Unbundling India’s Productivity Performance

(1)

UNIVERSITY OF GRONINGEN

Unbundling India’s

Productivity Performance

Jan Timothy Smith - S1894374

12/16/2015

Abstract

Keywords: Market distortions, Productivity, India

This paper investigates the presence of product and labor market distortions in India’s

(2)

1

Introduction

Central to development economics is the question of where differences in per capita income originate. A broadly based perspective views the primary source of these income differences to be productivity, commonly measured as Total Factor Productivity (TFP) (Hall and Jones, 1999). Naturally, the next question to ask is what drives differences in TFP between countries.

An important part of research on TFP has focused on the hampering effects of market distortions. Usually in the form of corruption and misguided government policies, market distortions can create imperfect competition. The resulting misallocation of resources

typically creates inefficiencies that can lead to lower TFP (Hsiew and Klenow, 2009; Inklaar et. Al, 2014).

The present article studies the role of market distortions in the development of India’s private economy, with a special focus on the manufacturing sector. There are several reasons why India is interesting to study in this context. First, after decades of slow growth after independence, India has witnessed a spectacular increase in growth during the last three decades. From 1980 until 2005, India’s aggregate real output growth averaged an impressive 5.8% per year (Verma, 2011) and GDP per capita more than doubled (Bosworth and Collins, 2008).

Second, the pattern of India’s rapid growth process has been unusual and is not yet fully understood by economists. Unlike in China and the Asian tigers, the structural

transformation process in India has bypassed the traditional step from being an agricultural economy to being an industrial economy; despite serious output growth, the manufacturing sector has never become India’s most important sector (Bollard, Klenow and Sharma, 2013). Manufacturing output as a share of total output has never been larger than 20% while the service sector now adds more than 50% of the value added in India (Bosworth and Collins, 2008).

Third, India has long had a range of protectionist government policies that are

suspected to have created severe market distortions in the manufacturing sector. Despite two large rounds of liberating market reforms in the 1980s and early 1990s there are still

(4)

3 this endeavor have been mixed however, leading Bollard, Klenow and Sharma (2013) to declare the growth in the Indian manufacturing Industry to be of ‘mysterious origin’.

The reason for these mixed results might be found in research on the effects of

distortions on TFP measurement. A set of studies that started with Hall (1988) brought to light that besides hampering productivity growth, market distortions can also bias the measurement of TFP. Normally, TFP is estimated through growth accounting. Under the assumption of perfect competition in product and factor markets and assuming constant returns to scale, growth accounting measures TFP as a residual; total output growth less the share-weighted averages of the growth rates of factor inputs (Solow, 1957). However, Hall (1988) showed that under imperfect competition in product markets the TFP measure would be

overestimated. To account for this bias, he adjusted the growth accounting model by adding a component that captures the markup of price over marginal cost.

Crépon, Desplatz and Mairesse (2005) in their turn showed that Hall’s extended model is still biased under imperfect competition in labor markets. Then, workers can, through their bargaining position, capture a part of the price-markup in the form of higher wages. If not accounted for, this ‘rent-sharing’ causes an underestimation of the markup, and thus a downward bias in the TFP measure. To correct for these two interacting biases, Crépon, Desplatz and Mairesse (2005) added an additional component to the model that accounts for the amount of rent-sharing in the economy. Their framework has given rise to a generation of other papers, studying the economies of France (Dobbelaere and Mairesse, 2013), Japan, Belgium, The Netherlands (Dobbelaere, Kiyota and Mairesse, 2015) and Chile (Benavente, Dobbelaere and Mairesse, 2009) The present study uses this framework to estimate the distortions in products and labor markets in India’s manufacturing sector.

(5)

4 This paper adds to the literature by answering several questions. First, the main aim of this paper is to establish whether distortions in products and labor markets are indeed present in India’s manufacturing sector. Second, we try to quantify these distortions and investigate whether they have changed over time; especially the extent to which they have been affected by the market reforms of the early 1990s. Third, using growth accounting, a large literature has found significant TFP growth in both the manufacturing and services sector (Bosworth and Collins, 2008; Bollard, Klenow and Sharma, 2013). We use the model to calculate an adjusted, less biased, measure of TFP that accounts for the product and labor market distortions.

In the following sections we will discuss (I) the literature on distortions in the product and labor markets in India, (II) the framework used to account for those distortions, (III) the data and methodology and (IV) the results. We will end with a discussion that places the results in the context of the literature on distortions in India and a some concluding remarks.

Literature Review

Product market distortions

The presence of distorted product markets in the Indian manufacturing industry and the resulting misallocation of resources have been well-established (Hsieh and Klenow, 2009; Bhalotra, 1998). Resource misallocations can create two sorts of inefficiencies. One is within-firm inefficiency, where the level of inefficiency in a representative within-firm differs from country to country (Hsiew and Klenow, 2009). Alternatively, resource misallocations can result in between-firm inefficiencies, where the level of efficiency differs between firms within the same country or industry (Restuccia and Rogerson, 2008). Due to possibly more severe

market imperfections, this could be especially true in developing countries. Indeed, Hsiew and Klenow (2009) have found much larger differences in firm efficiency levels in China and India than in the United States. Hsiew and Klenow (2009) even estimated that TFP would be 40%-60% higher in India and 30%-50% higher in China, if capital and labor would be

reallocated so that they would have the same level off variance in efficiency between firms as the United States.

(6)

5 Small-Scale Reservations Laws (SSRL). The SSRL restricted firm size in a significant

amount of industries that the government decided to keep in the hands of small firms. A second source of market distortions were extensive trade barriers in the form of tariffs and restricted FDI. While the License Raj and the SSRL legislation mainly affected the

manufacturing sector, the trade barriers affected the whole economy.

There have been two large rounds of reforms in Indian manufacturing policies. First there was a round of general pro-business reforms in the 1980s; price controls were removed and corporate taxes were reduced. These reforms favored incumbent firms. Next, in the early 1990s, under pressure from the IMF there was a round of reforms that specifically targeted India’s aforementioned industrial policies. The License Raj was abolished, a gradual lifting of SSLR started, tariffs were reduced and FDI became more liberalized. These reforms were pro-market and favored new entrants (Rodrik and Subramanian, 2004). The three policies are suspected to have created large product market distortions in the Indian manufacturing sector, with effects of them still lingering today. In the following section, the License Raj, SSRL and trade barriers will be discussed in relation to market distortions.

The License Raj was introduced in 1956. Under the License Raj, firms that wanted to produce a good needed a special permit to do so. The permit decided how much output a firm was allowed to have, which technology and machinery it could use and where it should be based. The license Raj has affected the incentives and abilities of the Indian manufacturing sector in various ways (Bhalotra, 1998). First, it has undermined TFP growth by protecting high cost firms and by undermining competition (Kathuria, 2013). Second, since the License Raj decided how much capital was allowed, it prohibited successful entrepreneurs from investing and thus decreased investments in capital. Third, it gave politicians the power to decide what a firm was allowed produce. This gives rise serious worries about corruption (Sharma, 2008). The license Raj was abolished in the 1990s.

The Small Scale Reservation Laws (SSRL) partly remained after the abolishment of the License Raj in the early 1990s. It reserved a number of industries for small-scale firms. At its highest point, more than 1000 industries fell under the restrictions from the SSRL and after the 1990s pro-market reforms, this number was still 830 (Williamson and Zagha, 2002). The dismantling of the SSRL happened very gradually and is still underway. The SSRL limited the scale of production for the better entrepreneurs, and diminished labor demand. García-Santana and Pijoan-Mas, (2014) argue that abolishing the SSRL would increase productivity by incentivizing capital deepening and by improving the allocation of production factors.

(7)

6 average firm-size in manufacturing increased by 10% and manufacturing output and TFP increased by 6.8% and 2%. Alfaro and Chari (2014) found a decrease of average firm size. However, this came as a result of an increase in the entry of new small firms; incumbent firms grew significantly after regulation and the largest firms kept dominating the industry (Alfaro and Chari, 2014). Alfaro and Chari (2014) interpreted this as an indication of decreased distortions.

Third and final, trade barriers. These consisted mainly of tariffs and restrictions on FDI. Trade reforms can create pressure on the price-cost margin by allowing foreign competitors to enter the market. The possibility that new firms can enter increases

competition and may reduce the price-cost margin of the firms. This foreign competition can also create stronger incentives for managers to innovate and adopt new technologies,

increasing TFP growth. Finally, trade reforms can improve productivity by giving firms the availability of better inputs (Sharma, 2008). The trade restrictions were largely diminished in the 1990s reform.

Looking into the effects of trade liberalization on the manufacturing industry, several authors found that liberalization created more competitive pressure for incumbent firms (Das, 2014; Topalova and Khandelwal, 2010). Harrison, Martin and Nataraj (2011) found that the lifting of the FDI restrictions increased firm entry and exit and improvements in within-firm productivity.

(8)

7 growth.

Additionally, Harrison, Martin and Nataraj (2011) rapported that aggregate output-weighted productivity growth already starts to increase after 1985, so before the 1990s reforms. Unel (2003) as well, found TFP in manufacturing to be higher during the 1980s than during the two decades before.

The results of the effects of the 1990s reforms had a more clear cut impact; generally TFP growth increased. First, several authors managed to link significant improvements in productivity to the lifting of the trade restrictions (Das, 2014; Topalova and Khandelwal, 2010; Sivadasan, 2006). Sivadasan (2006) estimated that the mean productivity per plant increased with 16% due to the liberalization of the FDI regulation and with 15.6% due to the liberalization of tariffs. Harrison, Martin and Nataraj (2011) found the increases in

productivity due to the liberalization of FDI and input tariffs to be 21.8% and 3.2%. The biggest drivers of this effect were increased firm entry and exit and improvements in within-firm productivity. Second, Sharma finds that the deregulation of the License Raj in the 1990s had a positive impact on TFP, labor productivity, output growth and industry structure

(Sharma, 2008). Finally Unel (2003) and Kathuria (2013) both find that TFP growth increased after the 1991 reforms. Kathuria (2013) attributes the increase in growth to process

technology acquisition, efficient utilization of resources and infrastructure development resulted in an increase in TFP growth.

(9)

8 re-allocation.

It is important to further investigate the question of how market distortions have developed after the 1990s reforms and how they have influenced measures of TFP. We aim to shed more light on the relationship between the reforms and post-1990s TFP growth by quantifying the distortions in the product market. Based on the literature, we state the following two hypotheses.

Hypothesis 1: There are significant distortions in the product markets of India’s

manufacturing sector.

Hypothesis 2: The distortions in the product markets of India’s manufacturing sector

have become less severe after the 1990s pro-market reforms.

Labor market distortions

In a globalized world, flexibility in labor use is important for competing with other businesses. Rigid labor laws can thus undermine competitiveness and growth (Bhandari and Heshmati, 2008). India’s labor laws are often viewed as very rigid. Indian workers are relatively well protected by the India’s labor laws. The Trade Unions Act of 1926 gives workers the right to form unions and engage in collective bargaining. The right to join a trade union is also secured in India’s constitution from 1950 (Nishith Desai Associates, 2015). Most research on distortions in the labor market has focused on the Industrial Disputes Act (IDA) from 1947. IDA protects and regulates the rights of employers and employees in case of disputes in the manufacturing sector. The IDA also protects workers against random firing. Since 1982, firms in the manufacturing sector with more than 100 workers can only fire workers with permission from the government. (D’Souza, 2010). These laws might increase the bargaining position from the Indian laborers.

(10)

9 growing dissociation in Indian manufacturing between output growth and employment

generation (Dutta Roy, 2003; Thomas, 2013; Bollard, Klenow and Sharma, 2013). The second particularity is the growth of India’s already relatively large informal sector. To evade labor market rigidities, the formal manufacturing sector outsources to the less productive informal sector (Erumban, Das and Aggarwal, 2012). Subcontracting has increased significantly since the 1990s (Basole, Basu and Bhattacharya, 2015). In 2005, 30% of the 17 million informal manufacturing firms worked on a subcontract (Erumban, Das and Aggarwal, 2012). In India, the differences in productivity between the formal and informal sector are large. The policy distortions in India are seen as an influential factor in maintaining and increasing these differences in productivity (Kathuria, Raj and Sen, 2013). De Vries, Erumban, Timmer, Voskoboynikov and Wua (2011) even found that in India, the positive effect on aggregate productivity growth by reallocation of labor across sectors disappeared when they made a distinction between the formal and informal sector. This was due to growth of the relatively labor intensive, unproductive informal sector.

Several authors have tried to link these patterns in employment growth to India’s labor laws but the results are mixed. Fallon and Lucas (1993) used the 1976 amendment of the 1947 Industrial Dispute Act of India to study its effect on labor demand. They argue that rigid labor regulations impose extra costs on adjusting the level of employment to changing demand. Furthermore, the rigid labor regulations increase the cost of having employees, which would lead to lower employment by the firm. They indeed find a substantial reduction in labor demand in firms that are affected by the new regulations. This is in line with the results from Bhalotra (1998) and Dutta Roy (2003), who find employment adjustment in the Indian labor market to be relatively rigid. They both find that on average it costs approximately five to six years to adjust employment to changes in demand. Bhalotra (1998) however, finds that these inertia effects already existed before the 1976 amendment. A remark here is that Fallon and Lucas’ (1993) methodology has been extensively criticized and their effects were inconsistent from state to state (D’Souza, 2010).

(11)

10 2010). Besley and Burgess (2004) themselves also find that labor regulation is no longer able to significantly explain change in formal manufacturing output and employment when they add a time trend variable. Additionally, they acknowledge that the effects they find might be due to a poor climate of labor relations, rather than due to labor regulations. However, they leave this open to interpretation.

Finally, Cai and Pandey (2013) find evidence that the labor regulations distort the labor market. They find that aggregate output per worker would increase by 2.3% if the regulations would be removed. Furthermore, without the regulations, the share of

employment in mid-sized companies would increase from 11% to 37%. This would leave the structure of the manufacturing sector in India quite similar to that of Korea, who does not have the same sort of labor regulations.

Thomas (2013) and D’Souza (2010) however, argue that the influence of the labor laws on formal manufacturing employment is overstated, especially since the 1980s pro-business and the 1990s pro-market reforms. They bring forth several arguments. First, the job security laws only apply to a small percentage of the manufacturing industry because 90% of the employees in manufacturing work in the informal sector. Here they are not protected by the labor laws (Thomas, 2013). Second, since the 1980s and especially after the 1990s employment has grown relatively fast in some years and has decreased in other years. For example, in the period 1986 – 1993, employment grew significantly with 2.1% (Bhalotra, 1998). This volatility could be an indication of a trend of increasing labor flexibility in the past thirty years.

A possible explanation for this increasing volatility could be found in research from D’Souza (2010). He finds that even though the regulations that protect workers are still in place, firms receive what he calls ‘tacit state support’ as the government weakly enforces the labor laws. The firms have been allowed to change to more flexible labor contracts, creating a situation where there is job security for the already employed, but not for the new contract workers. This is in line with Bhandari and Heshmati (2008) and Thomas (2013), who both found that after the 1990s reforms, India’s manufacturing industries have seen a rise in contract workers and a decline in permanent workers. These contract workers earn lower wages despite having similar labor productivity (Thomas, 2013).

(12)

11 period 1980 - 2008. Furthermore, the influence of these distortions on TFP will be quantified. The main thesis of the paper is that former estimates of TFP growth in India have been too optimistic, due to a failure to incorporate the distortions in both product and labor markets. Based on the literature, we state the following three hypotheses.

Hypothesis 3: There are significant distortions in the labor market of India’s

manufacturing sector.

Hypothesis 4: The distortions in the labor markets of India’s manufacturing sector

have become less severe after the 1990s pro-market reforms.

In the following section the background and intuition behind the framework that we use to estimate the distortions and the adjusted ‘purer’ TFP will be explained in more detail.

Empirical Implementation

The model

The framework we use is a generalized growth accounting model from a paper by Crépon, Desplatz and Mairesse (2005). They build on a model by Hall (1988), who already extended the standard growth accounting method to include distortions in the product markets. In 2005, Crépon, Desplatz and Mairesse in their turn extended Hall’s model by incorporating distortions in the labor market. It is this framework that we use to quantify distortions in product and labor markets, and their influence on the standard measure of TFP in India. In the following section we show how first Hall (1988) and then Crépon, Desplatz and Mairesse (2005) have built up this framework.

Hall (1988) started from a standard growth accounting equation of firm i in time t that operates in a competitive market. The logarithmic specification of this standard production function expresses the total output growth qit in the share-weighted averages of the growth

rates of capital kit, labor lit, intermediate inputs mit.

(1) ∆𝑞_𝑖𝑡 = ∆𝑎_𝑖𝑡+ 𝜀_𝑖𝑡𝐾∆𝑘_𝑖𝑡+ 𝜀_𝑖𝑡𝐿∆𝑙_𝑖𝑡+ 𝜀_𝑖𝑡𝑀∆𝑚_𝑖𝑡

(13)

12 markets, he argues firms charge a markup 𝜇, which can be calculated by dividing the firm’s price by its marginal cost: 𝜇 = 𝑃𝑖𝑡

𝐶_𝑖𝑡𝑄

.

Failing to incorporate this markup 𝜇, or price-cost margin in the growth accounting function, would result in a biased estimation of TFP. Next, Hall assumes that labor and intermediate inputs are variable factors, capital is quasi-fixed and firms maximize profits. Under these assumptions the elasticities of output with regard to the labor and intermediate inputs can be derived as follows:

(2) 𝜀_𝑖𝑡𝐿 = 𝜇_𝑖𝑡𝑠_𝑖𝑡𝐿 (3) 𝜀_𝑖𝑡𝑀 = 𝜇_𝑖𝑡𝑠_𝑖𝑡𝑀

where 𝑠_𝑖𝑡𝑋 (L, M) is the share of input factor X in total revenue. These shares are computed as the averages over adjacent years. It is assumed that 𝑠_𝑖𝑡𝐿 is equal to the marginal revenue of labor. This means that theelasticity of output with regard to labor 𝜀_𝑖𝑡𝐿 is thus equal to the marginal revenue of labor, multiplied by the markup 𝜇. Next, using (2) and (3) and assuming that the elasticity of scale is 𝜆𝑖𝑡 = 𝜀𝑖𝑡𝐾+ 𝜀𝑖𝑡𝐿 + 𝜀𝑖𝑡𝑀, the elasticity of output with regards to

capital can be expressed as:

(4) 𝜀_𝑖𝑡𝐾 = 𝜆_𝑖𝑡− 𝜇_𝑖𝑡𝑠_𝑖𝑡𝐿 − 𝜇_𝑖𝑡𝑠_𝑖𝑡𝑀 Inserting (2), (3) and (4) into (1) and rearranging the terms gives:

(5) ∆𝑞𝑖𝑡 = 𝜇𝑖𝑡(𝑠𝑖𝑡𝐿(∆𝑙𝑖𝑡− ∆𝑘𝑖𝑡) + 𝑠𝑖𝑡𝑀(∆𝑚𝑖𝑡− ∆𝑘𝑖𝑡)) + 𝜆𝑖𝑡∆𝑘𝑖𝑡+ ∆𝑎𝑖𝑡

which can be rewritten as:

(6) ∆𝑇𝐹𝑃𝑖𝑡 = (𝜇𝑖𝑡− 1) (𝑠𝑖𝑡𝐿(∆𝑙𝑖𝑡− ∆𝑘𝑖𝑡) + 𝑠𝑖𝑡𝑀(∆𝑚𝑖𝑡− ∆𝑘𝑖𝑡)) + (𝜆𝑖𝑡− 1)∆𝑘𝑖𝑡+ ∆𝑢𝑖𝑡

where ∆𝑇𝐹𝑃𝑖𝑡 is TFP as traditionally measured. Finally, by replacing (𝑠𝑖𝑡𝐿(∆𝑙𝑖𝑡− ∆𝑘𝑖𝑡) +

𝑠𝑖𝑡𝑀(∆𝑚𝑖𝑡− ∆𝑘𝑖𝑡)) by ∆𝑥𝑖𝑡 𝜇

, and assuming a constant 𝜇 and 𝜆, and Hall (1988) arrives at: (7) ∆𝑻𝑭𝑷_𝒊𝒕 = (𝝁 − 𝟏) ∆𝒙_𝒊𝒕𝝁 + (𝝀 − 𝟏)∆𝒌_𝒊𝒕+ ∆𝒖_𝒊𝒕

where the traditional measures of TFP (∆𝑇𝐹𝑃_𝑖𝑡) is decomposed into three components: (𝜇 − 1) ∆𝑥_𝑖𝑡𝜇 captures the markup 𝜇; (𝜆 − 1)∆𝑘_𝑖𝑡 captures the economies of scale 𝜆; and ∆𝑢_𝑖𝑡 is the residual that captures the adjusted, less biased measure of TFP (Hall, 1988).

(14)

13 when there is imperfect competition in labor markets. The reason is that in uncompetitive labor markets, laborers have bargaining power that they can use to extract a part of the extra rent that firms earn through their markup. Crépon, Desplatz and Mairesse (2005) assume that wages are determined contractually according to an efficient bargaining model. The amount of the rent they can extract depends directly on the laborers’ bargaining power 𝜃, with 𝜃 being a parameter between 0 (no bargaining power) and 1 (full bargaining power). This

‘rent-sharing’ gives the laborers an extra compensation in the form of higher wages. Their wages thus now compose of two parts: the wage they would have gotten had labor markets been competitive (the reservation wage), plus a share of the firm’s short run profit. This increases the share of labor in total revenue, 𝑠_𝑖𝑡𝐿, and decreases profits for the firm. Failing to account for this will bias the markup 𝜇 upwards, which would result in an underestimation of TFP. Therefore, Crépon, Desplatz and Mairesse (2005) assume an efficient bargaining model where both the wage and employment are bargained for between firms and their employees (Mc Donald and Solow, 1981). In this model, wage will lie somewhere between the marginal revenue of labor 𝑅𝐿 and the marginal revenue of labor when the intermediate inputs are paid for: (R-jM)/L, where j is the price of the intermediate inputs.

Using the shares of labor and materials 𝑠_𝑖𝑡𝐿 and 𝑠_𝑖𝑡𝑀, the elasticity of labor 𝜀_𝑖𝑡𝐿 and the markup 𝜇, Crépon, Desplatz and Mairesse (2005) incorporate the bargaining power 𝜃 into a new definition of 𝑠_𝑖𝑡𝐿:

(8) 𝑠_𝑖𝑡𝐿 = 𝜃_𝑖𝑡 (1 − 𝑠_𝑖𝑡𝑀) + (1 − 𝜃_𝑖𝑡) 1

𝜇𝑖𝑡𝜀𝑖𝑡

𝐿

By replacing (2) by (8), and by substituting (𝑠𝑖𝑡𝐿 + 𝑠𝑖𝑡𝑀− 1)(∆𝑙𝑖𝑡− ∆𝑘𝑖𝑡) for ∆𝑥𝑖𝑡𝜃, (7) becomes:

(9) ∆𝑻𝑭𝑷_𝒊𝒕 = 𝝆∆𝒙_𝒊𝒕𝝁 + 𝜸∆𝒌_𝒊𝒕+ 𝝎∆𝒙_𝒊𝒕𝜽 + ∆𝒖_𝒊𝒕 Where 𝜌 = (𝜇 − 1), 𝛾 = (𝜆 − 1) and𝜔 = 𝜇 𝜃

1−𝜃. The model decomposes the standard

measure of TFP (∆𝑇𝐹𝑃𝑖𝑡) into four, instead of three,components: (𝜇𝑖𝑡− 1)∆𝑥𝑖𝑡 𝜇

captures the markup 𝜇; (𝜆_𝑖𝑡− 1)∆𝑘_𝑖𝑡 captures the economies of scale 𝜆; and 𝜇 𝜃

1−𝜃∆𝑥𝑖𝑡

𝜃_{is a new term that}

captures the rent-sharing. Finally, ∆𝑢_𝑖𝑡 is the residual, that now captures the ‘pure’ measure of TFP, adjusted for distortions in both labor and product markets (Crépon, Desplatz and

Mairesse, 2005).

(15)

14 and 𝛾 > 0. The expectation for 𝜔 in India is more ambiguous. Although Fikkert and Hassan (1998) found constant returns to scale (CRS), Hulten and Srinivasan (1999) found evidence of diminishing returns to scale (DRTS). In case of DRTS, TFP might be biased downwards because the value of output would be greater than the marginal productivity of the inputs. Under CSR we expect that 𝜔 = 0, and under decreasing returns to scale we expect that 𝜔 < 0. The final step is to calculate the parameters for the markup 𝜇, the rent-sharing 𝜃, and the economies of scale 𝜆. This can be easily done by inserting the coefficients 𝜌, 𝛾 and 𝜔 into the model. To find the value of the parameter for the markup 𝜇, this means solving the

equation 𝜌 = (𝜇 − 1) for 𝜇. For the economies of scale 𝜆 this means solving the equation 𝛾 (𝜆 − 1) for 𝜆. Finally, when the markup 𝜇 is calculated, the estimate for the rent-sharing 𝜃 can be obtained by solving the equation 𝜔 = 𝜇 𝜃

1−𝜃 for 𝜃.

The model has several strengths (Crépon, Desplatz and Mairesse, 2005). The first is that the model does not only simultaneously estimate the distortions in both the product and labor markets, it also captures the interaction effect between the two. This is important because the labor market distortion is expected to bias the price markup downwards if not accounted for. Another advantage of the model is that it does not require to measure the user cost of capital to assess the magnitude of the markups. Also, it is not necessary to estimate the reservation wage. It is thus a more direct way of estimating both the mark-ups and the rent-sharing than other methods.

Data

To estimate the extended growth accounting model we use panel industry data from India KLEMS. The India KLEMS data set is developed under the broader umbrella of World KLEMS, which supports internationally comparable growth accounting research on growth and productivity. World KLEMS stimulates the creation of internationally consistent data bases of detailed industry data on inputs, outputs and productivity. A KLEMS database typically contains comprehensive, disaggregated industry level panel data on gross outputs, value added and five different inputs; K-capital, L-labor, E-energy, M-materials, and S-purchased services (KLEMS). The data thus allows for measuring productivity in two ways: based on value-added and based on gross outputs.

(16)

15 to continue into developing countries. The data base from KLEMS India was released in 2014. India KLEMS contains data on gross output, value added, labor services, capital services and intermediate inputs. It covers the entire Indian economy, both the registered (formal) and unregistered (informal) sector, for a relatively long period, (1981-2012). We have left out the years (2009-2012), because the crisis might otherwise bias the results. We thus use data from the 28 years long period (1981-2008). Additionally, to investigate developments in distortions over time we split our data into two sub-periods. We follow the literature and use the pro-market reforms of 1991 as a divide, creating a data set for the pre-reform period (1981-1991) and a data set for the post-pre-reform period (1992-2008) (Rodrik and Subramanian, 2004).

Following KLEMS EU, KLEMS India divides the economy into 26 industries. Of these 26 industries, four are from the public sector and we leave those out of the study. The reason is that we use the markup 𝜇 to estimate market distortions and the government does not aim to make a profit. The remaining 22 industries are the total private sector. We split them up into three broader sectors: manufacturing, other goods producing, and services. This leaves us with a rich, balanced panel of 616 data points.

Generally spoken, the main difficulties in creating a data base for the Indian economy are the lack of National Accounts Statistics data on the unregistered (informal) sector and inconsistency of the data. The KLEMS India database goes back as far as 1981. Not all the variables have been measured in the same way over this whole period. This means that in some cases values had to be inferred based on other data. Next we shortly discus the measurement of the outputs and inputs.

The output and input measures for KLEMS India were constructed by integrating several official Indian data sources. We provide some details on the measurement and qualities of the available output and input measures.

(17)

16 intermediate inputs into account. KLEMS India distinguishes three main categories of

intermediate inputs: Energy input, Service input and Material input. The data considers five energy types and 14 sorts of service inputs. All the other inputs are classified as materials. The main data source for these inputs are the input-output tables. Both commodities that are produced nationally and commodities that are imported are accounted for. To construct the intermediate inputs against constant prices, industry specific deflators have been used. There were several data issues with regards to the measurement of the intermediate inputs. First, the input-output tables are not consistent over the whole period. Second, the input-output tables are created with five year intervals, which makes it necessary to

interpolate by assuming constant shares. Third, no data on prices is available. This made it necessary for KLEMS India to assume all buyers pay the same price for each product. To measure labor input, India KLEMS measures labor services. They measure labor services by combining data on employment and data on education. This creates richer data that is more precise. KLEMS India first estimates persons employed at the industry level using Work Participation Rates (WPRs) from the Employment and Unemployment Surveys (EUS) that are constructed by the National Sample Survey Office (NSSO). This data was not available for all years and sometimes had to be extrapolated backward. The most liberal definition of employed was used when estimating total employment.

Next, this measure of employment is combined with data on education. This is done by estimating the composition of the labor in terms of education using five education

categories: up to primary, primary, middle, secondary & higher secondary, and above higher secondary. KLEMS India assumed that employees earn their marginal productivities and that marginal revenues are equal to marginal cost. Earnings are estimated based on data from the NSSO. Finally, the estimated composition of the labor force is combined with the number of employed persons per industries.

There are some limitations to this labor input measure. The first is that it uses employment instead of hours worked while the latter measure is more accurate. Second, the educational categories have not been defined consistently over the whole period; earlier years did not have a separate classification for higher secondary education. Third the definitions for employed and unemployed by the NSSO are not consistent over the whole period. Finally, data on wages for a part of the employed is missing.

(18)

17 data by asset type to account for heterogeneity in assets. A distinction is made between four categories of capital: construction, transport equipment, non-ICT machinery and ICT equipment. This last category exists of hardware, software and communication equipment. Because there was no comprehensive data base on ICT investment, multiple sources have been integrated to create its estimated value: the NAS for data on sectors, ASI for the formal manufacturing sector, NSSO for the unorganized manufacturing sector, Input-Output tables and CMIE’s Prowess firm-level database. The finally obtained estimates are consistent with the aggregate data from NAS. For all three categories of capital, non-ICT assets, asset-wise deflators from NAS were used with (1999-2000) as the base year.

Finally, this heterogeneous capital stock was adjusted for compensation for capital. Hereby the depreciation rate of the three asset types were derived for each industry. The capital services has been measured as a weighted sum of past investments with weights given by the relative efficiencies of capital goods at different ages. Following NAS, India KLEMS assumes the following asset lifetimes: 80 years for buildings, 20 years for transport equipment and 25 years for machinery and equipment.

A possible problem with the capital input is the depreciation rate, which is low (4%) compared to the literature. This might cause an overestimation of growth rate of capital. The labor share was calculated by taking the average over adjacent years of the value of labor compensation as a share of gross output in current prices. Similarly, the intermediate inputs share was calculated by taking the average over adjacent years of the value of the intermediate inputs in current prices as a share of gross output in current prices.

Quantitative analysis Basic patterns

In this section, we take a preliminary look at the data. First, we compare the pre-reform period (1981-1991) with the post-pre-reform period (1992-2008). We do this for the total private sector and per industry. Next, we discuss separately the manufacturing, services and other goods producing sectors. Table 1 reports the means, standard deviations, maximum values and minimum values of the main variables and TFP for the total private sector. The industries are pooled in an unweighted fashion; the relative size of the industries is not taken into account when estimating the model.

(19)

18 Table 1

Simple statistics in percentages on the main variables for the balanced panel data sample of 22 Indian sectors over the period (1981-2008).

Note: a) TFP = ∆𝑞𝑖𝑡 − 𝑠𝑖𝑡𝐿∆𝑙𝑖𝑡 − 𝑠𝑖𝑡𝑀∆𝑚𝑖𝑡 – (1 − 𝑠𝑖𝑡𝐿 − 𝑠𝑖𝑡𝑀)∆𝑘𝑖𝑡

b) ∆𝑥_𝑖𝑡µ = 𝑠𝑖𝑡𝐿(∆𝑙𝑖𝑡− ∆𝑘𝑖𝑡) + 𝑠𝑖𝑡𝑀(∆𝑚𝑖𝑡− ∆𝑘𝑖𝑡)

c) ∆𝑥𝑖𝑡𝜃 = (𝑠𝐿𝑖𝑡 + 𝑠𝑀𝑖𝑡− 1)(∆𝑙𝑖𝑡− ∆𝑘𝑖𝑡)

The average annual TFP growth for the whole (1981-2008 )period was a modest 0.51% for the private sector (Table 1). Real output in the Indian private sector has grown quickly, at an average of 7.53% per year. This growth in real output has been accompanied by 3.88% annual growth in labor, 8.44% annual growth in capital and 7.50% growth in

intermediate inputs. The intermediate inputs share is 60.52%. The labor share is 16.17%. This is quite low compared to for example China (Wang and Yao, 2002). The share weighted growth rates of labor and intermediate inputs to capital ratios is the variable related to the markup. The mean value for this variable is -1.41. The variable related to rent-sharing is the share weighted growth rate of labor to capital ratio, which has a mean value of 0.94.

The minimum and maximum values in Table 1 should be interpreted as the highest and lowest annual growth rates off all industries. For example, the minimum value for TFP is -34.04%. This is the growth rate of TFP in the wood and products of wood industry for the year 1992. The maximum value of TFP, 24.77%, is the growth rate of TFP in Hotels and Restaurants in the year 2000. The minimum and maximum values thus indicate year specific extreme values, not industry averages. Still, there seems to be a great deal of heterogeneity Variables: Annual growth rates and

factor shares in output value N Mean

Standard-deviations Min. Max. Industry real output growth rate ∆𝑞𝑖𝑡 616 7.53 7.73 -21.90 38.93

Labor growth rate ∆𝑙𝑖𝑡 616 3.88 4.08 -8.23 23.88

Capital growth rate ∆𝑘𝑖𝑡 616 8.44 5.83 -2.45 52.44

Intermediate inputs growth rate ∆𝑚𝑖𝑡 616 7.50 8.91 -26.86 47.61

Labor share 𝑠𝐿_𝑖𝑡 616 16.17 10.29 0.57 54.49

Intermediate inputs share 𝑠𝑀𝑖𝑡 616 60.52 22.63 13.02 94.51

TFPa _{∆𝑇𝐹𝑃}

𝑖𝑡 616 0.51 4.57 -34.04 24.77

Share weighted growth rates of ∆𝑥_𝑖𝑡µ 616 -1.41 6.81 -44.68 22.86 labor and intermediate inputs to

capital ratiosb

(20)

19 across industries regardless of the variable considered.

Table 2

Average annual growth rates of real output and MFP in percentages for the balanced panel

data sample of 22 Indian industries over the period (1981-2008).

In order to give an indication of heterogeneity between industries, Table 2 reports the yearly growth rates of TFP and real output per industry for the whole period, (1981-2008). Represented per industry, TFP growth ranges from -3.47% to 4.70% and real output growth from -0.52 to 13.85%. Industries with high TFP growth rates are Post and

Telecommunications (4.70%) and Financial Services (1.99%). Both these high growth industries are service industries that can be expected to be relatively IT intensive. These high growth rates are in line with the fact that the service sector has become the most important sector in the Indian economy. There are also industries where TFP declined, among which are Wood and Products of Woods (-3.37%) and Construction (-1.65%). The quickest growers in

Sectors Real output

growth TFP

Total private sector (sectors 1 - 22) 7.53 0.51 1. Agriculture, Hunting, Forestry and Fishing 2.72 0.00

2. Mining and Quarrying 5.70 -0.01

3. Food Products, Beverages and Tobacco 6.67 0.42 4. Textiles, Textile Products, Leather and Footwear 6.32 0.43 5. Wood and Products of Wood (20) -0.52 -3.37 6. Pulp, Paper, Paper Products, Printing and Publishing 6.78 0.37 7. Coke, Refined Petroleum Products and Nuclear Fuel 6.70 -0.52

8. Chemicals and Chemical Products 8.50 0.87

9. Rubber and Plastic Products 10.34 0.60

10. Other Non-Metallic Mineral Products 8.29 0.33 11. Basic Metals and Fabricated Metal Products 6.72 -0.43

12. Machinery, nec. 7.30 0.81

13. Electrical and Optical Equipment 10.34 1.16

14. Transport Equipment 7.92 0.36

15. Manufacturing, nec; recycling 11.14 1.50

16. Electricity, Gas and Water Supply 7.24 1.91

17. Construction 6.48 -1.65

18. Trade 7.27 1.49

19. Hotels and Restaurants 8.54 1.04

20. Transport and Storage 7.30 0.07

21. Post and Telecommunication 13.85 4.70

(21)

20 terms of real output are Post and Telecommunications (13.85%) and Manufacturing, nec; recycling (11.14%). There was one industry with a negative growth in real output: Wood and Products of Woods, the industry with the most negative growth in TFP. In line with what would be expected, the correlation between real output growth and TFP is 0.82 (Table A25), suggesting that they move in the same direction quite consistently.

Next, looking at heterogeneity between time periods, Table 3 shows the data for the total private sector split up over two periods: the pre-reform period (1981-1991) and the post-reform period (1992-2008). Table 3 indicates that on average, real output growth seems to have speeded up after the 1990s reforms. From 6.67% in the pre-reform period, to 8.08% in the post-reform period. This while after the 1990s reforms the growth in labor input decreased and the growth in capital stagnated.

Table 3

Average annual growth rates in real output for the total private sector over the period 1981 – 2008 and split out over the pre-reform period (1981 – 1991) and the post-reform period (1992 – 2008).

The only input for which the annual growth rate increased after the reforms were the intermediate inputs: from 6.59% before the reforms to 8.09% after the reforms (Table 3). The input share remained relatively stable, the labor share decreased a little and the intermediate Variables: Annual growth rates and

factor shares in output value

All years (1981-2008) Pre-reform (1981-1991) Post-reform (1992-2008) Industry real output growth rate ∆𝑞_𝑖𝑡 7.53 6.67 8.08

Labor growth rate ∆𝑙_𝑖𝑡 3.88 4.70 3.35

Capital growth rate ∆𝑘_𝑖𝑡 8.44 8.39 8.47

Intermediate inputs growth rate ∆𝑚_𝑖𝑡 7.50 6.59 8.09

Labor share 𝑠𝐿

𝑖𝑡 16.17 17.90 15.05

Intermediate inputs share 𝑠𝑀𝑖𝑡 60.52 59.63 61.09

TFPa ∆𝑇𝐹𝑃_𝑖𝑡 0.51 0.23 0.68

Share weighted growth rates of ∆𝑥_𝑖𝑡µ -1.41 -1.95 -1.07 labor and intermediate inputs to

capital ratiosb

(22)

21 input share increased a little. Interestingly, and in line with earlier research, annual TFP growth increased after the reforms. From 0.23% to 0.68% annually. The variable underlying the markup became less negative after the reforms, and the variable underlying the rent-sharing increased.

Now investigating heterogeneity between periods at the industry level, Table 4 reports the yearly growth rates of TFP and real output per industry split up over the pre-reform (1981-1991) and post-reform (1992-2008) period. The first thing we notice is that again there is a lot of heterogeneity between industries and time periods. For some industries TFP was positive in the pre-reform period and negative in the post-reform period. For other industries this was the opposite. Similarly, comparing the pre-reform period with the post-reform period, real output growth more than doubled in some industries (Hotels and restaurants; Post and

Telecommunication) while there was a strong decrease in real output growth in other industries (Electricity, Gas and Water Supply; Rubber and Plastic Products).

The second thing we notice is that the high TFP in Post and Telecommunications over the entire period is entirely due to TFP in the post-reform period, where TFP grew with an average of 9.11% per year. Before the reforms TFP in the Post and Telecommunications industry was negative, on average -2.10% per year. This increase in TFP is accompanied by a large increase in real output growth. TFP in financial services was more stable. The negative TFP in the Wood and Products of Wood industry was mainly due to the negative TFP of -5.57% in the post-reform period. Interestingly, real output growth strongly increased in the post-reform period. TFP in the Construction industry became less negative after the reforms, which was accompanied by a strong increase in real output growth.

A final interesting point is the correlation between TFP and real output growth. In the post-reform period this was 0.80 (Table A26). This correlation is quite strong, which is according to expectations. However, the correlation between TFP and real output growth was much lower in the pre-reform period, 0.48 (Table A25). The relationship between real output growth and TFP thus seems to have become stronger after the reforms.

(23)

22 Table 4

Average annual growth rates of real output and MFP in percentages for the balanced panel data sample of 22 Indian industries split up over the pre-reform period (1981-1991) and the post-reform period (1992-2008).

(1) Pre-reform period (1981 – 1991) (2) Post-reform period (1992 – 2008)

As for the private economy as a whole, real output in the manufacturing sector has grown quickly, with an annual average of 7.42% (Table 5). There was a small increase after the 1990s reforms, from 6.91% per year to 7.76% per year. This increase seems to be entirely due to an increase in the growth rate of intermediate inputs, which increased from 5.56% to 8.29% per year. The growth rates of the other two inputs, labor and capital, both decreased. Furthermore, TFP became negative after the 1990s reforms. Where in the pre-reform period, TFP grew at an annual average of 0.67%, this went to -0.11% after the reforms. The labor share is lower than for the private sector, 9.66%. The intermediate inputs share is higher than

Sectors Real output growth TFP

(1) (2) (1) (2)

Total private sector (sectors 1 - 22) 6.53 7.81 0.33 0.55 1. Agriculture, Hunting, Forestry and Fishing 2.49 2.86 0.11 -0.06

2. Mining and Quarrying 8.09 4.16 -0.94 -1.16

3. Food Products, Beverages and Tobacco 4.89 7.83 1.15 -0.05 4. Textiles, Textile Products, Leather and Footwear 4.82 7.30 -0.11 0.78 5. Wood and Products of Wood (20) -4.02 1.75 -0.79 -5.57 6. Pulp, Paper, Paper Products, Printing and Publishing 8.00 5.98 1.14 -0.12 7. Coke, Refined Petroleum Products and Nuclear Fuel 7.50 6.18 0.37 -1.10 8. Chemicals and Chemical Products 8.80 8.31 1.72 0.33 9. Rubber and Plastic Products 13.74 8.14 0.79 0.48 10. Other Non-Metallic Mineral Products 8.87 7.91 0.32 0.34 11. Basic Metals and Fabricated Metal Products 6.01 7.18 -0.17 -0.60

12. Machinery, nec. 7.07 7.45 -0.06 1.38

13. Electrical and Optical Equipment 9.14 11.12 1.85 1.49

14. Transport Equipment 7.93 7.92 -0.14 0.68

15. Manufacturing, nec; recycling 7.04 13.79 2.93 0.57 16. Electricity, Gas and Water Supply 10.55 5.10 1.52 2.17

17. Construction 3.49 8.42 -2.32 -1.22

18. Trade 4.70 8.93 -0.08 2.51

19. Hotels and Restaurants 4.42 11.21 -1.29 2.55

20. Transport and Storage 6.46 7.84 -1.08 0.81

21. Post and Telecommunication 6.40 18.67 -2.10 9.11

(24)

23 for the total private sector, 74.61%. The share weighted growth rates of labor and intermediate inputs to capital ratios is much less negative after the reforms; -1.11 in the post-reform period against 3.18 in the pre-reform period. The share weighted growth rate of labor to capital ratio is positive and quite stable.

Table 5

Mean values in percentages for the manufacturing sector for the period (1981-2008) and split out over the pre-reform period (1981-1991) and the post-reform period (1992-2008).

With an average annual rate of 9.40% over the whole (1981-2008) period, real output grew quicker in the services sector then it did in the manufacturing sector (Table 6). This difference is due to the staggering 11.29% with which the services sector has grown after the 1990s reforms. The increase in growth in the post-reform period is also witnessed in TFP. There is a large increase in annual average TFP growth from -0.44% in the pre-reform period to a whopping 3.35% in the post-reform period. Interestingly, growth in labor services

decreases a little in the (1992-2008) period, compared to the (1981-1991) period. This is surprising in light of the importance of human capital in the services sector. Consistently, the labor share also decreased in the post-reform period. The growth rates of capital and

intermediate inputs increase after 1991. The intermediate inputs share is relatively stable. The variable underlying the markup is only slightly negative. The variable related to rent-sharing Annual growth rates and factor

shares in output value

Manufacturing

(1981-2008) (1981-1991) (1992 - 2008) Industry real output growth rate ∆𝑞𝑖𝑡 7.42 6.91 7.76

Capital growth rate ∆𝑘𝑖𝑡 9.14 9.40 8.97

Labor share 𝑠𝐿_𝑖𝑡 9.66 10.97 8.82

TFPa ∆𝑇𝐹𝑃𝑖𝑡 0.21 0.69 -0.11

capital ratiosb

Share weighted growth rate of labor to capital ratioc

(25)

24 is positive and increased strongly, from 0.47 in the pre-reform period to 1.41 in the post-reform period.

Table 6

Mean values in percentages for the services sector for the period (1981-2008) and split out over the pre-reform period (1981-1991) and the post-reform period (1992-2008).

Finally, Table 7 shows the average annual growth split up over the pre-reform and post-reform period for the other goods producing sector. The real output rate is lower than in the other sectors, 5.54%. Furthermore, it decreases after the 1990s reforms. Of course, when comparing to other countries this growth rate is still quite impressive. The growth rate in labor and intermediate inputs decreases as well, but the capital growth rate increases. The labor share and the intermediate inputs share remain quite stable over periods. TFP is negative for this sector, -0.20. It is negative over the whole period but stronger so in the pre-reform period than in the post-reform period. The variable related to the markup is negative and the variable related to rent-sharing is positive.

To conclude, this preliminary analysis indicates that there is a lot of heterogeneity in the data. There are large differences between industries, where some industries grow quickly in both output and TFP, while other industries are in decline. There is also heterogeneity between time periods: for some industries and sectors growth increases in the (1992-2008) Annual growth rates and factor

Services sector

(1981-2008) (1981-1991) (1992 - 2008) Industry real output growth rate ∆𝑞_𝑖𝑡 9.40 6.48 11.29

Labor growth rate ∆𝑙𝑖𝑡 5.03 5.30 4.86

Capital growth rate ∆𝑘_𝑖𝑡 7.70 6.97 8.17

Intermediate inputs growth rate ∆𝑚𝑖𝑡 9.34 8.94 9.59

Labor share 𝑠𝐿

𝑖𝑡 27.11 30.70 24.78

TFPa ∆𝑇𝐹𝑃_𝑖𝑡 1.86 -0.44 3.35

capital ratiosb

(26)

25 period compared to the (1981-1991) period. For other industries and sectors this is the

opposite. In the next section we will elaborate on the estimation techniques we use to estimate the model.

Table 7

Mean values in percentages for the other goods producing sector for the period (1981-2008) and split out over the pre-reform period (1981-1991) and the post-reform period (1992-2008). Note: a) TFP = ∆𝑞𝑖𝑡 − 𝑠𝑖𝑡𝐿∆𝑙𝑖𝑡 − 𝑠𝑖𝑡𝑀∆𝑚𝑖𝑡 – (1 − 𝑠𝑖𝑡𝐿 − 𝑠𝑖𝑡𝑀)∆𝑘𝑖𝑡 b) ∆𝑥_𝑖𝑡µ = 𝑠𝑖𝑡𝐿(∆𝑙𝑖𝑡− ∆𝑘𝑖𝑡) + 𝑠𝑖𝑡𝑀(∆𝑚𝑖𝑡− ∆𝑘𝑖𝑡) c) ∆𝑥𝑖𝑡𝜃 = (𝑠𝐿𝑖𝑡 + 𝑠𝑀𝑖𝑡− 1)(∆𝑙𝑖𝑡− ∆𝑘𝑖𝑡) Estimation methods

The coefficients for the markup, rent sharing, elasticities of scale and pure TFP from equation (9) are estimated with five different techniques; Pooled OLS, Fixed effects, system GMM, difference-GMM, and Wooldridge-Levinsohn-Petrin (WLP). Before we discuss the strengths and weaknesses of these methods, the underlying econometric assumptions of the model will be discussed. The detailed supporting evidence from the tests used to examine them can be found in the appendix.

First, a skewness kurtosis test was used to test the normality of the dependent variable per industry (Table A15). For 9 of the 22 industries, the skewness kurtosis test was

significant, indicating possible problems with normality. This needs to be kept in mind when interpreting the results. Second, based on the correlational matrices, no problems with multicollinearity are expected. Third, the Breusch-Pagan test found heteroscedasticity in the Annual growth rates and factor

Other goods producing sector (1981-2008) (1981-1991) (1992 - 2008) Industry real output growth rate ∆𝑞𝑖𝑡 5.54 6.15 5.14

Capital growth rate ∆𝑘𝑖𝑡 7.07 6.89 7.19

Labor share 𝑠𝐿𝑖𝑡 23.62 24.42 23.11

Intermediate inputs share 𝑠𝑀

𝑖𝑡 44.84 45.19 44.61

TFPa _{∆𝑇𝐹𝑃}

𝑖𝑡 -0.20 -0.41 -0.07

Share weighted growth rates of ∆𝑥_𝑖𝑡µ -1.33 -0.32 -1.99 labor and Intermediate inputs

to capital ratiosb

(27)

26 data; also for the subsets of the different sectors. We will account for this by using robust standard errors when estimating the data. Fourth, there are no indications of problems with stationarity. A Fisher-type test strongly rejected the null hypothesis that the panels contain unit roots (Table A20 – A24).

Fifth and final, a serious issue to take into account when estimating TFP is endogeneity, or simultaneity bias. This problem theoretically arises because firms are

assumed to have at least partial knowledge about their own productivity level. They will take this knowledge into account when deciding how much they will produce and how much inputs they will use. The amount of inputs in the production function is thus not chosen independently, but depends on knowledge the firm has about its productivity (Olley and Pakes, 1996). This can cause endogeneity. Since all of our variables are directly or indirectly constructed based on the firm inputs, we have to take endogeneity into account when

discussing possible econometric techniques.

(28)

chi-27 squared test.

For both GMM techniques to be valid it is critical that the assumption holds that the (lagged) instruments are exogenous. Since we use robust standard errors to account for heteroscedasticity, the Sargan test, and the Hansen J test will be used to indicate whether the instruments can be assumed to be exogenous (Roodman, 2009). To keep the amount of instruments as low as possible, only one lag year will be used. Several lags have been tested and the second was chosen because it gave the best results.

Besides the tests for endogeneity, GMM-SYS and GMM-diff will additionally perform the Arellano –Bond test for both autocorrelation in the differenced residuals, AR (1), and for autocorrelation in levels, AR (2). Both test have the null hypothesis that there is no autocorrelation. AR (1) is expected to be significant since the errors are differenced, which creates an overlap between ∆𝑒𝑖𝑡 and ∆𝑒𝑖,𝑡−1. Both AR (1) and AR (2) will be reported

(Roodman, 2009).

Finally, we use the Wooldridge modification of the Levihnson-Petrin technique (WLP). This technique implements the moment conditions of the Levinsohn-Petrin approach into a method of moments framework. This creates robust standard errors. Like the GMM-diff and GMM-SYS, the WLP estimation uses lags of the endogenous variables to control for the endogeneity. Again the second lag was chosen because it gave the best results. To verify the robustness of the results we will discuss the results based on both the GMM techniques and WLP.

Before continuing to the estimation of (9), there are two possible biases (besides endogeneity) that we want to mention. First, a bias could be created by the presence of adjustment costs in inputs. The assumption that there are no adjustment costs could bias the use of inputs upwards. Firm level differences in adjustment costs will be averaged out by estimating average production function coefficients. However, country wide adjustment costs might still be a problem because research has found especially labor to adjust relatively slow (Bhalotra, 1998; Dutta Roy, 2003).

(29)

28

Analysis of the results

In this section we report the results from the estimation of (9) by SYS, GMM-diff, and WLP. For every estimation method we first report and explain the estimated coefficients 𝜌 for the markup, 𝛾 for the economies of scale and 𝜔 for rent-sharing. We also estimate the coefficient for the pure TFP1. Based on literature on distortions in both products and labor markets, we expect 𝜌 and 𝜔 to be positive. Since the literature has found constant and decreasing returns to scale, 𝛾 is expected to be zero or negative. Pure TFP is expected to be lower than the standard measure of TFP.

We start with discussing the coefficients 𝜌, 𝛾 and 𝜔 for the total private sector over the whole period (1981-2008). Then, we estimate the model split up over two periods: the pre-reform period (1981-1991) and the post-pre-reform period (1992-2008). The reason for this split up is that a significant part of the government policies that are suspected to have caused products and market distortions were abolished during the early 1990s reforms (Rodrik and Subramanian, 2004). We therefore expect the distortions and thus the markup and rent-sharing to be lower in the post-reform period. To examine heterogeneity between sectors, we finish the analysis of the total sector by calculating the TFP and pure TFP per industry. After discussing the total private sector we use GMM-SYS, GMM-diff, and WLP to estimate (9) for the three sub-sectors in the following order: manufacturing; services; and other goods

producing.

Finally, we focus on the manufacturing sector and insert the coefficients 𝜌, 𝛾 and 𝜔 into (9) in order to calculate and discuss the structural parameters for the markup 𝜇 , the economies of scale 𝜆, rent-sharing 𝜃 and pure TFP. We use these parameters to discuss hypotheses 1-4 from the literature review.

Consistent with the preliminary analysis from the Basic patterns section, we find a lot of heterogeneity in the data when looking at the broader picture. First, there is heterogeneity between time periods. Especially for the markup and rent-sharing, the coefficients generally differ strongly from the pre-reform period to the post-reform period. However, this difference is rather inconsistent. Second, the coefficients for the elasticities of scale are generally slightly negative and relatively stable over time. There are also large differences between industries, where TFP grows quickly in some industries and declines in others. Finally, there is

heterogeneity between sectors; the trends in the manufacturing sector are often the opposite from the trends in the services sector.

(30)

29 Now, we start the data analysis by investigating the total private sector. We estimate equation (9) for the whole period (1981-2008), the pre-reform period (1981-1991) and the post-reform period (1992-2008) (Table 8).

Table 8

Decomposition of the standard TFP growth: System-GMM and Difference-GMM regression for the total private sector

Standard errors in parentheses * p<0.01, ** p<0.05, *** p<0.1 (1) Total period (1981 – 2008) (2) Pre-reform period (1981 – 1991) (3) Post-reform period (1992 – 2008)

Table 8 reports the coefficients 𝜌, 𝛾 and 𝜔 for the total sector estimated through GMM-SYS and GMM-diff. Looking at the results for GMM-SYS, the first thing that should be noted is that, with one exception, the signs are of the expected direction. The coefficients 𝜌 for the markup and 𝜔 for rent-sharing are positive (except for the post-reform markup), and the coefficients 𝛾 for the elasticities of scale are consistently negative. This indicates decreasing economies of scale. Unexpectedly, none of the coefficients for the markup is significant. There are two reasons for this. First, the coefficients are smaller than one would expect from a developing country like India. Second, the errors are relatively large. This

VARIABLES System-GMM Difference-GMM

(1) (2) (3) (1) (2) (3) 𝜌 0.01 0.08 0.00 0.19* 0.29** 0.12 (0.05) (0.09) (0.09) (0.06) (0.12) (0.12) 𝛾 -0.11 -0.09 -0.14*** 0.04 0.05 0.01 (0.05) (0.08) (0.07) (0.10) (0.21) (0.19) 𝜔 0.36 0.37 0.32 0.26 0.08 0.21 (0.31) (0.74) (0.29) (0.40) (0.96) (0.47) Pure TFP 0.011*** 0.009 0.015 (0.01) (0.01) (0.01) Observations 594 242 374 594 220 352 AR (1) 0.000 0.002 0.001 0.001 0.004 0.001 AR (2) 0.194 0.93 0.063 0.213 0.94 0.108

Sargan test (p-value) 0.002 0.139 0.001 0.843 0.724 0.777

(31)

30 might be due to the relatively small amount of groups in our sample compared to the amount of instruments. The rule of thumb is that the amount of instruments should not exceed the amount of groups (Roodman, 2009). In the GMM-SYS, there are 22 groups and 157 (total period), 55 (pre-reform period) and 91 (post-reform period) instruments. The Hansen J test was not significant, indicating the instruments are exogenous. However, the Sargan test for the estimations for total period and for the post-reform period was significant, indicating that the instruments used might not have been exogenous. Additionally, the AR (2) test is

significant for the post-reform period, indicating possible autocorrelation between levels. Therefore, the results should be interpreted with caution.

The coefficients for the elasticities of scale are significant for the post-reform period, but only at the 10% level. The coefficient for rent-sharing is not estimated significantly for any of the coefficients. The coefficient for pure TFP was significant at the 10% level for the total period. In line with the literature, pure TFP growth increased in the post-reform period, relative to the pre-reform period.

Next, we look at the GMM-diff method (Table 8). We expect the GMM-diff estimates to be better than the GMM-SYS estimates for several reasons. First, as mentioned before, since GMM-diff uses less instruments versus the amount of groups the results might be less biased. Furthermore, both the Hansen J tests and the Sargan tests indicate that there are no clear problems with endogeneity. Finally, the AR (2) test indicates that there is no reason to worry about autocorrelation between levels.

When investigating the results for GMM-diff, the first thing we notice is that all the coefficients are positive, including the coefficients 𝛾 for the elasticities of scale. This means the model finds increasing returns to scale, in contrast to the results from the estimation with GMM-SYS. However, the positive coefficients for economies of scale are non-significant, meaning that we cannot attach much meaning to them. More interesting are the coefficients for the markup. The coefficient 𝜌 for the markup for the total period is .19 and significant at the 1% level. The coefficient for the markup for the pre-reform period is .29 and significant at the 5% level. Both are very high. Non-significant, but consistent with the literature, the markup is lower in the post-reform period than in the pre-reform period. The coefficients 𝜔 for the bargaining power are very low and non-significant.

(32)

31 level and higher than the markup in the pre-reform period. The coefficients for the economies of scale are all negative but significant. The coefficients for rent-sharing are and non-significant. The pure TFP over the whole period is 1.00% and significant at the 1% level. This is again quite low. Interestingly average TFP growth is much higher after the 1990s reforms. In the (1981-1991) period it is 0.10%, while in the (1992-2008) period it is 1.50% and highly significant.

Table 9

Decomposition of the standard TFP growth: WLP regression for the total private sector

Standard errors in parentheses * p<0.01, ** p<0.05, *** p<0.1

In order to explore heterogeneity between industries, we use the coefficients from the GMM-diff regression (Table 8) to calculate the pure TFP per industry. The results are

reported in Table 10. Since the constant in the GMM-diff estimation is ‘differenced out’, we calculated the pure TFP by entering the standard measure of TFP, the coefficients 𝜌, 𝛾 and 𝜔 and the mean growth statistics into the model, which we rewrite as:

(10) ∆𝒖𝒊𝒕= 𝑻𝑭𝑷 − 𝝆∆𝒙𝒊𝒕 𝝁

− 𝜸∆𝒌𝒊𝒕 − 𝝎∆𝒙𝒊𝒕𝜽

VARIABLES Total period

(1981-2008) Pre-reform (1981-1991) Post-reform (1992-2008) 𝜌 0.05 -0.02 0.11** (0.05) (0.05) (0.05) 𝛾 -0.06 -0.01 -0.07 (0.04) (0.09) (0.05) 𝜔 0.12 0.23 0.05 (0.18) (0.38) (0.22) Pure TFP 0.010* 0.001 0.015* (0.003) (0.006) (0.004) Observations 572 198 330 R-squared 0.010 0.006 0.038

(33)

32 Table 10

Estimates of standard TFP and pure TFP

For the 22 individual industries, the coefficients of the total private sector and the mean statistics of the individual industries were used (Table 10). The first thing we notice is that, according to expectation, pure TFP is consistently lower than the traditionally measured TFP estimate. Furthermore, the industries with the highest pure TFP are the same as the industries with the highest standard TFP: Post and Telecommunications (4.00%), Electricity, Gas and Water Supply (1.58%) and Financial Services (1.36%). Looking at sector averages, pure TFP is highest in the service sector. This is consistent with the expectations. The

Sector Industry Standard

TFP

Pure TFP Manufacturing Food Products, Beverages and Tobacco 0.42 0.23

Textiles, Textile Products, Leather and

Footwear 0.43 0.31

Wood and Products of Wood -3.37 -3.57 Pulp, Paper, Paper Products, Printing and

Publishing 0.37 0.18

Coke, Refined Petroleum Products and

Nuclear Fuel -0.52 -0.03

Chemicals and Chemical Products 0.87 0.17 Rubber and Plastic Products 0.60 0.36 Other Non-Metallic Mineral Products 0.33 0.00 Basic Metals and Fabricated Metal Products -0.43 -0.81

Machinery, nec. 0.81 0.78

Electrical and Optical Equipment 1.16 1.01

Transport Equipment 0.36 0.18

Manufacturing, nec; recycling 1.50 0.66 Services Wholesale and Retail Trade 1.49 1.07

Hotels and Restaurants 1.04 0.61

Transport and Storage 0.07 -0.57

Post and Telecommunication 4.70 4.00

Financial Services 1.99 1.36

Other goods Agriculture, Hunting, Forestry and Fishing 0.00 -0.31

producing Mining and Quarrying -0.01 -1.74

Electricity, Gas and Water Supply 1.91 1.58

Unbundling India’s Productivity Performance