What Constrains Indian Manufacturing?

India has undertaken extensive reforms in its manufacturing sector in the last two decades. However, an acceleration of growth in manufacturing, and a concomitant increase in employment, has eluded India. What might be holding the sector back? Using Annual Survey of Industries data at the three-digit level and difference in estimates this paper finds that the post-reform performance of the manufacturing sector is heterogeneous across industries. In particular, industries dependent on infrastructure or external finance, and labour-intensive industries have not been able to reap the maximum benefits of reforms. The results point to the importance of infrastructure development and financial sector development for the manufacturing sector's growth to accelerate further. They also emphasize the need to clearly identify and address the factors inhibiting the growth of labour-intensive industries.


Foreword
Observers have widely acknowledged and commented upon the fact that the Indian growth story is mainly about services and not manufacturing. It is indeed an anomaly of the Indian reform process that while the reforms have been primarily focused on propping up the manufacturing sector, the growth has accelerated primarily in the services sector.
Something seems to be holding back Indian manufacturing. What exactly is it? This paper relates the pattern of growth in India's manufacturing sector to cross-industry heterogeneity in an industry's dependence on infrastructure, financial sector, and in the use of labour relative to capital. Results in the paper show that the potential benefits from product market reforms might not be realized unless these are matched by enabling conditions, such as high quality infrastructure, availability of the right kind of labour, and financial markets that are deep enough. While some of the reforms are likely to be politically sensitive and consensus building may take time, others such as the provision of infrastructure-electricity, roads, ports and airports, are crucial to ensuring inclusive growth, and all efforts must be made to ensure provision of quality infrastructure.  1970 1974 1978 1982 1986 1990 1994 1998 2002 2006 Services Agriculture Manufacturing Section I: Introduction 1 Many emerging countries in recent decades have relied on a development strategy focused on promoting the manufacturing sector and the export of manufactured goods. These include many East Asian countries and most recently, the People's Republic of China. India, too, hoped for a dynamic manufacturing sector when it introduced substantial product market reforms in its manufacturing sector starting in the mid-1980s. But the sector never took off as it did in other countries. India no doubt has grown impressively in the last 15 years; but the main contribution to growth has come from the services sector rather than from the manufacturing sector. Moreover, in so far as subsectors within manufacturing have performed well, these have been the relatively capitalor skill-intensive industries, not the labour-intensive ones as would be expected for a labour-abundant country like India. What could be the reasons behind the rather lackluster performance of the manufacturing sector in India?
As Chart 1a shows, the manufacturing sector's share in GDP has been stagnant since the early 1990s despite several wide ranging reforms in this sector. Similarly, Chart 1b shows that the contribution to GDP growth has primarily been from the services sector and this contribution has been increasing overtime. Contribution to GDP Growth (%) 1951-60 1980-90 1990-00 2000-07

Chart 1 b : Sectoral Contribution to Indian GDP Growth, 1951-2007
Source: Authors ' calculations, Central Statistical Organization (India) Several hypotheses have been put forward to explain the lack of dynamism in India's manufacturing sector. Infrastructure-related bottlenecks are widely believed to be a part of the explanation. In particular, poor quality of power supply, road networks, ports and airports are believed to create significant disadvantages for Indian manufacturers by pushing up their costs of production, and making them uncompetitive in export markets. 2 Besides infrastructure, some key areas of policies remain unchanged. In particular, even though there have been extensive product market reforms, it has been widely observed that the labour market reforms to complement these have not been undertaken (see Panagariya 2006, andPanagariya 2008;Kochhar et al 2006). Moreover, credit constraints due to weaknesses in the financial sector may be holding back small and medium sized firms from expanding (see Nagaraj 2005;McKinsey 2006). 3 Finally, business regulations might have influenced key decisions of firms and potential investors. As the World Bank's Doing Business surveys of business regulations across the 3 world have found, the procedures and costs for starting and, especially, closing a manufacturing business in India are among the most cumbersome in the world. 4 It would be useful to empirically test the hypotheses related to the idea that various elements of the policy and institutional environment facing the manufacturing sector, either left untouched by the liberalizations of the 1990s or dealt with only partially, have emerged as significant bottlenecks to growth and employment generation.
One obvious way in which one can test for these hypotheses is to exploit the inter-state heterogeneity in the policy and institutional environment, including labour market regulations; financial sector development; and infrastructure for different states of India and then test whether the industrial performance has been better in the states with better policy and institutional framework. This is precisely what has been done in the existing literature to show the importance of labour market flexibilities in explaining the gains from product market liberalizations. Besley and Burgess (2004), for example, exploit state-level amendments to the Industrial Disputes Act (IDA) -arguably the most important set of labour regulations governing Indian industry -over 1958-1992, and code legislative changes across major states as pro-worker, neutral, or pro-employer. These legislative amendments are then used in the regression analyses of various outcomes in the manufacturing sector, including output, employment, investment and the number of factories. Besley and Burgess find that pro-worker labour regulations have had a negative impact on output, employment, and investment in organized manufacturing. 5 A related paper by Aghion et al. (2006) relates various dimensions of industrial performance to the extent to which an industry was covered by industrial licensing requirements, and state-level measures of the stance of labour regulations. They find that the effects of industrial delicensing were unequal across Indian states. In particular, delicensed industries located in states with pro-employer labour regulations grew faster in terms of both output and employment levels than those with pro-worker regulations. Similarly, Mitra and Ural (2007) show that industries experiencing larger tariff reductions grew faster in pro-employer states relative to pro-worker states. 6 4 Another possibility is that hysteresis in the pattern of development in Indian manufacturing implies that the relative profitability of capital-and skill-intensive activities remains higher than that of labourintensive activities despite the reforms of the early 1990s (Kochhar et al. 2006). Other factors often believed to be affecting the performance of Indian manufacturing are public ownership of enterprises, remaining small scale industries reservations, and stringent regulations on land use in India as discussed in Panagariya (2008). In recent years the availability of skilled labour has also emerged as a constraint on the growth of manufacturing and services. 5 While, in principle, the approach of Besley and Burgess (2003) has considerable merit, it is not without controversy. Bhattacharjee (2006), in particular, has argued that deciding whether an individual amendment to the IDA is pro-employer or pro-worker in an objective manner is quite difficult. Even if individual amendments can be so coded, the actual workings of the regulations can hinge on judicial interpretations of the amendments. Moreover, if noncompliance with the regulations is widespread, then even an accurate coding of amendments which takes into account the appropriate judicial interpretation loses its meaning. 6 There are some differences between Mitra and Ural and Aghion et al in terms of the states deemed to have pro-employer or pro-worker labor regulations. See Mitra and Ural (2007) for details.
In this paper we relate the pattern of growth in India's manufacturing sector to crossindustry heterogeneity in the reliance on infrastructure and financial sectors and in the use of labour relative to capital. In particular, we calculate the dependence of industries on infrastructure, on the financial sector, and the labour intensity of industries. Using Annual Survey of Industries (ASI) data at the three-digit level we employ difference in estimates to compare the performance of industries more dependent on infrastructure, on the financial sector and labour-intensive industries post-delicensing with that of the control group.
Our results indicate that the aggregate performance of the manufacturing sector masks important inter-industry differences. Quite interestingly, we find that the industries with greater need for infrastructure; greater dependence on the financial sector; and greater labour intensity have performed relatively worse in the post-delicensing period. Quantitatively, the results indicate that in the post-delicensing period the above median, infrastructure-intensive industries grew 10 per cent less than the industries below the median of infrastructure dependence. Similarly, industries above median in the distribution of financial dependence grew 18 per cent less than the industries below the median of financial dependence; and for labour intensity, industries with above median labour intensity grew 19 per cent less than the below median industries post-delicensing.
There are two ways in which one can interpret our results. First, one can use our results to identify which industries have not benefited much from reforms. Second, to the extent that the heterogeneity across industries on parameters such as infrastructure dependence is exogenous and determined by factors such as production technology, we can probably draw causal inferences as well. Thus, for example, we can claim that if industries dependent on infrastructure have not benefited as much from reforms it is because of the unavailability of adequate infrastructure; and similarly for industries dependent on the financial sector. For labour-intensive industries, an interpretation in terms of the limited supplies of labour would not be appropriate in the Indian context. A more natural interpretation would be to relate the relatively weak performance of labour-intensive industries to the quality of labour, skill mismatch and regulations on employment, which make the effective price of hiring labour too high.
In order to ensure that the results are not driven by outliers, the standard errors are corrected for heteroskedasticity and autocorrelation; and the estimates are not biased due to omitted variables, we conduct extensive robustness tests, and find our results to be robust to these sensitivity analyses.
There are two conclusions that one can draw from these results. First, product market reforms alone might not be sufficient to spur growth; for gains from these reforms to be maximized they may have to be complemented by reforms in other areas. Second, the potential benefits from product market reforms might not be realized unless these are matched by enabling conditions, such as high quality infrastructure, availability of the right kind of labour, and financial markets that are deep enough. There is no room for 5 complacency and efforts on a war footing should be made to remove these constraints if the manufacturing sector is to play the role that it did in the case of East Asian countries.
One point that needs to be noted about our analysis is that it is based on data only on organized (or registered) manufacturing sector and not the unorganized sector. This is primarily because of the unavailability of detailed data for the latter. A question that comes up then is whether the lack of quality data on unregistered manufacturing should preclude any analysis of the registered manufacturing sector. Though there is no denying the fact that the unorganized manufacturing sector is important when it comes to employment, its output, wages, and productivity are very low. In terms of policy objectives, improving the performance of registered manufacturing is a key aspect to making India a powerhouse in manufacturing.
The rest of the paper is organized as follows. In the next section, we summarize the Indian policy framework and lay out the stylized facts related to the performance of the Indian registered manufacturing sector. In Section 3, we summarize evidence from two different firm-level survey data, whereas Section 4 specifies the main econometric exercise, and presents and discusses our results. The last section concludes.

Section II: Stylized Facts and Preliminary Evidence II A: Indian Policy Framework
Since the early 1950s up until the early 1980s the evolution of India's manufacturing sector was guided by industrial and trade policies that protected domestic industry and gave the state a central role in investment decisions. While a strict regime of import and export controls defined trade policy, industrial policy worked through an elabourate system of industrial licensing. Under the Industries Development and Regulatory Act of 1951, every investor over a very small size needed to obtain a licence before establishing an industrial plant, adding a new product line to an existing plant, substantially expanding output, or changing a plant's location.
While the state-led import substitution policy framework had helped create a diversified manufacturing sector, industrial stagnation since the mid-1960s -increasingly blamed on the policy framework -led to some tentative steps aimed at liberalizing these regimes in the late 1970s and early 1980s (see Ahluwalia 1987Ahluwalia , 1991. Relaxations of the industrial licensing system were introduced and import licensing requirements were eased. However, by most accounts these reforms were marginal. Tariff rates as high as 400 per cent were not uncommon, non-tariff barriers remained widespread, and the industrial licensing regime continued to impose binding constraints to entry and growth for most firms.  1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 A verage T ariff (% ) More serious liberalization efforts began in 1985 with delicensing-the exemption from the requirement of obtaining an industrial license-of 25 broad categories of industries, which maps into 13 industries in our three-digit level data. The next major reform of the licensing regime came in 1991 when industrial licensing was abolished except in the case of a small number of industries (see Chart 2a and Table A4 for the time path of delicensing). This was also the year in which a decisive break was made with the trade policies of the past. The liberalization of 1991 included the removal of most licensing and other non-tariff barriers on the imports of intermediate and capital goods, the simplification of the trade regime, devaluations of the Indian rupee and the introduction of an explicit dual exchange market in 1992 (see Chart 2b).

Chart 2 a : Cumulative Share of Industries Delicensed
Source: Based on Aghion et al (2006) and extended by the Authors

Chart 2 b: Average Nominal Rate of Protection, 1988 to 1998
Source: Hasan, Mitra and Ramaswamy (2007) Despite these impressive reform measures there are certain areas in which there has been little progress. One area in which there has been no major policy change is in the labour regulations that apply to India's industry sector. According to Panagariya (2008), it is rigidities introduced by these (unchanged) regulations that are holding back the manufacturing sector in general and its labour-intensive sub-sectors in particular. Since the issue of India's labour regulations is one of the most contentious ones in the context of debates on economic reforms, some details are in order.
While India's labour regulations have been criticized on many accounts including, for example, the sheer size and scope of regulations, their complexity, and inconsistencies across individual pieces of regulation, a few specific pieces of legislation are the controversial ones. First, as per Chapter VB of the Industrial Disputes Act (IDA) it is necessary for firms employing more than 100 workers to obtain the permission of state governments in order to retrench or lay off workers. 7 While the IDA does not prohibit retrenchments, critics of the act argue that it is difficult to carry them out. Datta-Chaudhuri (1996) has argued, for example, that states have often been unwilling to grant permission to retrench.
Second, additional rigidities in using effectively a firm's existing workers are believed to stem from Section 9A of the IDA and the Industrial Employment (Standing Orders) Act, which pertain to procedures that must be followed by employers before changing the terms and conditions of work. While the two pieces of legislation seek to make labour contracts complete, fair, and legally binding they can constrain firms from making quick adjustments to changing conditions. In particular, worker consent is required in order to modify job descriptions or move workers from one plant to another in response to changing market conditions. In and of itself, this does not seem to be an unreasonable objective. The problem, according to some analysts, is that the workings of India's Trade Union Act (TUA) make it difficult to obtain worker consent. While the TUA allows any seven workers in an enterprise to form and register a trade union, it has no provisions for union recognition (for example, via a secret ballot). The result, according to Anant (2000), has been multiple and rival unions, making it difficult to arrive at a consensus among workers.
Similarly, hiring contract workers can enable firms to get around many of the regulatory restrictions on adjusting employment levels, productions tasks, etc.; however, it is argued that Section 10 of the Contract Labour Act, which empowers the Government to prohibit the employment of contract labour in any industry, operation, or process, limits this course of action. It is important to note that not all analysts agree that India's labour laws have made for a rigid labour market. In particular, a counter-argument to the views above is that the rigidity inducing regulations have been either ignored (see Nagaraj (2002)) or circumvented through the increased usage of temporary or contract labour (see Datta (2003) and Ramaswamy (2003)). 8 Ultimately, whether India's labour laws have created significant rigidities in labour markets or not is an empirical issue, as is the broader question of whether and to what extent various policies have been the main constraints on the growth of Indian manufacturing.
Another well-known constraint on growth is India's crumbling infrastructure. According to the Deputy Chairman of the Planning Commission, Mr. Montek Singh Ahluwalia, India needs to increase its investment in infrastructure from 5 per cent of GDP to 8 per cent of the GDP by the end of the Eleventh Five-Year Plan, yielding an investment of USD 400 billion in its infrastructure to sustain the current growth rates. One does not need any scientific evidence to show that infrastructure in India needs to be improved, as the erratic and costly electricity supply, congested roads, ports and airports are for all to witness. A recent OECD survey of Indian economy report compares Indian infrastructure with that of other countries and finds India to be badly lagging in most of the areas.
An interesting comparison in this regard is with the infrastructure in China. Total investment anticipated in infrastructure in the Tenth Five-Year Plan (2002)(2003)(2004)(2005)(2006)  Another area in which there has been rather slow progress on reforms is the financial sector (or the banking sector, more narrowly). Reform efforts in this area have been directed at deregulating interest rates; some dilution of public ownership of banks; and limited opening up of the sector to private domestic and foreign banks. However, as pointed out often, and most recently in the OECD Economic Survey on India (2007), some major challenges still remain. These include a very high share of public ownership in banks and low level of bank intermediation partly because of regulations on the allocation of credit, which require banks to allocate a substantial percentage of their total advances into government securities and other priority sectors. 9

II B: Performance of Indian Manufacturing
We look at a fairly long time series of data on Indian registered manufacturing from 1973-2003. 10 Below, we summarize some of the empirical regularities that we observe in the data on the various indicators of industrial performance and on employment-related variables. 11 Various panels of Chart 4 show that the growth of value added, employment, capital formation and factories has been stable throughout the last three decades and has not necessarily accelerated in the post-reform period. If anything, there is probably a stagnation starting sometime in the mid to late 1990s.
Panel B of the chart shows separately the employment of blue-collar workers and total employment. The trends seem to be broadly similar for both the variables. The data on contractual labour, is available only since 1998, but the trends show an increase in the share of contractual labour in total employment. The pace of growth of capital stock seems to be faster than that of employment. These different trends in employment and investment are probably reflected in the growth of labour productivity over time. Finally, the number of factories does not seem to have kept pace with the growth of value added. 9 In addition since the performance of the bank managers is not linked as tightly with the profitability of the banks, and is probably influenced more by the incidence of non-performing loans, they have little incentives to provide credit to the private sector. Hence they play extremely cautious and rather than lending to the private sector would rather invest in safe government securities (see . 10 Reference period for ASI is the accounting year of the industrial unit ending on any day during the fiscal year. Therefore, in ASI 2003-04, the data collected from the respective industrial units relates to their accounting year ending on any day between 1st April, 2003, and 31 March, 2004. When we say 2003 in the paper it refers to the fiscal year 2003-04. 11 The only comprehensive database available on Indian manufacturing is the ASI data which includes data on registered manufacturing, i.e. factories with more than 20 workers if not using power and factories employing more than 10 workers if using power. One caveat of using this data is that we are only looking at a fraction of total manufacturing. Registered manufacturing comprises 70 per cent of the total output being produced in the manufacturing sector but only 20 per cent of the total manufacturing employment.

Chart 2: Performance of Indian Manufacturing
The trends in these charts are also picked up in the table below. In Table 1, we estimate the trend growth rates for the aggregate values of various performance indicators pertaining to the manufacturing sector. The regression equations include the dependent variables in logs, and regress it on a linear trend and a dummy, which take the value 1 for  19731977198119851989199319972001 Panel F: Labor Productivity the post-1992 period, and zero otherwise. Thus, its coefficient measures the percentage change in the dependent variable post-1992 after accounting for its trend growth rate.
The results indicate a marginal pickup in the growth rate of value added post-1992; and in the rate of investment, but no significant improvement in the number of factories operating or in employment. 12 Thus, the data show that the aggregate value added has increased at about 6 per cent a year in the sample period, and there has been a modest annual growth acceleration of about ½ percentage point between 1993 and 2003. This modest pickup in value added is not accompanied by an additional growth in employment or in the number of factories. Employment has grown at the rate of 1.3 per cent a year, with no change in the rate of growth post-delicensing. New factories have come up at the rate of 2.7 per cent a year; with the rate decelerating post-1992. Investment rate, however, has been commensurate with the growth of value added. Investment accelerated by about 8.5 per cent post-1992. Poor performance of employment is a very important question in itself and we cannot do full justice to this issue here, hence propose to take it up in another paper.
Is this growth pickup impressive and does it imply that the reforms have paid off? When we compare this performance with the pace of growth in the manufacturing sector of many East Asian countries, including China, we realise that not only in terms of employment, but also in terms of value added, the performance of Indian manufacturing 12 has not been close to that of East Asian countries. For example, manufacturing value added in South Korea grew at an average annual real growth rate of approximately 17 per cent between 1960 and 1980; China's manufacturing sector witnessed an average growth rate of 12 per cent per year between 1990 and 2005.

Section III: Evidence from Enterprise Surveys
What lies behind this relatively lacklustre performance? Before turning to the main econometric analysis of this paper, it is useful to examine the views of managers based on two recent surveys of manufacturing firms: the Investment Climate Survey conducted by the World Bank and a survey of about 250 firms from some of the most labour intensive sectors, conducted by ICRIER (Field Survey on "How to Enhance Employment Generation and Exports of Labour Intensive Firms"). 13 The World Bank's investment climate survey (ICS) data consists of the responses of managers to a wide range of questions including those pertaining to managers' perceptions about how various regulatory and other factors influence their firms.
A key question asks about the extent to which various factors are considered "a problem for the operation and growth" of the surveyed firm's business. For each factor listed, respondents can reply in terms of a five-point scale: 0= no obstacle; 1= minor obstacle; 2= moderate obstacle; 3= major obstacle; 4= very severe obstacle. 14 It enables one to compare firms' responses about various factors, ranging from regulatory and governance issues to infrastructure-related concerns, in terms of how they influence firms' operations or growth prospects.
Chart 5 depicts the fraction of firms describing a given factor as a major or very severe obstacle in the 2005 ICS survey (similar patterns are observed in the 2002 ICS survey). Tax-related issues, incorporating difficulties with either the tax administration system or complaints about tax rates, are considered to be a major or severe constraint by more than 40 per cent of respondents. Of course, it is not easy to interpret this finding given what is probably a natural penchant among firms to want to pay as little as possible in taxes. Ignoring tax-related issues then, the situation with infrastructure can be seen to be a critical obstacle for operations and growth from the perspective of the firms. 15 Almost 40 per cent of respondents cite it as a major or severe obstacle. In addition to infrastructure, one fifth or more respondents cite governance issues (which include concerns with corruption) and the cost and access to finance as the major obstacles. Surprisingly, an almost equal per cent of respondents cite skills and labour regulations as major obstacles (around 15 per cent).
13 This survey was conducted by a team led by Dr. Deb Kusum Das. We thank him for sharing the data with us. 14 We are aware that the phrasing of this question may not be ideal since it lumps together operations and growth. It is quite possible, for example, that some aspect of industrial regulations may not be a problem for the operations of a firm, unless the firm tried to expand its operations. 15 Infrastructure includes electricity, telecommunications, and transportation. Disaggregating this variable shows that the concern with infrastructure is overwhelmingly driven by concerns with electricity.
Telecommunications are hardly considered a problem. Fraction reporting issue as 'major' or 'very severe' obstacle

Chart 3: Obstacles for Operations and Growth
A useful follow-up question entailed asking firms what constituted the single most important obstacle for firms' operation and growth. By far the biggest problem relates to infrastructure and within this, electricity was cited as the key issue. Indeed, 31 per cent of firms listed electricity as the source of their single most important obstacle to operations and growth. Another source of the views of manufacturing firm managers comes from the ICRIER survey of 250 enterprises engaged in manufacturing activities in five different sectors (apparel, bicycles, gems and jewelry, leather, and sports goods). It is useful to examine the results of this survey because it covers firms from some of the most labour-intensive manufacturing activities. Thus, to the extent that labour regulations create serious constraints and growth prospects of firms, the sample firms should be among those most affected. 16 Broadly speaking, the respondents find electricity and infrastructure in general; financing; and skilled labour availability to be the most serious constraints on growth. Just like in the ICS survey they also point to specific regulations, especially those related to taxes (and fiscal benefits, in general), among things that can be improved. 17 The chart below summarizes the responses of the firms. 18 In response to the questions on hurdles to increasing employment, a majority of the respondents reported shortage of labour (of mostly skilled and semi-skilled labour) as the key hurdle to hiring more labour. 19 Further, most of the firms (approximately 90 per cent) responded in the negative to being affected by any labour disputes or to having labour unions in their organization and/or any impact of the unions on their activities. 16 Not all the firms covered responded to all the questions in the survey. For the purposes of the present study, we focus on the responses relating to the questions on the hurdles to the growth of the firms. 17 A look at the specific responses makes it clear that the concern with fiscal issues is very narrowly defined and is more in the nature of a personal issue to the firms, to the extent that taxes directly affect their bottom lines. In response to the question what would you like to see changed to help you, majority of them answered that they would like the taxes to be lowered or subsidies from the government. 18 The survey also tried to find whether the technological changes are such that they are inhibiting employment growth. About two-third of the respondents acknowledged technological changes (either a lot or modest) taking place globally in their industry. Of those answering in affirmative to world-wide changes in technology in their respective industries, 70 percent of the respondents adopted new technology during the five years prior to the year of the survey; but the majority of them still find a gap between the technologies they used and those used globally. In general, however, there is no clear evidence on the impact of adoption of new technologies on labor. One potential explanation for the lack of a clear pattern could be that while adoption of new technology, on one hand, might be labor saving (substitution effect) and, on the other hand, growth result from adopting new technology might be expansionary and lead to hiring of more labor (growth effect). 19 Interestingly approximately 10 percent of the firms rue the lack of training facilities. This is consistent with shortage of labor or, more precisely, the shortage of the "right" kind of labor for the "right" kind of skills.

Chart 4: Areas for Improvement
Source: Authors' calculation using data from the ICRIER survey Taken together, there are some striking similarities in the results of these two very different surveys. First, infrastructure-related issues are very high on the list of constraints faced by firms. Indeed, ignoring tax-related issues, concerns with electricity seem to be paramount. Second, finance-related issues also seem be a problem, especially for the smaller firms. Third, surprisingly labour regulations do not show up as a significant concern for firms. Indeed, both surveys suggest that concerns with skillsrelated issues are more important than those having to do with labour regulations.
While the concerns with electricity are not surprising for anyone with some familiarity with the Indian industrial scene, the low ranking of labour regulations as obstacles for operations and growth is surprising. One interpretation of these results could be that labour regulations may not matter much to firms in practice. This could happen, for example, if noncompliance with labour regulations is not costly. Alternatively, firms may be able to "get around" restrictions on layoffs by hiring contract workers. A second interpretation of these results, however, is that labour regulations may not matter that much to incumbent firms. But it may matter to a non-incumbent investor contemplating entry into the manufacturing sector.
More generally, an investor's choice on which specific sectors (for example, services versus manufacturing) and subsectors (for example, a more labour intensive manufacturing industry versus a less labour-intensive one) to enter, as well as the production technologies, scale, and desired levels of employment to adopt, can be expected to be influenced by the regulatory framework. In this way, there may be an "exante" effect of the law that would be very difficult to capture through the surveys of 16 incumbent manufactures. In other words, deterred by specific elements of labour regulations such as Chapter VB of the IDA, potential investors, especially those contemplating large investments, choose to avoid investing in manufacturing altogether, or if they do invest in manufacturing, they avoid subsectors, product lines, or scales of production for which the regulations have most bite.
In what follows, we turn to an approach which has the potential for getting around the "selection" problem inherent in surveys of incumbent firms. In particular, we use industry-level data from India's organized manufacturing sector to examine the relative performance of industries with various characteristics.

Section IV: Econometric Analysis
We are interested in testing the variants of the following hypotheses: did industries that are more labour intensive; or industries that rely more on infrastructure; or industries that rely more on the financial sector for their financing needs grew less than the control group of industries in post-delicensing period? The econometric methodology is derived from Rajan and Zingales (1998), who analyse the effect of financial development on growth by comparing the growth of industries, which depend more on the financial sector in countries with greater financial depth with the growth of these industries in countries with shallower financial markets. Thus, if the financial sector indeed matters for growth then one would see higher growth in industries that rely on the financial sector in countries which have deeper financial sector, and vice versa. This technique gets to the causality issue much more cleverly than the alternative econometric ways to measure this relationship. The methodology has subsequently been used in several different contexts. 20 We use this technique to look at the constraints that Indian industry might have experienced post-delicensing. Hence we analyse the performance of the industries that rely more heavily on infrastructure; industries that depend on the financial sector; and the labour-intensive industries post-delicensing. An evidence of lacklustre growth in these industries is attributed then to the unavailability of inputs or factors that the respective industries rely more heavily on. Thus, if the infrastructure-dependent industries have not performed well post-delicensing then it is likely to be due to the fact that the infrastructure availability has not been adequate for these industries to avail of the maximum benefit from the reforms.

IV A: Construction of Variables
Reform Variables: As discussed in Section II, industrial and trade policy has seen wide ranging reforms over the period under study. While limited reforms were started from mid-1980s onwards, major policy changes were undertaken following the crisis in 1991. Some of the reforms introduced were more generic aimed at macroeconomic management; others were more specific to the industries. The reforms spanned several areas, including delicensing of industries; trade reforms and exchange rates reforms. In subsequent years these were complemented by the liberalisation of foreign investmentboth FDI and portfolio; dereservation of industrial sectors under small scale; financial sector liberalisation; and privatisation of public sector units.
In our econometric exercise we look at the effect of delicensing on Indian manufacturing industries. The reasons being that it is one of the most comprehensive programmes which covered almost all the industries, and the information on it is readily available. The fact that these reforms were undertaken at different points in time, allows us to include timefixed effects to account for unobservable but common macroeconomic shocks in the regressions. We do not have complete data for trade reforms, but we do control for it in the robustness exercises. 21 In robustness tests we also estimate a specification in which we include the interaction of industry characteristics with a post-1992 dummy in the benchmark specification to account for the reforms, which were more generic in nature, besides the delicensing. Results remain broadly unchanged and are presented selectively here.
Industrial Characteristics: Next, we define three industrial characteristics of various industries: labour intensity; dependence of industries on external finance; and infrastructure dependence. Rajan and Zingales (1998) assume that there are probably technological reasons why some industries depend more on external finance than others. We extend this reasoning to labour intensity and infrastructure intensity. To the extent that these two characteristics define input usage, the technological requirement assumption is perhaps as valid as for defining the external financing dependence. We briefly describe the various industrial characteristics below, further details are provided in the various tables in the Appendices.

Labour Intensity:
We define labour intensity as the ratio of total employment to capital stock. Since there are no comprehensive databases of employment at firm level we use the ASI industry level data to calculate this ratio.

Dependence on External Finance:
We calculate the external financial dependence of firms in two different ways and using two different databases. The first one uses the firmlevel data from the Prowess database published by the CMIE, and employs the same definition as used by Rajan and Zingales (1998). The second measure is calculated using the ASI data as the ratio of outstanding loans to invested capital. The index of external finance dependence does not correlate well across two databases and across different definitions. Neither of these correlates too well with the index calculated by Rajan and Zingales (1998) that was calculated for industrial data at two-digit level for US industries.
To the extent that our firm-level data (from Prowess) is only for listed firms whose access to financial markets is likely to be different from that of small and medium enterprises, and it might affect the cross-industry ranking, we use the financial dependence indicator calculated using the ASI data.

Infrastructure Dependence of Industries:
We calculate it as the ratio of expenses on distribution (i.e., storage and transportation) and power and fuel to gross value added using the firm level data. To the extent that we have data on expenses on fuel consumption in both the CMIE and ASI, we calculate an indicator just as the ratio of fuel expenditure to gross value added. These are highly correlated across the two databases; and with the indicator, which includes distribution expenses as well. Appendix B1 indicates which industries quality as below or above median for each of these characteristics.
In order to get around the concern that these characteristics would reflect the equilibrium conditions between the demand and supply of the respective inputs, we use the data from an earlier year (in general we use averages over the early1990s, but where data are available we confirmed that the industry characteristics are correlated highly with the ones that calculate using the data for earlier period) rather than contemporaneous data. Furthermore, to smooth out the noise in the data we use five-year averages of the relevant variables to calculate the industry indicators. We also confirmed, where possible, that the relative industry rankings across various characteristics do not change overtime. This robustness check gives credence to the belief that there are perhaps external technological reasons for why an industry uses more external finance; or uses more labour than capital; or depends more on infrastructure; and to the fact that using data from the early1990s is legitimate.
The questions that come to mind about these industry features are: are these capturing some other features of the industries; and how are the three features correlated with each other. In Tables B2 in the Appendix, we report correlations among these characteristics and some of the other features of the industries that we could calculate. As we can see from the table the various industry characteristics are not correlated significantly with each other. The exceptions include that the labour-intensive industries are negatively correlated with imports and financial dependence; and infrastructure dependence is negatively correlated with import and FDI intensity. Labour-intensive industries are also somewhat more export-intensive.

IVB: Empirical Results from the ASI Data
We begin by exploring the possibility that the overall performance of the manufacturing sector masks significant inter-industry heterogeneity. Are there certain industries which have not benefited as much from the reforms?
In Table 3 below, we find that the performance varies across different sectors. In particular, we identify industries, which depend more on infrastructure and the financial sector for their financing needs and the labour-intensive industries (see Appendix B1). We divide the industries into those belonging to above or median values for each industry characteristic and estimate separate regressions for industries below and above median values. We use log of value added as the dependent variable and control for industry and 19 year fixed effects and a dummy for delicensing which varies across industries and years in the regressions.  Table 3 show that the industries which are more infrastructure-intensive, on average, experience 15 per cent lower growth in value added post-delicensing (i.e., in the delicensed period relative to the earlier period), as compared with 33 per cent higher output growth in value added of industries which are less reliant on infrastructure.

Results in
Similarly, the industries more dependent on the financial sector or the labour-intensive industries have fared much worse than the industries that do not rely as much on the financial sector and are less labour-intensive. Thus, there seems to be significant heterogeneity in the performance of Indian manufacturing across industry groups.

IV C: Econometric Framework
We use the following econometric specification to analyse the impact of delicensing on various performance indicators: Y it = Σα i d i + Σβ t d t + γ (delicensing dummy it ) + δ (characteristic of industry i * delicensing dummy it ) + ε it (1) Where Y it is the outcome variable measured in log. As before we consider gross value added at constant prices, employment, capital stock, and number of factories as the outcome variables.
In equation 1, di's are industry-specific dummies and α i 's are their respective coefficients; d t 's are year-specific dummies and β t 's are their respective coefficients. The fixed effects account for the industry-specific omitted variables; and the year fixed effects control for year-specific shocks that are common to all industries. Since we are using industry fixed effects and year fixed effects in the regression equation the only additional variables we can include are the ones that vary with both industry and year. The next term in equation 1 is the delicensing dummy, which varies over time and industry. The dummy takes a value one from the year when the delicensing requirement for a particular industry was removed and remains one for the rest of the sample period. 22 We are interested in testing the variants of the following hypotheses: did industries that are more labour intensive (or industries that rely more on infrastructure or the financial sector for their financing needs) grow less than the control group of industries in the postdelicensing period? Testing for these hypotheses requires us to set up the regression equation for difference in differences estimates. Continuing with the specific hypothesis involving labour-intensive industries, consider the following possible cases for any given outcome variable: Essentially, we would like to test the hypothesis that (Θ L,Post -Θ L,Pre )-( Θ C,Post-Θ C,Pre ) is significantly different from zero. The coefficient δ in equation 1 allows us to do this. We use the interaction with each of the industry characteristics discussed above with delicensing separately and together in the regression equations.
How do we interpret a negative and significant coefficient for the interaction term of a particular industry characteristic, let's say infrastructure-dependent industries? The coefficient indicates that the industries which use infrastructure more intensively have grown less post-delicensing as compared to the industries which use infrastructure less intensively. Can it be interpreted as a causal relationship between the lack of infrastructure and performance? As mentioned in the introduction, to the extent that an industry characteristic is exogenous of performance, e.g., it is some sort of a technical requirement; or if we can have control for potential omitted variables, then we can probably claim causality in this result.
For exogeneity in our industry characteristics we use the data from the earliest possible period for which we have the data (in our case the early1990s). We control for omitted variables varying only over industries and over years by including the respective fixed effects. To rule out other potential omitted variables, we conduct extensive robustness tests as described later.
In Table 4, we present our results for the benchmark case as given by Equation 1. Coefficients on both the industry and year fixed effects have been suppressed from the table. In the results in column 1, the coefficient for delicensing shows a 12 per cent increase in value added per industry post-delicensing. Given that the average delicensing period is about 15 years, it amounts to a less than 1 per cent increase in value added per year in post-delicensing period. However, as we had seen in Table 3, certain industries did not fare as well during the post-delicensing period. Thus, when we control for the different effects on these industries separately, the post-delicensing impact on growth of the control group improves substantially.
In Columns 2-4, we include these characteristics with the interaction of delicensing one at a time. As expected, the performance of the control group goes up considerably. From Columns 2-4, we see that the industries more dependent on infrastructure, labour and external finance respectively have witnessed slower growth as opposed to their respective control group. In Column 5, we include them together and find that industries ranking higher on each of our three industry characteristics have fared poorly in the postdelicensing period. Finally, the last column is the same as Column 5 except that in this column instead of including the index of industry characteristics, we divide them into above and below median groups and include the interaction of the dummy variables that takes the value when an industry is above the median of the respective characteristic, with delicensing. Once again we find that (the results hold and)? industries above the median in each of the three characteristics have not done as well as the control group in the post-delicensing period. Quantitatively, the results, from Column 6 of Table 4, indicate that in the postdelicensing period the above median infrastructure intensive industries grew 10 per cent less than the industries below the median of infrastructure dependence. Similarly, industries above median in the distribution of financial dependence grew 18 per cent less than the industries below the median of financial dependence; and for labour intensity, industries with above median labour intensity grew 19 per cent less than the below median industries post-delicensing.
In Table 5, results are presented for dependent variable, number of factories (in log). The overall performance of Indian manufacturing seems to be even less impressive when we look at the number of factories. Overall there is no acceleration in the rate of expansion of factories post-delicensing. These results are on account of the fact that the performance has been particularly worse for the labour-intensive industries and industries dependent 23 on the financial sector. Once we control these as in the previous set of regressions, the performance of the control group (industries less dependent on infrastructure or financial sector or labour) is seen to be much better. The point remains that industries more dependent on external finance and labour intensive industries have fared much worse in the post-delicensing period in terms of new factories opening. Next, we look at employment and capital stock. The issues related to employment are manifold and much more complex, and all of these probably cannot be addressed in this paper. Some of the issues that merit attention include: why has growth not been employment-intensive; is technology displacing labour; how has the employment of unskilled vs. skilled workers evolved overtime; how is the skill premium changing overtime etc. For brevity, we look only at total employment here, which includes manual workers as well as supervisors; and regular as well as contract employees.
We look at two econometric specifications here. The first is the estimates from equation 1, results of which are reported in Table 6a and 7a, respectively, for employment and capital stock. From Column 1 of Table 6a, employment has increased by a mere 7 per 24 cent over the entire delicensing period. With the average delicensing period about 15 years, this translates into a less than ½ per cent increase in employment per year postproduct market reforms.
As expected, once we introduce heterogeneity based on industry characteristics, the performance of the control group improves rather substantially (Columns 2-5). The maximum increase is in the case of the industries less dependent on external finance. Notably, infrastructure and external finance dependent industries as well as labourintensive industries are the weakest performers in so far as employment generation in the post-delicensing period is concerned.
In addition, we estimate the following regression equation: In equation 2, E it refers to log of employment (or log of invested capital in real terms), d i 's are industry-specific dummies and d t 's are year-specific dummies as before. We also include the log of gross value added in the equation; the coefficient θ can be interpreted as the employment elasticity of output. It measures the percentage change in employment for a 1 per cent increase in output. The next term is the interaction of delicensing dummy with Y it. Its coefficient π gives the employment elasticity of output post-delicensing. Finally, we include the interaction of Y it , delicensing and industry characteristics. The coefficient λ measures the employment elasticity of output for the industry characteristic used in the interaction with post-delicensing. Thus, if we are including labour intensity in the interaction term in equation 2 then it measures the change in the elasticity of employment post-delicensing in labour-intensive industries. If it is positive it implies that the employment elasticity in labour-intensive industries has increased post-delicensing and so on.
Results on employment from specification 2 are in Table 6b. The results indicate that the employment elasticity of output is about 50 per cent on average, though there are differences across industries. The elasticity is higher for labour-intensive industries than for those dependent on infrastructure or finance. Results also indicate that there has been no change in the elasticity of employment post-delicensing, this is true on average for all industries, including the industry characteristics that we control for explicitly in our regressions.
These results have two implications: first, if growth were to accelerate in Indian manufacturing it would probably generate employment at the same rate as before; and second, in order to generate more employment in Indian manufacturing it is imperative that the labour-intensive sectors grow faster. As we mentioned earlier, aggregate employment masks several nuances related to different kinds of employment, but we do not have space to discuss them all here.  For analysing the patterns in investment we use both the specifications used for employment (Tables 7a and 7b). Thus, we look at the capital elasticity (i.e., how investment changes) of value added and compare it with the behaviour of investment post-delicensing. We also compare the investment behaviour across industries and see whether there are any patterns in the investment changes across industries postdelicensing. Here we find that the capital elasticity (Table 7b) is higher than employment elasticity. Across industries, infrastructure-and financially-dependent industries see higher investment than the labour-intensive ones as value added increases. Investment has also increased somewhat post-delicensing; quite interestingly this is on account of a higher investment in the labour-intensive industries. Thus, over time and especially postdelicensing, the labour-intensive industries seem to be substituting away from labour and adopting relatively more capital intensive technology! In addition, we find that industries, which are more dependent on external finance, see a decline in investment in the postdelicensing period.

IV D: Robustness of Results
We do extensive tests for the robustness of our results. These include checking the robustness to different time periods, to omitted variables, and to potential outliers. We account for the lags between policy and implementation; and also the possibility that the outcomes might be correlated by the industries or by the year of delicensing. While we do obtain small variations in coefficients and in the standard errors across these different specifications, overall the results are quite robust to various sensitivity tests. One result, which does seem a bit sensitive to some of the corrections for autocorrelation, is the result on infrastructure dependence. In some of the corrections for autocorrelations, the coefficients of the interaction between infrastructure and delicensing become less significant, but even here its effect holds at about the 20 per cent level of significance. Details on each robustness test follow. 29 Though in the methodology used here the omitted variables that vary only by industries or only by year have been accounted for through the respective fixed effects, the estimates remain susceptible to the omission of variables that vary over industry-year dimensions of the data. In particular, there might have been the following two types of omissions: first, the interaction of delicensing with industry characteristics other than the ones included; and second, the interaction of policy variables other than delicensing and their interactions with the industry characteristics included.
We explicitly control for only one of the major policy changes pertaining to Indian industries -delicensing. What about the other policy changes? In order to address these concerns we carry out two robustness tests. First, to control for the reforms which were more generic rather than specific to industries, we include in our regressions interaction of industrial characteristics with a post-1992 dummy. Second, we construct a trade policy measure which is industry-specific and interact it with industrial characteristics. Results that are dependent on infrastructure and external finance, and labour-intensive industries that have not benefited as much from reforms are fairly robust across these various specifications.
While we are unable to conduct these tests for some of the other reforms, the results are unlikely to change. The reason is that the reforms are highly correlated over time and across sectors. Thus, even if we get a somewhat different coefficient when we include interaction of industry characteristics with different reforms instead of delicensing, the basic message we want to bring home, that without sufficient infrastructure development, financial depth and progress on factors inhibiting labour intensive industries, Indian industry is unlikely to realise its potential would hold. For this argument it is really immaterial what kind of reforms we are talking about. Second, if we include the interactions of industry characteristics with different reforms measures, e.g. delicensing and trade reforms, in the same specification, then the coefficient for a particular policy measure would become weaker and probably even lose their statistical significance. Such a specification will be of little use since again the interest is in a composite reform measure rather than specific reform measure. Thus, even if the coefficients might be biased in the benchmark specification, to the extent that we do not really care about attributing it to delicensing per se, we are fine. For omitted industry characteristics we include other industrial characteristics in the regressions, such as export intensity or FDI intensity, interacted with delicensing and find the results to be robust.
We report results for some of the robustness tests in Table 7. The reported results are for the dependent variable log value added. In order to address the concerns related to autocorrelation we reduce the sample length to the period from 1980 onwards, Column 1, Table 7. We can restrict the period further but then we would start losing our control period. In the results reported in Column 2 of the table we calculate the standard errors corrected for Newey West adjustment. We also drop two industries, tobacco and petroleum (these industries typically show extreme measures on various accounts such as labour productivity and size) in column 3 and the results are unchanged. In Columns (4) and (5) of Table 7 we include the trade reform variable. Here as expected we find that the trade reforms have had a growth enhancing effect on Indian industries, and again the effect has varied across industries along the same dimension as we have seen in the earlier tables.
To control for the fact that there may be a lag between reform and actual implementation or the effect of the reform is felt with a delay, in Column 6 we lag the delicensing dummy by two years. In other words, if an industry was delicensed in 1991, we assume that the effect was felt two years later, i.e., 1993 onwards. Our benchmark results hold qualitatively.

Section V: Concluding Remarks
In this paper, we have analysed the performance of registered Indian manufacturing sector in India using data from the ASI. In line with some recent studies, we find that industrial performance has improved with industrial delicensing. However, our analysis also indicates that there is considerable heterogeneity in the response of industries to policy reforms. In particular, the industries more dependent on infrastructure; industries with greater dependence on sources of external finance to the firms; as well as those with high labour intensity have not performed well.
From a policy perspective, the important question then is what features of India's policy and institutional landscape explain this pattern? The ongoing policy debates in India suggest several leading candidates. In the case of infrastructure dependent industries, the inadequacy of public provision of infrastructure is probably the main culprit. Similarly, the failure to improve the Indian financial sector's ability to identify and finance creditworthy firms and investors may well lie behind the relatively weak performance of industries especially dependent on external finance.
A complementary analysis of two firm-level surveys of managers in the manufacturing sector lends further support to these arguments, especially in the case of infrastructure and finance. Taken together, the results of the World Bank's investment climate survey and ICRIER survey of labour-intensive manufactures support the notion that weak provision of infrastructure and finance has constrained the growth of the manufacturing sector.
As regards the weak performance of labour-intensive industries, certain elements of India's labour regulation may well be an important causal factor as argued by a number of observors of the Indian economy. In particular, certain elements of the IDA may have raised significantly the effective cost of hiring workers, thereby hitting the relative profitability of labour-intensive industries disproportionately. Since this is more likely to be the case for larger firms (due to the nature of the regulations), labour regulations may have led to relatively weaker performance of labour-intensive industries in two ways: first, by discouraging entry by large firms; and second, by reducing incentives among small firms to expand. 32 However, other factors cannot be ruled out --such as the policy of reserving a whole range of labour-intensive products for production by small-scale firms as recently as 2001. One way to make headway on this issue --i.e., establishing whether certain elements of the policy or institutional framework are causal drivers of the pattern of industrial performance we find -is to extend our analysis to the state level. To the extent that India's states present sufficient variability in the provision of infrastructure and finance and in the stance of labour regulations (as they actually apply to firms and not just on paper), carrying out this analysis at the state level should be very useful. We take up this issue in our forthcoming work.
In the meantime, our econometric analysis has served to highlight from where relatively weaker performance in India's manufacturing sector is coming. Unlike previous work that has highlighted mainly the role of labour regulations in influencing industrial performance, our econometric results interpreted in conjunction with perceptions of managers suggest that steps to improve infrastructure and the financial system should go a long way in improving manufacturing performance. Additionally, our results also point to the urgency of identifying the constraints on labour-intensive manufacturing in India and relaxing these.

Appendix A: Data Sources and Construction of Variables
We have primarily used ASI data at the three digit level. After the concordance from NIC87 and NIC70 into NIC 98 classification, we have data on 49 industries. Data are available from 1973-2003. Data in general seems good and comparable pre-and post-1998, when there was a change in the sampling framework. The following industries were excluded from the analysis. The first three (dressing & dyeing of fur, saw milling, and publishing) were excluded because of lack of data on infrastructure dependence from CMIE. The others that were dropped included processing of nuclear fuels and reproduction of recorded media. As noted by Aghion et al. (2006), processing of nuclear fuels is likely to be affected by noneconomic factors and hence we drop them from our sample. Finally, reproduction of recorded media was introduced as a new category in 1998. There is no data for this industry for the period between 1973 and 1998 and is, therefore, excluded from the sample. As the table below shows, we exclude less than 1 per cent of the registered manufacturing sector, whether we look in terms of employment or gross value added.

ASI
Blue-collared workers are defined to include all persons employed directly or through any agency whether for wages or not, and engaged in any manufacturing process or in cleaning any part of the machinery or premises used for manufacturing process or in any other kind of work incidental to or connected with the manufacturing process or subject of the manufacturing process.

Variable Data Source Description/construction Total employment ASI
Total employment is defined to include all blue-collared workers as defined above and persons receiving wages and holding clerical or supervisory or managerial positions or engaged in administrative office, store keeping section and welfare section, sales department as also those engaged in purchase of raw materials etc. or production of fixed assets for the factory and watch and ward staff.    1991 152,153,154,155,171,1721 73,181,182,192,202,221,22 2,233,241,269,271,272,289 ,313,314,332,333,351,352, 359,361,369  We used the data provided in Aghion et al (2006)  In addition, ICRIER holds public policy workshops to bring together policy makers, academicians, Union Cabinet Ministers, senior industry representatives, Members of Parliament and Media persons to try and forge a consensus on issues of major policy interest. In the effort to better understand global developments and experiences of other economies which have bearing on India's policies and performance, ICRIER invites distinguished scholars and policy makers from around the world to deliver public lectures on economic themes of interest to contemporary India.
ICRIER's highly qualified core team of researchers includes several PhD's from reputed Indian and foreign universities. At present the team has 18 economists. The team is led by Dr. Rajiv Kumar, D.Phil in Economics from Oxford University.