Transcript Chatman presentation.. - Nexus: Researching Networks, Economics

Do public transport
investments cause
agglomeration economies?
Daniel G. Chatman, Department of City and Regional
Planning, U.C. Berkeley
Symposium on Transportation Investment and
Economic Development
April 2, 2012 at U.C. Berkeley
How increasing travel speed
affects cities
• Increases accessibility, decreasing the costs
of accessing markets and of interactions
between firms and households
– UK def. of agglomeration; no spatial change
• May lead to relocation of economic activity
(or shaping of growth), creating or
intensifying agglomerations
– Depends on development/occupancy
responses
How might transit affect
agglomerations?
• Mostly, by making already-central locations
more accessible:
• …By increasing the number of workers that
can efficiently access/egress workplaces
and other locations
• …By reducing the amount of land required
for roads and parking, allowing for other
productive land uses
Agglomeration economies (AEs)
and AE mechanisms
• Increasing returns to agglomerating firms/
HHs, some of which are external to them.
– e.g. higher productivity per worker
• Various AE mechanisms e.g., firms join
cluster to find workers; attract more workers,
increasing labor pool size; other firms benefit
• AE mechanisms are of interest because not
all are likely to affected by travel, or travel
by all modes
How might transit influence
agglomeration economies?
• Question: mere spatial redistribution, or
(global) increase in productivity?
• Agglomeration economies are positive
externalities, so possibly undersupplied
• Transit might facilitate walking-based
interactions by increasing localized density
near stops
– Knowledge spillovers, transactions costs of
vertical disaggregation
How might transit influence
agglomeration economies?
Agglomeration mechanism
Likely facilitated by transit projects?
Input sharing
No, unless transit projects reduce road
congestion affecting freight
Knowledge spillovers
Indirectly (local firm concentrations;
speed of business travel?)
Labor market pooling
Yes, by increasing the size of the labor
pool within commuting distance
Reduced transactions costs Indirectly, by facilitating local and walkaccessible firm concentrations
How might transit influence
agglomeration economies?
Agglomeration mechanism
Likely facilitated by transit projects?
Input sharing
No, unless transit projects reduce road
congestion affecting freight
Knowledge spillovers
Indirectly (local firm concentrations;
speed of business travel?)
Labor market pooling
Yes, by increasing the size of the labor
pool within commuting distance
Reduced transactions costs Indirectly, by facilitating local and walkaccessible firm concentrations
Estimating transit’s effects on
productivity via agglomeration
• Collected data from all US metro areas
• Estimated the relationship between transit
and agglomeration, and between
agglomeration and productivity
• Used multiple measures of transit,
agglomeration, and productivity
• Employed various methods to control for
endogeneity and other causal factors
• Found very strong net “effects”
Formalization: Agglomeration as
a function of transit
EDi  Ti   Hi   Pi   X i
Pi  Ti   Hi  Pi ,t 1   X i
• ED: employment
density
• T: transit capacity
• H: highway capacity;
• P: population
• X: population
characteristics
Formalization: Productivity as a
function of agglomeration

 ij11
ij ij ij
 H ij
 HL  
YYij  A 1K11 H
log
 logijj   j ijAij ijlogA
log 
 
 Lij
Lij
Lij 1    Lij  1  

Yij
• Y: payroll or GMP
• L: labor supply
• Theta: rental price of
capital
• A: agglomeration
measure (employment
density or population)
• H: human capital



Data sources
• Initial approach: construct a panel of 366
metropolitan areas in the US (only 34 of
which have any rail capacity: 17 commuter
rail, 11 heavy rail, and 27 with light rail)
• Data were messy and required cleaning
• APTA, NTD, LEHD, Census, BEA, NTAD
Transit capacity measures
• Rail route miles (total, per capita, and per
urbanized area)
• Seat capacity (all transit, and rail only; per
capita, and per urbanized area)
• Revenue miles (all transit, and rail only;
total, per capita, and per urbanized area)
Agglomeration measures
• Employment density in the urbanized
portions of the Census-defined principal
cities of the metropolitan area
• Employment density in the urbanized
portions of the metropolitan area
• Population
• NOTE: No time-based measures here; only
distance based (and cruder)
.0015
0
.001
0
500
1000
1500
2000
Employment density - urbanized area
2500
0
0
2000
4000
6000
Employment density - principal city
8000
Productivity measures
• Gross metropolitan product (GDP for metro
area), total and per capita
• Payroll, total and per capita
Urbanized
area
employment
density
Central city
employment
density
Total track miles Negative
Positive
Track miles per
CBSA area
Positive
Negative
Freeway and
Not
arterial capacity statistically
significant
Positive
Population
Not
statistically
significant,
positive for
OLS
Positive
Urbanized
area
employment
density,
omitting NYC
Not
statistically
significant
Central city
Regression
employment diagnostics
density,
omitting NYC
Not
statistically
significant
Not
statistically
significant
(except 1 case
is negative)
Positive
Positive,
larger value
Positive,
larger value
Positive
Not
statistically
significant,
positive for
OLS
Good
instruments,
some overidentification
Urbanized area
over-identified
Notes on transit and
agglomeration models
• Heavy rail most influential; light rail
influential on central city employment
density
• Nonlinear effect: an additional mile of track
in an already-dense area has a bigger
absolute impact
• Little difference in two or four year lags
Findings: Agglomeration and
productivity
• Principal city employment density
significantly correlated with wages and GMP
per capita
• Population even more highly correlated
• No significant relationships with urbanized
area employment density
• Strong evidence of smooth nonlinearity in
productivity models
Industrial sub-sectors
• Manufacturing (NAICS 31-33) and finance
and insurance (52) payroll positively related
to industry-specific principal city
employment density – but only significant
in the case of manufacturing
• Health and social assistance (NAICS 62) per
capita wages negatively related to ownindustry employment density
Average
GDP per capita
Average
GDP per capita
payroll (wages)
payroll (wages)
elasticity
elasticity
elasticity
elasticity
Agglomeration mechanism
Total track miles
Track mile per sqm CBSA
area
Track mile per capita
Employment density (principal
city)
0.00090 0.00529
0.0011 0.0091
0.00246 0.00627
0.0031 0.0108
Track mile per sqm UZA
Rail revenue miles
Total revenue miles
Rail seat capacity per capita
Motor bus seat capacity per
capita
0.0015 0.0050
0.0056 0.0220
0.0013 0.0065
0.0119 0.0226
Population
0.004 0.0060
0.0153 0.0260
0.0036 0.0077
0.0327 0.0267
0.01262 0.02271
0.0145 0.0222
0.0214 0.047
0.0405 0.0831
0.0171 0.0466
0.0177 0.0142
0.01833 0.04180
0.0211 0.0409
0.0587 0.0556
0.1112 0.0984
0.0469 0.0552
0.0487 0.0168
Dollar value of elasticities
• Marginal dollar value effects range between
$5 and $50 per capita with variables held
at means
– Slightly more than one tenth percent increase
in the wage rate
• Across MSAs, multiplied across workers, net
“effects” are from $10m to $500m per year
Implications for policy
and future research
• Large metropolitan areas with dense
central cities might benefit more from rail
investments
• Constraints on employment densification in
central cities would lower these benefits
• Findings are subject to significant
refinement as we improve the models