Multiproduct Firms, Product Scope and Productivity: Evidence from India’s Product Reservation Policy Ishani Tewari, Yale School of Management† Joshua Wilde, University of South Florida†† September 2014

Abstract: We provide novel evidence showing product scope dynamics within a firm is an important dimension of productivity growth. This channel is identified by leveraging the gradual dismantling of an Indian regulation that “reserved” hundreds of products for manufacture in the small-scale sector. Following the removal of these product market restrictions, product churning and productivity rose. Multiproduct firms who were never in the reserved sector drive this increase, suggesting that the reservation policy constrained their ability to achieve the optimal product mix. Our findings underscore the importance of incorporating heterogeneity at the product-firm level in assessing the impact of size-contingent regulation.

JEL codes: O1, O25, O4, L5 Keywords: Productivity, Multiproduct Firms, India, Dereservation, Products

†

[email protected] †† [email protected]. Andrew Foster, Ross Levine, David Weil and

Ivo Welch provided helpful comments and discussion. PC Mohanan from the Ministry of Statistics and Programme Implementation and Sajeevan Gopalan from the Ministry of MSME were very helpful in procuring and helping with data. Financial support was provided by the Kauffman Foundation, the Rhodes Center for International Economics and Finance, and the

1

Brown COE Hazeltine Grant. We also thank seminar participants at the 2011 Productivity Conference IIT Bombay, Brown University, the 2012 AEA Meetings, the 2012 Yale China-India Conference, the 2013 New England Universities Development Consortium, and the 2013 Choice Symposium

I. Introduction Recently, an increasing focus on production patterns at the product-firm level has revealed some interesting intra-firm dynamics. This research shows that changes in product scope have large implications for overall growth at the industry and economy level. For example, product mix changes account for a substantial portion of output change over time, even larger than the role of entry or exit. In the U.S., Bernard et al. (2010) find that product churning accounts for a third of U.S output growth between 1972 and 1997. Goldberg et al. (2010) in a study on Indian multiproduct firms find that changes in product mix contributes 25% of the increase in manufacturing output between 1989 and 2003. Among Spanish firms, Doraszelski and Jaumandreu (2013) show that within-firm productivity gains contribute 25%-90% of industry productivity growth. 1 This mounting evidence has motivated theorists to incorporate heterogeneity, not just at the firm level, but at the product level into existing frameworks. Based on this feature, adjustments to policy changes which increase competition (typically trade liberalization in the current literature) manifest themselves at both the inter-firm and intra-firm 11

The primary source of this growth comes from a firm’s R&D activity to introduce new products and process innovations

2

margin (Bernard et al. 2012). A central result emerging from these models is that multiproduct producers are important drivers of productivity because of their ability to reallocate resources within the firm via product-line dynamics.

Despite this recognition for the relevance of the

intra-firm channel, there has been almost no work incorporating this feature into non-trade policy settings. A particularly important class of such policies is size-dependent regulation which has received much attention in work highlighting the importance of firm heterogeneity (Restuccia and Rogerson 2008; Guner, Ventura and Yi 2008; Garcia-Santana and Pijoan-Mas 2014; Garicano, Lelarge and Van Reenen 2012). However, there is no product-firm heterogeneity in these analyses. Its absence is the motivation for our work here. In this paper, we examine product churning as a driver of productivity during the unraveling of an important domestic Indian regulation—product reservation. For decades, the Indian government mandated that certain products, ranging from food items to chemicals, would be reserved for exclusive manufacture by small-scale enterprises. These items numbered over 800 and constituted almost 25 percent of manufacturing output. Starting in 1997, this regulation has been dismantled gradually, with different products being taken off the “reserved list” or being “dereserved” at different times. Dereservation was part of the “troika” of major Indian industrial policies along with trade tariffs and industrial licensing, and played a very important role in shaping Indian manufacturing. Dereservation in particular heavily influenced the evolution of the small-scale sector (Little et al. 1987; Mohan 2002). While the impact of other policies like tariffs, licensing, and FDI reform on growth and productivity have been studied extensively in the growth and development literature, this paper is among the first to rigorously study product reservations. Moreover, the evidence here is the first to support product churning as a margin of reallocative activity following industrial reform in India.

3

Leveraging plausibly exogenous variation generated by this policy change, we attempt to forge a causal connection between increased competition, product scope dynamics, and increases in productivity. Our empirical strategy relies on exploiting the timing of changes in the fraction of output produced in unreserved goods across different industries. Using this approach, we explore three main questions. First, we document the effects of this policy and the subsequent increase in competition on overall firm size and productivity. Second, we assess whether product scope dynamics (i.e. the addition and dropping of products) are an important margin of adjustment. And finally, we examine whether a firm’s past productivity affects product churning. The above questions are motivated by empirical and theoretical work that assesses the impact of globalization on the nature of product scope dynamics and its relation to firm productivity and size. Many of these models predict a reduction in product scope when competition pushes firms towards their “core competencies”, inducing them to shed their least productive products (Bernard, et al. 2011, Eckel and Neary 2010, Mayer et al., 2011). Other models, like Feenstra and Ma (2007) show opening up trade leads to fewer firms surviving in each country, but more varieties produced by each of those firms. In a similar vein, certain type of firms (such as those with higher productivity in Qiu and Zhou, 2013, or higher organizational capability as in Nocke and Yeaple, 2006) will expand scope. Thus the literature delivers contradictory results on important details such as whether product scope increases or decreases in the face of increase competition, and whether it is the least or most productive firms that alter their product lines. For example, Iacovone and Javorcik (2010) find scope reduction in Mexican firms post-NAFTA, and Baldwin and Gu (2009) only find scope reduction among small Canadian firms, whereas Berthou and Fontagne (2013) show the most

4

productive French firms increased the number of products produced after the establishment of the Eurozone. In this paper, we find that dereservation of products increased the size and productivity of firms. Multiproduct rather than single-product producers are the main drivers of these effects. For these firms, a 10 percentage point increase in unreserved share in an industry leads to a 3.5% increase in productivity, while there is no impact on single-product firms. This productivity increase is driven entirely by the sub-set of multiproduct firms who were never in the reserved sector. Product dereservation also results in substantial product additions and droppings among these multiproduct producers. Interestingly, if we look at adding and dropping of all products, we see an impact of dereservation only on product addition. We find that over the course of the policy, dereservation led to a 4.9% increase in the probability a firm would be a net adder. This echoes the findings of Goldberg et al. (2010) who also find that Indian firms tend to add products but not shed them. However, if we distinguish between products that belonged to the reserved sector and those which did not, product churning becomes apparent. The probability of adding and dropping dereserved products is higher after the policy, and disproportionately so for multiproduct firms. This evidence suggests that product reservation constrained the ability of multiproduct firms to achieve their optimal product mix and once the policy was removed, these firms were able boost productivity through product churning. Next, we examine whether a firm’s past productivity is an important correlate of its product churning. Higher productivity firms are more likely to respond to dereservation by adding a previously reserved product. And in line with the above findings, this is most likely in multiproduct firms, particularly those who were new entrants into the reserved sector.

5

The above findings complement and augment two strands of work. First, as we discussed above, there is a growing theoretical and empirical literature which stresses the importance of incorporating within-firm activity into traditional theories of firm productivity and industry dynamics. One paper that warrants some further mention is Goldberg et al. (2012) who are also interested in product line dynamics in the Indian context. They find that Indian multiproduct firms are very similar to U.S multiproduct firms in that they are larger and more productive than their single-product counterparts. Unlike the U.S, though, product additions but not product churning (product addition and dropping or creative destruction) is far more pervasive. The writers attempt to link trade policy to the intra-firm channel but are unable to find a connection. In contrast, we do find product churning in the aftermath of a separate, domestic policy. A second line of related work is the body of literature linking manufacturing productivity and growth to policies, and in particular to policies that restrict the size of firms. Recent work, for example by Guner, Ventura, and Yi (2008), Restuccia and Rogerson (2008), Hsieh and Klenow (2009), Garcia-Santana and Pijoan-Mas (2014) and Garicano, Lelarge and Van Reenen (2012) attempt to measure the aggregate productivity cost of distortions from the misallocation of capital, labor and managerial talent. Quantitative results show a large impact of size dependent policies, accounting for up to 50 percent of the productivity gap between some developing economies and the US. Garcia-Santana and Pijoan-Mas (2014) specifically examine dereservation in a Lucas span-of-control type model where the policy results in a misallocation of talent. Calibration results show dereservation boots output by 6.8% in manufacturing and 2% in the overall economy, and TFP by 2% and 0.75% respectively— quantitatively similar to our findings. Bollard et al. (2013) do not find evidence of increases in

6

productivity after dereservation. 2 In recent work, Martin et al. (2014) find a positive impact of dereservation on firm and district output growth. While all of these papers either feature no heterogeneity or heterogeneity only at the firm level, we underscore the importance of another dimension—heterogeneity at the product-firm level—in assessing the impact of sizecontingent regulation. The paper is organized as follows: Section 2 provides a brief background on product reservation along with descriptive evidence. Section 3 describes the empirical approach, data and variables. Section 4 presents the results and section 5 concludes. 2. Background on India’s Reservation Policy The Indian government has a long history of promoting small-scale industries. Postindependence, policymakers were inspired by the Gandhian ideals which viewed small industry as a means of generating employment and achieving social equity. The industrial agenda emphasized the need to encourage new ventures by individuals from classes, castes and communities that had historically contributed poorly to the nation's entrepreneurial activity. Indian small scale industry (SSI) was borne of the Industries (Development and Regulation) Act in 1951. Small scale industries have historically formed a large component of the Indian manufacturing sector, accounting for 40% of industrial production and 35% of total employment over the last three decades. Its overall contribution to GDP has been approximately 6%. SSI enterprises were beneficiaries of substantial institutional and financial assistance. Among the myriad protectionist measures benefiting SSIs, arguably the most extreme was product reservation, which was introduced in the Third Five Year Plan (1961-1966). Hundreds of products across the manufacturing sector were only allowed to be produced by small scale firms, 2

Some of this may be in part due to their measure of dereservation which is not based on product-level information.

7

insulating them from competition. 3 The types of products on the reserved list were varied, spanning many industrial sectors such as food, chemicals, electronics and textiles. Within the small-scale sector, the output share of reserved products was approximately 30% in 1987. Overall, reserved products constituted about 12% of Indian manufacturing output. There is considerable heterogeneity within industry sectors with reserved products forming 80% of output in hosiery and garments, 57% in certain wood products, and a negligible fraction in textiles. The reserved list as of July 2010 (Appendix Figure A) shows that currently reserved products are a diverse mix -- bread, wooden furniture, glass bangles, soap and steel furniture. There is no official documentation that describes how the reservation policy was formulated. An expert committee constituted in 1997 to review small scale industry in India's postliberalization period states, “the choice of products for reservation was necessarily arbitrary." Some observers think that the government's goal in early decades following independence was to create a labor intensive sector that would absorb abundant labor. However, there is no technical criterion for reserving specific goods say, based on optimal capital to labor ratios (which are difficult to ascertain in the first place). Since 1967, the "reserved" list has evolved, with a few more items being mostly added prior to 1997, at which point there were 821 products on the list. Starting in 1997, products were gradually removed from the reserved list, and as of 2010, only 20 products. Figure 1(a) shows the change in the reservation policy over the sample period. We plot the proportion of products in the economy that are unreserved by year from 2000-2010. We see that the policy started off gradually and accelerated greatly in the mid-2000s, continuing till the present day. The change varied not only across time but also across industries. Figure 1(b) gives a cross-sectoral snapshot 3

However, existing large enterprises that had been producing the products were allowed to continue production without being allowed to expand.

8

of the average proportion of products unreserved in 2000-2010. Reserved products are present in all two-digit sectors with the exception of tobacco products. Around 20% of products in apparel manufacturing are reserved while less than 5% in food and beverages. As in the case of reservation, the process of dereservation is also not well-understood. A reading of government documents, reports and media does not give clear-cut reasons as to why certain products were initially reserved or dereserved at certain times. A conclusive and definitive account of the dereservation process is not available. Conversations with ministry officials reveal no single reason as to why or when a given product was dereserved. Explanations range from competition from imports, technology requirements, needs to comply with regulation, no benefit to small producers or availability of unreserved substitute products. A product's path to dereservation tends to be lengthy and circuitous. A product is identified as a dereservation candidate by ministry or industry players (including manufacturers of reserved products themselves who find the investment ceiling constraining). Once identified, a series of meetings between "stakeholders" (such as trade associations or small firm groups and officials) takes place. After review up a chain of bureaucrats, the dereservation of a product is signed into law by the central government minister. Political economy factors, such as lobbying by interest groups, appear to be the prominent factor behind the sequencing. Qualitative support for the “random” nature of reservation and dereservation is reflected in the extent of reservation/dereservation both across and within product categories. We already noted the array of diverse product categories across sectors that were chosen for reservation. Appendix Figure B shows how even within a relatively narrowly defined industry like “vegetable oils,” we have many oils that were never reserved, and several (like sesame oil, mustard oil and rapeseed oil) which were. Among the latter, there is even variation regarding when they were dereserved.

9

It is difficult to think of some systematic and endogenous rationalization in this cross-sectional and time-wise pattern of reservation. While a conclusive and definitive account of the dereservation process is not available, we can use the data quantitatively to provide support for the identification assumption. We will discuss this briefly in Section 3. Efforts to promote SSEs may have impeded overall manufacturing growth. Between 1980 and 1998, the small-scale sector grew at a slower pace than the large sector (a 6% annual growth rate as compared to 9%, Mohan 2002). The limit on capital accumulation may have forced firms to under-invest in machinery or technology. For example, a shirt producer's minimum efficient scale is five hundred sewing machines, and factories of this size are common in countries like China and Sri Lanka. However, in India, the average non-exporting factory has only twenty machines (Garcia-Santana and Pijoan-Mas 2014). Our summary statistics in Table 1 show that units which produce reserved goods are smaller in terms of output, employment, investment, capital to labor ratio, output per employee and product scope than units which do not produce reserved goods. Figure 2 plots the correlation of average industry productivity and the share of output which is not reserved. There is a strong and positive correlation (p-value is < .01) implying that in industries with fewer products subject to the manufacturing constraints, firms tend to be more productive on average. In a similar vein, Figure 3 shows the change in the average output of firms producing four different goods that were reserved: knit cotton garments, toothpaste, paper tubes and sleeves, and transformers. Before their year of dereservation, there is no apparent trend in output for these firms. Afterwards, they appear to expand their output. However, these are simply correlations and do not imply causality -- we cannot conclude that producing reserved products causes a firm to be less productive, or that an industry with more

10

unreserved output is more productive. It could be the case that other observed or unobserved factors cause a factory to both produce a certain type of good as well as affect its performance. For example, low ability entrepreneurs may decide to produce a reserved good to avoid competition, in addition to being more likely to produce it poorly. Our goal in this paper is to forge a causal connection between dereservation and industry outcomes, and to mitigate endogeneity concerns by exploiting the pseudo-experimental features of this policy change. 3. Methodology, Data and Summary Statistics Econometric Specification and Identification Assumption In order to estimate the effect of product dereservation on firm size, product addition, and productivity, we use a classic difference-in-differences strategy. This specification exploits the changes in the reserved product list over time. Specifically, we estimate ϕ in the following regression:

𝑌𝑖𝑠𝑡 = 𝛼𝑠 + 𝛾𝑡 + ϕ𝐷𝑠,𝑡 + β𝑋𝑖𝑠𝑡 + 𝜀𝑖𝑠𝑡 ,

(1)

where Yist is the economic outcome of interest, such as gross value added, employment, fixed capital, productivity, or product switching behavior of firm i of industry s in year t. The main independent variable of interest, 𝐷𝑠,𝑡 , is the percentage of output within the 3-digit industry that

is accounted for by unreserved products. We include industry fixed effects, 𝛼𝑠 , to control for

any industry-specific factors (such as common technologies used), and a year fixed effect, 𝛾𝑡 to control for any year-specific shocks common to all industries. 4

4

Since our dependent variable is binary for some of our outcomes, some have wondered why we use a linear probability model in favor of another binary model such as probit. Our main reason is that given the large number of fixed effects in our regression, using probit would induce incidental parameters bias. Notwithstanding this problem, we have run all our result with probit and our LPM results are similar to the probit marginal effects in every instance, both in significance and magnitude. These results are available upon request.

11

To get an unbiased estimate of ϕ it must be the case that there are no systematic trends in the

product dereservation in industries or characteristics related to our outcomes of interest. To add some confidence for this assumption, in Table 2, we show that measures of dereservation within an industry are uncorrelated with observable pre-policy characteristics of the industry. Specifically, we regress the fraction of the industry’s unreserved output on several of its past characteristics: the two year lag of productivity, employment, capital, output and capital to labor ratio, as well as the output to labor ratio. 5 None of these variables successfully predict the fraction of unreserved output in an industry. This evidence lends further credibility to our argument that products were not chosen to be dereserved in industries which were observably different in size or productivity. Data and Construction of Key Variables The main dataset we use is the Annual Survey of Industries (ASI), the principal source of industrial statistics in India, for the years 2000 to 2010. The ASI is an annual census of all registered manufacturing plants in India with more than one hundred workers, combined with a random sample of registered firm with less. 6 The data is collected for the financial year, which runs from April 1st-March 31st. In our analysis, we assign each plant-year observation the year which corresponds with the end of the financial year. For example, data from the 2007-2008 ASI are given a year value of 2008. Sampling weights are used to provide a nationally representative picture of industrial activity in India. In the past few years, panel identifiers for factories have become available to researchers, enabling tracking changes in factories over time. Only the larger census firms appear in the sample for all 11 years, while the other smaller firms rotate in and out depending on the annual sample.

5 6

Other lag structures do not change this finding. Appendix Table A1 shows the result for three year lags. This definition has changed slightly over time. More details at http://www.mospi.nic.in/stat_act_t3.htm

12

The ASI also contains information on a variety of plant characteristics, such as industrial classification, items produced, total output, wages, workers, investment and value added, among others. Industries are classified using the Indian National Industry Classification system, hereafter NIC. This classification system has been modified periodically, and so we used an official concordance to convert all industry codes to the NIC2004 version. The ASI data has certain limitations like incomplete coverage of factories, under-reporting of workers in small factories and under-reporting of value added. Due to the widespread use of the ASI, these problems are well understood and documented. 7 However, these limitations should not be systematically correlated with the timing and sequence of dereservation, and therefore our estimates of the effect of dereservation should remain unbiased. Our main variables of interest from the ASI are output, material input, labor, and capital. Outputs and inputs are deflated by creating an index using the Wholesale Price Index (WPI) from the Handbook of Industrial Statistics. Labor is defined as the total number of employees, and while capital is measured by deflating the book value of capital by the WPI for machinery. We drop closed firms and firms with missing, negative or zero values for any of the main variables of interest. In addition, we drop firms in industries which are not present for the entire duration of the sample. Using these main variables, we use two measures for productivity: gross value added per employee, and a measure of total factor productivity. 8 To measure TFP, we use a chain-linked,

7

See Bollard et al. (2013) for a more in-depth discussion of the structure and shortcomings of the ASI. Gross value added is defined as gross output less gross inputs. Gross output is defined to include the ex-factory value, (i.e., exclusive of taxes, duties, etc. on sale and inclusive of subsidies etc., if any) of products and by-products manufactured during the accounting year, and the net value of the semi-finished goods, work-in-process, (represents the excess/deficit of value of semifinished goods or work-in-process at the end of the accounting year over that of the beginning of the year plus net balance of semi-finished fixed assets on factory’s capital account) and also the receipts for industrial and non-industrial services rendered to others, value of semi-finished goods of last year sold in the current year and sale value of goods sold in the same condition as purchased. Gross value added is gross output minus material inputs. 8

13

index number method similar to that suggested by Aw, Chen and Roberts (2001). This index expresses each firm’s inputs and outputs as deviations from an industry-specific representative firm in the base year (2000 in our sample). This hypothetical firm has input revenue shares equal to the average revenue shares in an industry, and input levels equal to the average of the log of the inputs. These averages are chain-linked for each year to account for shifts over time in these variables. The index, therefore, provides a measure of the difference in TFP for firm i in industry s in year t relative to the hypothetical firm in the year 2000 in industry s. Specifically: 𝑧 𝑡 𝑍 1 𝑧 ���� 𝑇𝐹𝑃𝑖𝑠𝑡 = (𝑞𝑖𝑠𝑡 − 𝑞 ����) ���� �������) ����) 𝑠𝑡 + ∑𝑟=2(𝑞 𝑠𝑟 − 𝑞 𝑠𝑟−1 − ∑𝑧=1 2 �𝜔𝑖𝑠𝑡 + 𝜔𝑠𝑡 �(𝑧𝑖𝑠𝑡 − 𝑧 𝑠𝑡 +

1 𝑧 𝑧 ������� ����� ∑𝑡𝑟=2 ∑𝑍𝑧=1 �𝜔 ���� �������) 𝑠𝑟 + 𝜔𝑠𝑟−1 �(𝑧 𝑠𝑟 − 𝑧 𝑠𝑟−1 , 2

(2)

𝑧 where 𝑞𝑖𝑠𝑡 is the log of gross value added (output less material inputs), 𝜔𝑖𝑠𝑡 is the revenue share

of input 𝑧, and 𝑧𝑖𝑠𝑡 is the log of input 𝑧. As discussed previously our two inputs are capital and labor. Bars indicate average values of variables for each industry-year. The revenue share of

labor is calculated as the total wage bill (deflated by the WPI) given in the ASI over gross value added. The revenue share of capital is given as one minus the revenue share of labor. We drop a very small number of firms which reported total wage bills higher than gross value added so that revenue shares would be strictly between zero and one. We also drop firms in industries which have less than 15 observations per year in order to reduce the noise in the productivity measure. 9 Central to our analysis are the variables which capture net addition/ net drops of products. In order to construct these variables, we exploit the panel structure of the ASI. These variables are binary and are equal to one when a firm has more/less products in the current period than in the last period (whenever it was last observed) and zero otherwise. We construct three variants of 9

Our results are robust to the inclusion of these firms.

14

this dummy: i) when a firm has more/less products in this period than in the last unconditional on the type of products added, ii) when a firm has more/less reserved products (product is currently or was reserved in the past) in this period than in last, and iii) when a firms has more/less “regular” products (a product never subject to reservation) this period than in last. When there are more (less) products the add (drop) dummies get value of 1 and 0 if the opposite occurs or the number of products stay the same. These constitute our measure of overall add/drop, reserved product add/drop and unreserved product add/drop. To construct our dereservation measure, we use information on the reserved list and timing of dereservations since 1997 from the Ministry of Small Scale Industries. Whenever a dereservation occurs, an official government notification is issued providing the description and the 9-digit ASICC product code of the item. Since the ASI also provides ASICC information on what items a plant produces, we can link the government dereservation order to the ASI to calculate which plants produce reserved items in each year. Making this link is not straightforward because at some point in time, product codes used by the ASI and the government diverge. A key challenge was to make the full and correct concordance between ASI and government notification product codes, which we have been able to do with detailed scrutiny of each product and helpful input from the MSME ministry. Since the ASI data collection ends on March 31st of a given year, any product dereserved after this date is treated as reserved for the current ASI financial year, but dereserved for the following. We exclude observations where all product codes are missing or all of them are lumped in uninformative categories (such as “other”) that do not give any information about the reservation status of the product. We then calculate the fraction of output produced by all firms in a given industry and year which was not reserved, and use this as our measure of the policy.

15

In some of our regression specifications, we divide the sample into different groups based on the production characteristics of the firm.

Since we are especially interested in the effect of

product addition and dropping in multiproduct firms, in many cases we estimate the effect of dereservation separately for multiproduct and single-product firms. In addition, we define a firm to be “reserved” if we ever observe them producing any product which is classified as reserved while they are producing it. 10 We then estimate the effect of dereservation on separately for “reserved” and “unreserved” firms in some instances. Summary Statistics Table 1 provides summary statistics for our sample, both for the overall sample and two sets of subsamples. The first subsample we call “Multiproduct Status”, and is divided between single-product firms and multiproduct firms. The second subsample division we call “Reservation Status”, and splitting the sample between reserved and unreserved firms. There are a few observations of note from this table. First, multiproduct firms tend to be larger than single-product firms in every size category, with 51.8% higher employment, 161% higher gross value added, and 228% higher fixed capital. In addition, it appears that labor productivity is higher in multiproduct firms compared to single product firms: they have higher gross value added per worker (32.8% higher) and more fixed capital per worker (24.5% higher). They are also much more likely to be net adders than net droppers (37.8% vs. 16.0%). While multiproduct firms and single product firms are roughly equal in our sample, (42.7% vs. 57.3%), there are 5.5 firms which have only produced unreserved products for every firm which has ever produced a reserved product. Firms which produce reserved products also are

10

For example, if a firm produces good A in 2000, and good A is eventually dereserved in 2005, we classify the firm as a “reserved” firm. However, a firm which begins to produce the same good A in 2007 will be considered “unreserved” since, although good A was reserved at one time, the firm never produced good A while it was reserved.

16

much smaller because of the reservation policy. For example, unreserved firms have 161% higher output than reserved firms, have 49.4% higher employment and 255% more fixed capital. Similar to multiproduct firms, they also have higher labor productivity: Gross value added per worker is 55.6% higher and capital per worker is 59% higher. Unreserved firms also experience higher rates of product churning: 32.3% of unreserved firms change products across observations in our sample (15.9% are net adders while 16.4% are net droppers), compared with 20.6% of reserved firms. 4. Results In Table 3 we estimate our difference in differences equation (1) to find the causal effect of dereservation on a series of size and productivity measures. For each variable, we show the results for the entire sample, as well as the single-product and multiproduct subsamples. We find that the dereservation policy led to increases in size. For example, since between 2000 and 2010 the fraction of unreserved output rose from 86.0% to 98.5%, we estimate that dereservation increased firm-level gross value added by 12.5% times 0.366 which is 4.6%.

Similarly, we find

that the policy increased firm-level employment by 5.0% and capital by 3.0%. This estimate is in line with other estimates of the effect of dereservation on output as in Martin, Nataraj and Harrison (2014) and Garcia-Santana and Pijoan-Mas (2014). Interestingly, we do not find that dereservation led to increases in productivity in the entire sample. However, looking at the single product and multiproduct subsamples unmasks interesting heterogeneity in the effect of dereservation on productivity by firm type. While single product firms did not experience an increase in productivity, multiproduct firms did. We find that the policy change resulted in a 4.4% increase in firm-level total factor productivity for multiproduct firms. 17

We also find that, with the exception of capital, multiproduct firms experienced larger increases in size as a result of the policy. For example, while gross value added in singleproduct firms grew by 3.3% as a result of the policy (.125*.267), in multiproduct firms it grew by almost double (6.3%). Growth in employment grew by 3.6% and 7.4% respectively. Capital does not follow this pattern: growth in capital in multiproduct firms is a statistically insignificant 1.6%, in contrast with a 4.3% increase in single product firms. This disproportionate expansion of capital by single product firms is perhaps not surprising since they are more likely to be producing reserved goods. Since the policy directly restricted capital itself, it is more likely that this was a particularly important margin of expansion for these capital-constrained producers. In Table 4 onwards we explore one particular channel that may be underlying these changes in size and productivity—addition and dropping of products. We investigate this mechanism by dividing the sample further to test whether the effect of dereservation is also heterogeneous not only on the firm type, but also on the mix of goods a firm produces. In Table 4 we find that the increases in productivity are solely driven by multiproduct firms which have never produced a reserved product. This is notable for three reasons. First, it seems to suggest that our results are not driven by the growth of small reserved firms which are no longer capital constrained by the policy. Second, these results are consistent with the story of large multiproduct firms (which cannot produce reserved goods due to the policy) expanding into the production of formerly reserved goods after the policy is lifted. Third, since we find no productivity growth in singleproduct firms which do not add products (and hence become multiproduct firms), this further suggests that product churning may drive the productivity results. To explore this further, we restrict our sample to firms which are only net adders in column (5) of Table 4. We find large and significant effects on productivity. Restricting to firms which

18

are net droppers in column (6), we also find an economically large effect, but it is not statistically significant. In column (7) we find no effect on firms which do not add or drop products. These results further point to product churning as a source of productivity change – firms which do not change their product lines do not increase their productivity as a result of the policy. In Table 5, we look at the effect of dereservation on product churning directly. We define two indicator variables – one for being a net adder of products, and another for being a net dropper of products, and test whether dereservation has changed the probability of being an adder or dropper. Panel A shows our results, by the reserved status of the product, for the entire sample, while Panel B shows the results for multiproduct firms only. From columns (1) and (2) in Panel A of Table 5, it would appear that dereservation has had no effect on the probability of being an adder or dropper in the entire sample. However, once we redefine our net adding and dropping variables to be for reserved or unreserved products only, we find that the zero effect in the overall sample actually masks some interesting heterogeneity. From Columns (3) and (4), we find that reserved products are much more likely to both be added and dropped after dereservation. This is consistent with the story of product churning in the face of increased competition for products which have been dereserved: small unproductive firms sheltered by the policy drop the now unprofitable formerly reserved products, while larger more productive firms add them. For unreserved products in columns (5) and (6), we find that there is either no change or a decrease in churning. These results become even more pronounced when looking at only multiproduct firms. In Column (7) of Panel B in Table 6, we find that firms are much more likely to add products after dereservation. Interpreting the coefficient, we find dereservation led to a 4.9% increase in the probability a firm would be a net adder (0.148*.125=1.85% percentage point increase on a base

19

of 37.8%). In addition, product churning among reserved products is also much higher than in Panel A. Using a similar calculation, we find that the probability of becoming a net adder of reserved products increased by 63.6% and of being a net dropper by 54.5%. Never reserved products are also dropped as firms adjust their products—there is a 12.1% increase in dropping over the course of the policy. Next, we explore whether there are particular types of firms which are more or less likely to expand scope in response to dereservation. Some of the existing literature argues that an increase in competition will reduce scope for all firms (Bernard et al. 2011, Eckel and Neary 2010). Others argue that more productive firms can more easily overcome the fixed costs associated with changing products, expanding product scope (Qiu and Zhou 2013). In Table 6, we test whether the effect of dereservation is stronger for productive firms by interacting our unreserved percent variable with lagged productivity as in Berthou and Fontagné (2013). Panel A shows the effect of dereservation on product churning in the full sample, and Panel B shows the effect only among multiproduct firms. As before, we find that there is significant churning among reserved products, especially in multiproduct firms. The interaction between the unreserved percent and lagged productivity is also positive and highly significant in both samples, indicating that higher productivity firms are more likely to add and drop products as a result of dereservation. 5. Conclusion In this paper, we assessed the relationship between product scope and productivity changes following policy reform. To do so, we exploited a potentially exogenous policy change which removed the restrictions on the production of certain products solely to SSEs – Indian product dereservation. Using data from the ASI to estimate a difference-in-difference estimation model, we found that the dereservation policy increase firm-level output by 4.6%, in line with other

20

estimates of the effect of product dereservation on output. We also provide evidence that productivity increased after dereservation, but only for multiproduct firms which expand into the formerly reserved sectors. Our evidence that dereservation increased product churning by multiproduct firms supports the idea that product adding and dropping is an important margin of within-firm reallocative activity following policy reform. Much of the current literature focuses on this mechanism in the context of globalization, but we provide novel evidence for product churning outside this typical trade setting in the aftermath of a policy that generally increased competitive dynamics. We also shed light on the question of whether more or less productive firms should be more likely to engage in product churning in the face of a change in competition. While we document that product churning primarily occurs among products which were formerly reserved in multiproduct firms, we also show that product churning is more intense in firms that are more productive. We hope this empirical evidence leads to incorporation of product scope dynamics in analyses of size-dependent policies which are widespread throughout the developed and developing world, and have sizable effects on productivity and growth.

References Aw, Bee Yan, Xiaomin Chen and Mark J. Roberts. 2001. “Firm-level evidence on productivity differentials and turnover in Taiwanese manufacturing.” Journal of Development Economics, 66(1): 51-86. Baldwin, J., & Gu, W. (2009). The impact of trade on plant scale, production-run length and diversification. In producer dynamics: New Evidence from micro data (pp. 557-592). University of Chicago Press. 21

Bernard, Andrew, Stephen Redding, and Peter Schott. 2011. "Multiproduct Firms and Trade Liberalization." The Quarterly Journal of Economics, 126(3): 1271-1318. Bernard, Andrew, Stephen Redding, and Peter Schott. 2010. "Multiple-Product Firms and Product Switching." American Economic Review, 100(1): 70-97. Bernard, A. B., Jensen, J. B., Redding, S. J., & Schott, P. K. (2012). The Empirics of Firm Heterogeneity and International Trade. Annu. Rev. Econ., 4(1), 283-313. Berthou, A., & Fontagné, L. (2013). How do Multiproduct Exporters React to a Change in Trade Costs?. The Scandinavian Journal of Economics, 115(2), 326-353. Bollard, Albert, Peter Klenow, and Gunjan Sharma. 2013. "India's Mysterious Manufacturing Miracle." Review of Economic Dynamics, 16(1): 59-85. Doraszelski, U., & Jaumandreu, J. (2013). R&D and productivity: Estimating endogenous productivity. The Review of Economic Studies, 80(4), 1338-1383. Eckel, C., & Neary, J. P. (2010). Multi-product firms and flexible manufacturing in the global economy. The Review of Economic Studies, 77(1), 188-217. Feenstra, R., & Ma, H. (2007). Optimal choice of product scope for multiproduct firms under monopolistic competition (No. w13703). National Bureau of Economic Research. García-Santana, M., & Pijoan-Mas, J. (2014). The reservation laws in india and the misallocation of production factors. Journal of Monetary Economics, 66, 193-209. Garicano, Luis, Claire Lelarge, and John Van Reenen. (2012). "Firm Size Distortions and the Productivity Distribution: Evidence from France." CEP Discussion Papers dp1128, Centre for Economic Performance, LSE.

22

Goldberg, Pinelopi, Amit Khandelwal, Nina Pavcnik, and Petia Topalova. 2010. "Multiproduct Firms and Product Turnover in the Developing World: Evidence from India." The Review of Economics and Statistics, 92(4): 1042-1049. Guner, Nezih, Gustavo Ventura, and Xu Yi. 2008. "Macroeconomic Implications of SizeDependent Policies." Review of Economic Dynamics, 11(4): 721-744. Hsieh, Chang-Tai and Peter J. Klenow. 2009. “Misallocation and manufacturing TFP in China and India.” Quarterly Journal of Economics, 124(4): 1403-1448. Iacovone, L., & Javorcik, B. S. (2010). Multi‐Product Exporters: Product Churning, Uncertainty and Export Discoveries. The Economic Journal, 120 (544), 481-499. Little, Ian. 1987. “Small Manufacturing Enterprises in Developing Countries.” World Bank Economic Review, 1(2): 203-236. Martin, L., Nataraj, S., & Harrison, A. (2014). In with the Big, Out with the Small: Removing Small-Scale Reservations in India (No. w19942). National Bureau of Economic Research. Mayer, T., Melitz, M. J., & Ottaviano, G. I. (2011). Market size, competition, and the product mix of exporters (No. w16959). National Bureau of Economic Research. Mohan, Rakesh. (2002). “Small-Scale Industry Policy in India: a Critical Evaluation.” In: Krueger, A. O. (Ed.), Economic Policy Reforms and the Indian Economy. Chicago: University of Chicago Press. Nocke, V., & Yeaple, S. (2006). Globalization and endogenous firm scope (No. w12322). National Bureau of Economic Research.

23

Qiu, L. D., & Zhou, W. (2013). Multiproduct firms and scope adjustment in globalization. Journal of International Economics, 91(1), 142-153. Restuccia, Diego, and Richard Rogerson. 2008. "Policy Distortions and Aggregate Productivity with Heterogeneous Plants," Review of Economic Dynamics, 11(4): 707-720.

24

Figure 1(a): Fraction of Output Unreserved by Year: 2000-2010 100%

Percent Unreserved

95%

90%

85%

80% 2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

Year

Note: The figure shows the share of output over our sample period in a three-digit industry constituted of products that were never under reservation or products that became dereserved. .

25

Figure 1(b): Fraction of Output Unreserved by Industry: 2000-2010 100%

Percent Unreserved

95%

90%

85%

80%

75%

70%

Note: The figure shows the aggregate share of output in a three-digit industry that is constituted of products that were never under reservation or products that became dereserved.

26

-1

-.5

0

.5

1

Figure 2: Average Industry Productivity and Unreserved Share

0

.2

.8

.6

.4

1

unrespct (mean) acr_tfp

Fitted values

Note: The scatter plot shows the correlation between an industry’s unreserved output share and average firm productivity across the sample period. The correlation is .1746, which is significant at