Engines of Growth in India Are modern services driving economic growth?

Engines of Growth in India

Master Thesis

M.Sc. in Economic Development and Globalization

Submitted by

Bente Jessen-Thiesen (S4089588) Supervisor: Prof. Dr. Bart van Ark Co-assessor: Prof. Dr. Bart Los

i

Engines of Growth in India

Bente Jessen-Thiesen 1

Abstract

In recent history, low and medium developed countries have become less dependent on industrialization as a major driver of modern economic growth, leaving economists wondering about the prospects of long-term economic growth in these countries. The service-led growth in India raises the question whether modern services are the engine of growth of a new growth path. In this paper, I analyse the performance of modern services for 32 Indian states and find that modern services have improved economic performance beyond the main hubs of economic activity. Conducting a shift share decomposition analysis reveals that modern services have improved in terms of within and between sector labour productivity. While for communication and finance the structural change affects LP growth dynamically, for business services static structural change accounts for the most part of LP growth suggesting that business services are unable to maintain rapid economic growth in the long-term. Testing the Kaldorian growth laws allows to conclude that only manufacturing and finance can be defined as engines of growth, while the overall dynamic of sectors acting as structural burdens is diverse.

Keywords: Structural Change, Kaldorian Growth Laws, Service-led growth, Labour Productivity, Shift Share Decomposition Analysis

JEL classification: O10, O14, O4, O53, L16

ii

Table of Content

List of Tables ... iii

List of Figures ... iv

Abbreviations ... iv

1. Introduction ... 1

2. Literature Review ... 2

2.1. Structural Change and Economic Growth ... 2

2.2. The stylized facts of an “engine of growth” ... 4

2.3. The changing characteristics of sectors ... 6

2.4. Existing Literature on Kaldorian Growth Laws and Shift Share Decomposition ... 8

2.5. The Indian Growth Path ... 12

3. Data ... 14

3.1. Value added ... 16

3.2. Employment ... 16

4. Service sectors at state level... 17

5. Drivers of labour productivity growth ... 20

6. Engines of growth ... 24

6.1. Methodology ... 24

6.2. Kaldorian Growth Laws ... 26

7. Summary of Analyses ... 33

8. Conclusion ... 35

Appendix ... 36

iii

List of Tables

Table 1: Data description: Sector disaggregation ... 15

Table 2: Effect of initial performance on growth of modern services (24 Indian states) ... 20

Table 3: Shift share decomposition (1983-2017): All 32 Indian states ... 22

Table 4: First Kaldorian Growth Law (1983-2017): All 32 Indian states ... 28

Table 5: First Kaldorian Growth Law (2000-2017): All 32 Indian states ... 29

Table 6: Second Kaldorian Growth Law (1983-2017): All 32 Indian states ... 31

Table 7: Third Kaldorian Growth Law: All 32 Indian states ... 32

0,001 Table A-1: Dataset description: Years covered ... 36

Table A-2: Original data sources for value added ... 38

Table A-3: Data description: Subgroups by VA per capita ... 38

Table A-4: Correlation modern services and total VA growth: 32 Indian States ... 40

Table A-5: Shift Share Decomposition (1983-1999): All 32 Indian States... 42

Table A-6: Shift Share Decomposition (1999-2017): All 32 Indian States... 43

Table A-7: Shift Share Decomposition (1983-2017): States with low VA per capita ... 43

Table A-8: Shift Share Decomposition (1983-2017): States with medium VA per capita ... 44

Table A-9: Shift Share Decomposition (1983-2017): States with high VA per capita ... 44

Table A-10: First Kaldorian Growth Law (1983-2017): All 32 Indian states, all sectors ... 45

Table A-11: First Kaldorian Growth Law, Frist Side Test (1983-2017): All 32 Indian States ... 46

Table A-12: First Kaldorian Growth Law, Modified Frist Side Test (1983-2017): All 32 Indian States ... 47

Table A-13: First Kaldorian Growth Law (2000-2017): All 32 Indian states, all sectors ... 48

Table A-14: Second Kaldorian Growth Law (1983-2017): All 32 Indian states, all sectors ... 49

Table A-15: Second Kaldorian Growth Law (Labour Productivity) ... 50

Table A-16: Second Kaldorian Growth Law (Employment) ... 51

Table A-17: Third Kaldorian Growth Law: All 32 Indian states, all sectors ... 52

Table A-18: Third Kaldorian Growth Law (1983-2017): Groups ... 53

iv

List of Figures

Figure A-1: Sectoral shares in VA in India (1980-2017) ... 39

Figure A-2: Shares in VA for service sectors in India (1980-2017) ... 39

Figure A-3: Growth of VA in modern services between 1983-2017 and VA per capita in 1983 ... 40

Figure A-4: Growth of VA in modern services between 1983-2017 and total LP in 1983 ... 41

Figure A-5: Growth of VA in modern services between 1983-2017 and LP in modern services in 1983 ... 41

Figure A-6: Sectoral share in employment in India (1983-2017)... 42

Abbreviations

DSE Dynamic Structural Change Effect EUS Employment and Unemployment Survey GDP Gross Domestic Product

GSDP Gross State Domestic Product

ICT Information and Communication Technology ISE Intra-Sectoral Effect

ISIC International Standard Industry Classification

KGL Kaldorian Growth Law

LP Labour Productivity

MOSPI Ministry of Statistics and Programme Implementation

NSS National Sample Survey

OECD Organisation for Economic Co-operation and Development

PC Population Census

PLFS Periodic Labour Force Survey

SC Structural Change

SSD Shift Share Decomposition

SSE Static Structural Change Effect

VA Value Added

1

1. Introduction

The question of why some countries are rich and others are poor has been central in economic theory for a long time. For the past 200 years, one fundamental answer to this question has been: industrialization. Advanced economies like European countries and the USA have reached economic growth through industrialization. Today’s newly developed countries, such as the Asian Tigers (Hong Kong, Singapore, South Korea, and Taiwan) and countries such as China, Thailand, and Vietnam, have followed suit in reaching rapid economic growth through industrialization. The common wisdom was reassured: Industrialization is key to rapid economic growth.

Recently, however, another trend has become apparent: Low-income countries in Africa, Asia and Latin America have turned their back on manufacturing production growth, deindustrializing at a far earlier stage than history had predicted. This trend has left economists wondering about the prospects of economic growth in low-income countries. According to the common wisdom of manufacturing being the engine of growth, a lack of manufacturing production would result in premature deindustrializing countries staying below their economic potential. Some scholars (Dasgupta and Singh 2005, 2006; Felipe et al. (2009); Gallouj and Savona 2009; Timmer and de Vries 2007, 2009), however, have taken the recent structural developments as a reason to challenge this understanding. Is manufacturing the only engine of growth? Do the growth-enhancing characteristics, traditionally ascribed to the manufacturing sector, still apply? May other sectors have the ability to foster economic growth equally? Historically, manufacturing-led growth has been exemplified by many fast growing and well-developed countries. An equally successful growth path based on services, on the other hand, is rather rare. One outstanding exception is India. With unusually high growth rates and a surprisingly small manufacturing sector, India is an epitome of service-led growth. By using disaggregated sectoral information on a state-wise level in India, in this paper I will take a closer look at the role of services in economic growth and whether services are an engine of growth. Using value added (VA) and employment data for 32 Indian states for the period 1980-2017, I will answer whether service-led growth has taken place in all states and how structural change (SC) has influenced labour productivity (LP) between sectors and states. By testing the Kaldorian growth laws (KGLs) I will examine whether services in India are “engines of growth”.

2

2. Literature Review

Factor productivity and the allocation of resources are fundamental determinants of economic success. Factor productivity obviously varies between sectors and regions. In many basic growth theories, the allocation of resources is assumed to be efficient. The theory on SC, however, has proven this broad assumption wrong. Resources, like labour, can be and are constantly reallocated between sectors and, if allocated efficiently, improve economic performance significantly. In the following of this chapter, I will outline the relationship of SC and economic growth in more detail. A brief analysis of the historical and recent patterns of SC will provide insight on what makes a sector an “engine of growth” and how these sectoral characteristics might have change in the past decades. Providing a broad overview on previous applications of the methods used in this paper, I will discuss the Indian growth path and my hypotheses for the following analysis.

2.1. Structural Change and Economic Growth

The generation of long-term economic growth and development has been a matter of broad interest and intense debate in academia. Traditional growth models often reduce economic growth to the improvement of productivity through technological change. Development economists, however, have found the allocation of inputs, such as labour, between sectors to have significant impact on economic growth as well. When analysing the performance of an economy, one fundamental question should be where the growth origins from. Has productivity increased? And if so, is it because a sector has improved its production processes causing within-sector productivity growth or because the economy has shifted its inputs more effectively?

Structural change is defined as the shift in the relative importance of sectors in terms of output or employment (Syrquin 1988). As suggested above, the LP in an economy can grow through to two different effects. First, technological change or the more efficient allocation of inputs across plants can reduce the labour needed to produce the same amount of output. In this case, the economy-wide LP grows due to within-sector productivity growth. The second effect is that labour moves from a low-productivity sector to a high-productivity sector. Though no nominal productivity improvement within each sector has occurred, the overall productivity of the country increases as high-productivity sectors grow in importance and with it the economic output. This effect is called growth-enhancing structural change (SC).

The earliest contributions to the matter of the effect of SC on economic growth were brought forward in the 1930s, when economists started differentiating between primary, secondary, and tertiary sectors (e.g. Fisher 1935, 1940). In his Nobel lecture in 1973, Kuznets named structural transformation2 as one of six characteristics of modern economic growth. More and more focus was set on the close relationship of the growth of manufacturing output and rising per capita income (Kuznets 1957; Clark 1940; Thirlwall 1983). In 1954, Lewis introduced a dual sector model, in which economic growth is generated through the shift of labour from the agricultural sector to the manufacturing sector. In his model, Lewis finds that the traditional agricultural sector absorbs large amount of surplus labour. Shifting this labour from agriculture into

3 manufacturing, will lead to LP growth in both sectors. Though Lewis’ dual sector model, naturally, simplifies the sectoral structure of an economy, it captures the fundamentals of growth-enhancing SC. Releasing labour from a low-productivity sector, like agriculture, and shifting it into high-productivity sectors, like manufacturing, increases overall LP and output. In history, many well-known examples of this growth-enhancing SC can be found. Beginning with the Industrial Revolution in the 18th and 19th century, today’s advanced countries like the United States or the United Kingdom have faced enormous economic growth by shifting labour from the agricultural sector to manufacturing production. Similarly, but slightly later, Japan followed their rapid growth path through industrialization and soon became the fourth biggest exporter of cotton yarn (Landes 1998). Most of the academic work on SC is based on these cases, but the growth effect of SC has strong examples in the recent history as well.

Between the 1960s and the 1990s, the so-called “Asian Tigers” (Hong Kong, Singapore, South Korea, and Taiwan) showed that becoming a developed country through industrialization was still possible. By exporting manufacturing goods, the countries reached enormous growth rates and managed to move from low levels of development to high development. The most recent examples for growth through industrialization are countries like China, Thailand, Malaysia, and Vietnam. In Malaysia, for example, the share of manufacturing in VA has increased from 19.8% in 1987 to 30.9% in 2000. In the same period GDP per capita has more than doubled3_{. Again,}

manufacturing-led growth as predicted by historical experience had taken place. Although these countries seem to confirm historical patterns, many other countries have not followed the same path. Before getting into the most recent developments of SC, a closer look into the regularities of SC as it has been experienced in the past, should be taken.

The correlation of sectoral shares and economic growth has been following clear pattern in history. As Lewis (1954) suggests and is unquestioned by almost unexceptionally all development economists, the growth of the share of agriculture in the total VA is negatively correlated with economic growth. As agriculture is the by far the most important sector in low-developed countries, moving out of this sector, will almost certainly increase the LP of the economy and thus lead to economic growth. While SC theorists have given only little attention to services, they found the share of manufacturing production to be positively correlated with output growth. Beginning in the 1950s, high developed countries have started to move out of manufacturing leading to the share of manufacturing in both employment and VA decreasing (Rodrik 2016). For lower levels of development, however, manufacturing share and total VA continue to be positively correlated (Herrendorf et al. 2014). The stylized relationship of manufacturing share and total VA growth, hence, suggests that countries keep industrializing until they reach a certain point of economic development, making the relation an inverted U-shape.

The relationship of service shares and total VA has been of only little attention in the first decades of SC theory and is still subject of debate. Some, like Kuznets (1957) and Kongsamut et al. (2001), suggest that the share of services remains relatively constant over the stages of development. Others, like Herrendorf et al. (2014) and Chenery and Taylor (1968), find that the share of services increases with the stage of development. Bringing these two perspectives

4 together empirically, Eichengreen and Gupta (2013) suggest that there are two waves in the development of service shares in VA. The first wave occurs when countries move from low income to middle income countries and is characterized by mostly traditional services evolving. The second wave occurs when countries move beyond middle income and leads to a growing share of modern services like IT and financial services. Today, “the positive association between the service sector share of GDP and per capita income is one of the best-known regularities in all of growth and development economics” (Ibid, p.96). Although modern services have evolved at a relatively early stage of development, India can be allocated to the second wave (Basu 2018).

When Kaldor (1966) proposed his growth laws, he based his findings on the, until then, positive relationship of manufacturing and total VA. By setting a number of stylized facts, Kaldor aimed to explain what is unique about the manufacturing sector as an engine of growth. I will discuss these “stylized facts” in the following subchapter.

2.2. The stylized facts of an “engine of growth”

As Dasgupta and Singh put in, “Kaldor was renowned as an apostle of industrialization” (Dasgupta and Singh 2006). In 1966, in a lecture presenting the “causes of the slow rate of economic growth of the United Kingdom”, Kaldor first brought forward a number of stylized facts on the sector economies which are today known as the “Kaldorian Growth Laws” (KGLs). The essence of the regularities Kaldor found is that manufacturing accumulates all the characteristics that are needed to be an “engine of growth”. The most important characteristics will be discussed now.

In the Lewis model, output growth is possible, because the manufacturing sector is able to extend production continuously. While limited resources bound the agricultural sector’s output, manufacturing production can increase their inputs. Its production is limited by demand. Historically, for most sectors output is limited by the domestic demand, but Kaldor claims that for the manufacturing sector this is not the case. The internationalization of trade has allowed the production of manufacturing goods to go beyond domestic demand, allowing countries to increase their manufacturing sector and make use of their comparative advantages through specialization. This means that the tradable manufacturing production can grow much more than the production of non-tradable goods while at the same time specialization allows for high LP. Historically, this great advantage of tradability is assigned almost exclusively to the manufacturing sector. As everybody’s daily consumption and recent trade statistics show, this clear distinction between tradable manufacturing products and non-tradable non-manufacturing products is a very rough and rather unrealistic generalization. Certainly, some agricultural goods but more importantly some services are. For the agricultural sector, a weaker rate of tradability is still reasonable but for services, the elasticity of demand is harder to generalize. While traditional services like transport service or community service are by nature not tradable, modern services like communication and business services are quite tradable at least since the “service transformation” in the 1970s/1980s (Ghani and Kharas 2010; Eichengreen and Gupta 2013).

5 has strong linkages to other sectors. When manufacturing production increases, forward and backward linkages to other sectors make the rest of the economy grow as well. To give an example, a tire factory in India will need to buy raw materials like rubber to produce its product. As India is one of the largest producers of natural rubber (Haan et al. 2003), the tire factory will probably buy its raw materials from a local rubber producer and thus foster economic growth in the agricultural sector. Once the tires are produced, services will be needed to sale the tires to consumers and the transport sector will be needed to carry them there. Just like that, the increase in manufacturing production has led to growth in other sectors. For manufacturing sector this storyline is easily understood, but for other sectors, like services, it is not as straightforward. Although other sectors also have inter-sectoral linkages, Kaldor finds these linkages to be strongest for the manufacturing sector. Hence, the manufacturing sector is best able to foster economic growth within and outside its own sector

A third characteristic of an engine of growth, besides being able to extend the sectoral output and fostering total VA growth beyond its own sectoral output, is that the LP of an engine of growth sector should be growing with the size of the sector. Kaldor argues that in the manufacturing sector technological and knowledge spillovers are stronger than in other sectors. The simplest view on SC is looking at the shift of workers from one sector to another with sectors having different levels of LP. This shift of labour towards more productive sectors is also called a positive static SC effect (SSE). Assuming these levels of LP to be untouched by the changes in employment, however, includes the assumption that sectors are subjects to constant returns to scale. As Lewis points out in his model already, this does not have to be the case, for example when surplus labour is absorbed in the agricultural sector. Kaldor assumes that the agricultural sector is subject to diminishing returns to scale, meaning that the LP of an additional worker is smaller than the average LP in the same sector. The SSE, which is larger if the difference in LP between two sectors is large, will be overestimated if a shifted worker does not adopt the same LP rate. At the same time, a sector with increasing returns to scale is able to increase its sectoral LP when the sectoral output grows. An additional worker would improve the LP of the sector even more and thus increase the previously observed SC effect. This second aspect of SC is called the dynamic SC effect (DSE). With increasing returns to scale, a larger manufacturing sector will lead to a higher LP and thus, the manufacturing sector has a positive DSE. Kaldor’s proposition of the manufacturing sector to be constantly growing in LP has recently been reassured by Rodrik (2013) who found evidence for the “unconditional convergence” of LP in the manufacturing sector. He states that it does not matter if a country is open to the international markets or not, the LP in the manufacturing sector will always converge with the rest of the world. Interestingly, the debate on which sectors are subject of productivity convergence suggests that the distinction between converging and non-converging sectors can be made differently than by splitting into manufacturing and non-manufacturing production. Inklaar and Diewert (2016) suggest that traded sectors are converging faster than non-traded sectors, which again raises the question of which sectors are tradable. Kinfemichael and Morshed (2019) find LP in the services to be converging.

6 labour should be moved out of this sector, but can agricultural workers be shifted to other sectors that simply? Workers in manufacturing production need to handle materials correctly. Workers in business services need to be literate and workers in transport services need to know how to operate vehicles. Sectors differ in the skill composition of their labour and the larger the differences in skill composition, the less mobile are worker across sectors. At the time of Kaldor, the skill requirements in manufacturing were relatively low compared to other high-productivity sectors. Shifting labour “from the farm to the factory” was accompanied by lower skill barriers than it was for other sectors, thus labour mobility was high between those two sectors. At the same time, manufacturing was able to not only absorb the “right kind” of labour but also large amounts of it, which allows manufacturing make use of the abundance of labour. With the rise of digitalization and robotization, the skill composition of sectors is likely to be changing. Not only might the manufacturing sector require higher skilled labour in the future, but also the characteristics of other sectors are changing. I will discuss these changes in the following subchapter.

2.3. The changing characteristics of sectors

Historically, the correlation of the share of manufacturing and economic growth has been inverted U-shaped, while service have been considered to be driven by domestic demand and are only growing in share on a higher stage of development (Eichengreen and Gupta 2013). Though the original work by Kaldor (1966) disagrees, the deindustrialization of high developed countries is not considered a “problem” in general. For most developed countries, the manufacturing share had been increasing at their early stages of development and when income per capita was sufficiently high, developed countries began to deindustrialize. It is the deindustrialization of today’s developing countries that economists are concerned about. Low and middle income countries all around the world are turning their backs on manufacturing activities into services “without having gone through proper experience of industrialization” (Rodrik 2016, p.2). This trend is what economists call “Premature Deindustrialization”. Studies have found that the turning point of industrialization to deindustrialization has historically been at a per capita income level of about 10,000 US$ in current prices and decreased to around 3,000 US$ recently (Dasgupta and Singh, A. 2006). The concerns raised with respect to premature deindustrialization rely on historical experience and the sectoral matching of the previously outlined characteristics. Voices have becoming stronger suggesting that these matches have changed.

For the longest time of SC theory, the service sectors have been considered “non-tradable, menial, low productivity and low-innovation” (McCredie and Bubner 2010). With the rise of information and communication technologies (ICTs) and the so-called computerization starting in the 1970s and 1980s, these characteristics have become more and more blurry. Modern services like communication, financial services, and business services are tradable, highly productive, and certainly not menial in some economies. Primo Braga et al. (2019) argue that service economy and the digital revolution are not sufficiently reflected in international policy making and that “(s)trong, sustainable and inclusive growth will not be achieved without due consideration of services” (Primo Braga 2019, p. 3).

7 the world economy is experiencing something that some call the “third industrial revolution” This transformation of sectors is driven by computerization, digitalization, robotization and globalization (Leighton 1970, Liu and Grusky 2013, Lind 2014). While services have previously been mainly domestic goods, Jensen and Kletzer (2005) estimate that theoretically, 70% of the professional and business employment in the US could be offshored. Indeed, service exports have been increasing substantially over the past decades and became more important in the economic structure of many developing countries (Ghani and O'Connell 2014).

The implications of the fundamental changes due to the third industrial revolution have been subject of debate in many areas of economic research. Economists try to ascertain what it implies for trade, employment, and economic growth in both developed and developing countries, but after all, the perception of growth-enhancing SC has remained focused on industrialization. In a very recent thought experiment, Baldwin and Forslid (2020) draw attention to what they call the “Globotics Transformation”. After the great transformation of labour shifting from “the farm to the factory” and the more recent transformation from “the factory to the office”, they argue that with globalization and robotization we are currently facing a new transformation that will have even stronger implication than the computerization transformation we have faced in the past decades. They ask what it would mean for economic development if “manufacturing becomes jobless, but most services are freely traded” (Baldwin and Forslid 2020).

8 The debate on whether and to what extent the characteristics of manufacturing and services have changed over the past decades, fundamentally affects the way, economists look at the developments of premature deindustrializing and service-led growth economies. A broad overview on recent studies of services as an engine of growth will be given now.

2.4. Existing Literature on Kaldorian Growth Laws and Shift Share Decomposition

The three stylized facts of the correlation of sectoral shares and economic growth have been tested by various scholars. Herrendorf et al. (2014) test the correlations for a number of developing and developed countries for the period 1970-2000. They find that the three patterns of SC, the negative relation of agriculture, the inverted U-shaped relation of manufacturing and the positive relation of services with GDP, indeed hold internationally. Looking closer into developing countries, Bah (2011) finds that the patterns of SC are quite different between today’s developing countries. He finds that, though for Asian countries the pattern seems to be relatively similar among themselves and the historical structural pattern, African and Latin American countries’ structural patterns are diverse. These findings provide a first insight on the relation of SC and economic growth, but to determine the origins of economic growth, a closer look into SC is needed. Though there are many other approaches to analysis SC and economic growth (e.g. Herrendorf et al. 2014; Reddy 2015; Foellmi and Zweimüller 2002), in this paper I will focus on the SSD analysis and the testing of the KGLs. An overview on what has been found for these two approaches will be given, now.

9 growth-enhancing. In their analysis of productivity growth of the manufacturing sectors in five Asian countries for the period 1973-1993, Timmer and Szirmai (2000) find that SC has not been growth-enhancing on average, but for India, both SSE and DSE have been influencing LP positively. They further find that the ISE is the most important driver of LP making up for between 85% to 91% of sectoral LP growth.

Timmer and de Vries (2009) conduct an SSD for 19 Asian and Latin American countries covering ten sectors. Again, they find that SC effects are relatively small compared to within-sectoral productivity growth. Contrary to traditional expectations, they find that LP growth in market sectors (trade, transport, communication, and finance) have a larger impact on overall LP growth than the manufacturing sector. For India specifically, Timmer and de Vries’ findings suggest that the contribution of manufacturing and services have been relatively equal in the 1960s, but while manufacturing productivity growth does not show a trend, the contribution of market and non-market services to LP growth has increased in the periods 1970-1979 and 1979-2004.

When Kaldor set the conditions for a sector to be an engine of growth, he directly and indirectly included the three components of the SSD in his growth laws. The claimed that the laws apply only to the manufacturing sector. As discussed above, this might have changed in the past decades. Testing for the KGLs allows to test if the stylized facts still hold for the manufacturing sector in a service-led economy and whether they also apply to other sectors like modern service sectors. The three KGLs can be summarized as following:

10 The first empirical tests of the KGLs had been conducted by Kaldor (1966) himself for 12 OECD countries. While he conducted the analysis for the period 1953-1963, Cripps and Tarling (1973) extended his work with data for the period 1950-1970. Since then, several studies testing (parts of) the KGLs for different geographical and developments stage groups, but only few studies have analysed services on a disaggregated level. In 2003, Wells and Thirlwall test the KGLs across Africa and find at there is some empirical support for the KGLs. They test the three laws for 45 African countries over the period 1980-1996 and conclude that the manufacturing sector is an engine of growth. Additional to the three original tests, Wells and Thirlwall conduct two “side-tests” for the first KGL with which they aim to overcome the possible bias resulting from the endogeneity of manufacturing output in total output. Both tests confirm the role of manufacturing as a driving force in output growth. For services, Wells and Thirlwall cannot find a significant correlation of sectoral output growth and total output growth. Dasgupta and Singh (2006) conduct a Kaldorian analysis for over 40 developing countries including India. They conducted their analysis for the period 1990-2000 and use two equations to test the three growth laws. They find the first KGL to hold robustly with a correlation coefficient of manufacturing below 0.5 which suggest that the larger the difference between manufacturing growth and GDP growth, the greater the latter (Dasgupta and Singh, A. 2006). Testing the first KGL for the agricultural sector and for the aggregated service sector, they find that the agriculture is not growing faster than the overall GDP, but the service sector is. Similar to the findings for the manufacturing sector, they find the explanatory power of the growth of service output to be high (R-squared is 0,98). Also, the inter-country variation can be explained relatively good by the variable (0,85). The correlation coefficient for services, however, is somewhat higher than for manufacturing, suggesting that manufacturing still grows faster than services relative to the overall economy. For the second and third KGLs, Dasgupta and Singh tests the relation of the sectoral output growth and sectoral employment growth with overall productivity growth. Though they find significant results in a similar analysis in 2005, Dasgupta and Singh’s results in 2006 are not robust. To loosen the assumptions, they decide to replace non-manufacturing employment by agricultural employment and now find robust results. When testing the law for the service sector instead of manufacturing, they again find the law to be true. Though the regression does not pass the normality test, they confirm the aggregated service sector as an engine of growth.

In an earlier work, Dasgupta and Singh (2005) conduct a cross-sectional analysis testing the first KGL for 29 Indian states for the years 1993 and 1999. The divide the economy into three sectors and find that for all three sectors VA growth is positively correlated to total VA growth. While agriculture and manufacturing provide a correlation coefficient smaller than one, the coefficient for services is larger than one. This suggests that statistically, services are growing slower than the overall economy.

11 bearing the fact that LP is significantly lower in informal manufacturing than in formal manufacturing. When analysing the growth of service and manufacturing with respect to the overall growth, the question of causality arises. Traditionally, one would argue that service grow because of manufacturing growth. As discussed above, inter-sectoral linkages from manufacturing to services exist but Dasgupta and Singh (2006) stress that this causality might hold for traditional services like transport or retail but is rather unlikely for modern services like IT and business services. Much more, for these services it could be the other way around, where services foster manufacturing growth.

Another great contribution in the recent testing of the KGLs for developing countries, has been presented by Di Meglio et al. in 2018. As the first, Di Meglio et al. have conducted their analysis for services on a disaggregated level. Instead of looking at the three sectors agriculture, manufacturing, and services only, they. split services into four sectors. Testing the KGLs for 29 developing countries of Asia, Latin America and Africa, Di Meglio et al. find that the KGLs hold not only for the manufacturing but also for business services.

For the first KGL, Di Meglio et al. (2018) find positive and significant results for all included sectors. All correlation coefficients are below zero, which indicates that statistically the sectoral output grows faster than the total output. Following Wells and Thirlwall (2003), they also conduct the two side tests. The only sectors passing these tests are the manufacturing sector and the business sector. It should be noted here that the business sector as Di Meglio et al. (2018) defines it, includes “Financial Intermediation”, which in the following analysis is a separate sector.

Instead of testing the third law directly, Di Meglio et al. turn to the SSD analysis. In fact, the two SC effects that can be detected by the SSD can be interpreted as the two elements of the third KGL as both approaches try to explain the origins of LP growth. For all 29 developing countries, Di Meglio et al. (2018) find that both, the overall SC effect, and the ISE are positive. The DSE, however, overall diminishes the LP growth of the economies suggesting that the labour shifted has been less productivity-enhancing than the static effect would have suggested. Looking at the sectoral level of the SSD, they find that the negativity of the DSE is the case for all sectors except for the sector “transport and communication”. The only sector for which both SC effects and thus the total SC effect is negative is the agricultural sector. For all other sectors, except “Other Industry” the overall SC effect is positive.

12 2.5. The Indian Growth Path

Although premature deindustrialization has many examples in the recent history, there are only few success stories for service-led growth. While other South Asian countries have been experienced it as well (Ghani 2010), the most prominent case of service-led growth is India. As mentioned before, the Indian growth pattern is different to what would have been predicted by history, but it also differs from the growth pattern of most other countries today. While India performs well in services, its manufacturing performance is remarkably poor. India’s economic growth has been service-led at least since independence (Ghose 2014), with especially modern services, like communication, finance, and business services, are increasing their share in total VA constantly. At the same time, the share of manufacturing in economic VA has never exceeded 19%4.

Many questions arise from the Indian experience. Why has manufacturing in Indian never really taken off? Which services have led to the economic growth and has it been taking place all over the country or isolated in the so-called brain hubs? To get an overview of the Indian economy, a look at the economic structure before the third industrial revolution should be taken. In the beginning of the 1980s, the Indian economy was far less service-oriented than it is today. Manufacturing accounted for around 15% of the total VA while service accounted for around 40%. Comparing these numbers to countries of a similar stage of development, the share of manufacturing in VA is not significantly smaller. Rather surprising, however, is that the share of services is small compared to similar countries (Kochhar et al. 2006). Over the past four decades the sectoral composition of VA has changes. While manufacturing has increased only little up to 18.3% in 2017, services, and especially modern services, have increased in shares substantially. While modern services had accounted from 16.5% in 1980, in 2017 they accounted for 22.6% of the total VA. In the same time span, the aggregated service sector grew from 40.1% to 51.4%.

The atypical sectoral structure of the Indian economy has various explanations. Strictly speaking, India has never truly experienced a deindustrialization as the manufacturing sector has just never really taken off (Dasgupta and Singh 2005). The low share of manufacturing can be explained by different structural bottlenecks. One of these bottlenecks is the lacking infrastructure which is important for the transport of goods but rather unimportant to export services. Another reason is that India is relatively far away from important markets like the US or Europe and thus bears relatively high transport costs. Though, especially after independence, policy makers tried to push the manufacturing production, the Indian manufacturing sector has remained fairly small. Kochhar et al. (2006) find that the manufacturing sector has been relatively skill-intensive in the 1980s already. By taking a closer look at the state-wise sectoral structure of the Indian economy, they also find that there is a direct link between the existence of a large skill-intensive manufacturing industry and the subsequent success of modern services. With the rise of digitalization and computerization in the 1980s, the rapid growth of modern services has begun in India. Today, India is known for its world-class service hubs (Kochhar et al. 2006). Cities like Bengaluru and Chennai are known for their successful software activities and cities like Mumbai are known for their large finance sector. But have services grown in all

13 parts of the countries? Kochhar et al. (2006) find that the growth of Indian state has been diverse. One possible explanation could be that state government have become more important for economic policy making since the 1990s, but they do not make a statement about whether services are the driver of this divergence. By looking state-wise sectoral data I will examine whether this is the case and which services have improved state performance most. For the analysis I propose the following hypothesis:

H1: The growth of modern service sectors has centralized in well-developed states in India. States with high labour productivity previous to the third industrial revolution have benefitted most from modern service sector growth.

The great success of modern services in India can be explained by many different factors. Basu (2018) argues that the low language barriers with important developed countries like the US and the UK are a major comparative advantage for Indian services. Eichengreen and Gupta (2013) highlight that the democratic social structure is an important commonality of countries that succeed in service exports at such an early stage of economic development. Through to high investments in tertiary education and their experience in skill-intensive manufacturing India also has an important basis for skill-intensive labour, which might have led to its comparative advantage in skill-intensive services (Ghani 2010, Kochhar et al. 2006, Basu 2018).

In the rise of services, two political turning points have had great impact. In the beginning of the 1980s, important reforms have been introduced by the India government. Through import liberalization, the “delicensing” of industries and tax incentives for exports, the strict regulation of economic activities were relaxed. Rodrik and Subramanian (2005) characterised these reforms as “pro-business” and indeed afterwards services started to expand (Cortuk and Singh 2011). A second wave of reforms between 1991-1993 led to the liberalization of market by reducing bureaucratic barriers, the reduction of public monopolies and important liberalizations of the financial and insurance sector. Dossani and Kenney (2007) find that in the 1980s and 1990s, it was mainly the software services that benefitted from the reforms, but by the end of the 1990s and the 2000s, they identify a “second wave of globalization” which lead to the rapid growth of offshoring activities like call centers.

The different services have certainly led to an increase in economic performance, but it is unclear what has driven the growth of output and LP. By conducting an SSD, I will test whether SC has influenced LP growth positively. As services tend to be highly productive, shifting labour into services is likely to have a growth-enhancing effect. Different to the straightforward assumption of Kaldor for the manufacturing sector, it is unclear whether services are dynamic sectors, meaning that they are subject to increasing returns to scale. Just as likely, the sectoral growth might lead to decreasing LP. For the SSD analysis, I therefore propose the following hypothesis:

14 Directly and indirectly, the three components of the SSD include the stylized facts laid out by Kaldor (1966). In the third part of my analysis, I will test whether his growth laws apply the manufacturing sector and the modern service sectors in India. Though on a low level, the manufacturing sector has improved its share in total VA overtime and is likely to still hold the characteristics which were appointed to it by Kaldor. Modern services have played a key role in India’s economic growth. Though likely to hold structural disadvantages compared to the traditional manufacturing sector, modern services drive economic growth in India. For the testing of the KGL for these sectors, I therefore propose the following hypothesis:

H3: Manufacturing and modern services are engines of growth for the Indian economy. Output growth of modern services are expected to be closely correlated to the overall economic performance. Labour productivity growth within the observed sectors as well as for the overall economy are driven by modern service sectors.

3. Data

For the following analysis of SC, the two variables, output and employment, are essential. As there is no readymade dataset for these two variables on a state level in India, I have created a dataset on my own. The dataset uses VA in 2011 constant prices in Lakhs as the variable for output and persons usually employed in principal status as a proxy for employment. While VA data is available as time series for the period 1980-2017, sectoral employment is available for eight years over the period 1983-20175. Both variables cover 32 of the 36 Indian states. A detailed overview on the state and sector coverage can be found in the Appendix Table A-1. The level of sectoral disaggregation of this dataset, goes beyond ordinary. While most studies on SC include only one service sector, some studies disaggregate services into four or five sectors. In the dataset created for this study, the economy is divided into 12 sectors, of which seven are service sectors.

The Republic of India is split into 28 states and eight union territories. Due to a lack of data and the reformations of states, the time periods and sectors covered in this dataset vary for some states. The following states and union territories have been excluded from the dataset: Telangana, Dadra and Nagar Haveli, Daman and Diu, Ladakh, Lakshadweep. The sectoral division is based on the International Standard Industry Classification (ISIC) Revision 3. For more detailed information on the service sectors, two sectors have been further disaggregated. First, the ISIC Rev. 3 sector “Transport, Storage and Communication” has been split into “Transport and Storage” and “Communication”. Second, the sector “Financial Intermediation” has been split into “Finance and Insurance” and “Real Estate, Renting and Business Services”. For the matter of flow of reading the sector names have been reduced to one of the components of the sector. An overview of the sectors and their composition is given in Table 1. Before discussing the construction process of the two variables in detail, a short comparison of the new dataset with the widely used India KLEMS dataset will be made.

15 To put the new dataset into context of other existing datasets, to my knowledge, there is no dataset for VA and employment on a state-wise level for India publicly available. On a country level the most popular database covering both variables is the 10-Sector Database by Timmer et al. (2015). For the period and country of interest, however, Timmer et al. rely on the India KLEMS database6. The India KLEMS database provides several indicators relevant for a sectoral productivity analysis on an all-Indian level. Assuming the KLEMS database to provide the most realistic numbers, it is useful to compare the new dataset to it. The VA data in my dataset is provided as a time series covering all years from 1980 to 2017. Comparing the all-India data with the KLEMS database, suggests a strong coverage for the past decades with recent coverage around 98%. For earlier periods, state-wise data is missing more often, which leads to a decreasing coverage of all-Indian VA. The lowest coverage of the dataset compared to the KLEMS database can be found for 1984 which is equal to 76%. For employment, the various definitions that can be applied make a comparison of datasets less straight forward. Using the KLEMS database as reference, my dataset covers around 90% of total employment for all years. The main source for employment of the KLEMS database are the Employment and Unemployment Surveys (EUS). Additionally, they use Annual Industry Surveys and Economic Surveys to estimate specific sectors. As will been discussed below, the EUS allow for various definitions of persons employed. Contrary to my dataset, the KLEMS database includes persons usually employed in subsidiary status. Consequently, my dataset structurally underestimates total employment compared to the KLEMS database.

Table 1: Data description: Sector disaggregation

Sector Definition

Agriculture Agriculture, hunting, forestry, and Fishing

Mining Mining and quarrying

Manufacturing Manufacturing activities, publishing Utilities Electricity, gas, and water supply Construction Construction

Trade Wholesale and retail trade, repair of motor vehicles, motorcycles and persons and household goods, hotels, and restaurants

Transport Transport and Storage

Communication Postal and Courier services, Information and Communication services Finance Financial intermediation, Insurance and Pension funding

Business Real Estate, Renting and Business activities

Public Services Public administration and defence, compulsory social security, Education, Health, and social work

Other Services

Other community, social and personal service activities, activities in private households as employers and undifferentiated production activities of private households, extraterritorial organizations, and bodies

6 _{Reserve Bank of India. The India KLEMS Database (July 2019). Available at:}

16 3.1. Value added

Value Added (VA) data has been extracted from the Ministry of Statistics and Programme Implementation (MOSPI) using Gross State Domestic Product (GSDP) at factor costs at constant prices. The original data has been reported in five different series with the time series overlapping. The most recent data is given in 2011 constant prices7. As there are conceptual differences between series, adjusting the base years to create one time series is not possible. Instead, I used the annual growth rates to extent the most recent time series. Assuming that more recent data is the most reliable, datasets including more recent years have been preferred over datasets covering a longer time period. Detailed information on the processing of the original data can be found in the Appendix Table A-2. Value Added is given in Indian Rupees (₹) in Lakhs which is equal to 100 thousand Indian Rupees (1 Lakh = 100.000). When available, the VA data provided includes information on organised and unorganised manufacturing activities.

For the following analysis, the 32 states have been split into three groups based on their relative level of VA per capita. Based on VA and estimated population data for 2017, the states have been split into three equally sized groups. The state composition of groups can be found in Table A-3.

3.2. Employment

To construct a dataset of state-wise sectoral employment, EUS provided by MOSPI have been used as the main source. The EUS are part of the National Sample Surveys (NSS) which have been conducted for a number of years in the period of interest. The time span between surveys varies, as does the extent for the NSS. The NSS used for this analysis are the 38th NSS (1983), 43rd _{NSS (1987), 50}th _{NSS (1993), 55}th _{NSS (1999), 61}st _{NSS (2004), 64}th _{NSS (2007) and 68}th

NSS (2011). Due to data irregularities the 62nd NSS (2005) and the 66th NSS (2009) have not been included. To estimate recent employment the Periodic Labour Force Survey (PLFS) conducted for 2017 is used.

The EUS provides employment information using different approaches. For this dataset, I use the employment definition of activity statuses, which does not reflect the hours worked by an individual but rather the number of persons employed. Again, there are different definitions of the activity status based on either annual, weekly, or daily activity. In this analysis, the usual status approach is used, which categorizes people based on the reference period of one year. For reasons of consistency only those persons working in “usual principal status” have been included, which means that persons working in “subsistence statuses” are excluded. In line with the VA data, manufacturing employment includes organised and unorganised activities. A major issue of the EUS are the conceptual differences between years, which lead to differences in the total numbers of employment. As for VA the conceptual differences seem not to have a large impact on the sectoral shares but rather on the total number of employments. As discussed by Das et al. (2018) total population predicted for the NSS is significantly lower than the total population found by the population censuses (PC). To overcome this problem, the total numbers of employment reported by the EUS needs to be normalized. To do so, I used the

17 worker population ratio (WPR) reported by the EUS reports combined with total population data reported by the PC to estimate total employment. Adjusting the sectoral employment to the total employment numbers thus provides the number of persons usually employed in principal status.

It should be noted that the PC years and the EUS years are not equal. The decennial population censuses have been conducted for the year 1981, 1991, 2001 and 2011. To estimate a time series, the growth rate of population between two censuses has been assumed to be linear. For years prior or after the included PCs, population has been estimated using the closest growth rates available. The PC reports and some EUS reports provide state-wise population for rural and urban population separately only8. Though this is usually not a problem for the calculation of total WPR, in some cases the rural-urban shares differ substantially between two censuses. In these cases, again I assume recent data to be most trustworthy and estimated shares based on the all-Indian development instead9_{. As information on WPR for the 43}rd _{NSS (1987) is missing,}

WPR therefore has been estimated using linear interpolation between 1983 and 1993.

The all-Indian WPR for this definition of employment between 1983 and 2011 varies between 40.4% and 35.3%. In 2017, the WPR decreases to 31.8% which, on the first sight, is a rather surprising finding. Indeed, there has be a debate on whether the decrease in work participation is due to methodological changes. Following the academic debate, suggests that the different surveys, PLFS and EUS are comparable and thus the decline reflects reality (Jajoria and Jatav 2020). Kannan and Raveendran (2019) find that the net job loss has taken place mostly by a decline in labour force participation of less educated women in rural areas and can be explained by various structural reasons like the shift of male population to urban areas or the decline of employment in public administration.

4. Service sectors at state level

India is one of the largest and most populated countries in the world with an estimated population of 1.3 Billion people and an overall VA of 119,698 Billion ₹ in 2017. Both, population and VA, have been growing quite rapidly over the past decades. Between 1980 and 2017 the economy-wide annual growth rate of VA is 6.68%. While the growth rates have been increasing between 1980 and 2000, in recent years, the extension of the Indian economy has been slightly slowing down. A similar picture can be drawn when looking at the growth of VA per capita. Over the full time period annual growth of VA per capita has been around 4.7% with growth rates constantly increasing till the 2000s and recently slowing down.

To give an idea of the overall economic development in India, over the past four decades the output of all sectors in India has been growing with agriculture accounting for the largest VA share until recently. While in 1980, agriculture accounted for around 45.7% of VA, in 2017 only 16% of total VA were generated in the agricultural sector. In Figure A-1, the trends of sectoral shares in VA are displayed for the agriculture, manufacturing, traditional and modern

8 _{In 1981, the total population is even further separated into rural female, rural male, urban female and urban} male.

18 services. Manufacturing and both service sectors have increased their shares over the four decades, but since the 1990s modern services is the only sectors with a constantly growing share. As discussed above, the dataset used for this paper allows to disaggregate the service sector into seven subsectors, which can be grouped into modern and traditional services. Modern services include the sectors communication, finance, and business, while traditional services include trade, transport, public services, and other services. Over the full period, most service sector shares have grown only slowly with average annual growth rates between -0.05% in business and 0.09% in transport (Figure A-2). Communication and finance, on the other hand, have experienced more significant growth in their shares with average annual growth rates of 5.9% and 4.4%. A closer look into subperiods helps to understand the evolution of sectoral shares better. In the 1980s, only the finance sector has been growing in share significantly. While some traditional services have been growing as well, communication has grown annually by only 0.4% and the share of business has even declined. The patterns change in the 1990s when communication started to grow extensively. In the 2000s, the growth of communication shares peaks with an annual growth of 11.5%. The finance sector experienced its fastest growth in the 1980s. Though slowing down, the growth between 1990 and 2017, remains rapid. Overall, communication and finance appear to be the drivers of the service-led growth in India. Interestingly, business services have not growing in share substantially. Over the full period their share has even declined. The disaggregated look at Indian service sectors provides an interesting insight to the structure of India’s service-led growth, but at least as interesting is the question of centralization of these economic activities across states.

Before looking into the development of service sectors across states, it should be noted that the growth of VA per capita varies quite strongly across states. In 1980, VA per capita in 2011 constant prices ranged between 7,341 ₹ and 52,728 ₹. While in all states VA per capita increased over time, levels have diverged. In 2017, the highest VA per capita was found for Goa (372,762 ₹) but also other states like Karnataka and Maharashtra which are known as India’s brain hubs (Bengaluru and Mumbai) have been growing substantially. The fastest growth rate for the full time period was found for Gujarat, which used to be one of the poorest states in 1980 and today ranks in the upper quarter in VA per capita. Its annual growth rate was 6.7%. Constant for all years, Bihar, one of the most populated states, has been the least developed. In 2017, Bihar’s VA per capita in 2011 constant prices was 28,760 ₹ which is only 8,77% of VA per capita in Goa. To set in context, converted into current US$, VA per capita in Bihar is about 1.58 US$ per day.

Regional inequality is a large topic for economists and policy makers in general and in the case of India. There is significant literature on the matter of regional divergence in India (e.g. Ghosh, Marjit, and Neogi 1998; Sachs, Bajpai, and Ramiah 2002; Kar and Sakthivel 2006; Chakravorty 2003). In this paper, I will not go deeper into the analysis of regional divergence in general but rather ask whether services specially have been growing unequally across states.

19 decreasing shares are equally distributed between the three groups of development levels10. For most of these seven states, the share of modern services has been above average in 1980 but is below average in 2017. The VA generated by modern services has increased in all seven states, what catches one’s eye, however, is that in all of them either traditional services or manufacturing have been growing rapidly. For example, in Manipur, the share of traditional services has increased from 37.96% to 53.67%. In Himachal Pradesh, manufacturing shares grew from initially 3.43% to 30.93% in 2017. The data shows that the growth of shares between states varies not only in service sectors. It should be kept in mind here, that the Indian states are large entities. Some states have a higher population than Germany, France, and the Netherlands together. Testing the relation of modern service and total VA, reveals that though the growth of VA in modern services in positively correlated with the growth of total VA, the growth of the share of modern services is negatively correlated to total VA growth (Table A-4).

The results suggest that modern services are correlated to overall economic growth, but they are not the only driver. It seems like states VA has been driven by other sectors as well. This is in line with the previous observation that states, which have not increased their share of modern services, have developed in other sectors. Looking at the sectoral shares provides an interesting overview on the sectoral emphases across the country, but to determine whether modern services have been growing unequally, the growth rates of sectoral VA should be analysed. For almost all states, communication and finance were the fastest growing sectors between 1980 and 2017. The average annual growth rates of VA of modern services in 1980-2017 vary between 4.1% in Jammu & Kashmir and 10.3% in Puducherry. In all three subsectors of modern services, large annual growth rates can be observed. In communication, the largest annual growth rate can be found in Kerala with 19%. The smallest growth rate was found for Assam with 9.1%. In West Bengal, finance has grown by 8.6%, which is the smallest annual growth rate for this sector. With 14.8% the largest annual growth rate in finance was found in Sikkim. Looking at these extremes reveals that there are significant differences between growth rates, but overall, communication and finance have been growing substantially in all states. Smaller but still impressive growth rates can be observed for business services. With 2.9% annual growth in Jammu & Kashmir and 10% annual growth in Puducherry, the spectrum of growth is similar, though on a lower level.

Overall, it seems that though states have benefitted from the boom of services differently, modern services have been growing faster than the rest of the state economies in almost all states. While some states like Kerala and Sikkim have experienced outstanding growth rates, even the least service-oriented states have experienced large growth rates. It becomes clear that service sector growth has driven states apart in terms of economic performance, but it is fair to say that all states have benefitted from services. Again, it should be noted that Indian states are large and thus state level data cannot give any information on the effect of services on inequality rural and urban population. However, the previously made hypothesis that services have played a growth-enhancing role in economic hubs like Chennai or Bengaluru only, does not hold. Service sector growth has occurred in most states, even in those states that have lagged behind in economic development big time.

20 Having laid out that services have benefitted all states in India, I will now test whether there is a pattern between the economic performance before the third industrial revolution and the gains from modern service growth. I expect to find that states with initially higher VA per capita have gained more from the boom of modern services. As they had higher initial LP adapting to the needs of modern services might have been easier. Table 2 presents the results for three regressions testing my hypothesis. The first column shows that the initial level of VA per capita in 1983 is significant and positively correlated to the growth of modern services since the beginning of the third industrial revolution. Similarly, the initial level of LP of a state positively affects the gains from modern services (column 2). When looking at the initial level of LP in modern services, however, a different result can be found. A high LP in modern services is negatively correlated to the growth of modern services between 1983-2017. A graphical representation of the three relationships can be found in the appendix (Figure A-3 to A-5). The results show that indeed, well developed states have gained more from the rise of modern services. As Kochhar et al. (2006) suggest the advantage of further developed states is not due to higher productivity in modern services itself, but rather caused by productivity advantages of the overall economy.

Table 2: Effect of initial performance on growth of modern services (23 Indian states)

(1) (2) (3) Variable 𝑣𝑎1983−2017,𝑚𝑠 = 𝛼 + 𝛽1∗ 𝑉𝐴𝐶 + 𝜀 𝑣𝑎1983−2017,𝑚𝑠= 𝛼 + 𝛽1∗ 𝐿𝑃𝑡𝑜𝑡𝑎𝑙+ 𝜀 𝑣𝑎1983−2017,𝑚𝑠= 𝛼 + 𝛽1𝐿𝑃𝑚𝑠+ 𝜀 VA per capita in 1983 0.0416** (0.0191) Adj. R² 0.177 LP in total economy in 1983 0.0141* (0.00730) Adj. R² 0.146 LP in modern services in 1983 -0.000468** (0.000182) Adj. R² 0.230

Notes: Standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1, OLS regression. Constant coefficient estimates are available upon request. The index “𝑚𝑠” indicates “modern services”, 𝑉𝐴𝐶 indicates VA per capita. Source: Own elaboration.

5. Drivers of labour productivity growth

21 even “job-loss growth” (Kannan and Raveendran 2019) would mean that SC in employment plays no or only a small part in LP growth. To get a clearer picture of the origins of LP growth in India, I will conduct an SSD analysis.

Over the period 1983-2017, total VA has grown by around 5.5% per year, while total employment has grown annually by around 0.5% only. The growth of LP in India, hence, has been enormous. When observing the sectoral shares of employment, the most significant trend is the declining share of agriculture, while employment shares of services and manufacturing change only slowly (Figure A-6). The pattern remains similar when looking at the states separately. In 1983, LP of the Indian states has been ranging on a relatively low level. With 1.74 ₹ Lakhs per person, Goa had the highest LP, while LP in Bihar was only 0.25 ₹ Lakhs per person. Over time, all states have improved their LP, but the relative performances have remained equal. While in 1983, labour in Goa has been about 7 times more productive than Bihar, in 2018 it has been around 7.5 times more productive. When looking at the LP in modern services, the patterns are different. LP has not grown in all states and the relative performance of states has changed quite often. Interestingly, the mean deviation of LP in modern services has decreased between 1983 and 201711. Having set this out, I will now examine what has driven the LP growth in modern services and how these effects have varied across states. In this part of the analysis, I conduct an SSD, which aims to explain the different drivers of LP growth by decomposing LP growth into three drivers accounting for LP growth due to productivity growth within sectors and between sectors, with the between sectors productivity growth being equal to the SC effect. The decomposition of the change in LP between the period (T,0) with T being the final period and 0 being the initial period, is given as

𝐿𝑃𝑇− 𝐿𝑃0= ∑ (𝐿𝑃𝑇,𝑖− 𝐿𝑃0,𝑖)∗ 𝑆0,𝑖 𝐿𝑃0 𝑁 𝑖=1 + ∑(𝑆𝑇,𝑖− 𝑆0,𝑖)∗ 𝐿𝑃0,𝑖 𝐿𝑃0 𝑁 𝑖=1 + ∑(𝐿𝑃𝑇,𝑖− 𝐿𝑃0,𝑖)∗ (𝑆𝑇,𝑖− 𝑆0,𝑖) 𝐿𝑃0 𝑁 𝑖=1 . .

𝑆𝑖 is the share of the sector 𝑖 in overall employment. The first term of the equation is the

within-sector productivity effect, also referred to as the intra-within-sectoral effect (ISE). It captures the LP improvements due to technological change, capital accumulation or the improvement of resource allocation across plants within the sector. The second and third term of the equation are the between-effects or SC effects. As has been discussed previously the SC effect can be divided into two. The first part, and the second term of the equation, is the SSE, which determines how effectively labour has been shifted from low-productivity sector to high-productivity sectors. The second SC effect, and the third term in the equation, represents the DSE. It includes the joint effect of changes in sectoral employment and productivity levels. The dynamic SC is positive if labour is moved to sectors of positive LP growth.

Now, looking at the results of the SSD for all 32 Indian states and the full time series, the first row of Table 3 shows the equation for the total economy, which is equal to the equation above. The results on sectoral level are given in the rows below. In the first column, the left side of the equation, LP growth, is given. As mentioned, all sectors have positive LP growth over the observed four decades. The following three columns (2, 3 and 4) display the three components. The row sums of these three columns is equal to the corresponding number of the first column.

11 _{Mean deviation measured based on 23 Indian states. Results: 𝑀𝐷}

1983= 8.12, 𝑀𝐷2017= 5.39