ABSTRACT Title of Document: MANUFACTURING SECTOR PRODUCTIVITY IN INDIA: ALL INDIA TRENDS, REGIONAL PATTERNS, AND NETWORK EXTERNALITIES FROM INFRASTRUCTURE ON REGIONAL GROWTH Adil Mohommad Doctor of Philosophy 2010 Directed By: Professor Charles R. Hulten, Department of Economics In this dissertation I examine sources of growth in the formal manufacturing sector in India, from 1970 to 2003. I consider both all-India trends and state-level trends in the growth of resource efficiency, measured by TFP, and the relative contribution of TFP growth to output growth in manufacturing, as compared to capital accumulation. At the state level, I also examine the relationship between per-capita income and trends in output per worker and TFP in the manufacturing sector. Finally, in a spatial econometric framework, I test for the presence and magnitude of network spillovers from infrastructure, including national and state highways, and electricity generation capacity, on manufacturing TFP levels across states. My work contributes to an on-going debate on the response of manufacturing sector TFP to the implementation of economic reforms in India, in the 1980s and 1990s. At the regional level, this dissertation addresses not only the literature on the causes behind rising income inequality across states, but also on the role of infrastructure on regional growth, restricting attention to the manufacturing sector. The results of this dissertation show that at the all-India level and at the state level, manufacturing sector TFP growth accelerated in India during periods of economic reform. The contribution of TFP growth to output growth increased in the 1990s relative to earlier periods, and exceeded the contribution of capital accumulation. At the state level, I find evidence of convergence in growth rates of output per worker and TFP in manufacturing. I do not find evidence of a significant correlation between output per worker in manufacturing and state per-capita incomes. Given the relatively small share of the manufacturing sector in state GDP on average, these results imply that the source of rising income inequalities across states may not be manufacturing. Finally, I find some evidence to suggest that there exist positive network spillovers from physical infrastructure on manufacturing sector TFP. The results suggest that doubling the stock of national and state highways, and electricity generation capacity can lead to a nine percent increase in manufacturing sector output. MANUFACTURING SECTOR PRODUCTIVITY IN INDIA: ALL INDIA TRENDS, REGIONAL PATTERNS, AND NETWORK EXTERNALITIES FROM INFRASTRUCTURE ON REGIONAL GROWTH By Adil Mohommad Dissertation submitted to the Faculty of the Graduate School of the University of Maryland, College Park, in partial fulfillment of the requirements for the degree of Doctor of Philosophy 2010 Advisory Committee: Professor Charles Hulten, Chair Dr. Harry Kelejian Dr. Wallace Oates Dr. Robert Schwab Dr Sonalde Desai ? Copyright by Adil Mohommad 2010 ii Dedication I dedicate this dissertation to the memory of the late Janaki Mani Iyer, my grandmother, who raised and nurtured me, and to the late Ustad Ali Akbar Khan, who gave me spiritual succor during testing days of my dissertation. iii Table of Contents Dedication ..................................................................................................................... ii Table of Contents ......................................................................................................... iii Chapter 1: Introduction ................................................................................................. 1 Chapter 2: Literature Review ...................................................................................... 16 Chapter 3: Sources of Growth..................................................................................... 26 Chapter 4: Infrastructure Spillovers ............................................................................ 50 Chapter 5: Summary and Conclusions ........................................................................ 68 Appendices .................................................................................................................. 73 Bibliography ............................................................................................................. 113 1 Chapter 1: Introduction This dissertation examines the sources of growth in the formal1 or "registered" manufacturing sector in India, between 1970 and 2003. It is an event study that addresses the on-going debate about the impact of economic reforms instituted in the 1980s and 1990s on the efficiency of resource use (typically measured as Total Factor Productivity or TFP2) in the manufacturing sector. Economic reforms coincided with an acceleration of overall economic growth3, and particularly with output growth in the manufacturing sector, which increased from about 6 percent in the 1970s to more than 8 percent on average in the post-reform period. However, there is disagreement in the literature about the contribution of TFP growth to this acceleration in manufacturing sector output growth. The economic and policy background against which this debate is situated helps to place this dissertation in context and motivates its questions. While rapid economic growth characterizes the overall achievement of the Indian economy in the past few 1 The formal sector refers to firms to that are registered under the Factories Act of 1948 and are subject to reporting requirements. For this reason, this sector is often referred to as the "registered" manufacturing sector. For the purpose of this dissertation, I use the terms "manufacturing" and "formal/registered manufacturing" interchangeably. 2 TFP is identified with the Solow residual, and is typically measured as the excess of value added growth over the income-share weighted growth rate of primary inputs. 3 The growth rate of GDP increased from about 3 percent per year in the 1970s, to between 5 and 6 percent in the 1980s and 1990s, and now over 7 percent in the 2000s. 2 years, the issue that is central in the minds of policy makers is sustainable growth. The latest budget document for the fiscal year 2010-11 states early in its preamble that the ?first? economic goal is to sustain a growth rate of over 9 percent per year or higher, over the medium term4. Considering that the incidence of poverty is still very high in India5, sustained high growth rates would play a crucial role in bringing down poverty in India, and for overall development. The budget document also highlights the ?second? economic goal of inclusive growth. In various forms, the goal of inclusive growth has been a consistent feature of Indian development policy. This goal is often articulated in the form of balanced regional growth, a central tenet of development policy right from the first Five Year Plan for the years 1951-19566. In the achievement of these objectives, namely sustained high growth balanced across regions, Indian policy makers have emphasized the importance of the manufacturing sector at various points, and physical infrastructure is understood to be a critical input in this regard (as discussed briefly below). This forms the policy backdrop against which the questions and findings of this 4 http://indiabudget.nic.in/ub2010-11/bs/speecha.htm 5 Estimates range from a low of 28.3 percent for 2004-05 as estimated by the Planning Commission to a significantly higher 37.2 percent estimated recently by the Suresh Tendulkar committee. There is a lively recent debate on the various poverty estimates due to their implications especially for food security policy. http://www.livemint.com/2009/12/13213528/The-poverty-estimates-debate.html 6 The Plan states that ?[T]he need for attaining a rate of investment ... which could form the basis of more rapid advances in the following years and lay the foundation for balanced regional development in the next planning period has been an important consideration in determining the development programme in this Plan.? 3 dissertation may be interpreted. The need for sustainable growth raises the question of how it may be achieved, and for this purpose, sorting out sources of growth is important from a development policy point of view. If output and income (in growth and level terms) are driven by TFP, this suggests a policy response that encourages innovation and diffusion of new technologies, whereas factor accumulation led growth suggests policies aimed at raising the rate of savings and investment. Secondly, if capital is subject to diminishing returns and innovation-led growth is not, this has implications for the sustainability of growth in the long run. Efficiency growth may serve as the engine that drives sustained growth rates. However, this requires appropriate policies that may serve to accelerate the creation and diffusion of new ideas and innovations that can drive efficiency growth. In this regard, India experienced economic reforms in the 1980s and 1990s that paved the way for a transition from a state-led growth model to a market oriented model. In conjunction with the observed acceleration in economic growth, this has led to an examination of whether economic reforms were coincidental with acceleration in output and efficiency growth, in particular within the manufacturing sector. On this issue, the debate on TFP growth in Indian manufacturing has largely been inconclusive. I re-examine this issue with methodological improvements in price measurement, and make use of a longer time series on manufacturing sector aggregates, that takes into several years into the post-1991 reform period. As a result of these efforts, I find that TFP growth in manufacturing indeed accelerated in the reforms era, both in the 1980s and again in the 1990s, and its contribution to 4 manufacturing output growth also increased. These results on TFP growth acquire importance when viewed against the literature examining cross-country sources. A growing body of evidence suggests that TFP is an important determinant of levels of development across countries, which justifies our focus. Hall and Jones (1999) find that the correlation between levels of TFP and output per worker (closely related to per-capita income) was 0.9, and that differences in TFP explain the major share of the difference in output per worker between the richest and poorest countries7. Klenow and Rodriquez-Clare (1997) show that more than half the difference in per-capita income levels and growth rates can be attributed to differences in TFP. More recently, Hulten and Isaksson (2007) show that differences in relative levels of TFP are the dominant factor accounting for differences in income per capita. They also find that TFP growth accounted for more than half the growth in output per worker in the Newly Industrialized Economies (NIEs) and other emerging Asian economies including India. This evidence contrasts the empirical findings of Mankiw, Romer and Weil (1992), and Allwyn Young (1995) that suggested capital accumulation as the main sources of growth8. 7 Hall and Jones find that output per worker was more than 30 times higher in the five countries with the highest levels of output per worker than in the five countries with the lowest levels. Of this difference, a factor of more than eight was attributable to TFP, and about two for both capital intensity and capital per worker, which explains the rest of the difference. 8 In this context, it is interesting to note that Hseih (1999) did not find declining rates of return to capital in the East Asian "miracle" economies, which should have been in 5 This study focuses to the manufacturing sector on account of the importance that has consistently been attached to it in India's development policy, despite its relatively small share in GDP9. The stated intention of policy makers and chambers of commerce to raise the share of manufacturing in GDP to 25 percent in the next few years signals this emphasis10. Limiting this analysis to the registered sector (as opposed to the entire manufacturing sector, which includes the informal/unregistered sector) is motivated by considerations of data. As the registered sector is subject to reporting requirements, a consistent time series on inputs and output are available, whereas data for the informal sector can problematic. In terms of coverage, registered manufacturing accounted for about 84 percent of total inputs, 82 percent of gross output, and 76 percent of gross value added in the entire manufacturing sector in 1999-00 (Ray 2004)11. The registered sector is not a major source of employment; the informal sector employs the major share of the total industrial workforce12. evidence if growth in these countries was indeed driven by capital accumulation as suggested by Young (1995). 9 About 17 percent of GDP, compared with 27 percent in Korea, and 43 percent in China (nominal terms). 10 http://machinist.in/index.php?option=comcontent&task=view&id=1237&Itemid=2 11 Ray uses ASI data for 1999-00 and National Sample Survey (NSS-56th Round) data for 2000-01. Note that Bosworth, Collins and Virmani (2007) place the share of organized manufacturing in sector GDP at 60 percent, not 76 percent as in the above study. 12 According to Bosworth et. al. (2007), about 98 percent of manufacturing sector 6 The all-India level growth-decomposition exercise facilitates comparison with the extensive literature on the timing of growth in registered manufacturing. In addition to this, I extend the sources of growth framework to the regional (state) level13. This extension is motivated by the sustained emphasis on balanced regional growth in Indian development policy. The second Five Year Plan (1956-61) had, among others, the following main objectives ? (i) growth led by rapid industrialization especially in basic and heavy industries and (ii) reduction in income inequalities and even spread of economic power, which was articulated in the form of balanced regional growth14. However, despite several years of emphasis on these objectives (extending up to the Eleventh Plan that is currently in sway) income inequalities have increased across Indian states. Figure 1 shows that the Gini coefficient for per capita state domestic employment is in the unregistered sector. 13 The all-India aggregates for formal manufacturing sector used in this study are constructed from state-level series, as the sum of 17 major states, that account for 98 percent of the formal manufacturing sector. 14 In early years of development planning, balanced regional growth was pursued through both state-led investment in industries as well as regulation/restrictions imposed on private activity, and through fiscal inducements. However, post-reform strategies place emphasis on private sector investment, and balanced growth objectives are articulated through policies such as establishment of Special Economic Zones (equivalent to the Export Processing Zones in China), whereby the state provides infrastructure to attract private industries. See Reddy, Prasad and Kumar (2009). 7 product has increased over time, according to Planning Commission estimates15. Moreover, the level of penetration of the manufacturing sector in state GDP is on average only around 12 percent, with most states falling below the average (Table 1). Against this backdrop, this study inquires into the role of the manufacturing sector with regard to regional income inequality. It focuses on convergence across states in growth rates of output per worker as well as TFP, and examines whether these productivity measures have any relationship with per-capita state incomes. The results of this study show that there is indeed convergence in the growth rates of output per worker and TFP growth across states, especially in the 1980s and also in the 1990s. In the case of TFP growth, again convergence is observed strongly in the 1980s, whereas in the 1990s it appears that TFP growth rates across groups of states are very similar if not identical. However, there is little correlation between a state's rank in terms of income per capita and its rank in manufacturing output per worker, indicating that the manufacturing sector has had little impact on income levels in states. Perhaps this is driven by the relatively low penetration of manufacturing in state economies. While I do find a positive correlation between TFP levels and per-capita income, I do not take this to indicate the effect of differences in manufacturing sector TFP levels across states on relative state incomes. TFP may well be influenced by levels of development, through the impact of income on health, education, and other 15 Eleventh Plan (2006-11), chapter 7. 8 environmental factors that are not directly related to the inputs and outputs of the manufacturing sector. In other words, the causation may run from income to TFP rather than vice-versa. An important conclusion from the findings on regional growth in manufacturing is that if one is to examine the causes for widening income inequalities across states, then one may have to look at other sectors, namely agriculture and services, in order to explain the causes for divergent levels of income per capita. The results show that levels of output per worker do not correlate with state per-capita income levels, and growth rates of output per worker have converged. Further, given the low penetration of manufacturing in state GDP, it is unlikely that income inequalities have widened on account of the manufacturing sector. In addition to the focus on regional inequalities, I also examine in parallel the contribution of TFP growth to state manufacturing sector output growth, relative to the contribution of inputs. This addresses the issue of sustainable growth, for even if manufacturing is currently small as a proportion of the state economy, its role is envisaged to expand, as mentioned above. I find that while TFP growth had relatively little role to play in the convergence of output per worker growth, its contribution to output per worker has indeed increased over time, and is now larger than the contribution of capital intensity to growth. 9 In the final part of this dissertation, I examine the impact of infrastructure provision on state manufacturing TFP levels in an econometric framework. Infrastructure provision is an important policy problem especially in India. Not only is it recognized as a bottleneck to high growth rates, in the context of balanced regional growth, infrastructure has consistently found a place of high importance development policy. In this regard, it is noteworthy that the 1st Five Year Plan (1951-56) recognized that in the absence of adequate ancillary services and social overhead capital, it would be difficult to woo investment to relatively less industrialized locations16. More generally, infrastructure is considered to be a critical input for development of the manufacturing sector. In the words of Albert Hirschman, private production activities cannot be undertaken without social overhead capital (infrastructure), including physical infrastructure such as roads and electricity, and services such as water, communications, public administration, education, and health. He assigns transportation and energy infrastructure to the "hard core of the concept" of social overhead capital, as "[I]t is widely assumed that enlarged availabilities of electric power and transportation facilities are essential preconditions for economic development practically everywhere." 16 First Five Year Plan, Chapter 29, paragraph 49. http://planningcommission.nic.in/plans/planrel/fiveyr/default.html 10 Not only does infrastructure have a role as a direct input in private production, but it may contribute to output growth via spillover benefits, over and above its contribution to growth as an input. To illustrate what is meant by such spillovers, take the case of physical infrastructure like roads17. Roads are a direct input for the transportation industry, when combined with trucks and drivers, produce transport services, which firms pay for, in effect paying for the infrastructure indirectly as embedded in the transportation services consumed by firms. In addition, good road networks may reduce travel time for all firms, as well as wear and tear on transport equipment, and help promote better inventory management, which may increase resource use efficiency for firms. Similarly, regular and high quality electricity may promote introduction of new technologies that require reliable power supplies, again increasing efficiency. Moreover, physical infrastructure is often a spatial network, and a reduction in congestion in one state (due to an improvement in the stock of infrastructure within that state), may also reduce congestion in an adjoining state since road networks and traffic flow are linked spatially. Stated another way, a state may derive spillover benefits from the infrastructure stock of its neighbours. We may term these kind of benefits from infrastructure as ?network spillovers?, which could be ?own? spillovers when a state?s own stock of infrastructure increases (say) manufacturing output within that state, or ?spatial? spillovers when output growth is impacted by changes in infrastructure stock and/or quality in adjoining states. It is these types of spillovers 17 This example is due to Hulten, Bennathan and Srinivasan (2006). 11 that are captured by TFP (for the registered manufacturing sector) whose presence we seek to establish. Two points are worth emphasizing here. Firstly, the type of network effect whose presence we seek to establish are pure spillovers, over and above the contribution to growth (in registered manufacturing output) via infrastructure?s role as an intermediate input. This should be kept in mind when interpreting the magnitudes of elasticities (of output to infrastructure stock) that we find in this study. Secondly, the effects, such as we do find here, only pertain to spillovers accruing to registered manufacturing, they exclude any such benefits that may accrue from infrastructure to any other sector in the economy. If spillovers of this nature are significant, this has implications for infrastructure policy. Given the special characteristics of infrastructure goods, determining the existence and magnitude of infrastructure spillovers can be a useful exercise. Infrastructure goods are typically in the nature of public goods or club goods, provided outside the market mechanism through government spending. For goods provided through the market mechanism, it is believed that the private and social marginal benefits are equated, and the price reflects the entire benefit from the good. On the other hand, infrastructure goods are typically provided on a cost-benefit basis, and if significant network spillovers are not accounted for, the direct rate of return may understate the total benefit, leading to under-provision. Especially in the presence of spatial spillovers, this factor may have a bearing on how decisions for 12 regional infrastructure provision are made. The literature on infrastructure spillovers on growth is marked by considerable debate, with estimates ranging from very high, such as Aschauer's (1989) estimates for the US, to faint or non-existent. Also, the results are mixed in a cross-country setting; we refer the reader to Chapter 2 for a fuller discussion of this literature. Based on the empirical analysis presented here, there appears to be some evidence of spillovers from highways and electricity infrastructure in Indian states. In this study I use the growth accounting methodology incorporating infrastructure spillovers developed by Hulten and Schwab (1991), and applied to Indian manufacturing sector data in Hulten, Bennathan and Srinivasan (2006), hereafter referred to as HBS. The main contributions of this study are as follows. Firstly, I extend the time period of the HBS study to take into account several years of data in the important post-1991 economic reform period. The HBS study extends from 1973 to 1993, which leaves an opportunity to explore in more depth what happened to manufacturing sector growth several years into the second wave of economic reforms. Secondly, I expend considerable effort to address the issue of price deflation, as it is clear from the literature that different assumptions about underlying prices can lead to qualitatively very different results on sources of growth, thus making it difficult to draw conclusions about the pattern of TFP growth in Indian manufacturing. The reader is referred to an extensive discussion of the methodology employed to construct price deflators in this study, in Chapter 3 (sections 2 and 3), and the 13 Appendix. Finally, I extend the work done by HBS in studying network spillovers from infrastructure on manufacturing sector growth, by implementing spatial econometric techniques developed by Kelejian and Prucha (1998, 1999), that allow us to test not only for the presence of own spillovers and spatial spillovers. To the best of my knowledge, this is the first such application of spatial econometric techniques to data from Indian states. WHY INDIA India is an important country to study from a development perspective, given that despite rapid economic growth, it still houses a third of the world's poorest, and ranks 122nd in the world on per-capita income. As a case study, its regions are heterogeneous in terms of geography, ethnicity, and socio-economic structure, but at the same time share common institutional features and statistical systems, that render it relatively free of problems that make comparisons and interpretation difficult in a cross-country setting. Thus, India is sufficiently large and diverse, and at the same time homogenous to the extent that may allow the findings of such a study to have some general validity (HBS 2006). India also underwent a shift in its approach to development policy, over a period that falls within the time-frame of this study. In the early years of post-Independence development policy, the main vehicle for implementing development strategies was central planning. The emphasis lay on import substitution, and development of 14 domestic capacity in heavy industries, as documented in the Second Plan (1956-7 to 1960-1), also known as the Nehru-Mahalonobis plan after its main architects. The Plan also placed emphasis on balanced regional growth that was to be pursued both through incentives and through restrictions on industrial scale and location. This period was characterized by strong presence of the state in production activities, directly through state owned enterprises and indirectly through industrial licensing, quotas and permits, and import controls, all of which combined to restrict the scope of activity in the private sector. On account of the restrictive nature of controls, this period is often referred to as the License-Quota-Permit Raj, evoking British colonial rule in India. Subsequently, economic reforms were initiated in the 1980s, largely aimed at the manufacturing sector, involving partial de-licensing and deregulation of industries, along with a measure of import liberalization18. These were followed by another wave of reforms, after a balance of payments crisis in 1990-91, in which virtually all manufacturing industries were de-licensed, along with further relaxation of controls on imports of capital goods and inflows of foreign direct investment. Conventional wisdom suggests that freeing up the private sector would results in absorption of new technologies through imports, and encourage innovation spurred by fewer entry 18 Ahluwalia (1991) noted that: "[T]he most important changes have related to reducing the domestic barriers to entry and expansion to inject a measure of competition in domestic industry, simplifying the procedures, and providing easier access to better technology and intermediate material imports as well as more flexibility in the use of installed capacity with a view to enabling easier supply response to changing demand conditions." 15 restrictions, which would be captured in higher TFP. The results on the timing of TFP growth thus provide indirect evidence on the efficacy of market-oriented reforms in India. To summarize, this dissertation consists of 3 inter-related sections. The first section is a sources of growth analysis of the registered manufacturing sector at the all-India level. The second section carries this analysis to the state level, and also examines the relationship between output per worker and TFP in manufacturing and per capita state incomes. The final section looks at a policy issue, namely infrastructure provision, and tries to ascertain the presence and magnitude of network spillovers from infrastructure on state level manufacturing growth, in a spatial econometric framework. The plan of this study is as follows. In Chapter 2, I review the literature on the three central questions of this dissertation. In Chapter 3, I undertake the sources of growth decomposition, laying out the model, describing the data, and discussing the all-India and state level results. Chapter 4 is devoted to network spillovers from infrastructure. This section includes data description and preliminary explorations, a description of the econometric model, and a discussion and interpretation of the results. Chapter 5 concludes. 16 Chapter 2: Literature Review 2.1. Sources of growth in all-India manufacturing Table 2 summarizes various results on trends in TFP growth (TFPG) in registered manufacturing, especially whether the acceleration (if any) coincided with economic reforms. Evidently the literature is not conclusive. Ahluwalia (1991) was one of the first to document a turn-around in TFPG from negative to positive in the 1980s, which she attributed to liberalization policies. Similar evidence of a turn-around was found by Dholakia and Dholakia (1994). However, Balakrishnan and Pushpangadan (1994) ? hereafter referred to as B-P ? pointed out methodological problems in Ahluwalia's procedure, and using an alternative procedure of estimating real value added19, showed that TFP growth in manufacturing decelerated in the 1980s, a result which stood at odds with the liberalization efforts of the day. 19 The debate over deflation methodology is really a debate over pricing intermediate inputs when estimating real value added from gross output data. In "single deflation", both gross output and intermediate input prices are assumed to grow at the same rate. But if intermediate input prices grow faster than output prices, real value added growth is under-estimated relative to its true rate (and vice-versa if intermediate input prices grow slower than output prices. B-P (1994) showed that output and input prices growth often diverge in the data. They adopted an alternative procedure known as double deflation, whereby gross output and gross intermediate inputs are separately deflated by two different price indices. The latter has to be constructed from several input price indices, to reflect the basket of intermediate inputs used by the manufacturing sector. They thus used input-output data on commodity flows to intermediate input consumption in the manufacturing sector for 1973-74, and derived a set of weights with which to combine intermediate commodity's price series into an aggregate intermediate price index, which was then used to estimate real intermediate input growth. However, double deflation is also not free of bias (see Rao 1996). 17 Since both the Ahluwalia and B-P procedure for estimating TFPG is potentially biased, Rao (1996) used the Tornqvist-Divisia procedure for estimating the productivity residual based on the Solow-Jorgenson-Griliches growth model that at the sectoral level, takes the form of a "KLEMS" model, an acronym for capital- labour-energy-materials-services. We refer to this measure as "total productivity" or TP, to distinguish it from the TFP concept that is based on value added and is suitable to analyze the aggregate economy20. TP is derived as the income share-weighted difference in real gross output growth and the growth rate of real primary and intermediate inputs, which includes energy, materials, and services. Unlike the value- added based model, productivity changes impact not only primary inputs, but also intermediate inputs in the KLEMS framework. This feature of the model makes it particularly suitable in the context of developing countries, where shocks to intermediate inputs can impact manufacturing output. Hence, economies from intermediate inputs may be a significant source of productivity growth. Using this framework, Rao also finds that TP growth decelerated in the 1980s. 20 At the level of aggregate economy, the consumption of intermediate goods is equal to their output, and thus intermediate flows cancel out. From the national income accounting identity, GDP equals gross domestic income (GDI), which equals the income of labour and capital. Hence TFP, which measures the excess of real value added growth over the income-share weighted growth of primary inputs, is appropriate at the level of the aggregate economy. However, at the industry/sector level, it is not necessary that purchases of intermediate inputs equal to sales of intermediate outputs. Thus the appropriate underlying production is a gross output function that includes intermediate inputs. Moreover, the efficiency parameter in the gross-output framework is typically modeled as augmenting not only primary inputs but also intermediate inputs, which can also be a significant source of efficiency. For a detailed discussion on this issue, see Hulten (2009). 18 A feature of both B-P (1994) and Rao (1996) is that their intermediate input price deflator uses fixed base weights, derived from one single year of input-output information for the Indian economy21. This is equivalent to a fixed-base weighting scheme, and thus subject to the criticisms that apply to such indices. Specifically, fixed weights derived from only one year of information on the flow of goods may miss out on changes in the underlying structure of inputs and outputs over time. To get around this problem, using state-level manufacturing data, HBS (2006) constructed a price deflator that varies by time and state22, and find that TP growth remained more or less unchanged in the pre and post-reforms periods. Hence, the spectrum of results covers acceleration, stagnation, and deceleration in manufacturing TFP in the 1980s relative to the 1970s and earlier. A big boost to liberalization efforts came in 1991, in the wake of a balance of payments crisis, leading to a bigger push towards market oriented economic policies, including widespread de-licensing and deregulation of industry, and further liberalization of the current account. A fresh literature emerged re-examining the timing of growth 21 Input-output tables for the Indian economy are generally available every five years, starting from 1973-74. 22 They derive implicit input prices at the 2-digit industry group level by assuming that within each industry group, the ratio of real input to real output is constant, and given this constant and the price of output, the implicit price of intermediate inputs at each industry group level can be determined, and aggregated to the all-manufacturing level using the relative weights of intermediate consumption in the given industry groups. 19 question, and again, while new methodological issues came to light, the results as to trends in productivity growth in manufacturing were inconclusive. For example, Unel (2003) uses the value added framework and finds manufacturing TFP growth to have accelerated in the 1990s compared to the 1980s. On the other, Goldar (2007), and Banga and Goldar (2007), using the gross output framework, find the opposite trend. Notwithstanding the mixed results, the potential importance of the role of services as an intermediate input was brought to light as a methodological issue. In previous studies, services prices were not included in measures of intermediate input prices, but over time, services have gained in importance as intermediate inputs, and therefore pricing intermediate inputs accurately requires that the weight of services be factored in. Banga and Goldar (2007) show that if the contribution of services is not taken into account, this leads to an overstatement of productivity growth in the post- 1990s period, perhaps on account of the faster growth of services-use in the 1990s relative to the 1980s. The results discussed above are confined to the registered or formal sector within manufacturing, consisting of firms registered under the Factories Act of 1948, which are subject to reporting requirements. Bosworth, Collins and Virmani (2007) point out that although this sector accounted for 60 percent of manufacturing output in 1999-00, bulk of manufacturing sector employment lies within the unorganized or informal sector, accounting for as high as 98 percent of manufacturing employment. For the manufacturing sector as a whole (including both registered and unregistered manufacturing), they find that TFPG is distinctly higher over 1980-2004, compared to 20 the 20 years prior. They also find TFPG slowed down in the 1990s, though it picked up again after 199923. Taken as a whole, the literature on All-India manufacturing TFPG is not conclusive about trends in the post reform period. However, it identifies a variety of issues that need to be addressed while estimating TFP growth, as highlighted above. In this study, we synthesize the lessons from the literature in a number of ways. Firstly, we adopt the Tornqvist-Divisia (T-D) index numbers procedure for estimating TP growth based on a gross output framework that includes intermediate inputs, and allows the productivity parameter to enhance all inputs, including intermediate inputs. Our use of this framework helps to guard against the biases that may creep into TFPG estimates due to single or double deflation. Secondly, we adopt a flexible weighting strategy for pricing intermediate inputs, and derive the weights by using several years of commodity-flows information for the Indian industrial sector. Thirdly, we incorporate the services sector in a KLEMS model, in view of its increasingly important role as an input. The methodology is described in more detail in Chapter 3 as well as the Appendix. 23 Bosworth et. al. (2007) assume fixed shares for capital and labour in value added, according to proportions that are observed in OECD countries (60:40 share for labour and capital respectively). This is important, since the growth rates of capital and labour may differ significantly in different periods, the choice of factor shares can have a strong influence on measured TFPG. Their choice may be justified owing to their coverage of both the organized and unorganized sector, as reliable data are difficult to find for the latter. In our case, we do not make this fixed shares assumption, since we deal only with registered manufacturing, which presumably has better quality data. 21 2.2. Regional growth and infrastructure Given the variation in regional development levels, development plans have consistently emphasized balanced regional growth24, and industrialization was given an important role in this objective. Prior to liberalization, industrial location policies were pursued with a combination of fiscal incentives to industrialize so-called "backward" areas, and industrial licensing and quota restrictions that prevented already industrialized regions from entering markets and expanding scale freely. Over time, it was recognized that this approach might be detrimental to overall growth, as articulated in the Sixth Plan (1980-85): ?[I]t should be generally accepted that the fulfillment of the objective [of balanced regional growth] required upgrading the development process in backward regions rather than curtailing the growth of these regions that have acquired a certain momentum.? Current planning strategies have evolved considerably, now emphasizing inter-state competition for investment, and improvement in governance to create an enabling climate for industrialization25. Despite the focus on balanced growth, and notwithstanding the shift in strategy, regional income inequalities have widened in the post-reform era, as acknowledged in the current (Eleventh) Plan. This has raised concerns that the shift in approach to 24 This objective finds mention in virtually all Five Year Plans. See Balanced Regional Development in India - Issues and Policies, Anita Kumar (ed.), 2006, for excerpts on balanced regional growth policies and problems, covering the First Plan to the Tenth Plan (spanning 1950-2007). 25 See Chapter 7 of the latest (Eleventh) Plan document. 22 regional development might contribute to widening income inequalities. A recent study by Misra (2007) focuses on inequality in the post-reform period, and finds that inequality in per-capita state domestic product (SDP) worsened over the 1980s and 1990s (Gini-coefficient rising from 0.14 in 1981 to 0.18 in 1995), and after 1995 this gap has not narrowed. This lends support to the rising inequality documented by the Planning Commission study as seen in Figure 1. In this context, infrastructure is often identified as both a cause and a solution for the problem for regional income inequalities. Testing for sigma convergence26 in income in Indian states, Ghosh and De (1998) found an increase in the coefficient of variation in state per-capita income, and a high correlation between an index of state infrastructure27 and per-capita income. They also find that the position of states relative to national average per-capita income, and the relative position of states in terms of infrastructure provision has remained unchanged over 1971-1994, implicitly attributing to lack of infrastructure provision, the failure of state incomes to converge. In another study, using state level manufacturing sector data from 1976-1992, Mitra, Varoudakis and Veganzones (1998) finding that infrastructure28 has a significant 26 Sigma convergence is said to occur when the dispersion (in income) falls over time. 27 Constructed by principal components. 28 Their measure is a composite indicator constructed using principle components, including core infrastructure, services infrastructure, and human capital. They include measures of electricity, roads, railways, vehicles per capita, postal system, primary and secondary education enrolment, infant mortality, bank branches per 1000 23 positive impact on TFP growth in 14 out of 18 industry groups. HBS (2006) find evidence of a positive elasticity of manufacturing sector growth to highways and electricity generation capacity. The interest in spillover effects from infrastructure was sparked by the finding by Aschauer (1989) that a one percent increase in the stock of public capital in the US led to a 0.4 percent increase in private output. Based on the stock of public capital and value of private output, this implied that an investment of US $10 billion in infrastructure would produce an additional US $7 billion in GNP the following year, and that public capital is 4 times more productive than private capita, at the margin. However, Munnell (1990b) estimated the direct contribution of infrastructure stock to private output using pooled state data, and found a significant, though much smaller elasticity of 0.15 percent. This difference in the aggregate and panel estimates was attributed to the presence of infrastructure externalities that may not be captured in state level data. However, Hulten and Schwab (1984, 1991) found no evidence that regional differences in productivity were driven by differences in stocks of public goods; in fact they found that productivity levels and growth rates were quite similar across reasons, implicitly ruling out any explanatory role for infrastructure. Holtz-Eakin and Schwartz (1995) also found no evidence of spatial spillovers from US national highways on state level population, and deposits and loans as percent age of income. 24 private output growth. There is some evidence of spatial spillovers in other countries. Pereira and Roca- Sagales (2002) estimate a 5.5 percent rate of return to public capital at the national level in Spain, and a significantly positive return in 14 out of 17 Spanish regions. Similar to Munnel's findings about the US, they also find that the sum of regional estimates (based on within-region stocks on infrastructure) account for less than half of the total effect of public capital, as estimated at the national level. Upon allowing for spatial spillovers, this sum exceeds the national estimate. In cross-country data29, Canning and Fay (1993) estimate normal to high returns from public capital in industrial countries, high returns in industrializing countries, and low returns in under-developed countries, suggesting that countries with relatively low infrastructure stocks get higher returns from additional stocks, as opposed to developed countries that already have dense infrastructure networks30. The evidence on developing countries suggests there may exist sizable infrastructure benefits within Indian states, since India as a whole is an infrastructure- shortage country. However, spillovers effects such as spatial spillovers may be 29 See Gramlich (1994) for a review of the literature. 30 This point was earlier made by Hulten, (commenting on Munnel 1990b), that "adding to an existing [infrastructure] network will rarely have the same return [as constructing the original network]: at the some point, the increasing returns to scale aspects of infrastructure are exhausted, and...marginal additions bring increasingly smaller benefits." 25 harder to detect. Indian states can be quite large in terms of area and certainly in terms of population, comparable to the largest European countries. For example, the population of one of the most industrialized states (Maharashtra) is over 96 million, and its land area is over 310,000 sq km. It compares with Germany, which has a population of 82 million, and a land area of more than 350,000 sq km. To put the problem in perspective, searching for spatial externalities across Indian states is equivalent to searching for them across large European countries, and it is possible that over such vast areas, these effects tend to diminish and may be difficult to detect. This may limit the strength of our results. 26 Chapter 3: Sources of Growth 3.1. Theory We adopt a non-parametric index-number approach to measuring manufacturing sector efficiency31. We implement the model developed for the US by Hulten and Schwab (1991), and applied to Indian manufacturing sector data by HBS (2006). The model relates gross manufacturing output to primary and intermediate inputs, and a Hicks neutral shift parameter that captures efficiency. This is the "KLEMS" framework, which is the form taken by the Solow-Jorgensen-Griliches framework for estimating resource efficiency as one moves from the aggregate economy to the sectoral level. As mentioned earlier, we refer to the measure of resource efficiency derived from this model as Total Productivity or TP. The shift parameter is modeled as a function of infrastructure stock32. The production function can be written as: 31 As opposed to assuming a specific functional form for the manufacturing sector production (gross output) function and estimating its parameters, we disaggregate the sources of growth based on the contribution of each input to real output growth. 32 Hulten and Schwab (1991) discuss the issues and implications of this formulation for the role of infrastructure. In this form, infrastructure is an "environmental" factor that can enhance the efficiency of any or all inputs, and captures its indirect or spillover impact on manufacturing output growth. Infrastructure can also directly contribute to growth as a paid input, and should thus be an argument in the production function. Suppose ),,,,()( ,,,,, titititiiti BMLKFtAQ = and this production function exhibits constant returns to scale over the private inputs. Then, under the assumption of competitive markets where each private factor is paid its marginal product, the growth rate of the residual can be expressed as ,tMtLtKtt MLKQA ?? ? ?? ???= pipipi 27 (1) ),,,(),( ,,,,, tititititi MLKFtBAQ = where tiQ , represents real gross output in state i in year t , tiK , and tiL , represent capital and labour stocks, and tiM , is real intermediate inputs, including materials, fuels and power, and services. )(?A captures exogenous changes in efficiency that shift the production function, and tiB , is the stock of infrastructure in state i at time .t The term also contains an independent time variable t to capture autonomous technical progress. We can give this function a specific functional form as follows: where )()( )()()( )()( )()( tQtP tXtPitQ tXtFiX i Q i i X i i iX tAtA ? ??=?=pi . But if the production function does not show constant returns to scale over private inputs (e.g., it shows constant returns over private and public inputs), then the omission public capital leads to two sources of bias. The first source is the contribution of public growth to output growth, analogous to the contribution to growth of private inputs. This is equal to .)( )( )()(, tQ tBtFiBti i iBtA ?=? The second source of bias comes from the fact that the price of private capital is not observed, but imputed by assuming constant returns to private inputs, and ignoring public capital. Thus, .)()()()()()()()( &tMtPtLtPtQtPtKtP iMiiLiiQiiKi ???= However, the residual attributed to private capital's share in income also includes the share of public capital, as an unpaid input )(tBi . Thus the share of private capital is over-estimated, introducing a bias. Note that if the production function does exhibit constant returns to private inputs, then this source of bias vanishes. In the case of Indian manufacturing, Fikkert and Hassan (1998) have shown that by and large Indian industry shows constant returns to scale, mitigating our concerns of bias from this source. There remains the possibility of infrastructure being a direct input into manufacturing. Like Hulten and Schwab (1991), I assume that the manufacturing sector buys intermediate inputs (services) from service-producing sectors such as transportation and communications, which in turn consume infrastructure as a direct input. Hence infrastructure is an indirect input for manufacturing industries. 28 (2) ,)( ,,,, ??? titititi MKLAQ ?= where .1=++ ??? Hence we assume that the production function exhibits constant returns in the private inputs. Taking logs, differentiating with respect to time, and rearranging, we obtain the Divisia index for TP growth: (3) ,MMKKLLQQAATP ??? ?? ???== ? ??? Under the assumption that factors are paid their marginal products, the elasticity parameters ??? ,, can be estimated as the share of input costs in gross output, i.e. ,pQwL=? pQMpM=? where Mp is the price of intermediate inputs, and ??? ??=1 is the share of private capital income33. This is a continuous time index of TP growth, but we use its discrete time approximation known as the Tornqvist-Divisia (TD) index: (4) ,lnlnln ,, ,,,, jijiMLKjtiti XQTP ????=? = pi where iQ is real gross output, jiX , is the level of (real) input j (labour, capital, and 33 Our estimate of capital stock is a Divisia aggregate of sub-components, based on ASI data on nominal investment in structure, and plant, machinery and equipment. Details are provided in the Data section of this chapter. 29 intermediate inputs), and ji,pi is the share of the jth input in gross output, calculated as the average of the share in two adjacent periods34. Let tp and t? be the price of output and intermediate inputs respectively in periodt . Divisia indices of real output and intermediate input growth were obtained as: )],ln()[ln()]ln()[ln(ln)5( 11,1,, ??? ???=? tttittitti ppQpQpQ and )]ln()[ln()]ln()[ln(ln)6( 11,1,, ??? ???=? tttittitti MMM ???? These growth rates are substituted back in (4) to obtain TP growth rates. An index is then constructed by incrementing (decrementing) using these growth rates and a base year. For the regional analysis, we also estimate TP levels using the procedure developed by Jorgenson and Nishimizu (1978) and generalized by Caves- Christensen-Diewert (1982), whereby the estimated growth rates of TP are applied to a base year estimate of the TP level of each state, which is relative to the all-India 34 For example, the share of labour in year t is calculated as: ][5.0 1,,, ?+= tLtLtL pipipi where pQwLL /=pi 30 average level. An estimate of the base year level is obtained as: ????? ???= MMMLLLKKKQQTPTP iiiii lnlnlnlnln)7( pipipi where starred variables are Divisia indices of all-India output and inputs, and jpi is the arithmetic mean of the share of expenditure on input j in state i , and the corresponding all- India share35. Finally, we normalize the state-wise estimates by 0TP , the average level in 1970, our base year. This base year productivity relative is then incremented (decremented) by the TP growth rates estimated above. 3.2. Data Data on manufacturing is taken from the Annual Survey of Industries (ASI) of the Central Statistical Organization (CSO), which covers all establishments (factories or units) registered under section 2m(i) and 2m(ii) of the Factories Act (1948). This includes factories employing 10 or more workers and using electricity, or 20 or more workers without electricity, on any day of the preceding 12 months. The population of eligible factories is split into a census sector and a sample sector for which the 35 This method was employed by HBS (2006). Banga and Goldar (2007) estimated TP levels normalizing by geometric averages of the Divisia indices of state output and inputs. We also experimented with this method and found very similar results to those obtained by the former method. 31 sampling probability has been revised at different points in time36. Therefore, while certain medium and large factories figure consistently in different survey years, those in the sample sector change over time. There have been three revisions to the industrial classification system since the first sample year (1970), which raises questions of data comparability across multiple survey rounds. This particular data set was compiled by the EPW Research Foundation (EPWRF), who attempting a concordance between the various series (NIC 1973, NIC 1987, NIC 1998), so that series obtained using NIC 1973-74 are as closely comparable as possible with data from subsequent rounds, at least at the 2- digit level. The data for the all-India manufacturing sector were constructed by aggregating 2- digit industries data at the state level, from 17 states37, excluding the north-eastern 36 Until 1986-87, the census sector was defined as 50 or more workers with power, and 100 or more workers without power, and the sample sector as 10-49 workers with power, 20-99 workers without power. Firms in the former group were enumerated every year, while those in the latter were enumerated every other year on the basis of a 50 percent probability sample. This methodology was replaced in 1987 by a new sampling method whereby any establishment employing in excess of 100 workers would be part of the census sector regardless of the use of power, whereas the rest would be part of the sample sector. There have been subsequent changes in the coverage of the census sector, but the definition of the sector as that employing more than 100 workers remains as of date. 37 These include Andhra Pradesh, Assam, Bihar, Gujarat, Haryana, Himachal Pradesh, Jammu and Kashmir, Karnataka, Kerala, Madhya Pradesh, Maharashtra, Orissa, Punjab, Rajasthan, Tamil Nadu, Uttar Pradesh, and West Bengal. In the subsequent econometric section, we leave out two mountain states of Himachal 32 states and other union territories that constitute a very small fraction of the manufacturing sector and do not have consistent time series data. In addition to ASI data on industrial aggregates, National Accounts Statistics (NAS) data on gross fixed capital formation was employed for constructing a capital stock series, and data from the input-output Transactions Tables (IOTT) for the Indian economy were used to construct prices deflators. How these additional data were used is described below. Construction of Price and Quantity Series Price Indices: We made extensive use of input-output information for the Indian economy to construct price deflators for ASI manufacturing output and intermediate inputs respectively. This approach builds on existing approaches for estimating real values of these quantities in the following ways. Firstly, with respect to output, the price deflator is a Divisia-type price index that weights the wholesale price index of each 2-digit ASI commodity by its share in total deliveries to final demand of all ASI-covered commodities. This is a refinement over using the official (CSO) wholesale price index for the aggregate manufacturing sector, since it excludes certain commodities (such as primary and processed primary goods) that are not covered by the ASI but are included in the official aggregate manufacturing price index. Secondly, with respect to intermediate inputs, we follow Pradesh and Jammu and Kashmir on account of issues of reliable infrastructure data and small share of manufacturing in state GDP in the case of Jammu and Kashmir. 33 Goldar (2004) and include services into the intermediate inputs category, as the share of services in inputs has growth considerably over time as shown below. Finally, our approach differs from the standard approach by using time-varying weights, making use of several different years of input-output information. Between 1973-74 and 1998-99, we consider 5 different rounds of input-output (I-O) commodity flow information as brought out by the Central Statistical Organization. The details of how I-O information was used to devise a set of commodity-wise weights to construct the aggregate price deflators are contained in the Appendix, while an overview of the method is provided below. In the existing literature, when I-O information has been used, the typical approach has been to use a single year of commodity flows data from the absorption matrix to derive weights, which overlooks changes in the input and output mix of the industrial sector especially when considering long spans of more than thirty years. The standard practice for deflating output or value added has been to use the overall manufacturing wholesale price index. We construct a Divisia-type output deflator as follows. Using I-O data over five survey rounds (1973-74, 1983-84, 1988-89, 1993-94, and 1998-99) we derive a series of time-variant weights for each commodity covered by the ASI, based on the share of a commodity's deliveries to final demand in total deliveries of all ASI commodities. Therefore, we obtain the proportion of output in the 25 2-digit ASI industry-groups delivered to final demand (consumption, investment, and exports). The publication of import-flow matrices in the 1989 and 1993 rounds makes it possible to identify the exact proportion of final demand met only by domestic output. 34 For other survey years, this information is not available, hence for years before 1989 (and after 1993), an approximation was employed, by imposing the proportion of deliveries to final demand from domestic output for 1989 on domestic production data from 1983 and 1973, and the proportion obtained from the 1993 data to the 1998 data. Using this method, we are able to obtain final-demand based weights for the ASI output price deflator. For years between I-O rounds, we apply linear interpolation. Table A.1 in the Appendix lists the two-digit industries covered in this paper, their weight in total deliveries to final demand, and the WPI series closest to the industry group. Figure 2 graphs the percentage difference between the official manufacturing and our Divisia output price deflator. From the input-flow matrix of the I-O tables, the values of flows of 21 input commodities/services to ASI industries were aggregated. Each input was then assigned a weight based on its share in total flows of all inputs to intermediate consumption by ASI industries. Weights for years between survey rounds were linearly interpolated (Appendix, table A.2 shows the estimated weights for each input). The use of I-O information for constructing an input deflator is common (see for example B-P 1994 or Rao 1996), but these studies use a single year of I-O information which amounts to fixed base weights, that might overlook changes in the structure of inputs. Flexible weighting guards against this. Table A.2 shows for example, that primary food articles and cotton textiles have declining shares in inputs, while mineral oils, chemicals, electricity, and services (especially transport, banking, and communication) have gained in importance. In Table A.3, we can see that the 35 share of services in intermediate inputs has doubled over time. Capital Stock: We construct an estimate of capital stock using the perpetual inventory method (PIM). The series is estimated as: ,)1( 0 1 ?? ? ?? ? +??= ? = ? ts st js t t IKK ? where K is gross capital stock, ? is the rate of depreciation, and I is gross investment. In our approach, we distinguish between structures (STR), and plant, machinery and equipment (PME), and obtain the final estimate of capital stock by summing the two series. An estimate of investment in each kind of capital stock was obtained as follows. ASI data provides only nominal investment figures, which have to broken down into real investment in either kind of capital. Using All-India National Accounts Statistics (NAS) data on aggregate investment (1974-1997) in PME and STR, we obtained the proportion of investment in PME and STR respectively by the public, private corporate and household sector. Since ASI data on investment is available by ownership, we apply the proportions by type of investment and type of ownership obtained from NAS data to the ASI series on investment by ownership, and add up the weights for each type of investment to obtain a final set of annual weights for splitting ASI nominal investment into investment in PME and STR (see Table A.4 in the Appendix). Based on NAS data, on average about 60 percent of nominal investment falls on PME and the remainder on STR for the period between 1974 and 36 1997. These averages were used as weights for years prior to 1974 and after 199738. Using these nominal series, Divisia growth rates of real investment in PME and STR were obtained separately, using the Electrical and Industrial Machinery WPI and Construction GDP deflator respectively, and applied to base-year level of stock to obtain our estimates of real investment. The base year stock was set at the deflated book value of gross PME and STR stock in 1970. Applying a depreciation rate of 3 percent for structures, and 15 percent for equipment, the aggregate capital stock series was obtained as ,)1( 0 1 , ?? ? ?? ? +???= ? = ? = jtjs st js t strpmejt IKK ? where j indexes the two types of capital39. Our method for constructing a series on capital stock is one among several alternatives that have been used for constructing this variable. However, while most studies employ the perpetual inventory method, they do not differentiate between 38 Rao (EPW, Jun 1994) provides data on book value of fixed assets by type, which shows 64 percent of ASI capital stocks in 1960 as PME and the balance as STR (in current prices). 39 Alternative approaches were also tried, such as a depreciation rate of 7.5 percent for PME, based on the idea that repair services in the developing economy may be cheaper, increasing the life of capital. Also, instead of deflating the split investment series, another option was to deflate the aggregate series by a Divisia deflator with flexible annual weights for PME and STR prices, and then split the series into two kinds of capital, apply the PIM, and re-aggregate for the final stock series. These approaches yielded very comparable stocks, so they were not eventually used. Note that the implied depreciation rate for the stock as a whole is just over 10 percent, twice the assumed value in some of the literature. 37 types of capital. Differences in the assumed rate of depreciation, the initial level of stock, and the appropriate investment deflator can yield different estimates. The typical assumption is a five percent depreciation rate, and a deflator which could be either the machinery and equipment price index, or the implicit price derived from current and constant price investment data for the aggregate economy (see for example Virmani 2004 or Goldar 2004). Our improvement consists of making a basic distinction between plant, machinery and equipment (PME) stock, and structures (STR), both for the assumed rate of depreciation and the price of investment. We are unable to use finer classifications of capital, or rental rates to accurately measure prices, due to the lack of data. Appendix Table A.5 shows the indices of real output and inputs obtained by these methods. We can observe the relatively rapid accumulation of capital stocks in mid-to-late 1990s, and the subsequent easing in investment. Several issues need to be highlighted in the measurement of labour input. The first is how to measure labour input. The definition of labour we use is referred to as `total persons engaged' in the ASI data, which apart from workers, managerial and other staff also includes unpaid members of the proprietor?s family who work in the factory. Thus, we include all employees that are directly or indirectly connected with the production process. We assume that the flow of labour services is proportional to the number of total persons engaged. There are other approaches in the literature that attempt to adjust the labour input for enhancements in labour quality over time, say via education. For example, Bosworth et. al. (2007) adjust for improvements in 38 labour quality by taking the average years of education in each of the three main sectors (agriculture, industry and services), and assume a seven percent return to each year of schooling so that ,LeL as=? where a is the assumed rate of return and s is the years of schooling. They are able to do so since they use quinquennial survey data for the entire economy, which contains details about educational attainments of the labour force. Lacking similar data on workforce characteristics within the ASI sector, as well as data on hours worked, we prefer to use the simpler formulation in which labour input is assumed to be proportional to the number of workers. Also, since we analyze state level trends, labour quality changes might differ by state, and to accurately adjust for these differences one would require state-by-state characteristics of the ASI workforce, which we do not have. Moreover, to the extent that improvements in labour quality are not paid-for, or occur on-the-job, they constitute enhancements in productivity and will be picked up by the productivity measure. A second important issue with labour in growth accounting is its share in total output. In our exercise, the contribution of labour to output is weighted by the share of payments to labour in total output. Labour income is measured by `total emoluments? that include wages and salaries, as well as the cash value of benefits. Under the competitive markets assumption, an input's share in total output measures its marginal product. This has important implications for growth accounting since incorrectly weighting a particular input can distort the results on productivity growth. Particularly for labour in developing countries such as India, the issue of measuring 39 the share of labour in output becomes problematic since a large number of workers are self-employed, whose income reflects partly a payment to their own labour and partly to their ownership of capital. Since the mixed-income sector in India is quite large (accounting for 45 percent of net domestic product (NDP) in 2002-03, and 79 percent of the income of the unorganized sector (CSO 2005), see Bosworth et. al. (2007)), assigning a share in income to wages equal to that observed in the formal sector may be misleading. In studies of the aggregate Indian economy, one expedient has been to simply assume a fixed income share for labour and capital, using observed shares in OECD countries, since these countries have relatively unchanged shares over time. However given, the structural differences between OECD and emerging economies, it is not clear whether this is a good assumption to make. In this study, we do not assume a fixed labour share in income. This is for several reasons. Firstly, this study deals only with the formal manufacturing sector, in which mixed income of the self-employed is not likely to be a big component, since the coverage of firms rules out very small firms in which mixed income is likely to have a significant share in output. Secondly, due to reporting requirements that ASI firms are subject to, the concern that the data are systematically incorrect are mitigated. What we observe in the data time is that the growth rate of labour has declined over time, as has its share in total output. In Table 3 we see that the share of labour costs in total value of output has fallen from about 11 percent on average in the 1970s, to about 6 percent in the 1990s, and the average annual growth rate has fallen from more than 4 percent to about half a percent in the 1980s and 1990s. The declining share of 40 labour income is ASI value added was also documented by Goldar (2004), falling from 42 percent in 1981 to about 28 percent in 1997. Goldar attributes the observed decline in the income share of labour to a labour saving bias in technological change, based on estimated elasticities from a translog production function. However, if labour is a relatively cheap and plentiful resource, it is not clear why the bias of technological change should be labour saving. There may be other factors at play that might lead to a falling share of labour in output. For example, if Indian industry was over-manned prior to economic reforms, then with liberalization of the manufacturing sector, an adjustment towards a more appropriate factor mix may have taken place, to correct for that over-manning, which would appear as a fall in the growth rate of labour. Moreover, with modernization of industry, diversification of the output mix, and improvement in quality of goods, the technology may call for greater capital intensity per se, compared to prior periods. Finally, the impact of reforms on the manufacturing sector employment also appears to depend on industry and state-specific factors. A recent study by Gupta, Hasan, and Kumar (2009) finds that labour intensive industries did not benefit much from reforms, and that states with relatively inflexible labour regulations have relatively lower growth rates in labour intensive industries, including employment growth. 41 3.3: Results 3.3.1. All-India manufacturing Figure 3 gives a snapshot of the growth rates of output and inputs. Average real output growth increased from about 6.5 percent over 1973-79 to 8.5 percent between 1990-99, accompanied by an acceleration in capital accumulation from six percent to 8-10 percent, over the same time frame. On the other hand, intermediate inputs growth declined from nine percent to 6.5-7 percent, while the growth rate of labour declined from almost five percent in the 1970s to between 0.5-1 percent in the 1980s and 1990s. Several factors acting in conjunction may explain the slow-down in labour growth. One possible cause may be a shift in the product mix, from labour-intensive to more capital intensive goods, which is supported by the observed increase in the rate of capital accumulation. Another explanation may be that Indian industries were relatively over-manned in the pre-reform era, especially in public sector enterprises, whereas the greater play of market forces led to an adjustment towards a more efficient level of labour in existing industries. A third hypothesis is that in the post- liberalization phase, Indian industry adopted relatively labour-saving technologies (Goldar 2007). 42 A number of works often report partial or single productivity measures, typically output per worker. In Table 3, we find that output per worker growth increased from an average of 2.1 percent in the 1970s, to an average of seven percent over the following two decades, even as capital intensity (ratio of real capital stock to workers) and capital-output ratio increased. We find that relative to the 1970s, capital intensity more than doubled in the 1980s, and was more than five times as high in the 1990s. The capital-output ratio almost doubled in the reform era. These trends might reflect the greater ease with which capital could be imported with the gradual elimination of import restrictions in the first wave of reforms. Turning to the sources of growth decomposition (Table 4), we clearly see that in the reform period, TP growth accelerated. It was negative in the 1970s, and turned around quite sharply to a positive average growth rate of 1.2 percent per year in the 1980s, and an even higher 1.8 percent in the 1990s. Column 4 and 5 compare the 1970s and 1980s. From these figures, it is easy to verify that capital accumulation contributed about 1.5 percentage points to output growth, which is not much higher than the contribution of TP growth. Both capital accumulation and efficiency growth accounted for about 30 percent of output growth, while bulk of the rest was due to intermediate inputs growth40. 40 The TPG numbers estimated here might look ?small? when compared to the growth rate of output in manufacturing, and when seen against the growth rate of aggregate output in the economy as a whole. However it must be kept in mind that these are growth rates of the residual of manufacturing output growth over the entire base of inputs, not just the growth rate of value added over the share weighted growth rate of labour and capital. In fact, one can obtain the corresponding TFPG figures from TPG (under the assumption of separability of the gross output function in terms 43 The negative growth rate of TP in the 1970s resonates with the widely held impression that Indian development policies were misguided, by erecting barriers to trade, propping inefficient industries under state guidance, and stifling private economic activity with cumbersome rules and regulations that raised myriad bureaucratic hurdles to get anything done (Rodrik and Subramanian 2004). The 1970s also witnessed two major oil price shocks that may have contributed to the observed decline in TP growth. Given restrictions on imports of technology and low levels of research and development domestically, firms may have found it difficult to substitute other inputs for more expensive energy, thereby sustaining negative productivity growth. While the sharp turnaround in TP growth in the 1980s has been attributed to initiation of economic reforms41, the 1990s witnessed deeper liberalization and market-oriented economic policies, along with almost complete de-licensing of the of value added and intermediate inputs), as TFPG = TPG/(1-b), where b = share of intermediate inputs in gross output. Based on the data in Table 3, (1-b) = 0.2, implying TFPG as high as six percent in the 1980s and nine percent in the 1990s. 41 Ahluwalia (1991) summarizes these changes as: "[T]he most important changes have related to reducing the domestic barriers to entry and expansion to inject a measure of competition in domestic industry, simplifying the procedures, and providing easier access to better technology and intermediate material imports as well as more flexibility in the use of installed capacity with a view to enabling easier supply response to changing demand conditions." 44 manufacturing sector. In this period42, TP growth increased to 1.8 percent per year, while capital accumulation slowed down, its contribution to average annual growth falling to roughly 1.25 percentage points. Hence, efficiency gains offset the decline in input growth in the 1990s to hold output growth to the same level achieved in the 1980s43. The overall contribution of TP to output growth increased from about 14 percent in the 1980s to 21 percent in the 1990s. As an additional check on our results, slightly different end-points were used to calculating the decadal average growth rates of TP. We find that the conclusion of accelerated TP growth in the 1980s and 1990s does not change (Table 5). 42 The year 1991 was left out from the calculation of average annual trends for two reasons, Firstly, it is the year in which reforms were introduced, and secondly, that year India suffered a balance of payments crisis, and may bias our conclusions about trends for the decade as a whole. Note that the averages in Table 4 included 1991, to show that its exclusion does not change the qualitative conclusions regarding trends in TP growth. 43 The results presented here are based on state-level manufacturing data, aggregated up from 1973 to 2003. These figures exclude the electricity, gas and water supply (EGW) sector, for which the ASI series were discontinued after 1997. We repeated this exercise using data including the EGW sector that allows us to cover a slightly different period from 1970-1997, based on published ASI data. In addition, data for the EGW sector were extrapolated forward to bring the coverage up to 2003. The results from this exercise are Table 4; qualitatively we find that TP growth accelerated in the 1980s and 1990s, and both capital accumulation and intermediate inputs consumption slowed down in the 1990s. We also find that output growth including the EGW sector was somewhat lower in the 1990s than in the 1980s, and labour growth was slightly negative. However, since the electricity sector in India has been in a process of major transformations since the late 1990s, the way these variables were measured before and after this process was initiated may not be strictly comparable. Hence, the results of this exercise may be seen as illustrative, and should be interpreted with caution. 45 Our results suggest that economic reforms can lead to gains in resource efficiency, especially since the reforms were in large measure aimed at freeing up the manufacturing sector from the various burdens of the license-quota regime. These results also mesh well with the evident rapid economic growth witnessed in the 1990s, unlike findings in the literature of a slowdown in TFP growth during the reform periods, which were described previously. The main difference in our approach is regarding the appropriate price deflator. Our methodology does not suffer from the biases inherent in "single" and "double" deflation methods described earlier, and the weights for the various inputs and output that make up the manufacturing sector at the 2-digit level are derived extensive input-output information spanning both pre and post-reform years, that allows us to pick up any changes in the industrial structure that may escape attention in a fixed-weights deflation approach. Figure A.1 in the Appendix (based on Table A.8) plots indices of TFP and TP derived under different assumptions and approaches to pricing intermediate inputs. On the left panel, we plot TFP indices derived under single deflation, double deflation, and Divisia-type deflation respectively. A feature of the price deflation method used for the TFP series is that they are based on fixed weights for the various commodities composing intermediate inputs (except in the case of single deflation where intermediate inputs are not separately deflated). We see that single deflation produces a TFP series that shows a decline in the late 1970s and, then remains relatively flat in the 1980s. On the other hand, double deflation produces a series that shows a steep increase in manufacturing TFP in the 1970, and sharp declines in the 1980s. Divisia type deflation by fixed-weights (Rao 1996) also 46 produces a series showing sharp increases in the 1970s but steep declines in the 1980s and in the early 1990s. On the right panel, where we plot indices of TP derived from Divisia type deflation using flexible weighting procedures. In comparison to the TFP series, these series are relatively smoother, and the index derived in this study clearly shows a rising trend in the 1990s. 3.3.2. Regional results Table 6 presents the sources of growth of real output per worker at the state level. To focus on convergence, the states are divided into three groups of roughly equal size, based on levels of output per worker in the base year (1970). The period under review is divided into the 1970s, and 1980s, and 1990s and beyond. In the first three columns of this table, we note that there is evidence in support of convergence in the 1980s, as output per worker grew faster in the middle and bottom group of states than in the top states. This process appears to continue in the 1990s, albeit only for the bottom group, and at a lower growth differential compared to that observed in the 1980s vis-a-vis the top group. The growth rates of the respective inputs per worker suggest that in the 1980s, faster growth of capital per worker and of materials per worker (especially for the bottom group) drove the convergence in output per worker. In the 1990s, for the bottom group, capital per worker continued to grow faster than for the top group, and TP growth was somewhat higher in the 1990s relative to the top group as well. TP growth appears to have played a smaller role in the observed convergence in the 1980s, as TP growth rates were very only slightly higher in the 47 bottom group relative to the top, in the 1990s. Even though the contribution of TP growth to convergence in output per worker is marginal, the contribution manufacturing output growth has been quite significant. In the 1980s, across the three groups, roughly 20 percent of the observed output growth could be attributed to TP growth, which increased in the 1990s to almost 25 percent. In contrast, the contribution to growth of capital per worker has stayed roughly constant at 13 percent of output growth for the two top groups, and fallen from 20 percent to 13 percent between 1980s and 1990s for the bottom group. (The remainder of output per worker growth is accounted for growth in material inputs per worker). Thus, TP growth had a larger contribution to growth compared to capital intensity, which echoes the cross-country findings of Hulten and Isaksson (2007)44. One can observe convergence across states in TP growth, which show that states that started off with lower initial levels of TP witnessed relatively faster TP growth rates in subsequent periods. We estimated TP levels by extending the methodology used for the all-India sources of growth analysis to state data45. In Table 7 we see that firstly, TP levels increased in all states at the end of the 1990s compared to the 1970s, 44 The Hulten-Isakkson study is based on TFP growth, using value added data, hence the contribution to growth can exceed 50 percent. Our growth accounting results are based on gross output, which if converted in value added terms, would yield similar ratios. 45 State-year time series of our estimates of TP levels are available in the Appendix Table A.7 48 which mirrors the results found at the all-India level. When we examine TP growth rates against initial levels of TP, we observe that over 1970-2003, middle and bottom ranked states (ranked by level of TP in 1970) had higher TP growth rates than the top group (Table 8). This catching up was most noticeable in the 1980s, whereas in the 1990s, the growth rates are virtually identical. A result to note is that the gap in levels between top and bottom, and top and middle states, fell from 13 percent to eight percent and from six percent to 5.3 percent respectively. Thus, although the gap in TP levels has narrowed across states, some gap appears to be persistent. Given the relatively small share of manufacturing value added in state GDP (average of 11.6 percent46 in 2003-04), output per worker and relative TP levels across states appear not to be strongly related to levels of state per-capita income across states. We ranked states at two different points in our sample (1970 and 2003, in Table 9 and 10 respectively) according to levels of output per worker, TP, and per capita income, to see if there was a noticeable overlap between the various ranks. For output per worker, we do not find this to be the case. The correlation between output per worker and income per capita ranks was 0.3 in 1970, and negligible in 2003. We do find a higher correlation for TP levels and income per capita, (between 0.65 and 0.70 in 1970 and 2003 respectively). However, this need not imply that higher levels of manufacturing TP cause higher per capita incomes, as they may be themselves be endogenous to per-capita income. 46 Reserve Bank of India data for the 17 states includes in this study. 49 The results on regionalization of manufacturing sector growth can be summarized as follows. Convergence in growth rates of output per worker is noticeable for states with lower initial levels of output per worker in manufacturing. This was especially evident in the 1980s and to a lesser extent in the 1990s. The main force behind convergence appears to have been faster growth in capital intensity and materials per worker, especially in the lowest ranked states (in terms of initial levels of output per worker). TP growth appears not to have had a major role in the observed convergence in output per worker; however, when TP growth rates are examined relative to TP levels in the base year, we find that states with lower initial levels of TP witnessed relatively faster TP growth in the 1980s. In the 1990s, TP growth rates appear to be almost identical, and the gap in TP levels in the 1990s is narrower than that in the 1980s, and TP levels have increased in all states between 1970 and 2003. Finally, there does not appear to be a noticeable relationship between output per worker in manufacturing and state income per capita, on account of the relatively small share of manufacturing in state GDP. One important implication of these results may be that the observed increase in income inequalities may not be driven by differences in manufacturing sector growth. Given that levels of output per worker are have little correlation with income per capita, the convergence in growth rates of output per worker, and the relatively low share of manufacturing in state GDP, the sources of widening income inequalities would have to be from other sectors, namely services and/or agriculture. 50 Chapter 4: Infrastructure Spillovers 4.1. Description of variables The task in this section of the dissertation is to explore the relationship between productivity in regional manufacturing (measured by TP) and the level of infrastructure. In particular we are interested in spillover from infrastructure, and these spillovers may exert an effect beyond the geographical boundaries of a given state. Our sample consists of 15 major states47, from 1970 to 2003. Infrastructure is widely understood to be a major supply-side bottleneck in India, particularly for the manufacturing sector. The Eleventh Plan (2006-7 to 2011-12) envisages infrastructure investment needs of more than US $500 billion (about 40 percent of India?s current nominal GDP) across 10 infrastructure sectors, in order to meet GDP growth targets of 9-10 percent. Almost 50 percent of the additional 47 Andhra Pradesh, Assam, Bihar, Gujarat, Haryana, Karnataka, Kerala, Maharashtra, Madhya Pradesh, Orissa, Punjab, Rajasthan, Tamil Nadu, Uttar Pradesh, and West Bengal are included in the sample. These 15 states comprise about 95 percent of the population the country. This set only excludes the small north-eastern states and Union Territories on account of incomplete data for both ASI manufacturing and infrastructure variables. Bihar and Madhya Pradesh were bifurcated in 2000 to form 2 new states, Jharkhand and Chhatisgarh. We continued to treat the two states as undivided, by aggregating the data from 2000 onwards for all the relevant variables. Although Himachal Pradesh and Jammu and Kashmir are included in our study of sources of growth in manufacturing, we exclude them from the econometric analysis as these states are mountainous, with limited additional road-building, and a large proportion of roads are built and maintained by the army and border security forces. Moreover, the data show downward jumps, in highway length, which could either be on account of destruction due to natural disasters, or due to other reasons. 51 investment is to be directed towards electricity and roads, with electricity itself accounting for more than a third of the total planned investment in all 10 sectors48. Given the relative importance attached to roads and electricity we restrict our attention to these stocks. Table 11 and Table 12.1 - 12.2 list the national and state- wise stocks of national and state highways, and installed generation capacity in electric utilities. Highways: The total road network of India is 3.34 million km, which is the second largest in the world, and at present carries 65 percent of freight and 85 percent of passenger traffic. Roads are classified on the basis of administrative status and by function. In this study we include National Highways (that are built and maintained by the central government), and State Highways (that are managed by states). We consider these two types of roads, and exclude local roads, since from an economic standpoint, highways are important, linking up national and state capitals, ports, rail- heads, and link with roads outside the country. Also, these roads are more likely to be associated with spillovers across state boundaries, than local roads. However, despite their economic importance, national and state highways are a much smaller subset of total road length, comprising 65,569 km, and 130,000 km respectively49. National highways thus comprise less than two percent of the total 48 http://infrastructure.gov.in/pdf/IBEF.pdf 49 Although the responsibility for maintaining national and state highways falls to different bodies (National Highways Authority of India, and state governments 52 network, but carry 40 percent of the total traffic. Since roads are mostly provided in the public domain, the allocation of funds made available for road building tends to favour the demands of the largest constituency. Moreover, roads construction is the responsibility of state governments (except in the case of national highways). Hence, local road projects such as rural roads and intra-district roads are likely to be preferred over other alternatives, such as multi-laning of state highways or pavement strengthening. Raw data are supportive of this conjecture. Between 1971 and 1991, one-half of the addition to the stock of surfaced roads (265,158 km out of 540,675 km) was local roads. In the second sub-period (1992-2003), total addition to the sum of National and State highways, and public works department (PWD) roads in 17 major states was 249,254 km (not differentiating between lane-km), of which 22,766 km were National highways, 6,453 km were State highways, and the balance 232,287 km were PWD local roads. As pointed out above, most of the traffic moves on highways although bulk of the additional roads is not in this category. Since 2001, the National Highway Development Project (NHDP) was launched under the aegis of the National Highway Authority of India (NHAI), to undertake major upgrades and add to the National Highway network of the country. In all, the programme has a total outlay of roughly US $40 billion. The first pieces of this network were the Golden Quadrilateral and The East West-North South corridor, comprising long stretches linking major metropolitan cities and state capitals along respectively), there may be significant overlap between the two categories, as they constitute a continuous physical network. For this reason, we considered the aggregate of both national and state highways in this study. 53 the route. We are unable to include the data from this project in our analysis as it falls outside our reference period. Electricity: The supply of uninterrupted, high quality power is a prerequisite for industrialization. However, India's electricity generation is frequently reported to fall short of demand. For the year 2009-10, peak power deficit is expected to reach 12.6 percent of demand, up from 11.9 percent the previous fiscal year50. There has been significant growth in captive or own-account power generation by firms over the years, which may be seen as a sign that utilities power is not sufficient for industrial users. Estimates of captive power generation range between 14,000 and 20,000 MW, amounting to 20-25 percent of the installed capacity of utilities. This is up from between 10-12 percent over 1972-92 (HBS 2006). We do not include captive generation in our analysis, as we lack adequate data at the state level. In the sample of 15 major states considered here, installed generation capacity in electric utilities alone has gone up from about 13,000 MW in 1970 to about 50,000 MW in 1991, and to more than 75,000 MW in 2003. Aggregate installed capacity was 104 GW in 2002, comparable to the generating capacity of UK and Germany, but in per capita terms, India's consumption (363 kWh per capita) is less than a thirtieth of that in the US, and about one-fifteenth of that in UK. 50 http://in.reuters.com/article/domesticNews/idINDEL44867220090710 54 Bulk of the generating capacity in India (about 60 percent in 2002) falls under the management and ownership of State Electricity Boards (SEBs). These SEBs suffer from routinely high losses, and face a number of problems. Firstly the average plant load factor (PLF) at which state-owned utilities run is only 65 percent, leading to high costs. Secondly, states cross-subsidize agriculture at the expense of industrial power, and vested interests make it increasingly hard to levy user charges on this politically powerful segment. For instance, in 2001-02, domestic and agricultural users consumed a total of 50 percent of generated power. Domestic users paid 195.6 paise51 per kWh, and agricultural users paid a subsidized rate of 41.6 paise per kWh. In contrast, industrial consumers used 29 percent of generated power and paid almost twice the domestic rate (378.7 paise per kWh). Industry thus subsidizes domestic and agricultural power consumption. Thirdly, there are large transmission and distribution (T&D) losses52, which amounted to 28 percent of generated power in 2001-02. In certain states, the losses are almost a half of generation, such as Haryana (40 percent), Rajasthan (43 percent), Andhra Pradesh (45 percent), and Orissa (51 percent). A few caveats are in order for the electricity variable. For our purposes, the ideal measure would be the actual amount of electricity transmitted to the manufacturing 51 One paise = 100th of one Indian rupee. 40 Indian rupees are roughly equivalent to one US dollar. 52 These losses are often on account of weak enforcement and outright theft of electricity, for which reason they are often referred to as ?theft and dacoity? losses. 55 sector. Installed capacity in utilities is only a proxy, as the amount of power actually generated and delivered can vary significantly from state to state and over time. Moreover, it excludes captive power generation. Thus, this measure may be noisy, and not give us an accurate picture of the impact of electricity on regional manufacturing productivity and growth. Secondly, the electricity sector has witnessed significant institutional changes since the enactment of the Electricity Act of 2003. Among other things, an important aspect of the changes envisaged under this Act is the setting up of a mechanism to help deepen and develop the power- sharing market, whereby electricity can be traded across regions in a quick and timely manner depending on varying demand and supply conditions. This may have a bearing on our results, since prior to this (including in our sample period), while power trading may have taken place across state boundaries, it may not have been rapidly responsive depending on demand and supply conditions. Thus, spatial spillovers from electricity generation capacity may be harder to detect in the data from Indian states. 4.2. Descriptive statistics and preliminary exploration Some of the problems in estimating infrastructure spillovers in our panel become evident from the descriptive statistics. One issue is the presence of a strong time trend in the productivity and infrastructure variables. As shown in Table 12.1, within each state, correlation between the time trend and the highways variable ranges from 0.47 to 0 96, and the correlation between electricity generation capacity and the time 56 trend ranges from 0.90 to unity (Table 12.2). This points to potential problems in distinguishing the time trend from the infrastructure variables, notably electricity, in our estimates. There is also a high degree of correlation among the infrastructure variables themselves, as shown in Table 13, which presents the average (across states) of cross- correlations among the independent (infrastructure) variables. We have defined two variants of the infrastructure variables; "own" stocks refer to levels of road and electricity generation capacity within a given state's borders, whereas "contiguous" stocks are the (appropriately normalized and weighted) level of roads and electricity in states adjoining a given state53. Both the own and contiguous levels of highways, and own and contiguous levels of electricity generation capacity are also highly correlated, apart from the high correlation of each of these with the time trend. This suggests that multicollinearity might lead to estimation problems, if we use within- state variation over time, to identify the relevant coefficients. In the cross-sectional dimension, there appears to be some evidence that states with higher stocks of infrastructure also have higher levels of TP. We present scatter plots of the state-wise time-average of TP levels stocks of highways and electricity generation capacity. Looking at own stocks, Figure 4 and Figure 5 show that across states, on average, states with higher TP levels also had bigger stocks of highways 53 The weighting procedure is described in the next section that lays out the econometric methodology. 57 and electricity. When we consider combined own and contiguous stocks, the relationship with roads still appears to be positive (Figure 6), whereas the relationship with electricity looks relatively flatter (Figure 7). Though the positive relationship suggested by these figures is not very strong, it may yet raise questions about the direction of causality, that higher productivity drives infrastructure levels, rather than vice-versa. We note here that the provision of infrastructure is not based on the requirements of any one sector alone; rather it serves a wider constituency including households, agriculture, and services, and hence is unlikely to be "caused" by purely changes in manufacturing sector productivity alone. The registered manufacturing sector that we are considering is itself a subset of the industrial sector, which accounts for less than a fifth of GDP. There is also the policy thrust of balanced regional growth, which would tend to push infrastructure provision towards relatively under-developed states, than to states with a relatively well-established manufacturing sector. 4.3. Model specification Recall that the Hicks-neutral shift parameter of the production function described in Section 3 was modeled as a function of infrastructure stock, and a time trend that captures the autonomous rate of technical change. We now give this parameter a specific functional form: 58 ,),()8( ,0,, ?? tititi BeAtBA = tiB . is the stock of infrastructure in state i )...1( Ni = at time t )...1( Tt = , the initial level of productivity is given by the constant iA ,0 , ? is the exogenous rate of technical change, and ? measures the infrastructure externality. The total productivity measure tiTP, derived from the regional growth decomposition exercise is our measure of A . Substituting tiTP, into (8) and take logs: )9( .lnlnln ,0, titi BtTPTP ?? ++= In this form, ? measures the elasticity of TP with respect to infrastructure stock within a state, and this equation forms the key building block of our estimating equations. In this form, the term ? captures the "own" spillovers from infrastructure. We employ the spatial econometric model developed by Kelejian and Prucha (1998, 1999), to expand the basic equation in order to take into account spatial spillovers from infrastructure. The spatial framework is useful for modeling various types of spatial effects, including spatial correlations in the dependent variable (known as the spatial lag model), spatial spillovers from the independent variable(s), and spatially correlated errors (known as the spatial errors models). A model may include any or all of these types of effects. In our application, we incorporate spatial spillovers from the independent variable (infrastructure), given that physical infrastructure is essentially a network and can thus exert an effect beyond where it 59 resides in space. The natural criterion for considering spatial spillovers in this context is adjacency, since states that share borders are most likely to experience spillovers from each other's infrastructure, than states not linked, or those at one remove from the immediate vicinity. Adjacency effects can be modeled by augmenting equation (9) in the following way. Let W be an N x N weighting matrix where the elements of row Ni ...1= assign weights to state i 's neighbours ,,, Njiij ?? based on some notion of `distance'. In this study, weights are assigned as 1, =jiw if ji, share a contiguous border, and 0 otherwise. The weighting matrix is then "row-normalized", such that ,1, 1 =? = ji N j w where jiw , is the weight assigned to state j by the state in row i . The value assigned to the diagonal is zero, that is, a state is not considered its own neighbour. In our scheme, weights are assigned in a uniform way, depending on the number of neighbours that a state has. However, the ideal weighting matrix for our purposes might take into account, for example, the density of traffic flow between two adjacent states to weight the neighbourhood stock of national and state highways, rather than uniformly weighing the stock of all neighbours. Similarly, we may prefer to weight the electricity stocks of adjacent states depending on actual power-trading data, or based on the sharing of transmission infrastructure. Unfortunately, these alternatives are not available to us on account of lack of data. Moreover, for our purposes, while the simpler weighting scheme may not be able to pick out how important some states 60 are in a neighbourhood to each other, it will not overlook them altogether. Hence as a first approximation of a weighting scheme for these types of variables, the uniform weighting may be adequate. Let tB denote an 1?N vector of the stock of infrastructure in each state for a given year t . We then define ][ tBW ? as an 1?N vector containing weighted sums of neighbourhood infrastructure stock in year ,t appropriately normalized54. This natural log of this term, ]ln[ tBW ? is the additional term included in equation (9) that controls for adjacent levels of infrastructure. The thi? element of this vector is the log of the weighted sum of (normalized) infrastructure stocks in state si? neighbours, for year t . Let 1? denote the coefficient of within-state infrastructure stock )(ln tB and 2? the coefficient of the `neighbourhood stock' ]ln[ tBW ? . In addition to these regressors, we include a time trend variable ? which captures exogenous technical change. Thus our estimating equations becomes (including the constant term): 34...1;]ln[ln][ln)9( 21 =+?++?+= tuBWBteeTP tttNNt ??? The error tu is a combination of a spatially correlated component and an idiosyncratic component, to capture the possibility that spatially contiguous states 54 In the case of national and state highways, we normalize by the area in square kilometers of the state, and for electricity, by population (per kilowatt of generation capacity). 61 may experience spatially correlated shocks. Let tu and t? be 1?N vectors of error terms, and tu is drawn from the following process: ,][)11( ttt uWu ??+?= where ))(,0(.. 2 NIdiit ?? ? 34,...,2,1=?? ?t . One implication of this assumption is ,)var( 2, ?? =ti and 0)cov( ,, =tjti ?? , unless ji = and .ts = The scalar parameter ? is assumed to be time-invariant, and represents the spatial correlation across error terms. This specification allows errors to be correlated within t across space, but uncorrelated over time. This variant is also known as the Spatial Errors Model, or SEM. We can now stack the data by year. Let ,)34,...,3,2,1( ?=T TI be an identity matrix, Te )1,...,1,1( ?= be a 1?T unit vector, D be an )1()1( ??? NN matrix of dummy variables, and TPln be the 1?NT vector of productivity levels in each state and Bln be the 1?NT vector of infrastructure stock in each state. This gives us: ;])ln[(ln][ln)13( 21 uBWIBeTeeTP TNTN +??++?+?= ??? where ,][ ?+?= uWIu T where u and ? are 1?NT vectors. I estimated this model using Jim LeSage's Spatial Econometrics Toolbox for Matlab, which is based on a GMM estimation approach as detailed in Kelejian and Prucha (1998, 1999). 62 4.4. Regression results The results of this exercise are presented in Table 14. The baseline regression of TP on a constant and the time trend shows the time trend is positive and significant, and implies an autonomous rate of technical change of approximately one percent per year, that remains fairly stable in the various specifications. HBS (2006) find lower estimates of 0.4 percent per year, but this may be attributed to our extension of the data to include several years after reforms in 1991, whereas HBS covers the period from 1974 to 1993. The remaining columns consider the infrastructure variables. In column 2 we introduce own stocks of national and state highways. The coefficient is positive and significant, and although the size of this coefficient varies somewhat in the different specifications, it remains significant in all specifications in columns 1-9 where the variable appears. The magnitudes of implied elasticities for own highways are small, ranging from 0.07 ? 0.1 percent in the different equations. Thus, a 10 percent increase in the "own" stocks of national and state highways would have a spillover impact of at most one percent increase in TP/output levels. But it is worth keeping in mind that these are spillover effects, over and above the direct contribution of infrastructure to manufacturing output as an intermediate input. In column 3 we consider only adjacent stocks of highways. The estimated coefficient is positive and significant, but the size of the coefficient is quite small. 63 When both own and adjacent stocks of highways are introduced into the equation (column 4), the adjacent stocks variable becomes negative (and significant), whereas the own stock variable becomes somewhat bigger than in column 2. When all infrastructure variables are considered together (column 9), contiguous highways have a positive impact, but the coefficient is not significant. Column 5 shows the effect of including only the own stocks of electricity generation capacity, along with the constant and time trend. Again, we find a small positive and significant impact, which remains significant and almost of the same magnitude when adjacent electricity stocks are introduced in the equation (column 7). In contrast, adjacent stocks of electricity appear to have a negative and significant impact in all specifications, though the magnitude of the effect is quite small, implying a negative elasticity of TP to adjacent stocks of electricity of roughly 0.3 percent, which is similar in magnitude to the positive elasticity of TP to own electricity stocks. Finally, we introduce both highways and electricity variables together. In column 8 we consider own stocks of highways electricity. Both variables have a positive and significant coefficient. In the final column, we control for all four infrastructure variables. Contiguous highways are not significantly correlated with TP, and own electricity, though positive, is no longer significant. Own highways continue to be positive and significant and adjacent electricity is still negative and significant. 4.5. Interpretation of results 64 Given that our estimates of spillover effects from infrastructure may suffer from problems due to features of the data, the results given above need to be interpreted with caution. Based on these results, there appears to be some evidence to suggest that own highway infrastructure may have positive spillover effects on TP, though as shown; the effect size is quite small, ranging between 0.07 ? 0.1 percent. At best, the results in column 9 of Table 14 would imply that ceteris paribus, if the stock of national and state highways within a state was doubled, the level of TP and manufacturing output would go up by around nine percent. Similarly the impact of own electricity generation capacity on TP also appears to be positive, though it is even smaller in size (elasticity ranging from 0.2 to 0.4 percent), and the coefficient is not significant in the full model including all infrastructure variables. In terms of adjacency effects, the coefficient on adjacent highway stocks changed signs between specifications, making it difficult to interpret. However, adjacent electricity stocks entered with a negative sign, the effect size implying a negative elasticity of around 0.3 percent. As pointed out earlier, various studies and planning documents have pointed out the scarcity of electricity in India, and one would expect that additional electricity in the immediate neighbourhood would have a positive impact on TP and manufacturing output. Thus, one would need to exercise even more caution in interpreting this particular finding. It is possible that part of this negative impact reflects the sorting of high productivity firms to states that are comparatively electricity-rich. This is especially likely if states cannot freely trade 65 electricity across state lines. In fact, attempts to deepen and develop the market for power trading are relatively recent in India, marked with the passing of the Electricity Act in 200355 as highlighted earlier. The lack of a well-developed market in power trading in prior periods may therefore explain in part our finding of a negative spatial spillover from electricity. A related issue is that the amount and quality of electricity available to manufacturing industry across states may be insufficient to spur innovations, especially of the type that require uninterrupted and high quality power supply. In other words, the first problem in India may be to achieve adequacy in electricity generation, given existing manufacturing technology. The current level of electricity available to manufacturing may be below what is required even for running existing technology efficiently, and thus we do not observe positive spillovers that are spurred by relative abundance. Finally, as mentioned earlier, generation capacity in utilities may not be a good proxy for the amount of electricity transmitted to the manufacturing sector. Installed capacity must be used to supply agriculture, services, and households, apart from manufacturing industries. Moreover, electricity transmission and distribution losses in India are quite high, as high as 50 percent in some states56. The substantial growth of own-account power generation by firms may reflect this insufficiency. Thus, our results should not be interpreted to imply that 55 http://www.electricityindia.com/powertrading.html 56 http://www.teriin.org/upfiles//pub/papers/ft33.pdf 66 electricity has no spillover benefits; rather the results may be pointing to a shortage relative to requirements, due to which the kind of spillovers we expect are not observed. Last but not least, it is worth remembering the sheer size of Indian states, which might rule out large spillover effects per se. 4.6. Other specifications This section contains results from other specifications that were also attempted, in addition to the spatial errors model that I chose to interpret. I considered a fixed effects estimator, and a "between" estimator, but the resulting coefficients either had implausible signs and magnitudes, or were not robust. Table 15 presents the results from the fixed effects estimator. Reading across the columns, the coefficient on own national and state highways stock is positive, but not significant in all specifications. On the other hand, the adjacent highways variable appears to have a much larger magnitude than own highway stocks. It seems counterintuitive that spatial spillovers from roads would exceed own spillovers, considering the fairly large size of Indian states. Also, both electricity variables enter with a negative sign, and both are highly significant. The magnitude of the coefficient on adjacent electricity appears to be too large to be plausible, as does the negative sign on own electricity stocks. Results from the "between" estimator are in Table 16. As can be seen, in this specification I did not find any evidence of spillover effects from infrastructure, which is contrary to earlier findings on Indian states by HBS (2006). Finally, I also 67 considered a variant of the spatial errors model, which included a full set of state dummy variables (Table 17). The inclusion of state dummy variables leads to the implausible result that own spillovers from highways are not significant in most specifications, whereas spatial spillovers from adjacent highways are positive, and significant. Similar to the findings from the fixed effects model, both own and adjacent electricity variables enter with a negative sign, and while the own electricity variable is in general not significant, the adjacent electricity variable is highly significant, and has a large negative sign, implying very high negative spillovers from neighbourhood stocks of electricity. 68 Chapter 5: Summary and Conclusions Industrial development is believed to hold the key to rapid and sustained economic growth in India. While this was well-recognized by policy-makers at an early stage as reflected in various planning documents, the approach to the problem has changed considerably over time. The issue of sustainable growth raises questions about what policies can help to achieve this objective. Concurrently, India is undergoing a transformation from a planned economy characterized by industrial regulation and licensing, administered prices, and restrictions on private activity to a more market- driven economy more encouraging of entrepreneurship. Focusing on the formal manufacturing sector, one of the main questions that we sought to answer is one that has triggered much debate, namely whether liberalization of the economy had a positive impact on productivity growth, and we found that the answer was affirmative. This contributes to a long-running debate generated by the contentious finding of some research that liberalization led to a distinct deceleration in productivity growth, especially in the 1990s, which was the era of significant economic liberalization. We highlighted the methodological problems that arose in the literature, mainly dealing with the appropriate deflator for manufacturing sector aggregates. Correcting for these, we found that at the All-India level, manufacturing sector TP growth increased substantially in the 1980s and 1990s, relative to the 1970s. This is in contrast to findings by Balakrishnan and Pushpangadan (1994) and Rao (1996) with respect to productivity growth in registered manufacturing in the 69 1980s over the 1970s, and in contrast to Rodrik and Subramanian (2004), Goldar (2004), and Banga and Goldar (2007), who report negative TFP growth rates for economy and for the manufacturing sector in the 1990s. Compared with these findings of slowing productivity growth, our results seem to be more plausible when seen against the observed high growth rates of the Indian economy the 1990s. In the second part of this study we examined trends in output per worker and manufacturing sector TP growth at the state level. We found evidence of convergence in output per worker growth, though the difference in growth rates was higher in the 1980s than in the 1990s, between the top group and bottom group of states, as ranked by initial levels of output per worker. The data indicate that this convergence was driven mostly by capital intensity and only to a small extent by differences in TP growth. Examining trends in TP growth in greater detail, we found that TP growth had a bigger contribution to growth in output per worker than capital intensity, which accords with the recent findings of Hulten and Isaksson (2007), where they find that in a cross-country setting, TFP growth accounted for a major share of output growth in the aggregate economy. In terms of TP levels, we find that all states have higher TP levels in 2003 than in 1970. One interesting finding was that states with lower initial levels of TP had higher rates of TP growth, especially in the 1980s. By the 1990s, however, TP growth rates appear to be more or less equal across groups of states (ranked by initial levels of TP). 70 Several studies have documented the increasing disparity in income levels across states, and we use our evidence to explore the relationship between trends in state- wise income per capita and manufacturing sector growth. We related trends in state income per capita with output per worker and TP levels in manufacturing, and found that there was little evidence that states with higher levels of output per worker in manufacturing had higher levels of per capita income. This is perhaps due to the relatively small share of manufacturing in state GDP. Moreover, state-wise growth rates of output per worker in manufacturing appear to have converged in the 1990s. Read together with the evident lack of correlation between levels of output per worker and per capita income, and the low penetration of manufacturing in state GDP, it appears that the rising income inequalities across states are not driven by manufacturing sector trends, but may perhaps be on account of differences in other sectors. Finally we addressed the issue of infrastructure and productivity growth. In pursuit of balanced growth, and for sustained high growth rates particularly in the industrial sector, infrastructure development has been identified as a vital input. While physical infrastructure plays a key role in manufacturing as an intermediate input, we hypothesized that roads and electricity may also affect manufacturing productivity through network spillover effects on manufacturing output via their impact on TP levels across states. We also incorporated the possibility of spatial spillovers. 71 Due to limitations in the data, it is hard to draw firm conclusions from the results of this exercise. However, despite the lack of robustness in the results, and problems with the underlying data on account of collinearity among various explanatory variables, the evidence does seems to suggest the presence of small spillover effects from a state's own stock of roads, the elasticity of TP to own national and state highways (in length per square kilometer of state area) of around 0.07 ? 0.1 percent. Own stocks of electricity generation capacity also appear to have a positive and significant impact on TP, though the effect is smaller than that estimated for highways. The elasticity of TP to additional electricity generation capacity (in kW per capita) was found to be around 0.02 ? 0.04 percent. Electricity generation capacity in neighbouring states appears to have a negative impact on state TP, i.e. spatial spillovers from electricity appear to be negative, contrary to our expectations. On reflection, this may be driven a variety of factors. It may reflect in part a dynamic sorting of high TP firms to areas with relatively abundant electricity, and may also reflect certain weaknesses in our measure of electricity. The ideal measure in this context would be the amount of electricity transmitted to the manufacturing sector, given the prevalence of leakages in transmission and distribution, installed capacity may be less than ideal for our purposes. Secondly, the market for sharing power across state boundaries is still evolving in India, thus one may observe the expected positive spatial spillovers as this market matures. On a related note, the relatively small size of positive spillovers from own-installed capacity might reflect the extent of electricity shortage; existing 72 power generation is short of demand to such an extent that additional capacity does not attain a level necessary to spur technological improvements of the type that require a relative abundance of high quality and uninterrupted power. 73 Appendices Appendix 1: Text tables and figures Table 1. Share of manufacturing in state GDP, 20001 (In percent) Gujarat 24.3 Andhra Pradesh 10.3 Haryana 18.7 Himachal Pradesh 9.6 Maharashtra 17.1 Assam 9.6 Tamil Nadu 15.5 Madhya Pradesh 9.5 Karnataka 14.5 Rajasthan 9.2 Punjab 12.9 West Bengal 8.5 Uttar Pradesh 11.1 Kerala 7.6 Bihar 10.9 Orissa 5.7 Mean 12.2 St. Dev. 4.8 1 Sample excludes Jammu and Kashmir on account of small size in GDP (2.8%) 74 Author(s) Period Deflation Method Ahluwalia 1965-85 1965-80 1980-85 1991 Single Deflation1 -0.3 3.4 Balakrishnan - Pushpangadan 1970-88 1970-79 1980-89 1970-89 1994 Single Deflation1 -0.71 1.47 0.38 Double Deflation2 3.5 0.19 1.84 Dholakia and Dholakia 1970-88 1970-79 1980-89 1970-89 1994 Single Deflation1 -1.69 1.89 -0.11 Double Deflation2 0.56 2.86 1.58 J M Rao 1970-92 1973-79 1980-92 1973-92 1996 Single Deflation1 0.33 1.51 1.14 Double Deflation2 6.6 0.02 2.09 Divisia deflator3 8.37 -1.78 1.43 Hulten and Srinivasan 1973-92 1973-82 1983-92 1973-92 1999 Divisia Deflation4 2.2 2.1 2.2 B. Unel 1979-97 1979-90 1991-97 1979-97 2003 Single Deflation1 1.8 2.5 1.8 B. Goldar 1979-00 1979-90 1991-97 1991-99 2004 Double Deflation2 2.14 1 1.57 Banga and Goldar 1980-99 1980-89 1990-99 1980-99 2007 Double Deflation2 1.3 0.5 0.8 This study 1970-03 1973-80 1981-90 1992-03 2010 Divisia Deflation5 -1.2 1.2 1.8 5 Divisia deflation with flexible Input-Output matrix weights 4 Divisia-type deflation with flexible weight price indices Table 2. Summary of Results for Registered Manufacturing Productivity Growth Productivity Trends 1 Based on manufacturing price index 2 Input deflator based on Input-Output 1973-74 weights for input commodities 3 Divisia-type deflation with fixed weight price indices using Input Output 1993-94 data 75 Table 3. Partial Productivity Measures Output per worker1 Capital intensity Capital- output ratio 1970-79 2.1 0.6 0.31 1980-89 7.1 1.4 0.44 1990-03 6.9 4.1 0.55 1 Average annual growth rate 1973 - 2003 1973 - 1990 1991 - 2003 1973 - 1980 1981 - 1990 1992 - 2003 Gross Output 7.5 7.5 7.5 6.2 8.4 8.4 Materials 6.9 7.6 5.9 8.1 7.3 6.5 Labour 1.4 2.0 0.6 4.3 0.4 0.5 Capital 8.8 9.3 8.3 6.7 11.1 8.2 TP 0.8 0.2 1.5 -1.2 1.2 1.8 Share in Nominal Gross Output Intermediates 78.2 77.6 78.9 76.5 78.5 78.9 Labour 8.0 9.4 5.9 10.6 8.5 5.8 Capital 13.9 12.9 15.2 12.9 13.0 15.3 Table 4. Sources of Growth (1973-2003) - All India Aggregates1 (In percent) 1 Excluding electricity, gas and water supply sector 76 (In percent) 1970-79 1973-79 1980-89 1990-97 1990-99 2000-03 1990-03 Series A1 -1.01 0.80 1.50 1.49 1.50 Series B2 0.03 1.17 1.70 Series C3 0.03 1.17 1.88 1.68 1.82 1 1973-2003, excluding electricity, gas, and water supply sectors. Table 5. Averages of TP Growth Rates 2,3 Including electricity, gas, and water supply sectors. Data for this sector are from ASI published series in Series B (1970-1997), and imputed for years after 1997 in Series C. 77 1970-79 1980-89 1990-03 1970-79 1980-89 1990-03 Top group1 2.7 6.8 8.0 3.0 6.1 6.2 Middle group1 2.4 7.6 7.2 3.4 6.8 5.7 Bottom group1 2.2 8.6 8.5 2.6 9.6 7.1 1970-79 1980-89 1990-03 1970-79 1980-89 1990-03 Top group1 6.0 5.0 7.0 -0.3 1.4 1.9 Middle group1 1.9 5.4 6.8 -0.1 1.6 1.6 Bottom group1 4.5 9.0 6.1 0.8 1.5 2.1 Table 6. Sources of growth of output per worker in state level manufacturing (Average annual growth rates) 1 Groupings are based on level of output per worker in 1970 Output per worker Capital per worker Material per worker Total productivity (TP) 78 % change 19702 20033 1970 2003 1 Andhra Pradesh 1.021 1.385 7 5 36 2 Assam 1.015 1.302 9 15 28 3 Bihar 1.011 1.367 10 7 35 4 Gujarat 1.089 1.376 2 6 26 5 Haryana 1.048 1.423 5 4 36 6 Himachal Pradesh 0.947 1.450 14 1 53 7 Jammu and Kashmir 0.859 1.345 16 10 56 8 Karnataka 0.985 1.249 11 16 27 9 Kerala 0.981 1.337 13 12 36 10 Madhya Pradesh 1.016 1.353 8 8 33 11 Maharashtra 1.141 1.440 1 3 26 12 Orissa 0.833 1.049 17 17 26 13 Punjab 1.052 1.442 4 2 37 14 Rajasthan 0.938 1.315 15 14 40 15 Tamil Nadu 1.040 1.338 6 11 29 16 Uttar Pradesh 0.982 1.328 12 13 35 17 West Bengal 1.052 1.353 3 9 29 2 Average over 1970-75 3 Average over 1998-03 TP level Rank Table 7. State-wise TP Levels and Ranks1 1 Data for electricity, gas and water supply sector based on extrapolations after 1997 79 (By tercile of TP level in 1970, in percent) Tercile 1970-03 1970-90 1991-03 1970-80 1981-90 1992-03 1970 2003 Top 1.1 0.6 1.8 -0.3 1.4 2.2 1.1 1.4 Middle 1.3 0.9 1.9 -0.4 2.1 2.3 1.0 1.3 Bottom 1.4 1.1 1.8 -0.4 2.6 2.1 0.9 1.3 % Gap in TP level Top - Bottom 13.0 7.9 Top -Middle 5.7 5.3 TP Levels2TP Growth Rates2 Table 8. State TP Growth Rates - Evidence of Convergence1 1 16 major states, excluding Jammu and Kashmir from original sample 2 Average of states in tercile 80 Output per worker rank1 TP rank1 Per capita income rank2 Per-capita income2 Top Maharashtra 1 1 2 2422 Punjab 2 4 1 2664 Haryana 3 5 3 2350 Bihar 4 10 16 911 Madhya Pradesh 5 8 11 1348 Middle Gujarat 6 2 4 1931 Orissa 7 16 12 1309 Uttar Pradesh 8 12 14 1274 Tamil Nadu 9 6 9 1494 Rajasthan 10 15 15 1210 Karnataka 11 11 7 1510 Bottom Assam 12 9 13 1277 West Bengal 13 3 5 1764 Andhra Pradesh 14 7 10 1372 Himachal Pradesh 15 14 6 1683 Kerala 16 13 8 1505 1 Based on 1970 levels 2 For 1980-81, Reserve Bank of India (in 1980-81 prices) Table 9. Manufacturing Output per Worker, TP, and Per-capita Income (1980) 81 Output per worker rank1 TP rank1 Per-capita income rank2 Per-capita income2 Top Gujarat 1 6 6 22491 Himachal Pradesh 2 1 3 24830 Maharashtra 3 3 4 24767 Bihar 4 7 16 7735 Madhya Pradesh 5 8 13 12226 Middle Uttar Pradesh 6 12 15 10324 Haryana 7 4 1 28071 Karnataka 8 15 9 18289 Rajasthan 9 13 11 15299 Orissa 10 16 14 11802 Assam 11 14 12 13734 Bottom Punjab 12 2 2 26891 Tamil Nadu 13 10 7 20570 West Bengal 14 9 10 17915 Kerala 15 11 5 22786 Andhra Pradesh 16 5 8 19062 1 Based on 2003 levels 2 For 2003-04, Reserve Bank of India (in 2000-01 prices) Table 10. Manufacturing Output per Worker, TP, and Per-capita Income (2003) 82 National and state highways (km) Electricity generation capacity (mw) 1973 112574 13896 1974 113465 15050 1975 114263 16772 1976 114932 18130 1977 117823 20322 1978 119788 22938 1979 119561 24581 1980 121388 26122 1981 122087 27543 1982 122448 30232 1983 124561 32468 1984 124849 35306 1985 127065 38636 1986 127626 40039 1987 128144 41859 1988 142943 44132 1989 150703 46507 1990 152306 48445 1991 155522 49667 1992 156516 50386 1993 157975 52565 1994 160806 55418 1995 161960 55971 1996 163217 57580 1997 165409 60580 1998 168498 63799 1999 179751 66703 2000 179813 67585 2001 185950 70184 2002 188851 74467 Roads 0.97 Electricity 1.00 Table 11. All-India Highway and Electricity Stock1 Correlation with time trend: 1 Sum of 17 states in current sample. 83 Andhra Pradesh Assam Bihar Gujarat Haryana Himachal Pradesh Jammu & Kashmir Kerala Karnataka Maharashtra Madhya Pradesh Orissa Punjab Rajasthan Tamil Nadu Uttar Pradesh West Bengal 1973 7771 3351 6293 9835 3457 3516 1202 2792 7973 17553 13310 3893 2839 10809 3597 10610 3773 1974 7812 3351 6293 9957 3752 3575 1205 2792 8054 17703 13310 3893 2839 10828 3610 10715 3776 1975 7812 3351 6294 10007 3772 3638 1232 2857 8058 17715 13855 3844 2869 10831 3610 10740 3778 1976 7812 3617 6302 10061 3779 3683 1238 2857 9522 17975 13932 3825 2877 9354 3610 10664 3824 1977 7812 3617 6307 10116 3779 3944 1238 2857 9583 20324 14074 3803 2877 9375 3610 10683 3824 1978 7815 3620 6309 10329 3788 3944 1302 2859 9620 21843 14074 3820 2877 9375 3663 10676 3874 1979 7815 3620 6309 10405 3791 4002 1302 2859 9770 21461 14134 3833 2877 9375 3678 10456 3874 1980 7815 3627 6309 10518 3791 4004 1281 2865 9770 21786 14168 4465 2877 9375 3679 10440 4618 1981 7815 3727 6309 10579 3791 3837 1281 2892 9781 21894 14179 4465 2877 9786 3681 10441 4752 1982 7815 3727 6309 10579 3791 3840 1281 2918 9781 22072 14319 4465 2877 9786 3681 10430 4777 1983 9422 4002 6309 10626 3791 3912 1281 2918 9880 22026 14382 4458 2877 9786 3681 10433 4777 1984 9422 4029 6309 10706 3791 3916 1281 2883 9880 21999 14411 4458 2877 9978 3697 10435 4777 1985 10898 4090 6310 10808 3792 4128 1336 2860 9880 22007 14393 4470 2877 9978 3719 10434 5085 1986 10976 4100 6310 10863 3792 4128 1336 2841 9880 22197 14393 4552 2940 9978 3747 10508 5085 1987 10976 4114 6310 10954 3792 4128 1336 2841 9880 22479 14509 4552 2940 9981 3764 10503 5085 1988 11018 4122 6310 10954 3792 4241 1336 2861 13167 33503 14589 4552 2927 9981 3768 10737 5085 1989 11238 4122 6310 16996 3792 4454 1336 2865 13223 32913 14650 4552 3146 10075 3768 12178 5085 1990 11238 4122 6310 18002 3792 4392 1336 2865 13228 33215 14686 4552 3146 10075 3779 12483 5085 1991 11238 4122 6310 20620 3792 4392 1336 2865 13279 33543 14693 4552 3146 10087 3917 12524 5106 1992 11238 4122 6310 20962 3792 4402 1336 2865 13285 34249 14711 4552 3146 9982 3922 12536 5106 1993 11333 4149 6310 21061 3792 4385 1336 3059 13285 34249 14731 5633 3154 9997 3931 12531 5039 1994 11726 4149 6310 21181 3792 4390 1336 3059 13392 34900 14731 5640 3154 11566 3931 12510 5039 1995 11726 4149 6310 21227 3792 4399 1336 3059 13392 34900 14736 5640 3154 12656 3935 12510 5039 1996 11726 4149 6310 21289 3792 4404 1336 3372 13392 35207 14765 5985 3154 12852 3936 12509 5039 1997 11755 4149 6310 21333 3792 4415 1336 4780 13392 35317 14765 6209 3364 12893 3936 12530 5133 1998 11813 4257 6739 21798 4074 4536 1336 4780 13403 35480 15169 6280 3364 13148 4493 12704 5124 1999 12403 4547 7271 22037 4497 4755 1426 4779 13255 36849 16783 6565 3494 14128 7897 13751 5314 2000 12265 4601 7731 21620 3822 4870 1426 5343 13223 36838 17136 6913 3494 13279 7897 14169 5186 2001 12063 4731 9107 21590 3822 4870 1511 5368 13399 36838 18400 7351 3719 13195 7926 16874 5186 2002 12239 4731 9107 21624 3822 4870 1511 5289 13399 37031 18400 7351 3719 13113 10936 16225 5484 Correlation with time trend: 0.94 0.94 0.58 0.91 0.47 0.96 0.88 0.75 0.91 0.95 0.80 0.92 0.90 0.76 0.64 0.83 0.87 Min 0.47 Max 0.96 (Sum of national and state highways; in Table 12.1. State-wise Highway Stock Over Time 84 Andhra Pradesh Assam Bihar Gujarat Haryana Himachal Pradesh Jammu & Kashmir Kerala Karnataka Maharashtra Madhya Pradesh Orissa Punjab Rajasthan Tamil Nadu Uttar Pradesh West Bengal 1973 670 197 604 954 504 51 82 625 967 1882 777 684 771 583 1654 1558 1333 1974 890 197 604 1173 557 53 94 625 967 2070 776 803 886 581 1654 1841 1279 1975 990 167 604 1361 617 52 102 755 1056 2480 895 863 999 581 1764 2087 1399 1976 1200 160 712 1595 617 53 101 1013 1056 2600 895 923 976 581 1764 2499 1385 1977 1520 162 890 1711 742 113 136 1012 1145 2822 1135 923 1246 684 1824 2868 1389 1978 1620 162 891 2216 978 113 171 1012 1145 3322 1318 923 1541 820 2119 3076 1511 1979 1930 162 890 2212 1081 114 206 1012 1335 3552 1528 923 1536 820 2329 3340 1611 1980 2240 228 941 2197 1141 129 206 1012 1470 3992 1630 923 1536 810 2329 3612 1726 1981 2240 333 940 2407 1141 127 206 1012 1740 4322 1631 1032 1586 810 2539 3751 1726 1982 2678 334 1051 2576 1213 128 206 1012 1875 4862 1841 1442 1704 1023 2539 3752 1996 1983 2830 349 1160 2786 1266 128 206 1012 2010 5572 2471 1134 1828 1167 2509 3916 2124 1984 3156 432 1374 3106 1311 134 205 1012 2220 5995 2681 1134 2315 1180 2509 4148 2394 1985 3366 432 1595 3316 1429 134 205 1309 2530 6884 2944 1235 2449 1230 2529 4355 2694 1986 3595 507 1595 3526 1553 135 207 1476 2530 7011 2997 1235 2459 1233 2795 4566 2619 1987 3614 507 1549 3863 1569 154 210 1477 2530 7272 3042 1394 2660 1212 3300 4887 2619 1988 3614 537 1549 3973 1787 274 235 1477 2530 7482 3087 1394 3048 1467 3642 5417 2619 1989 4064 537 1549 4220 1795 274 284 1477 2645 8208 3088 1574 3049 1722 3875 5527 2619 1990 4130 537 1549 4395 1780 272 262 1477 2970 8705 3298 1612 3049 1722 4089 5527 3071 1991 4224 537 1544 4737 1780 272 262 1477 2986 9207 3383 1612 3289 1732 4311 5179 3135 1992 4226 537 1548 4891 1780 272 262 1477 3052 9129 3533 1742 3499 1733 4315 5075 3315 1993 4727 577 1550 4938 1780 272 337 1484 3167 9339 3783 1742 3509 1943 4317 5575 3525 1994 5209 597 1765 4939 1780 274 362 1492 3377 9987 3864 1952 3509 1949 4737 6075 3550 1995 5210 597 1768 5329 1780 289 366 1492 3379 10000 3864 1692 3509 1985 5067 6069 3575 1996 5709 597 1983 5457 1780 300 366 1566 3385 10000 3873 1693 3509 1985 5723 6059 3595 1997 6208 617 1988 6520 1780 299 374 1771 3450 10500 3878 1693 3719 1985 5763 6169 3866 1998 6214 622 1988 6973 1780 299 374 1816 3973 11600 4094 1698 3929 2235 5988 6085 4131 1999 6255 622 1988 7223 1780 300 409 2118 4368 12400 4353 1993 3929 2487 6052 6053 4373 2000 6756 622 2108 7223 1990 326 417 2218 4465 12900 4373 2298 4529 2489 3521 6567 4783 2001 7238 622 2228 7352 1990 412 496 2239 4987 13200 4408 2298 4529 2999 3802 6600 4784 2002 7616 622 2228 7323 1990 612 496 2239 5197 13200 4500 2304 4533 3077 7146 6600 4784 Correlation with time trend: 0.99 0.95 0.98 0.99 0.94 0.91 0.97 0.95 0.98 1.00 0.98 0.96 0.99 0.97 0.90 0.97 0.99 Min 0.90 Max 1.00 (In megawatts) Table 12.2. State-wise Expansion in Electricity Generation Capacity Over Time 85 Table 13. Average of state-wise correlations1 TP Trend Own highways Contiguous highways Own electricity Contiguous electricity TP 1 Trend 0.89 1 Own highways 0.79 0.83 1 Contiguous highways 0.87 0.93 0.82 1 Own electricity 0.72 0.90 0.72 0.79 1 Contiguous electricity 0.77 0.95 0.75 0.82 0.93 1 1 Average of 15 states incuded in regression sample 86 Dependent variable: TP 1 2 3 4 5 6 7 8 9 Trend 0.0118 0.0104 0.0114 0.0103 0.0103 0.0128 0.0114 0.01 0.0115 (16.33)*** (15.96)*** (15.93)*** (16.11)*** (13.13)*** (15.61)*** (13.55)*** (14.70)*** (15.61)*** 0.0862 0.1129 0.0753 0.0908 (9.36)*** (9.11)*** (7.33)*** (7.68)*** 0.0234 -0.0249 0.0119 (3.78)*** (-3.17)*** 1.35 Own Electricity 0.0436 0.0416 0.0177 0.0048 (6.12)*** (6.05)*** (2.33)*** 0.63 Contiguous Electricity -0.0298 -0.0284 -0.0449 (-6.45)*** (-6.40)*** (-8.80)*** Constant 0.9228 1.203 0.9916 1.2166 1.3832 0.6147 1.0686 1.3541 0.8377 (63.79)*** (36.95)*** (42.93)*** (37.44)*** (18.06)*** (12.15)*** (11.94)*** (18.62)*** (8.60)*** Observations 510 510 510 510 510 510 510 510 510 Number of states 15 15 15 15 15 15 15 15 15 Lambda 0.44 0.42 0.45 0.41 0.49 0.52 0.53 0.42 0.50 t-stat (lambda) (8.60)*** (8.58)*** (8.64)*** (8.65)*** (9.43)*** (9.22)*** (9.75)*** (8.71)*** (9.73)*** R-squared 0.6484 0.70 0.66 0.70 0.67 0.67 0.70 0.70 0.75 Own National + State Highways Contiguous National + State Highways t statistics in parantheses; *** indicates significant at 1 percent, ** at 5 percent, and * at 10 percent level. error structure: u = lambda*Wu + e, W = weights matrix Table 14. Pooled Spatial Errors Model (SEM) 87 1 2 3 4 5 6 7 8 Trend 0.011 0.01 0.009 0.016 0.02 0.02 0.015 0.018 (25.38)*** (14.83)*** (14.22)*** (27.96)*** (28.22)*** (28.33)*** (23.81)*** (20.45)*** 0.038 0.012 0.047 0.031 (1.74)* (0.54) (2.30)** (1.57) 0.146 0.14 0.135 (4.01)*** (3.64)*** (4.02)*** Own Electricity -0.117 -0.044 -0.119 -0.051 (7.86)*** (2.77)*** (8.01)*** (3.27)*** Contiguous Electricity -0.221 -0.191 -0.188 (12.00)*** (9.03)*** (9.03)*** Constant 1.042 1.366 1.386 -0.316 -1.391 -1.549 -0.549 0.451 (14.58)*** (12.22)*** (11.75)*** (2.01)** (7.22)*** (4.43)*** (4.43)*** (4.43)*** Observations 510 510 510 510 510 510 510 510 Number of states 15 15 15 15 15 15 15 15 R-squared 0.78 0.79 0.79 0.8 0.83 0.83 0.83 0.83 Own National + State Highways Contiguous National + State Highways t statistics in parentheses; * significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent level. Table 15. Fixed Effects 88 1 2 3 4 5 6 7 8 0.106 0.15 0.092 0.131 (2.38)** (2.47)** (1.7) (2.02)* 0.023 -0.039 -0.02 (0.73) -1.06 -0.46 Own Electricity 0.052 0.06 0.019 0.026 (1.54) (1.72)* (0.51) (0.72) Contiguous Electricity -0.013 -0.025 -0.033 (-0.46) (-0.92) -1.13 Constant 1.445 1.189 1.471 1.641 0.997 1.478 1.588 1.403 (10.68)*** (13.33)*** (10.75)*** (4.89)*** (3.60)*** (3.88)*** (5.04)*** (3.50)*** Observations 510 510 510 510 510 510 510 510 Number of states 15 15 15 15 15 15 15 15 R-squared 0.3 0.04 0.36 0.15 0.02 0.21 0.32 0.46 Own National + State Highways Contiguous National + State Highways t statistics in parentheses; * significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent level. Table 16. Between Estimator 89 1 2 3 4 5 6 7 8 Trend 0.0123 0.0106 0.0113 0.012 0.0162 0.0169 0.0125 0.0161 (16.95)*** (13.14)*** (12.21)*** (12.21)*** (17.62)*** (15.92)*** (16.76)*** (13.94)*** -0.035 -0.0256 -0.0255 -0.0017 (-2.18)** (-1.53) (-1.44) (-0.10) 0.0799 0.0624 0.0762 (2.44)** (1.79)* (2.27)** Own Electricity -0.0078 -0.0179 -0.0102 -0.0233 (-0.58) (-1.45) (-0.74) (-1.86)* Contiguous Electricity -0.1272 -0.1323 -0.132 (-5.96)*** (-6.15)*** (-6.11)*** Constant 0.9318 1.3041 1.1725 0.9501 -0.3503 -0.5846 0.8558 -0.3771 (19.71)*** (11.47)*** (8.23)*** (7.03)*** (-1.51) (-2.05)** (5.93)*** (-1.25) Observations 510 510 510 510 510 510 510 510 Number of states 15 15 15 15 15 15 15 15 Lambda 0.70 0.71 0.71 0.61 0.63 0.64 0.58 0.62 t-stat (lambda) (24.71)*** (24.17)*** (24.10)*** (19.05)*** (18.70)*** (18.90)*** (18.03)*** (17.41)*** R-squared 0.91 0.92 0.92 0.90 0.92 0.92 0.90 0.92 all regression include state dummy variables for N-1 states Own National + State Highways Contiguous National + State Highways t statistics in parantheses; *** indicates significant at 1 percent, ** at 5 percent, and * at 10 percent level. error structure: u = lambda*Wu + e, W = weights matrix Table 17. Spatial Errors Model (SEM) - including state dummy variables 90 0.15 0.18 0.21 0.24 0.27 0.3 1993 1995 1997 1999 2001 2003 2005 Figure 1. Gini coefficient of income inequality across states over time1 (Based on per-capita gross state domestic product) Source: Planning Commission, Eleventh Plan document, Chapter 7 1 Solid markers indicate gap years. 91 -2 -1 0 1 2 3 4 19 74 -75 19 76 -77 19 78 -79 19 80 -81 19 82 -83 19 84 -85 19 86 -97 19 88 -89 19 90 -91 19 92 -93 19 94 -95 19 96 -97 19 98 -99 20 00 -01 20 02 -03 1 Manufacturing WPI based on official data from Central Statistical Organization (CSO) official series Figure 2. Divisia Output Price Index vs. Manufacturing WPI1 (Difference in percent) 92 0 2 4 6 8 10 12 Output Intermediates Labour Capital 1973-79 1980-89 1990-99 Figure 3. Growth Rate of Real Output and Inputs (In percent) 93 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2 1.25 1.3 -4 -3.5 -3 -2.5 -2 Av era ge T P l ev el (19 70 -03 ) Average (ln) own national and state highways per square km (1970-03) Figure 4. State-wise TP and National + State Highways (1970-03 average) 94 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2 1.25 1.3 -12 -11.5 -11 -10.5 -10 -9.5 -9 -8.5 -8 Av era ge T P l ev el (19 70 -03 ) Average (ln) own electricity generation capacity per capita (1970-03) Figuer 5. State-wise TP and Electricity Generation Capacity (1970-03 average) 95 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2 1.25 1.3 -4 -3.5 -3 -2.5 -2 -1.5 -1 Av era ge T P l ev el (19 70 -03 ) Average (ln) own + contiguous national and state highways per square km (1970-03) Figure 6: State-wise TP and National + State Highways (contiguous, 1970-03 average) 96 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2 1.25 1.3 -11 -10.5 -10 -9.5 -9 -8.5 -8 Av era ge T P l ev el (19 70 -03 ) Average (ln) own + contiguous electricity generation capacity per capita (1970- 03) Figure 7: State-wise TP and Electricity Generation Capacity (contiguous, 1970-03 average) 97 Appendix 2: ASI data definitions, construction of price and quantity indices 1. Data Definitions: Following are the terms and definitions employed by ASI in measuring state-level industrial aggregates. Note that all definitions are taken from ASI 1973-74 -- 2003-04 (Vol. II) published by the EPW Research Foundation. For all variables below, data for the year 1972-73 was not available, and was approximated by linear interpolation. Total Persons Engaged: was used as the measure of labour input. ASI defines Total Persons Engaged as all persons engaged by the factory whether for wages or not, in work connected directly or indirectly with the manufacturing process and include all administrative, technical, clerical staff as also labour engaged in production of capital assets for factory's own use. This is inclusive of persons holding supervisory or managerial positions or engaged in administrative office, store keeping section and welfare section, sales department as also those engaged in the purchase of raw materials etc. , and production of fixed assets for the factory and watch and ward staff. It also includes all working proprietors and their family members who are actively engaged in the work of the factory even without any pay and the unpaid members of the co-operative societies who worked in or for the factory in any direct and productive capacity. (Note that in summary reports published prior to 1979-80 total persons engaged was termed as employees. Lacking data on Total Persons Engaged before 1980, employee data was used for previous years). 98 Total Emoluments: measures payment to labour input. Emoluments include wages and salaries, and imputed value of benefits including those paid in kind. Total Inputs: is the current cost of fuels, electricity, materials, and services (such as freight and transport charges, communication costs, and insurance and banking costs) consumed in production. Capital Stock: series was built using the perpetual inventory method as in most studies. For this, ASI gross investment data at current prices was employed. For years prior to 1979, the gross investment series was not available, and was estimated using ASI depreciation and net fixed capital stock data57. Gross Value of Output: is the measure of output employed for estimating Total Productivity. This is the ex-factory value of ASI "product", which includes all goods except intermediates produced in the year, whether sold or not, and inclusive of fixed capital produced by the factory for its own use. 2. Electricity, Gas and Water Supply Sector: In order to exploit the entire available time series up to 2003 for all other sectors, aggregates for this sector were imputed from 1997 onwards, based on historical growth rates. The imputed aggregates were applied to both the All-India series, and the state-year series. Average growth from 1990-97 was chosen for extrapolation. It is arguable that a more recent average, such as 1993-97 57 Gross Investment(t) = Net Stock(t) + Depreciation(t) - Net Stock(t-1) 99 might have been appropriate, but the growth rate of employment, coming off a trough in the preceding five years, would likely over-estimate labour. The other growth rates from 1990-97 and 1993-97 appear comparable. Also, barring depreciation rates, the 1973-97 averages and 1990-97 averages seem comparable too. Secondly, the average share of the aggregates of this sector in each state's aggregates was estimated using state-industry data (available from 1979-1997), and the imputed values (additions to 1997 stocks) were split up between states based on this share. Imputed addition to capital stock was assigned according to EGW's average share in state net capital stock at book value. 3. Methodology for construction price deflators The approach to constructing price deflators in this study is to derive a set of weights for the aggregate output and aggregate intermediate input series respectively. These weights are then used to combine wholesale price indices corresponding to the ASI industries at the 2-digit level that were aggregated in this study to for state level (and all- India) aggregate series. Table A.1 lists the 25 ASI 2-digit industry groups that are included in our measure of aggregate gross output, the weight of the industry group in the output price index, as well as the name of the wholesale price index that most closely maps into the given industry group's output. Similarly, Table A.2 lists 21 ASI 2-digit industry groups that are included in our measure of aggregate intermediate inputs, the weights of each industry group in the input price index, and the corresponding wholesale price series that maps to the given 100 industry. The weights in both tables refer to years for which I was able to obtain I-O information for the aggregate Indian economy. The commodity-flow matrix (or absorption) matrix in the I-O tables details the deliveries of commodities to industries for intermediate consumption, as well as the deliveries of commodities to final demand in the aggregate economy. There are 6 categories of final demand: (i) Private Final Consumption Expenditure (ii) Government Final Consumption Expenditure (iii) Gross Fixed Capital Formation (iv) Change in Stocks (v) Exports and (vi) Imports. Strictly speaking, the commodity/industry table is not symmetric, in the sense that by -- products of given industry are treated as the principle output of the industry that produces this particular as the main product. Hence, row (commodity) totals are not equal to column (industry output) totals. This is not a problem for out price indices, as we are eventually interested in what prices industries receive for their output, whether it be their principle product or a by-product. An additional note on the I-O tables is that while for 1973-74 we only had a 60-sector classification, this expanded to 115 sectors in subsequent rounds. Not wanting to lose the earliest I-O information, I aggregated all the other years also up to the 60-sector aggregate, using the official aggregation provided in I-O tables. Also, although I have tried to create a price index that refers to only those industries covered in the ASI frame at the 2-digit level, the I-O information itself is based on economy-wide flows, not limited to the registered manufacturing sector. In other words, the weights derived from I-O information are also based on informal sector manufacturing. It is quite possible that the technology of the informal sector even within an industry group is quite different 101 from that of the formal sector, hence the weights are proxies at best. However, this concern is mitigated when one recalls that the formal sector accounts for bulk of the flows of intermediate consumption and gross output in the manufacturing sector as a whole (refer footnote 8 in text). Deriving weights for output prices: For each I-O year, I isolated the deliveries to final demand of the 25 ASI industries (commodities) covered in our sample. However, since commodity flows to intermediate and final demand also include imported commodities, these have to be netted out from the flows in order to price only domestic production. Using the import flow matrix (available in I-O tables from 1989-90 onwards, but not for previous rounds), it is possible to distinguish flows of imports to intermediate consumption and to final consumption (PFCE, GFCE, and GFCF) respectively58. Having isolated deliveries to final demand in this way, we then weight each industry according to its share in deliveries to final demand, by all 25 ASI industries. Hence, if ,Fiiiiii MXIGFCEPFCEY ?+++= where FiM is imports for final demand of 58 The import flow tables show no deliveries of imports to Change in Stock; all imports are exhausted between intermediate use and the three categories of final demand above. For years prior to 1989-90, a simple expedient is to calculate the fraction of imports-to- intermediate use over total imports from the closest I-O round for which data are available, and impose that fraction on total imports as given in the absorption matrix of the previous rounds as a best approximation. 102 commodity ,i then industry si? weight in the output price index iw ./ 25 1 iii YY = ?= Thus we obtain five sets of weights for industry output, one for each available I-O year. Table A. 1 shows how that the weights for output tend to be relatively stable over time. However, the weights for iron and steel, petroleum products, and miscellaneous manufacturing have gone up in a noticeable way. Deriving weights for input prices: For each I-O year, I isolated flows of the 21 input goods listed in Table A.2 to the 25 ASI industries in our sample. I excluded flows of these input commodities to all industries not covered in the ASI frame, so that the resulting input price deflator is as specific to ASI as possible. I did not in this case net out flows of imports to intermediate demand, as we are interested in prices paid by ASI industries for the consumption of intermediate goods irrespective of the source of origin. Then, the weight of commodity j in the input price index is simply /jj M=? , 21 1 jj M = ? where jM is the total delivery of commodity j to the 25 ASI industries: , 25 1 jiij mM = ?= and jim is the consumption of intermediate input j by industry .i This exercise shows that intermediate input structure has changed substantially over time. Table A.3 showed that the share of services inputs has increased quite substantially in the 30 year time frame. In addition we can see from Table A.2 that the share of primary articles has fallen dramatically, whereas the share of fuels, coal, chemicals, and 103 electricity has increased. This shows that there could be potentially severe pitfalls from using fixed weights for the input price index, if the growth rates of these commodities differ from each other. Using the weights derived in the above manners, aggregate output and input price indices could then be constructed by combining the change in individual industry price indices with the weights to derive a Divisia-type growth rate of the aggregate index: ],ln[lnln 1,, 25:1 Q ti Q tiii Q t PPwP ?= ??=? where Q tP is the aggregate output index and Q tiP, is the wholesale price index of commodity i. Similarly for intermediate inputs, we get ].ln[lnln 1,, 21:1 M tj M tjjj M t PPP ?= ??=? ? We set the value of each price index = ln(100) in the year 1993-94. We can then increment (decrement) this base value by the estimated change in log-levels as shown above. The exponent of this incremented (decremented) value yields the requisite price index. The index levels are given in Table A.6, along with the official manufacturing WPI series as published by the CSO. Note that for the purpose of estimating TPG and TP levels, we only need to know the growth rates of the aggregate price indices. 104 Appendix 3: Additional figures and tables 80 100 120 140 160 180 200 220 240 19 70 19 73 19 76 19 79 19 82 19 85 19 88 19 91 19 94 19 97 20 00 20 03 19 70 19 73 19 76 19 79 19 82 19 85 19 88 19 91 19 94 19 97 20 00 20 03 Double deflation 1/ Single deflation 1/ Divisia deflation (Rao) 2/ Divisia deflation (Hulten et. al.) 3/ Divisia deflation (current) Figure A.1 TFP and TP Indices Under Different Deflation Methods (1973=100) 1 Data from Balakrishnan and Pushpangadan (1994) 2 Data from Rao (1996) 3 Data from Hulten, Bennathan and Srinivasan (2006) Indices of TFP Indices of TP 105 No. ASI 2-digit industry Industry description 1973- 74 1983- 84 1989- 90 1993- 94 1998- 99 Wholesale price index series 1 20 Wood and wood products except furniture 0.003 0.004 0.001 0.004 0.004 Wood products 2 21 Furniture and fixtures 0.013 0.007 0.004 0.010 0.020 Wood products 3 22 Paper and paper products 0.003 0.004 0.003 0.009 0.009 Paper products 4 23 Printing, publishing, and allied activities 0.041 0.040 0.029 0.026 0.021 Paper products 5 25 Plastic and rubber products 0.041 0.048 0.043 0.052 0.046 Plastic and rubber products 6 26 Petroleum products 0.032 0.089 0.047 0.045 0.052 Mineral oils 7 28 Inorganic heavy chemicals 0.007 0.006 0.005 0.003 0.004 Chemicals and chemical products 8 29 Organic heavy chemicals 0.002 0.008 0.010 0.008 0.011 Chemicals and chemical products 9 30 Fertilizers 0.000 0.000 0.000 0.001 0.001 Fertilizers 10 31 Paints, varnishes and lacquers 0.004 0.005 0.004 0.005 0.006 Paints, varnishes, and lacquers 11 32 Other chemicals and chemical products 0.086 0.088 0.074 0.081 0.076 Chemicals and chemical products 12 33 Cement 0.000 0.000 0.000 0.001 0.001 Cement, lime, and plaster 13 34 Non metallic mineral products except cement 0.084 0.071 0.055 0.042 0.045 Non-metallic mineral products 14 35 Iron and steel industries and foundries 0.014 0.002 0.016 0.037 0.033 Iron and steel 15 36 Other basic metal industries 0.001 0.001 0.002 0.003 0.003 Basic metals and alloys 16 37 Metal products except machinery and transport equpiment 0.068 0.053 0.070 0.064 0.065 Metal products 17 38 Agricultural machinery 0.020 0.017 0.021 0.019 0.019 Non-electrical machinery and parts 18 39 Industrial machinery for food and textiles 0.024 0.020 0.015 0.010 0.010 Food and textile machinery 19 40 Other machinery 0.129 0.123 0.135 0.115 0.108 Electrical machinery 20 41 Electrical machinery, apparatus, and appliances 0.149 0.137 0.160 0.164 0.159 Electrical industrial machinery 21 42 Railway transport equipment 0.007 0.026 0.018 0.020 0.014 Transport equipment and parts 22 43 Other transport equipment 0.118 0.124 0.164 0.123 0.094 Transport equipment and parts 23 44 Miscellaneous manufacturing industries 0.129 0.104 0.081 0.110 0.145 Manufacturing wholesale price index 24 46 Electricity 0.014 0.003 0.023 0.032 0.036 Electricity 25 47 Gas and water supply 0.009 0.021 0.022 0.015 0.021 Electricity, gas and water GDP deflator1 Total 1.000 1.000 1.000 1.000 1.000 Input-Output table Table A.1. Output Weights in ASI Output Price Deflator 1From www.rbi.org.in 106 No. Commodity description 1973- 74 1983- 84 1989- 90 1993- 94 1998- 99 Wholesale price index series 1 Food (Primary Articles) 0.299 0.168 0.119 0.108 0.128 Food - Primary Articles 2 Eggs, Fish, Meat/Milk 0.031 0.026 0.034 0.024 0.026 Eggs, Fish, Meat 3 Logs and Timber 0.013 0.022 0.037 0.012 0.011 Forestry-GDP Deflator1 4 Coal 0.014 0.035 0.002 0.043 0.038 Mining - GDP Deflator1 5 Mineral Oils 0.067 0.138 0.104 0.104 0.091 Mineral Oils 6 Sugar/Edible Oil 0.037 0.035 0.011 0.012 0.025 Sugar 7 Beverages and Tobacco 0.005 0.004 0.001 0.001 0.002 Beverages and Tobacco 8 Cotton Textiles 0.069 0.085 0.065 0.052 0.039 Cotton Textiles 9 Wood and Wood products 0.014 0.007 0.000 0.008 0.010 Wood and Wood products 10 Paper and Paper products 0.023 0.024 0.029 0.023 0.022 Paper and Paper products 11 Leather and Leather products 0.008 0.005 0.006 0.006 0.005 Leather and leather products 12 Rubber and Plastic 0.014 0.014 0.013 0.018 0.015 Rubber and Plastic products 13 Chemicals and Chemical products 0.084 0.081 0.126 0.135 0.125 Chemicals and chemical product 14 Non-metallic mineral products 0.009 0.012 0.008 0.005 0.005 Non-metallic mineral products 15 Electricity 0.042 0.079 0.092 0.094 0.106 Electricity 16 Gas and Water Supply 0.001 0.002 0.005 0.002 0.003 Gas/Water Supply GDP Deflator1 17 Railway Transportation 0.022 0.024 0.029 0.026 0.023 Rail Transport GDP Deflator1 18 Other Transportation 0.030 0.028 0.042 0.088 0.059 Other Transport GDP Deflator1 19 Communication 0.000 0.004 0.011 0.009 0.008 Communication GDP Deflator1 20 Banking 0.025 0.024 0.038 0.035 0.062 Banking GDP Deflator1 21 Insurance 0.003 0.007 0.015 0.010 0.007 Insurance GDP Deflator1 Total 1.000 1.000 1.000 1.000 1.000 Input-Output table Table A.2. Input Weights in Intermediate Inputs Price Deflator 1From www.rbi.org.in 107 1973-1974 1983-1984 1989-1990 1993-1994 1998-1999 8.60% 9.80% 14.70% 17.50% 17% Table A.3. Weight of Services in Intermediate Inputs1 1Services include construction, transportation, communication, banking, insurance, gas and water supply. 108 Machinery and equipment Structures 1974-75 0.57 0.43 1975-76 0.60 0.40 1976-77 0.45 0.55 1977-78 0.48 0.52 1978-79 0.42 0.58 1979-80 0.51 0.49 1980-81 0.57 0.43 1981-82 0.55 0.45 1982-83 0.61 0.39 1983-84 0.60 0.40 1984-85 0.56 0.44 1985-86 0.62 0.38 1986-87 0.60 0.40 1987-88 0.58 0.42 1988-89 0.69 0.31 1989-90 0.66 0.34 1990-91 0.58 0.42 1991-92 0.70 0.30 1992-93 0.63 0.37 1993-94 0.72 0.28 1994-95 0.74 0.26 1995-96 0.63 0.37 1996-97 0.85 0.15 1997-98 0.70 0.30 Average 0.6 0.4 Table A.4. Type of Investment by ASI Industries1,2 1 Estimates based on National Accounts Statistics (NAS) data on All-India investment by public, private corporate, and household sector in these 2 categories, applied 2 Figures from 1970-1973, and 1998-2003 were set equal to average over 1974-1997 (Ratio) 109 Output Intermediate Labour Capital 1973 26.7 27.0 66.9 16.8 1974 26.6 28.3 70.2 16.1 1975 28.4 32.2 74.0 17.8 1976 32.5 36.0 76.9 17.6 1977 35.5 39.3 81.5 18.3 1978 39.1 42.6 83.2 20.1 1979 39.6 45.0 88.1 23.0 1980 40.4 46.5 89.6 26.1 1981 43.8 48.8 90.3 30.7 1982 49.0 53.2 92.9 34.6 1983 49.8 51.0 90.0 40.0 1984 53.7 55.7 89.2 44.5 1985 56.2 58.5 85.7 49.2 1986 58.3 61.2 84.8 52.9 1987 64.3 66.3 88.6 57.7 1988 71.0 73.5 89.2 61.9 1989 82.1 86.4 92.6 67.2 1990 89.6 92.9 93.4 74.2 1991 87.1 91.2 94.1 81.2 1992 95.2 99.3 100.2 90.1 1993 100.0 100.0 100.0 100.0 1994 113.5 110.1 103.6 115.5 1995 134.3 131.2 114.4 132.4 1996 135.1 126.8 111.9 147.8 1997 155.4 145.3 112.4 158.9 1998 161.9 145.3 108.3 174.5 1999 181.1 160.4 103.1 180.7 2000 174.8 156.3 100.7 186.1 2001 174.8 155.9 97.6 197.2 2002 202.2 177.2 100.1 199.6 2003 223.9 190.2 99.2 206.0 Table A.5. Index Numbers of Output and Inputs - All-India Aggregates1 1 Sum of 2-digit ASI data from 1973 onwards 110 (1993-94 = 100) Intermediate inputs1 Output (ASI industries)1 Output (overall manufacturing)2 1970-71 13.1 14.9 15.2 1971-72 13.7 15.8 16.6 1972-73 14.8 16.7 18.5 1973-74 17.5 18.5 21.2 1974-75 22.6 24.9 25.7 1975-76 23.2 26.5 26.0 1976-77 23.5 26.2 26.6 1977-78 24.9 27.4 27.2 1978-79 25.8 28.2 27.3 1979-80 29.2 33.0 32.8 1980-81 33.2 37.6 39.1 1981-82 38.2 41.4 41.1 1982-83 41.1 43.3 42.6 1983-84 44.2 45.4 45.2 1984-85 46.7 48.0 48.3 1985-86 50.9 52.3 51.2 1986-97 53.5 54.8 53.1 1987-88 56.9 57.6 57.0 1988-89 61.9 63.1 62.3 1989-90 66.1 68.2 69.3 1990-91 72.1 73.4 75.2 1991-92 81.1 83.0 83.6 1992-93 90.4 92.6 92.8 1993-94 100.0 100.0 100.0 1994-95 111.4 108.4 112.3 1995-96 120.8 118.0 122.0 1996-97 128.8 121.6 124.4 1997-98 137.4 125.2 128.1 1998-99 144.9 128.9 133.6 1999-00 152.6 132.1 137.2 2000-01 165.3 141.2 141.7 2001-02 171.8 145.9 144.3 2002-03 178.4 149.0 148.1 2003-04 188.7 153.1 156.5 Table A.6. Indices of Output and Intermediate Input Prices 1 Divisia index; author's estimate 2 Official manufacturing sector price index 111 Andhra Pradesh Assam Bihar Gujarat Haryana Himachal Pradesh Jammu & Kashmir Kerala Karnataka Maharashtra Madhya Pradesh Orissa Punjab Rajasthan Tamil Nadu Uttar Pradesh West Bengal 1970 0.990 0.952 0.996 1.097 1.040 1.036 0.756 1.022 0.992 1.140 0.987 0.853 1.054 0.928 1.021 0.970 1.025 1971 0.975 0.927 0.990 1.058 1.062 1.248 0.815 1.023 1.026 1.137 0.974 0.829 1.028 0.931 1.008 0.949 1.030 1972 1.034 0.999 1.015 1.100 1.056 0.960 0.886 0.992 1.018 1.161 1.035 0.869 1.046 0.958 1.049 0.981 1.065 1973 1.102 1.098 1.053 1.156 1.072 0.841 0.949 0.996 1.026 1.201 1.102 0.909 1.091 0.993 1.103 1.027 1.114 1974 1.005 1.126 1.052 1.092 1.034 0.726 0.950 0.950 0.949 1.132 1.034 0.818 1.039 0.926 1.058 0.963 1.062 1975 0.995 0.977 0.969 1.013 1.008 0.917 0.817 0.902 0.917 1.076 0.957 0.747 1.015 0.893 0.990 0.988 1.028 1976 1.014 1.062 0.937 1.060 1.058 1.000 0.816 0.933 0.949 1.106 1.015 0.859 1.048 0.930 1.048 1.029 1.059 1977 1.011 1.118 0.925 1.083 1.080 1.058 0.880 0.963 0.955 1.120 0.996 0.844 1.059 0.928 1.066 0.935 1.060 1978 1.022 1.063 0.960 1.074 1.104 1.019 0.982 0.983 1.045 1.145 0.989 0.865 1.088 0.966 1.090 0.956 1.086 1979 0.973 0.993 0.913 1.029 1.034 1.047 0.992 0.964 0.942 1.082 0.966 0.838 1.063 0.941 1.033 0.915 1.033 1980 0.957 0.911 0.866 1.016 1.020 0.975 0.922 0.961 0.911 1.063 0.966 0.785 1.059 0.898 1.025 0.883 1.032 1981 0.991 0.948 0.977 1.047 1.082 1.051 0.982 0.986 0.938 1.092 1.006 0.790 1.081 0.918 1.053 1.041 1.050 1982 1.066 0.987 1.020 1.079 1.114 1.753 1.054 1.016 0.974 1.113 1.038 0.795 1.097 0.922 1.095 1.043 1.080 1983 1.139 1.113 1.120 1.164 1.188 1.635 1.069 1.044 1.054 1.169 1.056 0.850 1.127 1.034 1.126 1.030 1.100 1984 1.140 1.187 1.031 1.130 1.107 1.278 1.128 1.097 1.010 1.172 1.035 0.800 1.125 0.977 1.149 1.027 1.106 1985 1.096 1.208 1.031 1.123 1.105 1.246 1.082 1.069 1.011 1.199 1.075 0.839 1.140 0.982 1.133 1.017 1.110 1986 1.093 1.175 1.027 1.137 1.114 1.199 1.059 1.068 1.021 1.186 1.017 0.868 1.136 1.025 1.138 1.103 1.112 1987 1.076 1.157 1.095 1.159 1.127 1.078 1.105 1.113 1.033 1.190 1.085 0.859 1.172 0.998 1.147 1.128 1.175 1988 1.110 1.090 1.149 1.159 1.140 1.223 1.004 1.103 1.041 1.221 1.086 0.964 1.147 1.008 1.169 1.125 1.110 1989 1.093 1.208 1.126 1.137 1.131 1.096 1.047 1.159 1.073 1.229 1.103 0.978 1.235 1.018 1.180 1.171 1.095 1990 1.144 1.179 1.119 1.160 1.184 1.225 1.110 1.095 1.117 1.256 1.146 0.961 1.196 1.069 1.201 1.177 1.147 1991 1.121 1.137 1.123 1.110 1.158 1.174 1.080 1.135 1.127 1.187 1.082 0.936 1.182 1.041 1.166 1.186 1.129 1992 1.136 1.122 1.107 1.217 1.120 1.126 1.101 1.107 1.129 1.240 1.126 0.927 1.216 1.068 1.163 1.165 1.122 1993 1.159 1.117 1.307 1.236 1.160 1.317 1.278 1.100 1.113 1.296 1.169 0.928 1.219 1.109 1.203 1.195 1.167 1994 1.239 1.148 1.125 1.295 1.233 1.245 1.123 1.139 1.186 1.323 1.188 0.956 1.266 1.188 1.221 1.246 1.175 1995 1.285 1.171 1.163 1.310 1.280 1.231 1.165 1.188 1.168 1.350 1.264 0.978 1.244 1.189 1.220 1.221 1.181 1996 1.251 1.117 1.265 1.316 1.329 1.318 1.169 1.201 1.234 1.333 1.231 0.937 1.318 1.174 1.243 1.283 1.266 1997 1.363 1.176 1.417 1.272 1.331 1.336 1.216 1.238 1.222 1.383 1.281 1.081 1.332 1.287 1.255 1.317 1.342 1998 1.300 1.248 1.380 1.335 1.344 1.335 1.278 1.326 1.219 1.374 1.280 0.974 1.383 1.218 1.278 1.267 1.277 1999 1.347 1.296 1.422 1.367 1.418 1.410 1.358 1.315 1.199 1.443 1.331 1.048 1.444 1.342 1.325 1.297 1.308 2000 1.346 1.191 1.250 1.344 1.388 1.446 1.329 1.316 1.198 1.415 1.343 1.033 1.388 1.324 1.342 1.300 1.318 2001 1.384 1.139 1.239 1.327 1.420 1.465 1.344 1.312 1.240 1.406 1.339 1.022 1.439 1.312 1.319 1.312 1.355 2002 1.416 1.423 1.420 1.399 1.460 1.529 1.380 1.354 1.300 1.464 1.371 1.081 1.460 1.325 1.341 1.363 1.408 2003 1.489 1.500 1.500 1.465 1.485 1.581 1.407 1.396 1.364 1.535 1.447 1.167 1.488 1.370 1.410 1.410 1.469 Table A.7. State-wise Time-Series Estimates of Total Productivity Levels 112 Single deflation1 Double deflation1 Divisia deflator2 Divisia deflator3 Current 1970 ? ? ? ? ? 1971 -4.3 -21.0 ? ? ? 1972 1.1 -2.8 ? ? ? 1973 -7.1 17.2 ? ? ? 1974 3.9 17.2 16.8 5.8 -3.6 1975 -3.4 -8.2 -14.6 -4.5 -5.1 1976 3.8 12.3 19.2 6.6 4.9 1977 1.3 8.6 10.9 6.5 1.1 1978 10.7 16.3 18.9 7.0 1.8 1979 -10.8 -4.2 -4.1 -7.7 -5.1 1980 -11.3 -11.0 -13.9 -4.4 -2.3 1981 6.4 15.6 24.0 9.2 2.5 1982 9.2 14.5 22.3 9.1 2.7 1983 7.0 -4.4 -2.4 -6.6 3.4 1984 -2.2 -1.1 2.0 0.4 -0.4 1985 2.6 -4.4 -14.2 4.7 -0.4 1986 -4.2 -7.6 -9.1 3.4 -0.5 1987 0.8 -1.0 4.7 1.5 1.9 1988 6.9 4.1 5.7 7.2 0.9 1989 ? ? -6.4 5.3 0.4 1990 ? ? 6.5 -1.1 1.6 1991 ? ? -6.0 -11.4 -2.6 1992 ? ? -19.5 5.9 0.3 1993 ? ? ? ? 2.8 1994 ? ? ? ? 2.7 1995 ? ? ? ? 0.5 1996 ? ? ? ? 1.5 1997 ? ? ? ? 2.3 1998 ? ? ? ? 2.9 1999 ? ? ? ? 3.2 2000 ? ? ? ? -1.7 2001 ? ? ? ? -0.4 2002 ? ? ? ? 4.0 2003 ? ? ? ? 4.2 Table A.8. TFP and TP Growth Rates Under Different Deflation Methods 1 Based on Balakrishnan and Pushpangadan (EPW 1994) 2 Based on Rao (EPW 1996) 3 Based on Hulten, Bennathan and Srinivasan (WBER 2006) 113 Bibliography 1) Ahluwalia, I (1991), Productivity and Growth in Indian Manufacturing, New Delhi, Oxford University Press 2) Balakrishnan, P. and K. Pushpangadan (1995), ?Total Factor Productivity Growth in Manufacturing Industry,? Economic and Political Weekly, March 4, 1995 3) Balakrishnan, P. and K. Pushpangadan (1994), ?Total Factor Productivity Growth in Manufacturing Industry: A Fresh Look,? Economic and Political Weekly, July 30, 1994 4) Bosworth, B. and Susan Collins (2007), ?Accounting for Growth: Comparing China and India,? NBER working paper 12943 5) Bosworth, B., Susan Collins and Arvind Virmani (2007), ?Sources of Growth in the Indian Economy,? NBER working paper 12901 6) Caves, D., L. Christensen and E. Diewert (1982a), ?Multilateral Comparisons of Output, Input, and Productivity Using Superlative Index Numbers,? The Economic Journal 92(365), 73-86 7) Caves, D., L. Christensen and E. Diewert (1982b), ?The Theory of Index Numbers and Measurement of Input, Output, and Productivity,? Econometrica 50(6), 1393- 1414 8) Dholakia, B. and R. Dholakia (1994), ?Total Factor Productivity Growth in Indian Manufacturing,? Economic and Political Weekly, December 31, 1994 9) Dholakia, B. and R. Dholakia (1995), ?Total Factor Productivity Growth in Indian Industry,? Economic and Political Weekly, July 15, 1995 10) Fikkert, B. and Rana Hassan (1998), ?Returns to scale in a highly regulated economy: evidence from Indian firms,? Journal of Development Economics 56, 51-79 114 11) Goldar, B (2004), ?Productivity Trends in Indian Manufacturing in the Pre-and Post- Reform Period,? ICRIER Working Paper 137 12) Goldar, B and Rashmi Banga (2007), ?Contribution of Services to Output Growth and Productivity in Indian Manufacturing -- Pre and Post Reforms,? Economic and Political Weekly, June 30, 2007 13) Gramlich, E. (1994), ?Infrastructure Investment: A Review Essay,? Journal of Economic Literature 32(3), 1176-1196 14) Gupta, P, Hasan, R., and Utsav Kumar (2008), ?Big Reforms but Small Payoffs: Explaining the Weak Record of Growth in Indian Manufacturing,? NCAER New Delhi 2008. 15) Hall, R. and Charles Jones, ?Why Do Some Countries Produce So Much More Output Than Other Countries,? Quarterly Journal of Economics 114, 83-116 16) Hirschman, A.O., (1958), The Strategy of Economic Development, New Haven, Connecticut: Yale University Press. 17) Hulten, C. (2009), ?Growth Accounting,? NBER Working Paper 15341 18) Hulten, C. and Anders Isaksson (2007), ?Why Development Levels Differ: The Sources of Differential Economic Growth In A Panel of High and Low Income Countries,? NBER Working Paper 13469 19) Hulten, C., E. Bennathan, and S. Srinivasan (2005), Infrastructure?, Externalities, and Economic Development: A study of Indian Manufacturing Industry,? World Bank Economic Report, 2005 20) Hulten, C. and R. Schwab (1991), ?Public Capital Formation and the Growth of Regional Manufacturing Industries,? National Tax Journal LXIV (4), 121-134 21) Hulten, C. and R. Schwab (1984), ?Regional Productivity Growth in US Manufacturing,? American Economic Review 74(1), 152-162 115 22) Holtz-Eakin, D., and Amy Schwartz (1995), ?Spatial Productivity Spillovers from Public Infrastructure: Evidence from State Highways,? NBER working paper 5004 23) Jorgensen, D. and M. Nishimizu (1978), ?U.S. and Japanese Economic Growth, 1952 -- 1974: An International Comparison,? Economic Journal 88(352), 707-726 24) Klenow, P. and Andres Rodriguez-Clare (1997), ?The Neoclassical Revival in Growth Economics: Has It Gone Too Far??, NBER Macroeconomics Annual 1997, Volume 12, pages 73-114 25) Mankiw, G., David Romer, and David Weil (1992), ?A Contribution to the Empirics of Economic Growth,? Quarterly Journal of Economics 107(2), 407-437 26) Mitra, A., A. Varoudakis and M. Venganzones (1998), ?State Infrastructure and Productive Performance of Indian Manufacturing,? OECD working paper 139. 27) Munnell, A. (1990a), ?Why Has Productivity Declined,? NEER, Federal Reserve of Boston, January/February 1990, 3-22 28) Munnel, A. with the assistance of Leah M. Cook (1990b), ?How Does Public Infrastructure Affect Regional Economic Performance,? NEER, Federal Reserve of Boston, September/October 1990, 11-32 29) Munnel, A. (1992), ?Policy Watch: Infrastructure Investment and Economic Growth,? Journal of Economic Perspectives 6(4), 189-198 30) Pereira, A. and O. Roca-Sagales (2002), ?Spillover Effects of Spanish Capital Formation: Evidence from Spanish Regions,? College of William and Mary, Department of Economics working paper 31) Rao, J. Mohan (1996a), ?Manufacturing Productivity Growth -- Method and Measurement,? Economic and Political Weekly, November 2, 1996 32) Reddy, P., Prasad, A. and Sampath Kumar (2009), ?Balanced Regional Development 116 of India through Special Economic Zones: An Empirical Study,? Journal of Social Science 20(1), 1-13 33) Unel, B. (2003), ?Productivity Trends in India's Manufacturing Sector in the Last Two Decades,? IMF working paper WP/03/22 34) Ray, S.N. (2004). ?Unorganized vis-?-vis Organized Manufacturing Sector in India,? Ministry of Statistics and Programme Implementation Seminar Series, November 2004, http://mospi.gov.in/mospi_seminarseries_nov04_3_6_final.pdf 35) Rodrik, D. and A. Subramanian (2004), ?From Hindu Growth to Productivity Surge: The Mystery of the Indian Growth Transition,? IMF working paper WP/04/77