, I have that ∂ y¯ dr ∂ y¯ dθ d y¯ = + , dτ ∂r dτ ∂θ dτ ∂ y¯ d y¯ ∂ y¯ = ( y¯ + τ ) − ∂r dτ ∂θ with θ = 1 − τ . Thus, the total derivative of average labor income with respect to changes in the tax rate is given by y¯r y¯ − y¯θ d y¯ = < 0, dτ 1 − τ y¯r with y¯r = 0 I have

∂ y¯ ∂r

and y¯θ =

∂ y¯ . ∂θ

Finally, substituting

d y¯ dτ

=

y¯r y¯ − y¯θ 1−τ y¯r

into y¯t − ytm + τt ddτy¯tt =

y¯r y¯ − y¯θ , 1 − τ y¯r ηr y¯ (1 − τ ) − ηθ y¯ τ , = ( y¯ − y m )(1 − τ ) + 1 − ηr

0 = y¯ − y m + τ

where ηr = y¯r ry¯ and ηθ = y¯θ θy¯ are the partial elasticities of average income. Solving the above equation for τ , yields τ= with m =

y¯ . ym

m − 1 + ηr m − 1 + ηr + mηθ  

The key insight of Meltzer and Richard (1981) is that an increase in labor income inequality raises taxation, since an increase in income inequality raises m and from (B.14) I have that dτ > 0. (B.15) dm I am interested in the consequences of higher capital income inequality as in Luo et al. (2017). To study this issue I consider an increase in the capital income earned by the individuals in the set K of all individuals with capital income above Q99% .7 The effect of the increase in capital income going to the top capital-income recipients is to reduce the gap between taxable mean and median labor income. Hence an increase in overall income inequality can coexist with a reduction in labor income inequality. dτ > 0, it follows that an increase in capital income inequality unambiguously Since dm lowers the tax rate chosen.

7

I focus on the 99% percentile because in the empirical section that follows I use the income share of the top 1% as our measure of capital income inequality.

Appendix B: Capital Income Inequality, Taxation …

223

Lemma B.1 Suppose the top capital-income recipients are sufficiently productive that they also earn labor income above the median labor income, and consider an increase in capital-income inequality represented by an increase in the capital income earned by the top capital-income recipients. Then the labor income tax rate τ falls as capital income inequality rises. This indicates that government size diminishes with increased capital income inequality, identical to Proposition 1 in Luo et al. (2017). If inequality increases such that the share of capital income going to the top income recipients increases, then the preferred tax rate falls because the (capital) rich are supplying less taxable labor income and hence the capacity of the median voter to redistribute is reduced. The key issue is the extent to which the median voter can effectively redistribute through the tax system. As discussed above there are good reasons to believe that taxation of relatively mobile capital is considerably more difficult than taxation of labor income. If the rich are rich primarily due to capital income, perhaps because of the rising capital share, and perhaps due to successful reclassification of their income streams, then the capacity of the median voter to redistribute is curtailed. Moreover if rising inequality translates into further reductions in the supply of taxable labor then it follows that the demand for redistribution will fall.

B.3 Capital Income Inequality and Growth I now turn to the effect of capital income inequality on economic growth via the channel of redistribution. Combining (B.10) and the total derivative of y¯ , I have Lemma B.2. Lemma B.2 The growth rate falls as the labor income tax rate τ rises, e.g., dg D d( = dτ γ+D

∞∞ 0

0

ei n[(1 − τ )ei , r, R i ] f (ei , R i )dei d R i + R) < 0. dτ

Thus all else equal, the higher is the labor income taxation, the lower is the growth rate. Proof For the average individual in (B.1) and (B.2), I have kt+1 = yt + Rkt − ct , dt+1 , = yt + Rkt − D γ kt+1 = yt + Rkt − . D Solving the above equation for kt+1 , yields

224

Appendix B: Capital Income Inequality, Taxation …

kt+1 =

D(yt + Rkt ) . γ+D

Combining the above equation and (B.8), the growth rate of k can be obtained kt+1 − kt , k   t∞  ∞ i D 0 0 e n[(1 − τt )ei , rt , R i ] f (ei , R i )dei d R i + R = − 1. γ+D

gt =

Again for a given labor income inequality, the political equilibrium τ and g are constant over time, so that the time subscript t is suppressed henceforth. Thus, the r y¯ − y¯θ , yields effect of taxation on growth, making use of ddτy¯ = y¯1−τ y¯r D d dg = dτ γ+D =

∞∞ 0

0

ei n[(1 − τ )ei , r, R i ] f (ei , R i )dei d R i + R , dτ

D k1 dy < 0. γ + D dτ  

From the properties of the g and τ functions derived above, I can obtain Lemma B.3. Lemma B.3 A more unequal distribution of labor income decreases growth, e.g., dg dg dτ = < 0. dm dτ dm

(B.16)

This indicates that labor income inequality is harmful for growth which is identical in spirit to Persson and Tabellini (1994). Now consider the consequences of higher capital income inequality and the mechanism analyzed above. Proposition B.1 The growth rate rises as capital income inequality rises. In direct contrast to Persson and Tabellini (1994) economic growth increases with increased capital income inequality. When income differences are driven by capital income, the capacity of the median voter to redistribute through taxation is reduced since the capital-rich supply less (taxable) labor. Such redistributive policies, financed by distortionary taxes, in principle, affect capital accumulation and growth-promoting activities which in turn is actually detrimental to growth. If capital income inequality increases such that the preferred labor income tax rate falls as the (capital-poor) median voter cannot effectuate redistribution, then the subsequent rate of economic growth increases because smaller size of redistributive policies are financed by less distortionary taxes. If declining distortionary taxes translate into further less restriction on aggregate capital accumulation then it follows that subsequent economic growth will increase.

Appendix B: Capital Income Inequality, Taxation …

225

References Diamond PA, Mirrlees JA (1971) Optimal taxation and public production I: Production efficiency. Am Econ Rev 61(1):8–27 Gordon R, Kalambokidis L, Slemrod J (2004) Do we now collect any revenue from taxing capital income? J Publ Econ 88(5):981–1009 Gordon RH, Slemrod J (1988) Do we collect any revenue from taxing capital income? Tax Pol Econ 2:89–130 Luo W, Pickering A, Monterio PS (2017) Inequality and the size of government. University of York Discussion Papers in Economics (17/02) Meltzer AH, Richard SF (1981) A rational theory of the size of government. J Polit Econ 89(5):914–927 Persson T, Tabellini G (1994) Is inequality harmful for growth? Am Econ Rev 84(3):600–621

Appendix C

Alternative Measure of Inequality and Growth: Evidence from OECD Countries

C.1 Data and Econometric Specification The empirical analysis examines a panel of OECD countries over the period 1965– 2010.8 Following Perotti (1996), the dependent variable is the average rate of growth of income per capita per five-year period as yearly growth rates incorporate shortrun cyclical disturbances. For example, this means that growth rate in period 2 is averaged over 1971–1975 and is regressed on explanatory variables measured during period 1 (1966–1970).9 This reduces yearly serial correlation from business cycles. The change from previous model is to include capital income inequality and labor income inequality instead of aggregate inequality. The final data set, with means, and standard deviations is contained in Table C.1. Figure C.1 depicts the top income share data for all nineteen countries, showing all countries averagely experienced a downward trend in the earlier years followed by a period of stasis or even slight increase since around 1990. The argument proposed in this chapter is the following: as the top income share increases, distortionary taxes fall and investment is facilitated, which is likely to result in more accumulation and higher growth. Figure C.2 depicts the raw correlation between the change in GDP per capita between 1960 and 2010 and the lagged top income share. Below I show that when I control for initial GDP per capita, human capital and market distortions, there is no evidence of a negative relationship between top income share and GDP per capita growth; on the contrary, the relationship is significantly positive in many specifications. 8

Specifically the countries included are Australia, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Japan, Korea, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, the United Kingdom, and the United States. Current data availability for the top income share precludes using other countries. 9 In practice, each explanatory variable is measured in 1970, except capital income inequality and labor income inequality, which are sometimes not available in a specific year and is taken from the year closest to 1970. © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 W. Luo, Inequality, Demography and Fiscal Policy, Applied Economics and Policy Studies, https://doi.org/10.1007/978-981-99-0518-8

227

228

Appendix C: Alternative Measure of Inequality and Growth: Evidence …

Table C.1 Descriptive statistics Obs Mean TOPINC UTIP y MEDU FEDU PPPI PROP65 SHARE OUTLAYS

738 746 969 190 190 864 720 548 861

7.88 35.18 20.71 3.02 2.64 85.88 12.79 67.97 39.98

Std. dev.

Min

Max

2.35 3.54 9.37 1.38 1.41 25.20 2.84 6.03 10.65

3.49 27.42 1.07 0.32 0.14 34.58 5.73 44.74 12.8

18.33 44.12 53.77 6.59 5.84 179.06 21.02 82.10 72.4

Notes T O P I N C is the top 1% income share—taken from the WID. U T I P is the University of Texas Inequality Project’s Estimated Household Income Inequality. Income y is real GDP per capita in $000s of 2005 prices—taken from the Penn World Tables. M E DU and F E DU are respectively the average years of secondary schooling in the male and female population aged over 25—taken from Barro and Lee (1996). P P P I is the price level of investment measured as the PPP of investment over exchange rate relative to the United States–taken from the Penn World Tables. P R O P65 is the proportion of the population aged 65 and above—taken from WDI database. S H A R E is the business sector labor share—taken from the OECD database. OU T L AY S denotes total government outlays as a percentage of GDP—taken from the OECD Economic Outlook database

-5

Growth rate of GDP per capita 5 10 0

15

Growth Versus Capital Income Inequality

5

10 15 Top 1% income share gyp

Fitted values

Fig. C.1 Growth and capital income inequality, 1960–2010

20

Appendix C: Alternative Measure of Inequality and Growth: Evidence …

229

-5

Growth rate of GDP per capita 5 10 0

15

Growth Versus Labor Income Inequality

25

30

35 UTIP gyp

40

45

Fitted values

Fig. C.2 Growth and labor income inequality, 1960–2010

As noted above previous empirical literature including both country dummies and period dummies has generally been unsupportive of the original Persson and Tabellini (1994) hypothesis. If the mechanism put forward in this chapter is important, and productivity differences and capital-income inequality are correlated with each other, then arguably previous analyses have suffered from an omitted variable bias. A measure of productivity heterogeneity, U T I P (taken from the University of Texas Inequality Project’s Estimated Household Income Inequality data), is therefore employed in the empirical analysis as in chapter two. As the top income share will also be picking up productivity-induced inequality, the correlation between productivity inequality as measured by U T I P and the income share of the top 1% is somewhat weaker than might be expected, depicted by Fig. 2.7 with smaller sample of OECD countries.10 Hence there is meaningful separate information in the two series.11 Figure C.3 depicts the raw correlation between the change in GDP per capita between 1960 and 2010 and the lagged productivity-induced labor income inequality, showing that there is no evidence of a negative relationship between labor income inequality and GDP per capita, in support of Forbes (2000). The analysis includes control variables following Forbes (2000). Controls include per capita GDP in constant chained PPP US$ (denoted y). Per capita GDP y and the resultant growth rates are taken from the Penn World Tables. Following most empiri10

Figure 6 in Luo (2017) depicts a scatter plot of the two series with full sample, exhibiting a correlation coefficient of around 0.49. 11 The argument proposed in this chapter is that the top income share is especially informative about capital income inequality rather than productivity-induced labor income inequality.

230

Appendix C: Alternative Measure of Inequality and Growth: Evidence …

cal studies of income distribution and growth (e.g. Alesina and Rodrik 1994; Persson and Tabellini 1994) human capital effects are also included, and are represented by average years of secondary schooling in the male and female population aged over 25 (denoted M E DU and F E DU ), drawn from the data set compiled in Barro and Lee (1996). These two schooling variables proxy for the stock of human capital at the beginning of each of the estimation periods. The price level of investment (the PPP of investment over exchange rate relative to the United States, denoted P P P I ) as used in Perotti (1996) is also employed in the regression analysis to capture market distortions that affect the cost of investment, also taken from the Penn World Tables. Finally, the country dummies are employed to control for time-invariant omittedvariable bias, and the period dummies are employed to control for global shocks that may affect aggregate growth in any periods but are not captured by other explanatory variables. It is clearly possible to include a set of additional variables. However, as in Perotti (1996) this chapter mainly focuses on this simple specification for three considerations. First, in order to estimate the impact of inequality on growth it is important to make as few discrepancy as possible relative to typical growth model. Second, as the number of observations is limited by the availability of inequality data, this simplified specification will help maximize the number of degrees of freedom. Third, since some control variables used in standard-growth model (e.g. government expenditure) may be endogenous, focusing on stock variables measured at the start of each periods instead of flow variables measured throughout each periods can reduce the potential endogeneity problem. To summarize, the growth model central to be estimated is G R O W T Hi,t = β1 T O P I N Ci,t−1 + β2 U T I Pi,t−1 + β3 yi,t−1 + β4 M E DUi,t−1 + β5 F E DUi,t−1 + β6 P P P Ii,t−1 + αi + ηt + u i,t (C.1) where i represents each country and t represents each time period, G R O W T H is average annual growth, αi are country dummies, ηt are period dummies, and u i,t is the error term.

C.2 Panel Estimation Table C.2 contains estimation results from fixed-effects panel regressions with average annual growth rate as the dependent variable. Column 1 examines the original Persson and Tabellini (1994) hypothesis using five-year periods, 1965–2005, applying the benchmark specification in Forbes (2000) with productivity-induced inequality (U T I P), and finding its coefficient to be positive, though insignificant. This positive sign coheres with the results in Forbes (2000). Column 2 further augments this specification with capital income inequality. The estimated coefficient for capital income inequality is positive, with a p-value of 2.0% and the estimated relationship is sizable: A one standard deviation increase in capital income inequality is

Appendix C: Alternative Measure of Inequality and Growth: Evidence …

231

statistically correlated with a 0.9% increase in average annual growth over the next five years,12 consistent with the theoretical reasoning given here. It is also noteworthy that the coefficient estimate for productivity-induced labor income inequality is now negative, though is still not statistically significant. Following Forbes (2000) results are also presented (in columns 4 and 5) using ten-year panels, and the results essentially duplicate those in columns 1 and 2, establishing that this observed short-term, positive relationship is not dampened over time. Column 6 of Table C.2 contains Arellano-Bond dynamic panel estimation results extending the specification used in columns 4 and 5 to include the lagged dependent variable (G R O W T H ). Here the positive relationship between capital income inequality and growth holds up, and indeed the coefficient estimate pertaining to labor income inequality is negative and significantly different from zero at the 10% level, consistent with the Persson and Tabellini (1994) hypothesis. This evidence suggests that previous tests of the Persson and Tabellini (1994) hypothesis were potentially hampered by the conflation of capital and labor income inequality. Columns 7–9 again test 1–3 using extended sample of 1965–2010 and duplicate their results. Most of the coefficient estimates of control variables agree with those traditionally reported in typical literature. As indicated by models considering conditional convergence, the coefficient on initial income level is negative and statistically significant. Note also that the opposite signs on the coefficients of M E DU and F E DU are in line with the findings in Barro and Sala-I-Martin (2003) and Perotti (1996), who obtain the results based on a larger sample.13 For a given male attainment, an increase in initial female attainment leads to less backwardness and thus slower subsequent growth since the economy converges toward steady state (see Barro and Sala-I-Martin 2003).

C.3 Further Estimation Previous work on the effect of income inequality on economic growth (Forbes 2000) discusses the necessity to deal with potential endogeneity. Following the specification by Forbes (2000), column 1 of Table C.3 applies difference GMM by Arellano and Bond (1991) to a panel covering 18 OECD countries during 1965–2010 in fiveyear periods. The basic difference GMM regression, eliminating the fixed effects and using lags of the endogenous variables as instruments, produces similar results presented in Table C.2, in particular, significant and positive coefficient on lagged capital income inequality. While heightening the concern is the problem of weak instruments in difference GMM, which led to the development of system GMM 12

Note, however, that it is unlikely that any country’s top income share could rise by this magnitude in a short period of time. 13 The insignificance of the coefficients of M E DU and F E DU indicates that human capital accumulation in OECD countries is not the only one crucial driving force of economy, which in turn underpins the inequality-growth argument proposed by this chapter.

0.119 (0.113) −0.459*** (0.155) 0.141 (0.612) −0.369 (0.774) −0.00908 (0.0127) 138 19 5-year 0.523

0.412** (0.161) −0.108 (0.154) −0.612*** (0.130) 0.433 (0.504) −0.201 (0.615) −0.00212 (0.00690) 125 19 5-year 0.631

−0.0324 (0.0911) 0.491*** (0.164) −0.189 (0.165) −0.828*** (0.127) 0.412 (1.029) −0.439 (1.181) −0.00397 (0.0124) 92 18 5-year 0.198* (0.0974) −0.557*** (0.100) 0.0947 (0.546) 0.0459 (0.731) 0.00215 (0.0109) 70 19 10-year 0.595

(4)

0.529*** (0.140) −0.0668 (0.111) −0.610*** (0.0964) 0.915 (0.532) −0.462 (0.781) 0.0121 (0.00954) 63 19 10-year 0.736

(5) −0.0442 (0.203) 0.539*** (0.187) −0.265* (0.143) −0.792*** (0.138) 1.038 (1.012) −1.051 (1.257) 0.0263** (0.0107) 32 17 10-year

(6)

0.116 (0.120) −0.336** (0.123) −0.172 (0.696) 0.277 (0.827) −0.00737 (0.0124) 154 19 5-year 0.524

1965–2010 (7)

0.424** (0.179) −0.114 (0.154) −0.544*** (0.113) −0.137 (0.514) 0.551 (0.595) 0.00171 (0.00746) 141 19 5-year 0.626

(8)

−0.00593 (0.0795) 0.563*** (0.124) −0.145 (0.142) −0.825*** (0.0961) −0.137 (0.843) 0.483 (0.905) 0.00220 (0.0110) 118 18 5-year

(9)

Notes Panel regressions of average annual per capita growth rate including fixed effects, L .T O P I N C, L .U T I P, L .y, L .M E DU , L .F E DU , L .P P P I , and robust standard errors clustered by country in parentheses. Year dummies are included in all regressions. Columns (3) and (6) contain Arellano-Bond estimation with lagged values of both the predetermined and endogenous variables as instruments. Columns (7)–(9) again test (1)–(3) using extended sample 1965–2010. *, **, and *** respectively denote significance levels at 10, 5, and 1%

Obs Countries Periods R 2 (within)

L.PPPI

L.FEDU

L.MEDU

L.y

L.UTIP

L.TOPINC

L.GROWTH

Table C.2 Panel estimation results with fixed effects 1965–2005 (1) (2) (3)

232 Appendix C: Alternative Measure of Inequality and Growth: Evidence …

Appendix C: Alternative Measure of Inequality and Growth: Evidence …

233

by Arellano and Bover (1995) and Blundell and Bond (1998), and could reinforce endogeneity bias. The perfect p-value of 1.00 for the Hansen test is a classic sign of instrumental proliferation.14 The remaining columns 2–5 of Table C.3 examine the sensitivity of the results to reducing the number of instruments. Column 2 firstly collapses the instruments. Columns 3 and 4 use two different lags from the instrument set, and column 5 combines the two modification. It should also be noted that the AR(2) test and the Hansen J test show that there is no further serial correlation, and the overidentifying restrictions are not rejected. As difference GMM can suffer from the problem of weak instruments, the rest columns of Table C.3 utilise system GMM, which augments the equation estimated by difference GMM, simultaneously estimating an equation in levels with suitable lagged differences of endogenous variables as instruments. Therefore, columns 6–10 mimic columns 1–5 whilst instead using system GMM and produce similar results, which reinforce the proposed theory. Throughout Table C.3 the positive coefficients on capital income inequality lose significance as the number of instruments falls.

C.4 Sensitivity Analysis Table C.4 tests the robustness and contains estimation results from fixed-effects panel regressions using five-year periods. Column 1 uses the same specification as column 2 of Table C.2 but excluding Asian countries (i.e. Japan and Korea) to examine whether the regional coverage of the sample affects the results. Apart from the regional coverage, not surprisingly, the representative of very poor countries is extremely limited due to the unavailability of the top income share statistics. Alternatively, the relationship between capital income inequality and growth may depend on the stage of development of a country. I split the sample into wealthy and poor countries based on initial income level in 1965, and then reestimate equation (3.18) for two groups (reported in columns 2 and 3). Note that no matter which sample selection is utilized, the relationship between capital income inequality and growth remains positive and statistically significant. Column 4 of Table C.4 includes the percentage of population over the age of 65 (denoted P R O P65) as an additional control variable following the argument of Perotti (1996). This demographic variable may be correlated with income inequality as among retirees both average income and inequality are lower. In turn, if the population in a country is older, then the demand for social security is potentially higher and hence, more taxation distortions and slower subsequent growth. The coefficient on this demographic variable is negative and statistically significant at the 5% level, sup14

More recent literature on weak instruments (in system GMM estimation in particular) has indicated that if instruments are weak, then inferences based on conventional Wald statistics can be misleading. However, this is still an open question, and we should not simply conclude that the system GMM estimator is not a useful tool for conducting cross-country growth empirics (see Bazzi and Clemens 2013; Kraay 2015).

−0.786*** (0.0912)

−0.726 (1.020)

−0.507*** (0.0958)

−0.457 (0.460)

0.832* (0.490)

0.000338 (0.00628)

118

18

5-year

4.44

0.366

Difference GMM

L.y

L.MEDU

L.FEDU

L.PPPI

Obs

Countries

Periods

Hansen test

AR(2) p-value

Estimator

Collapse

Difference GMM

0.322

9.57

5-year

18

118

0.0166 (0.0129)

Lags 1–1

Difference GMM

0.264

7.16

5-year

18

118

0.00672 (0.0117)

0.727 (1.121)

−0.766 (0.972)

−0.862*** (0.121)

−0.123 (0.240)

0.387** (0.188)

(3)

Lags 1–2

Difference GMM

0.300

5.55

5-year

18

118

0.000378 (0.00728)

0.971 (0.662)

−0.608 (0.533)

−0.651*** (0.104)

−0.160 (0.169)

0.480** (0.203)

(4)

(5)

(6)

(7)

Collapse & lags 1–1

Difference GMM

0.997

exactly identified

5-year

18

118

System GMM

0.749

3.31

5-year

19

141

−0.0177** (0.00724)

−0.000266 (0.0242)

Collapse

System GMM

0.710

4.52

5-year

19

141

−0.0166 (0.0134)

0.900 (0.827)

−0.162 (0.452)

3.536 (5.658)

−0.211* (0.110)

−0.204* (0.110)

0.211 (0.136)

−0.437 (0.745)

−0.170** (0.0788)

−0.0351 (0.0449)

0.115* (0.0626)

0.390 (0.492)

−3.286 (4.913)

−0.366 (0.947)

−0.0492 (0.437)

0.359 (0.341)

(8)

Lags 1–1

System GMM

0.863

11.93

5-year

19

141

−0.0220*** (0.00750)

0.195 (0.387)

−0.134 (0.499)

−0.134* (0.0770)

−0.0210 (0.0454)

0.0803 (0.0622)

(9)

Lags 1–2

System GMM

0.749

5.54

5-year

19

141

−0.0177** (0.00724)

−0.162 (0.452)

0.390 (0.492)

−0.170** (0.0788)

−0.0351 (0.0449)

0.115* (0.0626)

(10)

Collapse & lags 1–1

System GMM

0.720

2.34

5-year

19

141

−0.0139 (0.0147)

2.475** (1.258)

−2.115 (1.322)

−0.330** (0.164)

−0.469 (0.358)

0.459 (0.292)

Notes In columns (1)–(5) estimations use the difference GMM of Arellano and Bond (1991), with robust standard errors. In columns (6)–(10) estimations use the system GMM of Arellano and Bover (1995) and Blundell and Bond (1998), with robust standard errors. “collapse” stands for collapsed instruments; “lags” stands for restricting the number of lags used in generating instuments from the endogenous variables. Year dummies are included in all regressions. Endogenous variables used as instruments: L .T O P I N C, L .U T I P, L .y, L .M E DU , L .F E DU , L .P P P I . *, **, and *** respectively denote significance levels at 10, 5, and 1%

Method to reduce count

−0.285 (0.224)

−0.147 (0.169)

L.UTIP

0.882 (1.295)

0.441*** (0.167)

(2)

0.386** (0.160)

L.TOPINC

(1)

1965–2010

Table C.3 Difference and system generalized method of moments regressions 234 Appendix C: Alternative Measure of Inequality and Growth: Evidence …

113 17 Excluding Asia 5-year 0.634

0.304** (0.114) −0.0493 (0.121) −0.439*** (0.119) −0.131 (0.460) 0.339 (0.509) −0.000104 (0.00806)

63 9 Higher income 5-year 0.608

0.350** (0.146) −0.0460 (0.129) −0.455*** (0.0712) 0.786 (0.964) 0.0446 (1.018) 0.00300 (0.00851)

62 10 Lower income 5-year 0.769

0.466* (0.209) −0.179 (0.195) −0.676* (0.302) 0.451 (1.358) −0.855 (2.694) 0.00235 (0.0279)

(3)

5-year 0.659

125 19 Full

0.325** (0.132) −0.0611 (0.108) −0.616*** (0.107) −0.162 (0.706) 0.327 (0.711) 0.00495 (0.00863) −0.463** (0.216)

(4)

5-year 0.676

−0.0550 (0.0561) 108 18 Full

0.440** (0.171) −0.126 (0.134) −0.738*** (0.0999) 0.316 (0.617) 0.0651 (0.661) 0.00433 (0.00769)

(5)

127 17 Excluding Asia 5-year 0.637

0.265** (0.110) 0.000345 (0.113) −0.383*** (0.0634) −0.806 (0.480) 1.186** (0.505) 0.00234 (0.00736)

69 9 Higher income 5-year 0.582

0.295** (0.117) 0.00976 (0.143) −0.404*** (0.0986) 0.0346 (0.915) 0.806 (0.785) 0.00891 (0.00883)

1965–2010 (6) (7)

72 10 Lower income 5-year 0.773

0.474** (0.189) −0.184 (0.200) −0.611** (0.221) 0.130 (0.704) −0.0412 (1.126) 0.00305 (0.0234)

(8)

5-year 0.670

141 19 Full

0.328** (0.132) −0.0585 (0.0933) −0.573*** (0.0763) −0.716 (0.604) 1.202* (0.606) 0.00657 (0.00754) −0.461*** (0.130)

(9)

5-year 0.667

−0.0658 (0.0579) 124 18 Full

0.440** (0.175) −0.110 (0.152) −0.655*** (0.101) −0.259 (0.662) 0.783 (0.701) 0.00629 (0.00907)

(10)

Notes Regression specification is the same as column (2) of Table C.2, and robust standard errors clustered by country in parentheses. Year dummies are included in all regressions. Column (1) excludes Asian countries. Columns (2) and (3) respectively correspond to higher and lower levels of initial income in 1965. Column (4) includes L .P R O P65 as a further control, and column (5) includes L .S H A R E as a further control. Columns (6)–(10) again test (1)–(5) using extended sample 1965–2010

Periods R 2 (within)

Obs Countries Data

L.SHARE

L.PROP65

L.PPPI

L.FEDU

L.MEDU

L.y

L.UTIP

L.TOPINC

1965–2005 (1) (2)

Table C.4 Sensitivity analysis

Appendix C: Alternative Measure of Inequality and Growth: Evidence … 235

236

Appendix C: Alternative Measure of Inequality and Growth: Evidence …

porting the mechanism proposed. Further, inequality stemming from capital income is likely to be correlated with the labor share of income (denoted S H A R E). As in Facchini et al. (2017) a recent declining labor share has played a part in explaining the slowdown in the growth of government size and therefore, less distortions and higher growth. In fact, no matter whether I control for P R O P65 or the labor share, as in columns 4 and 5, the coefficient on capital income inequality is positive and statistically significant at the 5% level. Note also that throughout columns 1–5 of Table C.4 the coefficient estimates for labor income inequality are consistently negative (though not significant). If interpreted causally, the estimated effect of capital income inequality on growth remains sizable: An increase in T O P I N C by one standard deviation is associated with an increase in average rate of growth of GDP per capita by around 0.7%. The contribution of this empirical work is that it does not follow previous empirical studies using aggregate income inequality measure (i.e. Gini coefficient), but instead splitting the aggregate income inequality measure into capital income inequality and labor income inequality. If only labor income inequality is incorporated, then evidence shows its coefficient to be positive, in direct contrast with earlier crosscountry OLS studies (e.g. see Alesina and Rodrik 1994; Persson and Tabellini 1994; Perotti 1996; and Deininger and Squire 1998). When capital income inequality is controlled (as the main innovation of this chapter), the relationship between labor income inequality and growth is found to be negative, consistent with those paper conjectured.15 References Alesina A, Rodrik D (1994) Distributive politics and economic growth. Q J Econ 109(2):465–490 Arellano M, Bond S (1991) Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Rev Econ Stud 58(2):277–297 Arellano M, Bover O (1995) Another look at the instrumental variable estimation of error-components models. J Econ 68(1):29–51 Barro RJ, Lee JW (1996) International measures of schooling years and schooling quality. Am Econ Rev 86(2):218–223 Barro RJ, Sala-i-Martin X (2003) Economic growth. McGraw-Hill, New York Bazzi S, Clemens MA (2013) Blunt instruments: avoiding common pitfalls in identifying the causes of economic growth. Am Econ J Macroecono 5(2):152–186 Blundell R, Bond S (1998) Initial conditions and moment restrictions in dynamic panel data models. J Econometrics 87(1):115–143 Deininger K, Squire L (1998) New ways of looking at old issues: inequality and growth. J Dev Econ 57(2):259–287 Facchini F, Melki M, Pickering A (2017) Labour costs and the size of government. Oxford Bullet Econ Stat 79(2):251–275 15

The empirical model presented above incorporates panel estimation as well as difference and system GMM estimation (and some sensitivity check), which plausibly indicates that it is adequate for distinguishing the impact of a second order effort, and it could engage in the discussion of inequality and growth in current literature.

Appendix C: Alternative Measure of Inequality and Growth: Evidence …

237

Forbes KJ (2000) A reassessment of the relationship between inequality and growth. Am Econ Rev 90(4):869–887 Kraay A (2015) Weak instruments in growth regressions: implications for recent cross-country evidence on inequality and growth. World Bank Policy Research Working Paper (7494) Luo W (2017) Inequality and growth in the 21st century. University of York Discussion Papers in Economics (17/18). Perotti R (1996) Growth, income distribution, and democracy: what the data say. J Econ Growth 1(2):149–187 Persson T, Tabellini G (1994) Is inequality harmful for growth? Am Econ Rev 84(3):600–621

Appendix D

Demography and Taxation: Further Global Evidence

D.1 Data and Descriptive Statistics The main agenda in this section is to test the hypothesis proposed above—whether and how the ratio of income to expenditure taxes across countries and time systematically changes with the fraction of the population that is retired. This empirical analysis focuses on a panel of over 100 countries. Cross-country annual data on income and expenditure tax revenue are available over the period 1990–2014 from the World Development Indicators (hereafter WDI). This chapter also reports results from crosscountry regressions with data measured by within-country averages. Moreover I separately examine how different categories of tax measures respectively co-move with the fraction of retirees. Following Pickering and Rajput (2018), the main dependent variable is the ratio τ of income taxes to expenditure taxes, T = τcy , constructed by the ratio of taxes on income, profits and capital gains (as a percentage share of total tax revenue) to taxes on goods and services (as a percentage share of total tax revenue). Both are extracted from the WDI database. In practice (and also within countries) rates of tax vary with different types of income and goods, but the measure of ratio proposed is a way to capture the extent to which taxes are levied on income relative to expenditure. Due to the relatively small value in the data of taxes on goods and services in the case of some countries, following Pickering and Rajput (2018) I use the natural logarithm of T , ln(T ), in the below regression analysis. The argument proposed predicts that the extent of taxes on income relative to expenditure declines with an increased fraction of the retired population. The measure of the proportion of retirees in the population used is the percentage of the population over the age of 65 (denoted P R O P65), which is also taken from the WDI database. This measure of the retired fraction is preferable to the dependency ratio used by Razin et al. (2002). The dependency ratio includes children as well as retirees which would have different impacts on taxes as shown by Shelton (2008).

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 W. Luo, Inequality, Demography and Fiscal Policy, Applied Economics and Policy Studies, https://doi.org/10.1007/978-981-99-0518-8

239

240

Appendix D: Demography and Taxation: Further Global Evidence

-4

ln(income taxes/expenditure taxes) 0 -2 2 4

ln(income taxes/expenditure taxes) Versus Aging

0

5

10 Proportion over 65

(mean) lreltaxes

15

20

Fitted values

Fig. D.1 Correlation between aging and the composition of taxes, 1990–2014

Figure D.1 depicts a scatter plot of the logarithm ratio of income to expenditure taxes and P R O P65, showing a negative relationship, in support of the proposed theory. Following Pickering and Rajput (2018), one important determinant of the composition of taxes is the level of development, so I include the natural logarithm of GDP per capita in constant chained PPP US$ (ln(y)), taken from the Penn World Tables, as a first control in the regression analysis. As another measure of the development level and institutional capacity, hence OECD membership (denoted O EC D) is also employed as a further control variable. To fully capture demographic effects, the econometric analysis includes the percentage of the population between 15 and 64 years of age (denoted P R O P1564, also from the WDI database) as an additional control. Another potential determinant of the composition of taxes is inequality (Pickering and Rajput 2018), taken from the University of Texas Inequality Project’s Estimated Household Income Inequality data of Galbraith and Kum (2005), so the inequality measure (denoted U T I P) is also included as a control. Governments collect tax revenue through means beyond taxation on income and consumption. One important source is the revenue from import duties and tariffs due to openness (Rodrik 1998). Thus the trade share (the sum of exports and imports as a percentage of GDP— denoted T R AD E) is also employed in the regression analysis. Apart from these control variables the natural logarithm of the total population size (denoted ln(P O P)) is included as well, to some extent capturing any scale (dis-) economies related to particular forms of tax collection.16 These data are also taken from the WDI database. 16

For example, larger size of population means more sources of tax collection, which may lead to scale economies or scale diseconomies depending on its size.

Appendix D: Demography and Taxation: Further Global Evidence Table D.1 Descriptive statistics Obs Mean ln(T ) τy τc PROP65 PROP1564 UTIP ln(y) OECD TRADE ln(P O P) POLITY 2 YGAP

2091 2142 2141 4633 4633 1556 3652 5325 4263 5302 3790 4668

−0.261 22.45 29.06 7.01 61.20 42.79 8.58 0.138 86.94 15.07 3.04 0

241

Std. dev.

Min

Max

1.00 12.75 13.76 4.72 7.01 6.71 1.29 0.345 51.96 2.35 6.69 0.034

−4.00 0.349 0.024 0.335 45.29 22.75 5.03 0 0.309 9.11 −10 −0.609

5.09 66.72 89.22 25.08 85.81 59.96 11.73 1 531.7 21.04 10 0.505

Notes τ y denotes taxes on income, profits and capital gains as a percentage of revenue— taken from the World Development Indicators (WDI). τc denotes taxes on goods and services as a percentage τ of revenue— also taken from the WDI. T = τcy . P R O P1564 and P R O P65 are respectively the proportion of the population aged between 15 and 64, and 65 and above. U T I P is the University of Texas Inequality Project’s Estimated Household Income Inequality. y is real GDP at chained PPPs in millions of 2005 US dollars per capita—taken from the Penn World Tables. O EC D is a dummy variable denoting OECD membership. T R AD E is the sum of exports and imports as a percentage of GDP. P O P is the size of country population. P O L I T Y 2 is a measure of democracy provided by the Polity IV project, with −10 denoting the highest level of autocracy, and 10 denoting the highest level of democracy. Y G A P is the difference between the actual output and its trend value in percentage

There may also be cyclical movements in policy variables. To address this potential problem the regression analysis includes the output gap (the difference between aggregate output and its trend value in percentage—denoted Y G A P) as a further control. The policy variables may also be affected by the degree of democracy through various channels, so the democracy score (with −10 denoting the highest level of autocracy, and 10 denoting the highest level of democracy) provided by the Polity IV project is included as a final control (denoted P O L I T Y 2). Table D.1 contains descriptive statistics of the variables used in the regression analysis. Note that there is considerable dispersion in how countries raise their tax revenue. Over the sample period taxes on income on average represent a smaller fraction of total revenue than taxes on goods and services (Fig. D.2 depicts a scatter plot of the two series).17 This indicates that the capacity to raise revenue through income taxes is normally limited in those countries with low income. For instance in the OECD members, taxes on income are on average 32% of total revenue, whilst outside the OECD income taxes account for just 20% of total revenue on average (Fig. D.3 depicts a line plot of the two series). 17

For example, Macao has a very large size of expenditure taxes relative to income taxes.

242

Appendix D: Demography and Taxation: Further Global Evidence

0

Income taxes as a % of total revenue 60 20 40

Income Taxes Versus Expenditure Taxes

0

20 40 60 Expenditure taxes as a % of total revenue (mean) taxes_ipcg_revenue

Fitted values

Fig. D.2 Correlation between income taxes and expenditure taxes, 1990–2014

Fig. D.3 Income taxes in OECD and non-OECD countries, 1990–2014

80

Appendix D: Demography and Taxation: Further Global Evidence

243

The P R O P65 data cover 155 countries, and numerically range from 1.37 (Qatar) to 18.39 (Italy) on average, with higher numbers meaning greater proportion of the retired population. Notably, these data are positively correlated with GDP per capita across the sample, with a correlation coefficient of around 0.72. Richer countries have greater retired fraction of the population than poorer countries. This emphasizes the need to include controls for the level of economic development, or the retired fraction may become a proxy for other drivers of the policy variables.

D.2 Panel Estimation D.2.1 Baseline Estimation Columns 1 and 2 of Table D.2 contain estimation results examining the impact of population aging on the ratio of income to expenditure taxes using OLS. Column 1 is a simple specification with the fraction of the retired population (P R O P65) and a number of control variables using annual data, with robust standard errors clustered by country. Column 2 extends the regression of column 1 to include time effects. The use of time effects will substantially control for the potential problem of a secular trend. In these specifications the sign of the coefficient estimate relating to the fraction of the retired population is negative in all cases, and all are statistically significant at the 1% level. This is consistent with the theory—an increase in the retired fraction increases expenditure taxes relative to income taxes. Columns 3 and 4 repeat the analysis of columns 1 and 2 using country fixed effects panel estimation instead. The results using panel estimation support those already found. The estimated statistical significance of the fraction of the retired population is unaffected and even remains at the 5% level in column 3. Using the estimate from column 3 of Table D.2, a one standard deviation increase in the fraction of the retired population is statistically associated with a fall of 0.63 in the ratio of income to expenditure taxes, holding all else equal. The magnitude of this estimated correlation is sizable - implying more than a half of the raw standard deviation in the policy variables. One may argue that the existence of P R O P1564 counteracts the efficiency of P R O P65 as a measure of population aging. To address the potential counteractive effect of P R O P1564, columns 5 and 6 of Table D.2 again test columns 3 and 4 with full control variables except P R O P1564. Further, columns 7 and 8 instead use the ratio of the population above 65 to those between the ages of 15 and 64, R AT I O, as in Shelton (2008), to measure population aging, and mimic columns 3 and 4. The results similarly demonstrate an increased tendency to use expenditure taxes rather than income taxes as population aging increases.

PROP1564

(3)

Panel

No

No

0.422

Data

Fixed Effects?

Time Effects?

R2

0.435

Yes

No

Panel

87

796

0.000618 (0.00131)

0.172

No

Yes

Panel

87

796

0.00172 (0.610)

0.000637 (0.00212)

0.195

Yes

Yes

Panel

87

796

−0.541 (0.707)

−0.0110 (0.0121)

0.430 (0.948)

0.000294 (0.00232)

(5)

(6)

0.143

No

Yes

Panel

87

796

−0.964 (0.792)

0.178

Yes

Yes

Panel

87

796

−0.512 (0.736)

−0.0139 (0.0113)

0.535 (0.956)

−0.508 (0.598) −0.0183 (0.0112)

0.000895 (0.00235)

0.0825 (0.114)

0.853*** (0.196)

−0.00911 (0.0191)

−0.0722 (0.0625)

0.00121 (0.00207)

−0.0406 (0.0704)

0.842*** (0.202)

−0.0145 (0.0208)

−0.124** (0.0576)

0.130

No

Yes

Panel

87

796

−1.010 (0.817)

−0.0201* (0.0109)

−0.513 (0.610)

0.000832 (0.00202)

−0.0661 (0.0712)

0.755*** (0.189)

−0.0162 (0.0222)

−0.0647* (0.0338)

(7)

0.175

Yes

Yes

Panel

87

796

−0.515 (0.737)

−0.0144 (0.0114)

0.782 (0.980)

0.000792 (0.00243)

0.0674 (0.121)

0.832*** (0.191)

−0.00933 (0.0192)

−0.0231 (0.0354)

(8)

Notes Columns (1) and (2) contain results using OLS regressions of the composition of taxes, ln(T ), including P R O P1564, U T I P, ln(y), O EC D, T R AD E, ln(P O P), P O L I T Y 2, and Y G A P as control variables. Columns (3) and (4) contain results using Panel regressions with country fixed effects. Columns (5) and (6) again test columns (3) and (4) without P R O P1564 as a control. Columns (7) and (8) instead use the ratio of the population above 65 to those between the ages of 15 and 64, R AT I O, to measure population aging, and mimic columns (3) and (4). Robust standard errors are shown in parentheses. Standard errors are clustered by country. *, **, and *** respectively denote significant levels at 10, 5 and 1%

796

0.253 (0.989)

YGAP

87

−0.132 (1.028)

−0.0166 (0.0214)

POLITY 2

Countries

−0.937 (0.734)

−0.0189 (0.0217)

0.150*** (0.0562)

ln( P O P )

Observations

−0.0134 (0.0124)

0.137** (0.0575)

0.00112 (0.00133)

0.0703 (0.117)

TRADE

0.338 (0.221)

−0.0399 (0.0833)

0.311 (0.226)

−0.00986 (0.0175)

−0.0452** (0.0224)

1.024*** (0.226)

OECD

(4) −0.125* (0.0643)

1.072*** (0.231)

0.715*** (0.156)

ln( y )

0.713*** (0.151)

−0.000347 (0.0143)

−0.0126 (0.0184)

−0.0525** (0.0246)

−0.133** (0.0519)

UTIP

−0.00770 (0.0153)

−0.0755*** (0.0193)

−0.0716*** (0.0190)

PROP65

RATIO

(2)

−0.104*** (0.0243)

(1)

−0.0954*** (0.0231)

Dep Var: ln(T )

Table D.2 Basic estimation results—the composition of taxes (annual data)

244 Appendix D: Demography and Taxation: Further Global Evidence

Appendix D: Demography and Taxation: Further Global Evidence

245

D.2.2 Heterogeneity It is natural to investigate whether or not the results reported change with the degree of democracy, given that the theory proposed is based on the median voter framework. Table D.3 thus extends the regression results by splitting the sample by levels of democracy. Column 1 contains results for countries with stronger democratic credentials (i.e. with the democracy score P O L I T Y 2 of 7 or above over the sample period). Column 2 contains results for countries with weaker democratic credentials (i.e. with P O L I T Y 2 of less than 7). The democracy criterion is strengthened further in columns 3 and 4, and the sample is split according to P O L I T Y 2 ≷ 8. When the sample is separated it becomes clear that the negative relationship between the fraction of the retired population and the ratio of income to expenditure taxes holds only in the subsample of democratic regimes. This is in line with the theory, which assumes a complete franchise. In column 3 the p-value for the estimated coefficient for the fraction of the retired population is 1.1%, and the estimated effect is sizable: A one standard deviation increase in the fraction of the retired population is statistically associated with the policy variable ln(T ) which is smaller by 0.78, holding all else equal. Note that it is also of interest to ask whether there are other stories, related to the Laffer curve, explaining the recent tendency of increasing expenditure taxes. The Laffer curve suggests that when income tax rates increase from low levels, the tax revenue collected by government also increases. If tax rates keep increasing after a certain point, then it would cause people not to work as hard as before, thereby reducing tax revenue. One common explanation for the appearance of expenditure taxes is that income taxes were close to their Laffer curve peaks. Columns 5 and 6 of Table D.3 split the sample with stronger democratic credentials (P O L I T Y 2 ≥ 8) by τ y (determined by the median value of τ y ). If the story of the Laffer curve explains the results, when τ y increases from lower to higher levels, then the tax revenue generated by taxes on income relative to expenditure firstly increases and then declines. This indicates that the sign of the coefficient on P R O P65 should be reversed at lower and higher levels of τ y , holding all else equal. Statistical significance in columns 5 and 6 implies that the estimates are stable across these subsamples, which in turn supports the theory proposed. It is also natural to see whether the results vary with level of development. Columns 7 and 8 of Table D.3 split the sample by levels of GDP per capita (determined by the median value of GDP per capita). As can be seen in all cases, the ratio of income to expenditure taxes is negatively correlated with the fraction of the retired population. However, this negative relationship holds significantly only in the group of countries with higher income level. Rich countries commonly have larger fractions of the retired population, and therefore tax revenue collected by taxes on income is reduced relative to expenditure. In column 7 the p-value for the estimated coefficient for the fraction of the retired population is 4.2%, and the estimated effect is sizable: a one standard deviation increase in the fraction of the retired population is statistically associated with a reduction of 0.61 in the policy variable ln(T ).

0.0458 (0.102)

0.00210 (0.00265)

ln( y )

OECD

TRADE

ln( P O P )

−0.0232 (0.0206)

Yes

POLITY2 < 7

Panel

Yes

POLITY2 ≥ 7

0.255

Data

Fixed Effects?

Sample

R2

276

0.278

POLITY2 ≥ 8

Yes

Panel

56

496

−1.749 (1.405)

(4)

0.201

POLITY2 < 8

Yes

Panel

41

300

−0.146 (1.015)

−0.0106 (0.0213)

−0.981* (0.515)

0.000869 (0.00280)

−0.0501 (0.118)

0.875*** (0.299)

0.000338 (0.0211)

−0.0115 (0.0167)

−0.116 (0.0955)

(5)

0.248

POLITY2 ≥ 8 & High τ y

Yes

Panel

38

273

−0.451 (1.129)

0.397 (0.365)

0.715 (1.340)

0.00486 (0.00449)

0.892* (0.447)

−0.0231 (0.0169)

−0.106** (0.0496)

−0.111* (0.0580)

(6)

(7)

0.347

POLITY2 ≥ 8 & Low τ y

Yes

Panel

32

223

−1.664 (1.408)

−0.154** (0.0716)

0.214

High y

Yes

Panel

67

559

−2.434* (1.239)

−0.0249* (0.0142)

−0.955 (1.234)

0.000949 (0.00219)

−0.000143 (0.00262)

3.173** (1.352)

−0.000149 (0.0939)

1.171*** (0.412)

−0.0437* (0.0251)

−0.0253 (0.0289)

−0.129** (0.0623)

−0.0244 (0.0655)

1.595*** (0.296)

−0.00980 (0.0237)

−0.0774* (0.0433)

−0.124*** (0.0406)

(8)

0.215

Low y

Yes

Panel

31

237

−0.873 (0.850)

−0.00594 (0.0131)

0.747 (0.811)

0.00140 (0.00354)

0.967*** (0.328)

0.0199 (0.0189)

−0.100** (0.0432)

−0.128 (0.248)

Notes Regression specification is the same as column 3 of Table D.2. Columns (1) and (2) respectively correspond to higher and lower democracy levels (according to P O L I T Y 2 ≷ 7). Columns (3) and (4) instead respectively correspond to higher and lower democracy levels (according to P O L I T Y 2 ≷ 8). Columns (5) and (6) respectively correspond to higher and lower levels of τ y under the sample with higher democracy level (P O L I T Y 2 ≥ 8). Columns (7) and (8) respectively correspond to higher and lower levels of income

0.218

Panel

37

520

61

−0.133 (1.020)

−2.197 (1.351)

YGAP

Countries

−0.0100 (0.0219)

−0.0429 (0.145)

POLITY 2

Observations

1.833* (1.081)

−0.991* (0.531)

1.419 (1.122) 0.0149 (0.156)

0.00126 (0.00259)

0.000153 (0.00285)

0.0532 (0.0877)

1.647*** (0.455)

UTIP

−0.119** (0.0507)

−0.0559 (0.127)

0.00239 (0.0225)

−0.0252 (0.0198)

PROP1564

(3) −0.165** (0.0629)

1.628*** (0.465)

−0.0152 (0.0181)

−0.138*** (0.0442)

PROP65

0.882*** (0.308)

(2)

−0.0852 (0.108)

(1)

−0.161** (0.0623)

Dep Var: ln(T )

Table D.3 Estimation results—the composition of taxes (annual data)

246 Appendix D: Demography and Taxation: Further Global Evidence

Appendix D: Demography and Taxation: Further Global Evidence

247

D.2.3 Robustness As a robustness check, columns 1–6 of Table D.4 again use panel regressions with country fixed effects including full controls except P R O P1564, with robust standard errors clustered by country. Columns 1 and 2 respectively correspond to higher and lower democracy levels (according to P O L I T Y 2 ≷ 7), and columns 3 and 4 explore further stronger democratic requirement (i.e. P O L I T Y 2 ≷ 8). Columns 5 and 6 instead split the sample of countries according to higher and lower levels of income. The estimation results support the findings in Table D.3. Columns 7–10 then present regressions of the taxes composition on the ratio of old to young, R AT I O. The results presented in columns 1–4 of Table D.3 and columns 1–4 of Table D.4 clearly establish that the estimated effect is predominantly driven by countries with high P O L I T Y 2 scoring scale. Note that these data are not classically normallydistributed, since there is a cluster of countries scoring 10. While many countries are with ‘intermediate’ P O L I T Y 2 scores, indicating a substantial political volatility (i.e. democratic reversals). This is likely to create further volatility in fiscal policy decision. Columns 7 and 8 therefore split the sample according whether or not P O L I T Y 2 = 10, showing a ‘perfect’ democracy throughout the period. Both of the relevant coefficient estimates are negative and statistically different from zero. Notably column 9 again confirms that the estimated effect holds significantly in rich countries.

D.2.4 Income Taxes In Tables D.5 and D.6 results are presented respectively for income taxes, τ y , and expenditure taxes, τc , the numerator and denominator in the main dependent variable, using Panel regressions as above. In Table D.5 the findings for income taxes, τ y , are quite similar to the results found for ln(T ) though with lower significance levels. Increases in the share of retirees are generally found to be negatively correlated with the extent to which taxes are levied on income, but more so in the stronger democracies. In countries where full sample is included or P O L I T Y 2 ≥ 7, the estimated effect remains negative, though is not statistically significant. When the stronger democratic criterion (i.e. P O L I T Y 2 ≥ 8) is employed, the estimated effect increases and is statistically significant at the 10% level. When the sample is refined further to those countries with P O L I T Y 2 = 10 throughout the same period (in columns 6 and 7), and utilizing instead the ratio of old to young (R AT I O) measure of aging, the negative coefficient estimate is sustained, although statistical significance is in this instance low. If all control variables are excluded except ln(y), then the p-value of the coefficient estimate pertaining to R AT I O in column 6 improves to p = 0.023. Using the estimate of column 3, a one standard deviation increase in the fraction of the retired population is statistically associated with a reduction of 6.73 in

Yes

POLITY2 < 7

Panel

Yes

POLITY2 ≥ 7

0.148

Data

Fixed effects?

Sample

R2

276

0.196

POLITY2 ≥ 8

Yes

Panel

56

496

−1.603 (1.382)

(4)

0.200

POLITY2 < 8

Yes

Panel

41

300

−0.149 (1.028)

−0.0125 (0.0199)

−1.083* (0.561)

0.00121 (0.00284)

−0.0492 (0.114)

0.834*** (0.276)

−0.000954 (0.0208)

−0.124 (0.100)

(5)

0.206

High y

Yes

Panel

67

559

−2.397* (1.239)

−0.0250* (0.0139)

−1.166 (1.288)

0.00123 (0.00230)

−0.0112 (0.0805)

1.066*** (0.390)

−0.0456* (0.0272)

−0.123* (0.0622)

(6)

0.151

Low y

Yes

Panel

31

237

−1.512 (1.093)

−0.0125 (0.0128)

−0.121 (0.692)

0.00194 (0.00443)

0.679* (0.350)

0.0216 (0.0215)

−0.278 (0.229)

Yes

Panel

62

515

−1.027 (0.881)

Yes

Panel

67

559

−2.327* (1.241)

−0.0247* (0.0139)

−1.124 (1.308)

0.00111 (0.00224)

−0.0310 (0.0786)

0.954*** (0.355)

−0.0487* (0.0292)

−0.0701** (0.0343)

(9)

0.334

0.171

0.198

POLITY2 = 10 POLITY2 < 10 High y

Yes

Panel

31

281

1.918 (1.147)

−0.0125 (0.0124)

−0.619 (0.633)

0.00302 (0.00286)

−0.00386 (0.00260) −0.882 (1.215)

0.0289 (0.0949)

0.860*** (0.239)

−0.0227 (0.0227)

−0.138** (0.0599)

(8)

−0.117 (0.164)

0.999*** (0.192)

0.0254 (0.0184)

−0.0317** (0.0132)

(7)

0.115

Low y

Yes

Panel

31

237

−1.511 (1.066)

−0.0175 (0.0117)

−0.294 (0.641)

0.00129 (0.00432)

0.560 (0.367)

0.0259 (0.0234)

−0.0831 (0.214)

(10)

Notes Columns (1)–(6) use Panel regressions with country fixed effects without P R O P1564 as a control. Columns (1) and (2) respectively correspond to higher and lower democracy levels (according to P O L I T Y 2 ≷ 7). Columns (3) and (4) instead respectively correspond to higher and lower democracy levels (according to P O L I T Y 2 ≷ 8). Columns (5) and (6) respectively correspond to higher and lower levels of income. Columns (7)–(10) instead use the ratio of the population above 65 to those between the ages of 15 and 64, R AT I O, to measure population aging. Columns (7) and (8) respectively correspond to P O L I T Y 2 = 10 and P O L I T Y 2 < 10. Columns (9) and (10) respectively correspond to higher and lower levels of income. Robust standard errors are shown in parentheses. Standard errors are clustered by country. *, **, and *** respectively denote significant levels at 10, 5 and 1%

0.214

Panel

37

520

61

Countries

−0.149 (1.039)

−1.990 (1.341)

YGAP

Observations

−0.0125 (0.0203)

−0.186 (0.158)

POLITY 2

−0.0664 (0.168)

0.832 (1.335)

ln( P O P )

−1.121* (0.584)

0.00123 (0.00271)

TRADE

−0.0887 (0.0700)

0.195 (1.420)

−0.127* (0.0739)

OECD

1.080*** (0.289)

−0.0308 (0.0226)

0.000462 (0.00264)

−0.0541 (0.122)

1.044*** (0.310)

ln( y )

(3)

−0.117* (0.0631)

0.000656 (0.00294)

0.826*** (0.285)

−0.0290 (0.0250)

UTIP

0.000814 (0.0225)

PROP65

RATIO

(2)

−0.0968 (0.113)

(1)

−0.117* (0.0665)

Dep Var: ln(T )

Table D.4 Estimation results—the composition of taxes (annual data)

248 Appendix D: Demography and Taxation: Further Global Evidence

Appendix D: Demography and Taxation: Further Global Evidence

249

τ y , holding all else equal. Given that this is nearly about half of a standard deviation in the policy variable, the magnitude of the estimated correlation is sizable. This weak estimated relationship indicates that there is a very slight variation in income taxes within countries over the sample period. This in turn emphasizes the need and motivation to examine the relationship between population aging and the extent to which taxes are levied on income relative to expenditure, instead of income taxes only. In Table D.5 the results relating to the control variables are of some interest. One regularity is that consistent with Perotti (1996), Benabou (1996), and results in chapter two, who have generally challenged the Meltzer and Richard (1981) hypothesis, there is a clear negative relationship between income taxes and income inequality (though at weak significance levels). This indicates that a more unequal distribution of income implies divergence between mean and median income and so, under universal suffrage, reduces income taxes. In addition as shown by Besley and Persson (2014) there is a positive relationship with income per capita, which likely indicates greater potential to tax in richer countries. Further, trade is found to be positively associated with income taxes as in Rodrik (1998), which shows a greater potential to tax in countries with higher level of openness.

D.2.5 Expenditure Taxes Table D.6 contains estimation results relating to τc , the extent to which taxes are raised through expenditure on goods and services. In contrast to income taxes, increases in the share of retirees are generally found to be positively related to the extent to which expenditure taxes are used, and again this result is particularly strong in the stronger democracies. In countries where P O L I T Y 2 < 7, the estimated relationship is positive, though it is not statistically significant, whilst in countries where P O L I T Y 2 ≥ 7, the estimated effect is statistically significant at the 1% level. When the stronger democratic requirement (i.e. P O L I T Y 2 ≥ 8) is applied, the estimated effect remains positive and statistically significant at the 1% level, whilst in countries where P O L I T Y 2 < 8, the estimated relationship is found to be positive, though at a weaker significance level. The coefficient estimate is also positive and statistically significant when the sample is refined further to those pure democracies (P O L I T Y 2 = 10), and utilizing instead the R AT I O measure of aging in column 6. Using the estimate of column 5, a one standard deviation increase in the fraction of the retired population is statistically associated with a increase of 8.84 in τc , holding all else equal. As with ln(T ), this again represents more than a half of the raw standard deviation in τc , so this is still a sizable effect. There are some differences between the results relating to the controls for income taxes and expenditure taxes. For instance there is a clear positive relationship between expenditure taxes and income inequality, which is opposite to the findings in income taxes and implies that if the median voter becomes relatively poor, then he is likely to tax more on expenditure instead of income. Further in contrast to τ y there is a

−0.307 (0.307)

PROP1564

Yes

Full

0.090

Sample

R2

0.263

POLITY2 ≥ 8

POLITY2 ≥ 7

0.242

Yes

Panel

57

515

−16.67 (20.04)

0.414 (1.750)

30.57** (12.24)

0.0693** (0.0297)

−0.141 (0.753)

16.63** (6.548)

−0.312 (0.243)

−1.364** (0.658)

Yes

Panel

62

(3) −1.425* (0.792)

(4)

0.181

POLITY2 ≥ 7

Yes

Panel

62

539

−20.39 (17.85)

−1.271 (1.693)

14.85 (14.89)

0.0549** (0.0268)

−1.768 (1.347)

10.57** (4.100)

−0.378 (0.249)

−0.844 (0.727)

(5)

0.208

POLITY2 ≥ 8

Yes

Panel

57

515

−14.30 (19.53)

−0.517 (1.996)

19.48 (14.90)

0.0589** (0.0274)

−1.737 (1.372)

10.41** (4.071)

−0.394 (0.266)

−0.892 (0.699)

0.261

POLITY2 = 10

Yes

Panel

32

295

22.06* (12.83)

0.100

POLITY2 < 10

Yes

Panel

63

531

−16.25 (14.38)

−0.370 (0.255)

−5.042 (9.814)

0.0671** (0.0323)

−0.0111 (0.0284) −18.29 (15.34)

−3.185 (2.685)

5.903 (4.655)

−0.271 (0.284)

−0.429 (0.738)

(7)

−1.508 (2.706)

11.00*** (3.498)

0.412** (0.197)

−0.213 (0.173)

(6)

Notes Estimations contain results using Panel regressions with country fixed effects of income taxes, τ y , including P R O P1564, U T I P, ln(y), O EC D, T R AD E, ln(P O P), P O L I T Y 2, and Y G A P as control variables. Column (1) includes full data sample; Columns (2) and (3) respectively correspond to different democracy levels (P O L I T Y 2 ≥ 7 and P O L I T Y 2 ≥ 8). Columns (4) and (5) again mimic columns (2) and (3) without P R O P1564 as a control. Columns (6) and (7) instead use the ratio of the population above 65 to those between the ages of 15 and 64, R AT I O, to measure population aging, respectively correspond to P O L I T Y 2 = 10 and P O L I T Y 2 < 10. Robust standard errors are shown in parentheses. Standard errors are clustered by country. *, **, and *** respectively denote significant levels at 10, 5 and 1%

Panel

Fixed Effects?

−13.17 (12.89)

YGAP

Data

−23.05 (18.41)

−0.382 (0.238)

POLITY 2

539

0.147 (1.594)

−2.303 (9.922)

ln( P O P )

826

26.68** (12.16)

0.0410* (0.0211)

TRADE

89

0.0653** (0.0289)

−2.855 (2.394)

OECD

Countries

−0.0816 (0.680)

7.378 (4.489)

ln( y )

Observations

16.57*** (6.128)

−0.147 (0.247)

UTIP

−0.345* (0.201)

−1.377** (0.573)

−0.340 (0.687)

PROP65

RATIO

(2)

−1.275 (0.765)

(1)

Dep Var: τ y

Table D.5 Estimation results—income taxes (annual data) 250 Appendix D: Demography and Taxation: Further Global Evidence

Appendix D: Demography and Taxation: Further Global Evidence

251

negative relationship between τc and income per capita, which reflects the ability to collect revenue through taxes on income in particular.

D.2.6 Cross-Country Estimation Table D.7 presents OLS estimation results using cross-country averages to examine the effect of population aging on the ratio of income to expenditure taxes, with robust standard errors. This econometric analysis at least has the advantage of addressing potential cyclicality in the data. The estimated effect using cross-country regression still remains negative and is statistically significant at the 1% level when all controls except ln(y) are dropped or full controls are incorporated (in columns 1 and 2). Using the estimate in column 2, a one standard deviation increase in the proportion of the retired population is statistically associated with a reduction of 0.46 in the policy variable ln(T ). In Tables D.8 and D.9 results are again presented respectively for τ y and τc whilst using cross-country estimation. As in Razin et al. (2002) population aging leads to smaller income taxes, and the estimated effects are generally statistically significant and are moreso in the stronger democracies. On the other hand, population aging is found to be positively related with expenditure taxes but at weaker significance levels, while this relationship holds in regimes with higher democratic scores (i.e. P O L I T Y 2 ≥ 9). In the case of cross-country estimation in income and expenditure taxes, some results relating to the control variables are interesting. Notably the extent of democracy is positively associated with both τ y and τc . This means that revenue relied on τ y and τc is increasingly related with the stronger level of democracy. Further in line with Baunsgaard and Keen (2010), trade is negatively related to τc (though at weak significance level), which implies that globalization might constrain the capacity to raise revenue through taxes on goods and services especially more pressure in countries without alternative sources to collect revenue.

D.3 Conclusion The relationship between population aging and the composition of taxes is tested using international panel data, including the fraction of the population that is retired as an explanatory variable. Data for taxes composition and demography are all from the WDI database. Consistent with the theory, the extent of taxes on income relative to expenditure is found to be negatively associated with the fraction of the retired population.18 Moreover, income taxes as a proportion of total revenue fall with aging 18

This is in line with intuition: when the median voter is of working age, then population aging increases the demand for expenditure rather than income taxes in order to increase the tax burden on the retired population.

0.428** (0.195) −17.31*** (4.735) −0.740 (3.168) 0.0131 (0.0300) −12.34 (13.53) 0.488 (1.560) 26.43** (11.69) 521 61 Panel Yes POLITY2 ≥ 7 0.249

2.318*** (0.713) 1.447*** (0.515)

(2)

0.403* (0.216) −18.66*** (4.965) −1.030 (3.089) 0.0251 (0.0281) −14.34 (14.14) 0.661 (1.708) 26.49** (11.72) 496 56 Panel Yes POLITY2 ≥ 8 0.266

2.459*** (0.766) 1.476** (0.590)

(3)

0.467* (0.239) −10.97*** (3.446) 1.079 (2.270) 0.0222 (0.0320) 0.505 (16.22) 1.992 (1.567) 24.27** (11.90) 521 61 Panel Yes POLITY2 ≥ 7 0.165

1.861** (0.801)

(4)

0.498** (0.222) −11.88*** (3.439) 0.726 (2.175) 0.0350 (0.0304) −1.944 (16.70) 1.667 (1.862) 24.68** (11.37) 496 56 Panel Yes POLITY2 ≥ 8 0.183

1.872** (0.806)

(5)

−8.405 (22.93) 281 31 Panel Yes POLITY2 = 10 0.210

0.485* (0.251) 0.132 (0.473) −12.06** (5.128) 1.820* (0.964) 0.0710* (0.0359) 14.43 (21.48)

(6)

2.088** (0.905) 0.294 (0.270) −6.643** (3.143) −3.411 (4.018) −0.00922 (0.0335) 9.005 (12.52) −0.0686 (0.165) 8.095 (9.098) 516 62 Panel Yes POLITY2 < 10 0.109

(7)

Notes Estimations contain results using Panel regressions with country fixed effects of expenditure taxes, τc , including P R O P1564, U T I P, ln(y), O EC D, T R AD E, ln(P O P), P O L I T Y 2, and Y G A P as control variables. Column (1) includes full data sample; Columns (2) and (3) respectively correspond to different democracy levels (P O L I T Y 2 ≥ 7 and P O L I T Y 2 ≥ 8). Columns (4) and (5) again mimic columns (2) and (3) without P R O P1564 as a control. Columns (6) and (7) instead use the ratio of the population above 65 to those between the ages of 15 and 64, R AT I O, to measure population aging, respectively correspond to P O L I T Y 2 = 10 and P O L I T Y 2 < 10. Robust standard errors are shown in parentheses. Standard errors are clustered by country. *, **, and *** respectively denote significant levels at 10, 5 and 1%

Observations Countries Data Fixed Effects? Sample R2

YGAP

POLITY 2

ln(P O P)

TRADE

OECD

ln(y)

UTIP

RATIO

0.228 (0.222) −11.01*** (2.808) −1.702 (2.343) 0.0152 (0.0277) −1.007 (11.57) −0.0797 (0.184) 7.993 (7.769) 797 87 Panel Yes Full 0.139

PROP65

PROP1564

(1)

2.166*** (0.739) 0.869** (0.351)

Dep Var: τc

Table D.6 Estimation results—expenditure taxes (annual data) 252 Appendix D: Demography and Taxation: Further Global Evidence

146 146 Cross-country Full 0.150

0.422*** (0.120)

0.0486* (0.0253) 0.909*** (0.295) 0.0401 (0.441) 0.0000832 (0.00208) 0.161* (0.0947) 0.117 (0.120) 56 56 Cross-country POLITY2 ≥ 7 0.550

−0.0763** (0.0299) −0.0375 (0.0426)

(3)

0.0736*** (0.0213) 1.029*** (0.293) 0.187 (0.383) 0.000832 (0.00215) 0.228** (0.0982) 0.103 (0.179) 45 45 Cross-country POLITY2 ≥ 8 0.631

−0.0970** (0.0365) 0.0229 (0.0467)

(4)

0.0573** (0.0222) 0.863*** (0.304) 0.111 (0.448) −0.000364 (0.00225) 0.147 (0.0994) 0.115 (0.123) 56 56 Cross-country POLITY2 ≥ 7 0.540

−0.0819** (0.0314)

(5)

0.0695*** (0.0200) 1.051*** (0.299) 0.141 (0.392) 0.00107 (0.00218) 0.233** (0.0991) 0.120 (0.169) 45 45 Cross-country POLITY2 ≥ 8 0.628

−0.0933** (0.0357)

(6)

(8)

−0.0897*** (0.0231) 0.0387* (0.0227) 0.478*** (0.143) 0.302 (0.497) 0.00150 (0.00216) 0.0356 (0.0679) −0.0316 (0.0204) 27 84 27 84 Cross-country Cross-country POLITY2 = 10 POLITY2 < 10 0.869 0.397

−0.112*** (0.0216) 0.0686*** (0.0205) 1.662*** (0.314) 0.0468 (0.345) −0.00361* (0.00208) 0.256** (0.101)

(7)

Notes Estimations contain results using cross-country OLS regressions of the composition of taxes, ln(T ), including P R O P1564, U T I P, ln(y), O EC D, T R AD E, ln(P O P), and P O L I T Y 2 as control variables, with robust standard errors in parentheses. Column (1) includes full data sample, only with ln(y) as a control; Column (2) includes full data sample, with full control variables. Columns (3) and (4) respectively correspond to different democracy levels (P O L I T Y 2 ≥ 7 and P O L I T Y 2 ≥ 8). Columns (5) and (6) again mimic columns (3) and (4) without P R O P1564 as a control. Columns (7) and (8) instead use the ratio of the population above 65 to those between the ages of 15 and 64, R AT I O, to measure population aging, respectively correspond to P O L I T Y 2 = 10 and P O L I T Y 2 < 10. *, **, and *** respectively denote significant levels at 10, 5 and 1%

Observations Countries Data Sample R2

POLITY 2

ln(P O P)

TRADE

OECD

ln(y)

UTIP

RATIO

0.0327* (0.0180) 0.783*** (0.163) 0.505 (0.341) 0.000658 (0.00156) 0.0969 (0.0612) −0.0332* (0.0200) 111 111 Cross-country Full 0.436

−0.0976*** (0.0273) −0.0589** (0.0229)

−0.112*** (0.0276)

PROP65

PROP1564

(2)

(1)

Dep Var: ln(T )

Table D.7 Estimation results—the composition of taxes (cross-country data) Appendix D: Demography and Taxation: Further Global Evidence 253

147 147 Cross-country Full 0.110

4.349*** (1.417)

0.600 (0.498) 8.101** (3.183) 5.691 (6.844) −0.0362 (0.0516) 1.108 (1.517) 3.319* (1.734) 56 56 Cross-country POLITY2 ≥ 7 0.426

−1.369** (0.603) 0.160 (0.578)

(3)

1.022** (0.437) 15.09*** (3.905) 7.643 (6.798) −0.0525 (0.0530) 1.393 (1.552) 0.653 (3.498) 45 45 Cross-country POLITY2 ≥ 8 0.509

−2.090*** (0.753) 0.432 (0.835)

(4)

0.562 (0.445) 8.297** (3.352) 5.388 (6.615) −0.0342 (0.0501) 1.165 (1.462) 3.330* (1.707) 56 56 Cross-country POLITY2 ≥ 7 0.425

−1.345** (0.584)

(5)

0.944** (0.405) 15.49*** (4.132) 6.781 (6.940) −0.0481 (0.0521) 1.498 (1.577) 0.973 (3.337) 45 45 Cross-country POLITY2 ≥ 8 0.505

−2.019*** (0.739)

(6)

(8)

−0.930*** (0.311) 0.291 (0.335) 3.790** (1.832) 0.496 (5.879) 0.0396 (0.0296) 1.763** (0.828) 0.523* (0.281) 27 85 27 85 Cross-country Cross-country POLITY2 = 10 POLITY2 < 10 0.707 0.239

−2.046*** (0.528) 1.029* (0.553) 24.19*** (5.176) 8.585 (6.872) −0.118** (0.0573) 0.644 (1.954)

(7)

Notes Estimations contain results using cross-country OLS regressions of income taxes, τ y , including P R O P1564, U T I P, ln(y), O EC D, T R AD E, ln(P O P), and P O L I T Y 2 as control variables, with robust standard errors in parentheses. Column (1) includes full data sample, only with ln(y) as a control; Column (2) includes full data sample, with full control variables. Columns (3) and (4) respectively correspond to different democracy levels (P O L I T Y 2 ≥ 7 and P O L I T Y 2 ≥ 8). Columns (5) and (6) again mimic columns (3) and (4) without P R O P1564 as a control. Columns (7) and (8) instead use the ratio of the population above 65 to those between the ages of 15 and 64, R AT I O, to measure population aging, respectively correspond to P O L I T Y 2 = 10 and P O L I T Y 2 < 10. *, **, and *** respectively denote significant levels at 10, 5 and 1%

Observations Countries Data Sample R2

POLITY 2

ln(P O P)

TRADE

OECD

ln(y)

UTIP

RATIO

0.194 (0.309) 6.620*** (1.846) 7.656 (4.713) 0.0155 (0.0248) 1.870** (0.823) 0.485* (0.274) 112 112 Cross-country Full 0.307

−1.372*** (0.424) −0.407 (0.326)

−0.585 (0.365)

PROP65

PROP1564

(2)

(1)

Dep Var: τ y

Table D.8 Estimation results—income taxes (cross-country data) 254 Appendix D: Demography and Taxation: Further Global Evidence

(2)

(3)

−2.525** (1.000)

−1.044 (0.847)

0.507* (0.294)

111

ln(P O P)

POLITY 2

Cross-country

Full

0.056

Countries

Data

Sample

R2

Cross-country

0.221

Full

45

0.463

0.619

Cross-country POLITY2 ≥ 9

Cross-country

35

35

3.354 (3.612)

−2.773** (1.335)

−0.00373 (0.0201)

−3.487 (4.175)

−9.850*** (3.076)

−0.913*** (0.283)

-0.438 (0.615)

POLITY2 ≥ 8

45

(4) 0.868* (0.503)

(5)

0.463

POLITY2 ≥ 8

Cross-country

45

45

0.173 (1.775)

−2.521** (0.954)

−0.0394 (0.0314)

−2.787 (3.679)

−7.985** (3.339)

−0.793** (0.344)

0.309 (0.360)

(6)

0.610

POLITY2 ≥ 9

Cross-country

35

35

2.672 (3.486)

−2.942** (1.305)

−0.00992 (0.0198)

−3.364 (4.409)

−9.743*** (3.266)

−0.839** (0.314)

0.860* (0.485)

0.733

POLITY2 = 10

Cross-country

27

27

−3.371*** (1.100)

0.00990 (0.0197)

1.783 (4.634)

−14.83*** (3.894)

−0.703* (0.363)

0.543* (0.314)

(7)

0.186

POLITY2 < 10

Cross-country

84

84

0.454 (0.298)

−0.0294 (1.058)

−0.0328 (0.0294)

−0.763 (7.102)

−3.142* (1.849)

−0.281 (0.372)

0.614* (0.353)

(8)

Notes Estimations contain results using cross-country OLS regressions of expenditure taxes, τc , including P R O P1564, U T I P, ln(y), O EC D, T R AD E, ln(P O P), and P O L I T Y 2 as control variables, with robust standard errors in parentheses. Column (1) includes full data sample, only with ln(y) as a control; Column (2) includes full data sample, with full control variables. Columns (3) and (4) respectively correspond to different democracy levels (P O L I T Y 2 ≥ 8 and P O L I T Y 2 ≥ 9). Columns (5) and (6) again mimic columns (3) and (4) without P R O P1564 as a control. Columns (7) and (8) instead use the ratio of the population above 65 to those between the ages of 15 and 64, R AT I O, to measure population aging, respectively correspond to P O L I T Y 2 = 10 and P O L I T Y 2 < 10. *, **, and *** respectively denote significant levels at 10, 5 and 1%

147

147

Observations

111

−0.0396 (0.0338)

−0.0281 (0.0227)

TRADE

0.162 (1.852)

−2.756 (3.849)

−2.458 (4.306)

ln(y)

OECD

−0.791** (0.368)

0.0154 (0.522)

0.307 (0.394)

−7.999** (3.375)

−0.322 (0.305)

0.793** (0.369)

0.295 (0.420)

−6.691*** (2.139)

UTIP

RATIO

−2.566 (1.653)

0.953** (0.369)

PROP65

PROP1564

(1)

Dep Var: τc

Table D.9 Estimation results—expenditure taxes (cross-country data)

Appendix D: Demography and Taxation: Further Global Evidence 255

256

Appendix D: Demography and Taxation: Further Global Evidence

in support of the Razin et al. (2002) hypothesis, whilst expenditure taxes as a proportion of total revenue increase with aging. The empirical results hold across various econometric specifications employed. In particular the fact that the results found hold significantly in countries with strong democratic credentials is supportive of the mechanism proposed in this chapter. References Benabou R (1996) Inequality and growth. Nat Bureau Econ Res Macroeconomics Ann 11:11–74 Besley T, Persson T (2014) Why do developing countries tax so little? J Econ Perspect 28(4):99–120 Galbraith JK, Kum H (2005) Estimating the inequality of household incomes: a statistical approach to the creation of a dense and consistent global data set. Rev Income Wealth 51(1):115–143 Meltzer AH, Richard SF (1981) A rational theory of the size of government. J Polit Econ 89(5):914–927 Perotti R (1996) Growth, income distribution, and democracy: what the data say. J Econ Growth 1(2):149–187 Pickering A, Rajput S (2018) Inequality and the composition of taxes. Int Tax and Publ Fin 25(4):1001–1028 Razin A, Sadka E, Swagel P (2002) The aging population and the size of the welfare state. J Polit Econ 110(4):900–918 Rodrik D (1998) Why do more open economies have bigger governments? J Polit Econ 106(5):997–1032 Shelton CA (2008) The aging population and the size of the welfare state: is there a puzzle? J Publ Econ 92(3–4):647–651