83146 DECEMBER 2013 • Number 131 Highways and Spatial Development Ejaz Ghani, Arti Grover Goswami, and William R. Kerr This Economic Premise examines the link between highways and spatial development. The Golden Quadrilateral (GQ) highway project in India—5,846 km of highways linking four major urban hubs—improved the connectivity and market accessibility of districts close to the highway compared to those more removed. Non-nodal districts located within 0–10 km from the highway experienced substantial increases in entry of new enterprises. The highways facilitated a more natural sorting of industries that are land and building intensive, improved efficiency in the manufacturing industries, and encouraged decentralization of urban transformation by making intermediate cities more attractive. Understand- ing these patterns is important for policy makers, because well-targeted infrastructure projects can improve resource reallocation, accelerate spatial development, and also promote shared prosperity. Adequate transportation infrastructure is an essential ingre- of roads connecting many of the major industrial, agricul- dient for economic development and growth. Beyond simply tural, and cultural centers of India. Figure 1 maps the GQ facilitating cheaper and more efficient movement of goods, network. people and ideas across places, transportation infrastructure Ghani, Goswami, and Kerr (2013) studied how proxim- impacts the distribution of economic activity and develop- ity to the GQ network in non-nodal districts affected the orga- ment across regions to the extent that agglomeration econo- nization of formal manufacturing activity using establish- mies and efficient sorting can be realized, the levels of compe- ment counts, employment, and output levels, especially tition among industries and concomitant reallocation of among newly entering plants that chose their locations before inputs toward productive enterprises are achieved, and much or after the upgrades. This work on the organization of the more. Rapidly expanding countries like India and China of- manufacturing sector also considers industry-level sorting ten face severe constraints in terms of their transportation and the extent to which intermediate cities in India are be- infrastructure. Many business leaders, policy makers, and aca- coming more attractive to manufacturing plants. The study demics report inadequate infrastructure as a critical obstacle examines the impact on the sector’s performance through to sustained growth that must be resolved with public fund- measures of average labor productivity as well as through total ing—but to date, there is limited understanding of the eco- factor productivity (TFP). nomic impact of those infrastructure projects. Many researchers have shown that transport investments This note summarizes the main findings of the study play an important role in spatial development, city competi- conducted by Ghani, Goswami, and Kerr (2013) on the im- tiveness, and urbanization. Henderson (2010) finds that in- pact of the GQ project on organized manufacturing in In- dustrial decentralization in the Republic of Korea is attribut- dia. The GQ project sought to improve the connection of able to massive transport and communications infrastructure four major cities in India: Delhi, Mumbai, Chennai, and investments. Baum-Snow et al. (2012) show that transport Kolkata. The GQ system comprises 5,846 km (3,633 miles) infrastructure aided the decentralization of industrial produc- 1 POVERTY REDUCTION AND ECONOMIC MANAGEMENT (PREM) NETWORK    www.worldbank.org/economicpremise Figure 1. Map of GQ and NS-EW tion and population in Chinese cities. Several other studies sity districts to districts that are less congested, allowing indus- find positive economic effects in non-nodal locations due to trial activity to spread more equally across space. transportation infrastructure in China (for example, Banerjee The Ghani, Goswami, and Kerr (2013) study provides et al. [2012] and Roberts et al. [2012]), while Datta (2011) important contributions to the literature. First, and perhaps finds similar results for India. Duranton and Turner (2012) most important, this study is the first to bring plant-level data show that transportation investments increase city population to the analysis of highway projects. This is not feasible in the in the United States. Desmet et al. (2012) have argued that most-studied case of the United States, because the major manufacturing in India is slowly moving away from high-den- highway projects mostly predate the United States’ detailed 2 POVERTY REDUCTION AND ECONOMIC MANAGEMENT (PREM) NETWORK    www.worldbank.org/economicpremise census data. As a consequence, state-of-the-art work like Chan- Panel A of table 1 reports the base results. All estima- dra and Thompson (2000) and Michaels (2008) use aggregate tions include the initial level of activity in the district for the data and broad sectors. The later timing of the Indian reforms appropriate outcome variable as a control to flexibly capture allows utilization of detailed plant data, providing more in- issues related to economic convergence across districts.1 The sight on many dimensions such as entry behavior and distribu- columns of table 1 list dependent variables: columns 1–3 tions of activity. An example of the resulting benefit is the present measures of total activity in each district; columns improvement in allocative efficiency for industries initially 4–6 present measures of new entry specifically; columns 7 positioned along the GQ network after the reforms. Second, and 8 present the average productivity measures; and col- existing work mostly identifies how the existence of transpor- umns 9 and 10 present wage and labor cost metrics. The first tation networks impacts activity, but Ghani, Goswami, and row in the base results of panel A shows increases in nodal Kerr go a step deeper and also discuss the likely impacts from district activity for all metrics. The higher standard errors of investments improving existing networks. The returns to this these estimates, compared to the rows beneath them, reflect latter type of investment are very large and growing. the fact that there are only nine nodal districts. These results are not emphasized, given that the upgrades were built around Data and Methodological Framework the connectivity of the nodal cities.2 The GQ project began in 2001, was two-thirds complete by The primary emphasis is on the highlighted row, which 2005, and mostly finished in 2007. The study analyzed re- considers districts that are 0–10 km from the GQ network, peated cross-sectional surveys of organized manufacturing but are not nodal districts. Columns 1–3 find increases in establishments carried out by the government of India at the aggregate activity of these districts. The coefficient on those points in time. Ghani, Goswami, and Kerr (2013) stud- output is particularly strong and suggests a 0.4 log point in- ied the organized sector surveys that were conducted in crease in output levels for districts within 10 km of the GQ 1994–95 and covered several surveys from 1999–2000 to network in 2007–9 compared to 2000, relative to districts 2009–10. This coverage shows the performance of Indian more than 50 km away from the GQ system. The estimates manufacturing before the GQ upgrades began in 2001 as well for establishment counts and output in districts 0–10 km as during and after the GQ upgrades. The work on the GQ from the GQ network exceed the employment responses. was officially complete in 2005 (at the 90 percent mark) and These employment effects fall short of being statistically sig- 97 percent complete by 2007. nificant at a 10 percent level, and this is not due to small The core sample examines plant-level data from 311 dis- sample size, as there are 76 districts within this range. Gen- tricts. The key focus is on non-nodal districts very close to the erally, the response around the GQ changes favored output GQ network and on comparing them to districts that were over employment. farther away. The study specifically compares non-nodal dis- Columns 4–6 examine the entry margin by quantifying tricts 0–10 km from the GQ network to districts 10–50 km levels of young establishments and their activity. The study away (and in some specifications, with additional concentric finds much sharper entry effects than the aggregate effects in rings to 200 km away). Additional sources of variation come columns 1–3, and these entry results are very precisely mea- from the sequence in which districts were upgraded, differ- sured. The districts within 0–10 km of the GQ have a 0.8– ences in industry traits within the manufacturing sector, and 1.1 log point increase in entry activity after the GQ upgrade differences in the traits of non-nodal districts 0–10 km from compared to districts more than 50 km away. Columns 7 and the GQ network. 8 report results for the average labor productivity and TFP in the districts 0–10 km from the GQ network. These average Impacts of Highway Upgrades values are weighted and thus primarily driven by the incum- Long-differenced estimations compare district activity in bent establishments of the districts. In general, analysis shows 2000, the year prior to the start of the GQ upgrades, with dis- an increase in labor productivity for the district as a whole trict activity in 2007 and 2009 (average across the years). that is also evident in a comparison of columns 2 and 3. On About 95 percent of the GQ upgrades were completed by the the other hand, no TFP-level growth is apparent. Columns 9 end of 2006. The analysis uses two years after the conclusion and 10 show an increase in wages and average labor costs per of most of the GQ upgrades, rather than just the final data employee in these districts. point of 2009, to be conservative. Most outcome variables are For comparison, the third row of panel A provides the expressed in logs, with the exception of TFP, which is ex- interactions for the districts that are 10–50 km from the GQ pressed in unit standard deviations. Estimations report ro- network. None of the effects on the allocation of economic bust standard errors, weight observations by log total district activity that were seen in columns 1–6 for the 0–10 km dis- population in 2001, and have 311 observations representing tricts are observed at this spatial band. This isolated spatial the included districts. impact provides a first assurance that these effects can be 3 POVERTY REDUCTION AND ECONOMIC MANAGEMENT (PREM) NETWORK    www.worldbank.org/economicpremise Table 1. Long-Differenced Estimations of the Impact of GQ Improvements, 2007–9 versus 2000 Log levels of total activity Log levels of young firm activity Log Log cost Log labor Total factor average per Plants Employment Output Plants Employment Output productivity productivity wage employee (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) DV: Change in manufacturing trait listed in column header Base spatial horizon measuring effects relative to districts 50+ km from the GQ network (0,1) 1.467+++ 1.255+++ 1.413+++ 1.640+++ 2.004+++ 2.468+++ 0.138 1.971+++ 0.382+++ 0.393+++ Nodal (0.496) (0.464) (0.480) (0.499) (0.543) (0.621) (0.111) (0.195) (0.065) (0.069) district (0,1) District 0.364+++ 0.235 0.443+++ 0.815+++ 0.882+++ 1.069+++ 0.199+++ 0.163 0.121++ 0.130++ 0-10 km (0.128) (0.144) (0.163) (0.161) (0.198) (0.277) (0.074) (0.195) (0.055) (0.056) from GQ (0,1) District -0.199 -0.325 -0.175 -0.238 -0.087 -0.281 0.157 0.286 0.098 0.095 10-50 km (0.185) (0.222) (0.293) (0.237) (0.314) (0.455) (0.126) (0.280) (0.091) (0.094) from GQ Source: Authors' compilation. Notes: Long-differenced estimations consider changes in the location and productivity of organized-sector manufacturing activity in 311 Indian districts from 2000 to 2007–9 from the Annual Survey of Industries (ASI). Explanatory variables are indicators for distance from the GQ network that was upgraded starting in 2001. Estimations consider the effects relative to districts more than 50 km from the GQ network. Column headers list dependent variables. Young plants are those less than four years old. Labor productivity is total output per employee in district, and TFP is weighted average of Sivadasan (2009) approach to Olley-Pakes estimations of establishment-level productivity with repeated cross-section data. Outcome variables are winsorized at their 1 percent and 99 percent levels, and entry variables are coded at the 1 percent level where no entry is observed to maintain a consistent sample. Estimations report standard errors, have 311 observations, control for the level of district activity in 2000, and weight observations by log total district population in 2001. linked to the GQ upgrades rather than to other aspects, such To better establish the timing of these reforms, the study as regional growth differences. team also constructed two dynamic specifications. First, they Ghani, Goswami, and Kerr (2013) test many variations separately estimated effects for each calendar year to see on these themes. The first trial shows results after controlling whether the growth patterns appear to follow the GQ up- for other district traits, including: national highway access, grades hypothesized to cause them. Effects are measured rela- state highway access, broad-gauge railroad access, and district- tive to the 1994 period and tend to confirm the right timing, level measures from the 2000 census of log total population, as in figure 2, for the output levels of young firms. In an alter- age profile, female-male sex ratio, population share in urban native dynamic specification, the study team identified the areas, population share in scheduled castes or tribes, literacy sections of highway that were completed earlier than others. rates, and an index of within-district infrastructure. The in- Results show that the effects are largest in the district’s where clusion of these controls in the long-differenced estimation is the work was completed by March 2003, followed by those akin to including time trends interacted with these initial co- finished by March 2006, and then the last sections to be built. variates in a standard panel regression analysis. The inclusion of these controls substantially reduces the coefficients for the Figure 2. Dynamics of Log New Output Growth nodal districts. More importantly, the increased activity for 3.0 GQ upgrades 80% complete 0–10 km districts remains quite statistically and economi- start cally important. These patterns also hold when using alterna- 2.5 tive distance bands, including state fixed effects, and similar 2.0 tests. 1.5 Also critical to the analysis is the placebo setting. Sec- tions of the North–South and East–West (NS-EW) highway 1.0 were scheduled to be upgraded at the same time as the GQ 0.5 network, but the work was postponed. The study finds that the districts within 10 km of the NS-EW highway show no 0.0 response. In addition, the results generally confirm the ordi- -0.5 nary least square findings with straight-line instrument vari- -1.0 able estimates that connect nodal cities, which helps with 99 00 01 02 03 04 05 06 07 08 09 particular concerns about the endogenous weaving of the net- 19 20 20 20 20 20 20 20 20 20 20 work toward certain districts with promising potential. Source: Authors' illustration. 4 POVERTY REDUCTION AND ECONOMIC MANAGEMENT (PREM) NETWORK    www.worldbank.org/economicpremise The GQ upgrades also appear to have facilitated a more GQ project in India upgraded the quality and width of 5,846 natural sorting of land- and building-intensive industries km of highways linking four major cities in India. In the pro- from the nodal districts into periphery locations. This general cess, this upgrade improved the connectivity and market ac- urban-rural or core-periphery pattern is evident in many cessibility of districts lying close to the highway compared to countries and is associated with efficient sorting of industry those more removed. Non-nodal districts located within placement. Moreover, this feature has particular importance 0–10 km from the GQ network experienced substantial in- in India because of government control over land and build- creases in entry levels and ambiguous productivity conse- ing rights, leading some observers to state that India has tran- quences. Dynamic specifications and comparisons to the NS- sitioned from its “license Raj” to a “rents Raj.” Given India’s EW highway system mostly confirm these conclusions. The distorted land markets, the heightened connectivity brought GQ upgrades also appear to have facilitated a more natural about by the GQ upgrades may be particularly important for sorting of industries that are land and building intensive from efficient sorting of industry across spatial locations. These the nodal districts into periphery locations and have im- patterns suggest that the GQ upgrades may have helped with proved allocative efficiency in the manufacturing industries the efficient sorting of industries across locations. Ghani, Go- located along the GQ network. The upgrades also appear to be swami, and Kerr (2012) find that infrastructure aids efficient encouraging decentralization by making intermediate cities sorting of industries and plants within districts, and these pat- more attractive to manufacturing entrants. terns show a greater efficiency across districts. Many studies The study by Ghani, Goswami, and Kerr (2013) contrib- have warned about the misallocation in the Indian economy utes to the literature on the economic impacts of transporta- (for example, Hsieh and Klenow [2009]), and these results tion networks in developing economies that is unfortunately suggest that better connectivity across cities/districts may re- quite small relative to its policy importance. Understanding duce some of these distortions. the impacts of a large-scale infrastructure project on econom- The upgrades also appear to encourage decentralization ic activity and the pattern of development is important for by making intermediate cities more attractive to manufac- policy makers and regional analysis because these impacts turing entrants. For instance, moderate-density districts— identify how infrastructure investments shape the spatial like Surat in Gujarat or Srikakulam in Andhra Pradesh—that growth of regions within India and the distribution of indus- border the GQ highway registered a more than 100 percent trialization and income. This study provides quantitative esti- increase in new output and new establishment counts after mates of the likely impact of other highway development proj- the GQ upgrades. On the other hand, the GQ upgrades are ects in India, and the work on the relative impacts across not linked to heightened entry or performance in low-densi- districts by distance to the network offers insights into the ty areas. These results suggest that the improved connectivi- distributional impacts of these infrastructure projects. This ty enables manufacturing establishments to efficiently lo- type of empirical analysis is an essential input for urban plan- cate in intermediate cities, but that localization economies ning and economic policy, which govern the distribution of prevalent for the sector continue to preclude entry in low- economic activity and industrial development of a country. density places. The study’s results on improved spatial sorting can also help Importantly, and the subject of ongoing research, the up- guide policies for promoting stronger productivity growth grades are also associated with better allocative efficiency in and better allocation of scarce resources in India. the organized sector. Allocative efficiency measures the ex- About the Authors tent to which the employment of an industry is contained in the industry’s most productive plants. India generally com- Ejaz Ghani is Lead Economist in the Economic Policy and Debt pares very poorly to advanced economies, such as the United Unit of the Poverty Reduction and Economic Management States, on this dimension. Industries that were initially posi- (PREM) Network at the World Bank. Arti Grover Goswami is a tioned along the GQ show improved allocative efficiency Consultant Economist at the World Bank. William Kerr is an compared to industries initially positioned on the NS-EW Assistant Professor at Harvard Business School. system. This is encouraging for the competitive dynamics in- Notes duced by better infrastructure. 1. In general, however, the estimates show very little sensitiv- Conclusions ity to the inclusion or exclusion of this control. This note summarizes the Ghani, Goswami, and Kerr (2013) 2. Because the coefficients are being measured for each band study on the impact of a large-scale highway project on eco- relative to districts more than 50 km away from the GQ net- nomic activity in the Indian formal manufacturing sector us- work, the inclusion or exclusion of the nodal districts does ing establishment-level survey data from 1994–2009. The not impact the core results regarding non-nodal districts. 5 POVERTY REDUCTION AND ECONOMIC MANAGEMENT (PREM) NETWORK    www.worldbank.org/economicpremise References Ghani, E., A. Goswami, and W. Kerr. 2012. “Is India’s Manufactur- ing Sector Moving Out of Cities?” World Bank, PRE Working Banerjee, A., E. Duflo, and N. Qian. 2012. “On the Road: Access Paper No. 6271, Washington, DC. to Transportation Infrastructure and Economic: Growth in ———. 2013. “Highway to Success in India: The Impact of the China.” NBER Working Paper No. 17897, Cambridge, MA. Golden Quadrilateral Project for the Location and Performance Baum-Snow, N., L. Brandt, V. Henderson, M. Turner, and Q. Zhang. of Manufacturing.” World Bank PRE Working Paper No. 6320, 2012. “Roads, Railroads and Decentralization of Chinese Cit- Washington, DC. ies.” Working Paper, Brown University. Henderson, V. 2010. “Cities and Development.” Journal of Regional Chandra, A., and E. Thompson. 2000. “Does Public Infrastructure Science 50 (1): 515–40. Affect Economic Activity? Evidence from the Rural Interstate Hsieh, C., and P. Klenow. 2009. “Misallocation and Manufacturing Highway System.” Regional Science and Urban Economics 30 (4): TFP in China and India.” The Quarterly Journal of Economics 457–90. 124 (4): 1403–48. Datta, S. 2011. “The Impact of Improved Highways on Indian Michaels, G. 2008. “The Effect of Trade on the Demand for Skill: Firms.” Journal of Development Economics 99 (1): 46–57. Evidence from the Interstate Highway System.” Review of Eco- Desmet, K., E. Ghani, S. O’Connell, and E. Rossi-Hansberg. 2012. nomics and Statistics 90 (4): 683–701. “The Spatial Development of India.” World Bank Policy Re- Roberts, M., U. Deichmann, B. Fingleton, and T. Shi. 2012. “Evalu- search Paper No. 6060, Washington, DC. ating China’s Road to Prosperity: A New Economic Geography Duranton, G., and M. Turner. 2012. “Urban Growth and Transpor- Approach.” Regional Science and Urban Economics 42 (4): tation.” Review of Economic Studies 79 (4): 1407–40. 580–94. The Economic Premise note series is intended to summarize good practices and key policy findings on topics related to economic policy. They are produced by the Poverty Reduction and Economic Management (PREM) Network Vice-Presidency of the World Bank. The views expressed here are those of the authors and do not necessarily reflect those of the World Bank. The notes are available at: www.worldbank.org/economicpremise. 6 POVERTY REDUCTION AND ECONOMIC MANAGEMENT (PREM) NETWORK    www.worldbank.org/economicpremise