Executive Summary
Researchers at NYU Tandon have developed MATSim-NYC, the first open-source, multi-agent simulation testbed for New York City. By simulating the daily itineraries of over 8 million synthetic agents, the model reveals that proposed congestion pricing plans could reduce car trips by twice as much as previously estimated. However, the simulation also highlights a stark divide in benefits between Manhattan residents and the rest of the city, raising important questions about social equity.
The Challenge of Modeling Urban Congestion
New York City is on the verge of implementing landmark congestion pricing, but predicting its impact is notoriously difficult. Traditional "static" travel demand models are like looking at a photograph; they show where people are, but struggle with the "movie" of urban life—how traffic spillbacks at 8:00 AM cause a traveler to leave earlier the next day, or how the surge in For-Hire Vehicles (FHVs) like Uber and Lyft changes the math for subway riders.
Existing tools often lack sensitivity to time-of-day dynamics or the "cross-elasticities" of emerging mobility options like Citi Bike. To design fair and effective policies, city agencies need a "virtual testbed" that can simulate individual decisions and traffic physics simultaneously.
Methodology: The MATSim-NYC Framework
The researchers utilized the Multi-Agent Transport Simulation (MATSim) framework, which treats every traveler as an "agent" with a specific daily itinerary (e.g., home -> work -> gym -> home).
Key technical innovations include:
- Mesoscopic Queue Modeling: Instead of complex fluid dynamics, the model treats road links as queues with specific capacities, allowing for large-scale simulation without excessive computational requirements.
- SPSA Calibration: They used Simultaneous Perturbation Stochastic Approximation to iteratively tune road capacities until the simulation matched real-world traffic counts.
- Scaling: Due to the massive computational load of NYC, the team used a 4% sample population (~320,000 agents) and adjusted network capacities accordingly.
Data Integration and Study Area
The model is a "mashup" of several high-fidelity datasets, including a synthetic population of 8 million agents based on Census/ACS data and the Regional Household Travel Survey (RHTS). Road geometry was sourced from OpenStreetMap (OSM), while transit schedules were derived from MTA GTFS feeds. The model was validated against INRIX speed data, NYCDOT bridge/tunnel volumes, and MTA subway turnstile counts.
Simulation Results and Policy Implications
The simulation suggests a car trip reduction of 127,000, nearly double the 59,000 predicted by previous static models. This suggests that when you account for agents actively avoiding congestion and switching modes, the policy is much more "potent."
- **The $14 Threshold:** Increasing the peak fee from $9.18 to $14 actually decreased citywide consumer surplus. While it incrementally benefited Manhattanites by further clearing their streets, it reached a point of diminishing returns for the rest of the city.
- Winners and Losers: The model reveals that while the average benefit is positive, roughly 37-40% of the population suffers a net loss in "traveler surplus" due to the fees. Benefits for those making trips within Manhattan were 110%+ higher than for those outside.
- Revenue: The simulation forecasts annual revenues between $1.09B and $1.55B, aligning with the MTA's targets for funding transit improvements.
Limitations and Future Work
The authors acknowledge several "shortcuts" taken for efficiency, such as "teleporting" walking and biking modes (simulating travel time but not physical presence on the network). Furthermore, the Value of Time (VOT) was set at a flat $29/hr, which may overestimate the surplus for lower-income residents. Future work aims to implement dynamic congestion pricing, where fees fluctuate in real-time based on actual queue lengths, and to better account for income-based impacts on social equity.
Summary updated