INDIA POLICY FORUM VOLUME 15 2018 ` 1395 789353 287191 9 ISBN 978-93-532-8719-1 Contents I

é 2018 15 EDITED BY BOSWORTH SHEKHAR SHAH, BARRY MURALIDHARAN KARTHIK Jain Sajjid Z. Chinoy and Toshi ExportsWhat and What Explains Drives India’s the Recent Slowdown? Evidence and New Implications Policy Dean Spears ClimateVulnerability Quantifying India’s Moore and Charity Troyer Rohini Pande, Erin K. Fletcher, Descriptive Evidence and a in India: Work and Women Policies Review of Potential Impact of Tax Breaks on Household Financial Breaks Saving in India of Impact Tax Jain Dhruv Ahmad, Florian Blum, and Junaid Gupta, Poonam Growth Story India’s and Mark Budolfson, Kuruc, Kevin Melissa LoPalo, Radhika Pandey, Ila Patnaik, and Renuka San Ila Patnaik, Radhika Pandey, VOLUME

NCAER INDIA POLICY FORUM VOLUME 15 2018 ` 1395 789353 287191 9 ISBN 978-93-532-8719-1 Contents i

15 2018

EDITED BY Shekhar Shah Barry Bosworth Karthik Muralidharan

NATIONAL COUNCIL OF APPLIED ECONOMIC RESEARCH New Delhi ii INDIA POLICY FORUM, 2010–11

First published in 2019 by

SAGE Publications India Pvt Ltd B1/I-1 Mohan Cooperative Industrial Area Mathura Road, New Delhi 110 044, India www.sagepub.in

SAGE Publications Inc 2455 Teller Road Thousand Oaks, California 91320, USA

SAGE Publications Ltd 1 Oliver’s Yard, 55 City Road London EC1Y 1SP, United Kingdom

SAGE Publications Asia-Pacific Pte Ltd 18 Cross Street #10-10/11/12 China Square Central Singapore 048423

Library of Congress Serial Publication Data applied for

Disclaimer: The papers and the associated comments represent the views of the individual authors and do not imply agreement by the Governing Body, officers, or staff of NCAER or by members of the IPF panels.

ISSN: 0973-4805 ISBN: 978-93-532-8719-1 (PB)

All rights reserved. No part of this book may be reproduced or utilized in any form or by any means, electronic or mechanical, including photocopying, recording, or by any information storage or retrieval system, without permission in writing from the publisher.

Published by Vivek Mehra for SAGE Publications India Pvt Ltd, typeset in 10.5/13 pt Times by AG Infographics, Delhi, and printed at Avantika Printers, New Delhi. Contents iii

15 2018

Editors’ Summary ix

Radhika Pandey, Ila Patnaik, and Renuka Sané Impact of Tax Breaks on Household Financial Saving in India 1 Comments by Rajnish Mehra and M. Govinda Rao 28 General Discussion 34

Poonam Gupta, Junaid Ahmad, Florian Blum, and Dhruv Jain India’s Growth Story 39 Comments by Sudipto Mundle and Dilip Mookherjee 94 General Discussion 101

Melissa LoPalo, Kevin Kuruc, Mark Budolfson, and Dean Spears Quantifying India’s Climate Vulnerability 107 Comments by Navroz K. Dubash and Shreekant Gupta 139 General Discussion 146

Erin K. Fletcher, Rohini Pande, and Charity Troyer Moore Women and Work in India: Descriptive Evidence and a Review of Potential Policies 149 Comments by Pranab Bardhan and Farzana Afridi 199 General Discussion 213

Sajjid Z. Chinoy and Toshi Jain What Drives India’s Exports and What Explains the Recent Slowdown? New Evidence and Policy Implications 217 Comments by Surjit Bhalla and Kenneth Kletzer 244 General Discussion 255 iv INDIA POLICY FORUM, 2010–11 Contents v

PURPOSE AND ORGANIZATION This 15th India Policy Forum 2018 Volume comprises papers and highlights of the discussions at the India Policy Forum (IPF) held in New Delhi on July 11–12, 2018. The IPF is organized by NCAER, the National Council of Applied Economic Research, India’s oldest and largest, independent economic think tank. The IPF promotes economic policy and original empirical research on India. It commissions original papers and policy-focused expert reviews, the latter usually based on robust, original research. It provides a unique combination of intense scholarship and policymaker engagement at the IPF Conference, held every year in July in New Delhi, leading to this international journal. The IPF in that sense is a lot like an open editorial board meeting, one of its unique characteristics. An international Research Panel of India-based and overseas scholars with an abiding interest in India supports the IPF with advice, active conference participation, and the search for innovative papers that promise fresh insights. An international Advisory Panel provides overall guidance. Members of the two IPF panels are listed below. Papers appear in the annual IPF Volume after revisions based on IPF discussants’ comments, a lively floor discussion, and the editorial guidance provided by the editors of the India Policy Forum Volume. To allow readers to get a sense of the richness of the conversations that happen at the IPF, edited discussants’ comments are included here, as is a summary of the floor discussion on each paper. The IPF also features an annual, invited lecture. The Annual IPF Lecture for 2018 was delivered on July 10 by Professor Avinash Dixit, Professor Emeritus, Princeton University. In recent years, the IPF has featured an IPF Policy Roundtable that allows timely discussion of a topic of current policy relevance: this year’s topic was “India’s Healthcare Reforms: Getting to Health for All.” In 2018, the IPF added a concluding roundtable discussion on the policy and research ideas emerging from the conference. In celebration of its 15th Anniversary, the 2018 IPF also featured an extended conversation on July 12 with the outgoing Chief Economic Adviser to the Government of India. Full videos of all IPF sessions, the Annual IPF Lecture, the two roundtables, and the special 15th anniversary event are available on NCAER’s website (www.ncaer.org). The 2018 IPF program, presenta- tions, and videos can be viewed by scanning this SPARQ code: vi INDIA POLICY FORUM, 2010–11

ADVISORY PANEL* Shankar N. Acharya Indian Council for Research on International Economic Relations Viral V. Acharya Reserve Bank of India Isher J. Ahluwalia Indian Council for Research on International Economic Relations Montek S. Ahluwalia Former Planning Commission of India Pranab Bardhan University of California, Berkeley Jagdish Bhagwati Columbia University & Council for Foreign Relations and NCAER Barry Bosworth Brookings Institution Willem H. Buiter Citigroup Stanley Fischer Former Board of Governors of the U.S. Federal Reserve System Vijay Kelkar NIPFP and India Development Foundation Mohsin S. Khan Atlantic Council Anne O. Krueger SAIS, Johns Hopkins University Ashok Lahiri 15th Finance Commission Rakesh Mohan Yale University Arvind Panagariya Columbia University Raghuram Rajan University of Chicago Booth School of Business Urjit R. Patel Former Reserve Bank of India Shekhar Shah NCAER Nicholas Stern London School of Economics & Political Science Lawrence H. Summers Harvard University

RESEARCH PANEL* Abhijit Banerjee Massachusetts Institute of Technology Kaushik Basu Cornell University and NCAER Surjit S. Bhalla Oxus Investments Pvt. Ltd and NCAER Sajjid Z. Chinoy J.P. Morgan Mihir Desai Harvard Business School Shantayanan Devarajan World Bank and NCAER Esther Duflo Massachusetts Institute of Technology Subir Gokarn Brookings Institution India Center Maitreesh Ghatak London School of Economics Jeffrey S. Hammer Princeton University and NCAER Vijay Joshi Merton College, Oxford Contents vii

Devesh Kapur SAIS, Johns Hopkins University Kenneth M. Kletzer University of California, Santa Cruz K. P. Krishnan Government of India Robert Z. Lawrence Harvard Kennedy School Rajnish Mehra Arizona State University Dilip Mookherjee Boston University Karthik Muralidharan University of California, San Diego and NCAER Ila Patnaik NIPFP Lant Pritchett Harvard Kennedy School Indira Rajaraman Former 13th Finance Commission Tarun Ramadorai Imperial College, London and NCAER M. Govinda Rao Former14th Finance Commission, NIPFP, and NCAER Ajay Shah NIPFP Nirvikar Singh University of California, Santa Cruz Rohini Somanathan Delhi School of Economics Arvind Subramanian Harvard Kennedy School Arvind Virmani Chintan

*All affiliations are as of April 2019.

PARTNERS NCAER gratefully acknowledges the generous support for IPF 2018 from HDFC Ltd, Reliance Industries Ltd, State Bank of India, and Citibank N.A. Their support reflects the deep commitment of the enlightened leadership of these organizations to rigorous, independent, economic policy research that helps promote more informed policy debates and evidence-based policymaking in India.

CORRESPONDENCE Correspondence about papers in this IPF Volume should be addressed directly to the authors (each paper contains the e-mail address(es) of the corresponding author(s)). All author affiliations in the papers are as of the IPF Conference. Unsolicited manuscripts are not accepted for review because all papers are commissioned. Feedback on the IPF Volume may be sent to: The Editors, India Policy Forum, NCAER, 11 Indraprastha Estate, New Delhi 110002, or by e-mail to [email protected]. More information on the IPF is available on www.ncaer.org. viii INDIA POLICY FORUM, 2010–11

THE IPF TEAM NCAER is responsible for development, planning, organization, and publication for the India Policy Forum. The Editors and the IPF Panels are deeply grateful to the following NCAER staff for their dedication and hard work on the 2018 IPF:

Anwesha Pandey (until Jan. 2019) Special Assistant to the Director General Namrata Ramachandran Special Assistant to the Director (from Jan. 2019) General Sudesh Bala Team Lead and overall coordination Anupma Mehta Editing Jagbir Singh Punia Publication and production Praveen Sachdeva Conference production and graphics P. P. Joshi Hospitality and logistics Sangita Chaudhary Team assistance Khushvinder Kaur Team assistance Editors’ Summary

he India Policy Forum (IPF) marked its 15th Anniversary with its Tconference in New Delhi on July 10–11, 2018. The primary goal of the IPF is to promote original policy and empirical research on India. The annual IPF Conference provides a unique combination of intense scholarship and commentary on the research and a focus on its policy implications. The revised papers are published in this journal and benefit from a wide international readership. Over the years, interest in India has grown to the point where there is now much more original research on India appearing in international economic journals than was the case when the IPF started. The IPF has also changed, making room for more policy-focused review articles that define the best policy advice based on the most robust, empirical research available. The IPF has for some years now also featured a roundtable discussion of a key policy challenge dominating the discourse around the time of the IPF. In 2017, we also added a concluding roundup discussion to gather the key policy and research ideas emerging from the two days of the IPF. In celebration of the 15th Anniversary of the IPF, the 2018 IPF featured an extended conversation between the outgoing Chief Economic Adviser to the Government of India, Arvind Subramanian, and one of the three IPF editors. This journal edition of the India Policy Forum contains the five 2018 IPF Conference papers, the paper discussants’ comments, and a summary of the floor discussion. Videos of the entire Conference, including the two panel discussions, the Annual IPF Lecture, and the special 15th Anniversary event are available on www.ncaer.org by scanning the QR Code at the end of this Editors’ Summary.

Impact of Tax Breaks on Household Financial Saving in India

Indian households tend to hold a large fraction of their wealth in non-financial assets such as real estate and gold. Tax policy has been used as a lever in India for incentivizing saving in financial assets as well as encouraging saving for the long term. This is done through tax breaks for specified financial products such as fixed deposits, small saving instruments, pension and provident funds, and insurance through Section 80C of the Income Tax Code.

ix x INDIA POLICY FORUM, 2018

International evidence broadly suggests that tax breaks of specific financial products lead only to a substitution effect—households shift investments to those assets that qualify for a tax break. Several direct tax government committees in India—the Shome Committee in 2001, the Kelkar Committee in 2002, and the Malegam Committee in 2015—have raised similar concerns about the structure of tax breaks in India. Empirical work in India on this question is largely absent. This is mainly because empirical analysis of these questions is best done using survey data that studies investor-level decisions over time, and requires some exogenous variation in eligibility (or some other instrument) to clearly identify the effect of the tax break. Unfortunately, such Indian microdata have not been available. This paper by Radhika Pandey, Ila Patnaik, and Renuka Sané for the first time presents micro as well as macro evidence on what tax breaks have done for financial saving. They ask questions such as: Is it that overall financial saving has increased? Or that the tax policy has channeled household saving into specific products? For a macro level analysis, this paper uses aggregate national accounts data to study how financial saving has evolved with changes in tax breaks. It enumerates the changes since 2001 in the taxation of savings instruments for individuals and their impact on financial saving. For a micro perspective, the paper studies household portfolios for the financial year 2016–2017 using the CMIE Consumer Pyramids Household Survey (CPHS). This allows the authors to compare households that fall under the tax bracket with households that do not. The paper finds that financial saving rose steadily from about 11 percent of GDP in 2001 to a high of about 17 percent in March 2007. The years between 2003 and 2005 did not see any tax breaks, and yet there was a rise in financial saving. Since 2007, there has been a tax break announced almost every year. And yet, financial saving had fallen to about 11 percent of GDP by March 2013. The authors also look at the new GDP series and their conclusion remains the same. Financial saving as a percentage to GDP remained roughly constant through the tax breaks of 2012 to 2015 and had increased only slightly as of March 2016, which may be because of the increase in the overall tax exemption limit from `100,000 to `150,000 in the 2014–15 budget. The association between tax breaks and financial saving appears weak from the national income data. Considering the share of various savings instruments in overall household financial saving, the paper finds that bank deposits constitute the bulk of household financial saving, though its share in the overall household financial savings has seen a dip from 2011–12 onward. Investments in provident Shekhar Shah, Barry Bosworth, and Karthik Muralidharan xi and pension funds have seen a gradual rise, though they still constitute a small proportion of overall saving. Looking to household financial portfolios from the micro data, the paper finds a big difference between those households that are taxed versus those that are not taxed. In 2016–17, a much larger proportion of taxed households claimed to have outstanding investments in fixed deposits, insurance, small savings, and pensions, all of which are instruments covered under Section 80C. For example, 88 percent of the households that are not taxed claim to have outstanding investments in fixed deposits, as opposed to 96 percent of taxed households. The next instrument of choice is insurance with 50 percent of non-taxed households possessing insurance, as opposed to 90 percent of taxed households, followed by provident/pension funds, where the difference is much more stark—7 percent of non-taxed households as opposed to 55 percent of taxed households. To control for the income effect as distinct from the tax effect, the paper further compares households that have the same household income, but differ in the tax treatment because at least one individual in the household has income that makes her fall in the tax bracket. A disaggregated analysis of salaried households suggests that the impact of tax incentives on saving is highest for salaried households in the income bracket of `350,000 to `500,000. Those who fall under the tax bracket in the `350,000–`400,000 income category are almost 7 percent more likely to have invested in insurance relative to those who do not fall under the tax bracket. Those taxed in the `400,000–`450,000 income category are 5 percent more likely to have invested in insurance. The results suggest that tax breaks do not have an impact on increasing overall financial saving, and at most lead to a substitution effect, that is, they lead to households channeling saving into the tax-exempt products without increasing the amount of overall saving. The authors recommend that these considerations should influence the design of tax policy.

India’s Growth Story

India has attained much economic success in the last three decades. Yet an economic deceleration in recent years has generated worried commentaries about India’s growth outlook. This paper by Poonam Gupta, Junaid Ahmad, Florian Blum, and Dhruv Jain offers a long-term perspective on India’s growth experience. Looking back at the last 50 years, the authors analyze India’s long-term growth patterns in different ways, and compare India’s growth experience with that of other large, emerging market economies. xii INDIA POLICY FORUM, 2018

The authors note the following stylized facts. First, India’s long-term economic performance has been impressive. Despite variations around the long-term growth rate, average growth over any continuous 10-year period has steadily accelerated and has never reversed for a prolonged period. Economic growth has also become more stable—both due to growth rates stabilizing within each sector, and due to the transition of the economy toward the services sector, which has a more stable growth rate. Second, the long-term growth experience has been balanced and diversified in the sense that acceleration and stability are evident across the India states; and for the most part, growth is not concentrated in a few sectors. Growth acceleration has been characterized by productivity gains, and not just by an increase in factor inputs. Third, assessing the period since the early 1990s more closely, the authors note three distinct phases of growth: a period of slow acceleration from 1991 to the early 2000s; a short period of rapid growth, with certain features of unsustainability, during 2004–08; and a corrective slowdown that started with the Global Financial Crisis (GFC) in 2008. The period of growth acceleration was marked by a rapid increase in the rate of investment, financed by high credit growth and a surge in capital flows, while the slowdown has reflected most profoundly in investment, credit, and exports. Fourth, even as the economy has slowly reverted to the trend growth rate and stabilized in recent years, the revival is not yet firmly anchored in investment, exports, and the industrial sector. Recovery in investment and credit has been more protracted in India than in other countries, and India has lost share in the global export market. This may have implications for accelerating growth to India’s potential, and for enhancing the potential growth itself. The authors maintain that the economy seems to have settled down to a 7–7.5 percent growth rate. Attaining a growth rate of 8 percent or higher on a sustained basis will require contributions from all domestic sectors and support from the global economy. It will also require a concerted reform and policy momentum, wide enough in scope, which succeeds in reversing the slowdown in investment, credit supply, and exports. Private investment in India is constrained by several factors. The paper points to issues related to past leverage as well as subdued market demand. Going forward, derisking the private sector is important, as also the need to ensure an environment of policy certainty. Understanding and relieving the generic, spatial, and sector-specific constraints to investment growth will be important. Banks dominate the Indian financial system, and public sector banks dominate the banking sector. The last few years have been challenging, as Shekhar Shah, Barry Bosworth, and Karthik Muralidharan xiii credit growth remains subdued and the stress on asset quality continues. Bank credit growth has consistently declined since the GFC, after increasing briskly for a few years before that, and the ratio of gross non-performing assets to advances of public sector banks has increased. Under its current balance sheet situation, the banking sector does not seem well equipped to help finance a higher growth or investment rate, and suitable reforms will be needed to reverse this equilibrium. Reconsidering the ownership balance and incentive and governance struc- tures in banking may be important to improve the allocative and operational efficiency of the sector. The government is exploring different options to resolve the problem of high non-performing loans. Some of these measures, such as mergers within the public sector banks, have been used earlier, while others are more novel in the Indian context, including setting up a bad bank, or the aggressive use of bankruptcy procedures in loan recovery. It would be useful to consider the merits of these options in the light of cross-country experiences. Other issues that need specific attention are building risk assessment capabilities within the regulator and the banks, and developing and strengthening the personal bankruptcy framework. While private investment is likely held back primarily by domestic factors, export growth is constrained by both domestic and external factors. Export growth was an important driver of GDP growth prior to the GFC, but its contribution to growth has diminished since. India has barely managed to keep pace with the growth in world exports since the GFC, reflected in its stagnant or even declining share of world exports, and a declining export- to-GDP ratio. The factors that may help India improve its competitiveness include an infrastructural boost to bring it at par with other manufacturing hubs of the world, and reforms in land, labor, and financial markets and in the educational system to assure the continued competitive supply of key production inputs such as labor, land, finance, and skills. In addition, the authors suggest that issues related to a competitive exchange rate, enhancing bilateral and regional trade integration, evading the temptation to cave in to the rhetoric on trade protectionism, and embedding more deeply in global value chains all have great significance and require objective discussion and assessment. The authors conclude that there seems to be only limited room to ease fiscal, monetary, or exchange rate policies to boost GDP growth in the midst of complex and persistent structural constraints. Given the structural nature of weak exports and investments, the effectiveness of transitory countercyclical policies is likely to be limited. Even if used, these can provide only temporary reprieve, as by their very nature, countercyclical policies ought to be used temporarily and should be reversed within a reasonable period of time. xiv INDIA POLICY FORUM, 2018

Quantifying India’s Climate Vulnerability

Global average temperatures are projected to rise significantly over the next century, causing the frequency of extreme temperature days to increase, among many other impacts. Global warming can be mitigated through reduced carbon emissions today. However, especially for a developing country with many poor people such as India, the economic cost of reduced emissions may have large welfare effects. In order to make decisions about India’s best response to the current global trajectory, policymakers need information about the size of the damages likely to befall future Indians under different levels of warming. Especially in developing countries, evidence on the magnitude of the population’s vulnerability to climate change is scarce. This is particularly problematic since developing countries are likely to be disproportionately vulnerable to changes in climate: baseline temperatures are warmer on average, and poorer people are less able to protect themselves through adaptive technology. In their paper, Melissa LoPalo, Kevin Kuruc, Mark Budolfson, and Dean Spears examine, quantify, and discuss India’s vulnerability to climate change. They begin by reviewing recent microeconometric evidence of the impacts of extreme temperature—and importantly its interaction with humidity—on several economic outcomes. The recent literature in climate economics seeks to identify the causal effects of extreme weather, netting out associations between climate and other determinants of well-being. To achieve this, researchers generally compare outcomes on days or months with different weather at the same place, often within the same time of the year. This leads to questions like: Was infant mortality different in Uttar Pradesh in a June that was warmer than the average Uttar Pradesh June? Papers using this strategy find detrimental effects of extreme heat on a wide variety of variables such as conflict, health, and productivity. Recent work in this field by the authors suggests that the effects of extreme temperature days are significantly worse in areas of the world that are also humid, since humidity impairs the body’s ability to cool itself through sweating. The authors maintain that these findings have serious, and until recently under- appreciated, implications for the effects of climate change in the humid region of South Asia. To provide a sense of the magnitudes involved, the authors apply estimated effect sizes on infant mortality, labor productivity, and GDP to present and project temperature and humidity data from India. They show that the hot, humid areas of North India are particularly vulnerable to these effects and that the impacts will grow dramatically more severe Shekhar Shah, Barry Bosworth, and Karthik Muralidharan xv if climate change is allowed to unfold unabated. This geographic area contains the particularly populous states of Uttar Pradesh and Bihar, where the population may be already particularly vulnerable due to low incomes and poor baseline health. The authors then move to aggregating the economic consequences of temperature using a macroeconomic model. Specifically, an Integrated Assessment Model (IAM), originally built by William Nordhaus, is modified to predict the impacts of various warming scenarios on Indian well-being. To present total damages in a politically relevant metric, an equivalently bad “near-term” economic disaster is solved for within the model (i.e., how bad would a near-term economic collapse need to be to match the aggregate damage of climate change?). IAMs allow for such a quantification of climate damages by summing the well-being of multiple generations, allowing climate change to harm future generations through a “damage function” and incorporating the near-term consumption benefits provided by carbon- emitting technologies. These calculations rely on trade-offs across people living at different times, and with different levels of incomes. The authors draw on an active literature at the intersection of economics and philosophy to discipline their choice of modeling parameters, providing sensitivity tests along the dimensions of uncertainty that remain. The authors find that under a “business as usual scenario,” the economic damage from climate change will be as large as a 29 percent decline in per-capita GDP for 20 consecutive years. However, there are significant returns to reducing the amount of realized warming. Restricting the amount of warming to 3.5 degrees, a scenario in which the Paris Accord pledges are successfully implemented, would reduce these climate damages for India by over two-thirds. The authors re-run these projections under the counterfac- tual scenario in which India immediately stops emitting CO2 to understand the gains from unilateral abatement. The damages do not shrink by much, so that India’s vulnerability requires a global solution. The paper suggests that there is one way in which India could be helping itself through unilateral CO2 mitigation: the averted damages from air pollution that would result from reducing greenhouse gas emissions. The authors review the emerging literature in economics that has found the health effects of air pollution to be substantial. This consensus is particularly relevant to India, where 14 of the world’s 20 most polluted cities are located. In fact, the benefits to concurrent health are potentially large enough to justify mitigating a large share of emissions regardless of the benefits for climate change. As such, these health co-benefits will ultimately play a crucial role in deciding the optimal level of emissions reduction for India. xvi INDIA POLICY FORUM, 2018

The authors conclude with a discussion of the challenging political situation facing India. Given the magnitude of the problem and India’s inability to avert damages through unilateral abatement, the authors suggest that this should be a foreign policy priority. One promising option they suggest is to reduce India’s emissions to the level that would be optimal for Indians considering only the benefits from reduced air pollution, perhaps focusing on substituting away from coal. In this way, India might take leadership through its proactive reduction of emissions, giving it leverage to bring other large emitters to the table. Another possibility is to craft a creative concession, economic or otherwise, that India would be willing to trade for reduced emissions on the part of other global actors. Making such a concession would be deeply unfair to India, but given the grave vulnerability of its population to the ill-effects of climate change and current global inaction, the authors suggest it may very well be India’s best response.

Women and Work in India: Descriptive Evidence and a Review of Potential Policies

In spite of its rapid economic growth, education-related gains, and a significant fertility decline, India’s women are conspicuously absent from the labor force, with an overall female labor force participation (FLFP) rate of 28 percent or lower. Worldwide, economic growth in low- to middle-income countries is associated with women opting out of manual, low-paying jobs, while economic growth in middle- to high-income countries is associated with opting back into the labor force as suitable, higher paying jobs become increasingly available. India’s experience is starkly different: FLFP1 rates remain low and have even fallen in recent years, and FLFP is far lower than in countries with similar income per capita. Besides being a potential source of inefficiency, women’s limited economic engagement may have important ramifications for women’s education, timing of marriage and childbearing, and women’s decision-making power in the household. This paper by Erin Fletcher, Rohini Pande, and Charity Troyer Moore, examines possible constraints on FLFP in India and potential policy interventions that could increase it. Since observed FLFP levels reflect both supply and demand factors, determining causation, and thus the appropriate

1. The authors calculate the LFP rate by dividing the number of individuals in the working age population (ages 15–70) employed in wage labor, own-account work, casual labor, unpaid labor, self-employment, or as an employer, plus those unemployed and seeking work by the entire working-age population (15–70 years) not currently enrolled in school. Shekhar Shah, Barry Bosworth, and Karthik Muralidharan xvii prioritization of policy responses, is difficult. That said, the authors maintain that a descriptive understanding of potential factors constraining FLFP is a necessary first step to designing and evaluating pilot reforms to remove barriers. To this end, the paper utilizes the 68th Round of the National Sample Survey (NSS) to highlight five features of Indian women’s market engagement. Alongside this descriptive analysis, the paper examines evidence from recent research that seeks to provide causal estimates of policies and other factors relevant to increasing FLFP in India. The authors suggest, first, that a large proportion of Indian women counted out of the labor force express willingness to take on work. If all these women who stated they would take work actually did, we would see a 21 percentage point rise in the FLFP rate, substantial, given the low rates of participation overall. Women currently out of the labor force who are willing to take a job tend to be more educated, slightly more likely to live in rural areas, and not SC or ST. Further, almost 45 percent of rural, educated women who report their primary activity as domestic duties also report that they would accept work. Second, women have more trouble matching up to jobs than men. They report seeking or being available for jobs longer than men when unemployed, and women who did work reported spending more time unemployed than men. Part of this delay may reflect supply–demand mismatches: women counted out of the labor force who said they would take on work in the NSS generally preferred regular, part-time work, in contrast to the largely full- time work undertaken by women already in the labor force. Third, wage gaps and unexplained wage gaps—typically interpreted as evidence of gender-based discrimination in the labor market—are relatively higher in fields with greater female representation. However, women are seeking out relatively better and higher employment, evidenced by recent growth in women’s relative representation in manufacturing, a sector with relatively lower gender gaps, which could reflect responsiveness to these wage gaps. Fourth, at all levels of education, women who have completed some vocational training are more likely to work than those who have not. Latent female workers interviewed in the NSS also identified significant skills gaps in areas in which they would like to work. While the skilling–LFP relationship is clearly not causal, it highlights an interesting possibility worth further investigation, inasmuch as training programs may help women overcome both skills gaps and job-matching frictions. Finally, women are doing relatively well in terms of representation in specific jobs, namely, education and work provided by the government’s job guarantee program, the Mahatma Gandhi National Rural Employment xviii INDIA POLICY FORUM, 2018

Guarantee Scheme (MGNREGS). Both sectors have specific features that may increase demand for female labor, and MGNREGS has additional features aimed to attract women workers. Their relative success suggests that further study here is useful. The paper’s review of causally identified research examines features that may increase FLFP in India and is consistent with the descriptive evidence: a variety of constraints keep women from participating fully in the labor force, including information about jobs, norms, women’s lack of autonomy, and lack of access to part-time work and local work outside of agriculture. From their review, the authors identify effective methods to alleviate these constraints and encourage more women to join the labor force. Taken together, the descriptive analysis and evidence review suggest key areas to focus research inquiry, some also overlapping with the Government of India’s policy priorities. These include access to local work, perhaps even through supplementing local public sector employment. Government programs such as MGNREGS, and the National Rural Livelihoods Mission that aims to connect SHG members to access to training and local work, are also promising. Additional understudied, but policy-relevant, areas relevant to FLFP include women’s time use, household production technologies, safety, and the role of income transfers on women’s economic engagement. The authors urge that in addition to undertaking research focused on policy solutions to India’s low FLFP, increasing the frequency of data collected about Indian women’s economic activities and time use, improving data collected relevant to government initiatives that can influence FLFP, and ensuring that data are released regularly and transparently are additional important steps. Raising women’s visibility in the labor force through improved data collection and use can catalyze important dialogue and initiatives aimed to engage women more actively in the economy.

What Drives India’s Exports and What Explains the Recent Slowdown? New Evidence and Policy Implications

The paper maintains that the role of exports in India’s growth dynamics over the last two decades has been consistently under-appreciated. India’s export surge in the mid-2000s—when exports grew at almost 18 percent a year in real terms in the mid-2000s causing the exports-to-GDP ratio to more than double between 1999 and 2013—was a key driver of the high GDP growth of that era. Shekhar Shah, Barry Bosworth, and Karthik Muralidharan xix

In contrast, exports have collapsed in recent years, growing at less than 4 percent between 2012 and 2017, before staging some recovery in 2018. Furthermore, GDP growth has suffered a double whammy. Not only has gross export growth abated sharply, but the “domestic value add” of India’s exports has progressively fallen over the last decade. This should not be unexpected. As India integrates into the global economy, and slowly starts getting absorbed into global value chains, the domestic content per unit of exports should be expected to fall. Normally, this is more than offset by the “scale” benefit of integrating into global value chains. But India has still not benefited from the latter. This double whammy—lower gross export growth and the reduced domestic value-add per unit of exports—has meant that the fall in export growth can explain almost the entire GDP slowdown in recent years, compared to the mid-2000s. Given this macro backdrop, the question is: Why have exports slumped in recent years? There are several competing explanations. First, global growth has slowed, compared to the mid-2000s, and trade linkages have become far more tenuous (“deglobalization”) which could be responsible for the slowdown. Second, India’s broad trade-weighted exchange rate (36-country REER) appreciated almost 20 percent between 2014 and 2017, and ostensibly hurt export competitiveness. Third, India witnessed successive (but presumably transient) supply shocks in the form of GST and demonetization in 2016 and 2017. All these factors could have contributed to the export slowdown. This paper by Sajjid Chinoy and Toshi Jain thus seeks to answer three questions in disentangling these proximate causes of the export slowdown. First, what are the determinants of India’s exports and, in particular, the “income” (global growth) and “price” (exchange rate) elasticities of exports? This is important because the existing literature throws up varied and inconclusive findings on the exchange rate elasticity of India’s exports, thereby leaving open a key policy question: How do exchange rate movements impact export competitiveness? Second, how heterogeneous are these income and price elasticities across sectors, and how have they changed across time, given the dynamism and fast-changing composition of India’s export basket? Third, how much of the recent export slowdown is attributable to deglobalization, to exchange rate appreciation, and to the twin shocks of GST and demonetization? The authors find, first, that both global growth and exchange rates are important determinants of India’s export dynamics. The findings of strong xx INDIA POLICY FORUM, 2018

“income” elasticities should be unsurprising, and would be consistent with previous research. Importantly, however, they find a large exchange rate elasticity—in contrast to previous work—which is robust across different methodologies and choice of variables. This should be a worthwhile addition to the debate on whether exchange rates do matter in determining the competitiveness of Indian exports. That said, the authors also find that both “income” and “price” elasticities have attenuated in recent years, though are still quantitatively significant. Attenuating income elasticities are at the heart of the de-globalization hypothesis and, therefore, unsurprising. Falling price elasticities are surprising, prima facie, but, the authors maintain, not once we consider the progressively lower domestic value add of India’s exports. Even accounting for this attenuation, however, exchange rate elasticities are shown to be quantitatively significant. Third, the authors find significant heterogeneity in income and price elasticities across sectors, which explains the changing composition of India’s export basket. They observe that India’s new age exports, such as software services, engineering goods, and pharmaceuticals, are found to have the highest “income elasticities.” In contrast, India’s traditional exports (textiles, leather, gems and jewelry) are found to be much less sensitive to global growth. Similarly, price elasticities of India’s new age exports are also correspondingly higher than those of traditional exports. High income and price elasticities for India’s new age exports suggest they are both discretionary (reflected in their cyclicality) as well as in highly competitive sectors (reflected in their high price elasticities). The authors maintain that their model can explain a significant fraction of the export slowdown in recent years, thereby confirming that both global demand and exchange rate dynamics have posed meaningful headwinds to exports in recent years, rather than just the presumed temporary factors associated with demonetization and the introduction of the GST. In particular, the paper finds that the sharp 20 percent appreciation of the 36-country REER between 2014 and2017 likely hurt export competitiveness. In turn, the authors propose that the real appreciation itself was the inevitable outcome of the large, positive terms-of-trade shock that India experienced from lower oil prices, suggesting that India was afflicted by the Dutch Disease. Interestingly, exports began to recover in 2018, in conjunction with some depreciation of the REER. The authors suggest several policy implications. If oil prices were to rise, and the positive terms-of-trade shock reverse, the equilibrium real exchange rate will likely depreciate further and help improve competitiveness. Shekhar Shah, Barry Bosworth, and Karthik Muralidharan xxi

Policymakers must not fight this depreciation, but simply ensure that it is calibrated and non-disruptive. Second, fiscal policy must not become expansive and offset the real depreciation engendered by a reversing terms-of-trade shock. More fundamentally, India must, through supply side reforms, seek to improve underlying external imbalances, specifically the current account deficit minus oil and gold, which have deteriorated by almost 30 percent of GDP over the last three years. Finally, with the global growth potential declining, protectionism on the rise, and income elasticities falling, India must simultaneously seek other growth drivers in the coming years.

The 2018 Annual IPF Lecture, the IPF 15th Anniversary Event, and the IPF Roundtables

Though not included in this Volume, the 2018 IPF featured the 2018 Annual IPF Lecture, delivered by Avinash Dixit, Padma Vibhushan awardee and Professor Emeritus, Princeton University. Dixit spoke on “How Can India Avoid Losing its Race to Prosperity?” The lecture was chaired by Dr Shantayanan Devarajan, Senior Director for Development Economics and the acting Chief Economist at the World Bank. Dixit began by comparing India’s economy with China’s, emphasizing that though such a comparison was usually unfair to India, the dynamics between the economies of the two countries had changed somewhat after China’s slowdown, and India’s acceleration and more favorable demo- graphics, which had raised hopes of India overtaking China. However, he cautioned that India needs to improve policymaking and capitalize on its advantages; otherwise, it would lose its race to eliminate poverty and raise the well-being of its people. He argued that the Indian demographic dividend was handicapped by various factors, including low FLFP, poor quality of education, poor transport infrastructure, inferior physical and “invisible” infrastructure, low productivity, and corruption, all of which deterred innovation and investment. He concluded the lecture with the hope that these issues would be resolved by reforming India’s institutions and policies that would transform India’s prospects. In celebration of its 15th Anniversary, the 2018 IPF also included a conversation with Arvind Subramanian, erstwhile Chief Economic Adviser, Government of India, on his four years in the Ministry of Finance as he prepared to leave the position. Karthik Muralidharan, Associate Professor, UC San Diego, and Non-Resident Senior Fellow at NCAER and one of us, the three editors of the IPF, led the conversation. xxii INDIA POLICY FORUM, 2018

Also not included in this volume, another key part of the 2018 IPF was the IPF Policy Roundtable on “India’s Healthcare Reforms: Getting to Health for All.” The Roundtable involved a detailed discussion of the ambitious Ayushman Bharat Yojana or National Health Protection Scheme, slated to provide health insurance for 500 million Indians as announced in the 2018 Budget. The Roundtable was moderated by Karthik Muralidharan, with panelists Abhijit Banerjee, MIT; Alok Kumar, Adviser, Health and Social Policy, NITI Aayog; and Jeffrey Hammer, NCAER and Princeton Woodrow Wilson School. Videos of the 2018 IPF papers, the IPF Lecture, the IPF Policy Roundtable and Roundup, and the conversation with Arvind Subramanian are all available on the NCAER website (www.ncaer.org). They can be viewed using the hyperlinks on the IPF program by scanning the accompanying SPARQ code on a smartphone: RADHIKA PANDEY* NIPFP ILA PATNAIK† NIPFP RENUKA SAN É ‡ NIPFP Impact of Tax Breaks on Household Financial Saving in India§

ABSTRACT Academic literature on the effectiveness of tax breaks on financial saving in India is scant. This paper, for the first time, presents macro as well as micro evidence on what has been achieved by tax breaks with regards to financial saving. It uses aggregate national accounts data to study how financial savings have evolved with changes in tax breaks. It studies household portfolios for the financial year 2016–17 using the Consumer Pyramids household survey. It finds that there is no link between tax breaks and overall financial savings. Households that fall in a non-zero tax bracket invest more heavily in the tax-incentivized products. These results have implications for the design of tax policy.

Keywords: Tax Policy, Tax Incentives, Saving, India

JEL Classification: E2, H2, H31

1. Introduction

ousehold saving1 in India was 32 percent of GDP in 2015–16 (RBI H2017b). However, this has not translated into financial investments. Net financial saving by households was only 7.8 percent of Gross National

* [email protected] † [email protected] ‡ [email protected] § The authors gratefully acknowledge comments from Barry Bosworth on an early draft of the paper. They have benefited immensely from comments from participants at the NCAER 2018 India Policy Forum, particularly from the chair and discussants, Arbind Modi, Rajnish Mehra, and M. Govinda Rao. 1. To distinguish between the use of the term as stock and flow, we use the term ‘saving’ to denote flow and ‘savings’ to denote cumulative stock. 1 2 INDIA POLICY FORUM, 2018

Disposable Income (GNDI) in the same year.2 Indian households, on average, tend to hold a high fraction of non-financial assets, with particularly high relative weights in real estate and gold (Badarinza, Balasubramaniam and Ramadorai 2017). A policy lever used in India, as in other countries, to influence saving into financial markets has been tax breaks for certain specified financial products through the Income Tax Code. Tax breaks are often motivated by the need to incentivize households to invest in long-term saving instruments to build assets to finance retirement consumption in old age. For example, new vehicles for individual retirement saving were created through tax legislation in the 1970s and 1980s in the USA. In fact, tax incentives are used by most OECD governments to encourage private retirement saving. The Indian tax incentives are also motivated by the need to promote long- term saving. Assets with exemptions include certain bank deposits, small saving instruments administered by the Government of India, insurance, and pension products. The standard framework of intertemporal utility maximization suggests that tax incentives alter the after-tax rate of return. This may encourage additional saving, that is, lead to saving that would not have happened had there been no tax breaks. However, it is also possible that this may only encourage diversion of saving to tax preferred instruments, in which case no new saving takes place. This is known as the “infra-marginal” effect, which describes saving that households would have done anyway, except that now they are in the product with the tax break. Theory remains ambiguous on which effect, the income or the infra-marginal, substitution, will dominate. Empirical evidence from international research on the impact of tax policy on household saving largely points to the infra-marginal effect (Duo et al. 2006; Gale and Scholz 1994; Engen, Gale and Scholz 1996; Engerlhardt and Kumar 2007; Gelber 2011; Poterba, Venti and Wise 1995). For example, Ochmann (2014) studies German data and finds that households with higher tax rates are found to have relatively greater demand for tax-privileged assets. Similarly, Chetty et al. (2014) study Danish households and find that tax subsidies induce relatively few individuals to respond. When individuals do respond, they primarily shift assets to the subsidized accounts. Academic literature on the effectiveness of tax breaks in India is scant (Das-Gupta 1990). There has, however, been a fair bit of discussion in the policy space on the optimal way to tax financial saving. While policy reports (Shome

2. Net financial saving by households is arrived at by subtracting financial liabilities from gross financial saving. Radhika Pandey et al. 3

2001; Kelkar, 2002; Malegam 2015) claim that there is only a ‘substitution effect’ in India, empirical work largely remains absent. This is mainly because empirical analysis on these questions is best done using survey data that studies investor decisions over time and requires some exogenous variation in eligibility (or some other instrument) to clearly identify the effect of the tax break (Bernheim 2002). Unfortunately, in India such historical micro-data is not available. This paper, for the first time, presents micro as well as macro evidence on what has been achieved by the tax breaks with regard to financial saving. Is it that overall financial saving has increased? Or is it that tax policy has channeled household saving into specific products? This question assumes importance as individual households are the major contributors to income taxes in India. Out of 49.8 million income tax returns filed for the assessment year 2017–18, 46.6 million returns were filed by individuals (Income Tax Department 2018). After deducting the returns that pay zero tax, 41 million individuals paid income tax in the assessment year 2017–18. This constitutes roughly 15–16 percent of the total households in the country.3 This is the universe affected by tax policy on financial saving instruments, and while it may appear small, this is the higher income earning section of the economy that has the potential to save large amounts and help towards financialization of the economy. These figures motivate an analysis of the impact of tax policy on household saving. For a macro-level analysis, the paper uses aggregate national accounts data to study how financial saving has evolved with changes in tax breaks. It enumerates the changes in the taxation of saving instruments for individuals that have taken place since 2001, and studies financial saving over the same period. We find that overall financial saving is not correlated with tax breaks. Financial saving has, in fact, fallen over the period. For a micro perspective, the paper studies household portfolios for the financial year 2016–17 using the CMIE Consumer Pyramids household Survey (CPHS). This allows us to compare households that fall under the tax bracket with households that do not. One caveat is that the CPHS data only asks if households have investments in specific instruments and does not report the rupee value of saving. The analysis, therefore, is restricted to understanding household investments on the extensive margin. To control for the income effect as distinct from the tax effect, we further compare

3. According to a report, there were 250 million households in the country (Ministry of Statistics and Programme Implementation 2018). However, according to a recent survey by Broadcast India, there were around 298 million households in the country in 2018. 4 INDIA POLICY FORUM, 2018 households that have the same household income, but differ in the tax treatment because at least one individual in the household has income that makes him/her fall in the tax bucket. We find that tax-incentivized households are more likely to invest in the products that are given a tax break, especially insurance. We also evaluate if there is a difference between the household characteristics of those who purchase insurance. If tax incentives matter, then household characteristics should not have any influence on investments in insurance for the group that is taxed. On the contrary, for the group that is not taxed, we should see certain characteristics such as age or household size influencing the decision to invest in insurance. We consider the subset of salaried households and regress the probability of purchasing insurance on the proportion of working members in the household, age, age square, gender, religion, caste, education, total annual income, and region (urban versus rural). We find that most household characteristics do not explain the investments in insurance of taxed households. This is not true for non-taxed households. The results suggest that tax breaks do not have an impact on increasing overall financial saving and at most lead to a “substitution effect,” that is, they lead to households channeling saving into the tax-exempt products without increasing the amount of overall saving. As well, tax breaks give a subsidy to better-off households which fall in the income tax net. This suggests that tax breaks end up benefitting those already at the upper end of the income distribution in India. The Government of India recently set up another committee to redraft the direct tax law.4 As the new committee deliberates on the direct tax law, there is need to ask what specific objectives the government wishes to meet through tax breaks, and whether the existing tax breaks are designed to meet those objectives. A cost–benefit analysis of such provisions can shape policy in generating saving into financial markets. The paper proceeds as follows. We describe a conceptual framework for the empirical analysis in Section 2. The institutional setting and the exact nature of the tax breaks are described in Section 3. Section 4 describes the data, while Section 5 presents evidence on the impact on savings. Micro-level evidence on household portfolios is presented in Section 6, and policy implications are discussed in Section 7. Section 8 presents the conclusion.

4. https://www.thehindubusinessline.com/economy/policy/new-task-force-set-up-to- redraft-direct-tax-law/article9969907.ece (accessed on June 3, 2019). Radhika Pandey et al. 5

2. Conceptual Framework

There is a large literature, theoretical and empirical, on the effect of tax treatment on saving. In this section, we reproduce the standard framework of intertemporal utility maximization with the objective of anchoring the empirical analysis.5 Consider a standard utility maximization of the following form:

T t ∑uctt()ρ (1) t=0

T −t st..∑CWt ββ≤ () (2) t=0

T −t WW()ββ= ∑ t (3) t=0

β=11+−im() (4)

where r<1 represents the rate of time preference, W is the value of lifetime resources, and the discount factor, i the pre-tax rate of return, and m the tax

rate. The associated level of saving (st) in this model is given by the difference between total income (including investment returns) and consumption. In this set-up, there are two factors that influence saving—changes in the tax rate m, and the pre-tax rate of return i through the after-tax rate of return i(1 – m). An increase in the after-tax rate of return has a substitution effect wherein there is a shift of consumption towards the future, leading to an increase in saving. The income effect leads to an increase in consumption in both periods, leading to a reduction in saving. Theory is silent on which effect dominates. When tax incentives are provided on specific products, then this reduces the tax rate applicable to saving below some threshold (contribution limit). Saving in a tax-deferred account is then a perfect substitute for other saving, as it generates a higher return. This is evident from the following example. Consider, once again, a two-period set-up where an individual consumes

Ci1 in period 1 and Ci2 in period 2. The individual could save in a tax- incentivized account (T) or a usual savings account (S). By saving in the usual savings account, the individual earns i. Returns to saving increase

5. The discussion is borrowed from the analysis of Bernheim (2002) and Chetty et al. (2014). 6 INDIA POLICY FORUM, 2018 to i + ψ in the tax-incentivized account, which translates to a net subsidy to the tax-incentivized account. Saving in the tax-incentivized account strictly dominates saving in the taxable account. Increases in the subsidy should affect saving through three channels: (a) a reduction in the price of the tax-incentivized account should lead to a substitution effect across the two accounts, (b) a reduction in the price of consumption in Period 2 relative to consumption in Period 1 should lead to an increase in saving, and (c) an increase in total lifetime wealth should reduce saving. Prior empirical research (Duflo et al. 2006; Engen et al. 1996; Engerlhardt and Kumar 2007; Gale and Scholz 1994; Gelber 2011; Poterba et al. 1995) suggests that the substitution effect dominates.6

3. Institutional Background: Section 80C

We turn next to understanding the institutional structure of the tax system and saving incentives in India. The Income Tax Act, 1961, is the primary law regarding taxation in India. Table 1 shows the tax thresholds as of 2016–17. Individuals with an annual income of less than `250,000 are exempt from income tax. The tax rate increases to 5 percent, 20 percent, and 30 percent, respectively, at higher income brackets. Since our focus is to analyze the impact of tax breaks on household saving, we look at those provisions of the Income Tax Act that provide for various tax deductions and exemptions for individuals. This is primarily possible through Section 80 (Sections 80C, 80CCC, 80CCD[1]) through which a deduction of `150,000 can be claimed from total taxable income of an individual in a particular financial year.7

TABLE 1. Tax Thresholds Annual Income Tax Rate (%) < `250,000 Nil Between `250,000–`500,000 5 Between ` 500,000–`1,000,000 20 > `1,000,000 30 Source: Budget document, 2016–17.

6. The dominance on total saving largely depends on the elasticity of intertemporal substitution. 7. Section 80CCE of the Income Tax Act, 1961, restricts the deduction under Sections 80C, 80CCC, and 80CCD(1) to `1.5 lakh. Radhika Pandey et al. 7

There are exemptions given on four categories of financial products. These include long-term savings, small savings schemes, fixed income products, and investment vehicles including equity products, and collective investment vehicles (Malegam 2015). The following instruments within these categories are eligible for the deduction:8

• Long-term instruments • Payment of life insurance premium to effect or to keep in force (premium restricted to 10 percent of the actual capital sum assured); • Payment made to effect or to keep in force a contract for a deferred annuity (including payment made by government as an employer); • Contribution to a provident fund (or superannuation fund); • Contribution to a pension fund set up by a mutual fund; • Contribution to the National Pension System of the Central Government; • Investment in notified fixed deposits with a mandatory lock-in period of five years. • Fixed income instruments • Subscription to such bonds issued by the National Bank for Agriculture and Rural Development; • Any subscription made to any such deposit scheme or pension fund set up by the National Housing Bank. • Small savings instruments • Investments in time deposits at the post office; • Subscription to any notified security of the Central Government, or saving certificates;9 • Any investment in an account under the Senior Citizens Savings Scheme Rules, 2004. • Equity instruments and collective investment schemes • Subscription to units of any Mutual Fund of Section 10(23D), referred to as equity-linked mutual funds; • Subscription to equity shares or debentures forming part of any eligible issue of capital.

The tax structure also taxes some financial instruments through capital gains taxes as well as a dividend distribution tax. For example, in the budget

8. https://taxguru.in/income-tax/income-tax-deductions-section-80c-eligible-investments- expenses.html (accessed on June 3, 2019). 9. National Savings Certificates are issued by the post office and have a minimum lock-in period of five years. 8 INDIA POLICY FORUM, 2018 of 2017–18, a capital gains tax was levied on equity investments through mutual funds, but not on unit-linked insurance plans. Until 2005–06, most of the deductions for individuals were part of Section 88 of the Income Tax Act. They were substituted by Section 80C. The 2005–06 budget speech, which brought in this change, said:

State must be neutral between one form of saving and another and allow the tax-payer greater flexibility in making savings/investments decisions.

The 2006–07 budget speech included bank fixed deposits for a term of not less than five years as an eligible instrument for tax deduction under Section 80C of the Income Tax Act. Having said this, the mix of Section 80C exemptions was more or less kept the same as that of Section 88. The exemptions have continued till date with minor changes between instruments. The overall limit was increased from `100,000 to `150,000 in the budget speech of 2014–15. Through the budget of 2015–16, investments in the National Pension System (NPS) were given an additional tax break of `50,000 through Section 80CCD (1B). From the perspective of “long-term saving,” insurance, pensions, and fixed deposits (for a period of five years and above) have been given prefer- ential treatment. Table 2 presents the changes in taxation specifically related to fixed deposits, insurance, and pensions over the last 15 years. A fixed deposit of a term of not less than five years was included in the list of instruments in the budget speech of 2006–07. The interest income from a bank FD is, however, fully taxable. Banks deduct tax at source (TDS) at the rate of 10 percent if the interest income for the year is more than `10,000. If the household is in a higher tax bracket, the additional tax has to be paid through self-declaration when filing returns.10 In the case of insurance, the first tax break since 2001 appeared in 2003– 04, when any sum that a beneficiary received from the insurance policy (including a bonus) was exempted from income tax. The next change was in 2012–13 when the exemption on deduction for life insurance premium was reduced to 10 percent of the actual capital sum assured, from the earlier 20 percent. Finally, in 2013–14, this exemption was increased to 15 percent for persons with disability and people with diseases or ailments. The first change provides an incentive to save more, while the second change is a reduction in the incentive. The third change improves the incentive to save into insurance, but is not applicable to everyone.

10. If the person has not provided his or her permanent account number (PAN), the bank will deduct TDS at the rate of 20 percent. Radhika Pandey et al. 9

TABLE 2. Changes to Fixed Deposits, Insurance, and Pension Taxation in India, 2003–04 to 2015–16 This table presents the changes in taxation specifically related to fixed deposits, insurance, and pensions over the last 15 years. Year Section Tax Changes Fixed Deposits 2006–07 80C Investment in a term deposit, for a fixed period of not less than 5 years, with any scheduled bank shall be eligible for deduction. 2012–13 80TTA Deduction of `10,000 can be claimed against interest income from a bank savings account. Insurance 2003–04 10D/88 Any sum received under a life insurance policy, including the sum allocated by way of bonus on such policy is exempt. Restricted to 20% of the actual capital sum assured. 2012–13 80C Deduction for life insurance premium, issued on or after April 1, 2012, shall be allowed for only so much of the premium payable as does not exceed 10% of the actual capital sum assured. This is a change from the 20% of capital sum assured earlier. 2013–14 80C A higher limit of 15% of actual capital sum assured has been provided for persons with disability and people with diseases or ailments. Pensions 2004–05 80CCD Mandatory NPS for new entrants to civil services from January 1, 2004. 2007–08 80CCD Individuals employed by “other employers,” and not just the Central Government, are now included under the purview of this Act. 2009–10 80CCD NPS extended to “self-employed” also. 2011–12 80CCE The contribution made by the Central Government or any other employer to a pension scheme shall be excluded from the limit of `1 lakh provided under Section 80CCE. 2015–16 80CCD Additional deduction of `50,000 for amount deposited by taxpayer in their NPS account. Source: Budget documents, various years.

In 2004, the government introduced the National Pension System (NPS), a defined contribution pension plan for new recruits to civil services. In April 2009, this was opened for citizens of India on a voluntary basis. Over the last few years, most of the tax breaks on pensions have revolved around the NPS. The big incentive to the NPS was in the budget speech of 2015–16 where an additional deduction of `50,000 (over and above the `150,000 overall limit) was provided for investing in the NPS. 10 INDIA POLICY FORUM, 2018

4. Data

We use both micro- and macro-level datasets to analyze the impact of tax breaks on household financial savings. Our analysis begins from 2001 since we want to analyze the impact of changes in tax policy governed by Section 80C of the Income Tax Act on household saving.11 Data on annual household saving are published by the Central Statistical Office (CSO).12 Information on financial assets and liabilities of the household sector are also published annually by the RBI as part of the ‘Flow of Funds (FoF) Accounts of the Indian Economy’. The RBI has been publishing estimates of household financial assets and liabilities five months ahead of the CSO’s release. We use the RBI data on “Changes in Financial Assets and Liabilities of the Household Sector (RBI)” for data on household financial saving (RBI 2017a). This data focuses only on the assets and liabilities of the household sector and provides us with a time series of financial assets such as insurance, and pension and provident funds.13 The table on “Gross Value Added and Gross Domestic Product” provides data on GDP. This is provided as two different series. The first one is the GDP at current prices using the 2004–05 base year series, while the second is sourced from the 2011–12 base year series. Unfortunately, it is not possible to combine these two series into one consolidated GDP series, as the base year change was accompanied by a comprehensive change in the CSO’s methodology for computing GDP.14 We also source data from the Consumer Pyramids Household Survey carried out by the Centre for Monitoring Indian Economy (CMIE) in three waves a year across India, with a total sample size of about 160,000 households.15 The survey asks questions on income, consumption, sources of credit, and choice of saving instruments. We restrict ourselves to the sample of households that are available throughout the financial year 2016–17, which leaves us with about 90,000

11. We source our data from CMIE Economic Outlook that provides a consolidated data release from government sources. 12. See Central Statistics Office (2012) for the sources and methodology of estimating the components of household saving. 13. Financial assets include currency, bank deposits, non-banking deposits, life insurance fund, provident and pension fund, claims on government, shares and debentures, units of Unit Trust of India, and trade debt. 14. For a discussion of the GDP measurement methodology, see Sengupta (2016). 15. To make the sample representative of the household population in India, adequate weights have been provided for the roughly 160,000 households. These weights are based on the Compound Annual Growth Rate from Census data from 2001 to 2011. Radhika Pandey et al. 11 households. We analyze household responses on income and consumption for the financial year 2016–17 (April 2016—March 2017). The responses on savings are from the first wave of 2017 (January 2017—April 2017). The survey asks households if they have outstanding investments in a particular product as of the survey date. We classify households with investments in financial products as those which have outstanding investments in at least one of the following:

• Bank fixed deposits; • Post office savings; • National Savings Certificate; • Kisan Vikas Patra; • Insurance; • Provident funds/pensions; • Mutual funds; and • Listed shares.

We classify households with some investments in physical assets as those which have outstanding investments in at least one of the following:

• Gold, and • Real estate.

The survey does not ask for the rupee value of saving in each product. As a result, we can only know whether the households have saved in some instrument, but cannot know how much. To analyze household portfolios on the basis of tax incidence, it is important to classify households on the basis of whether they fall under a non-zero income tax bracket. Since all questions are asked at the household level, it is possible that while total household income seems higher than the income tax threshold, each household member could individually earn less than the threshold. We therefore calculate the total income of each member of the household for the 2016–17 financial year. We then classify each household as “taxed,” or “not-taxed” if there is at least one member with annual income greater than `250,000, the income tax threshold.

5. Impact on Saving

If tax incentives affect financial saving, then we should see a rise in financial saving over the years, as there has been some tax break or the other given on financial products, especially since 2005. If, on the other hand, there is 12 INDIA POLICY FORUM, 2018

FIGURE 1. Household Financial Assets as a Percent of GDP 2001–17 20 15 10 5 0 5 10 15 % of GDP

Household financial assets as a percent to GDP (Base series 2004–05) Household financial assets as a percent to GDP (Base series 2011–12)

2001 2003 2005 2007 2009 2011 2013 2015

Source: RBI and CSO. a pure substitution effect, we should see that instruments that have received the most tax breaks see the largest share in total financial saving.

5.1. Overall Financial Saving Figure 1 shows the time series of household investment in financial assets (representing financial saving) as a percent of GDP. The black line shows the series using data from the 2004–05 base year, while the dashed line shows the data using the 2011–12 base year series. Financial saving had been rising steadily from about 11 percent of GDP in 2001 to a high of about 17 percent in March 2007. The years between 2003 and 2005 did not see any tax breaks, and yet there was a rise in financial savings. Since 2007, there has been a tax break announced pretty much every year (Table A.1 in the Appendix). And yet financial saving had fallen to about 11 percent of GDP by March 2013. The new series also shows roughly the same estimate. Financial saving as a percent to GDP remained roughly constant through the tax breaks of 2012 to 2015, and increased only slightly as of March 2016. This may be because of the increase in the overall limit of tax exemption from `100,000 to `150,000 in the 2014–15 budget. The association between tax breaks and financial saving appears weak. This is reflected in the fall in financial saving as a percent of household gross saving as well. Figure 2 focuses on the time series of the rupee value of financial saving and its share in overall household saving since March 2001. The top panel shows the rupee value of financial saving, while the bottom panel shows financial saving as a proportion of total household saving. The rupee value of saving has been rising since March 2001. It stayed stable between March 2008 and 2009. Since 2009, it has been rising (barring Radhika Pandey et al. 13

FIGURE 2. Total Financial Assets of Households

The figure shows the time series of overall household financial assets since March 2001. The top panel shows the rupee value of financial assets, while the bottom panel shows financial assets as a proportion of total household savings. 14,000 n llio 0,000 ` Bi 6,00 01

2,000 2001 2003 2005 2007 2009 2011 2013 2015 50 40 20 30

10 Financial assets as percent of gross household savings (Base series 2004−05) Financial assets as percent of gross household savings (Base series 2011−12) 0 % of Gross Household Savings 2001 2003 2005 2007 2009 2011 2013 2015

Source: RBI and CSO. a small dip in 2011). The share of household financial saving in total household saving has actually been falling since 2007. The new series suggests that since 2012, this has roughly stayed constant, with a slight increase in 2015—perhaps a response to the increase in the overall exemption limit to `1.5 lakh from `1 lakh described earlier. There appears to be no correlation between tax breaks and financial saving. Despite a continuous regime of tax breaks on one product or another, household financial savings have risen in some years, stayed stable in others, and have actually fallen in one year. The share of household financial saving in total household saving is lower in 2016 relative to 2001.

5.2 Composition of Household Financial Saving Table 3 shows the share of the various savings instruments in the overall household financial saving from 2011–12 to 2015–16. Bank deposits constitute the bulk of household financial saving, though their share in the overall household financial savings has seen a dip from 2011–12 onwards. Shares 14 INDIA POLICY FORUM, 2018

TABLE 3. Composition of Household Financial Savings 2011–12 2012–13 2013–14 2014–15 2015–16 (%) (%) (%) (%) (%) Currency 11.39 10.48 8.36 10.61 13.19 Deposits 57.95 56.97 56.01 50.94 43.59 Shares and debentures 1.77 1.60 1.59 1.58 2.72 Claims on government –2.35 –0.67 1.94 0.08 4.38 Insurance fund 20.98 16.91 17.17 23.81 17.50 Provident and pension fund 10.26 14.71 14.93 15.02 18.21 Source: CSO. and debentures constitute a small part of the overall financial saving of households. Investments in provident and pension funds have seen a gradual rise though they still constitute a small proportion of the overall saving. This could be an outcome of tax incentives announced in the pensions’ space. For example, the Budget of 2015–16 announced an additional deduction of `50,000 for contribution towards NPS over and above the limit of `1.5 lakh under Section 80C. The NPS was till recently operated on an EET (Exempt, Exempt, Tax) basis, wherein 60 percent of the balance that could be withdrawn as a lump sum16 was taxed (but the original contribution and any accumulation was tax exempt). In a recent decision, the Government has decided to make the 60 percent withdrawable amount tax-free. This move could further enhance the attractiveness of NPS as a saving product. Table 3 gives us a big picture of the problem of household financial saving. It is instructive to look further at the household data to gain an understanding of household preferences for savings instruments. When we look at the composition of household portfolios, we find a big difference in those households that are taxed versus not taxed (Table 4). In 2016–17, a much larger proportion of taxed households claims to have outstanding investments in fixed deposits,17 insurance, small savings, and pensions, all of which are instruments covered under Section 80C. For example, 88 percent of households that are not taxed claim to have outstanding investments in fixed deposits, as opposed to 96 percent of taxed households. The next instrument of choice is insurance, with 50 percent of non-taxed households having insurance, as opposed to 90 percent of taxed households, followed by

16. NPS investors have to use 40 percent of the corpus to buy an annuity and can withdraw the remaining 60 percent of the corpus. The annuitized amount is tax-free. 17. The fixed deposits may not be the ones with a tax break. Radhika Pandey et al. 15

TABLE 4. Household Portfolios by Tax Status (2016–17) Under the Above the Tax Threshold Tax Threshold Average household income (`) 161,731.7 551,579.4 Percentage of households with investments in % % Physical assets 99 99 Financial assets 92 99 Fixed deposits 88 96 Insurance 50 90 Provident funds 7 55 Small savings 10 23 Mutual funds/shares 0.3 3.4 Number of households 79,497 7,279 Source: CPHS data. provident/pension funds where the difference is much more striking; 7 percent of non-taxed households as opposed to 55 percent of taxed households.

6. Impact on Saving: Micro-level Evidence

We turn to estimating the effect of being in a taxable income bracket on the probability of having outstanding investments in specific financial products. We estimate a probit regression as follows:

Y = ti β1 + Xi β2 + εi where Y is an indicator for the latent variable Y*. In this case, Y = 1 if Y* > 0, or Y = 0 otherwise. This indicates if a household i has outstanding investments in a specific financial product. ti indicates if the household falls in the tax-paying bracket. Xi are the controls which include household characteristics such as age, gender, education, occupation, religion, and caste of the head of the household, the number of earning and non-earning members in the household. ε is normally distributed with mean zero, and variance 1. This analysis does not estimate the determinants of household saving, but instead is focused on understanding the differential in propensity to save in various financial instruments between households that face different tax burdens. Table 5 shows the results of investments in fixed deposits (Columns 1 and 2), insurance (Columns 3 and 4), pensions (Columns 5 and 6), and small savings (Columns 7 and 8). Columns 1, 3, 5, and 7 show the results without TABLE 5. Investments in Tax-incentivized Products This table shows estimates of probit model that explains a dummy variable that is “1” when investment is made in a tax-saving financial product and “0” otherwise. For each financial product, two models are estimated: one in which the explanatory variable is whether the household falls in the tax bracket, and two in which household characteristics such as gender, age, average number of earning members, education, religion, and caste are also controlled for. The results show that controlling for household characteristics, households in the tax bracket are more likely to have outstanding investments in the tax-favored financial products. Fixed Deposit Insurance Provident Fund Small Savings (1) (2) (1) (2) (1) (2) (1) (2) Taxed 0.570*** 0.465*** 1.282*** 0.786*** 1.600*** 1.037*** 0.526*** 0.336*** (0.027) (0.029) (0.02) (0.022) (0.016) (0.019) (0.017) (0.020) Constant 1.177*** 0.604*** –0.009** –2.617*** –1.476*** –4.574*** –1.264*** –1.970*** (0.006) (0.13) (0.004) (0.104) (0.007) (0.182) (0.006) (0.137) Observations 86,776 86,776 86,776 86,776 86,776 86,776 86,776 86,776 Log likelihood −30,324.6 −29,574.6 −57,491.6 −53,025.1 −25,160.9 −19,807.6 −30,314.2 −29,817.9 Additional controls NO YES NO YES NO YES NO YES Sources: CPHS data; Authors’ calculations. Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01. Radhika Pandey et al. 17

TABLE 6. Marginal Effects: Investments in Specific Instruments Instrument Marginal Effect Fixed deposit 0.065*** (0.003) Insurance 0.280*** (0.006) Pensions 0.202*** (0.006) Small savings 0.073*** (0.005) Source: CPHS data; Authors’ calculations. any controls, while Columns 2, 4, 6, and 8 control for all the household characteristics. The results show that controlling for socio-demographic variables, households that fall in the tax bracket are more likely to have outstanding investments in the four tax-favored financial products than households that are not in the tax bracket. This is true for all four savings instruments. The magnitude of the effect of the tax break is best understood by calculating marginal effects. Table 6 shows the marginal effects of being in a taxable bracket from the probit regressions in Columns 2, 4, 6, and 8, that is, the regressions with the controls. The marginal effect can be considered as an approximation of the effect of a unit change in the independent variable on the probability P(Y=1|X=x). The results show that being in a tax bracket leads to a 6.5 percent higher probability of investing in a fixed deposit, a 28 percent higher probability of investing in insurance, a 20 percent higher probability of investing in pensions, and a 7 percent higher probability of investing in small savings. These results strongly suggest that tax incentives have an influence on the composition of savings. However, one could argue that this is purely an income effect, and not a tax effect. Those in the tax bracket are also those with higher incomes and would have saved in these products anyway. We, therefore, only consider those in the salaried class, as there is less ambiguity about the tax liability of these households, and because of TDS, we see the highest compliance. We also only consider those households with incomes over `250,000, as this is where the incentives will begin to matter. In Table 7, we further divide these households into those that are taxed or not. There may be households whose income may be over the threshold, but no single member may have an income higher than the threshold. This household would not qualify as a taxed household. This gives us tax variation in the same income bracket. 18 INDIA POLICY FORUM, 2018

TABLE 7. Tax-incentivized Households in Salaried Class This table shows the number of households in the Consumer Pyramids data within each income bracket that have a tax liability and those that do not have a tax liability.

Household Income Between Under Tax Threshold Above Tax Threshold ` No. of Households 250,000–300,000 1,218 323 300,000–350,000 682 680 350,000–400,000 358 766 400,000–450,000 213 651 450,000–500,000 136 427 500,000–550,000 61 282 550,000–600,000 81 363 > 600,000 69 1,309 Source: CPHS data.

As an example, Table 7 shows that there are 1,218 households in the income bracket of `250,000–300,000 but since the individual members of such households earn less than `250,000, they are not taxed. This approach enables us to isolate the impact of tax policy on the savings of households. As we have only considered those in “salaried” occupations, we are able to control for the potential confounding effect of agricultural income, which is not taxed. We estimate a probit regression on each of the income categories separately. The regressions control for all the household characteristics. Table A.2 in the Appendix presents the results from the regressions. They suggest that there is a limited “tax incentive effect” on saving in insurance and pensions. Those households within the same income bracket that are exposed to the tax are more likely to have outstanding investments in insurance and pensions than households that are not taxed. The effect is strongest for insurance in the sense that households across the various income groups are more likely to have outstanding investments in insurance if they fall in the tax bracket. There is no difference in savings in FDs and small savings. At incomes of about `500,000, there is no difference in the probability of investing in pensions between the taxed and non-taxed households. Table 8 shows the marginal effects. Here also, we see that the effect is significant mostly for insurance and pensions for households in the income category of `350,000 to `500,000. For example, those which fall in the tax bracket in the `350,000—`400,000 income category are almost 7 percentage points more likely to have invested in insurance relative to those which do Radhika Pandey et al. 19

TABLE 8. Marginal Effects: Outstanding Investment of Salaried Households This table shows the marginal effect of being taxed on the probability of having outstanding investment in various saving products. The “tax incentive” impact is seen to be the most significant for insurance and pensions. Fixed Deposit Insurance Provident Fund Small Savings Income Category (`) (1) (2) (3) (4) 250,000–300,000 –0.023 0.033 0.038 –0.010 (0.10) (0.024) (0.033) (0.024) 300,000–350,000 –0.005 0.038 0.094 0.011 (0.030) (0.044) (0.030) (0.021) 350,000–400,000 0.007 0.069*** 0.163*** 0.023 (0.091) (0.021) (0.033) (0.026) 400,000–450,000 0.002 0.046*** 0.109*** 0.012 (0.071) (0.024) (0.039) (0.037) 450,000–500,000 0.003 0.028 0.083* 0.037 (0.344) (0.151) (0.049) (0.051) 500,000–550,000 0.001 0.028 0.005 0.025 (0.11) (0.17) (0.055) (0.055) 550,000–600,000 –0.001 0.075 –0.002 0.096 (0.252) (0.523) (0.064) (0.269) 600,000+ 0.002 0.066 0.069 0.039 (0.102) (0.040) (0.059) (0.054) Source: CPHS data; Authors’ calculations. not fall in the tax bracket. Those taxed in the `400,000—`450,000 income category are 5 percentage points more likely to have invested in insurance. Another way to test whether this is a tax incentive effect is to evaluate if there is a difference between the household characteristics of those which purchase insurance. If tax incentives matter, then household characteristics should not have any influence on investments in insurance for the group that is taxed. On the contrary, for the group that is not under the tax bracket, we should see that certain characteristics such as age or household size influence the decision to invest in insurance. We consider the subset of salaried households and regress the probability of purchase of insurance on the proportion of working members in the household, age, age square, gender, religion, caste, education, total annual income, and region (urban versus rural). The results are presented in Table 9. We find that most household characteristics do not explain the investments in insurance of the taxed households. This is not true for the non-taxed 20 INDIA POLICY FORUM, 2018

TABLE 9. Probability of Investment in Insurance of Salaried Households

This table shows the effect of household characteristics on the probability of investment in insurance for taxed and non-taxed households. The table shows that household characteristics drive the decision to invest in insurance for non-taxed households. These characteristics matter less for households having tax liabilities. Non-taxed Taxed Prop. working members –0.139*** –0.032 (0.020) (0.021) Age 0.030*** 0.003 (0.004) (0.005) Age2 –0.0003*** −0.00005 (0.00004) (0.00005) Gender: male 0.070*** 0.025 (0.016) (0.017) Annual income 0.00000*** 0.00000*** (0.00000) (0.00000) Religion: Muslim −0.209*** −0.053** (0.023) (0.025) Religion: Hindu –0.025 –0.008 (0.018) (0.015) Caste: intermediate 0.037*** 0.009 (0.014) (0.012) Caste: lower –0.091*** –0.013 (0.010) (0.009) Caste: not stated 0.075*** 0.012 (0.036) (0.033) Educ: school 0.186*** 0.021 (0.018) (0.031) Educ: diploma 0.245*** 0.017 (0.033) (0.036) Educ: graduate/above 0.281*** 0.005 (0.021) (0.032) Region: urban 0.033*** 0.017 (0.011) (0.013) Constant –0.361*** 0.829*** (0.093) (0.120) Observations 13,065 4,801 Log likelihood −7,973.669 −485.168 Source: CPHS data, Authors’ calculations. Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01. Radhika Pandey et al. 21 households. The probability of insurance investments is lower when the proportion of working members is higher, increases with age, and then decreases, is higher for those educated relative to those who are illiterate, and is higher for those in urban regions. The direction of coefficients is what one would expect. While income is statistically significant, the coefficients suggest that it has no economic impact on the decision to invest in insurance. Once the household is in the tax bracket, none of these variables has any effect. This suggests that tax incentives do play a role in the purchase of insurance. Our analysis suggests a “tax incentive” effect of investments in financial products. Controlling for household characteristics, households that have tax liability are more likely to have outstanding investments in tax-incentivized financial products.

7. Policy Implications

The analysis indicates that tax breaks have no effect on the overall financial saving of households. It also shows that households that are in a non-zero tax bracket are incentivized to save in a specific set of products, in particular, insurance products. In this section, we evaluate the policy implications of the existing system of tax incentives and its impact on household portfolios.

7.1. Low Tax Base The inability of tax breaks to have any effect on overall financial saving might come from the low tax base in India. As shown in Table 1, individuals with an annual income of less than `250,000 are exempt from income tax. The tax rate increases to 5 percent, 20 percent, and 30 percent, respectively, at higher income thresholds. Given low incomes in India, very few households actually fall into the tax bracket. Figure 3 shows the number of households in the tax bracket in the CPHS dataset. About 92 percent of our sample (79,497) households fall in the 0 percent tax-paying bracket, 7.6 percent fall in the 5 percent tax bracket, while less than 1 percent fall in the 20 percent and the 30 percent tax brackets. The numbers from the survey data consist of households that have at least one member in the household with annual income greater than the thresholds. It is quite possible that some of this is agricultural income and, therefore, exempt from tax. It is also possible that there is an inconsistency between income declared to the field investigators and the actual income of the household. 22 INDIA POLICY FORUM, 2018

FIGURE 3. Number of Households at Different Annual Income Levels

0% 5% Bracket 25% Bracket 25,000

20,000

15,000

Households 10,000

5,000

0 0 200,000 400,000 600,000 800,000 1,000,000

Source: CPHS data. Note: The dotted lines indicate income tax brackets with no income tax below `250,000, 5% thereafter, and then 25% and 30%.

Data from the Income Tax Department shows that 46.6 million individuals filed income tax returns in the assessment year 2017–18. This is less than 3.5 percent of India’s total population. Of these, 5.6 million (12 percent of individuals who filed returns) filed zero returns, that is, they had income less than `250,000 (Income Tax Department 2018). It is, therefore, not surprising that tax breaks given by the Government on specific financial products have had little effect on overall financial saving in the economy. Given the low reach of direct tax incentives, the Government should evaluate the benefits of these tax breaks vis-à-vis the cost of revenue forgone, or distortions in the market that get created. For example, the Government’s Economic Survey 2015–16 points out that tax incentives for household savings lead to fiscal loss, distort the interest rate structure “and merely help in mobilising funds to specified savings instruments.” The Survey also observes that the “real small savers” are outside the tax net and do not enjoy any form of tax subsidy on their savings.

7.2. Skew toward Insurance According to a study by Willis Towers Watson (2015), in the year 2014–15, life insurance accounted for 19 percent of total household financial assets in India, second only to the banking sector that holds 46.9 percent. Traditional Radhika Pandey et al. 23 endowment products that bundle savings and insurance account for 87 percent of the total business of `3.6 trillion in the life insurance market in India (IRDAI 2016). Among the class of tax-incentivized instruments, insurance remains very popular. The skew toward insurance becomes a concern in the context of the large-scale mis-selling scandals in the sector that have been witnessed over the last decade and half.18 Mis-selling of bundled insurance products (unit- linked insurance plans) has been estimated to have cost customers around US$28 billion between 2004 and 2011 (Halan, Sané and Thomas 2014). A committee set up by the Ministry of Finance has found that the problem of poor disclosures on products is highest in the context of endowment insurance products (DEA 2015). Audit studies have also provided evidence of poor sales practices, especially with regards to insurance products (Anagol, Cole and Sarkar 2017; Halan and Sané 2017). A committee formed by the insurance regulator on the sale of insurance products through banks has also admitted to mis-selling through banks (IRDA 2011). In such an environment of poor consumer protection, the role of tax breaks on specific products needs to be questioned. The channeling of savings into insurance and pensions may also not be useful in providing capital to firms, if regulatory mandates inhibit these sectors from investing in assets other than government bonds. For example, the IRDAI (Investment) regulations mandate that not less than 50 percent of the funds of insurers in the life insurance business need to be invested in government securities and other approved investments.19

8. Conclusion

While there is an active policy debate on the tax treatment of savings, empirical evidence on the impact of tax breaks on household savings in India is relatively scant. This paper aims to fill this gap. It presents macro- and micro- level evidence on the impact of tax breaks on household financial saving. The results suggest that overall financial saving is not correlated with tax announcements. Financial saving has, in fact, fallen in the period. Micro- level analysis of household portfolio data suggest a “tax incentive impact” on saving. We evaluate the probability of investments in tax-favored financial products for households having tax liabilities. We find that after controlling

18. For example, see Datta-Ray (2015) and Basu (2015) for commentary on mis-selling. 19. The Budget 2018–19 proposed that regulators should allow investments in below AA-rated bonds to encourage investment in corporate bonds. 24 INDIA POLICY FORUM, 2018 for household characteristics, the probability of investments in tax-favored financial products is higher for households that are taxed. A disaggregated analysis of salaried households suggests that the tax- incentivized impact on saving is highest for salaried households in the income bracket of `350,000 to `500,000. For households not subject to tax liability, household characteristics drive the probability of investments in insurance products. Thus, while the aggregate financial saving has remained stable, tax breaks have been influential in driving savings into specific products, such as insurance and pensions. The skew toward insurance becomes a concern in the context of the large-scale mis-selling scandals in the sector that have been witnessed over the last decade and half. The results suggest that policymakers should rethink what policy goals are being served by channeling saving into specific products.

References

Anagol, S., S. Cole, S. Sarkar. 2017. “Understanding the Advice of Commissions- Motivated Agents: Evidence from the Indian Life Insurance Market,” Review of Economics and Statistics, 99(1): 1–15. Badarinza, C., V. Balasubramaniam, and T. Ramadorai. 2017. “The Indian Household Finance Landscape,” India Policy Forum, 13: 1–55, New Delhi: National Council of Applied Economic Research. Basu, D. 2015. “Regulators versus Customers,” Business Standard, January 25. Available at https://www.business-standard.com/article/opinion/debashis- basu-regulators-versus-customers-115012500724_1.html (accessed on June 3, 2019). Bernheim, B.D. 2002. “Taxation and Saving,” in A. Auerbach and M. Feldstein (eds.), Handbook of Public Economics, Vol. 3, Chapter 18, pp. 1173–1249. Amsterdam: Elsevier Science B.V. Central Statistics Office. 2012. “National Accounts Statistics: Sources and Methods.” New Delhi: Ministry of Statistics and Programme Implementation, Government of India. Chetty, R., J.N. Friedman, S. Leth-Petersen, T.H. Nielsen, and T. Olsen (2014). “Active vs. Passive Decisions and Crowd-Out in Retirement Savings Accounts: Evidence from Denmark,” The Quarterly Journal of Economics, 129(3): 1141–1219. Das-Gupta, A. 1990. “Personal Taxation and Private Financial Savings in India,” Economic & Political Weekly, 25(6): 325–338. Datta-Ray, S.K. 2015. “A Ride You Can’t Bank on,” Business Standard, July 10. Available at https://www.business-standard.com/article/opinion/ Radhika Pandey et al. 25

sunanda-k-datta-ray-a-ride-you-can-t-bank-on-115071001254_1.html (accessed on June 3, 2019). DEA. 2015. “Report of the Committee to Recommend Measures for Curbing Mis-Selling and Rationalising Distribution Incentives in Financial Products,” Committee Chaired by Sumit Bose. New Delhi: Department of Economic Affairs, Ministry of Finance, Government of India. Available at https://www.finmin.nic. in/sites/default/files/Final_Report_Committee_on_Incentive_Structure_0.pdf (accessed on June 3, 2019). Duflo, E., W. Gale, J. Liebman, P. Orszag, and E. Saez. 2006. “Saving Incentives for Low- and Middle-Income Families: Evidence from a Field Experiment with H&R Block,” The Quarterly Journal of Economics, 121: 1311–1346. Engen, E.M., W.G. Gale, and J.K. Scholz. 1996. “The Illusory Effects of Saving Incentives on Saving,” Journal of Economic Perspectives, 10(4): 113–138. Engerlhardt, G.V. and A. Kumar. 2007. “Employer Matching and 401(k) Saving: Evidence from the Health and Retirement Study,” Journal of Public Economics, 91: 1920–1943. Gale, W.G. and J.K. Scholz. 1994. “IRAs and Household Saving,” American Economic Review, 84(5), 1233–1260. Gelber, A.M. 2011. “How Do 401(k)s Affect Saving? Evidence from Changes in 401(k) Eligibility?” American Economic Journal: Economic Policy, 3(4): 103–122. Halan, M. and R. Sané. 2017. “Misled and Mis-Sold: Financial Misbehaviour in Retail Banks?” Journal of Comparative Economics, 45(3): 429–444. Halan, M., R. Sané, and S. Thomas. 2014. “The Case of the Missing Billions: Estimating Losses to Customers due to Mis-Sold Life Insurance Policies,” Journal of Economic Policy Reform, 17(4): 285–302. Income Tax Department. 2018. Income Tax Return Statistics: Assessment Year 2017–18. New Delhi: Government of India. IRDA. 2011. Report of the Committee on Bancassurance. New Delhi: Insurance Regulatory and Development Authority. IRDAI. 2016. Annual Report. New Delhi: Insurance Regulatory and Development Authority of India. Kelkar, V. 2002. Report of the Task Force on Direct Taxes. New Delhi: Government of India. Malegam, Y.H. 2015. Report of the Working Group on Taxation Practices for Financial Instruments in India. New Delhi: Reserve Bank of India. Ministry of Finance. 2016. Economic Survey 2015–16. New Delhi: Government of India. Ministry of Statistics and Programme Implementation. 2018. India in Figures: 2018. New Delhi: Government of India. Ochmann, R. 2014. “Differential Income Taxation and Household Asset Allocation,” Applied Economics, 46(8): 880–894. Poterba, J.M., S.F. Venti, and D.A. Wise. 1995. “Do 401(k) Contributions Crowd Out Other Personal Saving?” Journal of Public Economics, 58(1): 1–32. 26 INDIA POLICY FORUM, 2018

RBI. 2017a. “Changes in Financial Assets/Liabilities of the Household Sector (At Current Prices),” Handbook of Statistics on Indian Economy. New Delhi: Reserve Bank of India. ——— 2017b. Gross Savings, Annual Report of the RBI. New Delhi: Reserve Bank of India. Sengupta, R. 2016. “The Great Indian GDP Measurement Controversy,” The Leap Blog, September 10. Shome, P. 2001. Report of the Advisory Group on Tax Policy and Tax Administration for the Tenth Plan. Constituted by the Planning Commission. New Delhi: Government of India. Willis Towers Watson. 2015. India Market Life Insurance Update December 2015. Available at https://www.towerswatson.com/en-IN/Insights/Newsletters/ Asia-Pacific/india-market-life-insurance/2015/March-2015-India-Market-Life- Insurance-Update (accessed on June 3, 2019).

Appendix Tables

TABLE A.1. Changes to the Tax Structure Increase in Tax Break Overall Insurance Pensions Other 2015–16 Yes NA Yes Yes (post office) 2014–15 Yes* Yes Yes NA 2013–14 NA Yes NA Yes (shares) 2012–13 NA No NA Yes (bank) 2011–12 NA NA Yes Yes (bonds) 2010–11 NA NA NA Yes (bonds) 2009–10 NA NA Yes NA 2008–09 NA NA NA Yes (bank) 2007–08 NA NA Yes NA 2006–07 Yes NA NA Yes (bank) 2005–06 Can’t say NA NA NA 2004–05 NA NA Yes NA 2003–04 NA NA** NA NA 2002–03 NA** NA NA NA 2001–02 NA NA NA Yes (annuity) Source: Budget documents. Notes: *Overall limit increased to `1.5L. **Change under Sec. 88, not under Sec. 80C. Radhika Pandey et al. 27

TABLE A.2. Outstanding Investment of Salaried Households This table shows estimates of a probit model that explains a dummy variable that is “1” when investment is made in a tax-saving financial product and “0” otherwise. The analysis is done on households having tax liabilities in various income brackets. The findings suggest that households subject to tax are more likely to invest in tax-incentivized saving products, in particular in insurance and pensions. Fixed Deposit Insurance Provident Fund Small Savings Income Category (`) (1) (2) (3) (4) 250,000–300,000 –0.022 0.032 0.037 –0.012 (0.017) (0.025) (0.032) (0.025) 300,000–350,000 –0.007 0.041** 0.092*** 0.010 (0.014) (0.018) (0.027) (0.021) 350,000–400,000 0.012 0.071*** 0.158*** 0.024 (0.016) (0.020) (0.030) (0.027) 400,000–450,000 0.006 0.050*** 0.112*** 0.012 (0.020) (0.023) (0.037) (0.035) 450,000–500,000 0.022 0.034 0.079* 0.036 (0.021) (0.030) (0.046) (0.044) 500,000–550,000 0.009 0.036 0.004 0.024 (0.021) (0.037) (0.055) (0.055) 550,000–600,000 –0.015 0.100** 0.0003 0.101 (0.030) (0.040) (0.065) (0.063) 600,000+ 0.039 0.069** 0.070 0.035 (0.024) (0.031) (0.053) (0.056) Source: CPHS data, Authors’ calculations. Comments and Discussion*

Rajnish Mehra† Arizona State University and NCAER

Introduction

Historically, savings for retirement has not been a major public policy issue, especially in developing countries where life expectancy and the working life span largely coincided. Increases in life expectancy have increased the number of years a household spends in retirement relative to its working years. Consequently, the issue of motivating time-consistent saving for retirement planning has appeared front and center in the public policy arena. To induce households to save more, governments the world over have resorted to tax- incentivized schemes. A key public policy question is: Do these incentives increase household savings or do they simply result in portfolio rebalancing? Tax-incentivized schemes increase the after-tax rate of return to a targeted asset class, thereby changing the price of future consumption relative to current consumption. This induces both an income effect and a substitution effect. Making future consumption cheaper induces households to save more. However, tax-free compounding increases the after-tax rate of return on assets; this will, in general, lead to portfolio rebalancing and may induce a household to save less. Economic theory tells us that the net effect of a change in the after-tax return on savings will, in general, be ambiguous.1 Hence, this issue needs to be investigated empirically.

* To preserve the sense of the discussions at the India Policy Forum, these discussants’ comments reflect the views expressed at the IPF and do not necessarily take into account revisions to the conference version of the paper in response to these and other comments in preparing the final, revised version published in this volume. The original conference version of the paper is available on www.ncaer.org. † I am especially thankful to Saurabh Bandyopadhyay for his helpful comments and research support. 1. It will depend on the elasticity of intertemporal substitution. We illustrate this in the context of a simple deterministic two-period (partial equilibrium) model where agents have preferences of the form

c1−γ −1 uc(,γ)= 1− γ Radhika Pandey et al. 29

This paper addresses this important question in the Indian context. At the macro-level, the authors examine aggregate national accounts data to study how financial savings have evolved with changes in tax breaks. They find no link between tax breaks and overall financial savings. At the micro-level, using data on household portfolios for the financial year 2016–17 from the CMIE Consumer Pyramids Household Survey, they find that households that are taxed invest more in the tax-incentivized asset classes. The two findings taken together suggest that, in the Indian context, tax-incentivized schemes result in household portfolio reshuffling rather than net additional investment in financial assets.

A Snapshot of Taxation in India Before discussing the paper, it is useful to examine the taxation landscape, as it puts the findings and the scope of the paper in context.2 In 2015–16, only 40.7 million individuals, less than 4 percent of the population, filed tax returns. After adjusting for individuals reporting income below the `250,000 taxation threshold, the authors conclude that the tax exemptions discussed in the paper are useful to about 18.9 million individuals, less than 2 percent of the population. Hence, any quantitative effects of the policy changes addressed in the paper are likely to be insignificant. The authors are fully aware of this, “It is, therefore, not surprising that tax breaks given by the government on specific financial products have had little effect on overall financial saving in the economy.”

with elasticity of intertemporal substitution 1/g; g ≥ 0. In the first period agents work, consume (c0) and save (s) at the after-tax return (r). In the second period (retirement), they consume

(c1) their savings. Households solve the following problem: max u(c0, g) + bu(c1, g) subject to c0 + s  Y c1  s(1 + r) The solution is:

β1/ γ s = Y. ()1++r 11− //γγβ1

Since ∂s <=00,, or >>01 if γ =<11 or , the effect of a change in the after-tax return on savings ∂r is ambiguous. 2. All the figures are from the paper. 30 INDIA POLICY FORUM, 2018

A Macro Perspective The paper begins by documenting the changes to capital taxation over the last 15 years and examining the effects of these changes on savings in financial assets (the stock of assets expressed as a fraction of GDP). Figure 1 in the paper (reproduced earlier) plots household financial assets as a share of GDP. The lack of any systematic and significant variation in this figure leads the authors to conclude:

There appears to be no correlation between tax breaks and financial savings. Despite a continuous regime of tax breaks on one product or another, savings have risen in some years, stayed stable in others, and have actually fallen in one. The share of financial savings in total savings is lower in 2016 relative to 2001.

While this may well be true, it is premature to base this conclusion on the evidence presented in Figure 1 of the paper as it confounds two effects: inflows to financial assets and a revaluation of existing assets. While the stock of financial assets in the economy increases due to inflows, it also changes due to a revaluation of existing assets, both due to tax subsidies and changes in economic conditions. The latter effects may be positive or negative. A better metric would be to examine if flows respond to tax incentives. Figure 2 presents the annual flows into financial assets as a fraction of GDP over the period 2003–17. In contrast to the stock values presented in Figure 1, inflows as a share of GDP have increased over time. I have not correlated the inflows with

FIGURE 1. Annual Flows into Financial Assets as Percent of GDP

14.0 12.0 10.0 8.0 % 6.0 4.0 2.0 0.0

2003–042003–042004–052005–062006–072007–082008–092009–102010–112011–122012–132013–142014–152015–162016–17

Source: RBI, CMIE, and IndiaStat. Radhika Pandey et al. 31 changes in tax incentives but this is something the authors could explore in their revision.

A Micro Perspective The authors next estimate the effect of being in a taxable income bracket on the probability of having investments in specific financial products. They do this by examining portfolio formation using the CMIE Consumer Pyramids Household Survey Data for 2016–173 and running the following probit regression:

* Yt=+iiββ12X +εi

The results are presented in Tables 5 and 6 in the paper. While Table 5 shows that the results are statistically significant, the economically interesting results are in Table 6. The first number, 0.065, implies that in moving from a non-tax-paying bracket to a tax-paying bracket, the probability of investing in a fixed deposit increases by 6.5 percent. The potential confounders in this include income, wealth (financial and non-financial assets), and financial literacy. For example, wealthy people both pay taxes and invest more. This mechanically induces a correlation between paying taxes and investing. The authors are aware of this and attempt to mitigate this by exploiting the differential tax status in income-matched households. They categorize households based on salary, change X from being an indicator variable to one representing category values and re-run the probit regressions. The results are reported in Table 8. In Tables 5 and 6, where the comparison is between households in the non-tax-paying bracket with households in the tax-paying group, the results are statistically significant. In contrast, in Table 8, where the comparison is between salaried tax-paying and non-taxpaying households in different salary brackets, the statistical significance largely vanishes (except for households in the `350,000–400,000 and `400,000–450,000 brackets). On the basis of this mixed evidence, it would be premature to conclude that there exists a “tax incentive” effect of investments in financial products, as the authors conclude.

3. The low tax base presents a challenge, which the authors acknowledge: “About 92 percent of our sample (79,497) households fall in the zero percent tax-paying bracket. 7.6 percent fall in the 5 percent tax bracket, while less than 1 percent fall in the 20 percent and the 30 percent tax brackets.” 32 INDIA POLICY FORUM, 2018

Concluding Comments The paper documents the response of Indian households to tax-incentivized financial instruments, a relatively new, topical, and important area of public policy research. It is work in progress. I look forward to a more nuanced parsing of its implications for tax policy.

Govinda Rao National Institute of Public Finance and Policy

This is an important paper by Pandey, Patnaik and Sané. There is very little research on the subject, except an old paper by Arindam Dasgupta in the early 1990s, when we were all talking about tax reform for the Chellaiah Committee, where he calculated after-tax rates of return. In that sense, it is a welcome addition to the empirical analysis. In India, tax policy is employed to pursue multiple objectives. There are very few studies analyzing the cost and efficacy of achieving these objectives. In that sense, this study is opportune, particularly because Arbind Modi is busy re-redrafting the tax laws because the first Direct Tax Code, which was drafted by him, has gathered dust. Then there was a second version, which again went into oblivion. This is the third version being attempted, and hopefully something will happen. Now a small clarification: tax incentives are given not just for increasing savings. There are three different objectives of this policy. One is superannuation, or encouraging people to save for their retirement. The second is to insure against the risk to life, and the third is to encourage people to have more financial savings as against physical savings. In that sense, looking at things only from the superannuation perspective is not really appropriate. Of course, the paper doesn’t do that. It examines whether tax breaks given to promote financial savings have been effective. As expected, the study concludes that tax breaks have not led to higher financial savings. Using CMIE’s consumer pyramids household data, the study concludes that tax-paying households invest more in tax-exempted financial products, and this is done by reducing other financial products. This has important implications for policy, raising the question of whether the tax incentive for promoting savings should continue at all. In fact, the first Direct Tax Code attempted to remove these incentives. So possibly Arbind Modi will return to that in his report. There is a bit of trickiness in the way in which the tax incentive is given, particularly to fixed deposits. For new entrants to the income tax bracket, the Radhika Pandey et al. 33 marginal gain from the tax benefit becomes zero after saving `150,000. But for existing tax-payers, if they have saved for five years, they can continue to avail of the benefit without adding any new savings; they can simply roll over the savings in tax-free instruments made six years earlier. In fact, if they have saved for five years, they can simply roll over what they did five years ago, and do not need to save additionally in order to get a tax benefit. Deductible savings have only a five-year lock-in period, after which they can simply be renewed. While there may not be any dispute regarding the overall conclusions drawn in the study, in my view, the analysis needs to be strengthened. The authors say that this is a work in progress, so possibly they will strengthen it further. The major conclusion that there is no relationship between tax breaks and financial saving may well be true. However, this cannot be merely inferred from the lack of correlation. We need a robust analysis of the determinants of household saving. The increase in investment by households in physical as against financial assets may simply be the result of a higher after-tax rate of return on physical assets. Given informal markets, it is difficult to calculate after-tax returns on physical assets. The after-tax return on land or immovable property depends on rental income plus capital appreciation, and underestimation of the value of transactions in land and buildings makes it difficult to estimate capital gains. So an increase in investment by households in land and buildings, as against financial assets, may simply be because of the higher after-tax return and lower volatility of the former. After-tax rates of return on various physical and financial assets may vary due to several factors, including the relative inflation rate, volatility in prices and returns, and the state of physical and financial markets. In financial markets, there is a lot volatility if there is a global crisis, and if there is a very high rate of inflation, people would possibly prefer to save more in physical assets rather than in financial assets. So the tax incentive may not be a factor in this case at all. Trends in the household sector’s investment portfolio show an interesting pattern over time. There has been a sharp reduction in bank deposits from 58 percent of the total to 43.6 percent as per Table 3 in the paper. At the same time, the share of provident and pension funds has shown a sharp increase from 10 to 18.3 percent. The authors could explore the possible reasons for this sharp change in the composition of financial savings. Is it due to the high after-tax return or due to the introduction of the National Pension Scheme? In fact, pension funds have longer lock-in periods, unlike fixed deposits, and yet the authors have noted a steady increase in the former over the given time period. 34 INDIA POLICY FORUM, 2018

The authors have used CMIE consumer data, but it is a bit tricky to use household data for separating out taxed from non-taxed households. One of the criteria for identifying taxed households is that at least one member of the family should have an income in excess of the threshold. Can we really categorize a household as tax-paying just because one member has an income of more than `2.5 lakh? Is it possible that the income of this particular member is agricultural income and therefore he is exempt from paying tax? One needs to find a better method of classifying tax-paying households. In the probit regression estimates, the coefficients of fixed deposits are consistently lower than those of the other financial assets. One possible explanation for this is that only a small proportion of the fixed deposits constitutes tax savings. People save for a variety of reasons, not just for saving taxes, and this may be particularly true in the case of fixed deposits. In other words, the tax-saving component of small savings is a very small proportion of the total fixed deposits. A major proportion of fixed deposits is actually for less than five years. It may thus be possible to make some inferences by simply looking at the time profile of fixed deposits. On the whole, this is an important and interesting paper, and I hope this will be the beginning of more incisive analysis by the authors.

General Discussion

Chaired by Arbind Modi Member, Central Board of Direct Taxes

Karthik Muralidharan noted how frustrating it can be talking about something important without data, and thanked the authors for putting the data together. He asked, given that most of the paper’s results were on the intensive margin of the volume of savings, rather than the extensive margin of bringing in new savers, whether it would be possible to also look at the extensive margin of the likelihood of using a formal financial savings instrument. He cited the 2016 IPF paper by Badarinza, Balasubramaniam and Ramadorai1 looking at the distribution of assets of a median household with zero financial savings. Apart from the literature examining the role of tax savings in the overall volume of savings, it would also be useful to think about the number of savers and tax incidence, and therefore where in the distribution one might want to target tax incentives.

1 Reference is at the end of this section. Radhika Pandey et al. 35

Muralidharan made the point about regressivity, something Rajnish Mehra had mentioned, also indirectly what Ila Patnaik was mentioning about tax revenue foregone or tax expenditures. His sense was that a lot of these incentives were regressive. For the most part, the tax incentives were a lump-sum transfer to people who would have already saved. So we could do a fuller analysis of the opportunity cost of these tax expenditures, how we might use the revenue foregone to attain the objectives on the extensive margin. For example, with all the Jan Dhan accounts that are inactive or don’t have much money in them, are there instruments that can crowd in savings on the extensive margin and reduce the regressivity of the tax incentives? This would be worth pursuing. Suman Bery noted that the IPF conference was meant to encourage and improve empirical papers, and so he appreciated the spirit in which Ila Patnaik had acknowledged the comments by the discussants. He also thought that the paper would be enriched by a discussion of the motivation behind the focus on financial assets: Rajnish Mehra located this motivation in terms of retirement. This is a big change from where tax incentives in India basically started, which was resource mobilization—the notion that financial savings would be funneled through the banking system to government projects. At the macro level, he felt that providing the rationale for focusing on financial savings would be important. At the micro level, Bery asked if greater fungibility across financial assets improves household welfare compared to the lumpiness of saving in physical assets. Referring to the Tobin-related literature on portfolio shuf- fling of households in rich countries, he thought the econometrics used the constraint of overall net worth. So he wondered if there was wealth data that the authors could use, making sure that gold was a part of the wealth constraint as seen by the household. He thought that exploring the overall portfolio allocation problem that Indian households face—for example, how they evaluated their wealth, how capital gains enter into this, as also indicated by Mehra—would enrich the paper. Mihir Desai appreciated the topic and the paper. First, the paper drew attention to the great need for access to tax return data for this kind of a study. In the USA, access to such data made possible the heavy oversam- pling of high income households, which needed to be looked at closely for such work. Second, the main story seems to be that substitution is about its effect on intermediation, which is to say that fixed deposits are going down and pension plans are going up. If it is just that, fixed deposits being held inside retirement plans, then there isn’t much happening, but if there are real changes in intermediation, that would be interesting. Third, he thought 36 INDIA POLICY FORUM, 2018 that the identification strategy using households with different tax incentives was quite nice, but he suggested doing more to show that households on other observable characteristics were similar. Finally, on the phase-out of tax incentives, the way the USA works is to limit revenue loss by phasing out the subsidy or phasing out the advantage to high income individuals. India seems to remove the incentive at a certain savings amount. Phase-outs by income would be a smarter way to limit revenue loss and limit substitutability (a 100-lakh income person is more likely to be substituting than a 10-lakh income person). Devesh Kapur had two comments for the Chair. Since this analysis is better done using tax return data, he wanted to know why the CBDT is neither doing the analysis itself nor releasing the data to allow others to do it, even though the issues were central to understanding the cost–benefit of tax incentives, a problem that CBDT itself would like to address. He asked a second question about the extent to which the way the government thinks about insurance-linked tax breaks is related to the government’s huge stake in the Life Insurance Corporation. He thought there was a clear conflict of interest here. Poonam Gupta noted, in line with Rajnish Mehra’s point, that inflation was a confounder in the regression since it is a tax on financial savings. Controlling for that could change the results quite substantially. Market returns would be endogenous to the policy changes being explored. She also wondered if we could see any impact of demonetization on household savings behavior. Sandhya Garg wondered if, in talking about national saving, there was a need to study the savings behavior of people who were not in the tax net, such as agricultural households and zero tax-bracketed ones. She also asked if the authors had considered non-linear estimation because a 1 percent tax reduction would mean different amounts of money available for saving for each income level. Rajesh Chadha asked if the analysis accounted for two types of households: first, those that had net taxable income and paid taxes, and we analyze their saving behavior. But, second, there must be households who are depos- iting money in provident funds and thereby lowering their taxable income and therefore not paying taxes. If we did not analyze their saving behavior, would this make a difference? Renuka Sané thanked everyone for the comments and responded to the ones on data. On the good suggestion about looking at the extensive margin, the Consumer Pyramid data were limited and did not give the rupee value of savings, just whether outstanding investments were in a particular product Radhika Pandey et al. 37 or not. NCAER’s 2004–05 and 2011–12 IHDS data, which is fortunately a panel, did provide a rupee value for savings, so the authors would look at that. Ideally, answering questions on whether cash holdings increased or whether a particular policy had an effect needs continuous panel data. India is seriously short of this kind of panel data: NCAER is the only institution putting out such high-quality national data. On the comment that Rajnish Mehra made on revaluation of assets, she said that they did worry about this for shares, which fluctuate in market value. They had drawn the graphs without excluding shares and saw a very similar pattern because shares were only 3 percent of total financial assets. But they would look carefully at Mehra’s flow data and examine what explained the differences. Ila Patnaik also thanked everyone for the comments, which they would seek to address in their revision. She thought the objective of tax breaks changing from encouraging savings in financial instruments to encouraging retirement savings and pensions, because of the changing demograph- ics of the population, was an important point and something they had not addressed. She hoped that the Chairman, in rewriting the Direct Tax Code, was spending some time thinking about it. She felt that this was related to the paucity of long-term savings: for example, where do we get the finances for infrastructure? So the issue links to the larger macro question of encouraging savings for retirement as well as to financing long-term investment, particularly in infrastructure. Arbind Modi (Chair) chose first to respond to the question on why the tax department does not release tax return data and said this is an old problem, something he had been contending with since he started working on tax policy. Initially, the problem was that the authorities did not have the data. In the last decade or so, a large volume of data has been built up. However, the data are fragmented to the point where the tax authorities themselves are not in a position to mine the data and use them against tax evasion. Only recently have they launched “Project Insight,” which seeks to create a comprehensive database. Once that is operational, the authorities will explore internally how to share the data, whether individual tax return data or some form of summary data. But that call had not yet been taken. Hopefully, in the next one or two years, we should see some movement on this, making possible higher quality work on tax policy. There is one further complication. There is now a lot of international pressure that data confidentiality needs to be maintained because of FATCA and other factors. So the authorities are looking at appropriate legislation that would ensure confidentiality while releasing the data for doing research. He felt quite optimistic about this. 38 INDIA POLICY FORUM, 2018

On the long-standing, thorny question of insurance and tax breaks, he said he had known this problem ever since he joined the Department in 1981. The initial tax incentive was only for life insurance, probably to enable the government to mobilize resources, and the government gave the benefit only to government-managed products. The rationale changed post mid-1990s from mobilization to saving. The Unit Trust of India (UTI) was the first one to be granted the tax exemption, but the UTI was more or less government managed. Then in the late 1980s, when the insurance business was opened to banks, the first was SBI, and that was also when the focus started shifting to savings rather than mobilization. By then, some of these products were so entrenched and vested interest so strong that reform of the savings-related tax provisions had become extremely contentious and difficult. The last time the authorities tried to clean this up was in 2016, and people burnt their fingers. Modi said that another attempt will surely be made in the latest DTC, but some officials involved are worried that this sincere effort may jeopardize the entire DTC. So this remains a difficult issue. The Chair concluded by thanking everyone for participating in this excit- ing discussion and for throwing light on some ways in which we could move forward: at least we know better where we stand. So hopefully we may be able to design better tax policies on the savings side.

Reference

Badarinza, Cristian, Vimal Balasubramaniam, and Tarun Ramadorai. 2016. “The Indian Household Finance Landscape,” India Policy Forum, Vol. 13, 2016–17. New Delhi: National Council of Applied Economic Research. POONAM GUPTA ∗ World Bank JUNAID AHMAD† World Bank FLORIAN BLUM‡ World Bank DHRUV JAIN# World Bank India’s Growth Story§

ABSTRACT India has attained much economic success in the last three decades. Yet economic deceleration in the recent years has generated worried commentaries about India’s growth outlook. In this paper, we offer a long-term macro perspective on India’s growth experience. Analyzing past five decades of data, we note that growth has slowly but steadily accelerated over this period, become less erratic, and has been well diversified across sectors and states. Assessing the period since the early 1990s more granularly, we note three distinct phases of growth. A period of slow acceleration from 1991 to early 2000s; a period of rapid growth with several features of unsustainability during 2004–08; and a corrective slowdown that started with the Global Financial Crisis in 2008. The slowdown was reflected most profoundly in investment, credit, and exports. Even as the economy recovered to a 7–7.5 percent growth rate, durably accelerating it to a higher level will require concerted policy momentum that succeeds in reversing the slowdown in investment, credit supply, and exports, and the support from the global economy. Maintaining hard-won macroeconomic stability, a definite and durable solution to banking sector issues, and realization of the expected growth and fiscal dividend from the Goods and Services Tax are some of the factors that can help attain a higher growth rate. The paper also includes a short annex on India’s new GDP series and comparisons with the old.

Keywords: Development, Economic Growth, India, Investment, Exports, Banking Credit, Macroeconomic Stability, GDP series

JEL Classification: E65, F40, O11, O47, O53

∗ [email protected], † [email protected], ‡ [email protected], # djain3@ worldbank.org § The authors gratefully acknowledge useful comments and suggestions from Barry Bosworth, Dilip Mookherjee, Sudipto Mundle, and participants at the NCAER 2018 India Policy Forum. 39 40 INDIA POLICY FORUM, 2018

1. Introduction

ndia has achieved much economic success in the last three decades. Since Ithe early 1990s, when reforms began, growth rates have accelerated slowly and become more stable. The economy has become more modern and globally integrated, macroeconomic stability has improved, and the average citizen is better educated and lives longer. Yet an economic deceleration in the recent years has generated worried commentaries about India’s growth potential. The questions being raised are: Is the deceleration in economic growth structural or cyclical? Is the Indian growth story over? What is the “new normal” for India’s growth outlook? What sets of policies, structural or cyclical, might be needed to revive growth?1 In this paper, we offer a long-term perspective on India’s growth experience. Looking back at the last 50 years, we analyze India’s long-term growth patterns in different ways and compare India’s growth experience with that of the other large emerging market economies.2 We note the following several stylized facts. First, India’s long-term economic performance has been impressive. Despite variations around the long-term growth rate, average growth over any continuous 10-year period has steadily accelerated and has never reversed for a prolonged period. The acceleration in the growth rate is consistent with India’s steadily improving proximate determinants of long-term growth. Economic growth has also become more stable—both due to growth rates stabilizing within each sector and due to the transition of the economy toward the services sector, which has a more stable growth rate. Second, the long-term growth experience has been balanced and diversified in the sense that acceleration and stability are evident across states; and for the most part, growth is not concentrated in a few uses or sectors, but is visible in most of its components—consumption, investment, and

1. National Accounts data used in the analysis are for the 2011–12 base year. For the years prior to 2011–12, the data are available from the 2004–05 series. We have back-casted these data following the methodology laid out in Appendix A. The analysis does not incorporate the back series with the 2011–12 base year, released by the Central Statistics Office (CSO) for the years 2004–05 to 2011–12, on November 28, 2018. However, Appendix B presents some of the key results if this data revision is considered. Additionally, the analysis is limited to 2016−17 and does not reflect the first revised estimates for the year 2017−18 released on January 31, 2019. 2. Years refer to fiscal years in the paper unless otherwise indicated. For example, 2015 refers to fiscal year 2014–15, which runs from April 1, 2014, until March 31, 2015. GDP refers to GDP at market price, unless otherwise indicated. Poonam Gupta et al. 41 exports; and across sectors. Growth acceleration has been characterized by productivity gains, and not just by an increase in factor inputs. Productivity gains are reflected in labor as well as total factor productivity (TFP). The contribution of productivity gains to growth has increased in the recent decades (Bosworth, Collins, and Virmani 2007). Third, we reconcile the long-term growth potential of the economy with the perception of an ongoing slowdown in the economy. We do so by dividing the post-reform period since the early 1990s into three phases and analyzing the growth rate over each phase. The first phase of growth acceleration lasted from 1991 to 2003, when gross domestic product (GDP) grew at an average rate of 5.4 percent a year. It marked a growth acceleration of 1 percentage point a year over the previous two decades. A short second phase of unusually high growth followed during 2004–08, when growth was aided by rapid global growth and easy global liquidity, and by the impact of important reforms that were undertaken in prior years.3 During this phase, GDP grew at an average annual rate of 8.8 percent, taking it temporarily above the trend growth rate. The period of growth acceleration was marked by a rapid increase in the rate of investment, financed by high credit growth and a surge in capital flows. A final phase of growth slowdown then ensued, aligning with the slowdown in the global economy and the onset of the Global Financial Crisis (GFC) in 2008–09, and continuing till date. The growth slowdown reflected most profoundly in investment, credit, manufacturing, construction, and exports. The period was initially marked by worsening macroeconomic stability due to the fiscal response to the crisis and the broader macroeconomic management of the economy. Macroeconomic stability has improved since then.4 The slowdown has aligned India’s growth rate to the trend growth rate of the pre-boom period. Fourth, even as the economy has slowly reverted to the trend growth rate and stabilized in recent years, the revival is not yet firmly anchored in investment, exports, and the industrial sector. Recovery in investment and credit has been more protracted in India than in other countries, and India

3. This observation is based on our back-casted estimates of NAS using the methodology described in Appendix A. The narrative will change quantitatively but not qualitatively if one uses the 2011–12 back series data for the years 2004–05 to 2007–08 released by the CSO. 4. Macroeconomic stability is measured as a period of low inflation, budget deficit, and current account deficit. 42 INDIA POLICY FORUM, 2018 has lost share in the global export market. This may have implications for accelerating growth to India’s potential and for enhancing potential growth itself. Finally, the steep growth slowdown in the few quarters during 2018 is not a continuation of the long-term growth dynamics. While the deceleration of growth to about 7 percent in recent years is structural, a further decline to below 7 percent in 2017–18 was an aberration.5 This additional slowdown can be attributed to temporary disruptions in economic activity due to the twin policy shocks, as businesses prepared for implementation of the Goods and Services Tax (GST), an important indirect tax reform, and as the economy adjusted to demonetization. There are indications that the economy is recovering, with growth accelerating in the last few quarters. Growth steadily accelerated to 7.1 percent in the second quarter of 2018–19, from 5.6 percent in the first quarter of 2017–18. Analyzing the past episodes of high growth path, we note that there have been few sporadic episodes in the last five decades when growth rates exceeded 8 percent, about once in each decade. Most episodes of acceleration lasted only one to two years and corrected sharply in ensuing years. In some of these, the high growth was due to a base effect of slow growth in previous years followed by an unusually good agricultural output (1976, 1989); in others, it was due to unsustainable macroeconomic policies (such as in 2010–11). Attaining a growth rate of 8 percent or higher on a sustained basis will likely require contributions from all domestic sectors and support from the global economy.6 It will require a concerted reform and policy momentum, wide enough in scope, which succeeds in reversing the slowdown in investment, credit supply, and exports. Maintaining the hard-won macroeconomic stability, a definite and durable solution to the banking sector issues, and realization of the expected growth and fiscal dividend from the GST are other key components of attaining a growth rate of 8 percent or higher. As

5. See Note 1 for the data used. 6. Arvind Panagariya has highlighted the importance of reviving bank credit to reach growth rates exceeding 8 percent: (https://blogs.timesofindia.indiatimes.com/toi-edit-page/ how-to-revive-bank-credit-government-should-to-begin-with-offer-psbs-bonds-in-return- for-equivalent-equity/). In a recent interview, Arvind Subramanian (Chief Economic Advisor) indicated that reaching growth rates of around 8.5 percent is conditional on a reform agenda that addresses the banking sector and other issues (https://www.livemint. com/Politics/OUuLehx0uBAO32P1xhSYAN/India-can-return-to-85-growth-rate-Arvind- Subramanian.html). Poonam Gupta et al. 43 pointed out by the World Bank’s Systematic Country Diagnostic for India, a reform focus on moving to a more resource-efficient growth path, making growth more inclusive, and enhancing the effectiveness of the Indian public sector can ensure that these rates are sustained in the decades to come, moving more and more Indians into a status comparable to that of the global middle class. This paper proceeds as follows. Section 2 focuses on India’s long-run growth dynamics. Section 3 summarizes three phases of growth experienced by India since the early 1990s. Section 4 discusses the ongoing slowdown in parts of the economy and the policy challenges in reversing the slowdown. Section 5 concludes the paper.

2. India’s Long-term Growth Dynamics

2.1. Accelerating and Stabilizing Growth Rates Below, we look at the trends in the pace of economic growth in India starting in 1971.7 The long-term average growth rate has accelerated slowly in India and despite significant variation around the long-term average, the growth rate has never reversed for a prolonged period (Figure 1). We ask whether the growth acceleration is unique to India or if it has also been the experience of other emerging markets. For this, we compare the linear trend in India with the trend in seven large emerging economies, Brazil, Russia, South Africa, Malaysia, Mexico, Turkey, and Indonesia, which we refer to as EM7.8 We estimate regressions of the following form:

GDP Growth TrendIndiaTx rend (1) it =+ββ01 ti++βε2 tit

The outcome variable in Equation 1 measures the 10-year rolling aver-

age of GDP growth in country i in year t. The coefficient of interest, b2, measures the difference in the slope of growth acceleration between India

7. The source of the data is the CSO. See Appendix 1 for details of the data used and on how we spliced the GDP series for different base years. 8. According to the World Development Indicators (WDI), these countries accounted for 12 percent of the world population (30 percent when India is included), 13 percent of world GDP (20 percent when India is included), and an average per capita income of $16,678 (in 2011 PPP $) in 2016. 44 INDIA POLICY FORUM, 2018

FIGURE 1. India’s Growth Rate Has Consistently Accelerated over the Long Run

1A: GDP Growth Has Accelerated Over the Long Run 1B: Per Capita Income Growth has Accelerated Too

13 y = 0.109*** x – 3.0 10 y = 0.13*** x + 0.50 11 t stat: 3.65, R² = 0.25 8 t stat: 4.40, R² = 0.324 9 6 7 4 5 2 % % 3 0 1 –2 –1 –4 –3 –6 –5 –8 –10 –7 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 Real GDP Growth Real per Capita GDP Growth

1C: A Clear Trend of Growth Acceleration over the Long 1D: In Terms of Relative Prosperity, the Indian Economy Run Is Evident in 10-year Averages Has Shown Convergence 4.0 9 10-Year Rolling Average: GDP Growth 3.5 8 3.0 7 2.5 2.0 6 % 1.5 5

% y = 0.1148*** x + 3.617 1.0 4 t stat: 21.7, R² = 0.93 0.5 3 0.0 2 1 1971 1974 1977 1980 1983 1986 1989 1992 1995 1998 2001 2004 2007 2010 2013 2016 India’s GDP as a Share of World GDP 0 India’s PCY as a Share of US's PCY

1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 2011 2014 2017 (Constant 2010 US$)

Source: Data are from the CSO and WDI. Note: In Panel C, the 10-year rolling averages of growth rate are for the current year and the preceding nine years. Years refer to fiscal years in Panels A–C and to calendar years in Panel D. *** p<0.01, ** p<0.05, * p<0.1. and the average EM7 country. We find that, as compared to a significant coefficient of 0.114 for India (Table 1, Column 2), the coefficient of a similar linear trend for the 10-year average growth rates for other large emerging markets is negative (column 1). Column 3 of the table shows that the difference between trend coefficients is statistically significant, with India having a significantly higher growth acceleration than other emerging market economies. Although the pace of growth acceleration has differed across sectors, India’s growth pattern has been broadly diversified. The pace of acceleration has been fastest in services, followed by industry, and there has been no evident pattern of acceleration in agriculture. The most remarkable achievement in agriculture has been the greater stability of growth, but Poonam Gupta et al. 45

TABLE 1. Trend in the Pace of Long-term Growth of India and EM7 Countries (1) (2) (3) GDP Growth 10–Year GDP Growth 10–Year GDP Growth 10–Year Variables Rolling Averages Rolling Averages Rolling Averages Trend –0.027** 0.114*** –0.027** (2.98) (19.95) (3.19) India * Trend 0.142*** (6.29) Countries EM7 India EM7 and India Country Fixed Effects Yes No Yes Observations 240 37 277 R-squared 0.478 0.930 0.550 Source: WDI and authors’ calculations. Note: Robust t-statistics (controlling for country-level clusters in Columns 1 and 3) are in parentheses. Columns 1 and 2 present estimates of a regression of real GDP growth, calculated as a 10-year rolling average, on a linear time trend. The 10-year rolling averages of growth rates are for the current year and the preceding nine years. *** p<0.01, ** p<0.05, * p<0.1. not necessarily a higher average growth rate (Figure 2). Consistent with the experience of other countries, the contribution of agriculture and allied activities in GDP growth has declined, while that of the nonagricultural sectors has increased. The exceptionally fast growth of the services sector in India has been accounted for, in a large part, by modern services, compris- ing financial services, communications, and the IT sector, as highlighted by Eichengreen and Gupta (2011). The fact that growth has not just accelerated but has also become more stable over time is reflected in its steadily declining standard deviation, and a declining coefficient of variation (Figure 3).9 Particularly remarkable is the sharp increase in the stability of GDP growth in the postreform period since 1991.10 Even if growth accelerated episodically in the decades prior to 1991, it was punctuated by large annual variations and often failed to sustain. Thus, growth has not just accelerated post-liberalization but has also become more stable.

9. While the figure only shows the coefficient of variation, the results are very similar for standard deviation. 10. Figure 3 also documents a decline in the coefficient of variation in the 1980s, coinciding with some acceleration in growth in the 1980s. 46 INDIA POLICY FORUM, 2018

FIGURE 2. Growth Rates Have Accelerated and Become More Stable across Sectors

2A: A Consistent Acceleration in Growth Has Not Been 2B: …But Is Evident in Industry Observed in the Agriculture Sector… 20 14 y = 0.09*** x + 3.63 15 12 t stat: 2.69, R² = 0.159 10 10 8 5 6 % 0 % 4 2 –5 0 y = 0.0217x + 2.4797 –10 –2 R² = 0.0029; t-stat–0.36 –15 –4 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 Growth Rate of Agriculture Growth Rate of Industry

2C: Acceleration in Services Has Been the Fastest 2D: Services Have Emerged as the Largest Contributor to 16 y = 0.097***x + 5.67 GDP Growth, Followed by Industry t stat: 3.63, R² = 0.21 14 Contribution to Growth 12 8 10 6 8 4

% 2 6

% 0 4 –2 2 –4 –6 0 –8 –2 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 Growth Rate of Services Agriculture Industry Services

Source: CSO data. Note: Years refer to fiscal years. Agriculture includes crop, livestock, forestry, and fisheries; the industrial sector includes mining and quarrying, manufacturing, electricity, gas, water and other utilities, and construction; services include trade, hotels, transport, communication and services related to broadcasting, financial, real estate and professional services, and public administration, defense, and other services.

2.2. Spatial Trends in Growth Next, we examine growth patterns at the state level.11 For this, we estimate the trend coefficient of Gross State Domestic Product (GSDP) growth for annual data since 1981. We use the following regression equation, where the subscripts s and t denote variations at the state and year level, respectively:

GSDP Growth Trende()quation 2 st =+ββ01 ts++γεit (2)

11. The states included in this analysis are Andhra Pradesh, Arunachal Pradesh, Assam, Bihar, Goa, Gujarat, Haryana, Himachal Pradesh, Karnataka, Kerala, Madhya Pradesh, Maharashtra, Manipur, Meghalaya, Odisha, Punjab, Rajasthan, Tamil Nadu, Tripura, Uttar Pradesh, and West Bengal. We construct a back-casted panel on annual real gross state domestic product growth for the years 1981 to 2016. Poonam Gupta et al. 47

FIGURE 3. India’s Long-term Growth Rate Has Become Increasingly More Stable

3A: Coefficient of Variation, 3B: Coefficient of Variation, GDP Growth, GDP Growth, India EM7 Median 10-Year Rolling Coefficient of Variation 10-Year Rolling Coefficient of Variation 160 160 140 140 120 120 100 100 80 % 80 % 60 60 40 40 20 20 0 0 1980 1983 1986 1989 1992 1995 1998 2001 2004 2007 2010 2013 2016 1980 1983 1986 1989 1992 1995 1998 2001 2004 2007 2010 2013 2016

Source: WDI data. Note: Coefficient of variation is calculated as the standard deviation divided by the mean for rolling 10-year periods. For EM7, it is the median of the cross-country series for every year.

To control for level differences in growth across states, we include state-level fixed effects in the regressions, denoted by s. A positive and significant coefficient estimate for b1 presents evidence of growth acceleration. We also allow the trend coefficient to vary by state characteristics (equation 3). The characteristics we consider are the share of agricultural/ non-agricultural sectors in the state economies; and an indicator for whether a state’s per capita GDP is above the median per capita GDP across all states. All characteristics are measured for the initial year 1981 and are thus time-invariant.

GSDP Growth Year Characterstic x Year (3) st =+ββ01 ts++βε2 tit

The coefficient b2 measures the differential growth acceleration between states with and without a given characteristic. Table 2 highlights that India’s growth acceleration is reflected in the growth of an average state. On average, state-level growth accelerated by 0.09 percentage points per year between 1981 and 2016 (Table 2, Column 1). Our analysis does not detect differences in growth acceleration across agricultural and non-agricultural states, and across income levels. We further examine whether the growth stabilization documented at the national level is visible at the state level by estimating regressions similar to above but replacing the outcome variable with a 10-year rolling coefficient 48 INDIA POLICY FORUM, 2018

TABLE 2. Trends in the Pace of Long-term Growth at the State Level (1) (2) (3) (4) (5) Growth Growth Growth Coefficient Coefficient Variables Rate Rate Rate of Variation of Variation Trend 0.089*** 0.099** 0.089*** –2.96*** –3.56*** (9.27) (2.47) (5.99) (4.53) (3.22) Ag. share (1981) × Trend –0.0002 (0.251) GDP per capita > Median × Trend –0.0012 1.28 (0.064) (1.03) State-fixed Effects Yes Yes Yes Yes Yes Observations 734 734 734 567 567 R-squared 0.062 0.063 0.063 0.628 0.636 Source: Authors’ calculation based on data from CEIC Data Company Ltd. Note: In Column 3, we allow the linear trend coefficient to vary between richer and poorer states, with richer states defined by an indicator variable that takes the value 1 if a state has above median per capita GSDP in the initial year 1981. of variation for each state.12 The last two columns in Table 2 confirm that the variability in state-level growth has consistently declined over time, and that there is no statistically significant difference in the pace of increased growth stabilization between richer and poorer states. To understand differences in the drivers of growth, we examine differential trends between agricultural and non-agricultural, and richer versus poorer states, in the investment share of GSDP. The results are presented in Table 3. The table highlights that there is no statistically significant trend in the investment share of GDP since 1991 (the earliest observation in our dataset). However, the aggregate results in Column 1 of the table mask significant heterogeneity. States with below median per capita GSDP in 1981 experienced a declining trend in the investment to GDP ratio since 1991 that is statistically significant, but not the states with above median per capita income levels.

2.3. Composition of GDP Looking at the composition of GDP on the use side, the main trend that emerges is that of a consistently declining share of consumption in GDP, particularly the share of private consumption, while the share of investment

12. The coefficient of variation is calculated as the standard deviation of GSDP growth over rolling 10-year periods (the year of observation and the nine years preceding it), divided by the sample average of GSDP growth over the same period. Poonam Gupta et al. 49

TABLE 3. Trends in Investment at the State Level (1) (2) Variables Investment per GSDP Investment per GSDP Trend –0.596 –1.417** (1.164) (2.239) Ag. share (1981) × Trend GDP per capita > Median × Trend 1.642* (1.761) State-fixed Effects Yes Yes Observations 400 400 R-squared 0.196 0.196 Source: Authors’ calculation based on data from CEIC Data Company Ltd and Reserve Bank of India. Note: In Column 2, we allow the linear trend coefficient to vary between richer and poorer states, with richer states defined by an indicator variable that takes the value 1 if a state has above median per capita GSDP in the initial year 1981. and exports has increased (Figure 4). While private consumption accounted for nearly four-fifths of GDP in the early 1970s, this share declined to about three-fifths in 2017.13 After a small increase over recent decades, government expenditure has stabilized at nearly 12 percent of GDP. Equally salient is the steady increase in the rate of investment until the mid-2000s. The rate of investment peaked at nearly 36 percent in 2007–08, but in the last few years it has declined to a rate more aligned with the long-term trend rate. Historically, public and private investment contributed approximately about an equal amount to total investment, but the role of public investment in growth has diminished over time. After peaking at nearly 13 percent of GDP in 1986–87, public and private investment started to diverge, with public investment accounting for approximately only a quarter of total investment in recent years. India has also become more integrated into the global economy, with its trade ratio—the ratio of exports and imports to GDP—adding up to about 40 percent of GDP in 2017, five times the ratio of 7.6 percent in 1971, yet lower than its peak value of 57 percent in 2014. Exports as a percentage of GDP tripled from 7.3 percent in 1991 to 22 percent in 2007 and were 25.5 percent of GDP in 2014. The contribution of net exports to growth has been muted with import growth exceeding export growth in a majority of years.

13. Despite its declining share, consumption growth has been a key driver of aggregate GDP growth, contributing on average 3.76 percentage points to growth annually between 1971 and 2017. 50 INDIA POLICY FORUM, 2018

FIGURE 4. Consumption Share in GDP Has Declined, While the Share of Exports and Investment Has Increased

4A: Share of Private Consumption in GDP Has Declined; 4B: Share of Private Investment in GDP Has Increased Government Consumption Has Been Stable while That of Public Investment Has Declined 100 40 80 35 30 60 25 %

% 20 40 15 10 20 5 0 0 1971 1974 1977 1980 1983 1986 1989 1992 1995 1998 2001 2004 2007 2010 2013 2016 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 Total Consumption as a % of GDP Investment Rate Public Sector Private Consumption as a % of GDP Private Sector Govt. Consumption as a % of GDP

4C: Share of Exports and Imports in GDP Has Increased 4D: Consumption Growth Remains a Key Contributor as the Economy Has Progressively Opened Up to Growth, Followed by Investment 35 Contribution to GDP Growth 30 8 25 6 20 4 %

% 2 15 0 10 –2 5 –4 0 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 Exports as a % of GDP Imports as a % of GDP Consumption Investment Net Exports

Source: CSO data. Note: Years refer to fiscal years. Investment rate is defined as gross fixed capital formation (GFCF) as a percentage of GDP. Net exports are the difference between exports and imports of goods and services.

2.4. Sources of Growth: Inputs and Productivity Growth To understand the underlying determinants of India’s growth trend, one may decompose GDP growth into input usage and TFP. A common and simple growth accounting exercise decomposes GDP growth into use of labor and capital, and TFP using a Cobb–Douglas production function. TFP is estimated as the residual after accounting for labor and capital:

∆∆A Y ∆∆K L t =−t ααt −−()1 t At Yt Kt Lt

More recent growth accounting exercises have extended this framework by considering other forms of the production function, a richer set of factor inputs, allowing not just for the quantity of inputs but also adjusting for Poonam Gupta et al. 51 differences in input quality. For example, Bosworth, Collins, and Virmani (2007) allowed for time-varying factor shares and nonunit returns to scale in the production function. Caselli and Coleman (2001) considered a Constant Elasticity of Substitution (CES) production function. Hall and Jones (1999) augmented labor inputs for human capital, traditionally measured as the working population, by defining labor input as a function in the years of schooling. Further extensions have attempted to account for differences in schooling quality (e.g., Klenow and Rodriguez-Clare 1997; Bils and Klenow 2000) and differences in the quality of the physical capital stock (e.g., Bosworth and Triplett 2007). Finally, various contributors have argued that TFP might differ across sectors, calling for the need to obtain estimates at the sectoral or industry level. For example, Bosworth and Triplett (2007) and Triplett and Bosworth (2003) account for industry-level growth in the USA. Empirical estimates for India highlight an acceleration of TFP growth in the early 1980s, followed by a further acceleration in the post-reform period. Bosworth, Collins, and Virmani (2007) construct growth accounts for India for the period from 1960 to 2004 and find evidence for a strong acceleration in TFP growth (Figure 5, Panel A). The contribution of TFP growth was highest in the post-reform period and remained a significant contributor to GDP growth until 2004, the latest year included in their analysis. They also find that India’s growth since 1980 was fueled by a rapid expansion of TFP in services, while productivity increases in Indian agriculture were modest, and industrial growth relied on employment increases and experienced comparatively low TFP gains. In addition, decomposing improvements in output per worker, Bosworth, Collins, and Virmani (2007) find that the reallocation of workers from less to more productive sectors contributed approximately 1 percent per year to output growth in the 1990s, but gained importance in the 2000s.14 The analysis by Bosworth, Collins, and Virmani (2007) is insightful. While an extension of the analysis for more recent years would be very useful, we are constrained by the scope of this paper and the effort involved in constructing comparable data. We, however, decompose GDP growth using the simple Cobb–Douglas production function with capital and (unskilled) a 1–a labor as inputs, and a constant capital share of 0.3: Yt = AtKt (Lt) , where a is assumed to be 0.30, and TFP is estimated as the Solow residual. Consistent

14. Bosworth and Collins (2008) compare the Indian growth experience with China and highlight the more significant role of TFP increases in the early years of growth for India, as growth in China depended more on capital accumulation. See also Young (1995) and Young (2003), who argue that accounting for biases in official deflators and the measurement of human capital, productivity growth in China was muted, and Brandt and Zhu (2010) for a more recent update of Young’s (2003) calculations. 52 INDIA POLICY FORUM, 2018

FIGURE 5. Sources of Growth—Inputs and Productivity

5A: Decomposition of Growth into Factor Inputs and Total Factor Productivity

Growth Accounts for India Decomposition of Output per Worker 8 7 7 6 6 5 5 4

% 4 % 3 3 2 2 1 1 0 0 1960–73 1973–83 1983–93 1993–99 1999–2004 1960–73 1973–83 1983–93 1993–99 1999–2004 Employment Physical Capital Education TFP Within Sector Growth Reallocation Effects

5B: Labor Productivity Growth in India: Reallocation and Within-Sector Gains

12 10 8 6

% 4 2 0 –2 –4 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 Within Sector Gains Reallocation of Labour Total Change in Productivity

Source: Figure 5A is from Bosworth, Collins, and Virmani (2007). Figure 5B is based on authors’ calculations, data are from the CSO. Employment statistics are estimates provided by the International Labor Organization, available for 1991 to 2017. Note: Years refer to fiscal years. with Bosworth, Collins, and Virmani (2007), the results further highlight that the growth momentum in India since the 1990s has been fundamentally supported by increases in TFP, which on average accounted for 60 percent of overall growth between 1990 and 2011, and has since again emerged as a key driver of growth. Our growth accounts also reflect that investment rates in India have slowed more recently, which has reduced the role of capital accumulation in driving growth. In addition, increases in labor inputs have only been a modest driver of aggregate growth in recent years, as the contribution of employment growth stabilized at around 1 percent a year after the financial crisis. Both the diminishing role of capital accumulation and the comparatively limited importance of human capital in driving growth contrast the Indian growth experience to that of East Asia, as especially China relied on strong investment and capital accumulation. Poonam Gupta et al. 53

Turning to the sources of labor productivity, India has experienced two significant boosts to labor productivity, the first one commencing in 1993 and the second one in 2003 (Figure 5B). The rate of productivity increase during these episodes is larger than that experienced by the East Asian countries during the periods of very high growth, but is smaller than the labor productivity increases realized in China, which increased output per worker by 8.5 percent between 1993 and 2004, compared to 4.6 percent in India.15 Gains in labor productivity may be attained due to the reallocation of labor toward sectors with higher productivity. Such reallocation can help overcome the misallocation of factor inputs to comparatively unpro- ductive sectors and firms.16 Alternatively, labor productivity gains may occur due to workers becoming more productive within their sectors, for example, due to labor-augmenting capital accumulation or technology improvements.17 We compare the contribution of labor reallocation across sectors and the within-sectoral productivity gains to explain aggregate improvements in labor productivity for data extending until 2015. Over India’s two phases of high labor productivity growth, within-sector productivity improvement has been the key driver of growth in labor productivity (Figure 5B). Until the early 2000s, reallocation contributed only approximately 1 percentage point to annual growth. Even though productivity increases driven by labor reallocation have grown in importance since the early 2000s, the contribution of labor reallocation to total labor productivity gain has remained relatively modest, at around 1.5 percent.18

2.5. Long-term Proximate Drivers of Growth In this section, we discuss the proximate factors that have likely contributed to India’s steady economic growth. The Commission on Growth and Development (2008) identified the following factors as the correlates of high

15. See Bosworth and Collins (2008) for a discussion. 16. See Hsieh and Klenow (2009) for a discussion of the potential magnitude of these effects in manufacturing. 17. Within-sector productivity gains are likely to be substantive on aggregate, as evidence from development accounting exercises points to the fact that cross-country differences in income levels are more likely to be explained by sectoral productivity differences instead of the sectoral composition of the economy (Caselli 2005). 18. This contrasts with earlier periods, for which the literature estimates that reallocation contributed approximately 1 percentage point to annual growth until 2001 (Bosworth, Collins, and Virmani, 2007). 54 INDIA POLICY FORUM, 2018 and sustained growth: openness to trade and knowledge, macroeconomic stability, high investment and saving rates, efficient market allocation wherein prices guide resources and resources follow prices, and an enabling institutional, administrative, and governance environment. A review of the literature indicates that several of these factors have likely been instrumental in India’s growth experience. First, even though India’s trade to GDP ratio was persistently low for a few decades after independence, it experienced unprecedented growth from the early 1990s until the GFC (Figure 6A). Second, India has, and will likely continue to, benefit from a growing working age population, with the share of the population in working age having increased by more than 10 percentage points between 1970 and 2016. Third, India has benefitted from an increase in the savings and investment rate, which continued until the late 2000s. Fourth, evidence indicates that financial development is not only a by-product of growth but can also foster growth and development through its effect on factor accumulation and productivity. After independence, India started off with comparatively low levels of financial development as measured by its credit to GDP ratio. It has since, however, experienced two significant and stable phases of growth, one ranging from approximately 1960 to 1980 and the other from early 2000s until the GFC (Figure 6B). Financial development is also evident in financial access to individuals: while the country retains a relatively low rank among the EM7 with regard to the coverage of bank accounts in the population, it has experienced among the highest expansion rates of bank account coverage between 2011 and 2014. Fifth, India is considered to have strong and reliable institutions and a comparatively effective bureaucracy. Building on the institutional view of economic development, India’s growth has likely been aided by its institutional base.

2.6. Demographic Structure Demographic structure is a central determinant of a country’s active human capital, which, in turn, determines growth and growth potential. In this section, we provide a brief description of India’s population structure in comparison to the EM7 from 1960 until today. For a more detailed discussion, see, for instance, Bhagat (2014) and Chandrasekhar, Ghosh, and Roychowdhury (2006). Figure 7 plots the share of different age groups in the total population for India and for the median of EM7 countries. While India initially benefitted from a higher working age population share in 1960 compared to the EM7, Poonam Gupta et al. 55

FIGURE 6. Proximate Determinants of Growth in India

6A: India Has Become More Integrated with the 6B: Financial Development in India Has Increased Over Global Economy the Years19 Trade Openness Domestic Credit Provided by Financial Sector (as % of GDP) 70 100 60 50 80 40 60 % % 30 40 20 20 10 0 0 1960 1964 1968 1972 1976 1980 1984 1988 1992 1996 2000 2004 2008 2012 2016 1960 1964 1968 1972 1976 1980 1984 1988 1992 1996 2000 2004 2008 2012 2016 India Median in EM7 Countries India Median of EM7

6C: Investment Rate Has Increased Over the Years 6D: Saving Rate Has Also Increased Investment Rate Saving Rate 40 40 35 35 30 30

% 25

% 25 20 20 15 15 10 10 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 India Median in EM7 countries India Median in EM7 countries

6E: India Has Strong and Reliable Institutions Rule of Law Index* Government Effectiveness** 0.1 1

–0.1 0.5

–0.2

–0.3 0

India Median of EM7 India Median of EM7

Source: Data are from WDI and IMF International Financial Statistics. The latter’s date range is 1960– 2011. *: World Bank Worldwide Governance Indicators, Kaufmann et al. (2010). Date Range: 1996–2017 (with gaps); **: Political Risk Services. Date Range: 1996 to 2017. Note: Bars show the average (of the median) over time in the case of EM7.

19. Domestic credit provided by the financial sector includes all credit to various sectors on a gross basis, with the exception of credit to the central government, which is net. The financial sector includes monetary authorities and deposit money banks, as well as other financial corporations, such as finance and leasing companies, moneylenders, insurance corporations, pension funds, and foreign exchange companies. 56 INDIA POLICY FORUM, 2018

FIGURE 7. India’s Demographic Structure

7A: India’s Working-Age Population Share Has 7B: ... While India Maintains a Higher Youth Share Increased Slower than in EM7… in Its Population.… Population Aged 15–64 (% of Total) Population 14 and under (% of total) 70 45

Median of EM7 40 65 India 35 30

% 60 % 25 Median of EM7 India 55 20 15 50 10 1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 2011 2014 2017 1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 2011 2014 2017

Source: Data are from WDI. this share has increased slower than in the EM7 (Figure 7A). This slower increase was not driven by a comparatively higher share of older people exiting the labor force but rather by a slow decline in the youth share in the population (Figure 7B). This finding has implications for future and current growth. On the one hand, a comparatively low labor force share means that dependency ratios are high, which, in turn, can constrain contemporaneous savings and investment. On the other hand, India still holds considerable potential to realize a demographic dividend when population dynamics change, which can imply favorable growth conditions in the future. Below we look at India’s growth record since the early 1990s more granularly in order to reconcile its positive long-term growth trend with the perceived growth deceleration in recent years.

3. Three Phases in India’s Growth Trajectory since the Early 1990s

Economic growth in India since the early 1990s has been characterized by the pace of domestic reforms, the global economic environment, and the stance of macroeconomic policies. We divide the record of the Indian economy in the last two-and-a-half decades into three phases. These phases are defined broadly by India-specific and global events. We identify a first phase of growth from 1991 to 2003, when GDP grew at an average rate of 5.4 percent a year, marking a growth acceleration of 1 percentage point a year over the previous two decades. A short phase of unusually high growth followed during 2004–08, when growth was aided by rapid global growth and excess global liquidity, and by the impact of Poonam Gupta et al. 57 important reforms that were undertaken in previous years. GDP grew at an average annual rate of 8.8 percent during these five years. A third phase of growth slowdown then ensued, aligning with the slowdown in global growth rates and the onset of the GFC in 2008–09.

3.1. The Phase of Rapid Growth Acceleration Watershed reforms were undertaken in India starting in the early 1990s after the balance of payments crisis in 1991. These reforms changed the economic structure and the regulatory framework of the economy in a profound way and helped accelerate annual GDP growth to 5.4 percent a year, marking a growth acceleration of 1 percentage point a year over the previous two decades (Figure 8).20 Starting with the devaluation of the rupee, reforms in the 1990s included industrial deregulation; opening of the economy to foreign direct investment and eventually also to other forms of capital flows; trade liberalization; tax reforms; reduction in financial repression through deregulation of interest rates and reduction in the statutory preemption of bank credit; and continued evolution and modernization of monetary policy, while reducing fiscal dominance.21 A short phase of unusually high growth followed during 2004–08, when growth was aided by rapid global growth and excess global liquidity, and

FIGURE 8. Three Phases of Growth

8A: Growth Accelerated to an Average Rate of 5.4% 8B: Real Per Capita GDP Growth Followed the During 1991–2003, followed by 8.8% During 2004–08, GDP Growth Trajectory and a Slowdown Thereafter 10 7.2 13 5.5 8.8 3.4 5.4 7.1 5 2.0 8 4.3

% 0

% 3

–5 –2

–7 –10 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 Real GDP Growth Average 1971–1990 Real per capita GDP Growth Average 1971–1990 Average 1991–2003 Average 2004–2008 Average 1991–2003 Average 2004–2008 Average 2009–2017 Average 2009–2017

Source: CSO data. Note: Years refer to fiscal years.

20. For a growth narrative of the decades prior to 1991, see Panagariya (2004) and Mohan (2008). 21. See Mohan and Kapur (2015) for a discussion of these reforms. 58 INDIA POLICY FORUM, 2018 by the impact of these important reforms. GDP grew at an average annual rate of 8.8 percent during these five years. Indications of high growth were visible in all major sectors of the economy, and in the sources of financing. Manufacturing growth was robust, the investment rate peaked at 36 percent, export volume increased rapidly, and India increased its share of the world exports markets for both goods and services to levels higher than ever before. Even though an impressive feat on growth, the period was characterized by unusually high credit growth; and, in synchronization with other emerging markets, an unprecedentedly large volume of capital flows. A third phase of growth slowdown then ensued, aligning with the slowdown in global growth rates and the onset of the GFC in 2008–09.22 During this period, global growth turned negative, global trade volume declined and remained suppressed for years thereafter, and global liquidity froze temporarily. Indian growth adjusted to a lower level. After the GFC, India’s growth drifted down to about 7 percent, and some of the same drivers of growth that had played a prominent role during the pre-crisis boom were the ones that accounted for the slowdown. The slowdown was most pronounced in investments and exports, both of which more than halved their contribution to growth. Below, we analyze the period of economic boom and the succeeding period of deceleration, situating it in a global context. We compare the Indian experience with that of the other large emerging markets and track its constituents. Comparing growth acceleration in 2004–2008 with that of other emerging countries, we note that the spurt in the growth rate that India experienced during this period was larger than in many other emerging countries (Figure 9). Starting from a modest level, its credit-to-GDP ratio increased rapidly, surpassing the levels in EM7 countries. The rate of investment in India also outpaced the rate in EM7 countries, and India’s share in world export markets increased at a pace faster than in other emerging markets. The growth exuberance and the “credit bubble” were partly financed by large capital inflows during this period.23

22. Mundle, Rao, and Bhanumurthy (2011) and Mohan and Kapur (2015) point out that the pace of economic growth in India had started drifting down even before the GFC manifested itself fully with the collapse of Lehman Brothers in September 2008. 23. Gupta (2016) notes a rapid pickup in capital inflows to India starting in the early 2000s. The surge in capital inflows during 2003–04 to 2007–08 was prominently evident in all forms of capital inflows—portfolio flows, FDI flows, and other flows. Capital inflows accelerated to an average $44 billion a year during 2004 and 2008, compared to $10 billion a year in three prior years. At their peak in 2007–08, capital flows exceeded $100 billion in one year. The pace of capital inflows mirrored global trends and were thus vulnerable to reversal. There was a sudden stop of capital flows in 2008–09, when capital flows declined precipitously to $7 billion. Poonam Gupta et al. 59

FIGURE 9. India Grew Faster Prior to the GFC, and the Correction Was Sharper after the GFC

9A: GDP Growth Was Far More Rapid in India Prior to the GFC 9B: Investment Growth in India Outpaced Growth in the EM7 before the GFC, and the Correction Was Sharper… 12 10 30 8 25 6 20 15

% 4

% 10 2 5 0 0 –2 –5 –4 –10 2000 2002 2004 2006 2008 2010 2012 2014 2016 2000 2002 2004 2006 2008 2010 2012 2014 2016 GDP Growth (India) GDP Growth (Median of EM7) Investment Growth (India) Investment Growth (Median of EM7)

9C: …As Was the Case with Credit Growth 9D: …and Growth of Exports

30 30 25 25 20 20 15 10 % % 15 5 10 0 –5 5 –10 –15 0 2016 2000 2002 2004 2006 2008 2010 2012 2014 2008 2000 2001 2002 2003 2004 2005 2006 2007 2009 2010 2011 2012 2013 2014 2015 2016 Domestic Credit Growth (India) Exports Growth (India) Domestic Credit Growth (Median of EM7) Exports Growth (Median of EM7)

Source: CSO and WDI. Note: Data in the figures are for calendar years. Credit growth is nominal.

In econometric analysis (not reported here for brevity), we find that investment growth had a sharper correction in India and picked up in the years when government expenditure grew, which is indicative of a boost through public rather than private investment. While credit to the private sector as a percentage of GDP remained resilient to the GFC in the initial years after the crisis, it has since declined, and the growth rate of private sector credit has been consistently lower than in comparator countries. Interestingly, as credit growth slowed in other countries in 2008 itself, in India it remained high until later. As we discuss further, export growth slowed in India due to the global slowdown in trade and India’s decreasing share in world exports.24

24. Some features of the economy during this period look similar to those pointed out in the literature as being associated with a credit boom, surge in capital flows and in investment levels. A significant percentage of such episodes result in growth slowdown (see, for example, Dell’Ariccia et al. 2011). 60 INDIA POLICY FORUM, 2018

3.2. Interstate Patterns of Growth during and after the GFC To further identify the characteristics of the slowdown after the GFC, we analyze how economic growth across Indian states was impacted by the GFC.25 Unsurprisingly, we see exactly the kind of economic cycle in economic growth, credit growth, investment, and manufacturing sector at the state level as is evident at the national level.26 The average (mean as well as median) growth rates of all of these variables increased prior to the crisis, during 2004–08, followed by a correction that started with the global economic slowdown in 2007–08; and precipitated when the GFC took hold, with the collapse of Lehman Brothers in September 2008. While GSDP growth recovered in postcrisis years, credit growth, investment, and manufacturing growth remained subdued. We ask whether there were any specific state-level characteristics that correlated with the impact of the GFC on the states. We define the states’ dependence on agriculture, the relative importance of manufacturing in economic activity, and the credit-to-GSDP ratio as an indicator of the states’ dependence on credit, and the rate of credit growth prior to the GFC (between 2004 and 2008) as an indicator of the prevalence of a credit boom in states in years prior to the GFC. While in our main specifications we compare states above and below median for these characteristics, in robustness tests we also define the states that are in top one-third or bottom one-third of the respective state characteristics, or include the continuous measure of these characteristics. We note that the growth cycle around the GFC was more pronounced in states less dependent on agriculture.27 Similarly, comparing states across different manufacturing shares indicates that the states with a larger manufacturing sector experienced a sharper growth slowdown (the figure is not shown for brevity). The dynamics of growth and investment also correlate with the states’ credit dependence, or the pace of credit growth prior to the

25. Data on Gross State Domestic Product are from the CMIE’s database on the states of India. While India has a total of 36 states and union territories, we restrict our analysis to the 20 large states, including Andhra Pradesh, Assam, Bihar, Chhattisgarh, Gujarat, Haryana, Himachal Pradesh, Jammu and Kashmir, Jharkhand, Karnataka, Kerala, Madhya Pradesh, Maharashtra, Odisha, Punjab, Rajasthan, Tamil Nadu, Uttar Pradesh, Uttarakhand, and West Bengal. Our sample covers the years 1990 to 2015 for all states except the younger states, that is, Jharkhand, Chhattisgarh, and Uttarakhand, for which credit data are only available from 2001 onward. 26. Due to the unavailability of data for exports for each state, we cannot confirm the patterns in exports. 27. See also Kumar and Subramanian (2011). Poonam Gupta et al. 61 crisis. During the GFC, growth and investment were impacted less adversely in states with less dependence on credit (Figure 10). To investigate these relationships more systematically, we test whether there were significant differences in growth trends before and after the GFC across states with higher and lower credit dependence and credit growth. In the spirit of a difference-in-difference approach, we estimate the following regression model:

GDP Growth StateType X Post GFC (4) it =+ββ01 it++γµti+εit

gt and mi denote year and state-level fixed effects, respectively. The coefficient of interest, b1, captures the differential trend after the GFC on states with a certain credit-related characteristic, that is, it measures the difference in GDP growth before and after the GFC in states with a given characteristic minus the difference before and after the GFC in states without the characteristic. Table 4 presents the results. Column 1 of the table shows that states with credit growth above median prior to the GFC had, on average, a 1.45 percentage point larger decline in GDP growth per year afterwards, compared to states with below median credit growth. A placebo test confirms that there is no statistically significantly negative relationship for years prior to the GFC. Similarly, Column 3 shows that the growth slowdown in states in the top tercile of the credit growth distribution was 2.39 percentage points

FIGURE 10. Differential Impact of the GFC across States

10A: Growth Cycle around the GFC was More 10B: Growth was More Resilient in States That Were Pronounced in Non-agricultural States Less Dependent on Credit 5 01 10 15 81 Growth Growth 6 24 05 2000 2005 2010 2015 2000 2005 2010 2015 Year Year Non-agricultural States Agricultural States Low Credit Growth 03–08 High Credit Growth 03–08

Source: Authors’ calculations based on data from CEIC Data Company Ltd and Reserve Bank of India. Note: Outcome variables are measured as medians across states with the relevant characteristic. All years are fiscal years. Agricultural and credit dependency are defined with reference to the fiscal year 1999–2000. Credit growth refers to the period between 2002–03 and 2007–08. 62 INDIA POLICY FORUM, 2018

TABLE 4. Impact of GFC on States with Varying Credit Growth and Credit Dependence (1) (2) (3) (4) (5) (6) GDP GDP GDP GDP GDP GDP Growth Growth Growth Growth Growth Growth High credit growth × –1.449* post-GFC (Median) (1.854) High credit to GDP × –1.26 post-GFC (Median) (1.485) High credit growth × –2.39** post-GFC (Tercile) (2.268) High credit to GDP × –0.620 post-GFC (Tercile) (0.710) Credit growth 2003–08 –0.448** × post-GFC (continuous) (2.584) Credit to GDP × post-GFC –0.053* (continuous) (2.026) Observations 320 320 208 224 320 272 R-squared 0.312 0.311 0.281 0.379 0.318 0.289 Source: Authors’ calculations based on data from CEIC Data Company Ltd and Reserve Bank of India. Notes: The table presents regression estimates of Equation 4. We include state-level fixed effects in the regressions to account for time-invariant state characteristics. We estimate the regression using data from 1999–2000 onward for the sample of the large Indian states. All specifications are estimated with heteroscedasticity robust standard errors. larger than for those in the bottom tercile. The estimates are statistically significant at the 10 and 5 percent levels, respectively. As a further robustness check on this result, we estimate Equation 4 using the continuous variable measuring credit growth between 2003–04 and 2007–08 as the state-level characteristic. Our estimates imply a reduction of 0.45 percentage points in GDP growth after the GFC for every additional percentage point increase in credit growth (Table 4, Column 5). Similarly, the level of credit dependence of a state’s economy was negatively correlated with changing growth trends around the GFC: Column 6 of the table shows that states that had an above median credit-to-GDP ratio in 2000 experienced slower GDP growth after the GFC, compared to states with a below median credit-to-GDP ratio.

3.3. Policy Response to the GFC and Macroeconomic Stability The impact of the GFC on different countries and the pace of recovery depended both on the preconditions, such as the pace of GDP growth, and credit and investment growth in the years prior to the crisis, and on the Poonam Gupta et al. 63 policy response to the crisis. The initial impact of the crisis is considered relatively mute on India (Acharya 2012) largely due to a prompt and rather large policy response to the crisis, including monetary policy easing, a large fiscal stimulus, and regulatory forbearance on banks (or what some have referred to as the “evergreening of loans”).28 Mohan and Kapur (2015) and Mundle et al. (2011) have persuasively argued that in the run-up to the 2009 general election, the fiscal stimulus, in fact, started prior to the GFC. The excessively stimulative policy response and the subsequent macroeconomic management of the economy, however, worsened macroeconomic stability and possibly prolonged the slowdown (Figure 11). The slow and delayed recognition and resolution of stressed bank assets subsequently added to the issues with impaired balance sheets. A fallout of these policies was that some of the macroeconomic indicators reached crisis proportions by 2013, as the general government deficit touched nearly 10 percent of GDP; inflation reached double digit levels; the current account deficit increased to 5 percent of GDP; and the

FIGURE 11. Macroeconomic Stability Deteriorated Significantly during and after the GFC and Has Improved in Recent Years

1.2

0.8

0.4

% 0

–0.4

–0.8

–1.2 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

Macroeconomic Stability Index India

Source: Data are from WDI. Note: Years refer to calendar years. The index is constructed as an average of standardized indexes of inflation (CPI inflation), current account deficit (percent of GDP), and fiscal deficit (percent of GDP).

28. Repo rates dropped from 9 percent in September 2008 to 3.25 percent in April 2009 and the Center’s fiscal deficit increased by 5.5 percentage points of GDP. 64 INDIA POLICY FORUM, 2018 quality of public expenditure possibly worsened due to a decline in the share of capital expenditure. Hence, it is unsurprising that as market sentiment turned against emerging markets in summer 2013, India was one of the most impacted economies during the “taper talk.”29 The tapering talk episode started on May 22, 2013, when Federal Reserve Chairman Ben Bernanke first spoke of the possibility of the US central bank reducing the pace of its security purchases. Even though this announcement had a sharp negative impact on many emerging markets, market commentary focused most on five countries, Brazil, Indonesia, India, Turkey, and South Africa, christened as “Fragile Five.” Within this group, India had the second-largest exchange rate depreciation and the second-largest decline in reserves. With the rupee depreciating by 18 percent at one point, bond spreads increasing, and equity prices falling, some were concerned that the country was heading toward a financial crisis.30 Eichengreen and Gupta (2014) show that the emerging markets that allowed their real exchange rate to appreciate and the current account deficit to widen during the period of quantitative easing experienced a larger impact of the tapering event. Basu, Eichengreen, and Gupta (2015) show that India had received large capital flows in prior years and had large and liquid financial markets that were a convenient target for investors seeking to rebalance away from emerging markets; and its macroeconomic conditions had weakened in prior years, which rendered the economy vulnerable to capital outflows and limited the policy room for maneuverability. India’s current account deficit increased from about 1 percent of GDP in 2006 to nearly 5 percent in 2013, and its real exchange rate appreciated markedly. Furthermore, the fiscal deficit increased, and inflation at about 10 percent was stubbornly high. The policy interest rate was already high, with the Reserve Bank of India (RBI) having raised it from 3.25 percent in December 2009 to 8.5 percent in December 2012.31 The underlying drivers of India’s reduced macroeconomic stability, specifically the factors contributing to the high fiscal or current account deficit, also contributed to increased economic and financial vulnerabilities. The increase

29. The period of the “taper talk” generally refers to that between May 22, 2013, and September 18, 2013. 30. See, for example, “India in Crisis Mode as Rupee Hits Another Record Low” (http:// money.cnn.com/2013/08/28/investing/india-rupee/) and “India’s Financial Crisis: Through the Keyhole” (https://www.economist.com/banyan/2013/08/18/through-the-keyhole). 31. If the increase in fiscal deficit was in response to the GFC, India seemingly overreacted. Its deficit increased by more than in many other large emerging markets, a corollary of which is that inflation also increased by more than in other countries. Poonam Gupta et al. 65 in fiscal deficit was due to an increase in current expenditure, rather than to a pickup in public investment; while the increase in expenditure was due to increased subsidies (on energy, food, and fertilizer) that added up to 2.3 percent of GDP in 2008–09, (an increase of nearly 1 percentage point of GDP over the previous year); as well as debt waivers, Pay Commission awards, and expansion of the National Rural Employment Guarantee Act from 200 to 600 districts. Some of the increase in its current account deficit, largely a mirror image of the increased current expenditure, was due to the diversion of private savings into the import of gold. This reflected a dearth of attractive domestic outlets for personal savings in a high-inflation environment, where real returns on many domestic financial investments had turned negative. These results highlight the importance of having in place a policy framework that limits vulnerabilities and maximizes policy space for responding to shocks. Elements of such a framework include maintaining a sound fiscal balance, a sustainable current account deficit, an environment conducive to investment, managing capital flows so as to encourage relatively stable longer-term flows and discouraging volatile short-term flows, avoiding exces- sive appreciation of the exchange rate, holding a large stock of reserves, and preparing banks and corporates to handle greater exchange rate volatility.

3.4. Current Cyclical Dynamics Next, we analyze the dynamics of the Indian economy in the last few quarters and put them in context with the long-term experience discussed earlier. Most recent commentaries on the Indian economy focused on an ominously declining growth rate over a five-quarter period, from 9.3 percent in Q4, 2015–2016 to 5.6 percent in Q1, 2017–18. Further, we analyze the growth rate of quarterly GDP and its decomposition for the period starting 2013–14 through Q2 2018–19 (Figures 13 and 14). Two points are noteworthy. First, growth in the two quarters of Q1, 2016–17, and Q2, 2016–17 averaged 7.9 percent, higher than the average growth rate in recent quarters, or recent years. It would be erroneous to treat these as a part of the deceleration phase. Hence, the discussion around a five-quarter phase of deceleration should really center only around the three quarters during Q3, 2016–17 through Q1, 2017–18, when growth rates at 6.8, 6.1, and 5.6 percent, respectively, deviated significantly from the trend. Incidentally, these quarters coincided with the twin policy shocks of demonetization and the implementation of the GST. Sectors such as manufacturing and construction were reportedly most affected by the implementation of the GST and demonetization; in addition, an investment slowdown and increase 66 INDIA POLICY FORUM, 2018 in imports also impacted growth during the three-quarter deceleration period of Q3, 2016–17 through Q1, 2017–18.32 Second, many economic indicators now firmly indicate that these events had a transient impact as the economy has been slowly recovering from them.33 Growth has since steadily accelerated to 7.7, 8.2, and 7.1 percent in the last three quarters spanning Q4, 2017–18 to Q2, 2018–19. Economic revival is also evident in high frequency indicators such as the Purchasing Managers’ Index (PMI) and the Index of Industrial Production (IIP). Both of these con- firmed a sharp slowdown in the months surrounding the introduction of the GST, but have recovered and have registered a consistent expansion in the recent months. While consumption and services continue to be the main drivers of growth in recent quarters (between Q4 2017–18 to Q2 2018–19), the contribution of GFCF (on the demand side) and manufacturing and construction sector (on the supply side) has increased steadily. Even as the investment rate broadly remains burdened by stressed balance sheets of banks and corporates (twin balance sheet issues), investment growth has picked up in recent quarters. Credit growth, and to a lesser extent exports growth, has also recovered in recent months after a protracted period of deceleration (Figure 12).34

4. Continued Challenges for the Indian Economy

In this section, we take a look at the past episodes of high growth rates in India, defined as those when growth reached 8 percent or higher and ask whether similar levels of growth rates seem feasible in the near term, and what kind of challenges may need to be overcome for growth to move to a higher rate trajectory. We also discuss whether there is any room, or rationale, for countercyclical policies to support growth and how the external environment is poised to support a higher growth rate in India.

32. Due to the GST-related uncertainties, producers destocked their existing inventories, while exports declined, and gold imports nearly doubled, as buyers front-loaded their purchases. Once the initial uncertainties abated, economic activity recovered, and new orders, including in manufacturing, reportedly picked up. 33. In our analysis, we see the transient impact of demonetization on financial, real estate, and professional services, and on construction, but not so much on other sectors of the economy. On the uses side, deceleration was more evident in an already slowing rate of investment; and in an escalated level of import of gold (possibly due to capital flight). 34. Outstanding credit by only SCBs; we don’t consider credit by nonscheduled banks or other financial corporations. The same applies to the discussion of bank credit under Section 4.3. Poonam Gupta et al. 67

FIGURE 12. High Frequency Data Suggest Some Uptick in Credit Growth and Exports

Bank Credit Growth Export Growth 20 30 20 15 10

% 10

% 0 5 –10 –20 0 –30 Feb–14 Feb–15 Feb–16 Feb–17 Feb–18 Aug–13 Aug–14 Aug–15 Aug–16 Aug–17 Aug–18 Nov–13 Nov–14 Nov–15 Nov–16 Nov–17 Nov–18 May–13 May–14 May–15 May–16 May–17 May–18 Jul–13 Jul–14 Jul–15 Jul–16 Jul–17 Jul–18 Apr–13 Apr–14 Apr–15 Apr–16 Apr–17 Oct–18 Oct–13 Apr–18 Oct–14 Oct–15 Oct–16 Oct–17 Jan–14 Jan–15 Jan–16 Jan–17 Jan–18

SCB Credit Growth 3MMA Total Exports 3MMA

Source: Credit data are from the RBI; exports data are from the Ministry of Commerce and Industry. Note: 3MMA refers to three months’ moving average.

4.1. Past Episodes of High Growth A review of the data since 1971 reveals that there have not been many episodes when annual growth exceeded 8 percent. There have been six episodes over the last five decades, for a total of eleven years (including two years when the growth rate was 7.9 percent), when the growth rate in each year neared 8 percent or higher. With the exception of a five-year period, 2003–04 through 2007–08, most of these episodes of high growth did not sustain for more than a year (Table 5). Rather, growth acceleration lasted for only one year and corrected sharply a year later (Figure 15).35 In some of these episodes, high growth was on account of an unusually good agricultural output (1976, 1989) due to the base effect of slow growth in the previous year. In others, it was an outcome of stimulative fiscal or other macroeconomic policies (such as in 2010–11) and hence proved to be unsustainable. The only durable episode that lasted from 2004 to 2008 was dependent on a comprehensive reform agenda, an unusual buoyancy in the global economy, and easy global liquidity.36 We note that the high growth rate attained during 2004–08 reflected in robust growth rates in all domestic sectors (Figure 16). In contrast, several sectors have lagged behind in the last decade. Economic growth has been increasingly driven by consumption (private and public) since 2009, while two important engines of growth, private investment and exports, have

35. This is not unusual, as cross-country experience shows that a large percentage of high-growth episodes unravel within years (Berg, Ostry, and Zettelmeyer 2012; Pritchett and Summers 2014). 36. See Panagariya (2018) for a discussion of how reforms undertaken in the 1990s and early 2000s translated into higher growth subsequently. 68 INDIA POLICY FORUM, 2018

FIGURE 13. Growth Recovered across Sectors since Q2, 2017–18

13A: Growth Slowdown Has Likely Bottomed Out… 13B: …As Agricultural Growth Has Picked Up GVA at Basic Price Agriculture 10 10 8 8 6 2.6 6 4 %

% 7.2 2 4 0 2 –2 –4 0 Q1 2014 Q2 2014 Q3 2014 Q4 2014 Q1 2015 Q2 2015 Q3 2015 Q4 2015 Q1 2016 Q2 2016 Q3 2016 Q4 2016 Q1 2017 Q2 2017 Q3 2017 Q4 2017 Q1 2018 Q2 2018 Q3 2018 Q4 2018 Q1 2019 Q2 2019 Q1 2014 Q2 2014 Q3 2014 Q4 2014 Q1 2015 Q2 2015 Q3 2015 Q4 2015 Q1 2016 Q2 2016 Q3 2016 Q4 2016 Q1 2017 Q2 2017 Q3 2017 Q4 2017 Q1 2018 Q2 2018 Q3 2018 Q4 2018 Q1 2019 Q2 2019

13C: Industrial Growth Has Revived… 13D: …and Services Continue to Do Well Industry 14 Services 14 12 12 10 10 8 8 9.0 % % 6 6 4 7.0 4 2 2 0 0 Q1 2014 Q2 2014 Q3 2014 Q4 2014 Q1 2015 Q2 2015 Q3 2015 Q4 2015 Q1 2016 Q2 2016 Q3 2016 Q4 2016 Q1 2017 Q2 2017 Q3 2017 Q4 2017 Q1 2018 Q2 2018 Q3 2018 Q4 2018 Q1 2019 Q2 2019 Q1 2014 Q2 2014 Q3 2014 Q4 2014 Q1 2015 Q2 2015 Q3 2015 Q4 2015 Q1 2016 Q2 2016 Q3 2016 Q4 2016 Q1 2017 Q2 2017 Q3 2017 Q4 2017 Q1 2018 Q2 2018 Q3 2018 Q4 2018 Q1 2019 Q2 2019

13E: Manufacturing Has Picked Up 13F: Construction Sector Activity Has Revived Construction 20 Manufacturing 14 12 16 10 8 12 6

% 4

% 8 2 3.6 4 8.6 0 –2 0 –4 –4 –6 Q1 2014 Q2 2014 Q3 2014 Q4 2014 Q1 2015 Q2 2015 Q3 2015 Q4 2015 Q1 2016 Q2 2016 Q3 2016 Q4 2016 Q1 2017 Q2 2017 Q3 2017 Q4 2017 Q1 2018 Q2 2018 Q3 2018 Q4 2018 Q1 2019 Q2 2019 Q1 2014 Q2 2014 Q3 2014 Q4 2014 Q1 2015 Q2 2015 Q3 2015 Q4 2015 Q1 2016 Q2 2016 Q3 2016 Q4 2016 Q1 2017 Q2 2017 Q3 2017 Q4 2017 Q1 2018 Q2 2018 Q3 2018 Q4 2018 Q1 2019 Q2 2019

Source: CSO data for fiscal years. Note: The averages indicated are for Q1, 2014–Q2, 2017. continued to under-perform. This trend is of particular concern as investment and exports are not just important direct sources of growth and productivity but also determine the technological capability as well as the competitiveness of a country’s production structure. Sustaining growth higher than that indicated by the trend growth rate of 7.0–7.5 percent will require contributions from all domestic sectors. Besides, at a time when the economy is fairly Poonam Gupta et al. 69

FIGURE 14. GDP Growth Recovered since Q2, 2017–18

14A: GDP Growth Recovered 14B: Investment Growth Picked up after Remaining Subdued GDP at Market Price Investment 10 20

16 8 12 6 [VALUE] 7.4 average % % 8 4.6 4 4

2 0

0 –4 Q1 2014 Q2 2014 Q3 2014 Q4 2014 Q1 2015 Q2 2015 Q3 2015 Q4 2015 Q1 2016 Q2 2016 Q3 2016 Q4 2016 Q1 2017 Q2 2017 Q3 2017 Q4 2017 Q1 2018 Q2 2018 Q3 2018 Q4 2018 Q1 2019 Q2 2019 Q1 2014 Q2 2014 Q3 2014 Q4 2014 Q1 2015 Q2 2015 Q3 2015 Q4 2015 Q1 2016 Q2 2016 Q3 2016 Q4 2016 Q1 2017 Q2 2017 Q3 2017 Q4 2017 Q1 2018 Q2 2018 Q3 2018 Q4 2018 Q1 2019 Q2 2019

14C: Private Consumption Growth Remains at Levels 14D: Government Consumption Growth Is Mostly at Seen in the Past Levels Seen in the Past Private Consumption Government Consumption 12 35 30 10 25 7.2 20 8 15 6.0 10

% 6 % 5 4 0 –5 2 –10 –15 0 –20 Q1 2014 Q2 2014 Q3 2014 Q4 2014 Q1 2015 Q2 2015 Q3 2015 Q4 2015 Q1 2016 Q2 2016 Q3 2016 Q4 2016 Q1 2017 Q2 2017 Q3 2017 Q4 2017 Q1 2018 Q2 2018 Q3 2018 Q4 2018 Q1 2019 Q2 2019 Q1 2014 Q2 2014 Q3 2014 Q4 2014 Q1 2015 Q2 2015 Q3 2015 Q4 2015 Q1 2016 Q2 2016 Q3 2016 Q4 2016 Q1 2017 Q2 2017 Q3 2017 Q4 2017 Q1 2018 Q2 2018 Q3 2018 Q4 2018 Q1 2019 Q2 2019

14E: Exports Growth Shows Modest Pick Up 14F: Import Growth Has Picked Up Exports 30 Imports 20 25 15 20 10 15 1.8 10 5

% 5 % 0 0 –5 –5 –10 –10 –3.7 –15 –15 –20 Q1 2014 Q2 2014 Q3 2014 Q4 2014 Q1 2015 Q2 2015 Q3 2015 Q4 2015 Q1 2016 Q2 2016 Q3 2016 Q4 2016 Q1 2017 Q2 2017 Q3 2017 Q4 2017 Q1 2018 Q2 2018 Q3 2018 Q4 2018 Q1 2019 Q2 2019 Q1 2014 Q2 2014 Q3 2014 Q4 2014 Q1 2015 Q2 2015 Q3 2015 Q4 2015 Q1 2016 Q2 2016 Q3 2016 Q4 2016 Q1 2017 Q2 2017 Q3 2017 Q4 2017 Q1 2018 Q2 2018 Q3 2018 Q4 2018 Q1 2019 Q2 2019

Source: CSO data for fiscal years. Note: The average growth rates indicated are for Q1, 2014–Q2, 2017.

TABLE 5. Episodes of “High Growth.” Growth Rate No. of Episodes & Duration Time Period* 6 episodes 1976, 1989, 2000, 2006–08, ≥ 8 total duration: 9 years 2010–2011, 2016 Source: CSO data and authors’ calculations. Note: *Fiscal years. 70 INDIA POLICY FORUM, 2018

FIGURE 15. Most Episodes of “High Growth” Lasted Only for a Year

11 10.3 10 9.3 9.8 9.6 8.2 9 9.1 8.8 9.3 8.5 8 7.9 7.9 7.4 7.1 7 6.6 6 5.9 % 5 6.2 4 4 3.8 3.8 3.9 3.9 3 2 1.7 1 1.2 0 1 1 1 1 1 1 1 1 1 1 1 1 T– T– T– T– T– T– T+ T+ T+ T+ T+ T+ T=2016 T=2010 T=2011 T=2004 T=2005 T=2006 T=2007 T=2008 T=2000 T=1989 T=1976

Growth before and after the Acceleration Period Average

Source: CSO data. This figure is based on NAS data in the 2004–05 series prior to the year 2011–12 as explained in Appendix A. The episodes will vary if the recently released 2011–12 back series is used for the years between 2005–06 and 2011–12. Note: Years refer to fiscal years.

FIGURE 16. Growth during 2004–08 Built on Robust Growth across Domestic and External Sectors

16A: Average Growth Across Uses during and 16B: Average Growth Across Sectors during and after the Boom Years of 2004–08 after the Boom Years of 2004–08

25 14 12.5 20.1 18.7 17.8 12 20 9.7 10.0 10 8.7 9.2 15 7.7 8 7.1 6.8 % 8.8 10 7.2 7.4 6.7 % 4.9 5.4 7.1 5.1 5.8 6 4.8 5 4 3.0 0 2

GDP 0 Exports Imports GVA Other Consumption Services Industries Agriculture Construction Capital Formation Manufacturing

2004–08 2009–17 2004–08 2009–17

Source: CSO data. Note: Data are in constant Indian rupees. Years are fiscal years. open, it will be difficult to sustain such levels of growth only with the support of domestic factors and will require support from the global economy. Following first a period of unstainable boom and then of economic slowdown, and the build-up of macroeconomic unsustainability, reforms have Poonam Gupta et al. 71 been designed and successfully implemented in a number of areas in recent years—a new inflation targeting framework has been implemented, energy subsidy reforms have reduced the level of subsidies, the level of fiscal deficit has been contained, fiscal federalism has been strengthened, and the quality of fiscal expenditure has improved. The impact of some of these reforms is evident in a significant improvement in macroeconomic stability.37 Besides, there have been continuous efforts to improve the business environment, to ease inflows of FDI, improve credit discipline through the introduction and strengthening of an insolvency and bankruptcy framework, and widen access to financial services. GST has been implemented, which has harmonized tax rates across states and goods and services, and has the potential to boost interstate trade, formalize the economy, and improve the tax base. The expectation is that these reforms will help sustain current growth rates while ensuring macroeconomic stability. In addition, reversing the slowdown in specific sectors will require a careful analysis of their causes, and implementation of policy actions that are timely, wide-scoped, and innovative. Maintaining the ongoing reform momentum, and widening its scope will help revive growth in private investment, credit, and exports, in order to sustainably attain growth rates exceeding 8 percent. Further, we offer some perspectives on the challenges that may have been holding down the potential in these sectors and the related policy issues.

4.2. Continued Subdued Rate of Investment Is Worrisome After increasing slowly but steadily over the last several decades, and rapidly during the period of high growth, 2004–08, saving and investment rates have been declining since 2009. The saving rate has declined since the GFC, after registering a large increase in prior years, and is evident in a decline in the household physical savings rate, household financial savings rate, and in government savings. In contrast, the corporate saving rate has increased during the same period (Figure 17). The investment rate has declined as well since the GFC, after registering a rapid increase in prior years. The decline is most evident in corporate investment and household physical investment rate. There is a divergence in corporate savings and investment rates; while the corporate savings rate has increased, its investment rate has declined (Figure 18).

37. A sharp decline in oil prices, starting in mid-2014, low global inflation, and continued easy global liquidity provided the conditions conducive for the implementation of some of these reforms. 72 INDIA POLICY FORUM, 2018

FIGURE 17. Contribution to Decline in the Average Saving Rate between 2007–08 and 2016–17

2 1.0

–2 –1.1

% –4 –3.9 –6 –4.8

–8

–10 –8.9 Total Savings Household Public Household Private Financial Physical Corporate

Source: Data are from CEIC Data Company Ltd. Note: The figures show the difference between the average rates in 2016 and 2017, over 2007 and 2008.

FIGURE 18. Trends in the Investment Rate

18A: The Investment Rate Has Declined since the GFC… 18B: …the Decline Is Evident in Household Investment…

Investment as % of GDP Household Sector Investment 40 18 35 16 30 14 12 25 10 % 20 % 8 15 6 10 4 5 2 0 0 2000 2002 2004 2006 2008 2010 2012 2014 2016 2000 2002 2004 2006 2008 2010 2012 2014 2016

18C: …and in the Private Corporate Sector 18D: While Public Investment Fell after GFC, It Has Increased Modestly in Recent Years Private Corporate Sector Investment as Public Sector Investment as a % of GDP a % of GDP 18 10.0 16 9.0 14 8.0 7.0 12 6.0 10

% 5.0 % 8 4.0 6 3.0 4 2.0 2 1.0 0 0.0 2000 2002 2004 2006 2008 2010 2012 2014 2016 2000 2002 2004 2006 2008 2010 2012 2014 2016

Source: Data are from CEIC Data Company Ltd and for fiscal years. Note: Investment rate is defined as GFCF in % of GDP. Poonam Gupta et al. 73

The investment slowdown pervades across several sectors of the economy, most prominently in manufacturing and construction (Figure 19). Overall, the investment rate declined by approximately 4.9 percentage points during 2007–08 and 2015–16, driven by manufacturing, with the investment rate declining by 3.7 percentage points, followed by construction. Investment rates declined in other sectors too, but increased in trade, hotels, and restaurants. In a cross-country comparison of the trends in investment rate, we note that the Indian experience differs from that of the other emerging markets in that the investment rate increased far more rapidly in India prior to the GFC than in other countries, and the decline after the crisis was steeper too. This cycle is evident in both public and private investment rates, as both increased in the few years prior to the GFC and declined thereafter. While private investment has continued to remain depressed in recent years, public investment rates have increased (Figure 20). Private investment in India is constrained by several factors. There are issues related to past leverage as well as subdued market demand.38 Going forward, de-risking the private sector may be important, as it may be to further ensure an environment of policy certainty. Understanding and relieving the generic, spatial, or sector-specific constraints to investment growth may be important too. Reviving private investment in areas such as infrastructure to finance India’s long-term investment needs would be useful.39

4.3. Reviving Bank Credit and Resolving Asset Quality Issues to Support Growth The last few years have been challenging for Indian banks, as the pace of credit growth remains subdued, and the stress on asset quality continues. Bank credit growth has consistently declined since the GFC, after increasing briskly for a few years before that. The annual average growth rate of bank

38. Deleveraging could be one reason behind the slow pace of investment growth—Indian businesses overinvested and overleveraged during the boom years. Yet, due to the slow pace of resolution, businesses have been unable to deleverage quickly and start investing afresh. There may also be sectoral constraints to investments in sectors such as construction, leather, infrastructure, telecom, and energy sector. If the investment slowdown is concentrated in export-oriented firms, it may be indicative of specific constraints related to the size of external markets and to their competitiveness. 39. The World Bank has recently suggested a “Maximizing Finance for Development (MFD)” approach to crowd-in private financing through the use of public instruments such as guarantees, and by removing policy or regulatory gaps. The idea is to leverage more private investment, while reserving scarce public financing for areas where private sector engagement is unavailable or not optimal. 74 INDIA POLICY FORUM, 2018

FIGURE 19. Decline in Average Investment Share of GDP between 2007– 08 and 2015–16

3 1.7 2 1 0 –1 –0.2 –0.1 –0.1

% –0.4 –0.4 –2 –1.2 –1.1 –3 –4 –5 –3.7 –6 –5.5 , g g d e s Total r etc. Agricultur Construction etc. Wate Manufacturin Restaurant Trade, Hotel, an Community, Social, Electricity, Gas, and Mining and Quarryin Financial, Real Estate Transport, Storage, etc.

Source: Data are from CEIC Data Company Ltd. Note: The figures show the difference between the average share in 2015 and 2016, and in 2007 and 2008. credit was 9.5 percent in 2014–17 compared to 26.3 percent in 2004–08. As a result, the credit to GDP ratio has declined in the recent years, after peaking at 56 percent in 2014, and after doubling within a span of 7 years from 25.5 percent in 2001 to 52 percent in 2008. In addition, after a decade-long declining trend, the ratio of Gross Non-performing Assets to advances (GNPA) of Scheduled Commercial Banks (SCBs) increased from 2.5 percent in 2007 to 9.3 percent in 2017 (Figure 21). The asset quality of SCBs deteriorated across sectors, with the largest deterioration in the industrial sector. The level of stressed assets (NPAs and restructured loans) has been above 10 percent since 2014. The RBI’s Asset Quality Review in late 2015 resulted in a large migration of restructured loans into NPAs and new NPA recognition (IMF and World Bank 2017). There is a predominance of banks in the Indian financial system, and that of public sector banks (PSBs) in the Indian banking sector. Banks account for 60 percent of financial system assets, while 70 percent of banking assets are held by the PSBs. The share of publicly owned banks has remained largely unchanged, even as ownership has decisively become more mixed in other hitherto majority government-owned sectors of the economy, such as aviation and telecom (Figure 22). There has been a distinct difference in the trends for credit growth and asset quality for PSBs and private banks. Credit growth has been slower and the pace of bad assets higher for PSBs. In the last few years, the pace Poonam Gupta et al. 75

FIGURE 20. Comparative Analysis of the Rate of Investment in India and in Emerging Market Developing Economies (EMDE)

20A: GFCF (% GDP) 20B: GFCF Growth Rates 30 40 IND EMDE (Median) IND EMDE (Median) EMDE (25–75%) EMDE (25–75%) 35 20

30 10 25 0 20 –10 15

10 –20 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016

20C: Public Investment (% GDP) 20D: Public Investment Growth Rates

40 IND EMDE (Median) 12 IND EMDE (Median) EMDE (25–75%) EMDE (25–75%) 30 10 20 8 10 6 0 4 –10 2 –20 0 –30 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016

20E: Private Investment (% GDP) 20F: Private Investment Growth Rates

30 IND EMDE (Median) 40 IND EMDE (Median) EMDE (25–75%) EMDE (25–75%) 25 30

20 20

15 10

10 0

5 –10

0 –20 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2014 2016 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012

Source: WDI. Note: The shaded area represents the post-Global Financial Crisis years. of growth of PSB credit (outstanding) has continued to decline, growing at 1.8 percent in 2017, the lowest in the last two decades. On the other hand, credit by private banks grew at double digit rates (Figure 23). There has been a long downward trend in the high NPA level since the late 1990s. The PSBs had a higher non-performing asset ratio at that time. As PSBs gradually reduced their NPAs, the NPAs continued to grow at private sector banks till the early 2000s. In the recent years, the ratio of 76 INDIA POLICY FORUM, 2018

FIGURE 21. Trends in Banking Credit at Scheduled Commercial Banks

Credit Growth Credit-GDP Ratio 35 60 30 50 25 40 20 %

% 30 15

10 20

5 10

0 0 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018

Source: Data are from the RBI. Note: Years refer to fiscal years.

FIGURE 22. Ownership Structure of the Banking, Aviation, and Telecom Industries

22A: Shares of the Private Sector Continue to Increase in Aviation and Telecom Industries

Traffic (% of Total Traffic of all Telephone Subscribers Scheduled Indian Airlines) 100 100 80 80 60 60 %

% 40 40 20 20 0 0 2000 2005 2010 2015 2001 2005 2010 2015

22B: But Remains Low and Sticky in Banking.…

Assets (% of Total Assets of SCBs) Advances (% of Total Advances of SCBs) 100 100 80 80 60 60 %

40 % 40 20 20 0 0 2000 2005 2010 2015 2000 2005 2010 2015

Source: Data are from the RBI (banking), Directorate General of Civil Aviation (aviation), data.gov.in, and TRAI (telecom). Note: Years refer to fiscal years. Poonam Gupta et al. 77

FIGURE 23. Growth of Outstanding Credit Has Declined in the Last Few Years; Decline Is Sharper for Public Sector Banks

35 30 25 20 % 15 10 5 0 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 Public Sector Banks Private Banks

Source: Data are from the RBI. Note: Years refer to fiscal years. non-performing assets has increased in PSBs. Besides, there is also a stark contrast between the profitability of PSBs and private banks. PSBs have continued to record negative profitability ratios since March 2016. The return on assets of PSBs was −0.1 percent in September 2017, and its return on equity was −2.0 percent, compared to 1.4 percent and 11.9 percent, respectively, for private banks (Figure 24).40 According to experts, the distorted incentive structure coupled with political compulsions has resulted in allocative and operational inefficiencies of a public sector-led banking sector, and to periodic loan write-offs, for example, to the agriculture sector, and underrecovery of the corporate credit. Reconsidering the ownership balance, and the incentive and governance

40. The allocative efficiency of the public sector-dominated Indian banking sector is considered to be low, holding back potential investments and economic growth. Banerjee, Cole, and Duflo (2004) characterize the Indian public banks as “lazy,” since the lending decisions of their managers are not based on the current or expected profitability of firms; they underlend to the private sector and overinvest in government securities. They explain that the employees of the PSBs are treated as public servants by law, and hence if they take decisions which result in direct financial gain to a third party, they may be held guilty of corruption. The bankers thus choose to lend less to the private sector; and disproportionately more to the government. Gupta, Kochhar, and Panth (2015) show that the Indian PSBs allocate a larger share of their assets to government securities; and in doing so they respond more to the level of the fiscal deficit than to market signals or even the SLR ratio. 78 INDIA POLICY FORUM, 2018

FIGURE 24. Gross NPAs to Gross Advances of the Banking System

16 Gross NPA to Gross Advances Ratio 14 12 10

% 8 6 4 2 0 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

Public Private SCBs

Source: Data are from the RBI. Note: Years refer to fiscal years. structure may be important in order to alter the perennial cycle of allocative and operational inefficiencies. Under its current balance sheet situation, the banking sector does not seem well equipped to help finance higher growth or investment, and suitable reforms will be needed to reverse this situation.41 In this direction, the government announced an unprecedented recapitalization of PSBs on October 24, 2017. The proposed measures include recapitalization of approximately `2.11 trillion (around $32 billion) over the next two years. The government proposed to fund it through budgetary provisions amounting to approximately `180 billion; recapitalization bonds to the tune of `1.35 trillion; and direct capital raising by banks from the market by diluting government share (an estimated `580 billion). The recapitalization ensured continuity and stability in the system, but needs to be followed by wider reforms. The government also implemented and strengthened a new Insolvency and Bankruptcy Code. While an important step toward changing the credit culture, the policy will take time to help clean balance sheets and is unlikely on its own to improve the capital adequacy of banks. A dynamic measurement of PSBs’ governance and financial performance metrics could be deployed to systematically address moral hazard concerns. Additional measures to durably enhance the stability and efficiency of the financial sector could include consolidation of PSBs; revising their incentive structure to align more closely with their commercial performance; ensuring a level playing field for private banks; and opening

41. These have been discussed in the Nayak Committee Report, in the Indradhanush Plan, and in the recently concluded joint IMF−World Bank FSAP Report (IMF and World Bank 2017). Poonam Gupta et al. 79 the space to greater competition. It will be useful to take a call on what part of the ongoing spike in non-performing loans (NPLs) is cyclical and what part is structural. While the former can possibly be reversed with a cyclical turnaround of the economy, or addressed better through cyclical solutions such as regulatory forbearance, the latter ought to be addressed through structural solutions such as altering the ownership mix. The issue also ties in with the pace of resolution: for the cyclical part of the problem, perhaps more patience is warranted until a cyclical recovery takes hold, whereas the structural issues are unlikely to get resolved on their own and will require fast and decisive actions. The government is reportedly exploring different options to resolve the problem of high NPLs. Some of these measures such as mergers within the PSBs have been used earlier; while others are more novel in the Indian context, including setting up of a bad bank, or the aggressive use of bankruptcy procedures in loan recovery. It would be useful to consider the merits of these options in view of cross-country experiences. Other issues that could be afforded specific attention are on building risk assessment capabilities within the regulator and the banks; and developing and strengthening the personal bankruptcy framework.

4.4. Making Exports Competitive Again While private investment is being held back primarily by domestic factors, exports growth is constrained by both domestic and external factors. Exports growth was an important driver of GDP growth prior to the GFC, and specifically during the pre-crisis boom years. Its contribution to growth has diminished since. Export growth has experienced two phases of deceleration since the GFC. The first of these culminated in negative export growth rates in 2009–10, and the second phase resulted in slow exports growth 2013 onward. Meanwhile, import growth also decelerated until recently and temporarily turned negative in 2015–16. India has barely managed to keep pace with the growth in world exports since the GFC, reflected in its stagnant or even declining share of world exports, and a declining export-to-GDP ratio. The slowdown in export growth is evident in merchandise and services exports, and extends to different export destinations. The slowdown is partly attributed to a decline in the prices of oil and commodities during 2014–16; but is also prominently reflected in the slow growth of global export volume, and in India’s declining share in it. We decompose the slowdown in India’s merchandise exports into a price and a volume effect and further decompose the latter 80 INDIA POLICY FORUM, 2018 into a slowdown in global trade volume, and India’s share in it, and into its exports destinations. We note the following:

• The initial export slowdown from India around the time of the GFC was primarily due to a decline in global trade, with export growth recovering temporarily after the initial decline. The slowdown in subsequent years, however, is both due to a decline in the prices of oil and commodities, and a decline in India’s trade volumes. Between mid-2014 and January 2016, the global prices of oil and metal, and agriculture prices declined sharply, dropping by about 73, 37, and 23 percent, respectively.42 • We find that Indian merchandise export growth has decelerated in both value and volume terms.43 While the deceleration in Indian export values is significantly sharper than the volume of exports, volume growth turned negative in 2009 and again in 2016. The decline in India’s trade volume is larger in comparison to the global decline in trade volume, resulting in India’s reduced share in global exports (Figure 25). This is indicative of the role of India-specific factors in determining the export slowdown, or deteriorating external conditions, specifically for India’s export basket. • India’s export basket remains broad-based but the slowdown has been pervasive. The share of services exports in 2016 was approximately 36 percent of the total exports, core merchandise exports (i.e., non-oil, non-gold exports) accounted for about half of all exports, oil exports accounted for 10 percent, and gold exports accounted for 4 percent. The export slowdown has been pervasive across merchandise and services exports (Figure 26). • Comparing the product-specific average growth of exports in the boom period preceding the global economic crisis, 2003–08, to a more recent period, 2012–16, we note that the total export growth rate declined by 26 percentage points during this period (Figure 27). Decline in exports growth was most pronounced for commodity exports such as mineral fuels and lubricants, reflecting the effect of declining commodity prices after 2014, but it also extended to other product groups.44

42. The figures present the decline from peak prices in June, March, and July 2014 for oil, agriculture, and metals, respectively, to the trough in January 2016, and are drawn from the World Bank’s Global Economic Monitor database. 43. We use trade data measured in current USD and focus on merchandise trade for data availability reasons. For a globally heterogeneous export basket, using national deflators (either from the USA or India) is unlikely to yield credible estimates of constant export values. Hence, we rely on volume indices to decompose trade into volumes and prices. 44. Growth rates are for export values in current USD. Poonam Gupta et al. 81

FIGURE 25. Export Growth: India and Global, in Value and Volume

25A: Value and Volume Indices for World Growth of Exports 25B: India’s Share in World Exports

30 Value Volume 2 Value Volume 1.8 20 1.6 1.4 10 1.2

% 1

% 0 0.8 –10 0.6 0.4 –20 0.2 0 –30 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

Source: Data are from WDI on export values and from UNCTAD on export volumes. Note: The years are calendar years. The global volume index is calculated as the weighted average of countries’ volume indices with weights equivalent to countries’ (value-based) share in global merchandise exports.

FIGURE 26. Exports of Goods and Services: India, 2001−18

26A: Services and Merchandise Export Growth Has 26B: India’s Share in Goods Exports Has Plateaued in Slowed Down Recent Years Growth Rate of Exports India’s Share in Global Exports 70 4.0 60 3.5 50 3.0 40 2.5 30 %

20 % 2.0 10 1.5 0 1.0 –10 0.5 –20 0.0 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2001 2003 2005 2007 2009 2011 2013 2015 2017 Goods Services Goods Services

Source: Data are from RBI and WDI. Note: Years are fiscal years. Calendar year 2017 in WDI data is considered as fiscal year 2017–18.

• India’s export destinations remained diversified and were equally affected. The largest share of exports from India is destined for the Middle East (approximately 20% in 2016), and within it, the largest share is exported to the United Arab Emirates. The USA is the second largest destination, accounting for 16 percent of India’s exports, followed by China (including Hong Kong), Sub-Saharan Africa, and Europe (Figure 28). We disaggre- gate Indian exports across its eight main trading partners and the rest of the world for 2003 to 2017.45 We note that the export slowdown experienced

45. India’s main trading partners, in order of share in total exports, are the USA, the United Arab Emirates, Hong Kong, China, the United Kingdom, Singapore, Germany, and Saudi Arabia. 82 INDIA POLICY FORUM, 2018

by India after 2013 was across destinations, as exports slowed to most of India’s main trading partners in the Middle East, the UAE, and Saudi Arabia, but also to the USA and to China and Singapore.

A significant improvement in the competitiveness of Indian firms would be the key to reinstating the increasing trend in India’s share of global exports.

FIGURE 27. Export Growth Rates before and after the GFC (Difference in average growth rates between 2003–08 and 2012–16)

Mineral fuel / lubricantsCrude materials excludingMachinery fuel / transportAnimal equipmen / veg oil / fatChemicals / wax / productsManufactured n.e.s. goodsBeverages and tobaccoFood and live animalsMiscellaneous manuf.Commodities articles n.e.s 20 10 0 –10 –20 % –30 –40 –50 Total –60 –70

Source: Data are from UN Comtrade database. Note: Export growth is in nominal USD. The years are calendar years. The bars denote the difference between the average growth rate between 2003 and 2008, and the average growth rate between 2012 and 2016. Total refers to the total difference in average growth rates.

FIGURE 28. Destinations of Indian Exports

28A: Contribution to Export Growth 28B: Contribution to Export Growth: Middle East and the USA 40 6 4 30 2 20

% 0

% 10 –2 –4 0 –6 –10 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

–20 Middle East (Saudi Arabia and UAE) USA 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 China (Mainland and Hong Kong) and Singapore Europe (Germany and UK) Middle East (Saudi Arabia and UAE) USA Rest of the World

Source: Data are from UN Comtrade database. Note: The years are calendar years. Export growth is in nominal USD. Poonam Gupta et al. 83

4.5. Leveraging External Conditions India is a large emerging market. Contrary to some perceptions, India has continuously, even though incrementally, integrated globally both in its trade and capital account. It has liberalized inflows and outflows of FDI, portfolio capital, and other forms of capital flows. Gupta and Masetti (2018) document capital flow measures across emerging markets and observe that, starting from a relatively low base, India has liberalized its capital account significantly in the last few years. As a result, it attracts a large share of the capital that flows to emerging countries. Thus, it is not surprising that a number of global developments, such as those related to growth in global growth or trade, liquidity or risk aversion in financial markets, or oil prices have implications for India’s economic growth, balance of payments, macroeconomic stability, and fiscal and monetary policy outcomes. External conditions have remained broadly supportive of growth in the last couple of years, but the projected near-term outlook underscores several challenges. The global growth outlook is projected to be less robust in the coming years due to several factors. These include a rise in policy uncertainty, underpinned by escalating trade tensions, imposi- tion of tariffs and retaliatory actions by some of the largest economies; monetary policy tightening by the US Federal Reserve Board and other advanced country central banks; and moderation of growth in advanced economies and China. A surprisingly robust increase in oil prices, defying most expert projections, has presented an additional headwind in the last couple of years. Being a net oil importer, India is sensitive to an increase in oil prices through a number of channels, the most prominent one being through higher a current account deficit, which bears the direct first-order impact of an increase in oil prices. This then filters into a subsequent impact on the exchange rate, inflation, and fiscal deficit, constraining the scope of growth-supporting fiscal and monetary policies, yet leading to an emotive political narrative. On the financing side, the US Federal Reserve Board has raised its policy rate seven times in the last three years starting in 2015 as per its pre- announced path. It raised its policy rate by 100 basis points in 2017 and by another 100 basis points in 2018. Looking forward, global interest rates are likely to rise at a slightly slower pace than previously expected. Even if not disruptive to financial markets in the short run, higher interest rates have started to tighten financing conditions for emerging markets, including for India. Hence, enhancing competitiveness in the domestic financial sector will be even more important to ensure affordable financing conditions. 84 INDIA POLICY FORUM, 2018

4.6. Limited Room or Rationale for Countercyclical Measures in the Presence of Structural Constraints to Growth There seems only limited room to ease fiscal, monetary, or exchange rate policies to boost growth in the midst of complex and persistent structural constraints to a higher growth level. Given the structural nature of weak exports and investments, the effectiveness of transitory countercyclical policies is likely to be limited. Even if used, these can provide only a temporary reprieve, as by their very nature, countercyclical policies ought to be used temporarily and should be reversed within a reasonable period of time. Besides, with inflation hovering in the vicinity of 4 percent, the current account deficit at 1.9 percent this year and projected to be at about 2.5 percent next year, the general government deficit at about 6.5 percent, the combined public debt at nearly 70 percent of GDP, and bond yields nearly touching 8 percent, there seems limited room to consider expansionary policies.46 If still considered by the government, it will have to be creative about generating fiscal space. One way to do so may be to generate resources domestically by considering a careful divestment of assets as per the recommendations of the National Institution for Transforming India (NITI). If instead the government wishes to borrow to finance enhanced infrastructure spending, it would be prudent to do so cautiously to minimize potential vulnerability.

5. Conclusion

In this paper, we offer a long-term perspective on India’s growth experience. We note that growth has slowly but steadily accelerated over the last 50 years, has become less erratic, and has been well diversified across sectors and states. Assessing the period since the early 1990s more granularly, we note three distinct phases of growth: a period of slow acceleration from 1991–early 2000s; a short period of unusually rapid growth, with certain features of unsustainability, during 2004–08; and a corrective slowdown that started with the GFC in 2008. The slowdown was reflected most profoundly in investment, credit, and exports. Even as the economy has now recovered to a 7–7.5 percent growth rate level,

46. As per our analysis, at current levels, general government public debt is sustainable, despite some rise in real borrowing rates in recent years, largely because of fast economic growth and continued fiscal consolidation by the central government. Poonam Gupta et al. 85 durably accelerating it to a higher level will require concerted policy momentum that succeeds in reversing the slowdown in investment, credit supply, and exports, and support from the global economy. Reversing the slowdown in specific sectors will require a careful analysis of their causes, and implementation of policy actions that are timely, wide- scoped, and innovative. The factors that may help India improve its competitiveness include an infrastructure boost to bring it at par with other manufacturing hubs of the world; reforms in land, labor, and financial markets, and in the educational system, to assure the continued competitive supply of key production inputs such of labor, land, finance, and skills. Besides, issues related to a competitive exchange rate, enhancing bilateral and regional trade integration, evading the temptation to cave in to the rhetoric on trade protectionism, and embedding more deeply in global value chains, all assume great significance and require an objective discussion and assessment.

References

Acharya, Shankar. 2012. “India after the Global Crisis,” Hyderabad: ICFAI Foundation for Higher Education (IFHE), 5 October. Bhagat, R.B. 2014. “The Opportunities and Challenges of Demographic Dividend in India,” Jharkhand Journal of Development and Management Studies, 12(4): 6099–6113. Banerjee, Abhijit V., Shawn Cole, and Esther Duflo. 2004. “Banking Reform in India,” Brookings Papers on Economic Activity, Economic Studies Program, The Brookings Institution, 1(1): 277–332. Basu, K., B. Eichengreen, and P. Gupta. 2015. “From Tapering to Tightening: The Impact of the Fed’s Exit on India,” India Policy Forum, 11: 1–66, New Delhi: National Council of Applied Economic Research. Berg, A., J.D. Ostry, and J. Zettelmeyer. 2012. “What Makes Growth Sustained?” Journal of Development Economics, 98(2): 149–166. Bils, M. and P. J. Klenow. 2000. “Does Schooling Cause Growth?” American Economic Review, 90(5): 1160–1183. Bosworth, B. and S.M. Collins. 2008. “Accounting for Growth: Comparing China and India,” The Journal of Economic Perspectives, 22(1): 45–66. Bosworth, B., S.M. Collins, and A. Virmani. 2007. “Sources of Growth in the Indian Economy,” National Bureau of Economic Research Working Paper 12901, Cambridge, MA: National Bureau of Economic Research. Bosworth, B.P. and J.E. Triplett. 2007. “The Early 21st Century US Productivity Expansion Is Still in Services,” International Productivity Monitor, 14: 3–19. Brandt, L. and X. Zhu. 2010. “Accounting for China’s Growth,” IZA Discussion Paper No. 4764, Bonn: Institute of Labor Economics. 86 INDIA POLICY FORUM, 2018

Bruno, M. and W. Easterly. 1998. “Inflation Crises and Long-Run Growth.” Journal of Monetary Economics, 41(1): 3–26. Caselli, F. 2005. “Accounting for Cross-Country Income Differences,” in Philippe Aghion and Steven Durlauf (Eds.), Handbook of Economic Growth, Volume 1, Edition 1, Amsterdam: Elsevier, pp. 679–741. Caselli, F. and W.J. Coleman. 2001. “Cross-Country Technology Diffusion: The Case of Computers,” American Economic Review, 91(2): 328–335. Chandrasekhar, C.P., J. Ghosh, and A. Roychowdhury. 2006. “The ‘Demographic Dividend’ and Young India’s Economic Future,” Economic and Political Weekly, 41(49): 5055–5064. Dell’Ariccia, G., P. Rabanal, C.W. Crowe, and D.O. Igan. 2011. “Policies for Macro-Financial Stability; Options to Deal with Real Estate Booms,” IMF Staff Discussion Notes, 11/02, Washington, DC: International Monetary Fund. Eichengreen, B. and P. Gupta. 2011. “The Service Sector as India’s Road to Economic Growth,” National Bureau of Economic Research Working Paper 16757, Cambridge, MA: National Bureau of Economic Research. ———. 2014. “Tapering Talk: The Impact of Expectations of Reduced Federal Reserve Security Purchases on Emerging Markets,” World Bank Policy Research Paper, Emerging Markets Review, Washington, DC: World Bank. Commission on Growth and Development. 2008. The Growth Report. Strategies for Sustained Growth and Inclusive Development. Washington, DC: World Bank. Gupta, P., K. Kochhar, and S. Panth. 2015. “Bank Ownership and the Effects of Financial Liberalization: Evidence from India,” Indian Growth and Development Review, 8(1): 109–138, April. Gupta, P. 2016. “Capital Flows and Central Banking: The Indian Experience,” World Bank Policy Research Working Paper Series 7569, Washington, DC: World Bank. Gupta, P. and O. Masetti. 2018. “Capital Flow Measures: Structural or Cyclical Policy Tools,” Policy Research Working Paper 8418, Washington, DC: World Bank. Hall, R.E. and C.I. Jones. 1999. “Why Do Some Countries Produce So Much More Output per Worker Than Others?” The Quarterly Journal of Economics, 114(1): 83–116. Hsieh, C.T. and P.J. Klenow. 2009. “Misallocation and Manufacturing TFP in China and India,” The Quarterly Journal of Economics, 124(4): 1403–1448. IMF and World Bank. 2017. “Financial Sector Assessment Program Report,” IMF Country Report No. 17/390. Kaufmann, Daniel, Aart Kraay, and Massimo Mastruzzi. 2010. “The Worldwide Governance Indicators: Methodology and Analytical Issues.” Policy Research Working Paper No. WPS 5430. Washington, D.C.: World Bank. Klenow, P.J. and A. Rodriguez-Clare. 1997. “Economic Growth: A Review Essay,” Journal of Monetary Economics, 40(3): 597–617. Kumar, U. and A. Subramanian. 2011. “India’s Growth in the 2000s: Four Facts,” Peterson Institute Working Paper Series 11–17, Washington, DC: Peterson Institute. Poonam Gupta et al. 87

Mohan, Rakesh. 2008. “The Growth Record of the Indian Economy, 1950–2008: A Story of Sustained Savings and Investment.” Economic and Political Weekly. XLIII(19): 61–71, May 10–16. Mohan, R. and M. Kapur. 2015. “Pressing the Indian Growth Accelerator: Policy Imperatives,” IMF Working Paper No. 15/53, Washington, DC: International Monetary Fund. Mundle, S., M.G. Rao, and N.R. Bhanumurthy. 2011. “Stimulus, Recovery and Exit Policy: G20 Experience and Indian Strategy.” Economic and Political Weekly, 46(29): 85–94. Nagaraj, R. and T.N. Srinivasan. 2016. “Measuring India’s GDP Growth: Unpacking the Analytics and Data Issues behind a Controversy That Refuses to Go Away,” India Policy Forum, 13: 73–128, New Delhi: National Council of Applied Economic Research. Panagariya, A. 2004. “Growth and Reforms during 1980s and 1990s,” Economic and Political Weekly, 39(25): 2581–2594. ———. 2018. “India: Three and a Half Years of Modinomics,” Columbia SIPA Working Paper 2018-01, New York, NY: Columbia University. Pritchett, L. and L.H. Summers. 2014. “Asiaphoria Meets Regression to the Mean,” National Bureau of Economic Research Working Paper No. 20573, Cambridge, MA: National Bureau of Economic Research. Triplett, J.E. and B. Bosworth. 2003. “Productivity Measurement Issues in Services Industries: Baumol’s Disease Has Been Cured,” Economic Policy Review, September, pp. 23–33. Young, A. 1995. “The Tyranny of Numbers: Confronting the Statistical Realities of the East Asian Growth Experience,” The Quarterly Journal of Economics, 110(3): 641–680. ———. 2003. “Gold into Base Metals: Productivity Growth in the People’s Republic of China during the Reform Period,” Journal of Political Economy, 111(6): 1220–1261.

Appendix A: National Accounts Data Splicing

The national accounts data used in this paper was obtained from India’s CSO. In January 2015, the base year was revised from 2004–05 to 2011–12. This revision makes comparing GDP data before and after 2011–12 (the first year for which the revised series is available) challenging, as the new series introduces conceptual and statistical changes. Some of these include updating the NAS methodology to the latest SNA 2008 system; changes in methodology and databases in the estimation of saving and investment; direct estimation in the value added of private corporate sector using financial returns; using MCA-21 instead of Annual Survey of Industries (ASI) to estimate manufacturing sector output (Nagaraj and Srinivasan 2016). 88 INDIA POLICY FORUM, 2018

We splice the new annual and quarterly GDP series backward, using a simple backcasting methodology. The data for the fiscal years prior to 2011–12, available under the old series (with base year 2004–05) were converted to 2011–12 base as explained further. Consider a variable Xt that needs to be spliced. We denote Xt in the new series as X* and X in the old series. Suppose data in the new series begin from period t. To obtain the * value of Xt–1, we simply apply the following formula:

**Xt−1 Xt−1 = Xt Xt

Intuitively, this series maintains a growth rate in the new series (captured X by t1 ) that is consistent with the old series. The resulting series thus Xt resembles a level shift to the old series with equivalent growth rates. We used this procedure to maintain the growth rates of GDP at market prices, gross value added (GVA) at basic prices, and their main subcomponents. A challenge that arises when matching growth rates of subcomponents (whose shares add to 1) is that residuals appear, driven by the fact that changes to the base year affect the estimated contribution of various sectors to the economy. The CSO typically divides the Indian economy intro three sectors: agriculture, industry, and services. Agriculture includes crop, livestock, forestry, and fisheries. The industrial sector is again split into four sub-sectors: mining and quarrying; manufacturing; electricity, gas, water, and other utility supply; and construction. The services sector is split into three subsectors: trade, hotels, transport, communication and services related to broadcasting; financial, real estate, and professional services; and public administration, defense, and other services. To preserve additivity, we generate a residual series for GDP at market prices. For GVA, we employ the service sector (in annual data), and the public administration and defense services sector (in quarterly data), as the residual. We conduct robustness checks to verify that our observed growth rates in the services sector are not driven by its selection as a residual.

Appendix B: National Accounts Data Update

The CSO released the back series of the National Accounts Statistics (NAS) with base year 2011–12 (new series) in November 2018.47 The back series was released for the fiscal years 2004–05 through 2010–11. Until this release,

47. http://www.mospi.gov.in/sites/default/files/press_release/Press-Note-28Nov2018.pdf. Poonam Gupta et al. 89 the data for the years prior to fiscal year 2011–12 were available with the base year 2004–05 (old series). The CSO reported several methodological changes/upgrades in line with the recommendations of the United Nations System of National Accounts (SNA 2008) while preparing the back series. These include: (a) using the Ministry of Corporate Affairs (MCA) database and public sector data in manufacturing and electricity, respectively, which was previously estimated by annual reports of private electricity companies and the Annual Survey of Industries (ASI); (b) use of sector-specific CPI’s for the non-financial services (new CPI from 2011–12 and CPI-IW before that) versus the use of CPI (AL) and CPI (IW) in the previous series; (c) use of sales tax data to project growth in the trade sector; and (d) using minutes of usage rather than telecom subscriber growth to estimate the communication sector activity. However, in some cases—due to the data availability limitations for earlier years—either the splicing method or ratios were used for estimates in the base year of 2011–12. In the new series, GDP growth is lower between 2005–06 and 2011–12. Average GDP growth in the old series (base 2004–05) from 2005–06 to 2011–12 was 8.2 percent, while average growth in the new series is 6.9 percent, about 130 bps lower. Moreover, growth is lower in each of the years from 2005–06 to 2011–12, ranging from 60 bps in 2009–10 to 210 bps in 2007–08 (Figure 29). Growth of GVA is also consistently lower in the new series. Average GVA growth in the old series was 8.5 percent compared to 6.9 percent in the new series over the same period. The differential in average growth between the two series on the supply side was even higher, at 160 bps. An implication of this is that average growth of Net Indirect Taxes (NIT) was higher by about 30 bps between 2005–06 and 2011–12.48 Changes in growth are largely attributable to changes in the deflators. There is little change in the level of nominal GDP between the old and the new series. The nominal GDP growth differential across the new and old series is very modest. Much of the reduction in growth of real GDP is because of changes in the deflator. We calculate that of the 130 bps difference in real GDP growth, about 100 bps is attributable to the change in the deflator while only 30 bps is attributable to changes in nominal growth. One implication of the new series is that the economy between 2006 and 2012 was characterized by weaker growth but higher inflation (as captured by the GDP deflator).

48. GDP is computed as the sum of GVA (supply side) and net indirect taxes (indirect taxes less subsidies). 90 INDIA POLICY FORUM, 2018

FIGURE 29. Growth Is Consistently Lower in the 2011–12 Series Compared to 2004–05 Series

29A: Real GDP Growth Is Consistently Lower… 29B:…As Is Real GVA Growth GDP Growth GVA Growth 12 12 10 10 8 8 % % 6 6 4 4 2 2 0 0 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2011–12 Series 2004–05 Series 2011–12 Series 2004–05 Series

29C: Percent Change in Deflators in the New Series Is 29D: Much of the Reduction in Real Growth in the Consistently Larger New Series Is Attributed to a Higher Deflator % Change in GDP Deflator 1.8 2011–12 v/s 2004–05 Series 12 10 1.5 Reduction in real 8 1.2 GDP/GVA Reduction due to % 6 0.9 higher deflator 4 0.6 Reduction 2 due to 0.3 nominal 0 0 GDP/GVA

2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 GDP GVA 2011–12 Series 2004–05 Series Deflator Nominal

Source: Data from the CSO. Note: Years refer to fiscal years.

Services deflator explains most of the changes on the supply side. On the supply side, growth of the services sector is considerably lower in the new series while the growth of the agriculture and industry sectors remain comparable (Figure 30). The services deflator explains most of the real growth differential across the two series. Within the service-sector deflator, the most notable changes are to the deflator in the financial services sub-sector. Figure 30 (Panel F) shows that the deflator in this sub-sector witnessed sharp changes under the new methodology.49 On the demand side, lower growth under the new series was concentrated within the consumption sector (mainly private). In contrast, gross fixed capital formation (investment), export and import growth were virtually

49. The CSO’s press release notes first that the methodology for estimating GVA of the financial sector (unorganized) was revised to include stock exchanges, stockbroking companies, and asset management companies under the coverage of financial corporations. Second, the deflator used is an index based on the implicit price deflator of the nonfinancial sector. Poonam Gupta et al. 91

FIGURE 30. Changes in Services Sector Growth Explain Most of the Differential on the Supply Side

30A: Agricultural Growth Remains Comparable… 30B: As Does Industry Growth Agriculture Growth Industry Growth 10 14 8 12 10 6 8 % % 4 6 2 4 2 0 0 –2 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2004–05 Series 2011–12 Series 2004–05 Series 2011–12 Series

30C: Growth of Services Sector Is Considerably Lower 30D: Among Modern Services, Financial Services in the New Series… Growth Has Been Significantly Revised Downward Services Growth Financial Services Growth 12 25

10 20 8 15 %

% 6 10 4 5 2 0 0 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2004–05 Series 2011–12 Series 2004–05 Series 2011–12 Series

30E: Percent Change in Services Deflator 30F: …of which Change in Financial Services Deflator Is Very Notable % Change in Services Deflator % Change in Financial Services Deflator 12 20 10 15 8 10 % 6 % 5 4 0 2 –5 0 –10 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2004–05 Series 2011–12 Series 2004–05 Series 2011–12 Series

Source: CSO data. Note: Years refer to fiscal years. identical across the two series. This implies that the drivers of growth have not meaningfully changed over the last decade (Figure 31). Notwithstanding the data update, the key results on growth acceleration, stability, and diversification across sectors remain robust and unchanged (Figure 32). In Table 6, we present the trend in growth of GDP, per capita GDP, and sectoral growth for both the old and new series. While the 92 INDIA POLICY FORUM, 2018

FIGURE 31. On the Demand Side, Consumption Growth Is Lower in the New Series

31A: Consumption Growth Has Been Consistently Lower 31B:…while Investment Growth Has Remained Virtually in the New Series the Same Consumption Growth Investment Growth 10 20

8 15 6 10 %

4 % 5 2 0 0 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 –5 2004–05 Series 2011–12 Series 2004–05 Series 2011–12 Series

31C: Export Growth Has Also Remained Virtually the Same 31D:…along with Import Growth Export Growth Import Growth 30 40 30 20 20

% 10

% 10 0 0 –10

–10 –20 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2004–05 Series 2011–12 Series 2004–05 Series 2011–12 Series

Source: CSO data. Note: Years refer to fiscal years.

FIGURE 32. Growth Acceleration and Stability (under the New Series)

32A: Growth Continues to Accelerate in the New Series… 32B: … and Has Become More Stable 10-Year Rolling Average: GDP Growth 10-Year Rolling Coefficient of Variation 8 160 7 140 6 120 5 y = 0.0903x + 3.7341 100 R² = 0.8851

% 4

% 80 3 60 2 40 1 20 0 0 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018

Source: CSO data. Note: Years refer to fiscal years. Poonam Gupta et al. 93

TABLE 6. Trends in GDP and Sectoral Growth with the Old and New Series A: Old Series (1) (2) (3) (4) (5) GDP GDP per Agriculture Industry Services Variables Growth Capita Growth Growth Growth Growth Trend 0.109*** 0.130*** 0.0217 0.0970*** 0.0961*** (3.881) (4.645) (0.364) (3.078) (3.409) Observations 47 47 47 47 47 R2 0.251 0.324 0.00294 0.174 0.205 B: New Series (1) (2) (3) (4) (5) GDP GDP per Agriculture Industry Services Variables Growth Capita Growth Growth Growth Growth Trend 0.0932*** 0.114*** 0.0184 0.0863** 0.0802*** (3.403) (4.188) (0.308) (2.664) (3.853) Observations 47 47 47 47 47 R2 0.205 0.280 0.00210 0.136 0.248 Source: CSO data and authors’ calculations. Note: t-statistics in parentheses. *** p<0.01, ** p<0.05, * p<0.1 indicate level of significance. coefficients based on the new series are slightly lower, they do not dampen any of the key results of this paper. GDP growth continues to accelerate and has become more stable (under the new series) and the growth remains diversified across sectors. Moreover, since the back series has not affected the estimates of national accounts in nominal terms significantly, the key ratios such as investment rate, savings rate, exports to GDP, imports to GDP, consumption to GDP, remain virtually the same.50 The long-term trends in these ratios are, therefore, unaffected.

50. Graphs suppressed for brevity. Available on request. Comments and Discussion*

Sudipto Mundle National Institute of Public Finance and Policy

This is an impressive paper with a detailed, careful analysis of data to address three issues: long-term growth dynamics, three phases of growth in the economy, and the challenges going forward for India to maintain a high rate of growth. I basically agree with most of what has been said in the paper. Indeed, some of us have also been asking the same questions and giving similar answers related to investment rate changes, degree of openness of the economy, and so on. In a sense, these are fairly conventional answers. Today, I want to present a slightly different way of looking at the growth story, mainly to encourage a discussion. But first let me start with a question about the data. When I was first asked to comment on this paper, I was curious about how the authors would tell a long-term growth story of the Indian economy when we actually don’t have comparable GDP data prior to 2011–12. The new GDP series was launched with 2011–12 as the base year. Usually when that happens, the Central Statistical Organisation generates the back series so that researchers, analysts, and others have a long comparable series to perform the time series analysis. In this case, this was not done, partly because the new data sources being used, particularly the corporate balance sheet data (the MCA 21 database) of the Ministry of Corporate Affairs are not available for the previous years. The authors have thus generated their own back series. Coincidentally, the National Statistical Commission (NSC) has appointed a committee, which I happen to chair, that has also been mandated, among other things, to generate the missing back series. This work has just been completed. Unfortunately, it will take a week or two of internal processing before the new back series is made available in the public domain by the

NSC for discussion and comments. Perhaps it will be available when the paper presented today is revised. Meanwhile, I can give a comparison of the two ways in which the backcasting has been done in this World Bank paper and by my NSC-appointed committee. In both cases, the growth rates of the old series have been largely preserved, which is desirable, and this chain of growth rates has been linked to the higher GDP level of the new 2011–12 series, as compared to the old (2004–05 base) series. The back series generated for this World Bank paper has not been presented in this paper, but my assessment is that in their back-casting, the level of adjustment is fully back loaded onto the old base year (2004–05), because after that the old series growth rates are being maintained to reach the higher GDP level of the new series in 2011–12. What we have done in the NSC committee is to maintain the old growth rates, but incrementally adjust the output level each year, distributed over the entire back series period, so that cumulatively the back series ends up with the same 2011–12 output as in the new series. This assumes a smooth, gradual change in the production structure as compared to the one-time, big-bang change in 2004–05, as implicit in the World Bank paper. We felt that the gradual adjustment assumption is more realistic. Let me now go to the first part of the paper, which discusses the long- term growth dynamics. Consider the following thought experiment. Take an economy with two sectors; one which is faster growing, less volatile, and has higher productivity, and another which has lower growth, higher volatility, and lower productivity. If the growth process is repeated over a number of years, after say 10–15 years, the inter-sectoral differences initially embedded in the economy will endogenously generate a significant change in the production structure of the economy. Collaterally, the growth rate of the aggregate economy will have accelerated, it will appear to be less volatile, and productivity will have gone up significantly even though there is no change in productivity, volatility, or the growth rate in either sector. I obviously have in mind agriculture and non-agricultural as the two sectors, but the same logic would apply at any level of disaggregation. These are exactly the three main long-term trends that have been identified in the paper: growth acceleration, rising productivity, and declining volatility. The reason why I am emphasizing this is not because other things like changes in technology, products, and policy, or other shocks are unimportant, but to make the point that the impact of all these other changes and shocks are imposed on top of the underlying long-term structural changes arising from the simple arithmetic of weighted average dynamics as I have explained. The underlying structural changes 96 INDIA POLICY FORUM, 2018 would have occurred in any case without any of the additional technological or policy changes. Endogenous structural change and its consequences are an important part of the growth story. Let me now refer to the forward-looking part of the paper, which discusses what is likely to come later. Based on the long-term dynamics and the three phases that have been analyzed at higher granularity, including the higher frequency data for the last couple of years, a number of factors have been identified as the drivers of growth. Along with factors like India’s low dependency ratio, the investment rate changes, or changes in the degree of openness, the authors also cite reforms. However, the endogenous structural change which I discussed, together with the evolution of the dependency ratio, the rise and subsequent decline of the investment rate, and changes in the degree of openness can by themselves completely explain everything that has happened without any reference to reforms. This includes not just the long-term growth trends but also three growth phases that have been identified: the period of slow growth prior to 1991, actually prior to the mid-1980s; the later gradual increase in growth followed by the sharp growth acceleration between 2003–04 and 2007–08; and finally the slowdown post the financial crisis. This is not to suggest that reforms are unimportant. Indeed the rise and decline of the investment rate and changes in the degree of openness were themselves driven, among other things, by the post-1991 reforms and some subsequent retrogression. However, to tell an interesting story of India’s growth, it is important to distinguish between the different layers of causality: endogenous structural change, macroeconomic trends such as changes in the dependency ratio, the investment rate or degree of openness, and reforms in the institutional or regulatory framework. These drivers operate at different levels of causality and need to be treated separately, including interactions among the drivers themselves.

Dilip Mookherjee Boston University

The first half of this paper provides a range of interesting facts concerning long-term growth in India over the past five decades: accelerated growth rates, especially since 2000, accompanied by a reduction in growth volatility. They show striking contrasts with seven other emerging market economies, and that accelerated growth was sustained and uniform across Poonam Gupta et al. 97 different states within India. The acceleration was accompanied by a rise in investment rates, the role of the service sector, of credit, trade, and foreign direct investment, and in productivity growth rates. Moreover, traditional “structuralist” school of macroeconomic factors such as agricultural growth, domestic demand, or public investment did not seem to matter. Neither did “structural transformation” from agriculture to industry, or human capital growth. The paper is very clear in showing this. The second part of the paper then moves to a granular short-term perspective of performance of the economy since the 2007 crisis. The theme and style of this section is somewhat at odds with the first part. The authors describe three phases in the evolution of the economy since 1991: pick-up of growth rates, unusually fast growth during 2004–08, the slowdown from 2008, and the recent recovery. The post-2007 period was marked by a slowdown in investment, exports, credit quantity, and quality. “Structuralist” factors played a more important role in these short-term movements. The tension between the two parts could be that the 1990–2007 acceleration was a one-time phenomenon, and that it will be difficult to sustain in the years ahead. I suppose this is the million rupee question—to what extent will it be possible to sustain the high growth rates of the past three decades? Having achieved 7 percent growth over a couple of decades is an extraordinary achievement by worldwide historical standards; the prospect of an acceleration to an even higher long-term growth rate is hardly likely. The relevant question instead is, will it be possible to sustain, say, 6 percent plus growth over the next two decades? What will this take? Or will we slip back to a 2–3 percent growth on par with other developed and middle income countries and our own pre-1990 record? For me personally, the record of 7 percent over the last three decades, which makes India’s growth record just one notch below the Chinese over the same period, seems rather miraculous. This is particularly in light of the myriad growth bottlenecks that afflict India: frictions in land and labor markets, weak governance (especially urban governance), poor infrastructure, lower human capital, and lower spending on higher education and R&D. Both countries have weak judicial contract enforcement mechanisms. My hunch is that the conventional discussions of the drivers of growth in academic and policy circles are missing the essence of the Indian growth story. Before I discuss this in more detail, let me discuss the credibility of the data used. The recent changes in national income accounts methodology do not seem to pose a major problem: while the changes may change measured 98 INDIA POLICY FORUM, 2018 levels, they are less likely to distort the measured growth rates from one year to the next, or the long-term growth estimates. The data quality for saving and investment rates is more worrying, as they are often measured as residuals in the national income data and suffer from multiple price deflation and aggregation issues. Even more worrying are problems in measuring activity in the unorganized sector. Some of the measured growth acceleration may reflect a progressive shift of activities from the unorganized to organized sectors and improving coverage of the unorganized sector over time. While others are more qualified to comment on these issues, my guess is that these may account for part of the higher growth rates of the past two decades. But they are unlikely to account for all of it. The newfound dynamism is all too palpable for it to be entirely a statistical illusion. Let me return to the question of the possible explanations for India’s growth acceleration. It is important to note that the facts pertain to second derivatives (growth acceleration) and moments (volatility) rather than the first derivative (the growth rate), which is what most growth theory is about. Neoclassical growth theory does not seem relevant, as it is driven by investment rates and productivity growth, which are treated as exogenous. For given investment and productivity growth rates, neoclassical theory predicts growth rates will fall over time, that is, the phenomenon of convergence. Growth acceleration is consistent with neoclassical theory combined with rising investment and productivity growth, as has been observed in India. But then we need to understand the sources of rising investment and productivity. Endogenous growth theory a la Romer (1986, 1990) or Lucas (1988) also does not seem that relevant, rooted as it is in externalities generated by R&D and human capital, respectively. But the broader idea of learn- ing and spillovers could be relevant. The post-1991 liberalization opened the economy to higher quality inputs embodying improved technology, thereby allowing a range of new products to be produced (Goldberg et al. 2010). Perhaps liberalization of entry of new firms paved the way for agglomeration externalities to be realized, involving the sharing of key inputs, market access, and diffusion of know-how among new entrepreneurs. As in China, these may have been facilitated by efforts to provide entry and investment stimuli via special economic zones (SEZs) since the early 2000s: the recent work of Hyun and Ravi (2018) finds evidence of increased night-light intensity (a measure of local economic activity) in the close vicinity of SEZs after they went into operation, rising investment, productivity, employment and wages in the formal sector, and shifts from the informal to the formal sector. Poonam Gupta et al. 99

Community-based networks may have been an important channel for spillovers. Production clusters in specific industries consisting of dense concentration of firms of small scale, marked by a high degree of specialization, extensive subcontracting, informal pooling of capital, and risks among entrepreneurs of similar social origins have been documented in specific industries that achieved high growth rates in China (Long and Zhang 2011, 2012) and India (Banerjee and Munshi 2004; Munshi 2011). The networks are based on different sources of “social capital”—clan and common place of birth in China (Deng et al. 2018; Greif and Tabellini, 2017; Peng 2004) and subcaste in India (Munshi 2014). High social and economic interdependence within these communities provided opportunities to overcome barriers to entrepreneurship—access to capital, know-how and contract enforcement mechanisms—owing to weak market and state institutions. Once some entrepreneurs from a certain community gain a foothold, thanks to the liberalization of entry and investment regulations, the dynamic community-based spillovers could have triggered the entry of subsequent waves of new entrepreneurs from the same community in a cascade-like manner, resulting in growth acceleration phases. Empirical analyses of such community-based growth are hampered by lack of data concerning informal firms and the social origins of their entrepreneurs, and the inherent difficulty of identifying and estimating across-firm spillovers. However, the recent availability of data in China on all registered firms since the late 1970s and the birthplace of their principal entrepreneurs has enabled me and my co-authors to test such a model of community network-based firm entry and growth for the entire Chinese economy (Dai et al. 2018). We estimate the contribution of these origin community-based spillovers of firm entry and capital stock invested in private firms between 1990 and 2009 at 40 percent nationwide after controlling for a large range of constant and time-varying destination-specific factors which include geography, local infrastructure, and government support. What are the limitations of such community-based growth mechanisms? These have been less studied, but various case studies are instructive. By their very nature, the growth spurts are restricted to specific communities and can result in fast transitions from pre-industrial occupations to industrial entrepreneurship over narrow intervals of time, followed by subsequent plateauing once the transition is completed for most potential entrants from the community. So they can result in short-lived growth spurts that are hard to sustain and are inherently uneven in their incidence. They have been restricted to low-end manufacturing industries and service sectors, where requirements for entrepreneurial know-how, education, and capital are low. 100 INDIA POLICY FORUM, 2018

Market size constraints and low quality of products can also limit growth. Sustaining high growth rates requires upgrading of quality and access to export markets, that is, higher technology, capital requirements, and verti- cal integration, for which traditional community-based clusters may be ill-suited. Similar problems seem to be afflicting the growth of the Indian IT sector, as they face the challenge of growing international competition, automation, and the need to upgrade to new kinds of services. Apart from this, the various constraints mentioned by the authors in the second half of the paper in raising investment, exports, and improving credit quality are also likely to matter. And so are many other constraints they do not mention: resources (energy, water, and land), pollution, poor urban governance, rising inequality, and resulting political pressure to redistribute via policies that reduce efficiency and investment incentives. Sustaining 6 percent plus growth will not be easy. But then we economists tend to be pessimistic by nature, forever tending to gloomy forecasts, unable to predict growth miracles in advance, and explaining them after they happen.

References

Banerjee, A. and K. Munshi. 2004. “How Efficiently Is Capital Allocated? Evidence from the Knitted Garment Industry in Tirupur,” Review of Economic Studies, 71(1): 19–42. Dai, R., D. Mookherjee, K. Munshi, and X. Zhang. 2018. “Community Networks and the Growth of Private Enterprise in China,” Working Paper. Available at http://people.bu.edu/dilipm/wkpap/ChinaoverallV23.pdf (accessed on January 24, 2019). Goldberg, P., A. Khandelwal, N. Pavcnik, and P. Topalova. 2010. “Imported Intermediate Inputs and Domestic Product Growth: Evidence from India,” Quarterly Journal of Economics, 125(4): 1727–1767. Greif, A. and G. Tabellini. 2017. “The Clan and the Corporation: Sustaining Cooperation in China and Europe,” Journal of Comparative Economics, 45(1): 1–35. Hyun, Y. and S. Ravi. 2018. “The Effect of Place-Based Development Policies: Evidence from Indian SEZs,” Working Paper 306, Boston, MA: Institute for Economic Development. Long, C., and X. Zhang. 2011. “Cluster-Based Industrialization in China: Financing and Performance,” Journal of International Economics, 84(1): 112–123. ———. 2012. “Patterns of China’s Industrialization: Concentration, Specialization, and Clustering,” China Economic Review, 23(3): 593–612. Lucas, R. 1988. “On the Mechanics of Economic Development,” Journal of Monetary Economics, 22(1988): 3–42. Poonam Gupta et al. 101

Munshi, K. 2011. “Strength in Numbers: Networks as a Solution to Occupational Traps,” Review of Economic Studies, 78(3): 1069–1101. ———. 2014. “Community Networks and the Process of Development,” Journal of Economic Perspectives, 28(4): 49–76. Peng, Y. 2004. “Kinship Networks and Entrepreneurs in China’s Transitional Economy,” American Journal of Sociology, 109(5): 1045–1074. Romer, P.M. 1986. “Increasing Returns and Long-Run Growth,” Journal of Political Economy, 94(5): 1002–1037. ———. 1990. “Endogenous Technological Change,” Journal of Political Economy, 98(5, Part 2): S71–S102.

General Discussion

Chaired by Subhash Garg Secretary, Department of Economic Affairs, Ministry of Finance

Pranab Bardhan noted that economists tend to relate only reforms to long- run growth. Since the mid-1980s, Indian reforms have been accompanied also by massive social change. Subordinate castes have become politically more assertive and economically more active. Barriers to entry have fallen, and rising caste groups have entered small business, often investing their agricultural surplus. Economists have not tried to quantify the impact of this social development on growth dynamics. Much of this impact is in the informal sector, for which more data should be available now. The NSS has recently released data on unincorporated enterprises for two years (2011–12 and 2015–16). We should correlate the data on these informal and household enterprises with their social composition, as was done by Banerjee and Munshi (2004) for peasants from a certain caste who were now in the knitted garments industry. Surjit Bhalla commented that on the investment rate going down, we need to be careful about the deflator. If we look at it in real terms, the price of investment goods had not gone up as much as the price of other goods and, therefore, the investment rate was declining in nominal terms. In real terms, the decline had been about 3 percentage points, but the rate was still quite high. On GDP measurement, given all the confusion, he was really looking forward to the paper on GDP that the Bank paper referred to. On what determines growth, he agreed with Dilip Mookherjee that a large part of the growth story in developing countries could be explained by the very large increase in human capital and educational attainment. 102 INDIA POLICY FORUM, 2018

Devesh Kapur noted on Dilip Mookherjee’s point on R&D that he was surprised to learn that in the last decade, India had been the fastest growing country globally in the number of papers published. A recent US National Science Foundation report shows that the share of global papers from India doubled from about 2.5 to 5 percent between 2006 and 2015. India was now the third, after China and the USA. India’s gross enrolment ratio in higher education is considerably higher than China’s at a comparable level of income when China was at India’s level of income. Thus, in terms of the numbers on human capital and R&D, India had shown a considerable acceleration in the last decade. Vijay Joshi suggested that the title of the paper should be changed to “Some Macro Aspects of India’s Growth Story” because it did not deal with a lot of things such as education, reform of the state, and of state capacity in institutions that were essential to sustained, high growth. Second, he suggested that in the discussion on short-term dynamics, there was no mention of how much of the deterioration in investment was due to increased fiscal deficits. Crowding out was not a good explanation, since this was a period of exceptional monetary easing as real interest rates were extremely low immediately after the global crisis. How much was it fiscal deficits, and how much, for example, was it the debt overhang? Third, the paper merely listed some factors for export growth. How important was the significant appreciation of the exchange rate? Could we have achieved different results if the oil price bonanza witnessed after 2014 had been used differently? It would help for the paper to answer these important questions. Fourth, he said that the investment rate diagrams in the paper showed that there was a fall in the corporate investment rate after the global crisis, but it was followed by remarkable stability. So most of the decline after 2011 seemed to be in household investment, not in corporate investment. Should that be believed, and if not, why not? Finally, he wanted to know how investment could be revived given the current stress in Indian banking. He thought the paper should have addressed some of these questions. Sajjid Chinoy complimented the authors on a comprehensive overview of India’s growth story. First, he also advocated deeper analysis of India’s investment dynamics in the last decade. We assume that it is large corporate investment that had declined, and fixing up the banking system could reverse this. Corporate investment fell starting in 2008, but had actually then stabilized, even picking up subsequently as a percentage of GDP. A closer look at the data showed that a lot of the decline was Poonam Gupta et al. 103 related to the fall in household investment, which included investment by small and medium enterprises, over the last five or six years, even well before demonetization and the advent of GST. The implication was that without changing the economic viability of these smaller enterprises, say through factor market reforms, we may not be able to go back to those higher investment rates. Second, he stressed that the last six or seven years have shown an interesting dichotomy in what the authors have called the golden period of growth; a strong synchronization between India’s growth and global growth and then a de-synchronization in both directions in the last five years. Even as the global economy slowed down for three or four years, the Indian economy actually accelerated sharply, but as the global economy started recovering in the last two years, India grossly under-performed. One of the proximate reasons for this could be that India’s exports were not responding to stronger global growth, which implied that in a stronger global environment the country was not getting the benefit of net exports, but had to bear the cost through higher commodity prices and the adverse terms of trade. The authors needed to do some thinking on this because this de-synchronization over the last 5–6 years was being seen for the first time in 15 years. Third, he said he was getting nervous about the talk of 8 and 9 percent growth over long periods. The entire high growth of 9 percent achieved earlier was attributable to exports, and the entire slowdown from 8.8 percent to 6.9 percent in the last five years was attributable to the slowdown in the value added growth of exports. If this is the new normal, how do we sustain 7 percent, leave alone 9, in this gloomy global situation? This made it important, as suggested by Vijay Joshi, to identify other drivers of growth in order to push growth without stoking external imbalances, which was not a constraint when export-led growth was being achieved. Abhijit Banerjee agreed with Sajjid Chinoy that the Indian household sector actually included what might be called firms elsewhere. One of the striking facts about India was how small firms were relative to the industry. The decline in investment in the household sector may actually relate to firms becoming bigger and therefore not considered household. During 2003–08, start-ups in India had started becoming bigger. So if there was a shift in household investment to more formal firms, was that good news or bad news? Banerjee also suggested that the paper should take on the question of the fiscal arithmetic of India and how things might square up with the adverse trends in deficits and government borrowing we were seeing. He thought the only way this could be done would be a return to the bad 104 INDIA POLICY FORUM, 2018 days of very high inflation and financial repression. He thought the paper should comment on this. Ila Patnaik wanted to draw attention to what Sudipto Mundle had brought up about the formal and informal sectors. She said that she would like to see an analysis in the paper about whether some of the recent initiatives, such as GST or demonetization, were having an impact on pushing informal firms into the formal sector. What did the data say about decline in informal sector jobs, how small informal firms were doing? Perhaps there was an increase in jobs covered by EPFO, as shown by Ghosh and Ghosh (2018). In many instances in the short-run, these changes could be disruptive and the sectors affected may not do well, but in the long run, they could signify an improvement in sector productivity. What did the paper have to say about these possible developments? Shantayanan Devarajan noted that the biggest surprise for him from the paper was the balanced nature of Indian growth it showed, both the balance across sectors and the balance across states. He wondered how that could be reconciled with the unmistakable increase in inequality during this period, since balanced growth should reduce inequality. He echoed Dilip Mookherjee’s point and asked if there was an underlying growth model for India that could explain both the balanced growth and the increase in inequality. Karthik Muralidharan echoed the point that the underlying growth model might explain this growth pattern. He didn’t think it needed much more data, but it would be amazing to see this done at the state level. This could provide a lot of clues to the questions being raised. India was just one case study, but looking at the state level variation could really shed light on correlates. Govinda Rao observed that there appeared to be an 8- to 10-year cycle in the Indian growth story, it seemed tied to the Pay Commission awards and, perhaps, even the cyclical movement of oil prices. For instance, the Pay Commission Award in 1977–78 was followed by skyrocketing oil prices in July 1978. This cycle was also seen in 1989–90, then in 2000, and again in 2007–08. The Pay Commission, expansion of the Mahatma Gandhi National Rural Employment Guarantee Act, and loan waivers, were the three major issues in the 2007–08 budget, resulting in a sharp rise in the fiscal deficit. The Global Financial Crisis came much later. His point was that the decline in our macro stability and investment had much more to do with the country’s domestic policies than the Global Financial Crisis, and the paper would benefit from considering these. Poonam Gupta et al. 105

References

Banerjee, Abhijit and Kaivan Munshi. 2004. “How Efficiently Is Capital Allocated? Evidence from the Garment Knitted Industry in Tirupur.” Review of Economic Studies, 71(1): 19–42. Ghosh, Pulak, and Soumya Kanti Ghosh. 2018. “Towards a Payroll Reporting in India.” A Study by Indian Institute of Management, Bangalore, and State Bank of India.

MELISSA LOPALO* Montana State University KEVIN KURUC† University of Oklahoma MARK BUDOLFSON‡ University of Vermont DEAN SPEARS# University of Texas-Austin Quantifying India’s Climate Vulnerability§

ABSTRACT This paper asks about the climate damages that Indian policymakers can expect. What is the likely magnitude of climate damages, and how sensitive are they to the level of warming? How much worse would climate damages be for Indians under, say, 5° of warming rather than 3°? Understanding the magnitude of climate damages and how rapidly they increase as temperature change increases is critical for finding the right climate mitigation policy. This paper provides projections of India’s climate vulnerability on the basis of new microeconomic and macroeconomic evidence. The authors’ quantifications show that India is highly vulnerable to climate damage. Their baseline macroeconomic approach suggests that climate change peaking at 5°C, rather than 3°C, would be as detrimental to Indian well-being as a reduction in GDP by about 18 percent for each of the years from 2020 to 2040. Such an equivalent threat to near-term economic outcomes would be an overriding policy priority if political leaders anticipated it. The authors’ microeconomic results suggest that this may be an underestimate, because it ignores humidity. Emerging evidence suggests that humans are especially vulnerable to exposure to high temperatures in contexts of high humidity; humidity is a previously underappreciated and unquantified reason way in which India may be more climate-vulnerable even than some hotter developing and middle-income countries.

Keywords: Climate Change, Climate Damages, India

JEL Classification: Q51, Q58, O11, H23, E10

* [email protected] † [email protected] ‡ [email protected] # [email protected] § The authors would like to thank Karthik Muralidharan, Navroz Dubash, Shreekant Gupta, and conference participants at the NCAER 2018 India Policy Forum for helpful comments and suggestions on the paper. 107 108 INDIA POLICY FORUM, 2018

1. Introduction

he Intergovernmental Panel on Climate Change (IPCC) predicts an Toverall increase in the earth’s temperature over the next century due to climate change caused by human greenhouse gas (GHG) emissions, calling it “virtually certain” that there will be more frequent hot temperature extremes and less frequent cold temperature extremes experienced over most land areas (Solomon et al. 2007). A large body of literature from the IPCC and other researchers has estimated or projected economic, health, and other costs of climate change, finding that the net effect on humans will be negative on balance and potentially very large (Field et al. 2014). Much of this literature focuses on developed countries. Less is known about the adverse effects of exposure to higher future temperatures on health and economic activity in developing countries and emerging economies. As leading economists recently argued in Science, the focus in the prior literature on rich countries is “problematic, both because developing countries currently represent the majority of the world’s population and GHG emissions and because the nature of impacts and context for policy choice could differ greatly relative to developed regions” (Burke et al. 2016). Exposure to extreme temperatures is often greater in developing nations, which are disproportionately located in the hotter tropics. Harms conditional on exposure could also be greater: the poor may be less resilient to weather’s impacts due to worse overall health. And poor populations may be less able to adapt by reducing exposure to extreme heat and humidity, such as via climate-controlled housing and indoor work. This paper asks about the climate damages that Indian policymakers can expect: what is the likely magnitude of climate damages, and how sensitive are they to the level of warming? In other words, how much worse would climate damages be for Indians under, say, 5° of warming rather than 3°? Understanding the magnitude of climate damages and how rapidly they increase as temperature change increases is critical for finding the right climate mitigation policy. Reducing emissions has costs, in part because emissions are a by-product of productive economic activity, and in part because cleaner fuel choices can be more expensive than carbon-emitting fuel choices. These costs are especially salient for a developing country such as India, where many households still lack reliable electricity, and where foregone economic growth implies an important loss of well-being for all Indians. Public economics has a straightforward theoretical answer to externality problems such as climate change, where one decision-maker’s action causes Melissa LoPalo et al. 109 external harm to other people. Policy should be chosen so that the marginal social costs of reducing pollution equal the marginal social costs of the harm that is being averted. Still, applying this simple theory is difficult. One difficulty lies in even knowing the quantitative extent of the harm. Because climate change will impact many people—rich and poor; urban and rural; men and women; voting age citizens and their young children and future descendants—understanding the total sum of the harm requires comparing unalike consequences for unalike people (Dennig et al. 2015; Edenhofer et al. 2014). Another well-known difficulty is the politics of collective action: the globally optimal policy package, if it could be enforced for the whole world, may importantly differ from what is in the interest of one country’s population, especially the people alive at one time. Under the 2015 Paris Agreement, global mitigation policy will be made through countries’ own bottom-up pledges (Budolfson et al. 2019; UNFCCC 2015). To know what to pledge, Indian policymakers need to know the stakes for India. Therefore, our research speaks to the question of what it would be rational for an Indian policymaker to choose in the self-interest of the full Indian population, present and future. As we will detail, when we tally the social costs of climate change, we consider only costs to the population of India. In short, we find that the cost of climate damages for India is likely to be very large. Although India’s climate vulnerability has been widely discussed in the prior literature, quantification is necessary for domestic analysis and policymaking. Moreover, emerging evidence suggests that Indians may bear an even greater share of global climate damages than has been previously understood. For example, because of the combination of heat and humidity of the Indian monsoon months, and because human bodies are more stressed by thermoregulation in humid air than in dry air, India may face a much larger early-life mortality burden from climate change than sub-Saharan Africa (Geruso and Spears 2018). Among the many tragedies of climate change is the fact that India and other developing countries have not been responsible for much of the world’s carbon emissions to date, but Indians nevertheless stand to lose much from climate change. Our quantification of these losses emphasizes the depth of the policy challenge: what is India’s best, rational response to this climate injustice? This paper reviews and integrates microeconomic and macroeconomic literature in turn. Our analysis emerges from recent collaborative academic research by its authors, especially microeconometric research by Geruso and Spears (2018) and LoPalo (2018) about the consequences of heat and humidity in combination, and macroeconomic research by Budolfson et al. (2018) 110 INDIA POLICY FORUM, 2018 about the dependence of optimal mitigation policy on the unknown future trajectory for economic development of poor, climate-vulnerable countries. But we are far from the first to raise these themes, and we build upon an accomplished literature at the intersection of environmental and development economics (Dennig et al. 2015; Edenhofer et al. 2014; Greenstone and Jack 2015; Hallegatte et al. 2016; Hijoka et al. 2014; Olsson et al. 2014). Section 2 considers microeconometric evidence. It considers causally well-identified effect estimates of harms of climate exposure and uses them to project future damages within India under alternative possible futures for climate policy and outcomes. Section 3 presents macroeconomic projections. In this section, we make a novel application of the Regional Integrated Climate-Economy (RICE) climate-economy model, which was developed originally by the Yale University economist William Nordhaus (Nordhaus 1992, 2010; Nordhaus and Boyer 2000). As a global model that explicitly represents different nations, RICE includes assumptions, on the basis of scientific literature, that explicitly represent India’s economy and quantify India’s climate vulnerability. We use RICE to illustrate India’s climate vulnerability by computing the magnitude of hypothetical near- term consumption losses to all Indians that would be an equal-sized loss to social welfare as climate damages. In other words, assuming a method for aggregating social harm across present and future Indians, what size of near-term economic disaster in the shape of consumption losses would be comparably bad as climate damages are projected to be? These results will be underestimates, because in using the RICE damage function, we conser- vatively ignore the new evidence of humidity-based damages in Section 2. Section 4 briefly builds upon Greenstone et al.’s (2017) NCAER India Policy Forum study of air pollution. In contrast with climate damages, which will not fully unfold until future decades, India’s population is already exposed to hazardous levels of air pollution today. The interaction between air pollution policy and carbon emissions policy is complex because particles in the air that harm human health can also reduce global warming by reflecting away sunlight. Recent analysis by Scovronick et al. (2019) considers the balance between these mechanisms: for India, the health damages from air pollution dominate the computation and offer a compelling reason to simultaneously reduce air pollution and carbon emissions. Our focus is on understanding and quantifying the damages that India can expect in most of the paper. We turn to policy implications in the concluding section, that is, Section 5. What should Indian policymakers do in response to these grim facts? Elsewhere, we have considered the easier question of what the globally optimal policy would be. In Budolfson et al. (2019), we use Melissa LoPalo et al. 111 the same RICE model to show that the best global emissions policy would take into account inequality in world economic development and the fact that richer countries are more capable of making emissions cuts. The globally impartial, welfare-maximizing policy would have the rich countries such as the USA very quickly decarbonizing, middle-income countries such as India phasing out carbon emissions more slowly over several decades into the 21st century, and the very poorest countries in sub-Saharan Africa perhaps continuing to produce some carbon emissions even in the early 22nd century. But knowing what the globally optimal plan would be may provide little practical guidance to the leaders of India, or any other developing or middle-income country. Decades of highlighting the injustice of developed countries’ emissions policies has done little to change them. Nor, as we show in Section 3, could India acting alone do much to reduce its own climate damages, even by entirely eliminating its carbon emissions. If India is to escape the climate damages that we project, it will require international policy change (in combination with appropriate and effective domestic investment by India in adaptation). Our findings highlight what others have also argued: India’s leadership must approach the challenge of formulating a best response to climate injustice with an understanding informed by the sober facts of the vulnerability of its population.

2. Microeconomic Evidence: The Consequences of Heat and Humidity

In this section, we introduce empirical evidence from microeconometrics about the effects of temperature and humidity on outcomes such as health and productivity. We then compute the implications of these estimates for future Indians, where the combination of heat, humidity, and poverty—especially in the subtropical states of North India—comes together to create a unique context of climate vulnerability. Temperature has been shown to affect many relevant outcomes, from human health, to crop yields, to the productivity of workers. Because researchers cannot observe the future climate, the only available research design is to compare populations and economies exposed to different weather outcomes (or the same population at different times). But simply comparing countries with hot climates to countries with cold climates to learn about the potential impact of climate change is problematic. Climate may be correlated with other variables that are otherwise correlated with economic outcomes. In this case, the researcher would misattribute the effect of these 112 INDIA POLICY FORUM, 2018 other variables to climate. To overcome such difficulties, an active literature in microeconomics uses short-term fluctuations in weather to make comparisons of outcomes on hot and cold days (or months) within the same place. This strategy allows researchers to learn about the impact of temperature and other weather variables separate from other correlated factors (Dell et al. 2014). This literature has documented impacts of weather fluctuations on outcome variables such as conflict (Hsiang et al. 2013), health (e.g., Barreca et al. 2016; Deschenes et al. 2009), and productivity (e.g., Burke et al. 2015; Hsiang 2010). We demonstrate the implications of estimates from this literature for India by using data on current temperature distributions and projections of future weather under various climate change scenarios and the effect sizes from these studies. We note, however, that no empirical study of past consequences of exposure to the weather can fully uncover the possible adaptation that future populations could implement to reduce their exposure to harm. In that sense, these estimates underscore the potential benefits of effective adaptation.

2.1. The Under-appreciated Importance of Humidity From a physiological standpoint, temperature is not the only weather variable that may be important for human well-being. One of the human body’s main mechanisms for cooling itself is sweating, which lowers temperature through evaporation. Sweating is particularly important at high temperatures. Humidity significantly interferes with evaporative cooling: when the air is saturated with more moisture, sweat evaporates more slowly, meaning that the body is less able to cool itself. The results could be dire: when exposed to a combination of heat and humidity that is too extreme, the human body cannot cool itself because neither radiative cooling nor evaporative cooling from sweat will be successful. Under unlikely but feasible bad case scenarios for climate change, as Sherwood and Huber (2010) compute, high heat and humidity could make spending several hours outdoors literally deadly in much of the land surface of the world where humans currently live, including much of South Asia. Recent econometric studies corroborate humidity as an important moderator of the effects of temperature on economic outcomes, even at less extreme levels of exposure. Barreca (2012) shows that hot and humid days are more dangerous than merely hot days in terms of health impacts in the USA. This has implications for the distribution of health outcomes: these results imply that mortality rates will increase more in hot and humid climates than in hot, dry climates as baseline temperatures increase. The Melissa LoPalo et al. 113 literature on temperature and economic outcomes focuses primarily on developed countries, as data on both weather and outcome variables are more readily available in these contexts. However, this evidence suggests that it may be particularly important to understand the impacts of temperature in developing countries: developing countries are disproportionately located in hot and humid areas of the world. In addition, more people in developing countries work outside and fewer have access to adaptive technology such as air conditioning. For these reasons, developing countries are viewed as more vulnerable to the impact of humidity.

2.2. Climate Change and Infant Mortality Motivated by this literature on human thermoregulation, several recent studies estimate the effects of heat, humidity, and their interaction in developing countries. Geruso and Spears (2018) merge Demographic and Health Survey (DHS) data on month of birth and timing of infant deaths with gridded global weather data in four continents. In each country, the DHS collects full reproductive histories from a nationally representative sample of women of reproductive age. These birth histories include the month of birth (and, when applicable, death) for each child, allowing the authors to match data on weather exposure to births occurring years before the interview. Because many babies are born in the same village in different years and months, their large sample of several million births (from every geographically coded DHS before 2015) allows the authors to identify mortality effects using surprise variation in the weather, while controlling for local seasonality, even specific to the village. Like Sherwood and Huber, Geruso and Spears examine the impact of weather using a variable called “wet bulb temperature,” which is a (nonlin- ear) combination of temperature and humidity that gives a more complete portrait of outdoor conditions than temperature alone. In this literature, ordinary temperature is sometimes called “dry bulb temperature,” to distinguish it. Wet bulb temperature is the reading that would be given by a thermometer wrapped in a wet cloth; it is always lower than dry bulb temperature for relative humidity less than 100 percent. Geruso and Spears examine the impact of wet bulb temperature semiparametrically, estimating the impact of replacing a day with a 60–70° wet bulb temperature with a day in nine other bins. They find that hot and humid days in the month of birth predict significant increases in the probability of infant death. Geruso and Spears estimate that an additional day in a month over 85° wet bulb (approximately 32°C at 80 percent humidity) predicts about half an additional infant death per 1,000 births. These extra infant deaths tend 114 INDIA POLICY FORUM, 2018 to occur in the first month of life, when neonates’ bodies are still developing the ability to regulate their own temperature. In Figure 1, we apply the estimate derived from that study to Indian weather data in order to visually investigate the implications for climate change in India. In Panel A of the figure, we first plot the historical implied effects, using average counts of experienced wet bulb degree days above 85° between 2000 and 2010. These weather data come from the Princeton Meteorological Forcing Dataset, which gives information on temperature, humidity, and other weather variables for every 0.25° latitude and longitude grid point.1 We multiply this count by the implied annual effect size. The resulting distribution shows how much lower infant mortality rates per 1,000 births in 2000–2010 would have been in each location if the 85° days were replaced by 60–70° days. The figure shows that these extremely hot and humid days in India have been virtually restricted to the northern states of Uttar Pradesh and Bihar. Moreover, infant mortality rates would be as much as 7 per 1,000 births lower in the most impacted areas if the days over 85° wet bulb were replaced with mild days. This accounts for nearly 10 percent of the infant mortality rates in those regions during the period studied, a non-trivial fraction. Panels B and C of Figure 1 project how climate change may alter the situation depicted in Panel A. Panel B uses projections of heat and humidity obtained from the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP). These data use the Hadley Centre Global Environmental Model (HadGEM2) to predict temperature, humidity, and pressure at a 1° latitude/longitude resolution. These types of projections generally categorize predictions into “Representative Concentration Pathways (RCPs),” which characterize different assumptions regarding the trajectory of future GHG concentration. Panel B uses predictions under RCP 8.5, a pessimistic scenario in which emissions continue to rise throughout the 21st century under assumptions of relatively high population growth and relatively slow income growth, technological change, and energy intensity improvements. Panel B shows the results of these projections for India by 2050. This map uses the same method as Panel A: it shows the increase in infant mortality rate due to the number of days with wet bulb temperatures above 85° using the same effect sizes from Geruso and Spears (2018). Under this scenario, the ill effects of heat and humidity both spread to new areas in India and worsen in already affected areas. In addition to Uttar Pradesh

1. These data are derived from a combination of observational weather data from sources such as satellites, weather balloons, and stations with a physics-based model that extends the data to observationally sparse areas. FIGURE 1. Infant Mortality Rate Increases from Extreme Weather

A: Historical Implied Increase in IMR B: Business-as-usual Scenario C: Aggressive Mitigation Scenario (RCP 8.5) Increase in IMR (RCP 2.6) Increase in IMR

Source: Princeton Global Meteorological Forcing Dataset; ISIMIP; Geruso and Spears (2018); and Authors’ calculations. 116 INDIA POLICY FORUM, 2018 and Bihar, Punjab and West Bengal become more severely affected by the types of hot and humid days that have been shown to affect infant mortality. Still, these types of hot and humid days will continue to be concentrated in northern India. Panel C explicitly computes what is at stake when moving from a “business as usual” (BAU) climate outcome (RCP 8.5) to one requiring more aggressive climate mitigation (RCP 2.6) by showing the differential change in infant mortality under these scenarios. Under RCP 2.6, GHG concen- trations peak mid-century and decline by 2100, representing an optimistic pathway for emissions. This differential infant mortality increase can be seen as the marginal cost of a bad climate outcome relative to a good climate outcome. Specifically, we calculate the excess number of 85° wet bulb days in 2050 under RCP 8.5 relative to RCP 2.6 and then calculate the increase in IMR using the same strategy (again, the thought exercise is the excess in IMR over a situation where the 85° days are replaced with 60–70° days). The result shows the excess IMR that could be prevented by achieving the RCP 2.6 pathway instead of RCP 8.5 and that the preventable deaths are largely concentrated in Uttar Pradesh, Bihar, and the eastern states. All three panels show changes in infant mortality rates and therefore do not take into consideration the current population numbers or population projections in each place. However, these estimates indicate that these deaths will be taking place in some of the most populous regions in India; as of the 2011 Census, Uttar Pradesh was the most populous state while Bihar was the third most populous, together accounting for about a quarter of India’s population. These two states also have the highest fertility rates in the country, implying that a large portion of future births will continue to occur in these especially climate-vulnerable regions.2 Furthermore, Geruso and Spears find that measures of wealth in the DHS do not significantly mediate the impact of wet bulb temperature on mortality, suggesting that even wealthy people in developing countries may be unable to avoid some of the effects of extreme heat and humidity.3

2. Total fertility rates were 2.74 and 3.41 in Uttar Pradesh and Bihar, respectively, in the 2015–16 National Family Health Survey, in contrast to 1.83 in Andhra Pradesh. 3. Some prior literature has found that air conditioners moderated the mortality effects of high temperature in the 20th-century USA (Barreca et al. 2016). This is plausible here as well, in part because air conditioners also reduce humidity. Geruso and Spears cannot test for this, however, because they study developing countries where air conditioner ownership is sufficiently rare to be not measured in the DHS. In the 2005–06 India Human Development Survey (IHDS), only a small fraction of a percentage of households reported owning an air conditioner. Melissa LoPalo et al. 117

2.3. Climate Change and Labor Productivity Infant mortality is an extreme form of climate vulnerability, but it is not the only relevant outcome likely to be affected by the increase in incidence of extremely hot and humid days. Another recent study suggests that this type of weather also significantly impacts labor productivity. LoPalo (2018) examines the impact of weather on a category of workers who are both significantly exposed to outdoor temperatures and possible to study using publicly available data: survey interviewers. In other words, LoPalo uses the DHS surveys to study the effects of exposure to the weather on enumerators as workers. She merges data from over 1.1 million interviews conducted in the DHS with data on temperature and humidity on the day of the interview and examines the impact of daily average wet bulb temperature on indicators of productivity such as number of interviews completed per hour worked as well as measures of data quality. Her analysis shows that, on days when wet bulb temperature exceeds 85°F, the number of interviews completed per hour declines by approximately 10 percent. The effects are driven by an increase in working hours rather than a decrease in interviews completed in a day; interviewing teams start earlier in the morning on these hot and humid days but do not complete their work earlier. She also finds that on hot days, the quality of work decreases: data quality problems are more common. In Figure 2, we perform a similar exercise as in Figure 1, using the effects from LoPalo (2018). These maps plot the annualized estimate of the effect of temperature on productivity (number of interviews completed per hour in this case) multiplied by the number of high wet bulb days in each grid point. As in Figure 1, Panel A of Figure 2 depicts the impact of 85° wet bulb days under current distributions. It shows the impact that replacing each 85° day with a 60–70° day would have on annual productivity per hour. Panel B of Figure 2 shows the same estimates projected on future distributions of temperature under RCP 8.5. Note that the distribution of wet bulb days is precisely the same as in Figure 1; what has changed is that the scale is now interpretable as an effect on productivity, rather than infant mortality. Finally, Panel C of Figure 2 shows the difference in impacts on productivity per hour under the RCP 8.5 versus RCP 2.6 scenario. Again, these figures show that the greatest impacts will occur in the densely populated areas of Uttar Pradesh and Bihar as well as north-eastern India. Infant mortality and labor productivity are only two examples of the wide range of outcomes that could be impacted by temperature. To get a full picture of the distribution of damages that could be caused by FIGURE 2. Labor Productivity Decreases from Extreme Weather

A: Historical Implied Decline in B: Business-as-usual Scenario C: Aggressive Mitigation Scenario Labor Productivity (RCP 8.5) Decline in Labor Productivity (RCP 2.6) Decline in Labor Productivity

Source: Princeton Global Meteorological Forcing Dataset; ISIMIP; LoPalo (2018); and Authors’ calculations. Melissa LoPalo et al. 119 climate change within India, it is also useful to consider the evidence on the impacts of temperature on GDP. Several papers have established correlations between climate and aggregate productivity as well as causal relationships between fluctuations in weather and measures such as GDP. One such paper is Burke et al. (2015), which estimates the impact of average annual (dry bulb) temperature on change in log GDP per capita using a panel of 166 countries from 1960 to 2010. Like other papers in the literature, they make comparisons within countries and difference out country-specific time trends. They find that production per capita is highest at around 13°C and declines sharply not only at higher temperatures but also at colder temperatures. They project these estimates to quantify damages under RCP 8.5. Colder regions such as Europe may see productivity benefits under climate change, but regions that are warmer on average experience large damages. Burke et al. also conducted a country-by-country exercise to examine the implications of climate change for individual contexts. India is one of the most severely affected countries in the world by their estimates. We use their estimates to conduct an additional exercise to visualize the implied distribution of effects within India. In Figure 3, we implement a simplified calculation to show differences in growth rates that might be expected under RCP 8.5 versus 2.6. For this illustration, we assume that local GDP per capita growth is determined only by annual average temperature, as estimated in Burke et al. We then calculate the changes implied by the projected average temperature in 2050 under RCP 8.5 and 2.6, respectively. Finally, we calculate the difference between the two rates, giving an idea of the distribution of impacts on GDP growth under the two emissions scenarios. The results suggest that the majority of India will be negatively affected under RCP 8.5 relative to RCP 2.6, with Madhya Pradesh especially impacted. The white and light gray areas in the map signify regions that will be positively impacted by warming: this occurs for areas with an annual average temperature of less than 13°C. A consistent theme across these results is that India is deeply vulnerable to global warming given the humid climate of South Asia. Within India, climate damages will tend to be greater in the places where the population is already more disadvantaged: we find that Uttar Pradesh, Bihar, Madhya Pradesh, and neighboring states tend to show more vulnerability in the projections presented earlier. Given current inequities, climate damages will not merely reduce the average well-being of the future Indian population; they are also projected to increase inequality by falling disproportionately on the most disadvantaged within India. 120 INDIA POLICY FORUM, 2018

FIGURE 3. GDP Changes from Global Warming

Source: ISIMIP; Burke et al. (2015); and Authors’ calculations.

3. Macroeconomic Projections: How Much Are Climate Damages Worth?

Section 2 documented that many important economic and social indicators are vulnerable to temperature and humidity. However, a critical question remains: How does one weigh these costs in total? How should policymakers aggregate the consequences of climate policy for the full Indian population, including people alive today and people who will not be born for decades to come? To answer this question, we develop an India-centric Integrated Assessment Model (IAM) by modifying a global IAM in wide use in the climate policy literature. We modify William Nordhaus’ RICE model. Because RICE considers only (dry bulb) temperature, not humidity, this section does too. The evidence in Section 2 suggests that these results will, therefore, Melissa LoPalo et al. 121 underestimate India’s climate vulnerability. Despite this, the model projects total Indian climate damages to be extremely large. Quantitatively, the damages under the case of no global GHG reductions are as costly—in a well-being sense—as a hypothetical reduction in GDP per capita of 25–30 percent for each of the next 20 years. An event of this magnitude would be a humanitarian disaster. However, as the model shows, these damages cannot be avoided by a reduction in India’s emissions alone.

3.1 Overview of IAMs IAMs are macroeconomic growth models with a climate component designed to quantify the economic trade-offs associated with carbon emissions. The most widely used IAMs (DICE/RICE, PAGE, and FUND) share the same conceptual structure (Hope 2011; Nordhaus 2017, 2010; Tol 1999). Economic production/consumption generates well-being for the people who consume, but also results in GHG emissions. GHG emissions enter a climate module designed to track the stock of CO2 and the resulting global temperature dynamics. Higher future temperatures then cause harm to future people according to a relationship called the damage function. To measure these trade-offs in a way that assesses the consequences for everyone, we use a standard social welfare function (SWF) that is additive across time. Equation (1) formalizes this.

Z 1 Wc();,ρ LL= ∑ t ttUc() (1) t=0 ()1+ρ

Total social welfare is the sum of utility in each period from today (t = 0)

until some end date (t = Z) generated by per capita consumption, U(ct) 1 multiplied by the population in that time, L , and discounted by , t t ()1+ρ which is a factor that makes future costs and benefits worth less to the social evaluation than nearer term costs and benefits.4

Temperature, Tt, does not enter the SWF directly because the IAM is constructed to deduct climate damages directly from the output available for economic use.

N G (2) YDt =−()1 ()TYtt

4. The utility function is assumed to have diminishing marginal returns, specifically of the c1η constant relative risk aversion form: , where η is the inequality aversion factor. 1η 122 INDIA POLICY FORUM, 2018

N Equation (2) defines net output in each period, Yt , as some fraction of gross G output, Yt . The fraction lost, D(Tt), is the damage function. This functional form implies that some output is either spent in adaptation efforts (and is therefore unavailable for consumption) or is destroyed from high temperatures. The idea of temperature directly destroying output may be difficult to conceptualize, but it approximates two more realistic interpretations: (a) that more inputs are needed for the same level of output (productivity declines) or (b) that more output is necessary to retain the same utility level (agents need to be compensated for the higher temperatures).5 We are interested in the trade-offs relevant for an Indian policymaker, so Equations (1) and (2) include only Indian inputs. For example, (1) is an India-specific SWF with projected Indian population and per capita consumption in each scenario. Climate damages are losses to total Indian consumption from a warmer planet. Costs and benefits for people living outside of India are not counted.

3.2. Social Costs of Emissions in an IAM In building toward aggregate damages, we start with a decomposition of the social cost of an extra ton of GHG emissions. This quantity is known as the social cost of carbon (the SCC). This decomposition has a convenient multiplicative form that allows us to highlight each potential channel for damages to increase or decrease. The most uncertain and contested of these channels are the damage function and the social discount rate. We take extra care in discussing these further. Mathematically, the SCC can be shown to be of the form presented in Equation (3) (Golosov et al. 2014).

∞ 1 ∆Uc()t ∆ct ∆Tt SCC = ∑ t Lt (3) t=0 ()1+ρ ∆ct ∆Tt ∆E0 The complex economic and atmospheric relationships we hope to capture can be simplified conceptually into five multiplicative terms (here we use the y notation for the change (D) in y that results from a one-unit change in x). x 1 i. : Pure social discount rate t ()1+ρ

5. This second interpretation is not exact because some fraction is saved rather than con- sumed, but it is close enough for expositional purposes. Melissa LoPalo et al. 123

ii. Lt: Population in time t Uc  iii. t : Increase in utility that results from an extra unit of per capita ct consumption Dc iv. t : Consumption equivalent losses that result from an increase in DTt temperature at time t (the damage function) DT v. t : Increase in temperature at time t that results from an extra unit DE0 of GHG emissions today

In this paper, population projections are taken from the United Nations World Population Prospects. The Indian damage function that we use (term iv above) was derived by Nordhaus (2010) by scaling up a global damage function to reflect a consensus that India is more vulnerable than a globally averaged damage function would imply. As documented in Nordhaus and Sztorc (2013) (and replicated in Figure 4), the global damage function is

FIGURE 4. Global and Indian Damage Functions

30 Tol Survey Global Damage Function Indian Damage Function IPCC Estimates 20

10 Yearly GDP Loss (%)

02468 Temperature Increase from Pre-industrial Levels (Celsius)

Source: Nordhaus and Sztorc (2013); Tol (2009). 124 INDIA POLICY FORUM, 2018 fit to the meta-analysis of Tol (2009).6 The fitted function is restricted to be quadratic and is calibrated over a range of estimates from 1° to 3° of warming.7 India’s damage function then takes the same functional form but lies above the global function at all points. A challenge present throughout the IAM literature is that it is especially difficult to know how costly climate damages would be beyond 3°C of warming. We continue to follow Nordhaus (2010) by assuming the cali- bration at lower temperatures remains informative at higher temperatures. This results in substantial—yet unavoidable—uncertainty over a range of potential outcomes. Subsequent work suggests this uncertainty is one-sided: the DICE/RICE damage function used here is very likely a lower bound for damages at high levels of warming.8 Specifically, Weitzman (2012) presents a convincing case that the DICE/RICE implied damages are implausibly low for warming greater than 3°. Likewise, Burke et al. (2015) estimate damages using a method less reliant on extrapolation and find a South Asian damage function nearly an order of magnitude larger than what we use here. Nordhaus (2017) himself has even adjusted damages upward in his most recent work.9 Beyond this, no damage function in the IAM literature—including the Nordhaus (2010) specification that we use—considers increases in wet bulb temperature. As documented in Section 2, the importance of humidity makes India more climate-vulnerable (relative to drier developing regions such as sub-Saharan Africa) in a way that has been previously omitted. In order to be grounded in the prior literature, our damage function, too, omits the potentially important role of humidity. Therefore, although the damage function remains a highly uncertain object, we are confident that our striking results are not driven by unrealistically pessimistic assumptions regarding the damages of climate change. Terms (i) and (ii) of the SCC quantify the relative importance of damages faced by further-future people compared with damages faced by nearer- future people. These terms reflect the two justifications for discounting over time: (a) because damages occur in the future, and (b) because damages are suffered by richer populations in the future. Some combination of

6. The damage estimates in Tol (2009) are designed to include the monetary costs of optimal adaptation as well as the costs of lost output/well-being. For example, the costs of sea-level rise include the cost of building dikes and levees where possible (adaptation) and the cost of damaged/lost landmass where not (residual damages). 7. While only calibrated on 1–3°, the damage function sits in the IPCC range of estimates for 4°. 8. See Diaz and Moore (2017) for an extensive review of aggregate IAM damage functions. 9. We use the Nordhaus (2010) version because it is a disaggregated model which allows us to pull India’s damage function directly. Melissa LoPalo et al. 125 these two factors determines how much we ought to value losses to future populations.10 This is important for our analysis because climate damages will unfold over coming centuries. A large body of literature in climate economics has recognized that optimal mitigation policy is substantially shaped by the choice of a discount rate: if the social evaluation assumes that the future does not matter, then it is unsurprising that models recommend unaggressive climate mitigation policy. Understanding the respective roles of these parameters is then critical to understanding our results. To reiterate, term (i) plays a simple role of discounting well-being just because it is experienced at a later date. The way term (ii) influences discounting, however, is less obvious.11 Term (iii) is the marginal utility of an additional unit of consumption. It is an uncontroversial consensus among social scientists that this changes with income: adding $1 to the budget of a poor person increases his/her well-being more than if we did the same for a richer person.12 Throughout this literature, economists use functions in which a single parameter, h, controls the importance of extra money to a poorer person, relative to a richer person. This parameter is known as the “inequality aversion” of the model. Inequality aversion is important for discounting in climate policy if we expect future economic growth: because future Indians will be richer than present-day Indians, future money losses are less important to policymakers than today’s money losses to a poorer population.

3.3. Social Welfare Parameter Choices There is a large body of literature documenting that differences in discount rates drive many of the academic disagreements on climate policy (Broome 2012; Dasgupta 2008; Greaves 2018; Nordhaus 2007; Stern 2006; Weitzman 2007). After careful review of this past work, we have come to agree with the authors who believe that total discounting cannot and should not be inferred from individual economic choices. In our view, r reflects the ethical choice

10. The exact way these come together to determine the total discount factor, d, under a constant rate of economic growth, g, is represented by the well-known Ramsey Equation: d = r + hg 11. Well-being is emphasized because r is a discount on utility, not goods. It may be reasonable (as we discuss in the following paragraph) to discount damages to future people because they will be wealthier, but this has nothing to do with r. 12. Nordhaus (2010) and other regionally disaggregated climate-economy models use a solution technique called “Negishi weights,” which results in a SWF that does not respect this cross-sectionally—$1 to a rich person is as socially valuable as $1 to a poorer person. We interpret Negishi weights as an attempt to solve for the model’s equilibrium, rather than a rejection of cross-sectional diminishing returns. 126 INDIA POLICY FORUM, 2018 of policymakers: are future Indians as important as present-day Indians? In contrast, inequality aversion h is, in principle, empirical: it reflects how human well-being increases with increasing levels of consumption. This parameter is unfortunately impossible to estimate in practice. Therefore, as in essentially every study in the IAM literature, we choose baseline values of r and h, and present robustness checks with other values. We believe the appropriate choice of r is 0.13 The list of authors that agree with this choice is long,14 and it follows from a simple argument that in the SWF, all Indians, regardless of year of birth, matter equally. Suffering is no less bad whether it occurs 50 or 100 years from now merely because one is further away from us in time. The parameter that governs the rate of change of marginal utility, h, stands on less firm grounding. We choose a level to match our prior work in Budolfson et al. (2018). To understand the baseline parameter we choose, h = 2, consider two people, one twice as rich as the other. If the poorer person realizes some consumption gain, our baseline value of h implies that the wealthier person would need to receive four times that gain for it to be as socially good. Zero inequality aversion, in contrast, would imply the richer person would just need the same monetary gain for it to be as socially good, an implication we find implausible. Because any choice is subject to disagreement, we will present robustness checks with additional h values that correspond to the income gains needing to be 2.5 and 5.5 times as large, respectively, rather than the original 4.15

3.4. Quantitative Results We can now quantify aggregate damages to India from climate change using the model and parameters just described. These damages are large, even though they do not include the humidity interactions described in Section 2. We quantify damages from climate change in terms of consumption-equivalent losses to current people: by what percent would per capita

13. In practice, some very small positive number is used to follow Stern (2006) who makes an adjustment for the exogenous risk of extinction. 14. Cowen and Parfit (1992), Stern (2006), Dasgupta (2008), and Broome (2012) are some notable examples. 15. The main objection to our resulting discount factor is that an individual’s saving behavior does not match what would be implied by the discount rate on goods we are using. We are not bothered by this. Even if we believed the SWF should be democratically determined (i.e., correspond with individual preferences), saving decisions reflect how individuals plan to allocate their resources to their own individual futures. Personal impatience is a different consideration from how society values the lives of future generations. Melissa LoPalo et al. 127 consumption need to be reduced for the next 20 years to match the total welfare losses associated with climate change? What reduction in near-term consumption would be just as bad, from the point of view of the SWF, as climate damages will be? Preventing a deep and sustained economic collapse would presumably always be a top policy priority, so this is a useful way to calibrate the policy importance of climate damages. In particular, the consumption loss that would be equivalent to climate damages is calculated as follows:

• Step 1: Exogenously warm the planet and compute India’s total well- being for all future periods under the resulting level of global warming. • Step 2: Re-run this scenario without climate damages and instead reduce per capita consumption for the first 20 years until total well-being from Step 1 is matched. • Repeat Step 1 and Step 2 for various possible global warming scenarios.

Without any further global mitigation policy, the economic collapse necessary to match projected climate damages is a 29 percent reduction in GDP per capita for each of the next 20 years. This would be a catastrophic loss. Figure 5 presents these near-term consumption equivalent damages under the baseline choices of r = 0 and h = 2 for a wide range of potential climate outcomes.

FIGURE 5. Near-Term Consumption Equivalent Losses

Projections under 30 Current Policy Removing All Indian Emissions 20 Near-Term Consumption Loss (%) 10

1 2 3 4 5 6 7 8 Temperature Increase from Pre-industrial Levels (C)

Source: Authors’ calculations. 128 INDIA POLICY FORUM, 2018

As shown in Figure 5, these damages have the potential to be extremely large. The rightmost point labelled on the curve corresponds to the global BAU scenario in DICE: no GHG restrictions are enacted beyond current policy, and mitigation comes only from private sector technological developments.16 Under this outcome, many decades of the Indian population would experience climate damages amounting to about 15 percent of GDP. Perhaps more important than the large level of damages is the slope of this function. At high levels of warming, changes in global temperature cause very large changes in Indian well-being. For instance, the planet is projected to warm by around 3.5° if the national emissions pledges in the Paris Accord are successfully realized (Reilly et al. 2015). Climate damages would be cut by two-thirds despite warming being reduced by less than one half. Global efforts to reduce warming are especially valuable to India in light of this damage convexity. A natural reaction to this quantification of India’s climate vulnerability may be to suggest that India should quickly and unilaterally decarbonize. The RICE model also allows us to assess the consequences of such a policy. For better or worse, over the coming decades, India’s emissions are projected to remain a small fraction of the global, historical stock of GHG emissions. The dot to the left in Figure 5 shows that the peak global temperature would decrease only slightly if India were to altogether eliminate its emissions. As a result, its climate damages would only slightly decrease. The global temperature would probably decrease by even less than shown in the figure because we do not model an endogenous response of other countries: India removing itself from aggregate energy demand would reduce prices and increase other countries’ energy use. The message of the RICE model is clear: India is highly vulnerable to climate damages and cannot eliminate the problem by reducing its own emissions.

3.5. Robustness Given the well-known importance and uncertainty over how to discount future costs, we report the robustness of our results to alternative choices of the inequality aversion h.17 Figure 6 plots how the results change with these higher and lower values of inequality aversion.18

16. This corresponds closely to the RCP 8.5 scenario. 17. As we feel much more confident in our choice of r, we believe this uncertainty is the result of not knowing how fast individual (and social) marginal returns to income diminish. 18. See the paragraph directly preceding this “Results” section for the discussion of h values chosen for sensitivity checks. Melissa LoPalo et al. 129

FIGURE 6. Robustness to Alternative Inequality Aversion Assumptions

η = 1.42 η = 2 80 η = 2.45

Near-Term Consumption Loss (%) 20

2345678 Temperature Increase from Pre-industrial Levels (C)

Source: Authors’ calculations.

Because the model assumes that future Indians will be substantially richer than present-day Indians, changes to h are extremely influential for how bad climate damages are perceived to be.19 Using smaller values of h (1.42 in this case) pushes the damages to very high levels (over 80 percent for 6°C of warming). But if h is large (2.45 here), total damages become notably smaller. In fact, this graph is conceptually bounded between 0 and 100 so these values span nearly the entire set of feasible outcomes. The fact that the results are heavily shaped by the choice of h is consistent with observations in Dasgupta (2008). However, our choice of h is not low relative to practice in the climate-economy literature so we take little comfort in the low damages associated with an unusually high value of h.20 This is especially true given the conservative damage function we use.

19. Assuming otherwise—that India will not experience rapid economic growth—would make climate damages even more important to social welfare because a poorer future population would experience the harm. 20. Although Dasgupta (2008) urges authors to consider larger values for this parameter, the most influential IAM results (Nordhaus 2010; 2017; Stern 2006) all use a value less than 2 (some going as low as 1). Micro evidence supports our choice as well: Carlsson, Daruvala, and Johansson-Stenman (2005) use hypothetical survey questions about the well-being of grandchildren and estimate h to be 2 for intergenerational inequality. Studies directly using governmental behavior in tax policy to infer h in other policymaking spheres find values between 1.3 and 2 (Cowell and Gardiner 1999; Stern 1977). See Dasgupta (2008) and Greaves (2018) for reviews on total social discounting. 130 INDIA POLICY FORUM, 2018

4. Health Co-Benefits

Although the focus of this paper is on climate vulnerability, this section introduces an important near-term vulnerability of the Indian population with impacts for climate policy: air pollution. Reductions in GHG emissions tend to lead to reductions in air pollutants, because both pollutants tend to share common emission sources (e.g. coal-fired power plants [Gupta and Spears 2017]). As a result, reductions in GHG emissions are likely to lead to improvements in current human health through improved air quality. These benefits are often called health “co-benefits” because they are additional benefits that come alongside the direct climate-related benefits of GHG reductions. Emerging research suggests that these health co-benefits may be large, especially for a nation such as India in which air pollution is one of the nation’s leading health problems. For example, according to recent data from the World Health Organization, 14 of the top 20 cities with the highest levels of particulate matter pollution in the world are in India (BBC 2018). Interestingly, these cities are all located in northern India, the same region with the highest level of population, fertility, and climate vulnerability in the country: seven of these 20 globally-most-polluted cities are in Uttar Pradesh and Bihar. Thus, health co-benefits have a critical place within India’s climate policy decision-making and are an additional source of benefits for India from GHG reductions. This is in part because large benefits occur quickly enough to be economically important even with high time discount rates: air pollution is already harming the population alive today (Scovronick et al. 2019). Furthermore—and of particular importance to Indian policymaking—these health co-benefits of GHG reductions can be almost fully captured by a large country such as India through unilateral domestic policymaking, as most co-benefits are realized domestically (in contrast to the fully globally dispersed climate-related benefits of GHG reductions). Co-benefits are also not as vulnerable to being negated by the non-cooperative economic and policy response of other nations (in contrast to climate benefits, which are vulnerable to “emissions leakage,” as discussed further, and can also represent a transfer of GDP from the mitigating nation to other non-cooperative nations). Globally, the benefits from preventing air pollution-related deaths alone may outweigh the mitigation costs of reducing carbon emissions. Shindell et al. (2018) examine the local health impacts of reducing emissions enough in the 21st century to achieve 1.5° warming rather than 2°, finding that the drop in air pollution could prevent around 150 million premature deaths, Melissa LoPalo et al. 131 mostly in Asia and Africa. They estimate the health impacts in individual metropolitan areas, showing that Indian metros such as Kolkata, Delhi, Mumbai, and Lucknow will be among the top beneficiaries in terms of number of avoided deaths. Similarly, Markandya et al. (2018) find that in some mitigation strategies, co-benefits of carbon emission reductions were almost double the costs in some areas, implying that mitigating enough to achieve 1.5° warming would have a net benefit for India, as well as China. Scovronick et al. (2019) find that optimal global mitigation results in immediate net benefits when climate costs, climate benefits, and co-benefits and co-costs are all jointly considered. A large literature, recently surveyed by Greenstone et al. (2017), highlights the large costs of air pollution to the health of Indians and people in other emerging nations. Among the reasons for these costs, burning coal may be especially important. For example, Gupta and Spears (2017) estimate the impact of coal plants in India on the health of people living in the same district by studying districts where a new coal plant opened between the 2005 and 2012 waves of the IHDS. Because the survey visited the same households at the beginning and end of the seven-year interval, Gupta and Spears can show that reported respiratory health worsened over time in the districts that acquired a coal plant, relative to the districts that did not. Tellingly, the result is respiratory specific: diarrhea and fever were unaffected. Moreover, other types of new power plants—such as solar or hydroelectric—are not associated with worsening health, which rules out that the result is spuriously due to electrification or economic activity. One reason that air pollution is so harmful is that the impacts extend to essentially everybody and are almost impossible to escape. In a recent South Delhi winter, Vyas et al. (2016) conducted an experiment regarding potential avoidance of these harms in an upper middle-class flat in Green Park. They used air quality monitors to test the effectiveness of commercially available air filters.21 Under ideal conditions—never opening room doors, even to the interior of the house—the filters made a difference, but much pollution remained. Under a reasonably normal schedule of opening doors, much of what the filters achieved was erased. Part of the problem—reflected in the fact that indoor air quality remained highly correlated with outdoor air quality—is that even upper middle-class flats in privileged neighborhoods often do not have window frames and door frames that prevent air from circulating. Perhaps unlike other contexts, such as drinking water pollution, even rich Indians have little scope for buying their way out of air pollution.

21. These included both a relatively affordable filter and an expensive one. 132 INDIA POLICY FORUM, 2018

In recent research, Scovronick et al. (2019) modify the same RICE model that we used in Section 3 to incorporate an air pollution module. Their objective is to optimize mitigation policy while considering both climate damages and the near-term harm to health from air pollution. The optimal policy balances countervailing forces: air pollution can be cooling, as particles reflect sunlight away from the earth. They find that the health co-benefits dominate and recommend more rapid climate mitigation than if air pollution were ignored. Indeed, once health benefits are co-considered, it may be globally economically optimal to limit temperature rise to approximately 2°C. This finding is especially relevant for India, where severe health costs of pollution are the inverse of considerable health co-benefits. Their result suggests that health co-benefits could even make aggressive mitigation policy rational for India on its own.

5. Conclusion: India’s Best Policy Response to Climate Injustice

Our quantifications show that India is highly vulnerable to climate damage. Our baseline macroeconomic approach suggests that climate change peaking at 5°C, rather than 3°C, would be as detrimental to Indian well-being as a reduction in GDP by 17.5 percent for each year from 2020 to 2040. Our microeconomic results suggest that even this may be an underestimate because it ignores the humidity of South Asia. Clearly, such a threat to near-term economic outcomes would be an overriding policy priority if political leaders anticipated it. If so, India’s climate vulnerability should be a top priority too. What is India’s best response to these facts? As we have argued elsewhere, the Intended Nationally Determined Contributions that richer pollut- ers (such as the USA and the EU) have submitted under the Paris Agreement are inadequate, inequitable, and unjust (Budolfson et al. 2019). We believe that the richer countries should substantially reduce their emissions—quickly and without receiving anything in return—and should substantially fund the climate mitigation and adaptation of poorer countries. But what should India do if they do not, as will presumably be the case? There is no easy answer to this question. Faced with the dilemmas of international cooperation, some analysts suggest that India should “do it alone”: either unilaterally eliminate/reduce its GHG emissions, or, oppo- sitely, pollute as much as necessary to get rich enough to reduce its vulnerability to climate damages. But India cannot do it alone and reduce emissions enough to escape. One reason is limits to state capacity of the sort that many developing countries Melissa LoPalo et al. 133 face. As Greenstone et al. (2017) summarized in their seminal NCAER India Policy Forum paper:

A necessary requirement for command-and-control regulation to work is a very well- informed regulator with the willingness and ability to systematically enforce fair penal- ties in cases of non-compliance. In the main, this has been lacking in India. Duflo et al. (2013) show how reliable data can be an elusive goal, and Ghosh (2015) identifies severe weaknesses in the enforcement mechanism.

Coffey and Spears (2017) make similar observations about a high-profile rural sanitation program: behavior change is difficult to promote; the small personnel per capita size of the Indian state limits capacity; and official statistics can be unreliable even on matters that are routinely measured by straightforward demographic surveys. Muralidharan’s (2016) NCAER India Policy Forum paper on public employment touches on some root causes and potential solutions to these issues of personnel and capacity. However, developing and promulgating sophisticated and detailed guidelines for the optimal regulation of emissions might, in this context, waste valuable time while having little impact. The larger reason that India’s emissions reductions would be inadequate is that there simply are not enough of them to tip the scales: as we computed in Section 3, even if India hypothetically fully eliminated its emissions while the rest of the world did nothing, it would still face almost as many degrees of warming. Worse still, it is unlikely that the rest of the world would be unaffected by India’s unilateral decarbonization. Instead, India, removing itself from global aggregate demand for fossil fuels, might end up lowering the price of carbon, so that some of India’s emissions reductions could be offset by increases in other countries, often called “emissions leakage.” Nor can India do it alone and escape through unrestrained GHG emissions to accelerate development. That is because the numbers do not realistically add up. Emissions are valuable, but they are not valuable enough to promote the economic growth necessary to enable India to escape via this strategy. Therefore, India’s best response to climate injustice may be first and foremost foreign policy, as well as domestic economic and health policy. The reason the question of what India should do is so challenging is that it depends on India’s power to influence other countries’ emissions. It’s worth noting, however, that the fact that India cannot unilaterally mitigate its vulnerability to climate change does not imply that it would not be individually rational for India to dramatically reduce emissions. As we discussed in Section 4, recent evidence suggests that the current health burden of air pollution, which is particularly heavy in India, justifies significant 134 INDIA POLICY FORUM, 2018 mitigation of emissions independent of the climate benefits. While the concern of emissions leakage applies to this strategy, if India mitigated emissions to a level that would be optimal only considering health co-benefits, the leadership that India would be taking in reducing its own deep vulnerability to damages from air pollution may give it more leverage to convince other international players to take action, putting the world on a path toward reduced warming. Another possibility—suggested by the large size of India’s climate damages—is that India may have the option of achieving its climate policy goals via strategic international interactions that accept a creative concession in other sectors of policymaking in order to achieve reductions in the emissions of richer countries. We make no suggestions about what sort of non-climate concession (perhaps even a non-economic, symbolic concession) would be effective to offer; we merely note that India’s climate vulnerability unfortunately suggests that a Pareto improvement could perhaps be found in the right packaging of a non-emissions concession from India, combined with large emissions sacrifices from rich countries. How might such a package be invented? Perhaps one desirable feature is to engineer such a package to have time consistency between the concessions India makes and the emissions reductions that developed nations make with antecedently agreed mechanisms for monitoring and adjustment in light of each side’s subsequent compliance. For example, one can imagine trade concessions from India in exchange for deep emissions reductions, where the continuation of those concessions is contingent on reciprocal compliance. Or, perhaps the right package involves a concession in symbolic diplomacy, security policy, or another dimension of international politics—with the concession explicitly linked to and contingent on emissions reductions from China, USA, the EU, and perhaps others. Or perhaps a different package altogether is the best—the point is merely to illustrate that opportunities may exist for multilateral agreements between India and other nations that have desirable properties. Inventing the right concession to offer would be only one challenge. Such a strategic concession would only make sense if high-emissions developed countries are sufficiently rational actors in international politics that they could be bargained with; perhaps they are not. The success of such a scheme would require international monitoring of rich country agreements, so India can be sure it is getting what it bargained for. Efforts to create such monitoring standards should therefore be fully embraced by India. Even in the absence of an agreement between India and high-emission countries, it is to India’s benefit that these data be transparently and consistently collected: its Melissa LoPalo et al. 135 vulnerability and low emissions per capita result in it having much to gain and little to lose. Calls for credibility in GHG accounting may constitute a new reason that it would be in the interests of the Indian state to contribute to a norm of accurate official statistics. It would be a moral tragedy if India must make such a strategic concession to protect Indians from the unjust emissions of rich nations. But climate change involves moral tragedies. If either (or both) strategic concessions or immediate health-improving emissions reductions are possible and required to slow global GHG emissions, it would be a mistake for India not to do at least what is in the interest of present and future Indians to protect them from the grave threat posed by unbridled climate change.

References

Barreca, A. 2012. “Climate Change, Humidity, and Mortality in the United States,” Journal of Environmental Economics and Management, 63(1): 19–34. Barreca, A., K. Clay, O. Deschenes, M. Greenstone, and J.S. Shapiro. 2016. “Adapting to Climate Change: The Remarkable Decline in the U.S. Temperature- Mortality Relationship over the Twentieth Century,” Journal of Political Economy, 124(1): 105–159. BBC. 2018. “India Cities Dominate World Air Pollution List,” Available at https:// www.bbc.com/news/world-asia-india-43972155 (accessed on June 8, 2018). Broome, J. 2012. Climate Matters: Ethics in a Warming World. New York: W.W. Norton & Company. Budolfson, M., F. Dennig, M. Fleurbaey, N. Scovronick, A. Siebert, D. Spears, and F. Wagner. 2018. “Optimal Climate Policy and the Future of World Economic Development,” World Bank Economic Review, 33(1): 21–40. Budolfson, M., F. Dennig, K. Kuruc, D. Spears, and N. Dubash. 2019. “Minimal Utilitarian Benchmarks for NDCs Provide Equity, Climate and Development Advantages,” Working Paper, May. Burke, M., et al. 2016. “Opportunities for Advances in Climate Change Economics,” Science, 352(6283): 292–293. Burke, M., S.M. Hsiang, and E. Miguel. 2015. “Global Non-Linear Effect of Temperature on Economic Production,” Nature, 527(7577): 235–239. Carlsson, F., D. Daruvala, and O. Johansson-Stenman. 2005. “Are People Inequality- Averse, or Just Risk-Averse?” Economica, 72(287): 375–396. Coffey, D. and D. Spears. 2017. Where India Goes: Abandoned Toilets, Stunted Development and the Costs of Caste. Noida: HarperCollins. Cowell, F. and K. Gardiner. 1999. Welfare Weights. London: Office of Fair Trading. Cowen, T. and D. Parfit. 1992. “Against the Social Discount Rate,” in P. Laslett and J. Fishkin (eds.), Justice between Age Groups and Generations, pp. 144–161. 136 INDIA POLICY FORUM, 2018

Dasgupta, P. 2008. “Discounting Climate Change,” Journal of Risk and Uncertainty, 37(2–3): 141–169. Dell, M., B.F. Jones, and B.A. Olken. 2014. “What Do We Learn from the Weather? The New Climate-Economy Literature,” Journal of Economic Literature, 52(3): 740–798. Dennig, F., M.B. Budolfson, M. Fleurbaey, A. Siebert, and R.H. Socolow. 2015. “Inequality, Climate Impacts on the Future Poor, and Carbon Prices,” Proceedings of the National Academy of Sciences, 112(52): 15827–15832. Deschenes, O., M. Greenstone, and J. Guryan. 2009. “Climate Change and BIRTH WEIGHT,” American Economic Review, 99(2): 211–217. Diaz, D., and F. Moore. 2017. “Quantifying the Economic Risks of Climate Change,” Nature Climate Change, 7(11): 774–782. Duflo, E., M. Greenstone, R. Pande, and N. Ryan. 2013. “Truth-Telling by Third- Party Auditors and the Response of Polluting Firms: Experimental Evidence from India,” The Quarterly Journal of Economics, 128(4): 1499–1545. Edenhofer O., et al. 2014. “Technical Summary,” in O. Edenhofer, et al. (Eds.), Climate Change 2014: Mitigation of Climate Change. Contribution of Working Group III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge, UK and New York, USA: Cambridge University Press. Field, C., V. Barros, K. Mach, and M. Mastrandrea. 2014. “Technical Summary,” in P. Aldunce, J. P. Ometto, N. Rajolijao, and K. Yasuhara (Eds.), Climate Change 2014: Impacts, Adaptation, and Vulnerability. Contribution of Working Group 2 to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge, UK and New York, USA: Cambridge University Press. Geruso, M. and D. Spears. 2018. “Heat, Humidity and Infant Mortality in the Developing World,” NBER Working Paper #24870, Cambridge, MA: National Bureau of Economic Research. Ghosh, Shibani. 2015. “Reforming the Liability Regime for Air Pollution in India,” Environmental Law & Practice Review, 4: 125–146. Golosov, M., J. Hassler, P. Krusell, and A. Tsyvinski. 2014. “Optimal Taxes on Fossil Fuel in General Equilibrium,” Econometrica, 82(1): 41–88. Greaves, H. 2018. “Discounting for Public Policy: A Survey,” Economics & Philosophy, 33(3): 391–439. Greenstone, M., S. Harish, R. Pande, and A. Sudarshan. 2017. “The Solvable Challenge of Air Pollution in India,” India Policy Forum, 14: 1–40. New Delhi: National Council of Applied Economic Research. Greenstone, M. and B.K. Jack. 2015. “Envirodevonomics: A Research Agenda for an Emerging Field,” Journal of Economic Literature, 53(1): 5–42. Gupta, A. and D. Spears. 2017. “Health Externalities of India’s Expansion of Coal Plants: Evidence from a National Panel of 40,000 Households,” Journal of Environmental Economics and Management, 86: 262–276. Hallegatte, S., M. Bangalore, L. Bonzanigo, M. Fay, T. Kane, U. Narloch, J. Rozenberg, D. Treguer, and A. Vogt-Schlib. 2016. Shockwaves: Managing the Impacts of Climate Change on Poverty. Washington, DC: World Bank. Melissa LoPalo et al. 137

Hijioka, Y., E. Lin, J.J. Pereira, R.T. Corlett, X. Cui, G.E. Insarov, R.D. Lasco, E. Lindgren, and A. Surjan. 2014. “Asia,” in V. R. Barros, et al. (Eds.), Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part B: Regional Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, pp. 1327–1370. Cambridge, UK and New York, USA: Cambridge University Press. Hsiang, S. M. 2010. “Temperatures and Cyclones Strongly Associated with Economic Production in the Caribbean and Central America,” Proceedings of the National Academy of Sciences, 107(35): 15367–15372. Hsiang, S. M., M. Burke, and E. Miguel. 2013. “Quantifying the Influence of Climate on Human Conflict,” Science, 341(6151): 1235367. Hope, C. 2011. “The PAGE09 Integrated Assessment Model: A Technical Description,” Cambridge Judge Business School Working Paper, 4(11), Cambridge: University of Cambridge. IPCC. 2007. Climate Change 2007: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge: Cambridge University Press. LoPalo, M. 2018. “Temperature, Worker Productivity, and Adaptation: Evidence from Survey Data Production,” Working Paper. Markandya, A., J. Sampedro, S.J. Smith, R. Van Dingenen, C. Pizarro-Irizar, I. Arto, and M. Gonzáles-Eguino. 2018. “Health Co-Benefits from Air Pollution and Mitigation Costs of the Paris Agreement: A Modeling Study,” Lancet Planet Health, 2: e126–133. Muralidharan, K. 2016. “A New Approach to Public Sector Hiring in India for Improved Service Delivery,” India Policy Forum, 12: 187–225, New Delhi: National Council of Applied Economic Research. Nordhaus, W.D. 1992. “An Optimal Transition Path for Controlling Greenhouse Gases,” Science, 258(5086): 1315–1319. ———. 2007. “A Review of the Stern Review on the Economics of Climate Change,” Journal of Economic Literature, 45(3): 686–702. ———. 2010. “Economic Aspects of Global Warming in a Post-Copenhagen Environment,” Proceedings of the National Academy of Sciences, 107(26): 11721–11726. ———. 2017. “Revisiting the Social Cost of Carbon,” Proceedings of the National Academy of Sciences, 114(7): 1518–1523. Nordhaus, W.D. and J. Boyer. 2000. Warming the World: Economic Models of Global Warming. Cambridge, MA, and London: MIT Press. Nordhaus, W. and P. Sztorc. 2013. “DICE 2013R: Introduction and User’s Manual,” Haven, CT: Yale University. Olsson, L., M. Opondo, P. Tschakert, A. Agrawal, S.H. Eriksen, S. Ma, L.N. Perch, and S.A. Zakieldeen. 2014. “Livelihoods and Poverty,” In C.B. Field, et al. (eds.), Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. Contribution of Working Group II to the 138 INDIA POLICY FORUM, 2018

Fifth Assessment Report of the Intergovernmental Panel on Climate Change, pp. 793–832. Cambridge, UK and New York, USA: Cambridge University Press. Reilly, J., S. Paltsev, E. Monier, H. Chen, A. Sokolov, J. Huang, Q. Ejaz, J. Scott, J. Morris, and A. Schlosser. 2015. “Energy and Climate Outlook,” MIT Joint Program on the Science and Policy of Global Change, Cambridge, MA. Scovronick, N., M.B. Budolfson, F. Dennig, F. Errickson, M. Fleurbaey, W. Peng, R. Socolow, D. Spears, F. Wagner. 2019. “The Impact of Human Health Co-Benefits on Evaluations of Global Climate Policy,” Nature Communications, 10(1): 2095. Sherwood, S.C. and M. Huber. 2010. “An Adaptability Limit to Climate Change Due to Heat Stress,” Proceedings of the National Academy of Sciences, 107(21): 9552–9555. Shindell, D., G. Faluvegi, K. Seltzer, and C. Shindell. 2018. “Quantified, Localized Health Benefits of Accelerated Carbon Dioxide Emissions Reductions,” Nature Climate Change, 8(4): 291–295. Solomon, S., et al. 2007. “Technical Summary,” in S. Solomon, D. Qin, M. Manning, M. Marquis, K.B. Averyt, M. Tignor, H.L. Miller, and Z. Chen (eds.), Climate Change 2007: The Physical Science Basis. Contribution of Working Group 1 to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, pp. 19–91. Cambridge, UK and New York, USA: Cambridge University Press. Stern, N. 1977. “Welfare Weights and the Elasticity of the Marginal Valuation of Income,” in M. Artis and R. Nobay (eds.), Proceedings of the Aute Edinburgh Meeting of April 1976. Basil Blackwell. Stern, N. 2006. Stern Review Report on the Economics of Climate Change. London: HM Treasury. Tol, R.S. 1999. “The Marginal Costs of Greenhouse Gas Emissions,” The Energy Journal, 20(1): 61–81. ———. 2009. “The Economic Effects of Climate Change,” Journal of Economic Perspectives, 23(2): 29–51. UNFCCC. 2015. “Adoption of the Paris Accord,” Available at https://unfccc.int/ resource/docs/2015/cop21/eng/l09r01.pdf (accessed on June 18, 2018). Vyas, S., N. Srivastav, and D. Spears. 2016. “An Experiment with Air Purifiers in Delhi during Winter 2015–2016,” PLoS ONE, 11(12): e0167999. Weitzman, M. L. 2007. “A Review of the Stern Review on the Economics of Climate Change,” Journal of Economic Literature, 45(3): 703–724. ———. 2012. “GHG Targets as Insurance against Catastrophic Climate Damages,” Journal of Public Economic Theory, 14(2): 221–244. Comments and Discussion*

Navroz K. Dubash Centre for Policy Research

I really enjoyed reading this paper. I am going to try to do three things. I am going to leave the econometrics to Professor Shreekant Gupta, but I am going to try and talk first about the policy context for the paper. What policy conversations are these authors trying to engage in? Then, second, how robust are the conclusions, and third, should we buy the conclusions? Why this paper? I read the paper in the context in which it is hard to get people in India to focus on climate change. So I have somewhat provoca- tively called climate change in India as a policy and political backwater. If you are a part of policy conversations in India and you talk about climate change, then the few stock reactions all lead to the conclusion that India shouldn’t really pay a lot of attention to climate change. One stock reaction is that we have a lot of other immediate issues that with guaranteed certainty are doing more harm. Just to pick an issue at random, there is sanitation, where Dean Spears and Diane Coffey (2017) have made a very compelling case, putting numbers and qualitative information about why we should worry about sanitation right now. Another stock reaction is the climate justice story, which says that since the North caused this problem, why should we focus our policy attention on this? Then there is the argument that growth allows better adaptation, and since mitigation would only constrain growth, mitigation therefore works against our interests, even on climate change. So we have this strange situation in which India is one of the countries most vulnerable to climate change and globally we have this discussion about an existential crisis. But within India, it is extremely hard to gain much political attention for climate change other than episodically between November 5 and December 5 when the annual Conference of the Parties (COP) to the 1992

United Nations Framework Convention on Climate Change happens in one of the global capitals of the world. So I see this paper as being part of a larger conversation about how we can steer climate change out of this kind of backwater. I think Kevin Kuruc put it very bluntly. If we can put up a big GDP number up there that gets people’s attention, or if we can hitch this wagon to an engine that has more momentum and can bring in issues and communities that already command attention—infant mortality, labor productivity, air quality, using the examples from this paper—then we can get somewhere. So I see this paper in the context of how to get attention to this issue. So what are the answers provided by the paper? The paper argues that climate change-related temperature changes can increase infant mortality, decrease labor productivity, and lower GDP per capita. The big, headline number is 29 percent GDP per capita decline for 20 years. Besides avoiding these losses, the air quality benefits of dealing with climate change are considerable. These are all compelling arguments. Then the question is: Do they hold up? Before actually getting to this, let me just say that if this is indeed a correct reading of the paper; it is a view I am very sympathetic to. My questioning of whether the paper has achieved this goal is by way of strengthening its arguments, because I think this is a worthy objective to achieve. So here are some questions that I have. On the microeconomic evidence, the question is whether data from within-place variations across time capture the kind of climatic discontinuities that we expect to see from climate change? Melissa LoPalo in her presentation said that, in fact, they can’t. I think she was right to put that caveat, but I would note that this actually strengthens their case that this is only a floor level of impact. So it is a useful set of arguments. The authors refer to a paper by Mani et al. (2018), which looks at the effects of temperature and precipitation changes on living standards in South Asia. The paper has lots of informative, pretty pictures down to the district level, and is very vivid and visual. IPCC documents too are full of these complicated headline graphs, and a lot of time and attention is given to such visuals. So I would encourage the authors also to visualize their data and locate the results in the larger deliberative, policy conversations around climate change in India. Getting to the GDP headline number, which is the part that attracted me the most, because if we can stand by this figure of a 29 percent per capita loss, it really cracks open the conversation. However, there are a couple of things that I would worry about. One of these is the damage functions, and Melissa LoPalo et al. 141 though Kevin talked us through the complexities, some people think they are a bit dodgy. It is hard to figure out whether the paper has drawn those curves right, especially if they are for a country rather than for the world. But I think the authors are right: and they are well within the bounds of the literature. The part that I am more worried about is the use of the term “current policy” in the discussion of their quantitative results and their chart on consumption losses. I think the paper may do itself a disservice and border on the sensational by putting out the 6o warming number, which I think is drawn from the RCP 8.5. The Paris pledges that are already on the table and which countries have committed to are estimated to get us to somewhere between 2.7o and 3.3o global warming. So even if we give or take a few degrees, we are still not near 6o. Admittedly, those are pledges, they are not policy yet, but there is a lot of momentum, and many of these pledges are already locked in. So I think at this stage to be talking about 6o warming is probably a little misleading. The numbers shift hugely as Kevin pointed out; 3o warming gives you about a 7 percent loss level and 2o gives about 5 percent. I thought it is actually interesting that when we go from 3o to 2o, which is a huge change and a huge effort in policy terms, you don’t see much reduction in GDP losses. I am curious if the authors have an explanation for this. The other point is about inequality aversion, which is one of the assumptions that the authors make. It is a complicated literature. The inequality aversion, as I understand it, is the relative utility a rich person gets as opposed to a poor person from an extra ton of carbon. As the graph that Kevin showed suggests, the numbers vary hugely with the assumption about inequality aversion. So the higher value that they use, I think 2.45 or therea- bouts, lowers the GDP per capita loss at 6o to substantially less than the 29 percent, which is a world of difference in capturing policymakers’ attention. The plea to the authors, as they revise this paper, is that the headline GDP number is a wonderful thing to aim for and could rescue climate change from the backwaters, but only if the range can be reduced and made more credible, and some of these concerns are taken care of. So should we buy the conclusions? The key conclusion that I am taking away from the paper is that if even India were to go to zero emissions tomorrow, it would not make much difference. That is an important conclusion that is consistent with the other cited papers. But the forward-looking conclusion here is: India’s climate damage justifies what they call creative concessions to richer countries to induce enhanced actions, and these are in India’s interests. I have a few problems with this argument. From an Indian point of view, the reality is that such an approach would bring narrative dissonance, since the dominant narrative in India has been about climate 142 INDIA POLICY FORUM, 2018 justice, and how we did not cause the problem. Even if this were rational from a policy point of view, it is very hard to imagine that we would be able to act on such an advice because of the dissonance with the dominant narrative in India. Further, I am not sure that this approach would work in rich countries. For rich countries to be interested in this conversation, India’s concession would have to be larger than the perceived competitiveness loss from any mitigation action. So this is not really a two-way relationship. It is a relationship that is driven, at least in the politics of it, by perceived issues around competitiveness. I do accept that what has changed in climate politics a little is a decrease in the extent to which people perceive competitiveness to be threatened by mitigation. The final observation is that rich countries are also almost certainly absolute losers from climate change. But, is what drives a country’s policy response its loss relative to others? The argument would be: I am going to lose a bit; India is going to lose more; therefore, I should expect payment from India. So is it an absolute or a relative loss that drives how countries think about compensation for climate action? For these reasons, I find the conclusion about creative concessions not the most persuasive implication of the empirical material and analysis that the authors have presented. Perhaps there should be an alternative story that focuses more on co-benefits. I think the climate debate has shifted. It is not about international negotiations, except as an ex-post stamp of approval of national actions and a way of amplifying and ratcheting those actions over time. What matters is what drives national politics. And what drives national politics in India on climate change is likely to be the potential, not always guaranteed, for some of these co-benefit actions in areas like air pollution. This argument is worth pushing. Co-benefits are presumably what has driven the renewable energy story: since renewable energy has become cost competitive, it is consistent with India’s energy security interests and is being promoted. It is not really a climate-driven story. Asking India to provide these creative concessions goes well beyond the much lower risk in just playing nicely at the international negotiations on some issues. I do think the narrative has changed in India, but historically we have held back on some things that I think we shouldn’t have been holding back on, like more rigorous transparency mechanisms for all countries and seeking a more robust technical expert review process. In the past, we have been very hesitant about legally binding obligations. These are likely to bind the North just as much as, if not more than, India, because India is actually well down the path of things like the renewable energy transition. Melissa LoPalo et al. 143

So framing the paper and its conclusions around issues related to India’s role in spurring collective action might actually be a bit truer to the analysis and messages that the paper very nicely lays out.

References

Coffey, Diane and Dean Spears. 2017. Where India Goes: Abandoned Toilets, Stunted Development and the Costs of Caste. Noida: HarperCollins Publishers. Mani, Muthukumara, Sushenjit Bandyopadhyay, Shun Chonabayashi, Anil Markandya, and Thomas Mosier. 2018. “South Asia’s Hotspots: Impacts of Temperature and Precipitation Changes on Living Standards,” South Asia Development Matters, Washington, DC: World Bank.

Shreekant Gupta Delhi School of Economics

The objective of the paper is to quantify damages from climate change for India and to examine the implications of this for greenhouse gas (GHG) abatement, that is, climate mitigation by India. The problem is that the two issues are pretty much unconnected—reducing GHG emissions by India in itself will do little to reduce the damage it will face due to climate change. The reason is obvious—climate change is a global externality and will not be affected much by abatement by one country alone. It is a global problem that requires global collective action. India, of course, can and should do its fair share by promoting renewables and cutting back on use of fossil fuels, which has co-benefits in reducing local air pollution and reducing dependence on imports (especially oil). But most importantly (as I argue later), India has to focus on policies and measures to adapt to climate change. Unfortunately, the paper has little to say on co-benefits and even less on adaptation. The bulk of it is devoted to discuss climate damages for India. Health co-benefits are talked about briefly (two pages in the 20-page draft version), and adaptation not at all. To be fair, the paper acknowledges that reducing India’s GHG emissions to zero will do little to limit global warming (Figure 5 in the Conference version) and states that “India is highly vulnerable to climate damages and cannot eliminate the problem by reducing its own emissions.” But in that case, the impetus for climate mitigation has to come from co-benefits, and adaptation has to be central to climate policy. It would have been nice if these two issues had been discussed in greater detail. 144 INDIA POLICY FORUM, 2018

On the structure of the paper, it is trying to say three things. First, a hotter and more humid world is going to be bad for India—no surprise there! Second, there are co-benefits of mitigating GHGs, that is, we will have cleaner air and better health, etc. Third, the paper has a normative discussion on what India should do vis-à-vis climate change. The first point, that is, a hotter and more humid world will be bad for India, is made in two ways. Since these are disjointed, let me call them 1(a) and 1(b), and discuss them separately. 1(a) is merely a review of microeconometric studies that argue that a hotter climate is bad—actually they have this tweak that it is not just hot that matters, it is hot and humid that matters. This according to them is a kind of a new idea (hardly so!). They cite studies that show that in a hotter and more humid world, there will be higher infant mortality. Another point they make from citing existing microeconometric studies is that a hotter and more humid climate will affect labor productivity. These things are well known. No surprises here. All of this is secondary literature, and the paper by Geruso and Spears (2018) discussed in detail is not even for India. If I have understood the paper correctly, there are no regressions or any meta- analysis in the paper, which simply is saying, look, from the microeconometric evidence, this hotter and humid (which is made a big deal of) is going to be bad for India. The results from these are married with projections on what the climate is going to be like. It is only in 1(b) which is totally disjointed from 1(a), where there is new material, namely, results from an integrated assessment model (IAM). There is little else in the paper that is new. In the co-benefits discussion as well, it is basically secondary literature that is cited and that too somewhat perfunctorily. In 1(b), the paper is taking Nordhaus’ RICE model, which is a multi- region IAM in which the world comprises 12 regions, of which India is one. In this model, you are saying, “let us look at the world in two ways.” The first is a world with no climate change. Then you are looking at a world where there is climate change, but you are not doing anything about it (business as usual). Obviously, a world with no climate change will have higher welfare as compared to a world where there is climate change and you have business as usual. In IAMs, “welfare” is defined as the sum of discounted utility of aggregate consumption (call it GDP) over time. Since RICE has 12 regions, there are 12 welfare functions/levels, one for each region. The focus here is only on welfare levels for India. The difference between the two levels of welfare (with and without climate change) quantifies climate damage for India. What the paper does is to set this amount equal to the loss in welfare if there had been no climate change but instead consumption (GDP) had collapsed during 2020–40. In other words, what the paper is doing is asking, Melissa LoPalo et al. 145 in a world with no climate change, by how much do we have to force consumption down so that we get the same welfare loss as if climate change had happened? This is confusing since in these models, the consumption path is generated endogenously. So presumably the authors have generated a consumption path without climate change and then manually reduced consumption levels for 2020–40 to get the same loss in welfare as in a world without climate change. It’s not clear. By the way, the paper’s discussion of the social cost of carbon (SCC) in this section is gratuitous. The consumption paths and welfare losses are generated by the model and have little to do with calculating SCC. Also, the paper says not using humidity and only looking at temperature in RICE will understate damages—not really—in a highly aggregate economy wide model. Extrapolating from microeconometric studies on humidity to such a model isn’t very meaningful. But more importantly, the paper is saying, if you don’t do anything about climate change, it is going to be a 29 percent drop in per capita consumption/ GDP, etc. Why that is an interesting question is because the impact of climate change is going to be much farther out in the future. This is the point that I would have started my comments with. As Navroz said, the paper is trying to quantify the impacts of climate change so that Indian policymakers get scared and do something about it. Unfortunately, that is incredibly naïve because it takes a lot to scare Indian policymakers. I could be saying the same thing about air pollution: that it is very bad, and people are dropping dead, so do something about it. But I don’t think it takes us anywhere: Indian policymakers don’t get scared, and we live with these issues. There is nothing in this paper about adaptation, which I find puzzling, because as the paper says, if India were to reduce its emissions to zero, it is not going to make any difference to the global temperature trajectory. But from there, the paper makes the puzzling leap of suggesting that even if it is not going to make any difference, India should mitigate. In doing so, the paper misses out on two opportunities. The first is that this IAM has an adaptation version called AD-DICE, which modifies a DICE model to build in adaptation. I don’t see any analysis, particularly for India, being meaningful until we build adaptation into an IAM. The second opportunity the paper has missed is that after talking briefly about co-benefits, that is, reducing GHG emissions will give co-benefits in terms of lower air pollution, this is not reflected in the conclusions, which simply says that India should mitigate. Let me just cite the way the Conference version of the paper ends. It says

It would be a moral tragedy if India has to make such a strategic concession to protect Indians from the unjust emissions of rich nations. But climate change involves moral 146 INDIA POLICY FORUM, 2018

tragedies. If such strategic concession or other action is required and possible, it would be a mistake for India not to do at least what is in the interest of present and future Indians to protect them from the grave threat posed by unbridled climate change.

Now, if Indian mitigation, bringing our emissions down to zero, is not going to do anything for the global temperature trajectory, then where is this coming from? If it is coming from the co-benefits of mitigation, then that should be formally modeled into the IAM. We simply can’t assert that. So basically what I find in this paper are four disjointed pieces—the first, which is microeconometric evidence, and the second which is an IAM, both of which tell us that climate change will be really bad for India. The third part is that there are co-benefits. And the last part is that India should mitigate. Let me say a little bit more about adaptation. As India gets richer, there is no reason why it shouldn’t adapt. Lee Kuan Yew once said that the greatest invention humankind ever made was the air conditioner. India is a humid country so I don’t see any reason why we should ignore such adaptation. The paper cites literature that found that air conditioners moderated the mortality effects of high temperature in 20th-century USA. It is a different matter that the unavailability of data on air conditioner ownership may hinder the testing for this. But I did go through the paper carefully and saw that there really was no discussion of adaptation or any attempt to model it. I would argue on the basis of theoretically rigorous work that India’s marginal dollar should be spent on adaptation, and not on mitigation. Perhaps this can be shown through the AD-DICE model. Or it can be shown by building the co-benefits story, so that mitigation would make sense when it is in our own interest in terms of improving local air quality. Otherwise, what we should be doing with the money is climate-proofing agriculture, or doing things on our coast to cope with climate change.

General Discussion

Jeffrey Hammer started the discussion by seeking a clarification on co- benefits. He wanted to know how climate change, or CO2, and the local climate, or PM 2.5, are correlated empirically. Indira Rajaraman asked if the infant mortality effect figures, which were estimated based on the Geruso and Spears paper, were for India alone or, as it appeared, for all countries in the DHS. Obstetricians do not have a uniform workload across the year and their peak load varies across different parts of the country. Parents are responding endogenously to climate change over Melissa LoPalo et al. 147 the year, avoiding giving birth on hot days that may kill newborn babies. This adaptive response of parents needs to be factored in. Since the authors have data on the month of birth in every place, this could help them assess the extent to which this response varies between wet-bulb incidence areas such as Uttar Pradesh and Bihar and other parts of the country. Mihir Desai’s first question was about the absence of sea level or coastal data in the paper, as sea-level consequences would differ by country, and would thus have different consequences about where populations reside. The second question was about migration, and why relocation was not being subsidized to deal with the localized effects of climate change, especially in states such as Uttar Pradesh and Bihar. In such cases, migration could be the most obvious response. Devesh Kapur noted that the paper suggested that India could perhaps make some concessions on other non-climate issues to prompt rich countries to do more on climate change. However, he wondered if India should do the opposite, that is, threaten damage to rich countries on issues they care about, by, for example, walking away from the CFC Treaty. CFCs affect ozone and have an inimical impact on temperate-zone countries. If India walks out of that treaty, it would signal that it would negotiate on ozone only when the others are ready to negotiate on climate change. He asked why India should not adopt a more hard-line strategy on other issues rather than the soft strategy being advocated in the paper. Rajnish Mehra noted, first, that most of the damage assessment functions are level effects. The implications would be different if they were measuring growth rate effects. He mentioned one of his papers on asset pricing implications of macroeconomic interventions where the growth path of the economy changes. If this happens with climate change as well, in those cases one cannot use standard valuation measures like net present value used by damage assessment studies like the RICE and DICE models because they are looking only at level effects. Second, during the Club of Rome debates, many argued that the world was coming to an end because of population growth. But it did not end, because there is a powerful adjustment mechanism in economics called relative prices, and technologies also evolve. He remarked that it was difficult to determine today what would work—spending money to abate emission today or putting money into R&D to abate emission ten years down the road. However, what was certain is that land prices will change everywhere, in Siberia and in Canada, and migrations will take place because different prices are prevalent in different parts of the world. This adjustment mechanism needs to be addressed in the paper. 148 INDIA POLICY FORUM, 2018

Shreekant Gupta advised the authors to use the Ramsey Rule when doing simulations with those parameters, which would enable them to arrive at appropriate values for ρ and η. Agreeing with Gupta, Rajnish Mehra said that including changes in G growth rates would change gross output, Yt , in the model, which, in turn, would make a huge difference in level changes. Avinash Dixit thanked the authors and the audience for a stimulating discussion of the paper.

Reference

Geruso, M., and D. Spears. 2018. “Heat, Humidity and Infant Mortality in the Developing World,” NBER Working Paper No. w24870. Cambridge, Massachusetts: National Bureau of Economic Research. ERIN K. FLETCHER* Harvard Kennedy School ROHINI PANDE† Harvard Kennedy School CHARITY TROYER MOORE‡ Harvard Kennedy School Women and Work in India: Descriptive Evidence and a Review of Potential Policies

ABSTRACT Sustained high economic growth since the early 1990s has brought significant change to the lives of Indian women. Yet female labor force participation has stagnated at under 30 percent, and recent labor surveys even suggest some decline since 2005. Using the 2011–12 National Sample Survey, we lay out five facts about female labor force participation in India. First, there is significant demand for jobs by women currently not in the labor force. Second, female non-workers have difficulty matching to jobs. Third, women are more likely to be working in sectors where the gender wage gap and unexplained wage gap, commonly attributed to discrimination, is higher. Fourth, vocational training is correlated with a higher likelihood of working among women. Finally, female-friendly employment policies, including job quotas, are correlated with higher female participation in some key sectors. Combining these facts with a review of the literature, we map out important areas for future investigation and highlight how policies such as employment quotas and government initiatives focused on skilling and manufacturing could be leveraged to increase women’s economic activity.

Keywords: Female Labor Force Participation, Jobs, India

JEL Classification: J16, J20, J48, O14, O15

∗ [email protected] † [email protected] ‡ [email protected] § The authors are particularly grateful for comments from Farzana Afridi, Pranab Bardhan, and Karthik Muralidharan, and for helpful input from many participants at the NCAER 2018 India Policy Forum. 149 150 INDIA POLICY FORUM, 2018

1. Introduction

ver the past four decades, India has experienced rapid population Oand economic growth, urbanization, and demographic change. Between 1990 and 2013, GDP growth averaged 6.4 percent (Figure 1); the share of agriculture in GDP roughly halved (from 33 to 18 percent), while that of services increased from 24 to 31 percent. Alongside, urbanization increased from 26 to 32 percent (World Bank 2018). At the same time, women’s education and childbearing patterns have changed: over the same period, total fertility fell from 4.0 to 2.5 children per woman (World Bank 2014a). Girls’ primary school enrollment has reached parity with that of boys, and universal enrollment1 was achieved in 2015 (Neff et al. 2012; UNESCO 2015). Between 1994 and 2010, the fraction of women aged 15–24 attending any educational institution more than doubled from 16.1 to 36 percent (Kapsos et al. 2014). However, despite this rapid economic growth, educational gains, and fertility decline, India’s women remain conspicuously absent from the

FIGURE 1. GDP per capita and Female Labor Force Participation in India over Time

Income and Female LFP: India 40 2500 2000 1500 FLFP India 1000 GDP per capita India 10 20 30 500 0 0 1990 1995 2000 2005 2010 2015 Year Female LFP GDP per capita, constant US$

Source: World Bank World Development Indicators.

1. As a fraction of the school-age population. Erin K. Fletcher et al. 151 labor force. Female labor force participation (FLFP)2 rates remain low and have even fallen in the recent years.3 This perceived decline persists even when we account for increased schooling, which delays entry into the labor force (Klasen and Pieters 2015). Figure 2 shows that FLFP in India is well below its economic peers, and the mismatch between economic growth and FLFP rates in India presents a puzzle. In this paper, we examine possible constraints on participation and potential policy interventions that could increase it, highlighting five descriptive facts relating to patterns of FLFP in India and incorporating a literature review of policy evaluations to identify promising policies worth further investigation.

FIGURE 2. The Cross-Country Relationship between Income and Female Labor Force Participation is U-Shaped, but India is a Major Outlier 1 Lao PDR Nepal Kazakhstan China Australia Singapore

Bangladesh Philippines Japan Uzbekistan Indonesia

India Pakistan Female: Male Labor Force Participation Ratio .2 .4 .6 .8 7 8 9 10 11

Log GNI Per Capita (2013)

Source: World Bank (GNI) and International Labor Organization (LFPR), 2013. Notes: Labor data for Ages 15+. Excludes the Middle East.

2. We calculate the LFP rate by dividing the number of individuals in the working-age population (ages 15–70) employed in wage labor, own-account work, casual labor, unpaid labor, self-employment, or as an employer, plus those unemployed and seeking work, by the entire working-age population (15–70) not currently enrolled in school. 3. Although estimates based on household surveys vary, from a low of 24 percent using the National Sample Survey (NSS; for 2011–12) to a high of 31 percent using the Indian Human Development Survey (for 2004), it is widely acknowledged that FLFP growth has been stagnant, and that some earlier gains have been reversed. 152 INDIA POLICY FORUM, 2018

Implementing effective, evidence-based policy to increase FLFP and increase women’s economic activity could have a large impact on economic growth. Recent evidence from the USA suggests that misallocation of talent in the labor market, whereby high-ability women are in low-skilled, low- return occupations, presents a significant hindrance to growth (Hsieh et al. 2013).4 Specifically, in the Indian context, Esteve-Volart (2004) shows that a 10 percent increase in the female-to-male ratio of workers, a proxy for discrimination-based differential access to labor markets, would increase per capita net domestic product by 8 percent. From an individual woman’s perspective, participation in wage work delays age of marriage and age at first childbirth (Sivasankaran 2014), increases her decision-making power in the household, and increases child schooling (Qian, 2008).5 Figure 3, on the basis of India’s most recent

FIGURE 3. Empowerment/Decision-Making Index by Education Levels using Women’s Report of Autonomy in Decision-Making on Various Expenditures 4 3. 2. 1. Decision-Making Index 0. –.1

Primary or Less Secondary Tertiary

Not Working Family Worker Self-Employed Wage Work

Source: 2015–2016 NFHS. See footnote 6. Notes: Includes ever-married women aged 15–49.

4. Hsieh et al. (2013) find that alleviating gender- and race-based talent misallocation accounted for 16 to 20 percent of US growth over the years 1960–2008. 5. Several observational studies find that women with more control over resources such as land report greater mobility, have children with better nutritional outcomes (Swaminathan et al. 2012), and are less likely to experience violence (Panda and Agarwal 2005). In addition, access to and information regarding female-specific labor market opportunities improves female educational attainment and delays age of marriage and childbearing (Heath and Mobarak 2015; Jensen 2012). Erin K. Fletcher et al. 153

National Family Health Survey (NFHS), shows that women who work, regardless of education level, have more say in household decisions.6 Women’s work also has positive spillovers: Sivasankaran (2014) shows that sisters of women with longer work tenures marry later, and villages that are exposed to more female leaders show lower rates of sex selection (Kalsi 2017). The recent trends in India’s FLFP, combined with their already low levels of participation, are increasingly seen as a challenge that requires policy intervention to ensure that these changes do not result in deterioration in women’s well-being and already low levels of empowerment. Although the justification for a policy focus on FLFP is clear, the fact that observed FLFP levels reflect both supply and demand factors makes determining causation, and thus the range of appropriate policy responses, difficult. To better understand these potential factors, we use household surveys to document key descriptive facts highlighting both the role of social and economic factors that affect labor supply, demand, and outcomes. Given our use of one cross-sectional survey, we primarily focus on the low level of FLFP, rather than the recent decline in rural FLFP. Then we discuss the implications for further investigation tied to existing high-profile policies and government programs. On the supply side, Indian households often require that women prior- itize housework and may even explicitly constrain work by married women (Bose and Das 2018; Sudarshan 2014; Sudarshan and Bhattacharya 2009). Societal expectations of a woman’s role as caregiver and caretaker of the household often mean that women who seek work encounter opposition from their peers and families, leading to lower participation. Women frequently internalize these views and may therefore suppress labor supply even in the absence of explicit constraints. Rustagi (2010) provides evidence that these norms per se have not significantly changed over the last two decades. There is also evidence that these norms are more binding among wealthier, upper-caste households, suggesting that economic growth alone may not alter their influence.7 Low urban FLFP is consistent with this possibility.

6. Figure 3 was made using questions from the 2015–16 NFHS data on women’s roles in household decision-making and women’s views on whether beating is not justified in each of a given set of situations. Using these questions, we create an “empowerment” or decision- making index through principal components analysis, standardized to be equal to zero with a standard deviation of one. 7. Here and elsewhere, we define social norms to be a set of beliefs or perceptions of what one’s community holds to be true or acceptable (Ball Cooper, Paluck, and Fletcher 2012). 154 INDIA POLICY FORUM, 2018

On the demand side, women face legal, normative, and economic constraints to work. Indian women are still subject to laws governing when (i.e., which shifts) and in which industries they can work. These rules may disproportionately affect women even as the economy grows: for example, female participation in export-oriented manufacturing jobs fell despite increased trade and reduced trade barriers during the 1990s, likely due to legal constraints on women’s working hours through factory laws (Gupta 2014). Though these laws may change soon, employers still may be less apt to hire a woman over an equally qualified man. As long as there exist norms against women’s market engagement, we expect to see gender-based discrimination in hiring, legal or otherwise, and gender wage gaps persist that cannot be explained by common sources of observable market variation in wages. Demand for labor of rural Indian women engaged in agriculture is also particularly vulnerable to seasonal and local labor market fluctuations, leading women who count themselves as workers to withdraw into domestic activities when other work is not available (Bardhan 1984). Overall, high, sustained economic growth in India has not necessarily brought more jobs (Bhalotra 1998; Chowdhury 2011; Kannan and Raveendran 2009; Papola and Sahu 2012). Jobless growth in sectors that employ more women or seem friendlier to women necessarily limits growth in FLFP. In the 1980s, jobless growth was evident in manufacturing (Bhalotra 1998), and there is some reason to believe women may have suffered from this relatively more acutely than males. Recent work highlights the lack of jobs to absorb women transitioning out of agriculture, which may further depress demand for potential female labor (Chatterjee, Murgai, and Rama 2015). Norms around women and work clearly affect both supply of, and demand for, female labor. Data from the World Values Survey (WVS) give insight into how norms in India may constrain women’s labor force outcomes, while also highlighting that norms alone can only partially explain India’s low FLFP. Figure 4 shows responses that highlight the prominence of gender- biased views on women’s roles in the economic and political landscape in countries comparable to India. These statistics suggest that views against women holding an equal footing in the classroom and market still persist in India and elsewhere, even among women (albeit to a more limited extent than in males). Interestingly, although India’s FLFP looks most similar to Pakistan, its norms-related responses look more in line with countries that have a significantly higher FLFP, suggesting variation in these views on women and work cannot fully explain India’s lagging FLFP. FIGURE 4. FLFP and World Values Survey Attitudes on Women and Work

Views of Women in the Workplace and FLFP

Working Women Do Not Have as Good Relations with Children Ratio of Female-to-Male Labor Force Participation Rate vs. Stay-at-home Mothers 70 .2 .4 .6 .8 % who agree 40 50 60 80 0 30

India Ratio Female: Male LFP Pakistan Indonesia Bangladesh India China Pakistan Bangladesh Indonesia Female Male Country

Men Should Have Preference for Jobs Men Make Better Business Executives 8 8 6. 6. 4. 4. 2. 2. % who agree % who agree 0. 0. India China Pakistan Indonesia India China Bangladesh Pakistan Indonesia

Female Male Female Male

Source: Attitudes from most recent World Values Survey for each country. Female-to-Male LFP ratios are 2016 ILO estimates. 156 INDIA POLICY FORUM, 2018

Our descriptive analysis, focused on the 68th Round of NSS data, highlights five features of Indian women’s market engagement important for understanding the constraints to higher FLFP and potential policy solutions. First, a large proportion of Indian women express willingness to take on work despite being counted outside of the labor force. There is a strong rural–urban divide in this statistic, as others have noted (Kapsos et al. 2014). Second, women have more trouble matching to jobs than men. They report seeking or being available for jobs longer than men when unemployed, and women who did work reported spending more time unemployed than males. Third, wage gaps and unexplained wage gaps—typically interpreted as at least partially reflecting gender-based discrimination in the labor market— are relatively higher in fields with greater female representation. Fourth, at all levels of education, women with vocational training are more likely to work than those without training. Finally, women are doing relatively well in terms of representation in specific jobs, namely, education and work provided by the government’s job guarantee program, the Mahatma Gandhi National Rural Employment Guarantee Act (MGNREGA); factors potentially driving this success should be investigated further. Alongside these descriptive features, we examine evidence from recent high-quality academic research that seeks to provide causal estimates of policies and other factors affecting FLFP in India. The review of this evidence again underscores the importance of access to jobs, networks, social norms, and the potential importance of policy interventions in women’s labor force decisions. Taken together, the descriptive analysis and evidence review suggest several key areas on which to focus research inquiry, some of which converge with the Government of India’s policy priorities. The government has already put in place programs and policies to increase women’s access to labor market opportunities, namely, increased funding to skills and vocational training programs and gender-based employment quotas. There is some diagnostic evidence and literature that support the implementation of these policies, but the immediate pressing need is for more rigorous research to better understand the causal mechanisms for how these policies might affect female employment. Rigorous testing would also allow for better targeting of policies, both in who is most affected and how they are applied to different groups. An area requiring urgent attention is that of improving data and evidence to better understand the constraints and solutions to India’s low FLFP. We outline specific steps related to data collection that can raise women’s visibility in the labor force and serve as a potential impetus for important dialogue and initiatives aimed at engaging them more effectively in the economy. Erin K. Fletcher et al. 157

2. Data and Diagnostics Methodology

2.1. Data Our primary data source is the employment module of the Indian NSS for 2011–12 (68th Round). Our analysis sample consists of 136,465 women and 131,542 men aged 15–70 who are not currently enrolled in school.8 We define and examine labor force participation (LFP) using the survey question on usual principal activity of each household member who meets our inclusion criteria, unless otherwise noted.9 The LFP rate is calculated using the sum of all individuals employed in wage labor, own-account work, casual labor, unpaid labor, self-employment,10 or as an employer, plus those who are unemployed and seeking work, divided by the working-age population (15–70) not currently enrolled in school.11

2.2. Descriptive Summary of FLFP in India The variation in FLFP across India is striking—at the state level, FLFP rates vary from below 20 percent of the male LFP rate to nearly 80 percent—and its cross-sectional relationship with income does not align with the standard economic development story. Figure 5 examines the relationship between the natural log of net state per capita domestic production, a proxy for per capita income, and the ratio of female-to-male LFP, for comparison to the cross-country estimates presented in Figure 2. Although Indian FLFP is low from a cross-country perspective (below the U-shape), Indian states do not follow any sort of U-shape themselves in the cross-section; instead, FLFP is generally flat, with outliers on the higher end where some states at middle and higher relative incomes are associated with higher FLFP. What explains the differences in FLFP across India’s states? Figure 6 shows the level of FLFP by state, on the left, and the unexplained

8. This nationwide survey includes 459,784 individuals from 100,957 households. We drop individuals who do not report marital status or employment type and weight the survey as instructed, unless otherwise indicated. 9. We use the question on “principal usual activity status” from Block 5.2 of NSS Schedule 10. 10. Own-account workers are self-employed individuals operating their own enterprises, largely without hiring labor. Self-employment generally refers to persons who work in their own enterprises, often with the help of hired labor or employees. Unpaid refers to unpaid family workers. Regular employees receive salary or wages on a regular basis. Casual workers receive a wage according to the terms of a daily or periodic work contract. 11. Though some analyses of LFP in India include secondary activity statuses (e.g., Kapsos et al. 2014), we limit the definition of LFP to usual principal activity. FIGURE 5. Indian States’ FLFP Is Relatively Flat across Income Levels

State Per Capita Income and Relative Female Labor Force Participation

.8 SK HP ML

.6 CG MZ AP AR

.4 NL TN AN KA MH RJ RJ UK GA Participation Ratio MN TR MP GJ PY Female: Male Labor Force OR .2 CH WB DL JH UP AS JK PB HR BR 0

10 10.5 11 11.5 12 12.5 Natural Log State Net Domestic Product Per Capita

Source: State Net Domestic Product per Capita from MoSPI for 2011–12; Ratio of female-to-male labor force participation rates computed using NSS Round 68 from 2011–12. FIGURE 6. Some States Have Higher FLFP than Others, after Controlling for Income and Education

Source: NSS 68th Round. 160 INDIA POLICY FORUM, 2018 component of each state’s FLFP after controlling for the state’s mean income and (dummied) education levels in cross-state regressions on NSS data. Strikingly, some states have both high FLFP and a large component that is not explained by their income or education levels. A key question for policy, then, is: What are the features of these states, such as Himachal Pradesh or Chhattisgarh, from which other states can learn? One potential explanation here is the more progressive gender norms typically thought to characterize these two states. Beyond the state-level differences, descriptive statistics on FLFP show a significant difference in how men and women interact with the labor market, as well as regional and inter-caste differences among women. Male LFP averages 96 percent, while FLFP averages only 27 percent, and, as documented elsewhere (Klasen and Pieters 2015), FLFP is lower in urban areas relative to rural areas. Further, 76 percent of women in urban areas report their primary activity as domestic duties compared to 67 percent in rural areas. Women in rural areas are more likely than their urban counterparts to work in unpaid family labor. Rates of wage work and self-employment for women are similar, but low, in rural and urban areas. Table 1 provides basic summary statistics related to FLFP in India, and Figure 7 highlights the diversity in district-level FLFP patterns. These urban–rural differences in FLFP are important, given the much higher education levels among urban women: over 60 percent of women in rural areas have at best a primary education, while this is only true for 30 percent of urban women. Yet higher education does not predict higher FLFP rates linearly. Instead, we observe a U-shaped relationship between education and FLFP (Figure 8), much like the cross-country relationship between income and FLFP (Figure 2). Women at very low levels of education are more likely to be in the labor force, with 20 percent of low-educated women in the labor force in urban areas and 28 percent in rural areas. Women with some secondary education have the lowest levels of participation (around 22 percent) and highly educated women again post higher levels of FLFP. The U-shaped relationship is the clearest for urban women and likely reflects an income effect, whereby women opt out of the workforce and into greater household production and leisure as household incomes rise, and then opt back into market work as the opportunity cost of remaining out of the labor force increases. This U-shaped relationship between education and work for women stands in contrast to male LFP, which increases with education and is nearly universal, excluding those currently enrolled. Figure 9 shows that the age profile for FLFP differs across rural and urban areas. Young urban and rural women are similarly likely to enter the TABLE 1. Summary Statistics Out of Labor Force but Out of Labor Force and In Labor Force Willing to Work Not Willing to Work Rural Urban Rural Urban Rural Urban Rural Urban Variable Males Females Females Females Females Females Females Females Females Females Age 38.591 38.249 38.057 38.558 37.838 36.720 31.324 32.170 39.894 40.715 (13.170) (13.535) (13.723) (13.224) (12.499) (11.889) (10.129) (9.834) (13.817) (13.155) Married 0.774 0.805 0.815 0.789 0.744 0.594 0.860 0.853 0.877 0.875 (0.418) (0.396) (0.388) (0.408) (0.436) (0.491) (0.347) (0.354) (0.328) (0.331) In labor force 0.960 0.263 0.289 0.223 1.000 1.000 – – – – (0.196) (0.441) (0.453) (0.416) 0.000 0.000 – – – – Less than primary education 0.259 0.435 0.509 0.317 0.521 0.308 0.403 0.262 0.526 0.318 (0.438) (0.496) (0.500) (0.465) (0.500) (0.462) (0.491) (0.440) (0.499) (0.466) Primary education 0.318 0.278 0.283 0.271 0.260 0.220 0.346 0.330 0.281 0.276 (0.466) (0.448) (0.450) (0.445) (0.439) (0.414) (0.476) (0.470) (0.450) (0.447) Secondary education 0.157 0.118 0.101 0.146 0.088 0.096 0.130 0.166 0.102 0.165 (0.363) (0.323) (0.302) (0.353) (0.284) (0.294) (0.336) (0.372) (0.303) (0.371) Certificate/Sr. secondary education 0.123 0.084 0.066 0.113 0.070 0.113 0.081 0.124 0.060 0.113 (0.328) (0.278) (0.248) (0.316) (0.256) (0.317) (0.273) (0.330) (0.238) (0.317) Tertiary education 0.143 0.084 0.041 0.153 0.060 0.263 0.040 0.118 0.031 0.128 (0.350) (0.277) (0.198) (0.360) (0.238) (0.440) (0.197) (0.323) (0.174) (0.334) (Table 1 Continued) (Table 1 Continued)

Out of Labor Force but Out of Labor Force and In Labor Force Willing to Work Not Willing to Work Rural Urban Rural Urban Rural Urban Rural Urban Variable Males Females Females Females Females Females Females Females Females Females Self-employed 0.394 0.061 0.069 0.049 0.237 0.220 – – – – (0.489) (0.239) (0.253) (0.216) (0.425) (0.415) – – – – Unpaid family worker 0.098 0.075 0.103 0.031 0.357 0.137 – – – – (0.297) (0.264) (0.304) (0.172) (0.479) (0.344) – – – – Wage worker 0.439 0.111 0.104 0.123 0.359 0.551 – – – – (0.496) (0.314) (0.305) (0.328) (0.480) (0.497) Domestic duties/Collection of goods 0.006 0.703 0.677 0.745 – – 1.000 1.000 1.000 1.000 (0.079) (0.457) (0.468) (0.436) – – 0.000 0.000 0.000 0.000 Unemployed/Other 0.064 0.050 0.049 0.053 0.047 0.092 – – – – (0.244) (0.218) (0.215) (0.223) (0.213) (0.288) – – – – N 131,542 136,465 83,936 52,529 24,238 11,705 18,462 11,088 38,319 28,049 Source: NSS, 2011–12. Notes: Standard errors in parentheses. Sample restricted to individuals aged 15 to 70 years not currently enrolled in school. Erin K. Fletcher et al. 163

FIGURE 7. FLFP by District Female Labor Force Participation Rate by District

Female Labor Force Participation

Source: NSS, 2011–12. labor market, but FLFP rates across rural and urban areas for women in their mid-20s and older diverge; the higher rural FLFP primarily reflects these women’s participation in agricultural activities. The cross-section does not allow us to separate cohort and secular trends, limiting the conclusions that can be drawn, but the relatively low FLFP among both rural and urban young women is particularly disturbing since these young women are not enrolled in school. It is also suggestive of a lack of opportunities (or acceptable opportunities) for young women in rural areas, in comparison to less educated older rural women, in general. 164 INDIA POLICY FORUM, 2018

FIGURE 8. Educational Profile of Labor Force Participation for Men and Women

Male LFP by Education Female LFP by Education 1 1 .8 .8 .6 .6 mean of ilolf mean of ilolf .4 .4 .2 .2 0 Rural Urban 0 RuralUrban

Source: NSS, 2011–12. Note: Includes individuals aged 15–70 not enrolled in school.

FIGURE 9. Age Profile of Labor Force Participation among Women by Geographic Location

Relationship Between LFP and Age by Rural/Urban .4 3 2. Share in Labor Force 1. 0. 15 25 35 45 55 65 Age Urban Female Rural Female

Source: NSS, 2011–12. Note: Includes women aged 15–70 not enrolled in school. Erin K. Fletcher et al. 165

Social norms surrounding female work are an important constraint on FLFP in India, as they may dictate that women are primarily caregivers and thus belong in the home. Although we do not observe a sharp M-shaped relationship between age and FLFP—exit at childbearing and re-entry as children get older—as in Japan or Korea (Kawata and Naganuma 2010; Lee et al. 2013), FLFP does show a drop-off among women in their early to mid-20s in urban areas, suggesting that marriage and family-related responsibilities may specifically limit women’s LFP. Household surveys show that 13 percent and 50 percent of women are not allowed to visit village markets or stores alone, respectively, so imagining that women face constraints on working outside the home is not a large jump. These social norms are linked to the caste system; upper-caste women are more likely to face restrictive norms (Field et al. 2013).12 Figure 10, using the NSS, shows FLFP age profiles by whether the household is identified as Scheduled Caste (SC), Scheduled Tribe (ST), Other Backward Class (OBC), or other Hindus and Muslims. Those identified as SCs are the most likely to be working at all ages. All other social groups are much less likely to be working, but particularly for the youngest cohorts. High-caste Hindus and Muslims post the lowest rates of FLFP at all ages, consistent with other research. Household responsibilities and childrearing duties are often cited as key constraints to women’s participation in the labor force. Figure 11 illustrates how FLFP varies for married and unmarried women with and without children in the household over the cross-sectional age profile. The biggest takeaway from this figure is that women who marry have low LFP across all

12. Social norms may also affect whether survey questions can adequately measure the full extent of female participation in the labor market. If women identify strongly with a non-labor market role, such as caregiver or mother, or feel they are expected to identify with that role, they may designate that as their primary activity, even if they spend time in remunerated activities. Other nationally representative datasets from India also show slightly different levels of overall FLFP. The first round of the IHDS, a survey undertaken in 2004–05, estimated overall FLFP in India at 31 percent (14.6 percent in urban areas and 39 percent in rural areas), compared to 35 percent as reported by the ILO for 2004 (World Bank 2014b). The difference in overall levels of participation may reflect that women do not necessarily identify with work as their primary activity, and the use of more probing questions and time-use data would result in more available information on the productive and even income-generating activities of women. Further analysis of the IHDS shows similar patterns to the NSS in the relationships between key variables such as age, urban/rural location, and social group, even while the levels of participation for these sub-groups tend to be higher in the IHDS. Trends over time shown in the NSS data and statistics collected by the ILO and World Bank are likely real, even if we are concerned that the actual level of participation is obscured by reporting biases. 166 INDIA POLICY FORUM, 2018

FIGURE 10. Labor Force Participation by Age, Disaggregated by Social Group

Relationship Between FLFP and Age by Social Group and Religion .5 4 3. 2. Share in Labor Force 1. 0. 15 25 35 45 55 65 Age SC ST OBC Other Hindus and Muslims

Source: NSS, 2011–12. Note: Includes individuals aged 15–70 not enrolled in school.

FIGURE 11. FLFP by Marital Status and Presence of Children in the Household

Proportion of Women in Labor Force by Marital Status and Presence of Children in Household 8 6. .4 2. Proportion in Labor Force 0. 15 25 35 45 55 Age

Not married, children in hh Not married, no children in hh Married, children in hh Married, no children in hh

Source: NSS 68th Round, 2011−12. Note: hh = Household. Erin K. Fletcher et al. 167 ages, suggesting that older cohorts have not entered the labor force even as children grow up. A second insight is that the largest differences in LFP are reflected in marital status rather than the presence of children in the household, particularly during prime working ages. As approximately 95 percent of Indian women aged 25 and older are married (or formerly married), lower FLFP dominates. Below we highlight additional key descriptive facts about India’s FLFP to build on some of these more well-established features. 1. A significant portion of out-of-labor-force women express willingness to work: although socially constrained labor supply may explain part of low FLFP, women do express willingness or desire to work. Among both rural and urban women, particularly of certain demographic groups, a significant portion would be willing to take on work if it were offered. More than 30 percent of the group of women engaged primarily in domestic activities— and counted outside the labor force—would like to work and thus constitute a potential addition to the labor force or latent labor supply.13 If all these women who stated they would take work actually did, we would see a 21-percentage point (78 percent) rise in the FLFP rate, substantial given the low rates of participation overall. Women currently out of the labor force who are willing to take a job tend to be more educated, slightly more likely to live in rural areas, and not belonging to the SCs or STs. Figure 12 summarizes how education, geography, and social group (SC, ST, OBCs, and general categories) correlate with willingness to work. The percentage willing to work is slightly higher in rural areas (32 percent of respondents) than in urban areas (28 percent). Among rural women, latent labor supply is generally higher among those with more education. Almost 45 percent of rural, highly educated women who report their primary activity as domestic duties also report that they would accept work. Inter-caste differences in reported willingness to take on work point to the importance of norms in latent labor supply, particularly in urban areas, as suggested by Klasen and Pieters (2015). Figure 12 shows that women from “Other” and “OBC” categories consistently express lower willingness to work than SC and ST women of the same education levels and geographic sector. Among urban women in the OBC/Other categories, willingness to work does not increase with education. In contrast, urban

13. While only 815 males in the entire NSS were categorized as belonging to the domestic worker category and were asked this same question, a similar percentage (35 percent) report being willing to take on work. 168 INDIA POLICY FORUM, 2018

FIGURE 12. Women’s Willingness to Take Work by Education Level and Social Group (Those Occupied with Domestic Duties Only)

Willingness to Accept Work by Housewives 6 4 2 0

Proportion of Housewives Willing to Work Primary Primary

Graduate/Post-graduate Graduate/Post-graduate Certificate/Sr. Secondary Certificate/Sr. Secondary

Rural Urban ST SC OBC Other

Source: NSS, 2011–12. Notes: Includes women aged 15–70 not enrolled in school.

SC and ST women have a relatively U-shaped expressed willingness to work, reflecting the typical income and substitution effects. Rural women’s willingness to work, in contrast, generally increases within caste as education increases, pointing again to the lack of jobs for women at higher education levels in rural areas. Unsurprisingly, among women who did not work, over 90 percent were primarily occupied with domestic duties in the previous year; 92 percent of these women said domestic duties were their principal activity in the previous year because they were required (needed) to perform these activities, with 60 percent of these women reporting that there was no other household member available to carry out these tasks. Only 15 percent reported social or religious constraints as the predominant reason they were required to spend their time focused on domestic duties. 2. Job matching is more difficult for females than males: Analysis of available data on job-seeking suggests that women experience greater difficulty matching to jobs that suit them than men. If women have preferences Erin K. Fletcher et al. 169 for non-agricultural jobs in rural and peri-urban areas, the lack of non- agricultural jobs for women may explain low FLFP, in general, and the decline in rural women’s LFP specifically (Chatterjee et al. 2015). The types of jobs women report wanting vary by age, but are primarily of a part-time nature, reflecting the demands of other household responsibilities, particularly in the context of marriage and childbearing. While 73 percent of women willing to take a job prefer regular, part-time work, 22 percent want regular, full-time work; the remaining 5 percent want a mixture of only occasional full or part-time work. The youngest women are most likely to report wanting a full-time job, while those in the middle age ranges are most likely to prefer regular part-time work (Figure 13). Yet preferences of those outside the labor force do not align with jobs women have. Figure 14 compares the type of work undertaken by female workers to the type of work preferred by women out of the labor force who report being willing to take on a job. Of women who work, just under 17 percent report working part-time, over six times the rate that males report but less than a quarter the rate expressed as preferred by willing women workers—again pointing to a potential lack of jobs that may suit women’s

FIGURE 13. Type of Work Women Counted Out of the Labor Force Would Accept by Age

Type of Work Women Would Accept, By Age .4 .6 .8 1

Full−time Regular part−time Occasional full−time Occasional part−time

Source: NSS, 2011−12. Note: Includes individuals aged 15−55 not enrolled in school. Excludes those in the labor force. 170 INDIA POLICY FORUM, 2018

FIGURE 14. Current Female Employment Distribution and Type of Work Preferred by Female Domestic Workers Who Say They Want Jobs

Type of Employment of Female Workers and Preferred Work by Women out of Labor Force Actual Employment, Preferred Type of Employment, Women in Labor Force Women out of Labor Force

1 Who Report they Would Take on Work 1 .8 .8 .6 .6 .4 .4 .2 .2 0 0

Regular Full–time Regular part–time Occasional full–time Occasional part–time

Source: NSS, 2011–12. Note: Includes women aged 15–45 not enrolled in school. Women asked question for the graph on the right are those occupied with domestic duties and counted out of the labor force but say they would take on work made available to their household. preferences or obligations. Although only 5 percent of women out of the labor force who report being willing to take on work say they would prefer occasional work, 16 percent of women who did work were not working regularly, nearly twice the rate reported by males. Although women who work may prefer different types of work than those that remain at home occupied with domestic duties, the fact that employed women are overwhelmingly situated in full-time work while those who would like to enter the labor force prefer part-time work points to important supply–demand mismatches relevant to low FLFP rates. Finally, the process of job search itself is gendered: among those counted in the labor force, women who did not work the entire previous year spent more time seeking a job or being available for a job than men. Women who did work report being without work slightly longer than men as well. And even a sub-set of women reporting they were solely occupied Erin K. Fletcher et al. 171 with domestic duties report this was because there was no work available for them.14 Consistent with the possibility that labor market conditions constrain women’s market activities, those women counted in the labor force in the NSS Round we use also report significant differences in time spent in work and domestic activities in the previous week based on the month in which they were surveyed.15 Taken together, these statistics point to a market less closely aligned with female job-seekers than males. However, despite their stated willingness to work, women reported searching for jobs with less intensity than men. One-third of women report not seeking a job when they were unemployed, compared to 18 percent of men. It is difficult to disentangle the reasons for this differential search. Social desirability bias, whereby respondents are unable or unwilling to report true answers on sensitive subjects due to their perception of what is right or acceptable for women’s work, may lead to under-reporting of women’s willingness to take a job or—probably more consequentially— actual activities undertaken in a job search (Fisher 1993). Lower expected success in job searches may also result in women searching for jobs with less intensity than men, and, again, norms may constrain labor supply even when women prefer to work. 3. Wage gaps and unexplained wage gaps are higher in fields with greater female representation: How do women tend to fare in sectors in which they are most likely to work? We examine this question looking at the first (primary) activity women report undertaking in the previous week and the daily wages they report for this activity. Activities are classified using India’s National Industrial Classification (NIC) codes from 2008.16,17 The graph on the left-hand side of Figure 15 highlights how economic activities

14. The NSS question covering latent labor supply reads, “In spite of your preoccupation in domestic duties, are you willing to accept work if work is made available at your household?” It is asked of individuals who say they are primarily occupied with domestic duties only or domestic duties and the free collection of goods. 15. We utilize the NSS current weekly activity status to regress time spent on work, and time spent on domestic duties on the month of the survey for women counted in the labor force, similar to the approach used by Bardhan (1984) for rural West Bengal. 16. Of the 8 percent of women primarily occupied with domestic duties who said they were not required to be occupied with these tasks, just under 20 percent reported they continued working on domestic activities because there was no other work available to them. 17. NIC codes, produced by the Central Statistical Organisation in India, classify economic activities at the group, class, and sub-class level. We collapse the two-digit numeric codes, known as divisions, further among similar types of activities without fully condensing to the much broader section categorization. A detailed mapping of the NIC codes to the collapsed codes is available in Table A.1. FIGURE 15. Gender Wage Gaps, and Unexplained Wage Gaps across Types of Work

Wage Gap and Proportion of Employees Unexplained Wage Gap and Female in Sectors that are Female Representation in Sectors 1.52 51 Unexplained Wage Gap 0 51 Female Wage as Proportion of Male Wage 5. 0. –. 0.1.2.3.4.5.6 0.1.2.3.4.5.6

Proportion of Workers in Sectors that are Female Proportion of Workers in Sectors that are Female

Manufacturing/Construction Services Ag/Forestry/Fishing

Source: NSS, 2011−12. Note: Daily wages have been calculated on the basis of pay for main activity reported in the previous week. The Y−axis on the right-hand graph shows the unexplained component of the male−female wage gap after controlling for worker marital status age, social group, education (secondary, tertiary), and state using Oaxaca−Blinder decomposition for each NIC sector of work. Erin K. Fletcher et al. 173 in which women represent a larger proportion of the workforce are also those in which gender wage gaps are larger, as measured by the female wage as the proportion of male wages. Overall, women tend to be less represented in the service sector, and manufacturing industry is an important employer of women. In other work, we have shown how the gender gap in LFP in the services sector is 19 percent in favor of men, but 1 percent in favor of women in manufacturing, and women’s relative representation in manufacturing grew from 15 percent to 25 percent between 2010 and 2012 (Prillaman and Moore 2016). These facts alone raise important questions about the future of female employment, given the often-cited narrative on the role of service sector jobs in women’s increased employment, particularly as countries continue to develop economically (Goldin 1995). Wage gaps alone, however, may simply reflect differences in the labor force composition across genders on the basis of easily observable characteristics, such as education. Oaxaca–Blinder decompositions can highlight the extent to which the gender wage gap is driven by these observable differences across genders (Blinder 1973; Oaxaca 1973). The right-hand side graph in Figure 15 plots the unexplained wage gap that remains within each NIC category after netting out observable differences in marital status, age, social group (SC, ST, OBC, Other), education (secondary and tertiary education), and state-fixed effects across workers by gender on the natural log of wages by gender. Importantly, the unexplained component of the wage gap also tends to be larger for sectors in which females represent a larger proportion of all employed in that sector. Stated differently, the sectors in which females tend to fare relatively better in terms of wage gaps are often those in which they are least represented. Sectors with the lowest unexplained wage gap tend to be in the service sector, although a good number of service sector jobs also perform relatively poorly on this measure. 4. Women with vocational training are more likely to work at all levels of education: Conditional on reporting they were willing to accept a job, the NSS asked a sample of women whether they have the requisite skills to take on the type of work they preferred. More than half of these out-of- labor-force women who were primarily occupied with domestic duties and stated they were willing to take on work said they did not have the skills required to undertake work in their desired fields (Figure 16). Interestingly, women who have attended skills or vocational training, whether formal or informal, are more likely to be working. Women who 174 INDIA POLICY FORUM, 2018

FIGURE 16. Women’s Stated Skill Deficits

Women in Domestic Duties Lacking Skills by Desired Type of Work .8 .6 .4 .2 Proportion of Willing with Requisite Skills 0

Other

Food processing Animal Husbandry Textiles/Tailoring Leather/Wood manf. Rural Urban

Source: NSS, 2011–12. Note: Includes women aged 15–70 not enrolled in school. have participated in skills (vocational) training have higher levels of FLFP, regardless of educational levels (Figure 17), though the U-shaped relationship between education and FLFP persists. Although noteworthy, skills trainees are likely positively selected on a variety of dimensions and this relationship should, therefore, simply draw attention to the need for additional investigation and testing. 5. Fields with female-friendly policies have higher female representation: Despite their overall low LFP, certain fields and occupations employ many women and, in some cases, more women than men. Figure 18 highlights fields with high numbers of women employed by rural/urban status. As expected, agriculture is the most common employer of working women, with approximately 55.6 million women working in agriculture in rural areas alone. The next most common is manufacturing of textiles, food, and other products, which is a significant employer of women in both rural and urban areas. Women are also frequently employed in construction across both geographies. Other common fields employing women across urban and rural areas in the service sector include education, retail trade, and home-based services. Erin K. Fletcher et al. 175

FIGURE 17. Labor Force Participation by Educational Attainment of Respondents on the Basis of Participation in Skills Training

Labor Force Participation by Skills Training Recipients .6 4 2. Proportion in the Labor Force 0.

No Training Formal or Informal Vocational Training

Source: NSS, 2011–12. Notes: Includes women aged 15–70 not enrolled in school.

Fields with the highest proportion of female workers are not necessarily those with the highest numbers of female workers, and only a few fields exceed 50 percent representation. These fields include domestic workers in both rural and urban areas and some limited manufacturing in rural areas. Notably, female representation and overall employment numbers are relatively high in education, some manufacturing, and limited services across both rural and urban areas. The Government of India has worked to implement gender-sensitive policies in certain industries and occupations to increase gender parity. Primarily, these have worked through quotas, which we discuss further in the policy section, but here highlight the sectors in which there are quotas and women have relatively high participation. The Mahatma Gandhi National Rural Employment Guarantee Scheme (MGNREGS) provides up to 100 days of paid unskilled work per rural household annually. In contrast to the national labor market, which is comprised of only 22 percent women overall, 54 percent of MGNREGS person-days FIGURE 18. Number of Females Employed by Type of Work

Fields with Highest Number of Female Employees

Rural Urban

55,673

3,000 3,000

2,500 2,500 Thousands Thousands 2,000 2,000

1,500 1,500

1,000 1,000

500 500

0 0 e

Education Education AgricultureFood Manu Retail trade Wood Manu AgricultureRetail trade Food Manu Textile Manu Construction Textile Manu Human healthConstruction

Civil eng. architecture Chem/Bio/Metal manu Chem/Bio/Metal manu Other home repair/services Domestic personnel/hhOther homeus repair/services

Source: NSS 68th Round, 2011−12. Note: The type of employment is that listed as first activity in the weekly time-use module for the sector. Erin K. Fletcher et al. 177 were completed by women in fiscal year 2018–19.18 MGNREGS uses a gender quota, requiring that at least one-third of person-days are worked by females, but the 33 percent requirement is clearly exceeded, and, therefore, cannot fully explain such high levels of female participation. Other potential reasons why MGNREGS attracts women include its wage parity policy, which may be particularly appealing for unskilled rural women accustomed to large gender wage gaps, and because it provides work for women near their households. The education sector is also a large employer of women in both rural and urban areas, as mentioned earlier, and the share of female teachers has risen over the past four decades (Chin 2005). One possible explanation for this rise is the implementation of Operation Blackboard in 1990, a government initiative to increase educational attainments, which included a de jure quota for the proportion of female teachers at 50 percent. This quota has not been rigorously analyzed, and female representation continues to fall short of the 50 percent mark. However, the fact that education is an important sector for female employment suggests that gender-sensitive policies directed at the education sector may be features relevant to women’s relatively high participation.

3. Evidence Review

Against the background of descriptive facts, we review recent academic literature to identify potential policy levers for increasing FLFP. India has been host to a number of rigorous academic studies that seek to tackle causality concerns; several of these exploit the varied conditions and policies in India’s states. We perform a selective review of rigorous papers with a strong causal identification strategy (i.e., quasi-experimental, Randomized Controlled Trial (RCT), experimental) from a list of top academic journals and working paper series over the years 2004 to 2017 from India, with select papers of particularly high relevance included from other countries in the region. The review methodology and included papers are summarized in Appendix (Table A.2). The literature confirms findings from the descriptive evidence above that women have limited access to the labor force. Norms, declining FLFP in rural areas due to a lack of access to part-time work and work outside of agriculture, job mismatch, and more are important constraints that we

18. According to the MGNREGA Report Dashboard, available at http://mnregaweb4.nic. in/netnrega/all_lvl_details_dashboard_new.aspx (accessed on April 10, 2019). FIGURE 19. Fields of Work with Highest Representation of Females

Fields with Highest Proportion of Female Employees (Excludes Agriculture) Urban Rural 32% 10000 10000

8000 8000 44% 40% 51% 38% 47% 6000 6000 Thousands

4000 Thousands 4000 39% 31% 70% 40% 2000 53% 34% 2000 49% 22% 44% 39% 37% 31% 0 26% 25% 0

Education Consulting Education LibrariesResearch Publishing/Media DomesticFood Personnel Manufacturing Waste Management Res. Care/SocialTextile ManufacturingWork Other Manufacturing Domestic Personnel Food Manufacturing Textile ManufacturingRes. Care/Social Work Human Health Activities Other Home Repair/Services Civil Eng., Architecture, and Related Employment Acts/Office Support

Males Females

Source: NSS, 2011–12. Note: Numbers above bars show percentage employees in the sector that are female. Type of employment is that listed as first activity in time use module for sector. Excludes agriculture. Erin K. Fletcher et al. 179 examine in more detail in this section. Randomized and quasi-experimental evaluations show that there are proven methods to alleviate these constraints and encourage more women to join the labor force, also described further.

3.1. Information Women often lack information about returns to work and access to adequate job opportunities. When coupled with restrictive social norms, lack of information may depress how and when a woman may work, but research shows that these norms are not immutable. Information, obtained via active recruitment or through family ties, can affect women’s work and family outcomes. Active recruitment of women by the business processing outsourcing sector increased FLFP in that sector and by 2.4 percentage points overall (Jensen 2012) and sisters of factory workers were more likely to delay marriage and childbearing (Sivasankaran 2014). In the Philippines, women who were encouraged to attend a job fair were more likely to be in formal and informal employment, though less likely to be self-employed (Beam 2016).

3.2. Job Location Where travel is difficult, costly, or constrained due to norms linked to mobility, proximity to jobs is an important constraint. Although evidence on importance of job proximity in India is low, in nearby Bangladesh, factory placement is predictive of who works. Women living in close proximity to garment factories were 6.5 to 15.4 percentage points more likely to be employed than women far away from them (Heath and Mobarak 2015). In Pakistan, the presence of a government school was associated with more private schools, which increased female employment as women primarily staff such schools (Andrabi et al. 2013).

3.3. Peer Effects Like information, role model or peer effects can have an impact on women’s participation. In areas where jobs that women prefer are not available, self- employment may provide opportunity and flexibility for women to enter the labor market, and having contacts and role models can lead women to take steps to grow their businesses. Business training on its own increases the likelihood that women will take out loans for self-employment (Field et al. 2013; 2016), but inviting a friend to business training has a positive differential impact in encouraging women to take out loans over and above business training itself, particularly for women most constrained by norms (Field et al. 2016). 180 INDIA POLICY FORUM, 2018

3.4. Economic Returns and Norms Formation Environmental and institutional features can shape FLFP and have lasting effects. Comparing districts with soils in need of significant hard labor to areas with soil that is more easily worked, Carranza (2014) shows that high FLFP is persistent across time; a 10-percentage point higher fraction of loamy to clayey soils (proxies for areas in which females would be less likely to provide agricultural labor) is associated with a 5.1 percent decrease in FLFP in India. Similarly, plough use, which is associated with soil type, is con- nected to historical FLFP in agriculture, which contributed to the formation of norms around women’s work (Alesina et al. 2013).

3.5. Discriminatory Laws Legal barriers to female employment—restrictions on working hours or differential skill levels—are key to understanding how a discriminatory policy may affect overall participation. These restrictions interact with other policies. Notably, Gupta (2014) shows that reductions in trade barriers in India actually reduced female employment. Though the author cannot show that these effects are directly linked to discriminatory policies, the factory laws, which prohibit women from working certain shifts, are a likely culprit.

3.6. Targeted Policies Equality-enhancing laws may also exert effects on FLFP. Note that traditional economic levers, such as tax policies and incentives, which have been shown to be important contributors to women’s labor supply decisions in developed countries, are likely not a major determinant of FLFP in India, where 10 percent of the population19 is part of the formal labor force. The Hindu Succession Act, which granted women in parts of India equal inherit- ance rights, differentially affected geographic, religious, and ethnic groups. Heath and Tan (2014) exploit this natural experiment to show that women in the affected groups were 9.7 percentage points more likely to be working and 5 percentage points more likely to be working outside the home. Cash and asset transfers to female-headed households where recipients often survive on less than two dollars per day have also been shown to increase welfare for women. Banerjee et al. (2011) show that productive

19. As estimated from NSS 68th Round data for the population ages 15–70 not in school. A respondent is considered to be part of the formal labor force if they have a written contract for a job they hold, thus providing a lower bound on population participation in a job where income taxes would be relevant to the household. Erin K. Fletcher et al. 181 asset transfers (namely, livestock) to very poor women in West Bengal, when paired with training and savings, resulted in increased consumption, at least in part through increases in small business activity as well as an increase in labor supply on the intensive margin. Other findings from Bandiera et al. (2009) show that such asset transfers led to increased business skills and increased time spent working. These intensive margin effects on LFP could improve outcomes for self-employed women by increasing self-employment income or profits. In nearby Sri Lanka, business training plus cash grants were more effective at increasing profitability of female-owned businesses (De Mel et al. 2014). Finally, research also shows how transfers of MGNREGS wages into a woman’s own bank account, rather than that of the household head, in an RCT in Madhya Pradesh, increased women’s work under MGNREGS. Beyond this expected impact, the intervention also highlighted the potential importance of gender-specific norms related to women’s work in the household: women who were granted access to their workfare wages also worked more in the private sector and undertook more economic activities overall. The authors attribute these changes to increases in women’s intrahousehold bargaining power that induced them to work despite the social costs incurred to men whose wives worked. Survey data collected three years after the intervention began point to the role of this policy in changing views on women and work: women viewed women’s work outside the home more favorably, and husbands thought the social cost paid when their wives were working was lower (Field et al. 2019). The study points both to the role that social norms can play in restricting women’s work and the potential of targeted policies to help overcome these constraints.

3.7. Quotas India has a long history of implementing quotas. Since 1982, a certain percentage of public sector jobs has been reserved for SCs and STs. Starting in 1987, Operation Blackboard required that 50 percent of teachers be women. Further quotas have been proposed; the Women’s Reservation Bill would reserve 33 percent of seats in India’s lower house of Parliament for women, but has been awaiting passage in the Lok Sabha since 2010. Few of these gender-based quotas have been rigorously evaluated, but perhaps the greatest wealth of knowledge we have on causal evidence to increase FLFP comes from the Indian Government’s experiment with quotas for female leadership at the local level. A 1993 law mandated that one-third of seats on village councils (gram panchayats) be reserved for women. In many Indian states, the choice of 182 INDIA POLICY FORUM, 2018 which councils would be reserved was in effect random, which allowed for a rigorous examination of the effects of quotas on various outcomes. Quotas were implemented on a village-by-village basis and a village reserved for a female head in one election was not reserved in the next. Several papers exploit the as-good-as-random variation in the rotating system of implementation to show the effects of gender-based electoral quotas on female participation in politics. Bhavnani (2009) shows that wards in Maharashtra that had been reserved for female heads once saw a 120 percent increase in the average number of female candidates in the subsequent election. In West Bengal, women living in villages that were twice reserved were 2.8 to 3.2 percentage points more likely to stand for office and 4.5 to 5.5 percentage points more likely to win (Beaman et al. 2009). The electoral program quotas exerted effects on FLFP, female time use, and entrepreneurship, in addition to their direct participation in politics. Women in areas with female leaders were 39 to 52 percent more likely to start businesses than those in areas without leaders (Ghani et al. 2014). Beaman et al. (2009) showed that the gender gap in career aspirations of adolescents closed by 32 percent in villages that had been reserved for two election cycles. The gender gap in adolescent educational attainment was completely erased in villages with a reserved female head, while girls spent less time on household chores. Female participation in the MGNREGS national workfare program increased following the election of female leaders. Female person-days worked in the program were higher by 6 percent in areas that were exposed to quotas (Bose and Das 2018).

4. High-Potential Research Areas

Given the descriptive evidence and existing research, and in light of India’s current policy priorities, what are the most important avenues for investigation and testing to increase FLFP? We highlight several important areas that merit additional investigation, building on our core characterizations of FLFP in India, further.

4.1. Access to Suitable Jobs As shown earlier, there is a significant mismatch in the composition of female jobs and the job preferences of out-of-labor force women who are willing to work. In addition, out-of-labor force women express a willingness to participate in market work, but women spend a longer time searching for jobs. The types of jobs women are willing to take are likely correlated Erin K. Fletcher et al. 183 with their life stage (married or not), geographic location, and education, but the general need to identify ways for them to access jobs they will take prevails. Overall, women (especially married women) prefer regular work—particularly regular part-time work—but few women working are in part-time jobs. Several areas of research could shed light on how to help women access jobs they are willing to undertake. First, job search costs are likely higher for women than for men, but more research is needed to understand the dimensions of that search. The literature suggests that access to information about jobs is a constraint and social norms often dictate that women spend much of their time engaged in domestic duties rather than looking for work. Norms may also restrict network size for women. More efficient search could be achieved through increased information about job opportunities. Further research should focus on understanding how to ensure women have information about jobs that helps them more efficiently match to jobs. Second, women out of the labor force who want work overwhelmingly say they would prefer regular part-time work. More research is needed to understand how policies or market forces that increase the availability of part-time or flexible work arrangements could incentivize greater female participation. More work is needed to connect the desire for part-time work to women’s time use, and subsequently how to promote socially acceptable, flexible childcare arrangements for working women to allow for labor market participation. Support for women’s self-employment, whether through more appropriate financing or training, would also likely suit many women, given the demands on their time in the household. An obvious policy linkage here is to the government’s National Rural Livelihoods Mission, which supports self-help groups (SHGs) and aims to eventually connect women’s groups to flexible work opportunities convenient to the groups. Other major initiatives, such as the Self Employed Women’s Association (SEWA), already support similar initiatives, with success. Again, women’s demographic characteristics matter: age and marital status are important predictors of labor force attachment. Our analysis suggests that marriage is a more significant correlate of women’s lower LFP than childbearing, and younger, out-of-labor force women with expressed willingness to work are more likely to prefer full-time work. Work opportunities have been shown to delay marriage, but there is little evidence on how to incentivize labor market attachment to persist post-marriage. Incentivizing full-time opportunities for younger, unmarried women is one testable solution; further research should explore how pre-marriage career experience affects post-marriage labor market decisions. 184 INDIA POLICY FORUM, 2018

While women may prefer part-time work in an unconstrained environment, it is also possible that particular technologies or costs restrict the choice set upon which they optimize. For example, women may state a preference for part-time work because their household duties require they spend hours cooking each day, searching for firewood, or even retrieving water. Technology relevant to household production has been relevant to increasing women’s employment in other settings (e.g., Dinkelman [2011] for electrification in South Africa). Additional research on how technologies can reduce time burdens on women in India may be useful. The extent to which environmental degradation may contribute to time poverty relevant to women’s labor force decisions is also an important area for study. A similarly important example relates to women’s actual and perceived safety: women may report preferring jobs close to home not simply because they enjoy short commutes but also because they and family members are concerned about their safety if they venture far from home. Recent work has highlighted that young women in India are willing to incur higher costs (and lower education gains) for higher safety (Borker 2018). Rigorous studies diving further into these issues are all likely going to be important in the coming years.

4.2. Government Priorities: Quotas, Investments in Skills and Manufacturing, and Income Transfers The Government of India has recently committed to increased investments in skills training, to promoting manufacturing employment, and to additional gender-based quotas in areas from police forces to corporate boards. These commitments, combined with our diagnostics and literature review, suggest they are fruitful areas for rigorous pilots and evaluations to better understand how they can support women’s economic activities. The scope for improving skills and vocational training is significant. Many skills and vocational programs have been shown to be relatively inef- fective (Blattman and Ralston 2015; McKenzie 2017); in India, some of us found that only one-fifth of trainees are employed one year after training in a major skills scheme in India (Prillaman et al. 2017). That said, the potential for such programs to support women, in particular, is high: many government-funded programs have gender quotas, and some programs incentivize placement and retention in a first job after training, which could serve as a crucial linkage connecting women to jobs. Our diagnostics show that women with skills training are more likely to be employed. Given concerns over Erin K. Fletcher et al. 185 selection into training, research that examines the causal impact of training on labor market outcomes, as well as studies focused on how programs can help women overcome search frictions may be useful. A desire for more training by out-of-labor force women also suggests that supporting training for women seeking non-traditional (part-time, and potentially home-based) work is an important area for further study. In addition, manufacturing employment for women has grown over the past ten years despite its generally slow overall employment growth (Nayyar 2009; Prillaman and Moore 2016), with women occupying 25 percent of manufacturing positions by 2012. An expansion of manufacturing employment may be particularly important in rural areas. As employment in agriculture is declining and an increasingly educated workforce lacks access to jobs, sector-specific investments to improve job quality and availability could benefit women. Here, research to better understand the factors driving wage gaps, and potential ways to level the playing field, are warranted. Although the literature on quotas provides solid evidence on how increasing women’s political representation can benefit women and girls, questions remain on whether and how employment quotas can help women. For instance, should they be applied universally or only to certain fields, are there associated negative externalities, and are quotas strictly better than other policies aimed to increase FLFP? We suggest better evaluation of gender-based employment quotas that are already in place, such as those associated with the national welfare scheme, MGNREGS, and Operation Blackboard20 as well as more rigorous comparisons to alternate policies. Finally, since discrimination may also play a significant role in FLFP—both in discouraging women from applying for jobs, and from obtaining jobs they apply to—quotas have the potential to put more women in visible positions and possibly change social norms around women and work. There may also be important opportunities for the government itself to provide more women, particularly those with relatively higher levels of education, with access to suitable jobs in their own communities while con- ferring the additional benefit of improved service delivery (Muralidharan

20. To our knowledge, there has only been one evaluation of Operation Blackboard’s policies, but it did not specifically address the quota. Chin (2005) shows that primary school completion rates improved for girls under Operation Blackboard, despite no significant changes in class size or number of teachers. Although we cannot attribute the effect on schooling directly to the quota and Chin offers no estimation of effects on female employment, we can take this as prima facie evidence that the program—including the quota—was important and should be evaluated in more depth. 186 INDIA POLICY FORUM, 2018

2016). Frontline public sector workers in health and nutrition, for example, are overwhelmingly women, and yet evidence suggests these workers are overburdened and generally understaffed (Kapur et al. 2017; Muralidharan 2016). Hiring more frontline workers in health, nutrition, education, and other important community services may be an important way to legitimize women’s work and increase FLFP. Beyond this, expanding public childcare seems an important avenue to increase women’s employment while providing other women with greater flexibility to participate in income-generating opportunities. A final area that has seen increasing attention is that of income transfers from the government to citizens, most recently in the form of a Minimum Income Guarantee or Universal Basic Income. The impacts of such a benefit directed to women are theoretically ambiguous. For example, although a transfer directed to women could compensate them for unpaid work in the household, it could also lead working women to decrease their labor supply (due to the income effect) or drop out of the labor force entirely. On the other hand, if women want to work outside the home, directly paying them in ways that allow them to access and control these funds may increase their intra-household bargaining power and help them negotiate within households to enter the labor force. The income could also be useful to investing in training or capital that likely deter women from self-employment or other economic activities. Making the transfers conditional on earning less than a certain amount of income, however, would likely suppress their labor supply. All this suggests that any direct transfers, whether directly for women or to their households, should be carefully designed and tested to understand their impact on women’s labor supply (on this, also see Field et al. 2019).

4.3. Data Collection and Transparency A major limiting factor to better understand the reasons for India’s low FLFP is lack of up-to-date data. Additional data collection through more regular employment surveys would be particularly valuable. More regular surveys, as are now undertaken in the Periodic Labour Force Survey, will help policymakers adjust programs and policies quickly in response to economic shocks. They can also help increase understanding of anomalies in the data, such as the uptick in India’s FLFP in 2004 and its subsequent decline, the cause for which remains unresolved in the literature. In addition, time-use surveys would identify how India’s 200 million women engaged primarily in domestic activities spend their days and clarify the extent to which they may already be involved in labor market activities. Erin K. Fletcher et al. 187

They would also help reconcile large discrepancies in FLFP as measured by different household surveys and would prove constructive to analysis of gender dynamics in household activities, if collected for several members of the same household. India is positioned to collect quality time-use data due to the lessons from a 1998 pilot of six Indian states and recent announcements by the government to implement such exercises. States and the Central Government can also play a role in coordinating data collection by trainers and employers involved in major employment- oriented initiatives mentioned earlier. Ensuring both requisite technological infrastructure, as well as appropriate incentives, are in place to collect high- quality data is an important step toward better understanding FLFP and how women can fit into the “Skill India” and “Make in India” programs. The government can also do more to systematically collect and track both short-term economic migration and contract labor, both of which involve women (and possibly increasingly so), but around which data collection is extremely limited, particularly in terms of gender disaggregation. Finally, in cases when data are collected—through both surveys and administrative data systems—promoting and incentivizing data sharing and transparency will facilitate a study of these important topics.

5. Conclusion

Despite increases in education, declines in fertility, and strong economic growth, India’s FLFP has declined over the recent years and overall is quite low for India’s income levels, suggesting that action is necessary to increase women’s labor market participation and attachment. The micro and macroeconomic implications of India’s low and declining FLFP are at once adverse and consequential, and must be better understood and addressed. Our simple descriptive analysis of NSS data points to significant constraints on FLFP driven by both social and economic factors on the supply and demand side. Many women counted out of the labor force and primarily occupied with domestic duties say they want not simply to work, but to work in a regular job. Further evidence suggests women search less, or less efficiently, for jobs even as they face greater discrimination in the marketplace. Many women additionally lack the skills required to undertake work they would like. Although skills training may be able to address this constraint, more research is needed to better understand how women can best benefit from the government’s current investments in skilling. 188 INDIA POLICY FORUM, 2018

Indian women also tend to opt out of the labor market at marriage, losing high-potential early career earnings and experience that may be important for their socioeconomic trajectories. Once in jobs, women are also often at a disadvantage: in fields where women enjoy higher relative representation, pay is less equitable across men and women. Yet some fields with important female-friendly measures, including quotas, equal pay, and work close to women’s homes, have successfully attracted female workers. The specific features driving this relative success in FLFP need to be better understood. In addition to undertaking research focused on the challenges outlined here, a key step to improve our understanding of how to increase women’s economic engagement is to increase the frequency of data collected about Indian women’s economic activities and time use, to improve data collected relevant to government initiatives that can influence FLFP, and to ensure that data are released regularly and transparently. Over the past several years, a growing set of researchers have turned their attention to India’s low, and apparently declining, FLFP. This trend is promising, but much more needs to be done to spur rigorous innovations in both the public and private sectors to increase women’s economic engagement. Finally, although this paper focuses on constraints and potential strategies to increase FLFP in India, it goes without saying that the goal of increasing this outcome is to improve women’s welfare overall. Women’s perceived welfare reflects a variety of factors, of which economic engagement is one factor among many. Any policies that aim to increase women’s economic engagement should aim to measure changes beyond simply LFP, to better understand their implications for welfare of women and their household members.

References

Afridi, F., T. Dinkelman, and K. Mahajan 2018. “Why Are Fewer Married Women Joining the Work Force in Rural India? A Decomposition Analysis over Two Decades,” Journal of Population Economics, 31(3, No. 4): 783–818. Alesina, A. F., P. Giuliano, and N. Nunn. 2013. “On the Origins of Gender Roles: Women and the Plough,” The Quarterly Journal of Economics, 128(2): 469–530. Andrabi, T., J. Das, and A. I. Khwaja. 2013. “Students Today, Teachers Tomorrow: Identifying Constraints on the Provision of Education,” Journal of Public Economics, 100: 1–14. Ball Cooper, L., E. L. Paluck, and E. K. Fletcher. 2012. “Reducing Gender-Based Violence,” in M. Ryan and N. Branscombe (eds.), Sage Handbook on Gender and Psychology, pp. 359–377. London: SAGE Publications. Erin K. Fletcher et al. 189

Bandiera, O., R. Burgess, S. Gulesci, and I. Rasul. 2009. “Community Networks and Poverty Reduction Programmes: Evidence from Bangladesh,” LSE Research Online Documents on Economics 58054, London: London School of Economics and Political Science. Banerjee, A., E. Duflo, R. Chattopadhyay, and J. Shapiro. 2011. “Targeting the Hard-Core Poor: An Impact Assessment,” Working Paper. Cambridge, MA: Massachusetts Institute of Technology. Available at http://economics.mit.edu/ files/6645 (accessed on 29 June, 2019). Bardhan, Pranab. 1984. Land, Labor, and Rural Poverty. Essays in Development Economics. Delhi: Oxford University Press and New York: Columbia University Press. Beam, E. 2016. “Do Job Fairs Matter? Experimental Evidence on the Impact of Job-Fair Attendance,” Journal of Development Economics, 120(C): 32–40. Beaman, L., R. Chattopadhyay, E. Duflo, R. Pande, and P. Topalova. 2009. “Powerful Women: Does Exposure Reduce Bias?” Quarterly Journal of Economics, 124(4): 1497–1540. Bhalotra, S. R. 1998, February. “The Puzzle of Jobless Growth in Indian Manufacturing,” Oxford Bulletin of Economics and Statistics, 60(1): 5–32. Bhavnani, R. R. 2009, March. “Do Electoral Quotas Work after They Are Withdrawn? Evidence from a Natural Experiment in India,” American Political Science Review, 103(01): 23. Blattman, C., and L. Ralston. 2015. “Generating Employment in Poor and Fragile States: Evidence from Labor Market and Entrepreneurship Programs,” Available at SSRN: https://ssrn.com/abstract=2622220 or http://dx.doi.org/10.2139/ ssrn.2622220 (accessed on 29 June, 2019). Blinder, A. S. 1973. “Wage Discrimination: Reduced Form and Structural Estimates,” The Journal of Human Resources, 8(4): 436–455. Borker, Girija. 2018. “Safety First: Perceived Risk of Street Harassment and Educational Choices of Women,” Working Paper, Available at https://giri- jaborker.files.wordpress.com/2017/11/borker_jmp.pdf (accessed on 29 June, 2019). Bose, N., and S. Das. 2018. “Political Reservation for Women and Delivery of Public Works Program,” Review of Development Economics, 22(1): 203–219. Carranza, E. 2014. “Soil Endowments, Female Labor Force Participation, and the Demographic Deficit of Women in India,” American Economic Journal: Applied Economics 6(4): 197–225. Chatterjee, Urmila, Rinku Murgai, and Martin Rama. 2015. “Job Opportunities along the Rural–Urban Gradation and Female Labor Force Participation in India,” Policy Research Working Paper 7412, Washington, D.C.: World Bank. Chin, A. 2005. “Can Redistributing Teachers across Schools Raise Educational Attainment? Evidence from Operation Blackboard in India,” Journal of Development Economics, 78(2): 384–405. Chowdhury, S. 2011. “Employment in India: What Does the Latest Data Show?” Economic and Political Weekly, 46(32): 23–26. 190 INDIA POLICY FORUM, 2018

De Mel, S., D. McKenzie, and C. Woodruff. 2014. “Business Training and Female Enterprise Start-Up, Growth, and Dynamics: Experimental Evidence from Sri Lanka,” Journal of Development Economics, 106(C): 199–210. Deininger, K., H. K. Nagarajan, and S. K. Singh. 2016. Short-Term Effects of India’s Employment Guarantee Program on Labor Markets and Agricultural Productivity. Policy Research Working Paper No. WPS 7665, Washington, D.C.: World Bank. Dinkelman, Taryn. 2011. The Effects of Rural Electrification on Employment: New Evidence from South Africa. The American Economic Review, 101(7): 3078–3108. Esteve-Volart, B. 2004, January. Gender Discrimination and Growth: Theory and Evidence from India,” DEDPS 42, London: Suntory and Toyota International Centres for Economics and Related Disciplines, London School of Economics and Political Science. Field, E., S. Jayachandran, R. Pande, D. Mel, and D. Mckenzie. 2013. “Do Traditional Institutions Constrain Female Entrepreneurship? A Field Experiment on Business Training in India,” American Economic Review, 103(6): 2196–2226. Field, E., S. Jayachandran, R. Pande, and N. Rigol. 2016. “Friendship at Work: Can Peer Effects Catalyze Female Entrepreneurship?” American Economic Journal: Economic Policy, 8(2): 125–153. Field, E., R. Pande, N. Rigol, S. Schaner, and C. Troyer Moore. 2019. “On Her Own Account: Experimental Evidence on How Strengthening Women’s Financial Control Changes Work Choices, Beliefs and Norms,” Working Paper. J-PAL, Cambridge, MA: Massachusetts Institute of Technology. Fisher, R. J. 1993. “Social Desirability Bias and the Validity of Indirect Questioning,” Journal of Consumer Research, 20(2): 303–315. Ghani, E., W. R. Kerr, and S. D. O’Connell. 2014, May. “Political Reservations and Women’s Entrepreneurship in India,” Journal of Development Economics, 108: 138–153. Goldin, C. 1995. “The U-Shaped Female Labor Force Function in Economic Development and Economic History,” in T. P. Schultz (Eds.), Women’s Human Capital and Economic Development, pp. 61–90. Chicago: University of Chicago Press. Gupta, A. 2014. “Effect of Trade Liberalization on Gender Inequality: The Case of India,” Working Paper, Chennai: Indian Statistical Institute. Heath, R., and A. M. Mobarak. 2015. “Manufacturing Growth and the Lives of Bangladesh Women,” Journal of Development Economics, 115: 1–15. Heath, R., and X. Tan. 2014. “Intrahousehold Bargaining, Female Autonomy, and Labor Supply: Theory and Evidence from India,” Working Paper. Available at https://dl.dropboxusercontent.com/u/12277691/HSA_HeathTan.pdf (accessed on 28 June, 2019). Hsieh, C. T., E. Hurst, C. I. Jones, and P. J. Klenow. 2013. “The Allocation of Talent and U.S. Economic Growth,” NBER Working Paper No. 1869, Cambridge, MA: National Bureau of Economic Research. Erin K. Fletcher et al. 191

Jensen, R. 2012. “Do Labor Market Opportunities Affect Young Women’s Work and Family Decisions? Experimental Evidence from India,” The Quarterly Journal of Economics, 127(2): 753–792. Kalsi, P. 2017. “Seeing Is Believing—Can Increasing the Number of Female Leaders Reduce Sex Selection in Rural India?” Journal of Development Economics, 126(C): 1–18. Kannan, K., and G. Raveendran. 2009. “Growth Sans Employment: A Quarter Century of Jobless Growth in India’s Organised Manufacturing,” Economic and Political Weekly, 44(10): 80–91. Kapsos, S., A. Silberman, and E. Bourmpoula. 2014. “Why Is Female Labour Force Participation Declining So Sharply in India?” ILO Research Paper No. 10, Geneva: International Labour Office. Kapur, D., P. B. Mehta, and M. Vaishnav. (Eds.) 2017. “Introduction,” Rethinking Public Institutions in India, pp. 1–39. New Delhi: Oxford University Press. Kawata, H., and S. Naganuma. 2010. “Labor Force Participation Rate in Japan,” Bank of Japan Working Paper Series, Tokyo: Bank of Japan Review 2010–J18. Klasen, S., and J. Pieters. 2015. “What Explains the Stagnation of Female Labor Force Participation in Urban India?” The World Bank Economic Review, 29(3): 449–478. Lee, S., J. Cho, S. Park, and S. S. Lee. 2013. “It’s More Than an M-Shape: The Political Economy of Female Non-Standard Workers in the Republic of Korea,” Asian Social Work and Policy Review, 7(1): 1–17. McKenzie, D. 2017. “How Effective Are Active Labor Market Policies in Developing Countries? A Critical Review of Recent Evidence,” IZA Discussion Paper Series 10655, Bonn: IZA Institute of Labor Economics. Muralidharan, Karthik. 2015–16. “New Approach to Public Sector Hiring in India for Improved Service Delivery,” India Policy Forum, 13: 187–236, New Delhi: National Council of Applied Economic Research. Nayyar, G. 2009, September. “The Nature of Employment in India’s Services Sector: Exploring the Heterogeneity,” Discussion Paper No. 452, Oxford: Department of Economics, University of Oxford. Neff, D., K. Sen, and V. Kling. 2012. “Puzzling Decline in Rural Women’s Labor Force Participation in India: A Re-examination,” Working Paper No. 196. Hamburg: German Institute of Global and Area Studies. Oaxaca, R. 1973. “Male–Female Wage Differentials in Urban Labor Markets,” International Economic Review, 14(3): 693–709. Panda, P., and B. Agarwal. 2005, May. “Marital Violence, Human Development and Women’s Property Status in India,” World Development, 33(5): 823–850. Pande, R., D. Ford, and E. K. Fletcher. 2015. “Female Labor Force Participation in Asia,” Working Paper. Cambridge, MA: Center for International Development, Harvard University. Papola, T., and P. P. Sahu. 2012. “Growth and Structure of Employment in India: Long-Term and Post-Reform Performance and the Emerging Challenge,” 192 INDIA POLICY FORUM, 2018

New Delhi: Institute for Studies in International Development, Available at http:// isid.org.in/pdf/ICSSR_TSP_PPS.pdf (accessed on 29 June, 2019). Prillaman, S. A., and C. Troyer Moore. 2016, February. “Skill India and Make in India: Can They Empower India’s Women?” Cambridge, MA: J-PAL and Evidence for Policy Design, Harvard University. Prillaman, S. A., R. Pande, V. Singh, and C. Troyer Moore. 2017. “What Constrains Young Indian Women’s Labor Force Participation? Evidence from a Survey of Vocational Trainees,” Technical Report, Cambridge, MA: J-PAL and Evidence for Policy Design, Harvard University Qian, N. 2008, August. “Missing Women and the Price of Tea in China: The Effect of Sex-Specific Earnings on Sex Imbalance,” Quarterly Journal of Economics, 123(3): 1251–1285. Rustagi, P. 2010. “Changing Patterns of Labour Force Participation and Employment of Women in India,” The Indian Journal of Labour Economics, 56(2): 215–241. Sivasankaran, A. 2014. “Work and Women’s Marriage, Fertility and Empowerment: Evidence from Textile Mill Employment,” Working Paper, Cambridge, MA: Harvard University. Sudarshan, R. M. 2014. “Enabling Women’s Work,” ILO Asia-Pacific Working Paper Series, New Delhi: DWT for South Asia and Country Office for India International Labour Organization. Sudarshan, R. M., and S. Bhattacharya. 2009. “Through the Magnifying Glass: Women’s Work and Labour Force Participation in Urban Delhi,” Economic and Political Weekly, 44(48): 59–66. Swaminathan, H., R. Lahoti, and J. Suchitra. 2012, June. “Women’s Property, Mobility, and Decision-Making: Evidence from Rural Karnataka, India,” IFPRI Discussion Paper No. 01188. Washington, D.C.: International Food and Policy Research Institute (IFPRI). UNESCO. 2015. “Education for All 2000–2015: Achievements and Challenges.” EFA Global Monitoring Report. Available at https://en.unesco.org/gem-report/ report/2015/education-all-2000-2015-achievements-and-challenges (accessed on 28 June, 2019). World Bank. 2014a. “Fertility Rate, Total (Births per Woman)”. World Bank Data. Available at https://data.worldbank.org/indicator/sp.dyn.tfrt.in (accessed on 28 June, 2019). ———. 2014b. “Labor Force Participation Rate, Female (% of Female Population Ages 15+) (Modeled ILO Estimate),” World Bank Data. Available at https:// data.worldbank.org/indicator/SL.TLF.CACT.ZS (accessed on 28 June, 2019). ———. 2018. “Urban Population (% of Total),” World Bank Data. Available at https://data.worldbank.org/indicator/sp.urb.totl.in.zs (accessed on 28 June, 2019). Erin K. Fletcher et al. 193

Appendix

TABLE A.1. This table maps the original NIC codes to the condensed codes used in this paper. Condensed version Original NIC code Accommodation Accommodation Advertising, market research Advertising & market research Agriculture Crop & animal prod., hunting & related service activities Arts/Entertainment/Sports Sports Act. & Amusement & Recreation Act. Creative arts & entertainment activities Chemical/biological/metal Manufacture of other non-metallic mineral products manufacturing Manufacture of coke & refined petroleum products Manufacture of pharmaceuticals, medicinal, chemical, & botanical products Manufacture of rubber & plastic products Manufacture of chemical & chemical products Manufacture of basic metals Manufacture of paper & paper products Manufacture of metal products, except machinery & equipment Civil engineering, architecture, Architecture & engineering act., tech. testing & analysis tech testing, analysis Civil engineering Computer programming Computer prog., consultancy & related act. Construction Specialized const. activities Construction of buildings Consulting Act. of head offices mgt. Consultancy act. Domestic personnel/household use Act. of households as employers of domestic personnel Education education Electricity, gas, AC supply Electricity, gas, steam, & air condition supply Electronic manufacturing Manufacture of computers, electronic & optical products manufacture of electrical equipment Employment acts/office support Employment activities Office administrative, office support & other business support act. Equipment repair Repair & installation of machinery equipment Equipment/vehicle manufacturing Manufacture of motor vehicles, trailers, & semi-trailers Manufacturing of other transport equipment Manufacture of machinery & equipment N.E.C. Financial/info services Other financial activities Information service activities Financial service act. except insurance & pension funding Food manufacturing Manufacture of food products Manufacture of beverages Manufacture of tobacco products Food service Food & beverage service activities Forestry/fishing Fishing & aquaculture Forestry & logging Gambling Gambling & betting act. Human health activities Human health act. Insurance, pensions Insurance, reinsurance, & pension funding except compulsory social security Legal, accounting Legal & accounting activities (Table A.1. Continued) 194 INDIA POLICY FORUM, 2018

(Table A.1. Continued)

Condensed version Original NIC code Libraries Libraries, archives museums, & other cultural act. Media production Printing & reproduction of recorded media Mining Mining of coal & lignite Mining of metal ores Extraction of crude petrol. & natural gas Other mining & quarrying Mining support service activities Other Act. of extra territorial org. & bodies Activities of membership org. Other home repair/services Other personal service act. Repair of computers & personal & household goods Other manufacturing Other manufacturing Other science/tech Other prof. scientific & tech. activities Postal/courier Postal & courier activities Public administration/defense Public admin. & defense, compulsory social security Publishing/media Program & broadcasting activities Publishing activities Motion picture/video & TV prog. prod and related activities Real estate Rental & leasing act. Real estate act. Research Scientific research development Residential care, social work Residential care activities Social work act. without accommodation Retail trade Retail trade, except of motor vehicles & motorcycles Security/building services Services to buildings & landscape act. Security & investigation activities Telecoms Telecommunications Textile manufacturing Tanning & dressing of leather and manufacturing of related stuffs Manufacture of wearing apparel Manufacture of textiles Trade/repair vehicles Wholesale & retail trade, repair of motor vehicles & motorcycles Transport Air transport Land transport & transport via pipelines Warehousing & support activities for transportation Water transport Travel/tours Travel agency, tour operator, & other reservation service act. Veterinary Veterinary act. Waste management Remediation act. & other waste management services Waste collection, treatment & disposal act. material recovery Sewerage Water collection/supply/ Water collection, treatment, and supply treatment Wholesale trade Wholesale trade, except of motor vehicles & motorcycles Wood manufacturing Manufacturing and prod. of wood except furniture and other related items Manufacturing of furniture Source: NIC Codes based on the National Industrial Classification (2008), Central Statistical Organisation, Ministry of Statistics and Programme Implementation, Government of India. TABLE A.2. This table presents the results of a review of top-tier journals in economics, including both general interest and field journals, and academic working papers over the years 2004–17. We include only papers with strong causal identification strategies such as a natural experiment caused by a policy change or a randomized control trial. Paper Area of Study Context Strategy for Assessing Impact LFP Estimate A. Information and Job Location Jensen (2012) North India Information provision on RCT: Compare FLFP in villages exposed to Women in villages visited by recruiters were 4.6 (Haryana) job opportunities recruiters for business process outsourcing percentage points more likely to be employed in jobs. the BPO sector and 2.4 percentage points higher overall. Heath and Bangladesh Location of textile Natural experiment: Compare women on Women in close proximity to garment factories Mobarak manufacturing firms the basis of proximity to garment factories. were 6.5 to 15.4 percentage points more likely to (2015) be employed. Sivasankaran South India The role of longer Natural experiment: Compare outcomes on An additional month of contract length increased (2014) (Tamil Nadu) duration-work contracts the basis of exposure to wage and contract length of employment by 0.5 months. policies. Andrabi et al. Pakistan The role of primary and Natural experiment: Compare teacher Areas with government schools were 20 to 27 (2013) secondary education in jobs in areas where schools were built to, percentage points more likely to have a private determining skill profiles where they were not built to see effects on school, which employs, on average, four women. job opportunities for women. Afridi et al. India Increasing education Parametric and nonparametric Changes in women’s education over time explain (2018) level in rural areas decomposition using Blinder (1973) and about 21.8% of the total decline in FLFP. Oaxaca’s (1973) technique to decompose Women’s own education and that of the men in the change in employment rates of women their household accounts for between 87% and over time on the basis of the data from 95% of the overall decline in FLFP in 1987–99. employment and unemployment rounds of In the 1999–2009 decade, they explain 25–37% India’s of the total decline in women’s LFPR. In both NSS in 1987–88, 1999–2000 and decades, education is the largest contributor to 2009–10. the decline in women’s LFPR. (Table A.2. Continued) (Table A.2. Continued)

Paper Area of Study Context Strategy for Assessing Impact LFP Estimate Beam (2016) Philippines Job fair Randomized encouragement design: Attending the job fair causes a 10.6-percentage (Sorsogon Measure the impact of attending a job fair point increase in being employed in the formal Province) on employment outcomes. sector (pooled men and women). Attending the job fair increases likelihood of female being employed in informal sector by 11.4 percentage points and decreases likelihood of female being self-employed by 16.0 percentage points. B. Information via Quotas Beaman et al. East India Gender electoral quotas Natural experiment: Compare number of Women in villages that were twice reserved were (2009) (West Bengal) women in elected positions in villages 2.8–3.2 percentage points more more likely to exposed to female leader quotas. stand for office and 4.5–5.5 percentage points more likely to win. Bhavnani West India Gender electoral quotas Natural experiment: Compare number of Number of women standing for election was (2009) (Mumbai) women in elected positions in villages 120% (0.5 candidates to 1.1 candidates) higher exposed to female leader quotas. inward that were once reserved compared to never reserved. Ghani et al. India Gender electoral quotas Natural experiment: Compare number Women in exposed states were 39–52% more (2014) of women-owned small enterprises in likely to start own businesses. states exposed to female leader quotas at different times. Bose and Das Northern Workfare program Natural experiment: Compare women’s Number of female person-days worked under (2018) Indian (Uttar gender quotas employment in areas with political NGREGA was 6% higher in administrative units Pradesh) positions reserved for female leaders. with female leaders. Deininger et India Workfare program Panel data analysis: 4,000 panel Program increases wages both for male and al. (2016) gender quotas households in 232 villages from 17 Indian female participants and also brings a shift from states. farm to non-farm and salaried employment in female labor supply.

C. Control of Resources and the Ultra-Poor Heath and India Property and lifetime Natural experiment: Rollout of Hindu Women in treated group (Hindu and affected by Tan (2014) unearned income Succession Act varied exposure to female HSA) were 9.7-percentage points more likely to be control of assets by state and time. working, 5 percentage points more likely to work outside the home. Banerjee et al. East India Asset transfers and RCT: Compare small enterprise activity Recipient households increased work by one hour (2011) (West Bengal) small enterprise activity in households given productive asset per day. transfers-to those not receiving transfers. Bandiera et Bangladesh Asset transfers to RCT: Compare labor force activity by Increase in self-employment and quality of jobs al. (2009) ultra-poor women given asset transfers to those not among those women receiving transfers; 1% receiving transfers. increase in hours worked. D. Peer Effects Field et al. Western India Business training and RCT: Evaluate interaction between Women who received business training were (2013) (Ahmedabad) microcredit randomized business training and social 13-percentage points more likely to take out norms. loans. Field et al. Western India Business training, RCT: Evaluate effectiveness of business Women who received business training with a (2016) (Ahmedabad) microcredit, peer training when combined with existing friend increased working hours by 17% and were networks social networks. 5.3 percentage points more likely to take out a loan from SEWA. Carranza India Soil type Natural experiment: soil types vary by Women in areas with a 10-percentage point higher (2014) district. fraction of loamy to clayey soils is associated with a 5.1% decrease in FLFP as agricultural workers (1.5 percentage points of rural FLFP average). Paper Area of Study Context Strategy for Assessing Impact LFP Estimate Beam (2016) Philippines Job fair Randomized encouragement design: Attending the job fair causes a 10.6-percentage (Sorsogon Measure the impact of attending a job fair point increase in being employed in the formal Province) on employment outcomes. sector (pooled men and women). Attending the job fair increases likelihood of female being employed in informal sector by 11.4 percentage points and decreases likelihood of female being self-employed by 16.0 percentage points. B. Information via Quotas Beaman et al. East India Gender electoral quotas Natural experiment: Compare number of Women in villages that were twice reserved were (2009) (West Bengal) women in elected positions in villages 2.8–3.2 percentage points more more likely to exposed to female leader quotas. stand for office and 4.5–5.5 percentage points more likely to win. Bhavnani West India Gender electoral quotas Natural experiment: Compare number of Number of women standing for election was (2009) (Mumbai) women in elected positions in villages 120% (0.5 candidates to 1.1 candidates) higher exposed to female leader quotas. inward that were once reserved compared to never reserved. Ghani et al. India Gender electoral quotas Natural experiment: Compare number Women in exposed states were 39–52% more (2014) of women-owned small enterprises in likely to start own businesses. states exposed to female leader quotas at different times. Bose and Das Northern Workfare program Natural experiment: Compare women’s Number of female person-days worked under (2018) Indian (Uttar gender quotas employment in areas with political NGREGA was 6% higher in administrative units Pradesh) positions reserved for female leaders. with female leaders. Deininger et India Workfare program Panel data analysis: 4,000 panel Program increases wages both for male and al. (2016) gender quotas households in 232 villages from 17 Indian female participants and also brings a shift from states. farm to non-farm and salaried employment in female labor supply.

Paper Area of Study Context Strategy for Assessing Impact LFP Estimate De Mel et al. Sri Lanka Business training versus RCT: Evaluate the impact of (a) Existing business owners: Management (2014) business training + cash business training solely and practices improved in both interventions but grant business training coupled slightly higher in training + cash. Training with cash grant on existing only doesn’t improve business outcomes but business female owners and training + cash increases capital stock by potential start-ups. `10,000 and profits temporarily; (b) Potential start-ups: Training only increases business ownership rate by 12 percentage points and training + cash increases it by 29 percentage points points in the short run, both have no long- term impact. Training only increases work income by `1,494 (significant) and training + cash increases it by `697 (not significant). Note: Refer to Carranza (2014). The FLFP percentage estimate is determined by taking the percentage change in FLFP and dividing by the total FLFP in rural areas from the NSS. Comments and Discussion*

Pranab Bardhan University of California, Berkeley

In general, I agree with most of the points made in this paper about the characteristics and trends described and policy issues raised. My comments and suggestions below therefore are mainly supplementary and rather piecemeal.

• The paper simultaneously discusses both low and declining female labor force participation (FLFP)—the two aspects should be separated more clearly for analysis. The factors explaining them can be different. For example, gender norms or social expectations which may explain low FLFP may not be used as easily in explaining declining FLFP, even when those norms and expectations are not immutable. • Across countries, it is still not clear to me why India and Pakistan have such low FLFP, even as compared to their poorer South Asian neighbors such as Nepal and Bangladesh. Hindu or Muslim cultural norms in general are not enough to explain why India and Pakistan are so much of an outlier. • For the participation rate, NSS usual status data are used in the paper, but given the fragmentary and part-time nature of a great deal of women’s work, one should also make full use of the NSS current status data, where the reference period is the previous week. Detailed time disposition data are available for such current activities. • A statistical analysis of the number of days in work in the reference week (not just the number of women in work) can yield some valuable insights. For example, in my old work on a statistical analysis of the NSS household level data for rural West Bengal—reported in my book Land, Labor and Rural Poverty (1984)—I found the following demand side factors significant in explaining variations in the number of days in work:

a. Rainfall pattern in the area, with better rainfall areas having higher FLFP; b. Lean or busy season (even though imperfectly captured in the NSS sub-round variations), with women entering the current labor force in the busy season and withdrawing in the lean season; and c. A “discouraged, dropout” effect in seeking work, controlling for other factors, seen in households with more male members unemployed, where the number of days of female work participation was lower. • For examining the puzzle of the declining FLFP in NSS data in the face of education gains and fertility reduction, one should try to cross- check with the (scanty) panel data available, for example, IHDS data for 2004–05 and 2011–12. • Some additional explanations for the decline in FLFP worth examining are as follows: • With environmental degradation, collection activities mainly done by women, for example, of water and firewood, may take up increasing amounts of time in the day, leaving less time for “gainful” work. • With income and education improving, the same oppressive, dead-end, low-status jobs which women have been working on for generations are now less acceptable (this is an example of how declining FLFP can be welfare-improving). • In many lines of activity, with possibly worsening job prospects for the men in the family, the discouraged dropout effect on women may get stronger. • Mechanization of agricultural operations, particularly in female labor-intensive tasks such as harvesting, threshing, and food processing, among others, may be impacting the FLFP. • Perception of the increasing lack of safety for women in public places may be reducing FLFP. • On ecological factors like soil quality discussed in Section 3.4 in the paper, a related issue may be the particular crop grown. For example, cultivation and post-harvest operations for rice are more female labor- intensive than for, say, wheat. • On the adverse impact of trade liberalization on female employment discussed in Section 3.5, it may be less due to factory laws, and more due to the wiping out of low-productivity informal enterprises— which have more women workers—as a result of foreign competition (Nataraj 2011). • Regarding the explanation of why the gender wage gap is distinctly higher in sectors where more women are represented (see Figure 15 in Erin K. Fletcher et al. 201

the conference paper), could it be that in industries such as garments or bidi-making, where the majority of workers are women, men mostly do the supervisory–managerial work, and the gender wage gap partly reflects the wage gap between production and managerial work? • Here are a few brief suggestions on some additional policy issues: • The paper points to the latent female labor supply: large numbers of women currently in domestic work express willingness to work, but mostly for part-time work. As the NSS question on this suggests (see Footnote 14 in the conference paper), such part-time work has to be dovetailed with domestic work. Often the work, such as sew- ing, tailoring, animal husbandry, food processing, basket-making, and other handicrafts, may have to be brought home. There are special policy issues here involving credit, provision of supplies, marketing and transportation, organization of cooperatives and self-help groups, among others. • The idea of community kitchens (such as “amma canteens” in Tamil Nadu and “Indira canteens” in Karnataka) and community day- care centers needs to be tried on an all-India scale. An important special effect of this is not just on an adult woman’s outside work participation but also on the schooling of her elder daughter. • Extension services, specially oriented to women, located in nearby community centers or panchayat offices, is imperative, not just with respect to new production technology but also on information relating to job search.

References

Bardhan, P. 1984. Land, Labor and Rural Poverty. New Delhi: Oxford University Press. Nataraj, S. 2011. “The Impact of Trade Liberalization on Productivity: Evidence from India’s Formal and Informal Manufacturing Sectors,” Journal of International Economics, 85(2): 292–301, November.

Farzana Afridi Indian Statistical Institute

There has been a dramatic increase in women’s labor supply in the US and Europe since the beginning of the 20th century (Goldin 2006). Research has underlined the importance of increasing levels of education of women and falling fertility accompanied by more favorable gender wage ratios in raising 202 INDIA POLICY FORUM, 2018 women’s workforce participation. In contrast to the Western experience, female labor force participation (FLFP) in India has been low and either falling or stagnant for the last few decades, despite a decline in gender gaps in education, falling fertility, and high economic growth (Afridi et al. 2018). By some estimates, raising women’s participation in the economy to the same levels as men’s can raise GDP by as much as 27 percent (Lagarde and Solberg 2018). The paper by Fletcher, Pande, and Moore in this volume of the India Policy Forum is, therefore, not only timely but also imperative for finding policy solutions to address this issue. The authors provide an excellent and thorough summary of the issues surrounding FLFP in India. My comments focus primarily on the exposition of the observed levels and trends in FLFP in India with the objective of highlighting the policy measures that would be effective in addressing this issue. I, therefore, classify my comments and suggestions into two broad groups: (a) describing FLFP in India, and (b) policy recommendations emerging from the data analysis in the paper.

1. Describing FLFP in India

1.1 Distinction between Levels and Trends in FLFP The authors have used NSS survey data to describe the status of women’s work in India quite exhaustively. However, the paper’s data description tends to oscillate between the discussion of levels and trends in FLFP. I suggest that the authors distinguish between the two up front and clearly, since the policy prescriptions for addressing low levels and declining trends in FLFP may be quite different. To elaborate, FLFP in India has been historically low (Figure 1), lower in urban areas as opposed to rural areas. However, it is well documented now that while FLFP has declined in the recent decades in rural areas, in the urban areas, it has been mostly stagnant (Figure 1). While low levels of FLFP may be due to patriarchy and cultural factors that prevent women from working, the trends that suggest a decline may be related to the structural changes, or lack thereof, in the Indian economy. I elaborate on these points further, but the essential point is that the authors should clarify how their policy suggestions aim to address levels or trends, or both, while clearly distinguishing between the two.

1.2. Spatial and Sectoral Variation in FLFP The reasons for the significant difference in FLFP in rural and urban India are likely to vary spatially and so then would the policy prescriptions. Erin K. Fletcher et al. 203

FIGURE 1. LFP by Gender and Location (NSS) 91 8. .3 .4 .5 . 67. 12 0. Male Female 1987 1999 2011 Conf. interval (a) Rural 91 8. 7. .3 .4 .5 .6 12 0. Male Female 1987 1999 2011 Conf. interval (b) Urban

Source: Afridi, Dinkelman, and Mahajan (2018).

In rural areas, on average, FLFP is higher perhaps due to relatively greater destitution, which leads to higher willingness or need for women to work. In urban areas, there is less destitution, as a result of which social norms or stigmas against women’s work are likely to become more binding (a la Goldin). Moreover, the majority of the women in rural areas classify themselves as self-employed in agriculture (Figure 2), followed by those engaged 204 INDIA POLICY FORUM, 2018

FIGURE 2. Rural, Married FLFP by Sector (NSS)

LFPR (UPSS): Rural, Age 25–65

50.0%

40.0%

30.0%

20.0%

10.0%

0.0% 1987 1999 2009 Year

Agriculture Manufacturing Construction Service

Source: Author’s calculations. Note: UPSS = Usual Principal and Subsidiary Status in casual work. The predominant role of agriculture in employing women in rural areas, hence, cannot be ignored. Furthermore, we see that the decline in FLFP is almost completely due to a reduction in self-employment (Figure 3). Thus, any policy prescription for addressing FLFP should be conducted in the backdrop of the spatial and sectoral variations in both the levels and trends in women’s workforce participation in the country.

1.3. Women’s Demographic Characteristics In 2011, only 20 percent of rural, married women in the age group of 15–60 years were in the labor force, 30 percentage points lower than for unmarried women. While workforce participation rates among urban unmarried women went up by 11 percentage points between 1999 and 2011, the rate has remained stagnant for married women at 20 percent for the past 30 years (NSS, various years). I suggest that the authors distinguish between workforce participation by marital status as there is a large marriage pen- alty on women’s work in India (discussed further). I would also urge the authors to update the analysis to the 2014–15 NFHS to allow access to the latest data available on this issue, given that the last NSS survey data are available only until 2011. Erin K. Fletcher et al. 205

FIGURE 3. FLFP by Type of Employment (NSS)

LFPR (UPSS): Age 25–65 40% 35% 30% 25% 20% 15% 10% 5% 0% 1987 1999 2009

Year

Self-Employed Casual Salaried

Source: Author’s calculations. UPSS = Usual Principal and Subsidiary Status

To summarize, I suggest that the authors clearly distinguish the spatial and demographic variation in the observed levels and trends in women’s workforce participation over the last few decades in India. While their existing exposition is exhaustive, a more organized structure of the data descriptions would be easier for the reader to follow, and, more importantly, make it possible to distinguish between policy measures that target these constraints.

2. Policy Recommendations

I suggest that the authors classify the policy recommendations into two broad categories—those that address supply side or household factors and those that could loosen demand constraints to address economic and structural factors.

2.1. Supply Side Constraints • Emphasizing the role of cultural and social norms: Cultural norms underlying the traditional role of men and women in Indian households manifest themselves in the significantly greater time spent by women 206 INDIA POLICY FORUM, 2018

in home production than men, irrespective of their level of education and thereby potential wage earnings. This leads to a higher elasticity of labor supply for women relative to men, and low substitutability of female labor with male labor in home production. Using the only detailed time-use data available for India (NSS, Time Use Survey, 1998), Afridi, Bishnu, and Mahajan (2018) find that across education levels, women spend significantly less time at work than men (Figure 4). On average, 15–60 year-old married women in urban India spend a mere 9.36 hours at work per week, while their male counterparts spend 58.71 hours. As women go from being illiterate to completing higher secondary schooling, work hours show a declining trend and then jump up at the “graduate and above” level. Despite the rise in the time spent at work by women at the highest education level, the average weekly hours only reach 13.32. In contrast, across education levels, women spend significantly more time on domestic work than men (Figure 5). On average, 15–60 years old married women in urban India spend 51.85 hours on domestic work per week, while their male counterparts spend 4.18 hours. The average weekly hours of domestic work increase for women— although at a declining rate—up to the higher secondary schooling level and then fall by 14.25 percent at the “graduate and above” level but still remain at 47 hours per week. The domestic work hours of men do not vary significantly by education. While men and women spend comparable time on leisure, across education levels women spend significantly more time on childcare than men. When there are children below five years of age in the household, married women in urban India spend an average of 12.68 hours per week on childcare. The corresponding figure for men is 2 hours (Figure 6). When there are children below 14 years of age in the household, the figures are 9.91 and 1.73 for women and men, respectively (Figure 7). The gender gap in childcare hours does not vary significantly across levels of education in both cases. While the figures here are for urban areas, these conclusions hold for rural India as well. To summarize, women disproportionately bear the burden of domestic work in the household and hence face time scarcity. It appears that childcare, which is a large component of domestic work, is a key constraint on the FLFP, even for educated women. Therefore, to enhance the FLFP, increasing women’s education is perhaps not Erin K. Fletcher et al. 207

FIGURE 4. Time Spent on Work by Urban, Married Women and Men Aged 15–60 Years

Urban Women 15–60, married 10 15 20 25 30 35 40 45 50 55 60 05

Illiterate =G raduate Education level

Mean 95% Conf. interval

Urban Men 15–60, married 10 15 20 25 30 35 40 45 50 55 60 5 0 Illiterate =G raduate Work Mean 95% Conf. interval

Source: Afridi, Bishnu, and Mahajan (2018). 208 INDIA POLICY FORUM, 2018

FIGURE 5. Time Spent on Domestic Work by Urban, Married Women and Men Aged 15–60 Years

Urban Women 15–60, married 10 15 20 25 30 35 40 45 50 55 60 05 Illiterate =Graduate Domestic work

Mean 95% Conf. interval

Urban Men 15–60, married 10 15 20 25 30 35 40 45 50 55 60 05 Illiterate =Graduate Domestic work

Mean 95% Conf. interval

Source: Afridi, Bishnu, and Mahajan (2018). Erin K. Fletcher et al. 209

FIGURE 6. Time Spent on Childcare in Households with Under-5 Children, by Urban, Married Women and Men Aged 15–60 Years

Urban Women 15–60, married 10 15 20 25 30 35 40 45 50 55 60 05 Illiterate =Graduate Education level

Mean 95% Conf. interval

Urban Men 15–60, married 10 15 20 25 30 35 40 45 50 55 60 05 Illiterate =Graduate Education level

Mean 95% Conf. interval

Source: Afridi, Bishnu, and Mahajan (2018). 210 INDIA POLICY FORUM, 2018

FIGURE 7. Time Spent on Childcare in Households with Under-14 Children, by Urban, Married Women and Men Aged 15–60 Years

Urban Women 15–60, married 10 15 20 25 30 35 40 45 50 55 60 05 Illiterate =Graduate Education level

Mean 95% Conf. interval

Urban Men 15–60, married 10 15 20 25 30 35 40 45 50 55 60 05 Illiterate =Graduate Education level

Mean 95% Conf. interval

Source: Afridi, Bishnu, and Mahajan (2018). Erin K. Fletcher et al. 211

sufficient. Policy should focus on the provision of reliable and acces- sible childcare arrangements for working women. Further, flexible working conditions for women can enable them to balance work and home better while we simultaneously chip away at social norms that are very sticky and less responsive to policy. In addition, there are several policy initiatives that are more broadly aimed at poverty reduction, but which can also free women’s time from home production. For instance, technological changes that reduce women’s time in household chores (e.g., subsidized LPG, and improved and universal access to electricity) have been shown to release women’s time from home production (Dinkelman 2011). Such policies should be encouraged and emphasized from the perspective of women’s time scarcity. • Restrictions on women’s mobility: While the authors acknowledge the concerns regarding women’s safety, I would suggest they also empha- size the role of providing basic infrastructure to improve women’s access to and safety in public spaces. For instance, improvements in the frequency and quality of public transportation, street lighting and regular safety audits could significantly improve women’s mobility and thereby their workforce participation.

2.2. Demand Side Constraints In the current version of the paper, the authors have not elaborated on the possible demand side constraints that may have been a factor in the low level and declining trend in FLFP in India. I would encourage the authors to do so in their revised draft and consider the policy recommendations as follows.

• Create more good jobs: India needs policies that create good (read formal sector) jobs which women, even with relatively low levels of education, can engage in (e.g., in agriculture and manufacturing). This is linked to the issue of lack of jobs in general in India, and also to a special focus on the greater disadvantage women face in accessing formal sector jobs. Policies that aim at improving farmers’ access to new technology, credit and markets in agriculture, and fostering the growth of manufacturing are measures that are on the policy radar, but lack a gender lens. • Encourage flexible work hours and piece rate work for women: Several surveys suggest that women prefer work that allows them to balance household and work for pay. Encouraging employers to provide flexible work hours and/or piece rate work to women in the 212 INDIA POLICY FORUM, 2018

manufacturing sector would be one such measure that could increase women’s workforce participation. • Provide home-based work: Policies that bring work closer to women’s homes would address both their time poverty as well as safety concerns. Programs such as the NREGA in rural areas have been shown to increase FLFP wherever it has functioned well. In the manufacturing sector, contractors often provide factory-based work to women in residential units around the industrial towns. However, women are typically exploited in these transactions in terms of extremely poor remuneration for the work they do. • Reduce gender gap in wages and earnings: Gender gaps in wage earnings in India are well documented and typically higher than in developed countries. Along with lack of decent jobs, a prime factor in the perceived low returns to FLFP in India is the lower wages women receive for the same kind of work they do as men. Legal provisions that make gender discrimination untenable, and their effective enforcement, would be essential for long-run improvements in the perceived returns to women’s work.

Finally, I would suggest the authors exercise caution on the following:

• The authors interpret the response to the question on “willingness to work if made available at household” in the NSS as an unconditional statement of women’s willingness to work. In my opinion, since this question is conditional on availability of work close to home, it highlights the point I made previously about the constraints women face due to the gendered division of labor within the home. • The authors tend to interpret the observed relationship between vocational training and FLFP as causal. I would caution against this interpretation because women who chose to take up vocational training may already be predisposed to working. It is not obvious that vocational training per se improves their LFP. • The authors fleetingly suggest extending political quotas for women to jobs in the public sector. While political quotas at the local government level have had, unarguably, a benign effect on women’s political participation, it is neither clear that there are enough public sector jobs to go around for gender quotas nor is it certain that it would lead to a significant increase in FLFP given the supply side constraints discussed earlier. Erin K. Fletcher et al. 213

References

Afridi, Farzana, Taryn Dinkelman, and Kanika Mahajan. 2018. “Why Are Fewer Married Women Joining the Workforce in Rural India: A Decomposition Analysis over Two Decades,” Journal of Population Economics, 31(3): 783–818. Afridi, Farzana, Monisankar Bishnu, and Kanika Mahajan. 2018. “Home Production, Social Norms and Women’s Labor Supply in India.” Paper presented at the 13th Annual Conference on Economic Growth and Development at Indian Statistical Institute, Delhi. Dinkelman, Taryn. 2011. “The Effects of Rural Electrification on Employment: New Evidence from South Africa,” American Economic Review, 101(7): 3078–3108. Goldin, Claudia. 2006. “The Quiet Revolution That Transformed Women’s Employment, Education, and Family,” American Economic Review, 96(2): 1–21. Lagarde, Christine and Erna Solberg. 2018. “Why 2018 Must Be the Year for Women to Thrive,” Paper written for World Economic Forum Annual Meeting 2018, held at Davos-Klosters, Switzerland. January 23–26.

General Discussion

In response to Pranab Bardhan’s comment that it is hard to understand the trends in labor force participation (LFP) on the basis of norms, Dilip Mookherjee pointed to the stigma underlying males or in-laws in the family preferring the women not to work, but allowing them to do so if the household was very poor. But as the household’s earning or wealth improves, there is an income effect with the women pulling out of the labor force, which can lead to a declining labor force participation rate (LFPR). He also referred to his experiment in West Bengal giving loans to low-income households, which had led to women withdrawing from the labor market. But the time allocation data showed that the women were not spending more time in leisure or on household chores, but were running self-employed businesses. Was this income effect empowering women? The women said they preferred to be self-employed at home and to socialize with other women during work. However, it is possible that the norm effect was also kicking in, with the family dissuading the women from working outside the home, but not minding their running a business from home. It is hard to separate the two effects, and difficult to make welfare judgments about female empowerment by looking just at female labor force participation (FLFP). Devesh Kapur agreed with Pranab Bardhan and Farzana Afridi that the role of technology was under-emphasized in the paper, especially 214 INDIA POLICY FORUM, 2018 technological change that is displacing women from agriculture or construction and easing workloads through more readily available cooking fuels and kitchen implements. This should be empowering for women. Surjit Bhalla suggested that India had undergone a large expansion in female education, catching up with men in education, and the implication of that should be an increase in the female labor force participation rate (FLFPR). In his earlier work on the emerging middle class, he had also found a backward bending supply curve for women, possibly because of status or cultural reasons, since India and Pakistan are the outliers also in this IPF paper. He also suggested that over the next 20 years, the picture in the world would be about declining male labor force participation rates (MLFPRs). Abhijit Banerjee highlighted the need to be careful in using LFP as a welfare outcome for women: it is important to also consider who is making the choice to participate or not, and how that choice is being affected by policy interventions, for which there is usually no clear theory. Microcredit may often directly contribute to women being made to start a business to serve some particular goal of their husbands. He then referred to a RCT done in Mirzapur, Uttar Pradesh, which looked at self-improvement interventions for women. When women say that they want to work at home, are they also speaking about particular things they feel they can do, pointing to a lack of self-belief that they cannot go out and work, say, in a factory? Once the RCT self-efficacy treatments were done, women could be convinced that they could go out and participate in the labor force. Mihir Desai noted that the paper had missed the opportunity to use the district-level variation in the FLFP, which cannot be fully explained by the urban–rural gap, as done in the paper. He thought that MLFP was not nearly as variable at the district level. This presented an opportunity. Rohini Somanathan cautioned against getting hung up on LFPRs and jumping to issues of efficiency and welfare without thinking them through. She contended that if women’s wages were doubled, we could certainly get to much higher FLFPRs, but it is not clear at all if that would improve efficiency. Thinking this through requires consideration of which markets don’t work, why we may be in a bad equilibrium, and what the source of inefficiency is. For example, corresponding to the figure of 30 percent of women saying that they wanted to work as cited by the authors, she noted that it would be useful to know that number for men as well. Finally, she said that policy interventions may work for women at certain ages to increase participation and welfare, but women might be better off out of the labor force at other ages. Erin K. Fletcher et al. 215

Ratna Sudarshan praised the paper for its emphasis on women’s part-time work, noting that it was perhaps the first time she was seeing this emphasis. But she cautioned against this leading to gender-conflictive situations where existing work is merely being re-allocated. Instead, the imperative is to expand work and job opportunities for women, and upgrading work that is already being done. Among younger, more educated women, there is actually a search for newer, different types of part-time work. Second, she stressed the importance of developing a more realistic narrative in the paper about women’s work in India, one that brings family and marriage also into the picture and helps develop a sense of identity and purpose in women’s work. Part-time work is very central to that narrative, so the paper could develop that further. Renuka Sané maintained that safety is a very big issue when it comes to FLFP, but is not discussed enough. She related this issue with the comment on how and why women prefer to work closer to home. She advised caution in interpreting a policy recommendation out of that statement, remarking that instead of recommending bringing work closer to home for women, we should perhaps focus on making it safer for women to travel to work. Anushree Sinha noted that government policies like UJALA that promote electrification and infrastructure, in part to save women time for care work, may actually be reinforcing traditional gender roles. The argument is made that women now have facilities to save time for care work, which is their role anyway, and they can do market work in whatever residual time is saved. It is important to devise ways to facilitate sharing of both care work, outside work, and unpaid work between men and women, and to take into account more the choices that women actually want to make. Premila Nazareth shared work she had done for the International Finance Corporation on women’s participation in the Indian mining sector. Mining firms were keen to use women more at all levels, but for mining engineers there was a law that women were not permitted to work in underground mines. Furthermore, the website of the Ministry of Labour and Employment mentioned that women were not allowed to work at night on the shop floor in factories. So these two laws that were supposed to protect women actually were holding them back from participating in a fuller way in a key growth sector. The long-term result was that women who had trained in mining were instead going into IT, so that the top CEO-type jobs were going to men in mining and manufacturing. On a more individual level, there is a need to change the narrative of how we think about women’s work, which in India involves not only childbearing and child-rearing but also looking after the 216 INDIA POLICY FORUM, 2018 elderly. She appealed to economists to usher a change in the way we think about this and to bring women into the workforce. Rajnish Mehra was surprised that India’s and Pakistan’s FLFPR were so different from others, including neighboring countries. He asked if similar studies have been done on Bangladesh and Nepal, particularly studies comparing FLFP on two sides of the border. He also noted that India exhibited a lot of variation across states in fertility, agricultural productivity, education levels, and health outcomes. He wondered if this could be exploited to examine any systematic relationships about LFP. SAJJID Z. CHINOY* JPMorgan Chase TOSHI JAIN† JPMorgan Chase What Drives India’s Exports and What Explains the Recent Slowdown? New Evidence and Policy Implications§

ABSTRACT The role of exports in India’s growth dynamics over the last two decades has been consistently under-appreciated. India’s export surge in the mid-2000s was a key driver of the high GDP growth experienced during that time. Conversely, the sharp fall-off in exports growth (both gross exports and domestic value addition) over much of the last six years has been an important driver of India’s GDP growth performance in recent years compared to the mid-2000s. So why have Indian exports slumped in recent years? To ascertain this, we analyze the determinants of India’s exports and estimate their “income” and “exchange rate” elasticities during 2004–17. We find that (a) both global growth and exchange rates are important determinants of India’s export dynamics, (b) these elasticities have attenuated in recent years (consistent with de-globalization and lower domestic value addition in India’s export basket), and (c) there is a significant heterogeneity of elasticities across sectors, which explains the changing composition of India’s export basket. Against these results, our model explains the significant slowing of exports in recent years. In particular, we find that the sharp 20 percent appreciation of the real exchange rate between 2014 and 2017 has impinged on export competitiveness. We posit that the real appreciation was the inevitable upshot of the large, positive terms-of-trade shock that India experienced from lower oil prices, suggesting that India was afflicted by Dutch disease. Exports began to recover in 2018 in conjunction with some depreciation of India’s real effective exchange rate (REER).

* [email protected] † [email protected] § This paper was presented at the India Policy Forum in July 2018. In subsequent months, a back-casted GDP series for the previous decade was released. However, only annual data are available in the back-casted series. Since the analysis in this paper is based on quarterly data, the authors continue to use the old series. To be sure, all recent revisions to the new GDP series have been incorporated. Finally, use of either series does not change the qualitative results and implications of the paper. The authors gratefully acknowledge comments from participants at the NCAER 2018 India Policy Forum, particularly from the discussants, Surjit Bhalla and Kenneth Kletzer. 217 218 INDIA POLICY FORUM, 2018

Assuming oil prices continue rising and the positive terms-of-trade shock continues to reverse, the equilibrium real exchange rate will depreciate further and help improve competitiveness. Policymakers must not fight such a depreciation, but ensure that it is calibrated and not disruptive. Fiscal policy must not become expansive and offset this real depreciation. More fundamentally, India should pursue supply side reforms to improve its current account deficit, and, with export prospects dimming, seek other growth drivers for the foreseeable future.

Keywords: India, Exports, Exchange Rate, Income Elasticity, Price Elasticity, Dutch Disease

JEL Classification: F10, F14, F32

1. Introduction and Motivation

ver the last decade, there has been a growing appreciation of India’s Ofinancial integration with the rest of the world. Foreign direct investment (FDI) levels have increased in the recent years and are a major source of financing for the economy’s current account deficit (CAD). Similarly, foreign portfolio participation into the equity and debt markets has progressively increased and has provided much-needed liquidity and a more diversified investor base. More generally, asset prices in India are increasingly correlated with global asset prices—the ultimate manifestation of India’s progressive financial integration. Paradoxically, however, this integration is most often appreciated during adverse shocks. It is in the “sudden stops” or “sudden outflows” of capital that India has episodically witnessed (the Global Financial Crisis in 2008, the European sovereign debt crisis in 2010–11, the taper tantrum in 2013, during each of which the exchange rate came under pressure and domestic financial conditions tightened) that the full extent of India’s financial integration with the world has been truly appreciated. In contrast, there is much less appreciation of India’s global integration on the real economy side. There is still a perception that India’s economic prospects are governed by its large domestic market, and that trade—both exports and imports—are peripheral to growth prospects. We believe that to harbor this perception, however, is to live in the old reality. The shares of exports and imports have risen materially as a share of GDP over the last two decades. Total exports/GDP doubled from 11.3 percent of GDP in 1999 to 25.4 percent of GDP by 2013. Since then, exports have slowed against the backdrop of de-globalization witnessed around the world. Sajjid Z. Chinoy and Toshi Jain 219

FIGURE 1. India Exports Share in GDP

% of GDP 28 25 22 19 16 13 10 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18

Source: Ministry of Statistics and Programme Implementation (MoSPI).

FIGURE 2. India Total Exports and Imports

% of GDP 60

20 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18

Source: MoSPI.

Despite that slowing, however, exports/GDP are currently still at 20 percent of GDP—almost twice the level that existed at the start of the millennium and at a level similar to that of Indonesia (Figure 1). This real integration also extends to the import side. Total imports to GDP increased from 13.1 percent of GDP in 1999 to 28.4 percent of GDP in 2013, before slowing in recent years. Consequently, total trade (exports plus imports) as a percentage of GDP rose from 24.4 percent in 1999 to 54 percent in 2013 before ebbing in recent years (Figure 2). All this has brought welcome exposure to global markets, along with the technological transfer and productivity growth that comes with it. 220 INDIA POLICY FORUM, 2018

2. The Quiet Revolution: From Textiles to Auto Parts…

Quite apart from the under-appreciated quantitative role of exports in economic activity, it’s important to appreciate the structural change and the two revolutions that have marked this sector over the last two decades. The first, more visible, was the surge in service exports. Back in 2003, service exports constituted 30 percent of the total export basket (Figure 4). But in a matter of just four years, service exports rose to 40 percent of the total basket, reflecting the software and BPO revolution around the world in which India was a major participant. However, after that step jump, service exports have plateaued at 40 percent of the total basket over the last decade. During this time, there has also been a quiet revolution occurring on the manufacturing side. Back in 2003, textiles, leather, and gems/jewelry, India’s traditional exports, constituted nearly 60 percent of the merchandise export basket (ex-petroleum). But their share has fallen secularly, and currently they account for just 40 percent of the basket. In contrast, engineering goods exports, mainly auto parts and capital goods, have grown at an average annual pace of almost 20 percent for 13 years, such that their share of exports in the manufacturing export basket has leapt from 20 percent to 35 percent in just 12 years (Figure 3). In a sense, therefore, India’s exports have become much more “high-tech” over the last two decades and also improved in technological content, quality, sophistication, and complexity (Anand et al. 2015). By 2015, engineering goods, electronics, and pharmaceuticals/chemical products constituted almost 60 percent of the non-oil merchandise basket. Equally, however, one could lament that all of this

FIGURE 3. India Sectoral Change in Manufacturing Exports

% change of manufacturing export basket, 2003 to 2015 15

–5 Engg Services Gems Leather

Source: Ministry of Commerce (MoC), JPMorgan research. Sajjid Z. Chinoy and Toshi Jain 221

FIGURE 4. India Export Basket

% of exports ex oil 70 Services Manufacturing 60 50 40

30 Primary 20 10 0 00 0201 0403 0605 0807 1009 1211 1413 16 1715

Source: MoC, RBI. growth has occurred in the capital-intensive sectors, when the need of the hour is job creation in labor-intensive sectors, whose share has reduced in the export basket.

3. Exports Drove the Growth Boom Pre-crisis

Perhaps the best appreciation of India’s increased openness is gleaned by the role that exports have played in India’s growth dynamics over the last two decades—both the sharp growth acceleration in the pre-Lehman period and the slowdown post the crisis. India’s growth surged in the mid-2000s with GDP growth averaging 8.8 percent between 2003 and 2008.1 What is less known is that this was achieved largely at the altar of surging export growth during this period of “hyper-globalization” around the world. The domestic value added of India’s exports grew at 16.6 percent during this period,2 whereas domestic consumption (public and private) averaged less than half of that at 7.2 percent (Figure 5). Consequently, the surge in private investment witnessed at the time (with gross fixed capital formation growing at 16.2 percent a year for 5 years) was largely responding to the buoyancy

1. As discussed earlier, we use the old GDP series for the 2000s since the back-casted series does not have the quarterly data, which are the unit of analysis for our study. Also, the back-casted series is only available from 2005–06 and not before that. 2. The domestic value added is estimated on the basis of the study by Veeramani and Dhir (2017). 222 INDIA POLICY FORUM, 2018

FIGURE 5. GDP, Exports, and Consumption Growth (Average for 2003–07)

% one year ago (oya)

20 17.8 16.6 16 12 8.8 7.2 8 4 0 * GDP Exports Consumption DVA Exports

Source: MoSPI; *Veeramani and Dhir (2017) for Domestic Value Addition in Exports average for the period 1999 to 2012.

FIGURE 6. Investment and Exports Growth

% oya, 3-year moving average (3YMA)

25 Exports 20

5 Investment

0 98 99 00 01 02 03 04 05 06

Source: MoSPI. of external demand rather than domestic demand. Exports were driving investment, and this is also seen in the close correlation between exports and investment during those years (Figure 6). India had briefly turned East Asian.

4. Slowing Exports Explain the Bulk of the Slowdown

The slump in exports over much of the last six years explains a large fraction of India’s growth slowdown vis-à-vis the mid-2000s. Between 2012 and 2018, for example, GDP growth has averaged 7.1 percent versus 8.8 percent Sajjid Z. Chinoy and Toshi Jain 223

FIGURE 7. GDP Growth

% oya 8.8 9

8 7.1 7

5 2003–07 2012–17

Source: MoSPI. in the boom years of 2003–073 (Figure 7). Between these periods, however, exports have suffered a double whammy. Not only has the growth of gross exports abated sharply but the growth of domestic value added in India’s exports has progressively fallen over the last decade (Figure 8). According to Veeramani and Dhir (2017), the domestic value content has fallen from about 86 percent in FY2000 to about 65 percent in FY2013 (Figure 9). This should not be unexpected. As India integrates with the global economy, and starts getting absorbed into global value chains, the domestic content per unit of exports should be expected to fall. Normally, this is more than offset by the scale benefit of integrating further into global value chains. But India has still not benefitted from the latter. In India’s case, therefore, not only has gross export growth slowed but the domestic value content per unit of exports has progressively declined. Consequently, the domestic value content of exports grew at 16.6 percent between 2003 and 2007, but at just 3.4 percent between 2012 and 2017 (Figure 8). Given its share in GDP, slowing exports account for a 200 bps slowing of headline GDP growth across the 2003–07 and 2012–17 periods. In other words, slowing exports alone (a slowdown in gross exports and the progressively lower domestic value added) can account for the entire slowdown in growth across these two time periods.

3. The figure of 8.8 percent is based on the old GDP series because the back-casted series is only available from 2005–06. But even if one were to combine the two series—use the earlier one up to 2005–06 and then the back-casted thereafter, average GDP growth in that period is almost 8 percent—which is still higher than growth over the last seven years, when exports have slowed materially. 224 INDIA POLICY FORUM, 2018

FIGURE 8. Exports Value Added Growth

% oya

20 16.6 16

8 3.4 4

0 2003–07 2012–17

Source: Veeramani and Dhir (2017).

FIGURE 9. Ratio of Domestic Value Added to Gross Exports

0.90

0.85

0.80

0.75

0.70

0.65

0.60 99 00 01 02 03 04 05 06 07 08 09 10 11 12

Source: Veeramani and Dhir (2017).

5. Scope of the Study: Three Questions

The natural question to ask, therefore, is: What has contributed to the sharp slowdown in Indian exports? Disentangling different impulses is important because there are different proximate factors at play. First, global growth has slowed and trade linkages have become more tenuous. Second, India’s broad, trade-weighted exchange rate (the 36-country real effective exchange rate) appreciated almost 20 percent between 2014 and 2017. Third, India witnessed successive (but presumably transient) supply shocks in the form Sajjid Z. Chinoy and Toshi Jain 225 of demonetization and GST in 2016 and 2017, respectively. All these factors have potentially contributed to the export slowdown. We attempt to answer three questions in this paper. First, what are the determinants of India’s exports and, in particular, what are the income and price (exchange rate) elasticities of exports, and how have they changed over time? This first question is important in light of the varied and inconclusive findings of income and price elasticities in the extant literature and the role of other factors (e.g., supply bottlenecks) that some have identified. Our second question is: How heterogeneous are these elasticities across different sectors, given the dynamism and changing composition of India’s export basket? The latter is important because previous work (Chinoy and Aziz 2010) has found that the elasticities vary sharply across sectors. Our third question is: How much of the recent export slowdown can our empirical exercise explain? In particular, can the recent disappointment be explained just by external demand and price factors? Or are they explained by factors outside the model, which would suggest transient shocks from demonetization and GST? Disentangling the slowdown is important because it can throw light on the durability of the export slump and also inform the policy response.

6. Price and Income Elasticities: Review of the Literature

The literature analyzing the determinants of India’s exports is relatively sparse, which is not unsurprising given the unappreciated role of exports in India’s growth dynamics. For the literature that does exist, studies try and estimate the price (exchange rate) and income elasticities of India’s exports—consistent with the approach found in cross-country studies. “Income elasticities” are measured by estimating the sensitivity of real exports to external demand (proxied differently in different studies through global growth, partner country growth, and global exports), while “price” elasticities are measured by estimating the impact on real exports from movements in the real exchange rate. Raissi and Tulin (2015) is the most recent study that estimates both income and price elasticities at an industry level from 1990 to 2013, and it finds a statistically and economically significant role for both income (1.3 percent) and price (–0.99) elasticities. The study also finds that supply constraints are a determinant of exports and have constituted a drag in the case of India. However, their analysis ends in 2013 and has, therefore, missed the bulk of the export slowing over the last six years. It also does not focus 226 INDIA POLICY FORUM, 2018 on whether these elasticities have changed over time and the impact of de-globalization on India. In contrast, IMF (2012) finds a much lower price elasticity. The long-run elasticity is estimated at –0.1 for the full sample period (1982–2011) and somewhat higher at –0.2 for the post-1990s period, but much lower than what Raissi and Tulin find for about the same time period. The corresponding long-run elasticities on external demand are found to be 2.9 and 2.2 for the respective periods. Kapur and Mohan (2014) get very different results in their 2013 IPF paper. Using annual data for 1980–81 to 2007–08, they find a long-run income elasticity of 1.1–1.4 and a price elasticity of –0.2 to –0.6—which is in between both the earlier studies. However, when they use the quarterly data to estimate the same elasticities post the reform period (presuming a structural break after the reforms), they find much higher elasticities, with that of external demand rising to 1.6–1.9 (when proxied through world exports) and to 2.6–3.6 (when proxied through world GDP). Similarly, REER elasticities rise significantly to –1.1 to –1.5 in the post-reform period. Importantly, however, the analysis stops in 2008, so we don’t know if these elasticities still hold and how they have changed over time. Chinoy and Aziz (2010), using quarterly data for 1996–2008, find a positive and statistically significant impact of external demand (real GDP growth in partner countries) on exports, with the estimated coefficient even larger at 4.6. However, the coefficient on the REER, while correctly signed (–0.6), is statistically insignificant at the aggregate level, even though it is economically and statistically significant for some sectors. All this suggests that there is a pressing need for updated work in this area. First, even the most recent of these studies uses data only up to 2013. So there are no estimated export elasticities over the last five years, precisely when forces of de-globalization were in the ascendancy. It is important, therefore, to ascertain whether and how these elasticities have held up and how they have evolved over time. Second, the extant literature is inconclusive on the role of the exchange rate and price elasticities. Some studies find a strong impact and other studies find a minimal impact. Ascertaining the role of the exchange rate is particularly important in recent years, given the sharp appreciation of India’s trade-weighted exchange rate between 2014 and 2017. All this makes a compelling case for new work in this area to analyze what has contributed to India’s export performance in the recent years and how much of the recent export slowdown can be explained by traditional determinants (global demand, exchange rate), and what fraction cannot Sajjid Z. Chinoy and Toshi Jain 227 be explained, and is therefore potentially attributable to idiosyncratic, and presumably transient, forces related to GST and demonetization.

7. The Approach

We use the quarterly time series data from 4Q2004 to 4Q2017 to estimate the sensitivity of India’s non-oil exports to both global demand and the trade-weighted real exchange rate, controlling for potential bottlenecks, since supply constraints are also found to be a determinant in some earlier work (Raissi and Tulin 2015). Our sample period starts from 2004 because the RBI’s 36-country, CPI-based REER that we use in our baseline model, as well as the export unit value indices from the IMF, date back only to 2004.4 We estimate the standard equation (Aziz and Li 2007; Hooper, Johnson, and Marquez 2000) linking real exports to global demand and the REER, and augment it to control for commodity prices (just in case the unit value deflators do not fully deflate out commodity price effects in the dependent variable), the global financial crisis, supply constraints at home, proxied through quarterly data on stalled projects, and the cost of capital. This log- linear model has been the dominant and empirically successful specification in the literature, given its relatively undemanding data requirements and straightforward interpretation. In particular, we estimate the following equation: ln(X) = a + b ln(GG) + p ln(REER) + m ln(CRB) + l(STALLED) + q(GFC) + j(GSec) + e (1) In our baseline equation, the dependent variable X is real non-oil exports (the sum of quarterly real merchandise and service exports), GG = real global GDP, REER is the 36-country, broad, trade-weighted REER produced by the RBI,5 CRB is the global commodity price index, GFC is a dummy for

4. We use the quarterly data, as do many other studies, since the use of annual data, as suggested by our discussant, would severely limit our data points, degrees of freedom, and the power of the model. 5. As a robustness test, we also use the BIS REER, as suggested by our discussant. This does not change our results at all. Also, if REER movements were simply reflecting productivity changes in the economy, productivity-adjusted exchange rates would not be changing. These, in turn, would result in the measured REER having little or no impact on export volumes. But this is precisely what we want to test in our analysis. The fact that measured REER has a significant impact on export volumes, ceteris paribus, suggests that actual REER changes are not moving in tandem with productivity changes, that is, the actual and equilibrium REER deviate, which is exactly what one would expect, and a plethora of other studies find. 228 INDIA POLICY FORUM, 2018 the global financial crisis, STALLED is a measure of stalled projects—our proxy for binding supply constraints—and GSec are real government bond yields, which are our proxy for the cost of capital. As part of our robustness checks, we replace global growth by: (a) global exports and (b) India’s trade-weighted partner country growth. As noted above, we also replace RBI’s 36-country REER with the REER created by the Bank for International Settlements (BIS).

8. Estimation Methodology

We find that our key variables X, GG, and REER are non-stationary I(1) (Table 1), we run both the Engel–Granger and Johannsen–Juselius maximum eigenvalue cointegration tests and find that the variables are cointegrated at order 1 (Table 2). Consequently, we estimate the aforementioned equation using both a dynamic ordinary least squares (DOLS) model and a vector error correction model (VECM) to estimate the long-run price and income elasticities of exports. The DOLS approach is a robust, single-equation approach which corrects for regressor endogeneity by the inclusion of leads and lags of first differences of the regressors, and for serially correlated errors by a generalized least squares (GLS) procedure. This is important because exports and the real exchange rate are both simultaneously determined, and DOLS controls for this endogeneity. DOLS is the preferred estimation technique for small sample sizes, given its efficiency and robustness properties. Furthermore,

TABLE 1. Unit Root Tests Null Hypothesis: Global growth has a unit root t-Statistic Prob Augmented Dickey–Fuller test statistic 1.11 1.00

Null Hypothesis: Real exports (ex-oil) has a unit root t-Statistic Prob Augmented Dickey–Fuller test statistic –1.66 0.45

Null Hypothesis: REER RBI has a unit root t-Statistic Prob Augmented Dickey–Fuller test statistic –2.46 0.13 Source: Authors’ calculations. Sajjid Z. Chinoy and Toshi Jain 229

TABLE 2. Cointegration Tests 1) Engle–Granger Variables: X, REER, GG Value Prob. Engle–Granger t-statistic –3.5 0.04 Engle–Granger z-statistic –17.0 0.09

2) Unrestricted Cointegration Rank Test (Maximum Eigenvalue) Variables: X, REER, GG No. of CE(s) Eigenvalue Prob. None* 0.4 0.0 At most 1 0.2 0.1 At most 2 0.0 0.2 Source: Authors’ calculations. Note:* Max-eigenvalue test indicates 1 cointegrating eqn(s) at the 0.05 level.

Monte Carlo experiments show that with a finite sample, DOLS performs well relative to six other asymptotically efficient estimators, including Johansen’s (1988) vector error correction maximum likelihood estimator (Stock and Watson 1993). Lag length for leads and lag has been chosen using the Akaike Information Criteria (AIC), subject to a maximum of four leads and lags given the limited length of the dataset. As a robustness check, we also estimate the exports equation using the VECM approach

∆L(Xt) = a1 + a2(ECMt–1) + ∑a3∆L(Xt–i) + ∑a4∆L(GGt–i)

+ ∑a4∆L(REERt–i) + GFC + ut (2) where ECM term captures the deviation of exports from the long-run equilibrium (L(X) – c1L(GG) – c2.L(REER)).

9. Empirical Results

9.1. Income Elasticity Our baseline results find a strong and statistically significant impact of global growth on India’s real exports. It is important to caveat, however, that because the variables are cointegrated, the estimated coefficients should be interpreted as long-run elasticities between these variables. As our baseline estimation reveals (Column 1 in Table 3), a 1 percent increase 230 INDIA POLICY FORUM, 2018

TABLE 3. Estimation of Income and Price Elasticity (Using Customs Data) Dependent variable: Log Real Nonoil (Customs Data Deflated) Method: Dynamic Least Squares (DOLS) Sample Period: December 2004 to December 2017 Results using 4 leads and 4 lags Baseline Robustness check 1 2 3 4 5 6 Cointegration regressors LN(REER)–RBI –1.44*** –2.0*** –2.3*** –1.4*** –1.4*** LN(GG) 2.6*** 2.9*** 3.3*** 2.6*** 2.6*** LN(WE) 3.4*** LN(REER)–BIS –2.2*** Deterministic regressors GFC 0.3** 0.3*** 0.4*** 0.2* 0.3** 0.1 LN(CRB) 0.2** 0.2 0.1 STALLED –0.004 G-Sec real rates –0.01 Adjusted R-squared 0.81 0.84 0.81 0.73 0.80 0.82 Source: Authors’ calculations. Note: *,**, and *** indicate statistical significance at 15%, 10%, and 5%, respectively. in global growth increases India’s real exports by 2.6 percent. The result is robust to how external demand is proxied. If instead of global GDP growth GG, we use global real export growth (WE) from the World Bank’s World Integrated Trade Solutions (WITS) database, the elasticity increases from 2.6 to 3.4 (Column 3 in in Table 3). Therefore, no matter what the definition of external demand, the income elasticity estimates are economically and statistically significant. The result is also robust to different measures of exports (i.e., dependent variable). If we use real export growth from the GDP accounts, the elasticity of global growth increases from 2.6 to 3.2 in the baseline model and remains very statistically significant (Table 4). Similarly, all the different definitions of global growth continue to remain economically and statistically significant under the alternative dependent variable choice. So the result is robust across different definitions of the dependent and independent variable. How does this magnitude compare to earlier studies? Our estimated coefficient is at the lower range than found by Kapur and Mohan (2.6–3.6) and much lower than found by Chinoy and Aziz (4.6). Recall, however, that both those studies ended in 2008. In comparison, Raissi and Tulin (2015) find Sajjid Z. Chinoy and Toshi Jain 231

TABLE 4. Estimation of Income and Price Elasticity (Using GDP Data) Dependent variable: Log Real exports (GDP) Sample Period: December 2004 to December 2017 Results using 4 leads and 4 lags Robustness check 1 2 3 4 5 6 Cointegration regressors LN(REER)–RBI –1.6*** –2.2*** –2.2*** –1.6*** –1.5*** LN(GG) 3.2*** 3.5*** 4.2*** 3.2*** 3.1*** LN(WE) 3.6*** LN(REER)–BIS –2.8*** Deterministic regressors GFC 0.2** 0.2*** 0.3*** 0.2** 0.3** 0.1 LN(CRB) 0.2*** 0.2** 0.2 G-Sec real rates –0.01 Adjusted R-squared 0.89 0.92 0.91 0.86 0.88 0.89 Source: Authors’ calculations. Note: *,**, and *** indicate statistical significance at 15%, 10%, and 5%, respectively. an income elasticity of 1.6, but their analysis extends till 2013. Prima facie, therefore, this suggests that income elasticities have potentially declined over the last decade, consistent with the de-globalization hypothesis. To test this, we divide our sample into two sub-samples from 2005–11 and 2011–17 and estimate the coefficients separately in each sample period to ascertain if the former have changed over time. We do find that income elasticities have declined over the last decade (Figure 10). The income elasticity for the 2005–11 period is estimated at 3, while that for 2011–17 falls to 2.4. This pattern is robust to how external demand is proxied (world exports, partner country growth) and confirms the hypothesis that India has not escaped the de-globalization bug, in that any level of global growth translates into correspondingly lower export growth over the last six years. The IMF (2013) estimates that the decline in trade growth globally (manifested in lower income elasticities) is attributable to weaker investment growth in developed economies (on account of the changing composition of growth) and the fact that investment growth suffered sharply in commodity-exporting developing countries on the back of the fall in commodity prices. All that said, while the elasticity has reduced from 3 to 2.4, the decline is not as large as commonly presumed. Global growth still has a significant bearing on India’s exports. 232 INDIA POLICY FORUM, 2018

FIGURE 10. Export Elasticities

3.0 2005–11 2011–17 3.0 2.4 2.0

% 1.0 –1.9 –1.1 0.0

–1.0

–2.0 Income Elasticity Price Elasticity

Source: Authors’ calculations.

9.2. Price Elasticity Apart from ascertaining whether income elasticities have fallen over the last decade, a key question of interest is whether, and to what extent, price (exchange rate) elasticities matter for Indian exports, particularly given the heterogeneous and dated findings of this phenomenon in the extant literature. As our baseline result reveals [Column 1 of Table 3], we find a strong and statistically significant impact of the exchange rate on non-oil exports between 2004 and 2017. Every 1 percent appreciation of India’s 36-country REER reduces real exports by 1.4 percent. Furthermore, the result is robust to the choice of REER, how the dependent variable is proxied, and the time period under consideration. In the baseline model, we use the RBI’s 36-country CPI-based REER. If, however, the RBI REER model is replaced with the BIS REER, the elasticity actually increases to 2. Both versions of the REER (RBI and BIS) also remain economically and statistically significant if the dependent variable is proxied by GDP-exports instead. Finally, the statistical significance is not an artifact of this particular time period. If one were to start the sample a year or two later and/or end a year or two earlier, the result is significant. This is a salient finding because the literature thus far has been very mixed on the role that exchange rates have played in impacting exports. Some studies (IMF 2012) find a very muted impact of the REER on exports. Others (e.g., Kapur and Mohan 2014) find a much larger elasticity (1.1–1.5) but their study ends in 2008. IMF’s (2015) analysis extends to 2013, but they find a smaller elasticity at 0.9. Sajjid Z. Chinoy and Toshi Jain 233

All this has led to the perennial questions of: How much do exchange rates matter for Indian exports? And how has this relationship changed over time? Our results hope to end that debate. We find a large and significant relationship between the broad, trade-weighted exchange rate and non-oil exports using data all the way till 2017. The next question, however, is whether these elasticities have changed over time. As in the case of external demand, we estimate the coefficient across two sub-samples: 2005–11 and 2011–17. As in the case of income elasticities, we find that price elasticities have also come down in recent years. The estimated price elasticity between 2005 and 2011 is 1.9 but falls meaningfully to 1.1 between 2011 and 2017 (Figure 10). But this, too, should not be a surprise. As noted earlier, the domestic value added of India’s exports has progressively fallen over the last decade (Figure 9, Veeramani and Dhir 2017). This is to be expected. As India integrates with the global economy, and starts getting absorbed into global value chains, the domestic content per unit of exports should be expected to fall. The implication, however, is that the role of the exchange rate should be expected to attenuate over time as exporters become more naturally hedged at the margin.

9.3. Other Controls As a robustness check, we also control for other factors such as global commodity prices to account for the possibility that real exports have not been fully deflated. In particular, we run the DOLS regression controlling for commodity prices—using the global commodity price CRB index—as a deterministic regressor. We find that the results do not materially alter with the inclusion of commodity prices. However, as commodity prices are statistically significant only in certain cases, they are not part of the baseline model. Some studies find that supply constraints matter for exports. Consequently, we test this hypothesis by including a proxy of supply constraints, the number of stalled projects (STALLED) as a fraction of projects under implementation, in our equation. However, we do not find STALLED to be a statistically significant determinant of exports, even though it is correctly signed. Finally, real government bond yields—our proxy for the cost of capital—are statistically insignificant.

9.4. Fitted Value The baseline DOLS model estimates the cointegration equation well with an adjusted R2 of 81 percent (Table 3). The model captures both actual real non-oil exports and the model’s fitted value. 234 INDIA POLICY FORUM, 2018

9.5. Vector Error Correction Model As a robustness check, we also estimate the long-run trade elasticities on the basis of a VECM. The results are presented below. Lags of the model are based on the AIC criteria with maximum four lags given the short sample size. The estimates based on the VECM approach are broadly in line with the estimates based on DOLS. In particular, the full sample income elasticity is estimated at 3.1 and the price elasticity is estimated at 2.2 (Table 5).

10. Sectoral Results

Having estimated the income and price elasticities, and examined how they have evolved over time, we move to our second question: How different are these elasticities across sectors? We find that India’s new-age exports-like services (which are mainly software services), engineering goods, and pharmaceuticals are found to have the highest “income elasticities” over the full sample period (Table 6 and Figure 11). In particular, the elasticity of pharmaceuticals and engineering

TABLE 5. Vector Error Correction Estimates (VECM) Sample Period: December 2004 to December 2017 Cointegration Equation with Log (real exports ex oil) LOG(GG) 3.1*** LOG(REER) –2.2*** Source: Authors’ calculations. Note: *,**, and *** indicate statistical significance at 15%, 10%, and 5%, respectively.

TABLE 6. Sectoral Elasticity Estimates Method: Dynamic Least Squares (DOLS) Sample Period: December 2004 to December 2017 Sectoral Pharma Engg Services Textile Leather Gems Cointegration regressors LOG(GG) 4.0*** 3.7*** 3.1*** 2.6*** 2.2*** 0.5 LOG(REER)–RBI –3.7*** –3.1*** –2.2*** –2.1*** –2.0*** –0.8** Adjusted R-squared 0.89 0.75 0.82 0.90 0.90 0.84 Source: Authors’ calculations. Note: *,**, and *** indicate statistical significance at 15%, 10%, and 5%, respectively. Sajjid Z. Chinoy and Toshi Jain 235

FIGURE 11. Export Elasticities (2004–17)

Income elasticity 4.0 Price elasticity

2.0

% 0.0

–2.0

–4.0 Engg Gems Leather Pharma Textiles Services Source: Authors’ calculations. goods is estimated at 4 and 3.7, respectively. Services follows suit with an elasticity of 3.1. In contrast, India’s traditional exports (textiles, leather, gems, and jewelry) are found to be much less sensitive to global growth. Textiles’ elasticity is estimated at 2.6, that of leather at 2.2, and that of gems and jewelry is just 0.5 (and not statistically significant), suggesting that it is not “cyclical” in nature. Instead, we find that gold prices are a bigger determinant of gems and jewelry exports. Similarly, price elasticities of India’s new-age exports are also correspondingly higher than the traditional exports. In particular, the price elasticity is found to be the highest for pharmaceuticals (3.7), followed by engineering goods (3.1), and services (2.2). Price elasticities for textiles (2.1), leather (2.0), and gems and jewelry (0.8) are all lower and statistically significant. Interestingly, while gems and jewelry are not elastic to global growth, they are sensitive to price changes. High income and price elasticities for India’s new-age exports suggest they are both discretionary (reflected in their cyclicality) as well as in highly competitive sectors (reflected in their high price elasticities). As in the case of the overall basket, we break down each sector into two sub-samples to assess how these sectoral elasticities have changed over time. Tables 7 and 8 show our main findings.

• Income and price elasticities for all sectors declined in the period 2011–17 versus the period 2004–11 period, consistent with the findings of the overall basket (Tables 7 and 8). • The new-age sectors (pharma, engineering goods, and services) have seen a much sharper decline than India’s traditional exports. This has 236 INDIA POLICY FORUM, 2018

TABLE 7. Sectoral Elasticity Estimates (2004–11) Method: Dynamic Least Squares (DOLS) Sample Period: Dec 2004 to Dec 2012 Sectoral Engg Pharma Services Textile Cointegration regressors LOG(GG) 5.4*** 4.7*** 3.1*** 1.6*** LOG(REER)–RBI –4.9*** –4.5*** –2.3*** –1.0*** Adjusted R-squared 0.52 0.64 0.78 0.71 Source: Authors’ calculations. Note: *,**, and *** indicate statistical significance at 15%, 10%, and 5%, respectively.

TABLE 8. Sectoral Elasticity Estimates (2011–17) Method: Dynamic Least Squares (DOLS) Sample Period: Dec 2011 to Dec 2017 Sectoral Engg Textile Services Pharma Cointegration regressors LOG(GG) 2.4*** 1.1*** 0.8** 0.3 LOG(REER)–RBI –1.3*** –0.4** 0.2 0.7 Adjusted R-squared 0.99 0.99 0.73 0.88 Source: Authors’ calculations. Note: *,**, and *** indicate statistical significance at 15%, 10%, and 5%, respectively.

meant a greater convergence for both price and income elasticities across sectors in the latest period. • However, some coefficients are statistically insignificant and perversely signed. We believe this is due to a very small sample size, which is unsuitable for the long-run relationship implicit in the DOLS and VECM models. We therefore have slightly lesser conviction when breaking the dataset into sub-samples.

11. Can the Model Explain the Sharp, Recent Slowdown in Exports?

We turn to our third and final question, which is, whether, and to what extent, these factors can explain the recent slowdown of exports. As discussed earlier, an important recent macroeconomic puzzle is why India’s exports Sajjid Z. Chinoy and Toshi Jain 237 have slowed so sharply in recent years. It is important to analyze the recent slowdown because there are several proximate factors at play. First, notwithstanding the recent volatility, average global growth in recent years is still meaningfully lower than the pre-crisis period and should have had a depressing effect on export volume growth. Second, India’s broad 36-country REER appreciated almost 20 percent between 2014 and 2017 (discussed in more detail below) and should have posed a headwind to exports. Third, India was buffeted by back-to-back adverse supply shocks in the form of demonetization and GST that are hypothesized to have adversely impacted exports by disrupting domestic supply chains. Our objective here is to try and disentangle these effects. First, how much of the slowdown can our model explain? The more we can explain, the less the recent slowdown is ostensibly attributable to supply shocks associated with demonetization and GST. Second, within the model itself, how much is attributable to weaker global growth versus a more appreciated real exchange rate? Our approach involves using the conventional “out-of-sample” testing. In particular, we run the model from 2004 to 2014 and then compare the model’s “out-of-sample” forecasts with the actual out-turn from 2015 to 2017. What we find is that the model comes close to explaining the actual outturn. In particular, between 2015 and 2017, real exports averaged just 1.1 percent, versus an average export growth of 8.6 percent within the sample. The model’s “out-of-sample” forecasts show a substantial deceleration of growth to 3.6 percent (Figure 12). So global growth and REER dynamics were themselves able to explain a significant deceleration of

FIGURE 12. Export Growth: In and Out-of-sample

% oya 10 8.6 8

6 3.6 4

2 1.1

0 In Actual Out of sample (2015–17) sample (2005–14) (2015–17) Source: Authors’ calculations. 238 INDIA POLICY FORUM, 2018 export growth, with exports forecasted to less than half of their in-sample growth. That said, the model does not capture the full slowdown in the period 2015–17. This suggests that factors outside the model (e.g., demonetization/ GST) were temporarily responsible for depressing export growth below what global growth and exchange rate dynamics would have suggested. All told, therefore, the model is able to explain a substantial deceleration in export growth between 2015 and 2017. Part of this has to do with the sustained appreciation of the REER during that time. From 2015 to 2017, the REER appreciated, on average, by 4.4 percent a year. Using the estimated elasticities, the real appreciation pulled down export growth by 7.7 percent per year, which is an appreciable drag on export growth the last few years.

12. Role of REER: Dutch Disease

What all this suggests is that (a) slowing global growth is not the only reason why India’s export growth slumped between 2015 and 2017, and (b) the exchange rate has been an important determinant of India’s exports with the cumulative 20 percent real appreciation between the start of 2014 (when the exchange rate had stabilized after the taper tantrum) and the end of 2017, posing a significant headwind to exports (Figure 13). From a policy perspective, it is important to ascertain what factors underpinned this large real appreciation.

FIGURE 13. 36-Country Trade-Weighted Real Effective Exchange Rate (REER)

Index (36 currencies) 125 120 115 110 105 20% appreciation from 2014–17 100 95 09 10 11 12 13 14 15 16 17 Source: RBI. Sajjid Z. Chinoy and Toshi Jain 239

We would argue that this real appreciation was inevitable. How so? The collapse in oil prices in 2014 served as a large, positive terms-of-trade shock for India. Economic theory would argue that a positive terms-of-trade shock should manifest itself in a more appreciated real exchange rate. The intui- tion is straightforward. To the extent that windfall gains from a positive terms-of-trade shock (either higher export prices or lower import prices) are spent, the price of non-tradables should rise vis-à-vis the price of tradables and drive some real appreciation. We have previously estimated that India witnessed large windfall gains from the collapse in oil prices (3.1 percent of GDP across FY15 and FY16, of which two-thirds was estimated to have been spent—Table 9). So the collapse in oil prices should have put upward pressure on actual and equilibrium real exchange rates in India. The only choice Indian policymakers had was whether to accommodate this real appreciation through nominal appreciation or relatively higher inflation. Operationally, this manifested itself in a collapse of the CAD (because of oil) and, therefore, a larger balance of payments surplus that was putting upward pressure on the rupee. This was compounded by FDI flows almost doubling after the new government came to power in 2014. All this exacerbated the terms-of-trade shock such that the CAD collapsed from 2014 to 2017 exerting sharp and sustained appreciation pressures. In a sense, this is akin to the “Dutch Disease” problem, a term developed to describe the situation in the Netherlands in the 1960s where a discovery of gas deposits in the North Sea, and the income boom that followed, led to a real appreciation of the exchange rate that crowded out manufacturing exports. The “Dutch Disease” phenomenon has since been broadened to include the effects of positive terms-of-trade shocks, something that appears to have afflicted India over the last four years: the collapse in oil prices resulted in a large, positive terms-of-trade shock that drove up the actual

TABLE 9. Estimated Impact of Oil on Indian Growth Oil Prices ToT Shock Boost to Growth ($/barrel) (% of GDP) (% of GDP) FY14 105 FY15 85 1.0 0.7 FY16 46 2.1 1.3 FY17 47 0.0 0.0 FY18F 58 –0.6 –0.4 Source: Authors’ calculations. 240 INDIA POLICY FORUM, 2018 and equilibrium real exchange rate which, in turn, has likely reduced the competitiveness of India’s exporting sector.

12.1. Still Waters Run Deep The slowdown in India’s non-oil exports has meant that India’s underlying external imbalance has deteriorated far more than the headline numbers suggest. On the surface, India’s CAD looks relatively benign at 1.9 percent of GDP in 2017–18, and expected to widen, but to less than 2.5 percent of GDP in 2018–19—only half the level of the CAD of 4.8 percent of GDP in 2012–13 that pre-dated the taper tantrum. But the real story lies below the surface: that the fall in oil and gold imports is masking a sharp and sustained deterioration in India’s underlying external imbalances. If one excludes net oil and gold imports, India runs a sizeable current account surplus. But that current account surplus has seen a sharp and sustained deterioration, declining by almost 3 percentage points of GDP between 2014 (when the economy had stabilized post the taper tantrum) and 2017 (Figure 14). Underlying imbalances have, therefore, markedly widened and, in 2017, were at their weakest level in more than a decade. This starkly confirms that India’s underlying competitiveness has reduced in recent years, and the sharp real appreciation from 2014 to 2017 likely played a part (Figure 15). That appreciation, however, was an inevitable consequence of the large, positive terms-of-trade shock that India was the beneficiary of, suggesting that India was, indeed, afflicted by the Dutch Disease phenomenon.

FIGURE 14. Current Account Ex Oil and Gold

% of GDP, 4-Quarter Moving Av. 6

1 05 07 09 11 13 15 17 Source: RBI. Sajjid Z. Chinoy and Toshi Jain 241

FIGURE 15. Current Account Ex Oil, Gold, and REER

% of GDP, 4-Quarter Moving Av. Index, reverse scale 6 95 REER 5 102

4 109

3 116 Current account balance REER 2 ex oil, gold appreciation 123 1 130 05 07 09 11 13 15 17 Source: RBI.

13. Policy Implications

At least four crucial policy implications follow from this discussion—two short-run and two long run. First, the 50 percent increase in crude prices over the last 18 months will have the symmetrically opposite impact on the equilibrium real exchange rate. It partially reverses the earlier positive terms-of-trade shock and will thereby induce some actual and equilibrium real exchange rate depreciation. This should help mitigate some of the pressures on India’s tradable sector, thereby reversing the Dutch Disease. It is important, therefore, that policymakers do not fight this real depreciation or attempt to change the end- point (since this is an equilibrium phenomenon), but simply use reserves to ensure that the new equilibrium is reached in a calibrated and non-disruptive manner, so as to avoid self-fulfilling panic and overshooting. Policymakers seem to be doing that. Since the start of 2018, the 36-country REER has depreciated by 6 percent (Figure 16). So the first implication is to let the real exchange rate gradually depreciate as the terms-of-trade shock reverses. Interestingly, since the start of 2018, coincident with the real depreciation, exports have re-accelerated, but this has also been helped by the temporary disruptions associated with the adoption of the GST likely fading. Second, it is crucial that fiscal policy (at both the central and state levels) does not become more expansive. Already, the total public sector borrowing requirement has widened over the last two years. The more expansive fiscal policy is in India, the more real appreciation it will induce, and thereby offset the real depreciation that will naturally occur from the positive 242 INDIA POLICY FORUM, 2018

FIGURE 16. 36-Country Trade-Weighted Real Effective Exchange Rate (REER)

Index (36 currencies) 125 120 115 110 105 6% depreciation since 2018 100 95 09 10 11 12 13 14 15 16 17 18 19 Source: RBI. terms-of-trade shock reversing. Fiscal expansiveness will, therefore, indirectly contribute to impinging on tradable sector competitiveness. These are, however, short-term policy implications. The broader policy implications are more important and medium-term in nature. First, India needs to dramatically improve underlying trade competitiveness, quite apart from exchange rate dynamics, by boosting infrastructure and total factor productivity, and assimilating into global value chains. Second, India will need to look for new “growth drivers.” Exports boosted growth in the mid-2000s during the period of hyper-globalization. But with global growth softening (compared to that era), growing fears of protectionism around the world, and “income elasticities” reducing in recent years, the global economy is unlikely to provide the tailwinds it did in the mid-2000s. As such, either India will need to improve its competitiveness to the point that it increases its market share in the exports arena, thereby capturing a bigger slice of a stagnant/shrinking pie, or look for new growth engines domestically. Either which way, Indian policymakers have their work cut out.

References

Anand, Rahul, Kalpana Kochhar, and Saurabh Mishra. 2015. “Make in India: Which Exports Can Drive the Next Wave of Growth?” Working Paper WP/15/119, Washington, DC: International Monetary Fund. Aziz, Jahangir, and Xiangming Li. 2007. “China’s Changing Trade Elasticities,” Working Paper WP/07/266, Washington, DC: International Monetary Fund. Sajjid Z. Chinoy and Toshi Jain 243

Chinoy, Sajjid, and Jahangir Aziz. 2010. “India: More Open Than You Think,” Economic Research, New York: JP Morgan. Hooper, Peter, Karen Johnson, and Marquez. 2000. “Trade Elasticities for G-7 Countries,” Princeton Studies in International Economics No. 87, Princeton, NJ: Princeton University. Kapur, Muneesh and Rakesh Mohan. 2014. “India’s Recent Macroeconomic Performance: An Assessment and Way Forward,” Working Paper WP/14/68, Washington, DC: International Monetary Fund. Raissi, Mehdi and Volodymyr Tulin. 2015. “Price and Income Elasticity of Indian Exports—The Role of Supply-Side Bottlenecks,” Working Paper WP/15/161, Washington, DC: International Monetary Fund. Stock, J. H. and M. W. Watson. 1993. “A Simple Estimator of Cointegrating Vectors in Higher Order Integrated Systems,” Econometrica, 61(4): 783–820. Veeramani, C. and Garima Dhir. 2017. “Domestic Value Added Content of India’s Exports: Estimates for 112 Sectors, 1999–2000 to 2012–13,” IGIDR WP-2017- 008, Mumbai: Indira Gandhi Institute of Development Research. Comments and Discussion*

Surjit Bhalla PM’s Economic Advisory Council and NCAER

The paper is rich and one shouldn’t expect less from the authors. The major claims in the paper relate to the questions of why Indian exports have slumped in recent years and, relatedly, why export growth has not responded to faster global growth over the last year. The long-term nature of currency depreciation and its relationship with export growth means that we are talking about a phenomenon occurring over the last decade post-GFC. The paper suggests that its empirical model can explain the significant slowing down of exports; that global demand and exchange rate dynamics have posed meaningful headwinds for Indian exports in recent years; and that temporary factors such as demonetization and GST may have had little role to play. Their model is a straightforward one: real, non-oil exports are a function of prices, proxied by the real exchange rate, and global GDP and our trade growth. I have no problem with this specification, except the definition of the real effective exchange rate (REER) because it does not control for Balassa–Samuelson productivity effects. Several important pieces of evidence presented in the paper are for the golden era of 2003–07, when export growth went up to double digits upward of 20 percent per annum, and the bronze era of the last 4–5 years, when exports slowed down to less than 8–10 percent. The reason I am emphasizing the incorporation of productivity effects is that during 2003–07, the Indian rupee was a lot more undervalued than today, and this can explain part of the divergence between then and now. Further, I will show that the earlier period does not quite stand up to scrutiny as having been that glorious time for export growth.

In addition, I have a serious problem with the use of quarterly data for explaining long-term performance. It can lead to confounding and surprising results, one of which is the large price elasticity found by the paper. In general, very different results are often reached by authors with a minor increase or decrease in the sample size of the quarterly data. On the basis of the quarterly data for 12 years, 2005–16, the RBI’s monthly “Monetary Policy Report” for April 2018 reported that a consolidated fiscal deficit increase of 100 basis points would lead to a rise in inflation by 50 basis points in the same year. In a note on the RBI paper, I show how adding one extra year of 2004 quarterly data reduces statistical significance, and if more years are added all the way till 1996 (the start of the quarterly data), significance completely disappears. Just the addition of four quarters changed the entire conclusion. A similar problem may be plaguing this paper. Real exchange rate data for a large cross-section of countries is available on the BIS website; these data are monthly, and this hypothesis can easily be tested. A lot can be learnt from the use of just annual data before deep diving into the econometric complications of small sample size and dynamic least squares. The use of quarterly data is fraught with difficulties and the results can be fragile. I did not have their full model specification, which the paper now provides; but in their draft paper, I had problems in obtaining a negative elasticity for REER, and instead obtained a positive elasticity of REER with export growth using annual data. The dependent variable was the same, non- oil exports, and the independent variables were global growth and REER, as measured by BIS (see Table 1 for the underlying data). There may be good reasons for my getting the opposite empirical result as the paper. This has to do with deflators. I’ll illustrate this by investment. The share of investment in GDP declined from something like 38 percent at its peak to something like 27 percent today, a decline of 11 percentage points (Figure 1). However, the share of real investment to real GDP (CSO data) is only about 4 percentage points. So are we really talking about investment going from 34 to 30 percent, which has caused the entire slump in GDP growth (and maybe even caused export growth to decline)? I think we need to worry about that result. In contrast to the paper, I find that Indian exports have performed really well. This pertains to the trends in Indian goods and services exports as a percentage of our GDP and notice that the REER very correctly goes up, and that trade in goods and services peaked at something like 17 percent per annum and then declined (Figure 2). This confirms what the paper is saying—the REER goes up in the recent period, and export growth declines. 246 INDIA POLICY FORUM, 2018

TABLE 1. Indian Exports and REER: 1998–2017 Non-oil Export Growth REER Year Real Dollar REER36 BIS (Y-o-Y in %) 1998 4.3 –4.5 – 91.5 1999 13.4 8.9 – 90 2000 15.5 15 – 94.2 2001 0.5 –2.4 – 96.2 2002 19.7 17.5 – 92.8 2003 6.8 16.7 – 92.7 2004 25.9 24.3 – 93.6 2005 20.9 19.5 102 96.6 2006 15.6 16.2 100.7 95.8 2007 5.7 20.6 108 102.1 2008 19.3 22.3 101.8 97.1 2009 –13.4 –11.5 101.1 91.6 2010 24 34.2 111.7 102.3 2011 17.3 21.4 111.6 102.4 2012 –0.1 –5.2 105.6 95.9 2013 5.3 6.2 104.5 91.4 2014 –0.7 –2.1 106.7 92.8 2015 –3.7 –7.1 112.2 100 2016 4.8 2.9 113.3 101 2017 1.5 7.5 119.2 105.4 Source: Official data and author’s computations.

FIGURE 1. Investment Leads FIGURE 2. Trends in Indian Capital Formation! Goods and Services Exports

45 % of GDP 40 30 110 35 25 105 30 20 100 25 in % in 20 in % 15 95 15 10 90 10 5 85 5 REER-BIS (2010 = 100) 0 0 80 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2019 2001 2002 2003 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2000 2004 2005 2006 2007 2020 Fiscal Year Fiscal Year Gross Investment/GDP (Current Prices) Goods & Services Goods Gross Fixed Capital Formation/GDP (Constant 2010 Prices) Services REER-BIS

Source: Official data and author’s computations. Source: Official data and author’s computations. Sajjid Z. Chinoy and Toshi Jain 247

However, what is not quite so apparent is that the period 2000–07, the period of the largest increase in India’s trade shares, was also a period that witnessed a 10 percentage point increase in the REER! During the boom time for Indian exports, you obtain a 10-percentage point increase in the REER and strong double-digit export growth. I would be very, very worried, with this “perverse” result. In addition, I want to show what has happened to the deflator for world manufacturing (Table 2). The price deflator for that has not moved at all. This is only available from 2007 onward and has not moved at all from 2007 to 2017. Then I want to show how the rest of the world is performing in terms of real exports in constant 2010 US dollars using World Bank data. As

TABLE 2. Basic Statistics on Export Growth ∆Manufacturing Non-oil Export Growth World Manufacturing Deflator Year Real Dollar Deflator India World (Y-o-Y in %) 1998 4.3 –4.5 – 5.2 – 1999 13.4 8.9 – 3.3 – 2000 15.5 15 – 4.3 – 2001 0.5 –2.4 – 3.9 – 2002 19.7 17.5 – 1.9 – 2003 6.8 16.7 – 4.7 – 2004 25.9 24.3 – 6.7 – 2005 20.9 19.5 – 4.8 – 2006 15.6 16.2 – 5.6 82.8 2007 5.7 20.6 89.1 5 7.6 2008 19.3 22.3 96 7.5 7.8 2009 –13.4 –11.5 90.8 1.1 –5.2 2010 24 34.2 92.5 6 1.9 2011 17.3 21.4 99.9 7.6 8.1 2012 –0.1 –5.2 97.8 6.2 –2.1 2013 5.3 6.2 98.6 3.7 0.8 2014 –0.7 –2.1 98.4 3 –0.2 2015 –3.7 –7.1 88.8 –0.3 –9.7 2016 4.8 2.9 87.8 1 –1.1 2017 1.5 7.5 91.1 2.9 3.8 Source: Official data and author’s computations. 248 INDIA POLICY FORUM, 2018 regards how the rest of the world is working, you find that the Indian trade share as a percentage of world trades has constantly gone up. Let us look at what happened to our trade shares between 2000 and 2007 as compared to other countries in the world. I am only looking at Asian countries and Latin America, and taking out sub-Saharan Africa, Communist Russia, and the advanced countries. I am looking at the change in our shares in world export, change in our export shares, and world shares. The highest improvement between 2000 and 2007 was recorded by India at 0.8 percentage points (from 0.93 to 1.77 percent) and slightly behind India was Korea, also at 0.8 percentage points. Now let us look at the period 2007–16, when we did disastrously, whereas the country that registered the highest increase in trade shares was Korea again at 0.8. Which country had the second highest increase in trade shares in this sample of all countries (mind you I took out countries with populations of less than 5 million)? It was India, which saw an increase in share of 0.7 percentage points (from 1.77 to 2.43 percent). What about Vietnam that has done so well in export growth—it had an increase in share of 0.5 percentage points; Mexico had 0.3, Chile –0.1, and so on and so forth. My point is that between 2007 and 2017, our share of exports in world exports was the second highest, and we think that our share declined because of price elasticities. I don’t buy that and I don’t think that the data substantially support that. Now if you are looking for what has caused this, look at Table 3. It has the exports-to-GDP share in 2016 for various countries. It has a change in exports share for 2007–16, and the last column is from the World Bank’s Ease of Doing Business Tables, which shows the taxation level. In other words, how much of the profits are retained by the firm after payment of taxes and wages? Vietnam does very, very well. They have taken over the textile world, and their share is 61. The shares of Korea, Bangladesh, and India are 67, 66, and 45, respectively. If you were to have 10 percent depreciation, we will go from 45 to 50; but your competitors change as well, so the relative real exchange rate (the one which matters for competitiveness) does not change much. On the other hand, just a 5 percentage point change in overall taxes, and corporate taxes, among others, will bring about a much larger change in export growth because of a large improvement in competitiveness (assuming that other countries do not change their rates of taxation). Now the big takeaways. A small sample size yields dangerous results, and REER movements behave perversely. The decline in export shares is not much of an indicator for export performance between 2007 and 2016. The decline in most of the investment share can be explained by Sajjid Z. Chinoy and Toshi Jain 249

TABLE 3. Exports-to-GDP Share of Various Countries Profit Retained Exports/GDP Δ (Exports/GDP) (after Taxes & (2016) (2007–16) Wages in %) Country Share Rank Share Rank Share Rank Vietnam 94.1 7 22.2 2 60.6 40 Cambodia 66.4 12 10.9 8 79 74 Korea 44.5 24 2.1 19 66.8 54 Lao P.D.R. 27.1 52 1 21 74.7 71 Bangladesh 16 75 –0.7 25 65.6 52 India 19.6 69 –0.8 26 44.5 13 Thailand 70 11 –1.1 29 73.2 70 Hong Kong 238.2 1 –3 36 77.2 72 Nepal 12.2 79 –3.9 37 70.5 65 Myanmar 21 64 –4.4 38 66.9 55 Sri Lanka 21.7 61 –4.9 39 44.8 14 Pakistan 9.9 81 –5.4 40 67.1 57 Indonesia 18.4 71 –8.3 45 70.3 63 Philippines 27.4 51 –8.6 46 57.1 31 China 21.6 62 –15.9 56 32.1 7 Papua New Guinea 36.9 35 –17.6 59 60.7 41 Malaysia 71.5 10 –37 62 60 38 Source: Official data and author’s computations. low increases in the investment deflator versus the overall. India’s share in real world exports is today at its peak, at 2.6 percent versus 1.8 percent in 2007 (Figure 3). The problem is low share. What is the cause? We are looking for an explanation of bad export performance in all the wrong places. If you want to look at what determines GDP growth, it is very simple. This was done as part of the great demonetization exercise which, people keep repeating, has done a lot of things. I am simply looking at what happened to growth in India. The demonetization effect is defined as GDP minus mining, minus manufacturing, minus public administration and defense. Manufacturing is taken out because of what is called the “twin balance sheet problem.” These data go all the way back to 1980, and this is the real interest rate as defined by the SBI lending rate and deflated by the CPI. There is a close negative relationship between the two—high real rates mean lower GDP growth. If FIGURE 3. Trends in Indian Goods and Services Exports

% of World’s Exports 4 110

3.5 105

100 2.5

2 95 in %

1.5 90 BIS–REER (2010 = 100)

85 0.5

0 80 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 Year Goods & Services (Constant 2010 Prices) Goods Services REERBIS

Source: Official data and author’s computations. Sajjid Z. Chinoy and Toshi Jain 251 you are looking for exports performance, look at taxation—high rates of taxation lead to lower export growth.

Kenneth Kletzer University of California, Santa Cruz

I plan to talk about this paper a bit differently than Sajjid did in his presentation. I want to talk less about the subject and more about the paper with a focus on revisions and future research. First, this is a really nice paper. It is an interesting and important subject for India. One thing I particularly liked about the presentation was that it shows that the authors have already made a lot of progress in revising the paper. Sajjid offered quite a lot more material in the presentation than there is in the draft. The additions are very useful, and I had already planned to suggest some of them here. Let me begin with the motivation for the paper. In his discussant’s comments, Surjit Bhalla noted that we saw a real boom in exports between 2003 and 2008. The paper concentrates on two periods, 2003–08 and 2012–17. The rapid growth of India’s exports in the first period and the slowdown in the second have been discussed several times already. An important observation is that while India grew more rapidly than other emerging markets during 2003–08, exports growth was even higher, outperforming output growth. The authors focus on how India’s share of world exports fell as the global export growth rate declined by 3 percent following the global financial crisis. The share of exports in India’s GDP fell from 25 percent to 19 percent between 2013 and 2017. India’s share of global exports rose again to 2 percent. The improvement in the terms of trade and real exchange rate appreciation over the four-year period from about 2014 to 2018 is particularly important for the main policy implications raised by the authors in the paper. This is the reverse of the real exchange depreciation that took place during the boom years of 2003 to 2008. In addition to showing the effect of a boom in global demand on India’s export share, the data in the paper suggest that the depreciation of the real exchange rate made a major contribution to the rise in export growth. The first part of the paper shows that the decline in the growth rate of GDP happens to equal the decrease in the growth rate of value added in exports between the periods 2003–08 period and 2012–17. However, this does not mean that the decline in export growth explains that decrease. That doesn’t mean that I disagree with Sajjid’s emphasis on the importance of 252 INDIA POLICY FORUM, 2018 the boom in export growth in the years before the financial crisis either. Rather, they are doing a partial accounting exercise and even if you do a complete accounting exercise as Poonam Gupta presented in her paper on India’s Growth Story yesterday, you may not be able to infer causality. The argument that a drop in export growth explains the drop in output growth requires a more systematic analysis, however important the fall in global demand for Indian goods and services may have been. Another point worth noting is that imports are not mentioned in the comparison of growth between the two periods. Imports were not constant, and net exports grew between 2012 and 2017 (as shown in Anand et al. 2015). An increase in net exports should contribute to output growth. The authors link the substantial decline in investment to the fall in export growth, but I think that the accounting needs to be more thorough. After the financial crisis, both investment and imports began to rise as export growth remained slow. The figures in the paper illustrate a significant increase in consumption between 2003–08 and 2012–17, while the shares of investment and exports in output fell. Turning to the empirical analysis, I have a different take on the empirical analysis than did Surjit. In particular, I appreciate the authors’ efforts to work with as many data points as possible by using quarterly data. This is especially helpful because they want to compare export growth across sub-periods of a 15-year full period. The estimates show how the elasticities of exports with respect to global output growth and the relative prices of tradable goods vary across industries. The estimates show that exports are fairly responsive to real exchange rate movements. Compared to the cited papers, the data here provide significantly more variation in relative prices and in global income over the period. This is an important contribution. Let’s turn to how the authors interpret the slowdown. Their interpretation wasn’t emphasized in the presentation, but it is emphasized in the paper. They focus on the relationship between oil prices and the real exchange rate for India. Just graphing the data, and not even running a regression, shows that India’s real exchange rate and the price of oil have been closely related since 2000. The real exchange rate depreciated in the years before the global financial crisis as the price of oil rose more than threefold in current dollars from 2003 to 2008. The rupee steadily appreciated in real terms after 2012 as the price of oil fell. In his presentation, Sajjid raised a new question: whether the sensitivity of India’s trade flows to the terms of trade has changed as Indian industries became more integrated with global value chains. In particular, he suggested that the decrease in export value added might contribute to an increasing response of exports to the terms of Sajjid Z. Chinoy and Toshi Jain 253 trade. Looking at just the two time periods for India, one in which the terms of trade improve and the other in which they deteriorate does not provide sufficient data to consider the effects of global value chain integration, that is, across the two time periods, there may be a progressive or a sporadic change in the underlying character of trade. We really cannot tell whether the effects of real exchange rates or the terms of trade on export value added are asymmetric or symmetric in the sample. The core argument in the paper is that the large drop in oil prices is anal- ogous to a capital inflow and the outcomes can be interpreted in terms of the Dutch disease. The post-2012 decrease in oil prices provided an exogenous positive shock to net exports. As India was paying less for oil imports, the current account balance net of oil and gold also improved. The authors point out that the improvement in the current account over this period is similar to a surge in capital inflows. Falling oil prices improved the terms of trade vis-à-vis oil exporters and increased value added in non-petroleum traded goods industries. As a consequence of the rise in the value of output net of oil input, domestic demand should rise for both traded and non-traded goods. Like a capital inflow surge, the sharp fall in oil prices starting in 2012 ought to have caused a rise in the demand for non-traded goods requiring a rise in non-traded output. The appreciation of the real exchange rate follows from the necessary increases in equilibrium relative prices of non-traded goods. Thus, as the terms of trade improve and the real exchange rate appreciates, production moves from import-competing industries toward export industries and non-traded goods production. The Dutch disease concerns the net decline in exports that can occur because traded goods production as a whole decreases. The authors adopt this explanation for the reduction in Indian export growth in the global recovery. The behavior of the current account is consistent. As shown in the presentation, the non-oil and gold current account surplus falls as the real exchange rate appreciates. I would like to see more direct evidence that the production of non-traded goods increased, while overall traded goods production or employment contracted. On balance, however, I think that the emphasis on global demand and relative prices, that is, slower global export growth and real exchange rate appreciation, to try to explain India’s export growth after the crisis, is well placed. It is also consistent with the elasticity estimates. A substantive question I have concerns the importance of this interpretation for macroeconomic policy. In his presentation, Sajjid said very little about monetary policy, but the paper highlights the monetary policy implications. The regression results do suggest that real exchange rate movements significantly affect export growth and, as well, net exports. The 254 INDIA POLICY FORUM, 2018 observation that falling oil prices and capital inflow surges have parallel effects suggests that the RBI should intervene in the foreign exchange market or perhaps accumulate reserves to promote export growth. The authors raise the possibility that monetary policymakers were resistant to depreciation of the nominal exchange rate. That should not have necessarily happened in an inflation-targeting regime. At a first pass, open economy new Keynesian models show that the central bank of a small open economy should target domestic inflation. This prescription might be modified in the presence of incomplete exchange rate pass through or when the country can influence its terms of trade. And if the country can manipulate its terms of trade, an inflation-targeting central bank can use monetary policy to influence the nominal exchange rate, hence the real exchange rate and terms of trade. The point most relevant here is that foreign exchange intervention, or accumulating reserves, does not need to be in conflict with a domestic inflation-targeting regime. However, real appreciation can occur under a given inflation target, so that changing the inflation target, as suggested in the paper, might be warranted. I think we could learn more about what is going on in the export industries by looking to micro data. Although it is not in the paper, in his presentation, Sajjid talked about using micro data to understand export growth better. The macro data do not tell us much about how the real appreciation or fluctuations in global demand affected exporters; firm-level data might. The contraction in global economic growth after the financial crisis, the subsequent recovery of global export demand, and the real exchange rate could have affected exporting at either the extensive or intensive margins at the firm level. The effect of the real appreciation at the extensive margin of firm entry into exporting is probably very important. As we learn from the trade literature, firms encounter fixed costs entering export markets, more productive firms enter these markets, and productivity growth can be higher for exporting firms. How producer dynamics in export industries varied across industries would help us to understand the importance of the decline in export growth rates for policy intervention. Going back to the Dutch disease interpretation reminds us that a concern about real appreciation in newly resource export-abundant countries led to persistent declines in manufacturing employment. An episode of real appreciation that reduces industry growth temporarily might have persistent effects on productivity growth. The primary mechanism for temporary movements in the terms of trade or global demand to impact productivity growth in the new export industries for India could well be a decrease in the rate of entry into exporting. If as entry falls, industries realize fewer dynamic Sajjid Z. Chinoy and Toshi Jain 255 economies of scale, then macroeconomic responses to real appreciation could be quite important. Another hypothesis that could be addressed using micro data was made in the presentation. This is the observation that India’s move toward a greater integration in global value chains is reducing domestic value added in exporting. It would be good to know how productivity and value added evolved during the period of rapid growth, 2003–08, and after 2012–17. More detailed data might also help us to understand what is happening at the firm level during the exporting boom as well as recently. A related question is whether the allocation of foreign direct investment across export industries has changed. My last bullet point is that the macro analysis indicates the slowdown is concentrated in certain industries—pharmaceuticals are one, engineering is another, services less so. It cannot tell us why. We are aware that India is facing a lot of competitive pressure in pharmaceuticals, particularly from China, and this may also be occurring in some of the engineering industries. This could be a reason for the observed sensitivity to relative prices in these sectors. The price sensitivity in pharmaceuticals suggests that maybe it is a particularly competitive industry.

General Discussion

Chaired by Anup Wadhawan Commerce Secretary, Government of India

Vijay Joshi identified two problems with the single equation model—one econometric, and the other substantive. What is called the price elasticity is actually not: it is the exchange rate elasticity of exports. He argued that on the economic side, the single equation model mixes up the demand and supply. A simultaneous equation model with a demand equation for exports and a supply equation for exports would strengthen the results, as it would help estimate a foreign price elasticity of exports and a home supply elasticity of exports. On the econometric side, this exercise could be repeated with a different concept of the real exchange rate, that is, the price of tradables to non-tradables. If the paper could construct such a series, that would then accord better with the second half of the paper. On policy, the paper rightly says that real exchange rate appreciation is an equilibrium appreciation, and we should therefore also not resist a real exchange rate depreciation. So should the government have done something 256 INDIA POLICY FORUM, 2018 about the real exchange rate appreciation? There is a debate on this in India. Joshi felt that the lag between exchange rates and exports is long and variable, which necessitates having a competitive real exchange rate in place for some time before getting an export response. Therefore, the objective of the authorities should have been to keep the real exchange rate reasonably stable in the period when it appreciated very rapidly. This would have been possible through a traditional mixture of sterilized intervention and targeted capital controls, implying that India has gone too far in the direction of liberalizing capital controls. Prerna Prabhakar wondered why, since the model used in the paper had established a positive income elasticity, the impact of the global financial crisis was actually opposite to what one would expect. Her second query was whether the analysis for out-of-sample forecasting done in the paper could help in assessing a possible impact of demonetization on the decline in export growth. Sudipto Mundle sought two clarifications. One pertained to the argument raised in the paper that exports completely explained the growth story. In this context, the earlier discussion on the paper in this IPF on “India’s Growth Story” (Gupta et al. 2018–19) had also flagged investment, and he wondered if investment behavior could also be explained by exports. The second question was about looking for new drivers of growth. It would be useful for the paper to dig deeper into the role of global value chains and how India could be more of a central player there. Anup Wadhawan (Chair) concluded by thanking everyone for a most interesting session in which he had learnt a lot. He suggested that the main instrument available to a policymaker is domestic competitiveness, a policy environment that encourages investment and entrepreneurship, and state-of- the-art, cost-effective infrastructure, leading to a low-cost economy.

Reference

Gupta, Poonam, Junaid Ahmad, Florian Blum, and Dhruv Jain. 2018–19. “India’s Growth Story,” India Policy Forum, Volume 15. New Delhi: National Council of Applied Economic Research. iv INDIA POLICY FORUM, 2010–11 iv INDIA POLICY FORUM, 2010–11 é 2018 15 Moore and Charity Troyer Rohini Pande, Erin K. Fletcher, Descriptive Evidence and a in India: Work and Women Policies Review of Potential Jain Sajjid Z. Chinoy and Toshi ExportsWhat and What Explains Drives India’s the Recent Slowdown? Evidence and New Implications Policy EDITED BY BOSWORTH SHEKHAR SHAH, BARRY MURALIDHARAN KARTHIK Radhika Pandey, Ila Patnaik, and Renuka San Ila Patnaik, Radhika Pandey, on Household Financial Breaks Saving in India of Impact Tax Jain Dhruv Ahmad, Florian Blum, and Junaid Gupta, Poonam Growth Story India’s and Mark Budolfson, Kuruc, Kevin Melissa LoPalo, Dean Spears ClimateVulnerability Quantifying India’s VOLUME

NCAER INDIA POLICY FORUM VOLUME 15 2018 ` 1395 789353 287191 9 ISBN 978-93-532-8719-1