Evaluating the Impact of Urban Transit Infrastructure:
Evidence from Bogotá’s TransMilenio
Nick Tsivanidis
University of California, Berkeley & IGC
6th IGC-World Bank Urbanization Conference
September 2019
Urban Transit Infrastructure
Empirical Questions:
1. What are the aggregate effects of improving urban transit?
Urban Transit Infrastructure
Empirical Questions:
1. What are the aggregate effects of improving urban transit?
• 2.5 billion people will move into cities by 2050, most in developing countries
Urban Transit Infrastructure
Empirical Questions:
1. What are the aggregate effects of improving urban transit?
• 2.5 billion people will move into cities by 2050, most in developing countries
Urban Transit Infrastructure
Empirical Questions:
1. What are the aggregate effects of improving urban transit?
• 2.5 billion people will move into cities by 2050, most in developing countries
Urban Transit Infrastructure
Empirical Questions:
1. What are the aggregate effects of improving urban transit?
• 2.5 billion people will move into cities by 2050, most in developing countries
Urban Transit Infrastructure
Empirical Questions:
1. What are the aggregate effects of improving urban transit?
• 2.5 billion people will move into cities by 2050, most in developing countries
Urban Transit Infrastructure
Empirical Questions:
1. What are the aggregate effects of improving urban transit?
• 2.5 billion people will move into cities by 2050, most in developing countries
Urban Transit Infrastructure
Empirical Questions:
1. What are the aggregate effects of improving urban transit?
• 2.5 billion people will move into cities by 2050, most in developing countries
Urban Transit Infrastructure
Empirical Questions:
1. What are the aggregate effects of improving urban transit?
• 2.5 billion people will move into cities by 2050, most in developing countries
2. How are the gains distributed across the low- and high-skilled?
Urban Transit Infrastructure
Empirical Questions:
1. What are the aggregate effects of improving urban transit?
• 2.5 billion people will move into cities by 2050, most in developing countries
2. How are the gains distributed across the low- and high-skilled?
• Bogotá in 1995: low-skilled 25% more likely to commute using informal bus...
• Which were 32% slower than cars
Regression
TransMilenio: World’s Most Used Bus Rapid Transit System
Opened across 3 phases in 2000s
TransMilenio: World’s Most Used Bus Rapid Transit System
Opened across 3 phases in 2000s
Similar speed to subways, but
Faster and Cheaper to build
TransMilenio: World’s Most Used Bus Rapid Transit System
Opened across 3 phases in 2000s
Similar speed to subways, but
Faster and Cheaper to build
Currently being built in many developing
countries
TransMilenio: World’s Most Used Bus Rapid Transit System
Opened across 3 phases in 2000s
Similar speed to subways, but
Faster and Cheaper to build
Currently being built in many developing
countries
Combine with detailed tract-level data
to examine impact
Approach of This Paper
1. New Commuter Market Access approach from general equilibrium theory to measure
effects of transit infrastructure within cities
Approach of This Paper
1. New Commuter Market Access approach from general equilibrium theory to measure
effects of transit infrastructure within cities
• Individuals: Access to Jobs. Firms: Access to Workers
Approach of This Paper
1. New Commuter Market Access approach from general equilibrium theory to measure
effects of transit infrastructure within cities
• Individuals: Access to Jobs. Firms: Access to Workers
• Advantages vs Standard Distance-to-Station Approach
Approach of This Paper
1. New Commuter Market Access approach from general equilibrium theory to measure
effects of transit infrastructure within cities
• Individuals: Access to Jobs. Firms: Access to Workers
• Advantages vs Standard Distance-to-Station Approach
• Regression Framework: Log-linear reduced form between CMA and outcomes
Approach of This Paper
1. New Commuter Market Access approach from general equilibrium theory to measure
effects of transit infrastructure within cities
• Individuals: Access to Jobs. Firms: Access to Workers
• Advantages vs Standard Distance-to-Station Approach
• Regression Framework: Log-linear reduced form between CMA and outcomes
2. Quantitative general equilibrium model of a city:
• New Features: Low/High-skill workers + Multiple transit modes
Approach of This Paper
1. New Commuter Market Access approach from general equilibrium theory to measure
effects of transit infrastructure within cities
• Individuals: Access to Jobs. Firms: Access to Workers
• Advantages vs Standard Distance-to-Station Approach
• Regression Framework: Log-linear reduced form between CMA and outcomes
2. Quantitative general equilibrium model of a city:
• New Features: Low/High-skill workers + Multiple transit modes
3. Quantification+Counterfactuals:
• Quantify welfare effects through value of time savings (VTTS) + realllocation and
general equilibrium effects
Main Results
1. Aggregate Effects: Large gains, worth the cost
• Welfare " 1.63%, Output (net of costs) " 1.44%
• VTTS accounts for 60-80% of welfare gains, remainder by reallocation+GE effects
Main Results
1. Aggregate Effects: Large gains, worth the cost
• Welfare " 1.63%, Output (net of costs) " 1.44%
• VTTS accounts for 60-80% of welfare gains, remainder by reallocation+GE effects
2. Distributional Effects: High and low skilled benefit about the same
• Higher public transit use of low-skilled offset by differences in commuting elasticities
and GE effects
Main Results
1. Aggregate Effects: Large gains, worth the cost
• Welfare " 1.63%, Output (net of costs) " 1.44%
• VTTS accounts for 60-80% of welfare gains, remainder by reallocation+GE effects
2. Distributional Effects: High and low skilled benefit about the same
• Higher public transit use of low-skilled offset by differences in commuting elasticities
and GE effects
3. Key Policy Implication: Large gains to integrated transit + land use policy
• Average welfare gain 19% higher under more accommodative zoning policy
• Revenue from Land Value Capture scheme covers 10-40% of const. costs
Roadmap
1. Empirical Approach & Results
2. Quantification and Counterfactuals
Simple Model to Guide Empirics
• Ingredients:
• Many discrete locations indexed by i = 1, . . . , N (e.g. blocks or census tracts)
• Locations differ in amenities, productivities, commute times, floorspace
• Individuals decide where to live and work
• Firms in each location decide how much labor+commercial floorspace to hire
• House prices and wages adjust to clear markets
Simple Model to Guide Empirics
Individuals: Choose between pairs of where to live i and work j that depends on:
• Residential Location Characteristics: Amenities, house prices in i
• Workplace Location Characteristics: Wages in j
• Pairwise Commute Characteristics: Cost of commuting from i to j
Simple Model to Guide Empirics
Supply of Residents: Depends on amenities ui , house prices rRi and access to well-paid jobs
Ri (RCMA)
⇣ ⌘✓
LRi / ui rRi 1 Ri
Simple Model to Guide Empirics
Supply of Residents: Depends on amenities ui , house prices rRi and access to well-paid jobs
Ri (RCMA)
⇣ ⌘✓
LRi / ui rRi 1 Ri
Supply of Labor: Depends on wages wj and access to workers Fj (FCMA)
LFj / wj✓ Fj
Simple Model to Guide Empirics
Supply of Residents: Depends on amenities ui , house prices rRi and access to well-paid jobs
Ri (RCMA)
⇣ ⌘✓
LRi / ui rRi 1 Ri
Supply of Labor: Depends on wages wj and access to workers Fj (FCMA)
LFj / wj✓ Fj
Computing CMA: Unique values of RCMA and FCMA can be recovered from data (LFj , LRi )
and parameterization of commute costs (e.g. commute times computed in ArcMap).
Distance-Based Treatment Effect: Close vs Far
Distance to TransMilenio Line
<500m from line
>500m from line
Distance-Based Treatment Effect: Close vs Interm. vs Far
Distance to TransMilenio Line
<500m from line
500m - 1km from line
>1km from line
Residents: Change in lnRCMA
Hot: Larger increase
Cool: Smaller increase
Emp Dist Emp by Ind TM Map
Firms: Change in lnFCMA
Hot: Larger increase
Cool: Smaller increase
Res Dist Coll Share
Reduced Form Representation
Equilibrium can be written as:
ln YRi = R ln Ri + eRi
ln YFi = F ln Fi + eFi
where
⇥ ⇤ ⇥ ⇤0
• ln YRi = ln LRi ln rRi and ln YFi = ln LFi ln rFi are changes in
endogenous outcomes
• R . F are reduced form coefficients capturing direct+indirect effects of CMA on outcomes
• eRi , eFi are structural errors containing changes in amenities/productivities
Isomorphisms
Data
Dataset Source Year Variables
1993,
Population General Census/DANE Residential Population by Education Group
2005,2015
1995, 2005,
Commuting DANE Mobility Survey Trip-diaries (trip and person characteristics)
2011, 2015
Property value and characteristics, land use,
Housing Cadastral Department 2000-2013
land and floorspace area
General Census 1990, 2005 Employment and industry (universe of estab.)
Employment
(Firms) Business Registry
2000, 2014 Employment and industry (formal estab.)
(Chamber of Commerce)
Employment DANE Household Surveys Worker demographics and employment
2000-2014
(Workers) (ECH/GEIH) characteristics
Commute
City Maps - Times by mode computed in ArcMap
Times
House Prices Times Rel Speeds TM Use Inc Employment Congestion Trip Char Image
Establishing Causal Impact of BRT
• Challenge: BRT routes chosen by government, may be correlated with other drivers of
economic activity
• Approach:
1. Predict TransMi routes using (i) historical tram and (ii) least cost construction routes
2. Exploit opening across 3 phases to show no impacts until lines open
3. Use changes in accessibility due to new lines >1.5km away
• Additional Outcomes: In paper, look at effect on commute distances, wages and
gentrification
CMA Captures Differential Response Across Space
Residential Floorspace Prices vs RCMA
.04
Change in Log Residential Floorspace Price,
.02
residualized
-.02
-.04
-.1 -.05 0 .05 .1
Change in Log RCMA, residualized
Res Pop vs RCMA
.06
Change in Log Residential Population,
.04
.02
residualized
-.02
-.04
-.06
-.1 -.05 0 .05 .1
Change in Log RCMA, residualized
Commercial Floorspace Price vs FCMA
Change in Log Commercial Floorspace Price, .04
.02
residualized
-.02
-.04
-.1 -.05 0 .05 .1
Change in Log FCMA, residualized
Employment vs FCMA
.1
Change in Log Establishments,
.05
residualized
-.05
-.1
-.1 -.05 0 .05 .1
Change in Log FCMA, residualized
Roadmap
1. Empirical Approach & Results
2. Quantification and Counterfactuals
Extended Model
To speak to distributional consequences, paper then develops model with multiple types of
workers, firms and transit modes
Extended Model
To speak to distributional consequences, paper then develops model with multiple types of
workers, firms and transit modes
Summary of Identification:
1. Mode Choice Parameters: Responsiveness of mode choices to differences in commute
times
2. Commuting Elasticity: Responsiveness of change in commute flows to changes in
commute times
3. Agglomeration Externalities: Responsiveness of change in productivities + amenities to
exogenous shift in supply of residents and labor across city provided by CMA instruments
Aggregate Impacts of TransMilenio
Panel A: Aggregate Gains
Output 1.82%
Average Welfare 1.63%
Rents 1.91%
Panel B: Costs vs Benefits
Capital Costs (mm) 1,137
NPV Operating Costs (mm) 5,963
NPV Total Costs (mm) 7,101
NPV Net Increase Output (mm) 26,808
Net Increase Output 1.44%
Note: Table shows the (negative of the) value of the percentage change in each variable from removing
Open City
Welfare Decomposition
• Theoretical Result: In an efficient equilibrium, the first order welfare impact in the full GE
model is simply the VTTS
• Empirical Question: How important are reallocation + GE effects?
Welfare Decomposition
• Theoretical Result: In an efficient equilibrium, the first order welfare impact in the full GE
model is simply the VTTS
• Empirical Question: How important are reallocation + GE effects?
Average Welfare Inequality
First Order Approximation (VTTS) 1.308 -0.172
General Equilibrium 1.628 0.085
percentage change in average welfare and inequality from adding TransMilenio moving the equilibrium without it. Each entry is computed by
nitial equilibrium) and then adding it back in under the different approaches. The first row is the first order welfare approximation using the CM
mation from Implication:
• proposition X. TheReallocation + general
third line is the full GE effects areresponse.
equilibrium important for large shocks + distributional
consequences
Policy Counterfactuals 1: Network Components
1. Geography Matters: Low-skilled benefit most from lines connecting where they live with
areas of dense employment
2. Large Returns to Complementary Services: “Feeder” network increases welfare more
than any other line
Policy Counterfactuals 2: Land Value Capture
• In Bogotá, change in transit w/o complementary change in zoning laws
• ) No significant response in housing supply to TM details
Policy Counterfactuals 2: Land Value Capture
• In Bogotá, change in transit w/o complementary change in zoning laws
• ) No significant response in housing supply to TM details
• Land Value Capture:
• “Development Rights Sale” - Gvt sells permits to build at higher densities near stations
• Successful in Asian cities to (i) finance construction and (ii) increase housing supply
Policy Counterfactuals 2: Land Value Capture
• In Bogotá, change in transit w/o complementary change in zoning laws
• ) No significant response in housing supply to TM details
• Land Value Capture:
• “Development Rights Sale” - Gvt sells permits to build at higher densities near stations
• Successful in Asian cities to (i) finance construction and (ii) increase housing supply
• 2 Policies: Allocate the same amount of new floorspace permits via
1. Increase density by 30% within 500m of stations
2. Increase density proportional to predicted change in CMA
Policy Counterfactuals 2: Land Value Capture
Gvt. Rev Gvt. Rev
Avg Welfare Inequality Closed City Open City
Baseline 1.63% 0.09%
LVC-Distance 1.71% 0.03% 5.72% 17.82%
LVC-CMA 1.93% 0.01% 10.21% 41.07%
Note: Gvt revenue is fraction of construction costs.
1. Average welfare gain 19% larger under LVC
2. Welfare + Revenue Gain greater under CMA-based scheme
3. Low-skilled benefit the most
Conclusion
• My Contribution:
• Develop new empirical approach to measure effects of transit
• Quantitative model to assess aggregate and distributional effects across groups
• Combine rich microdata + construction of world’s largest BRT to assess causal impact
• My Findings:
1. Investments in transit such as BRT have large aggregate net benefits to cities
2. Low- and high-skilled benefit about the same ) less precise policy tool to target the poor
than implied by standard approach
3. Complementary change in zoning policies ) maximize returns from these investments
