Transport investment: urban form, trade and growth Vernon Henderson - - PowerPoint PPT Presentation

Transport investment: urban form, trade and growth

Vernon Henderson Brown University July 2012 Lectures at EIEF, Rome

Motivation

• World Bank: about 20% of lending to transport infrastructure

(more than social lending)

• Huge investments: China: $200b (2007) in infrastructure

investments per year. Much in cities

• Within cities
• Shapes cities for decades/centuries to come (LA vs. NY; Shanghai vs. Beijing)
• Little known about impacts in developing countries
• Mayors/planners:

(1) Optimal configurations: transit, radial and ring highways?

(2) How much to spend?: fundamental law of congestion (3) Sprawl/compactness and environmental footprint (3) Impact of specific types of investments?

• city shape
• urban growth

Motivation

• Across cities/districts
• Who grows / has higher income: on vs. left off the network
• Issue of general equilibrium effects
• Hinterland development (penetration roads, partial equilibrium)
• Overall network and effect on trade costs and total welfare
• Can be general equilibrium
• Optimal spending
• Positive analysis of impacts: perhaps benefits?
• Effect on comparative advantage

– More transport: produce heavier stuff

Past literature

• Limited

– Early location theory and network literature – 1980-1990’s literature on the effect of public investment on growth

• No identification
• NEG theory gives a framework
• Empirical literature we look at is all from last 5
• years. Starts with Baum-Snow (2007) focused on

identification of causal effects

– Smaller picture?

Outline 1– Across cities districts

• Effect on inter-city/district trade, prices, and either growth or

income

– Places that have better access have better outcomes in a later cross- section

• Income levels up but not growth rates (Banerjee, Duflo, Qian 2012)
• Trade flows up (usual NEG virtuous circle); shift to heavy stuff (Duranton

and Turner 2012)

• Inference tough: effect of placing a transport ray & then

instrument with historical lines

– Modern lines follow historical; historical built with an eye to trade and linking places with a competitive advantage

• Look before and after transport construction
• Price gaps narrow (export vs. import point), exports rise, real incomes rise

(Donaldson 2010)

• But who got treated (not random)
• Other exogenous change in transport costs

– Effect on income levels of oil price rise in places near versus distant from coastal markets places (Storeygard 2012)

Outline 2– Within cities

• Effect of transport investments within cities

– Urban form and decentralization

• “Sprawl” and spatial reorganization of production

– Baum-Snow (QJE, 2007); Baum-Snow, Brandt, Henderson, Turner & Zhang (2012)

• More efficient commuting and within city movement of goods

– Gains in welfare, growth » Duranton and Turner, RES 2012

• Inferences reasonable

– No random assignment: try for pseudo-randomization – Growing (or not growing) cities (good vs. bad unobservables) receive investments

• Instruments: historical lines/plans built between cities for

national defense or even trade, but with no intent to improve modern intra-city efficiency or facilitate decentralization (suburbanization)

Inter-city/district trade: Income effects

• “On the road…..” Banerjee, Duflo, Qian NBER, 2012

– If “quasi-randomly assigned better access, do you have better long run outcomes?”

• Rails connecting traditional major cities and treaty ports in

late 19th and early 20th century (but then later roads, also (maintained) canals & rivers)

– Look 1986-2006 for a sub-set of counties – Look at places near and further from straight lines connecting these nodes, to assess impact of future rail (and road) construction – Exclude nodes. Each county’s distance to line, segment city, later rail, river, coast, border » Not just trade, but public service delivery » Benefits differential between places limited by factor mobility (vs. Donaldson)

Neat instrument

Treaty ports: Shanghai, Ningbo, Fuzhou, Guangzhou

Interpretation

• (Not) mode (rail) specific? Vs. reduced form?

– 2-3 modes (?); one instrument: distance to straight line

• Approximation since not use same controls
• Same instrument for several modes, vs. just rail

effect?

1 1 1 1 1

ln(dist to mode for county ) = ln(area )+ ln(dist hist line ) ln(GDP pc ) ln(area )+ ln(dist hist line ) ˆ ˆ mode effect on ln(GDP pc ) /

ij i p i i ij i p i i

j i a b a b c X e j b b δ ε λ + + = + + + ≈

• Move from 25th to 75th percentile of distance to line: GDP pc drops by 19%

Many controls for access

Evaluating magnitudes

• Net general equilibrium effects?

– Winners over losers

• If factors mobile why see any GDP pc differences?

– Simple model: labor, land, capital. Same returns to labor and capital if mobile

• Then just differences in return to land [labor/land ratio down

to equalize VMP of labor]

• Want total GDP (per unit land)?

– Goods composition (4 or more factors of production)

• Low skill production off network?

“Roads and trade”

Duranton, Morrow and Turner, 2011

• NEG virtuous circle model: reduced transport costs increase

demand for different varieties everywhere

• “Propensity” to export related to in-city transport– kms of roads

within city

• Value and weight of trade flows between 65 regions in USA for 2007

City i produces one variety

• f each sector k

Issues in estimation

• current highway distance. Other R’s have kms of roads

within city

• For kms of roads within city, roads allocated to growing cities

(China) or as make-work to poorly performing cities (USA)

– Instrument with kms from 1947 plan, historical rails & exploration routes within city

• Old routes: cheaper to build more roads in city (e.g. right-of-way, bed)
• Built for inter-city trade, not movement within city (but looking at

inter-city trade)

• Between-city highway distances. Issue of highway placement:

serve regions with comparative advantage in trade and with historical cultural links?

– Instrument with historical rail distances (built for agriculture and natural resource extraction)

• City i uses j’s inputs historically and were provided better links

historically and today

ij

D

Trade flow equations

OLS: small insignificant effects. Actual roads over-allocated to non-exporters and poorly performing cities (make-work and rules)

“Railroads of the Raj”

Donaldson 2010

• British colonial rails: military vs. extractive purposes

– Incredible change in transport costs

• India poor roads and no canal system
• Cool things: conceptual framework and transport modeling
• Conceptual framework:
• Usual NEG preferences, Eaton-Kortum (2002) Ricardian

comparative advantage model. Amount of output z from a unit of land of commodity k in region 0 is realization of draw Z

0 ,

iid draws by , ,&

k k

A

• k

j θ

• In autarky produce things badly for consumption; with trade can

focus just on what do well. As trade costs fall, buy more products from

• ther places (as they are lowest cost producer).

Transport

• Roads, rivers, coast, rails

– Rivers not canals; silted and yearly variation in quality (flooding and water level drop) – Coast: steamships after 1840 for major ports

• GIS network analysis of modes, used between

region o and f, with nodes and arcs

– Shortest path vs. lowest cost.

Assume cost prop. to distance. Unit costs:

• 1. Solve for

, , . For each vector , pick cost minimizing combo (network) of modes over different parts of route from to . Then pic

rail road sea river

• f

α α α α α = k to minimize sq residuals in price equation. α

Some results

• Look at salt prices with several types with unique
• rigin and many destinations

– Differential in destination prices fully reflect transport costs differentials from origins

• Modal differences: rails, road, sea and river

LCRED: Effective distance in railway units

Welfare

• Section on trade flows– ignored here
• Look at welfare.

– Simple in model (only land in production) – Rents equal nominal ag. output per unit land

• Local yields and national prices of grains (?)

– Deflate by price index (prices for products)

• includes transport cost based on prior estimates
• Change in “rents” when rail introduced in a

district; 235 districts from 1870-1935:

– Change in output composition vs. price index

“Farther on down the road”

Storeygard, 2012

• Sub-Saharan African coastal nations where

primate city is a port

– Idea is export agricultural products to port for export – But sold and serviced in local interior cities each serving an agricultural hinterland

• Take road network as given

– Ask what happens to city incomes if transport costs rise exogenously?

Road systems in Tanzania

Dar es Salaam

Cost change:

Elasticity of VTC per unit distance with respect to price of fuel =. 35

How measure income: Lights!

• 287 cities over 17 years (92-07)
• Cities are light blobs with census populations

Henderson, Storeygard and Weil, AER 2012

• elasticity for low-middle

income countries is just over 0.3

• LD and FE same

Lights at night

Henderson, Storeygard and Weil, AER 2012

• In many countries poor GDP data, and no sub-national data
• Lights reflect economic activity (mostly use of electricity)
• Source

– US air force satellites

– Annual measures from NOAA for 1992-2008

– Each satellite observes every location on planet between 8:30- 10pm local time – Use on dark half of lunar cycle; filtered (cloud cover, forest fires, auroral activity); below Arctic Circle

• Measure

– Intensity as 0-63 annual digital number for every 30-second

• utput pixel (.86 sq. km at equator)
• Averaged over valid evenings of year (and satellites)
• Average over pixels (weighted by land area of pixel)

Korea

• Over time growth
• South vs. north
• Fishing

11/2010 33 Night Lights. Henderson Storeygard and Weil, Brown University

Interpretation Issues

• controls for other national trends (given nature of oil price rise)
• lights produced with electricity and may use fuel
• sub-sample: access and/or use of hydro power,
• where oil produced (sub-sample: remote from oil production)
• paved roads or not (endogenous)
• distance to primate if on paved; otherwise distance to nearest

top quintile city in country

STOP

“Roads, railways and decentralization of Chinese cities” Baum-Snow, Brandt, Henderson, Turner, Zhang, 2012

• Understand the forces driving urban spatial form in a

developing country context (vs. Baum-Snow (2007) for USA)

• Specifically:
• Estimate the effects of highway and rail networks
• n decentralization in Chinese cities since 1990
• Various configurations: radial rays, rings, total

kilometers of network

• Examine effects of transport infrastructure on

decentralization of:

• population
• industrial GDP

Motivation

• 1. Compact cities (vs. sprawl)

– Investments affect spread: lower commuting, rents, density – But externalities?

• “Green cities”: energy and environmental foot-print

– Fuel under-priced? – Under-invest in mass transit, buses (trans millennium)?

• fundamental law: elasticity of VMT with respect to

transport expansion = 1

– But welfare improving: more movement, better schedules

• Chinese perspective: Use of ex-urban farm land: “food

security”

– Value of food is more than consumption: national defense » Subsidize food production?

Motivation

• 2. Spatial (re-) organization of industry for economic growth:

– How best use information rich environments of central cities? – Early on: industrialization of central cities

• Maoist plan but has real reasons

– Rich information environment: import & adapt foreign technologies – Better infrastructure and institutional development

– Later: industry decentralizes (Duranton & Puga, 2001)

• Cheaper land and labor.
• Central cities: professional and business service centers

– Transport infrastructure key to facilitating

decentralization

• Hansen 1987, Lee and Choe 1990, Henderson Kuncoro and Nasution 1996
• Beyond case studies, and address causality

Preview of Results

• Highway rays
• each additional radial highway displaced about 4.0 percent
• f central city populations to suburban regions
• No affect on industrial decentralization (yet)
• Highway kms
• No affect conditional on rays (form of system)
• Railway rays
• Each railroad ray displaced 26 percent of industrial GDP from

the central city to suburbs

• No affect on population
• Ring roads in “suburbs”
• Decentralization of both population and GDP
• feed to highway and rail rays
• Buses and trolleys in central city retard decentralization

Context

• 1990 central cities (“city proper” in 1990) for 2005 prefectures

– for urbanized prefectures in Han China

• After 1990, big within prefecture movements of rural

populations into central cities (urbanization).

– Migration from outside the prefecture limited by hukou system

• Over 10% annual growth in real GDP per capita

– Huge increases in demand for housing

• In 1990 no limited access highways in China

– Little commuting

• Goods shipped by rail (1990: 5% ton-kms by “highway”) .

Context

• Urban reform in early 1990’s: create “land market” and move

industry to market basis.

• Couldn’t really decentralize before 1990: rural vs. urban
• Reforms:

– Start to separate living from work place: start of commuting

• Limited cars and no scooters in urban districts

– Factories sell off land and move to ex-urban areas; FDI

• 1990 cities not recognizable today

OLD

Study Area of China

Figure 1a: Study area. The yellow area includes the prefectures included in our study. Red lines indicate prefecture boundaries. Green indicates the extent of constant boundary 1990 prefectural cities. Figure 1b: Beijing and surrounding prefectures. Counties and prefectures are drawn based on 2005 boundaries for each. The green counties make up the 1990 city prefecture city, the central city. The yellow counties are counties converted to urban districts after 1990. The dots are the city centers, the locations of which are discussed later.

• Use counties (urban districts)
• Constant boundary:
• 1990 central cities

(expand)

• 2005 prefectures

(changes over time)

• Promoted cities

Table 1: Growth in Aggregate Lights, Population and GDP by Location, 1990-2010 3 periods of growth ’90 to ’ 00, ‘05, ‘10

City Proper Prefecture Remainder

Panel A: 2010 Sample of 210 Prefectures Lights Population Lights Population (Mean in 1990) (982,333) (2,995,989) 1990-2010 105% 55% 172% 5% Panel B: 2000 Sample of 257 Prefectures

Lights Population Lights Population (Mean in 1990) (955,683) (2,953,557) 1990-2010 102% NA 165% NA

Panel C: Sample of 108 Prefectures With GDP Data GDP Industrial GDP GDP Industrial GDP (Mean in 1990) (20.5) (12.3) (17.5) (7.0) 1990-2005 530% 417% 605% 794%

Figure 1a: Population growth by location, 1990- 2010

Figure 1b: Industrial sector GDP growth by location, 1990- 2005

Beijing Area

Beijing Prefecture Tianjin Prefecture Langfang Prefecture Baoding Prefecture Zhangjiakou Prefecture Chengde Prefecture Tangshan Prefecture Qinhuangdao Prefecture

Effective urban area is neither a county nor prefecture

(Brightest cell)

Digitization of Transport Infrastructure Information, Beijing 2010

• We have digitized large scale national maps from 1924, 1962, 1982, 1990, 1999,

2005 and 2010

• Results are detailed enough for our purposes, measuring roads and rails within a

few km of their actual locations

• In 1962 and 1999, we use all classes of roads reported on the map whereas in 2010

we use only express highways and national highways

Measuring Radial Road/Rail Capacity

• Draw rings 5 and 10 km

from the CBD. The contribution to radial index is the minimum of the number of times the network crosses each ring [6 (inner), not 8 (outer)].

Measuring Circumferential Road Capacity

• Draw rings 5 and 9 km from the CBD. Draw rays heading west and northwest from the CBD

(45o). The index for this quadrant and distance is the minimum of the number of times roads cross these rays.

• Repeat for each of the other 3 quadrants
• Repeat for 9-15km and 15-25 km distance bands [min. angle of ring road to ray is 54o (vs. 90)].
• Sum the results for intersections outside the central city,
• our index = 1 if this sum is positive, 0 otherwise

2005

• gap in center South and going North
• connect cities; no spurs

Shipping by rail vs. roads: Expressways

1999

Rails 2005

Additional Data

• Demographic data from the 1982, 1990, 2000 and 2010 censuses
• GDP data from various printed yearbooks and Michigan Online

China Data Archive

• Three samples:
• 1. For 1990-2000 CC population growth: Full sample of 257 (of

286) Han Chinese prefectures

• Start with a pop of more than 50,000 urban residents (lose 26)
• Have suburban area (lose 3)
• 88 promoted

2. For 1990-2010 CC population growth: Limited coverage in

the 2010 census data available to us: use a sample of 210

• 3. For 1990-2005 CC industrial GDP growth: CC sample of 205

for GDP and 187 for industrial sector GDP

• Boundary changes with no rural GDP numbers
• Because of limited GDP data availability, use lights at night

data as an alternate measure of GDP

Empirical Strategy for Roads

For the effects of roads, we estimate FD regression (adjust Baum-Snow, 2007):

C = “1990 definition city proper” P = “2005 definition prefecture” r = measure of roads y = outcome of interest x = control variables

• No highways in 1990. Assume (i.e., )

1

A <

1 2 1990 1 1 1990 2 2 1990 1990 1990 1 1 1990 2 2 1990 1990

ln ln (Alonso-Muth) ln ( ) ( ) ( ) ( ) (planning) ln ln ln

tC t tP t t C P t C t t P P t t

y A Ar A y B x y A A A A r A A y B B x y A A r Ar A y A y B x B x δ ε δ ε ε = + + + + + = + ∆ + + ∆ + + ∆ + + ∆ + + ∆ = ∆ + ∆ + ∆ + ∆ + ∆ + ∆ + ∆ + ∆

1990

= r

t t

∆ = r r

Empirical Strategy for Railroads

When evaluating the effects of railroads, we start with (from above)

measure of railroads:

• Little rail construction after 1990 and can’t confidently identify

new and lost rails; No instruments for measured changes in rails.

• Fortunately, in 1990, most Chinese cities were still completely

planned economies:

• little commuting
• Rural-urban legal separation: SOE’s can’t move out
• As a result, our estimates showing how market forces respond to

the railroad network are comparable to those for highways.

1 1 1990 2005 1

0; ; estimate A A r r A ∆ + = ≈

1 1 1990 2 2 1990 1990

ln ln ln

t C t t P P t t

y A A r A r A y A y B x B x ε ∆ = ∆ + ∆ + ∆ + ∆ + ∆ + ∆ + ∆ + ∆

1 2 2 1990 1990

ln ln ln

t C t t tP P t t

y A A r A y A y B x B x ε ∆ = ∆ + + ∆ + ∆ + ∆ + ∆ + ∆

Main identification issue

• Highways built to serve growing (central) cities. These

cities would then look like they are centralizing because of the new roads when in fact the reverse is true

• As a solution, we instrument for 1999, 2005 or 2010

highways and rails with corresponding 1962 measures ( )

1990 1990 1990 1990 1990 1990

( , | , , , ln ,ln ) ( , | , , , ln ,ln )

t t t P P t t t P P

Cov z r x q x y y Cov z x q x y y ε ∆ ∆ ≠ ∆ ∆ ∆ =

z

Using the 1962 Road Network for Instruments

• Strong instruments: New highways follow old paths. Cost issue:

Right of way and access points (city gates) already in place.

• Valid strategy if conditional on control variables, there are no

unobservables correlated with the 1962 network that drive decentralization

• vs. 1982, 1924
• 1962 road network was of very poor quality
• 2 lane maximum; not all paved
• primarily used for local shipping of agricultural goods from

rural hinterlands to nearby prefectural cities (not commuting)

• Empirically, 1962 roads are predicted by 1990 agricultural

population outside of the city proper

• Might affect decentralization directly
• Important control variable in all regressions

Controls

• Land area: CC & pref

– some geography: rain, elevation changes, distance to coast

• 1990 [92] base pref. population and lights (not GDP)
• 1990 [92]-200x pref. growth in population and light
• Issues:

1. Agricultural hukou population outside CC

• source of internal CC pop. growth, correlated with ’62 roads,

+ highly correlated with tot pop

• 2. Issue of exogeneity of change in prefecture light &

population

ln &

t P t

y x ∆ ∆

Using the 1962 Rail Network for Instruments

• 1962 rail network

– Private and foreign ventures, Japanese, civil war, Russians – Post 62, third front into Sichuan region

• Used primarily for

– Long-distance shipping of raw materials and fuels – Trade of industrial products (short or long haul) – Not related to access to suburban sidings

• There were more 1962 railroads in provincial capitals.

– Nodal shipping points – Control for this indicator in all regressions

Table 2: First Stage Regressions 2010 Hwy Rays 1999 Hwy Rays 2005 Rail Rays 1999 Hwy Rings (1) (2) (3) (4) Highway Rays in 1962 0.37*** 0.32*** 0.022

• 0.0043

(0.080) (0.079) (0.082) (0.018) Railroad Rays in 1962 0.24* 0.17** 0.50*** 0.0045 (0.13) (0.078) (0.057) (0.023) Highway Rings in 1962 0.55

• 0.56

0.013 0.44*** (1.04) (0.37) (0.26) (0.12) Log Central City Area 0.039 0.16 0.069

• 0.073***

(0.16) (0.11) (0.12) (0.023) Log Prefecture Area 0.25 0.23

• 0.13
• 0.030

(0.20) (0.14) (0.16) (0.043) Log(1990 Agric. Hukou Pop 0.0098 0.36** 0.19 0.0059 Outside Central City) (0.14) (0.14) (0.12) (0.020) Log(1992 Prefecture Lights) 0.44***

• 0.042

0.030

• 0.0087

(0.15) (0.13) (0.17) (0.032) ∆ log(Prefecture Population)

• 0.18

1990-2010 (0.60) ∆ log(Prefecture Population) 1.76***

• 1.55***

0.15 1990-2000 (0.48) (0.41) (0.13) ∆ log(Prefecture Lights)

• 0.15

0.020

• 0.46

0.044 1992-20xx (0.53) (0.32) (0.36) (0.061) Provincial Capital Indicator 1.91*** 1.44*** 0.096 0.13 (0.42) (0.46) (0.24) (0.096)

Observations 210 257 205 257 R-squared 0.31 0.29 0.29 0.16

Table 3: OLS Relationships Between Highway Rays and Central City Populations Classic bias ∆ ln(CC Pop), 1990-2010 ∆ ln(CC Pop), 1990-2000 (1) (2) (3) (4) Highway Rays in 2010 0.014

• 0.011

(0.011) (0.0073) Highway Rays in 1999 0.022*** 0.011 (0.0076) (0.0073) Log Central City Area

• 0.12***
• 0.063***

(0.021) (0.016) Log Prefecture Area 0.031 0.016 (0.030) (0.014) Log(1990 Agric. Hukou Pop 0.070** 0.035* Outside Central City) (0.029) (0.018) Log(1992 Prefecture Lights) 0.021 0.016 (0.026) (0.012) ∆ log(Prefecture Population) 0.79*** 0.80*** 1990-20xxa (0.11) (0.085) ∆ log(Prefecture Lights) 0.090* 0.072** 1992-20xx (0.046) (0.034) Provincial Capital Indicator 0.087 0.0017 (0.059) (0.032)

Observations 210 210 257 257 R-squared 0.01 0.56 0.03 0.39

Table 4: IV Estimates of Effects of Highway Rays on Central City Population ∆ ln(CC Pop), 1990-2010 ∆ ln(CC Pop), 1990-2000

(1) (2) (3) (4) (5)

Highway Rays in 2010

• 0.030
• 0.046**
• 0.038*

(0.030) (0.022) (0.022) Highway Rays in 1999

• 0.047***
• 0.042***

(0.014) (0.013) Log Central City Area

• 0.12***
• 0.13***
• 0.054***
• 0.052***

(0.020) (0.020) (0.016) (0.016) Log Prefecture Area 0.043 0.058* 0.032*** 0.055*** (0.027) (0.034) (0.011) (0.012) Log(1990 Agric. Hukou Pop 0.076*** 0.064* 0.065*** 0.053*** Outside Central City) (0.028) (0.035) (0.020) (0.018) Log(1992 Prefecture Lights) 0.036 0.037 0.013 0.0053 (0.027) (0.030) (0.016) (0.019) ∆ log(Prefecture Population) 0.79*** 0.75*** 0.91*** 0.91*** 1990-20xxa (0.088) (0.092) (0.095) (0.098) ∆ log(Prefecture Lights) 0.083* 0.076 0.073* 0.050 1992-20xx (0.045) (0.053) (0.038) (0.039) provincial capital indicator 0.16** 0.16** 0.097** 0.097** (0.081) (0.076) (0.042) (0.039)

Log (precipitation) 0.029 0.029** central city elevation range 8.1e-06

• 0.000038**

prefecture elevation range 1.3e-07 8.9e-06 log(distance to coast)

• 0.0086
• 0.0084

Observations 210 210 210 257 257

1st stage (Kleibergen-Paap rk Wald) 30.3 23.2 26.3 17.0 15.2

Comments on Table 4

• Radial highways retard centralization
• LR equilibrium model. Coefficients for 90-00

and 90-10 the same.

– But highway definition changes

• Robust to samples

– Drop big or small – Drop CBD-gap cities: stronger – Drop cities where CC expanded

• Coefficients drop 25%: interpretation and idea of ATE

Other infrastructure forms for population decentralization

• kms of highway
• rails

Production decentralization

• Industry
• Highways?
• Rails?

– Rays vs. kms.

Table 6: IV Estimates of Transport Infrastructure Effects on Production Decentralization ∆ln(CC Ind Sect GDP) ∆ln(CC GDP) ∆ln(CC Lights)

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Highway Rays in ‘05

• 0.028

0.0049 0.015 (0.083) (0.048) (0.029)

Railroad Rays in ‘05

• 0.26**
• 0.17***
• 0.044*

(0.11) (0.064) (0.026)

log(‘05 km of railroads

• 0.82***
• 0.54***
• 0.22*

in prefecture)

(0.24) (0.15) (0.12)

Log Central City Area

0.12* 0.11* 0.11 0.015 0.015 -0.0011 0.071*** 0.076*** 0.072*** (0.066) (0.068) (0.065) (0.047) (0.048) (0.047) (0.024) (0.023) (0.023)

Log Prefecture Area

• 0.29** -0.31***

0.22

• 0.20** -0.20***

0.17

• 0.090* -0.087** 0.040

(0.13) (0.11) (0.17) (0.077) (0.070) (0.11) (0.048) (0.042) (0.088)

Log(‘90 Agric. Hukou

0.28*** 0.31*** 0.29*** 0.13** 0.17*** 0.16*** -0.0004 0.024 0.030

Pop outside CC

(0.11) (0.11) (0.11) (0.056) (0.063) (0.063) (0.042) (0.036) (0.038)

Log(‘92 Pref. Lights)

• 0.010

0.054 0.15 0.072 0.090 0.15* 0.020 0.020 0.063 (0.13) (0.11) (0.14) (0.076) (0.063) (0.082) (0.040) (0.032) (0.041)

∆ log(Pref. Pop.)

1.67** 1.37** 0.80 0.51** 0.31

• 0.0054 -0.053 -0.088
• 0.33*

1990-2000

(0.82) (0.69) (0.64) (0.24) (0.20) (0.23) (0.17) (0.14) (0.17)

∆ log(Pref. Lights)

0.31 0.14 0.046 0.30** 0.17 0.081 0.93*** 0.89*** 0.81***

1992-2005

(0.22) (0.28) (0.24) (0.15) (0.17) (0.16) (0.080) (0.079) (0.085)

provincial capital

• 0.40**
• 0.31
• 0.22
• 0.057

0.036 0.12

• 0.11*
• 0.066 -0.033

(0.19) (0.21) (0.19) (0.097) (0.13) (0.13) (0.065) (0.051) (0.052)

Observations 187 187 184 205 205 202 257 257 248

First stage F

16.7 48.3 16.4 16.9 76.4 25.8 35.8 90.1 20.9

OLS rays coefficient for industrial GDP: -.06

Rays: highways vs. rails

• Why do highway rays lead to population decentralization?

– Some high income suburbanize (we think) – Local hukou people from rural sector can commute (full or part time): less need to move into CC

• Why do highway rays not affect production decentralization?

– Highways for long distance transport only accelerates in last 7-8 years

• If rails lead to production decentralization, why not lead

(indirectly) to population decentralization? i.e. , live near work place Muted by complexity

• Not absolute job loss in city: compositional change

– Huge growth of business & financial service sector overall , focused on CC

• Separation of where work and live: commuting now “huge”
• Increasingly employ those who would never have lived in center.

– Inter-provincial migrant workers (not so much locals) in the factories in dorms or rural housing: never would go to CC to begin with.

Table 9: Robustness

∆ln(CC Pop) ∆ln(CC Pop) ∆ln(CC Ind GDP) ∆ln(CC GDP) 1990-2010 1990-2000 1990-2005 1990-2005

(1) (2) (3) (4)

Panel A: Instrumenting for Prefecture Population Growth Highway Rays in 2010 or 1999

• 0.033
• 0.042***

(0.034) (0.015) Railway Rays in 2005

• 0.25***
• 0.14**

(0.094) (0.057)

N 210 257 187 205

First stage F 2.49 8.63 8.01 2.65 Panel B: Placebo Regressions Using Outcomes 1982-1990 Highway Rays in 2010

• 0.017

(0.04) Highway Rays in 1999

• 0.012

(0.028)

N 237 237

First stage F 7.28 12.6

• potential bias in panel A:
• include growth control: more negative
• exclude: bias to zero
• instrument
• 1982 stock (outside own prefecture) of rural (supply)
• vs. urban hukou population (competition) within 300 kms.
• not Bartik

“Growth” (level effects) of cities

• Highways enhance urban efficiency
• Au and Henderson (RES, 2006)

– Cities export manufacturing varieties produced with labor, capital, and non-traded locally produced diverse service/component inputs

1 1 / (1 ) 3/2 1 1 1 1 effective labour, market potential unit cost: after commuting

• trans. local

services

( ) ( )

y

x

VA Q c R MP r A N atR N

γ α ε γ ρ β σ α ω α α α α − + + − − − − − − −

= −            

“Urban growth and transportation”

Duranton & Turner RES, 2012

• 2003-1983 growth in population & employment



1 1 1 ,( ) 1 base value steady-state : speed of convergence 1 2 2 ,( ) 2 3 3 ,( ) 5 4 3 it it it i t x it it it it it i t x it it it i t x it it

n n A ar c x n r r A r c x n r A c x c z c n

λ

λ ε θ η ε ε

+ − + − −

− = + + + + − = + + + + = + + + +    

• Convergence to stationary population (vs differentiate LR equilibria)
• Slow adjustment of labor migration (and investments in roads)
• No g.e. effects: draw people from rural sector where alternative

utility is fixed

• No market potential effects

Issues

• Now roads are measured by total kilometers of road

within in 1983 metro area

• Roads allocated to growing (or depressed) places

– Instrument with 1947 highway plan (connect principal metro areas to serve national defense & connect border points). Kms within metro area – Also 1898 rails and routes of major explorations from 1528-1850 [strength– places easier to build roads]

• Treats base year (lagged dependent variable) as

exogenous…..

– x controls include 1970, 60,..20 population levels; – but worry about modern era shocks (deindustrialization, population skill composition, movement S&W)