Online Appendix for: Competition and the welfare gains from transportation infrastructure: Evidence from the Golden Quadrilateral of India Jose Asturias SFS in Qatar, Georgetown University Manuel García-Santana

Roberto Ramos

Universitè Libre de Bruxelles, ECARES

Bank of Spain

January 20, 2015

Contents A Data Appendix

2

A.1 Details on Plant-Level Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2

A.2 Computation of Prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2

A.3 Unit Misreporting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3

A.4 Details on Data Preparation of the Difference-in-Difference Specification

. . . . . .

3

A.5 Details on Construction of Graph . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4

B Constructing Iceberg Costs

4

C Direct Measures of Transportation Prices

5

D The Firm-Level Linear Relationship Between Labor and Sectoral Shares

7

E Sensitivity Analysis

9

This online appendix is a companion to the main article and it is organized as follows. In Section (A) we provide details about the manufacturing data that we use, our procedures to create 1

the pricing data, and our data preparation for the difference-in-difference exercise. In Section (B) we document how we construct the iceberg costs that we plug into the model based on our estimates of equation (15) of the main text (the monopolists price equation). In Section (C) we provide further details on the transportation prices that we collected and how we compute the implied iceberg costs. In Section (D) we derive the relationship implied by the model between labor and sectoral shares in the case of goods that are produced only in one state.

A Data Appendix A.1 Details on Plant-Level Data The Annual Survey of Industries (ASI) consists of two parts: the ASI census and the ASI sample. Plants with 100 or more workers are categorized as the census sector, which means that all plants are surveyed. In order to account for the rest of the population of registered plants, all plants with fewer than 100 employees are randomly sampled. The sample frame is carefully designed: all plants are stratified at the sector-industry 4-digit level of NIC and at least 1/5th of the plants in each strata are selected for the sample. The purpose of NSS is to cover all manufacturing production units that are not covered by ASI. This survey is conducted every five years by the Indian Ministry of Statistics: 1989-90, 1994-95, 2000-01, 2005-06, and 2010-11. It is a good data set for very small plants because the sampling universe is the set of Indian households. We use the waves with reference year 2000-01 and 200506. These two surveys correspond to the module 2.2 of the 56th and 62th Indian National Sample Survey. The data reported by the plants is carefully collected and monitored by the National Sample Survey Organization, which is part of the Ministry of Statistics and Programe Implementation. When plants report their records: (i) they are initially verified by the field staff; (ii) the information is manually inspected by senior level staff; (iii) the data is sent to the data center where it is verified again before it is entered in the computers; and (iv) once the data is entered, the members of the IT team look for anomalies and check consistency with previous surveys. A.2 Computation of Prices We compute prices of non-imported inputs consumed in the manufacturing production process for the years 2001 and 2006.1 Prices of every input in every district are computed as the total purchase value over total quantity consumed. Each input is identified by the 5-digits Annual Survey of Industries Commodity Classification (ASICC). We do not consider input items whose description refer to “other” or “non elsewhere classified” products.

1

The survey asks plants to distinguish between imported and domestic intermediate inputs.

2

A.3 Unit Misreporting We identify firms that appear to report quantities in different units than intended, which generates big changes in the scale of prices for some goods. Figure I shows the example of Chlorophos (ASICC 31611). The average log price of input is 5.6 for some firms and 12.3 for others. This is due to the fact that some firms report quantities of this input in tons as intended by the survey, whereas others in kilograms. Hence the difference in average log price of 6.7 can be explained by a denominator multiplied by 1,000 (ln(1000) = 6.9). As this is a source of error, we sort every product by price (from low to high) and identify jumps in prices in which the ratio of one price over the previous one is more than 20. If that happens, we consider those goods as different products. We use separate fixed effects for the set of observations before and after the jump in prices when estimating equation (15). Kothari (2013) used a similar strategy as outlined in Appendix C. A.4 Details on Data Preparation of the Difference-in-Difference Specification We use the 2000-2001 and 2005-2006 rounds of both the Annual Survey of Industries and the National Sample Survey in order to study the evolution of prices as a result of the construction of the GQ project.2 For each round and district, we compute the price of each product as a weighted average of the prices paid by the plants using that product as intermediate in that district. Each price is calculated as the value of consumption of the input over the quantity consumed. We exclude those goods that present some unit misreporting, as explained in the preceding section. We observe the price of 912 products that were consumed in the same district in both 2001 and 2006. There is a total of 323 districts. Additionally, using ArcGIS, we compute the shortest straight-line distance from each district to a completed stretch of the GQ in March 2001 and March 2006. We then compute several treatment dummies taking the value 1 if the district is within a certain distance of the GQ and zero otherwise. We consider this set of distances: 15, 25, 50, 100, 150, 200, and 300 kilometers from the GQ. Treated districts are those for which the treatment dummy changes between 2001 and 2006. The control districts are those that did not gain further access to the network infrastructure between 2001 and 2006. Following the previous discussion, we exclude nodal districts (Delhi, Mumbai, Chennai, and Calcutta) as well as a few contiguous suburbs identified by Datta (2012) that were on the GQ as a matter of design rather than fortuitousness (Gurgaon, Faridabad, Ghaziabad, Gautam Buddha Nagar, and Thane). Finally, we exclude the few districts that were within 50 kilometers of an upgraded portion of the GQ in 2001. The reason is that we want to compare the evolution of prices in districts that were treated in 2006 with districts that were not treated in 2006. Our benchmark administrative division is that of 2001, hence districts in 2006 that were carved 2

Although the GQ was not completed until 2011, we use 2006 as the last year of treatment due to the fact that

the National Sample Survey does not have information on inputs in the 2010-2011 round. Note that 91 percent of the GQ was finished in 2006.

3

out from existent districts in 2001 are assigned to their original district. Then, using highway maps and the ArcGIS software, we compute the shortest straight-line distance from every district to the nearest completed stretch of the GQ in March 2001 and March 2006. Of the 127 stretches of the GQ (5,846.64 km), 16 (769 km) were completed by March 2001 and 114 (5,303.17 km) by March 2006. That is, in 2001, only 13 percent of the GQ was completed, whereas in 2006, 91% of the network was finished.

Figure I

2

4

6

Log Price 8

10

12

Price Computed for Input Chlorophos (ASICC 31611)

2

3

4 5 Log Employment

6

7

Figure I shows the price of input good Chlorophos (ASICC 31611), computed as value over quantity consumed, against the size of the firm. The change in scale for some prices shows the unit misreporting in some observations.

A.5 Details on Construction of Graph The geospatial data on the national highway system of India was provided to us by ML Infomap. In order to convert the national highway system into a graph, we used Network Analyst in ArcGIS. The nodes of the graph are the most populous city of a district. Cities that are no immediately on the road are mapped to the closest straight-line point to a National Highway. In addition, the graph has nodes at any point in which the road changed from being treated to non-treated (upgraded vs. not upgraded). The nodes are connected by an arc if it is possible to travel from one node to the other without passing another node.

B Constructing Iceberg Costs In this section, we explain more in detail how we construct the iceberg costs that we plug into our model. As stated in the main text, we exploit the variation in prices charged by monopolists across locations in order to infer transportation costs. We identify more than 200 plants that account for at least 95 percent of total sales of each product. We then estimate the elasticity of prices paid 4

by plants consuming these products with respect to the effective distance between the place of consumption and the location of the corresponding monopolist (see equation (15). An example is illustrated in Figure II. The blue square represents the location of the monopolist producing the good raw jute (ASICC 65106), which is the industrial term for jute fiber. This is a vegetable fiber used for the production of twine, rope and matting. The production of this good is concentrated in West Bengal, as well as in Bangladesh. In our data, we find a monopolist located in the district of North 24-Parganas, in the state of West Bengal, represented by the blue square. Also, we find close to 60 plants that report the usage of raw jute as in intermediate input, geographically located in different districts of India (the red dots of the map). Hence, the variation of prices paid by these plants as a function of the effective distance with respect to the monopolists gives us the identification to estimate the iceberg costs. Table (I) of the main text reports the results of estimating equation (15). As discussed in the main text, although the overall pattern is increasing, our estimates for the dummies of effective distance deciles are not monotonically increasing. In order to avoid having non-monotonic iceberg costs to effective distance in the model, we assume that the relationship between iceberg costs and effective distance is given by a discrete cubic function g(xod ), where xod is the decile of the distribution of log distance to which the effective distance between origin o and destination d belongs. We find the parameters of g(xod ) that best fit the coefficients implied by the regression. This is: τdo = g(xod ) = 0.9747 + 0.1493x − 0.0239x2 + 0.0014x3 ,

(1)

where xod is a discrete variable that indicates the decile of effective distance. For example, it takes value 2 if the the log effective distance between origin o and destination d falls in the second percentile of the distribution of log effective distance. Panel (A) of Figure (III) shows transportation costs as a function of the deciles of the distribution of log effective distance, both the ones implied by the estimated coefficients of regression (15) and ones implied by our functional form from equation (B).

C Direct Measures of Transportation Prices We collected data on transportation prices charged by GIR Logistics, one of the biggest transportation companies in India.3 In particular, we have collected the price of transporting a standardized shipping container of size 20 ft x 8 ft x 8.5 ft for approximately 900,000 origin-destination Indian city pairs. We now show how we construct measures of transportation costs using information on transportation prices. We want to compare the transportation costs that we estimated using ASI-NSS 3

See www.girlogistics.in for more further details.

5

data with the ones implied by these direct measures of transportation prices. We first pick from the sample of 900,000 origin-destination Indian city pairs the same pairs that we use in the regression that we run with the ASI-NSS data. We then run the following regression, which is the equivalent to equation (15): log pod = βlog Effective Distanceod +

X

δo + od

(2)

o

where pod is the price of transporting a container from district d to district o, δo are a set of districts of origin fixed effects, and od is the error term. Table (I) shows the regression results. In column (1), we show that a 10 percent increase in the effective distance is associated with a 5.77 percent increase in the transportation price. In column (2), as we do in the table (I) of the main text, we consider a more flexible specification in which we allow for non-linearities in the relationship between transportation prices and effective distance. As in our benchmark regression, we include dummies associated to the ten deciles of the distribtution of effective distance, and find that the highest deciles are associated with larger increases in the transportation prices. We find, for instance, that the price of transporting a container to destinations that fall in the second decile of effective distance (around 280 km) are around 43% higher than the price of transporting it to the first decile (around 70 km). From transportation prices to iceberg costs: We now explain how we use the regression coefficients implied by column (2) of table (I) to construct bilateral iceberg costs between origindestination pairs in India. Note that, in order to construct iceberg costs, we need to pinn down the shipment value. To this end, we find a shipment value such that the implied average iceberg cost is the same as the average iceberg cost that we constructed using out benchmark regression (1.26). We obtain a shipment value of around 230,000 rupees (around $4,000). Furthermore, we normalize the iceberg cost of the first decile to one. Panel (B) of Figure (III) shows the implied iceberg costs as a function of the deciles of the distribution of log effective distance (dash line). We also plot the iceberg costs that we use as our benchmark (see Section B of this appendix for details). The overall shape of the two sets of iceberg costs is similar, especially at the right tail of the distribution of effective distance. Importantly, note that both funcions display a convexity starting at the 6-7th decile of the distribution. The biggest difference between the two sets of iceberg costs is at the 3rd decile, where the iceberg cost that we estimate using ASI-NSS data is 1.24, which is a 14.71% bigger than the one estimated using the direct measures on transportation prices.

6

Table I Impact of Road Distance and Infrastructure Quality on Transportation Prices (1)

(2)

Dep. Variable: Log price at district of destination Log Effective Distance

0.577∗∗∗ (0.030)

Log Effective Distance 2nd decile

0.435∗∗∗ (0.064)

Log Effective Distance 3th decile

0.669∗∗∗ (0.067)

Log Effective Distance 4th decile

1.027∗∗∗ (0.064)

Log Effective Distance 5th decile

1.319∗∗∗ (0.060)

Log Effective Distance 6th decile

1.446∗∗∗ (0.063)

Log Effective Distance 7th decile

1.528∗∗∗ (0.060)

Log Effective Distance 8th decile

1.631∗∗∗ (0.062)

Log Effective Distance 9th decile

1.818∗∗∗ (0.061)

Log Effective Distance 10th decile

2.036∗∗∗ (0.064)

District of Origin Fixed Effects

YES

YES

Observations R-squared

1,801 0.687

1,801 0.758

Table I shows the estimation of equation (2). The dependent variable is the log price of charge by GIR Logistics at a destination. The variable of interest is the effective distance between the district where the container is shipped from and the district of destination. Effective distance is defined as the lowest cost path between both districts, taking into account road distance and infrastructure quality. Specifically, going across the Golden Quadrilateral reduces road distance 48 per cent, relatives to roads not in the Golden Quadrilateral. The lowest path is computed by means of road networks and applying the Dijkstra’s search path algorithm. Column (1) uses a linear specification of effective distance, whereas column (2) estimates a non-linear specification, using 10 deciles of effective distance. District of origin are included. Robust standard errors are in parenthesis. Significance levels: ∗ : 10%; ∗∗ : 5%; ∗∗∗ : 1%.

D The Firm-Level Linear Relationship Between Labor and Sectoral Shares The optimal pricing decision of the firm is given by: Wo o (j, k) pod (j, k) = o d τ o, d (j, k) − 1 ao (j, k) d 7

where od (k, j)

=

and

1 ωdo (j, k) θ

+ (1 −

1 ωdo (j, k))

!−1

γ

,

po (j, k)1−γ ωdo = PN PdK o . 1−γ k=1 pd (j, k) o=1

Multiplying by ldo (j, k) on both sides of the equation and re-ordering terms: τdo Wo ldo (j, k) od (j, k) − 1 = . pod (j, k)cod (j, k) od (j, k) We now introduce additional notation to define the price that the firm sets before charging transportation costs. Let the price set by the firm at the gate of the factory be denoted: p˜od (j, k)

pod (j, k) = . τdo

This is the price that we can compute in the data when using firms’ reported sales and physical units. Using this definition, we can write the firm’s inverse of the markup as: Wo ldo (j, k) od (j, k) − 1 = , p˜od (j, k)cod (j, k) od (j, k) where

Wo ldo (j,k) p˜od (j,k)cod (j,k)

is the labor share of firms’ total revenue at destination d before transportation

costs are charged. Using the expression for the firm’s elasticity: Wo ldo (j,k) p˜od (j,k)cod (j,k)

= =



ωdo (j, k) 1θ

+ (1 −

ωdo (j, k)) γ1



−1

−1

ωdo (j, k) 1θ + (1 − ωdo (j, k)) γ1

1 − ωdo (j, k) 1θ − (1 − ωdo (j, k)) γ1 = 1 − ωdo (j, k) 1θ −

1 γ



+ γ1 ωdo (j, k),

which yields the following linear relationship between the firms’ labor share and sectoral share: Wo ldo (j, k) 1 1 1 =1− − − ωdo (j, k) o o p˜d (j, k)cd (j, k) γ θ γ !

(3)

Goods produced only in one state For those goods that are produced only in one location (location o for instance), the expression for firms’ market share becomes: ωdo (j, k)

pod (j, k)1−γ (τdo ) 1−γ p˜od (j, k)1−γ (τdo ) 1−γ p˜od (j, k)1−γ p˜od (j, k)1−γ = PK o = PK = o 1−γ PK o = PK o . o 1−γ o 1−γ p˜d (j, k)1−γ (τd ) ˜d (j, k)1−γ ˜d (j, k)1−γ k=1 pd (j, k) k=1 (τd ) k=1 p k=1 p

Note that ωdo (j, k) will be constant across different destinations. Then, summing equation (3) across destinations we get: Wo lo (j, k) 1 1 1 = 1 − − − ω o (j, k) o o p˜ (j, k)c (j, k) γ θ γ !

where: lo (j, k)

=

p˜o (j, k)co (j, k) = ω o (j, k)

PN

o d=1 ld (j, k) PN ˜o (j, k)cod (j, k) d=1 p PN d o p˜ (j,k)cod (j,k) d=1 d

= PK PN k=1

8

d=1

p˜od (j,k)cod (j,k)

E Sensitivity Analysis We now examine the sensitivity of our results by considering versions of our model in which we change the value of some of the crucial parameters. We first examine the implications of setting a lower value for the elasticity of substitution within sectors, γ. We next look at the effects of decreasing the value of the elasticity of substitution across sectors, θ. Lastly, we study a version of the model in which firms productivity shocks within sectors are uncorrelated across states. In all cases, we keep constant the rest of parameters which we estimate outside the model, and re-calibrate the ones that we calibrate in equilibrium (, i.e, labor endowment for each state i, Li , and the shape parameter of the Pareto distribution, α). In order match the fact that the top 5% of plants in manufacturing account for 89% of value-added, the model needs a shape parameter of the Pareto distribution α of 2.12 in the case of θ = 1.28, 2.00 in the case of γ = 10, and 5.75 in the case of uncorrelated draws (vs the 2.55 of our benchmark calibration). Overall, we find that the aggregate gains are remarkably stable across specifications. The fractions of gains that are pro-competitive do not change remain quantitatively relevant in the case of lower γ and lower θ. However, pro-competitive gains disappear in the case of uncorrelated productivity draws across states. Table III provides the numbers generated by the model in the three different scenarios. A lower elasticity of substitution across sectors: We set θ = 1.28, which is the value estimated by Edmond, Midrigan, and Xu (2014) using Taiwanese data. In this economy, there is more misallocation than the benchmark: the allocative efficiency index ranges from 0.87 to 0.91 across states, whereas in the benchmark calibration it ranged from 0.94 to 0.97. The reason is that the lower θ implies that firms with large market shares charge higher markups, increasing the dispersion of markups. In this specification, pro-competitive gains increase to 0.39% (vs 0.35%). Interestingly, the fraction of pro-competitive gains decline to 16.5% of the gains (vs 17.1%). At the state-level, pro-competitive gains are up to 26% of the overall gains (vs 26%). One has to be cautious when comparing these numbers since some crucial parameters are not constant across the two economies. For instance, the Pareto shape parameter becomes 2.12 (vs 2.55 in the baseline case). A value of 1.28 for θ would imply a too low elasticity for monopolists compared to the one we estimate using our Indian data. A lower elasticity of substitution within sectors: We set γ = 10, which is the value used by Atkeson and Burstein (2008) and Edmond, Midrigan, and Xu (2014). In this specification, the gains from the GQ increase to 2.32% (vs 2.05% in the benchmark). However, pro-competitive decline to 0.24% (vs 0.35%) and the fraction of gains that are pro-competitive drop to 10% (vs 17%). At the state-level, pro-competitive gains are up to 21% of the gains (vs 26%). 9

When setting γ = 10, the model generates a too weak cross-sectional relationship between markups and sectoral shares compared to the one we measure in our data. Uncorrelated productivity draws: Finally, we examine how our results change if we have uncorrelated productivity draws across locations. We find that aggregate gains increase to 2.56% (vs 2.05%). In this economy, pro-competitive gains are zero, which is consistent with the findings of Edmond et al. (2014). When we calculate our similarity index for this economy, we find a value of 0.25. This is lower than the 0.37 we obtained in our baseline calibration, and much lower than the 0.44 that we measure in the data. Thus, the degree of face-to-face competition that firms face is too low in the case of uncorrelated draws.

10

Figure II Location of Monopolist of Raw Jute and Plants Using the Good as Input

Figure II shows the geographic location of the identified monopolist of raw jute (ASICC 65106) -blue square, in West Bengal- and the plants that consume this good in the production process -red dots-.

11

Figure III Estimated Transportation Costs (B) ASI-NSS (smooth) vs GIR Logistics reg. 1.8 1.6

GIR logistics reg.

Iceberg Cost 1.4

1.8 Iceberg Cost 1.4

1.6

(A) ASI-NSS reg coeff. vs ASI-NSS smooth

ASI-NSS reg. (smooth)

1.2

1.2

ASI-NSS reg.

1

1

ASI-NSS reg. (smooth)

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

Decile (Effective Distance)

Decile (Effective distance)

Figure (III) shows the estimated transportation costs as a function of the deciles of the distribution of log effective distance. Panel (A) compares the function implied by the estimated coefficients in the regression (15) with the one implied by equation (B). Panel (B) compares the transportation costs implied by equation (B) with the ones implied by estimates using GIR Logistics data on transportation prices.

12

Table II Descriptive Statistics: ASI & NSS plants Mean

Percentiles

Mean

Percentiles

(Std. Dev)

25

50

75

(Std. Dev)

25

50

75

(1)

(2)

(3)

(4)

(1)

(2)

(3)

(4)

Panel A: ASI 2000-01 (Obs = 41, 096; plants = 171, 743)

ASI 2005-06 (Obs = 57, 304; plants = 179, 918)

Number of Employees

46.51 (382.07)

10

18

42

50.65 (347.60)

10

20

Gross Value Added per Worker (thousands of rupees)

191.87 (686.32)

27.19

63.20.

128.48

286.75 (1104.25)

11.82

57.10

Number of Products per plant

1.53 (1.11)

1

1

2

1.53 (1.12)

1

1

Panel B: NSS 2000-01 (Obs = 152, 494; plants = 17, 024, 108)

48 136.16 2

NSS 2005-06 (Obs = 82330; plants = 16, 953, 555)

13

Number of Employees

2.17 (2.55)

1

2

2

2.11 (5.59)

1

2

Gross Value Added per Worker (thousands of rupees)

16.23 (17.09)

4.18

8.56

17.52

23.12 (47.64)

4.80

9.54

Number of Products per plant

1.04 (0.26)

1

1

1

1.05 (0.28)

1

1

2 20.74 1

Table II shows descriptive statistics of Indian plants for the fiscal year 2000-2001 and 2005-06 according to NSS and ASI. Panel A shows statistics of plants in the Annual Survey of Industries (ASI). Panel B shows statistics of the National Sample Survey (NSS).

Table III Gains from the GQ: Sensitivity state ηw

γ = 10 ηRic ηT oT

ηae

ηw

θ = 1.28 ηRic ηT oT

Uncorrelated draws ηw ηRic ηT oT

ηae

ηae

14

India

2.32

1.83

0.26

0.00

0.24

2.37

0.96

1.02

0.00

0.39

2.56

1.26

1.30

-0.00

0.00

Maharashtra

1.83

1.63

-0.34

0.15

0.38

1.96

0.79

0.48

0.18

0.50

1.91

0.99

0.92

-0.00

0.00

Gujarat

2.97

2.23

0.24

0.08

0.43

3.10

1.39

1.06

0.07

0.58

3.06

1.53

1.54

-0.00

0.00

Tamil Nadu

1.65

1.57

-0.15

-0.01

0.24

1.74

0.76

0.61

-0.01

0.38

1.95

1.00

0.93

0.01

0.01

Uttar Pradesh

2.52

1.90

0.51

-0.07

0.18

2.52

1.02

1.21

-0.10

0.40

2.80

1.34

1.46

-0.01

0.01

Karnataka

3.34

2.36

0.83

-0.07

0.22

3.41

1.56

1.34

-0.08

0.59

3.75

1.79

1.95

-0.00

0.00

Andhra Pradesh

2.09

1.64

0.43

-0.09

0.11

2.05

0.72

1.09

-0.09

0.34

2.41

1.15

1.27

-0.01

-0.00

West Bengal

4.17

2.74

1.27

-0.12

0.27

4.14

1.85

1.98

-0.20

0.51

4.55

2.11

2.45

-0.02

0.01

Haryana

1.39

1.41

0.05

-0.15

0.09

1.40

0.55

0.78

-0.17

0.24

1.88

0.94

0.91

0.02

0.01

Jharkhand

5.02

3.05

1.74

0.03

0.20

5.17

2.22

2.74

0.03

0.19

5.17

2.51

2.66

-0.01

0.02

Rajasthan

3.30

2.27

1.26

-0.25

0.02

3.23

1.35

1.99

-0.28

0.18

3.75

1.73

2.01

0.00

0.01

Madhya Pradesh

1.56

1.34

0.25

-0.07

0.04

1.44

0.38

1.06

-0.12

0.11

1.79

0.89

0.92

-0.01

-0.00

Orissa

2.48

1.79

0.54

0.02

0.13

2.46

0.92

1.28

0.00

0.24

2.69

1.33

1.37

-0.01

0.00

Punjab

0.70

0.90

-0.21

-0.03

0.04

0.70

-0.13

0.59

0.11

0.13

1.09

0.55

0.49

0.03

0.01

Himachal Pradesh

0.55

0.93

-0.25

-0.15

0.03

0.49

0.02

0.58

-0.19

0.09

0.84

0.48

0.38

-0.04

0.01

Chattisgarh

0.40

0.81

-0.46

-0.01

0.05

0.31

-0.09

0.30

-0.04

0.13

0.61

0.43

0.21

-0.02

-0.00

Kerala

1.10

1.06

-0.00

-0.02

0.07

1.06

0.02

0.70

0.09

0.25

1.51

0.72

0.76

0.03

0.00

Uttaranchal

1.09

1.10

-0.01

-0.04

0.04

1.01

0.04

0.76

0.07

0.14

1.43

0.70

0.70

0.02

0.01

Delhi

1.46

1.21

0.17

0.00

0.08

1.48

0.10

0.88

0.26

0.24

1.93

0.84

1.04

0.05

0.00

Assam

0.50

0.76

-0.34

0.03

0.05

0.39

-0.23

0.45

0.06

0.11

0.75

0.41

0.35

-0.01

-0.00

Goa

8.19

4.52

4.16

-0.40

-0.10

7.70

3.50

5.16

-0.74

-0.22

8.47

3.88

4.69

-0.11

0.01

Bihar

5.66

3.27

2.66

-0.26

-0.01

5.41

2.32

3.51

-0.43

0.01

5.89

2.73

3.24

-0.08

0.00

Jammu and Kashmir

-0.22

0.42

-0.51

-0.10

-0.03

-0.22

-0.61

0.44

0.02

-0.08

0.13

0.13

-0.04

0.04

0.01 -0.01

Meghalaya

0.78

0.90

-0.23

0.06

0.05

0.56

-0.13

0.50

0.03

0.16

1.08

0.54

0.48

0.07

Tripura

-1.52

-0.28

-1.42

0.15

0.03

-1.15

-0.83

-0.63

0.22

0.09

-1.33

-0.46

-0.84

-0.03

0.00

Manipur

-1.42

-0.24

-1.44

0.23

0.03

-0.57

-0.14

-0.63

0.12

0.08

-1.34

-0.62

-0.86

0.14

0.00

Nagaland

-0.97

-0.04

-1.16

0.20

0.04

-0.41

-0.49

-0.37

0.35

0.10

-0.69

-0.22

-0.54

0.06

0.00

Sikkim

4.08

2.30

1.61

0.16

0.01

3.96

1.17

2.36

0.32

0.11

4.34

2.21

2.26

-0.13

-0.00

Mizoram

-1.05

0.17

-1.42

0.18

0.02

-0.45

0.02

-0.60

0.06

0.07

-0.58

0.25

-0.84

0.01

0.00

Arunachal Pradesh

-1.22

0.00

-1.53

0.28

0.02

-0.57

0.00

-0.71

0.06

0.07

-0.84

0.00

-0.94

0.10

-0.00

Table III shows the level of the four different components of the Holmes, Hsu, and Lee (Forthcoming) index for the 29 Indian states for before and after the construction of the GQ for the three different specifications of our sensitivity analysis; wn is the relative wage (note that we have excluded labor endowment for presentation purposes); Pnpc is the aggregate price index in state n if all firms charged marginal cost; µbuy represents the expenditure-weighted average markup charged on goods purchased in n state n; µsell represents the revenue-weighted average markup charged on goods produced in state n; Pn is simply the aggregate n price index in state n.

References Atkeson, A., and A. Burstein (2008): “Pricing-to-market, trade costs, and international relative prices,” The American Economic Review, 98(5), 1998–2031. Datta, S. (2012): “The impact of improved highways on Indian firms,” Journal of Development Economics, 99(1), 46–57. Edmond, C., V. Midrigan, and D. Y. Xu (2014): “Competition, markups, and the gains from international trade,” Discussion paper, National Bureau of Economic Research. Holmes, T. J., W.-T. Hsu, and S. Lee (Forthcoming): “Allocative Efficiency, Markups, and the Welfare Gains from Trade,” Journal of International Economics. Kothari, S. (2013): “The Size Distribution of Manufacturing Plants and Development,” Working Paper.

15