Simple vs. Optimal Congestion Pricing

Devansh Jalota, Sharon Di, Adam N. Elmachtoub

TL;DR

While "perfect" dynamic congestion pricing is theoretically superior, this paper proves that simple, static tolls—like those used in New York City and London—are remarkably effective. By analyzing canonical traffic models, the authors show that static tolls typically capture 80-90% of the revenue of complex dynamic schemes while keeping social costs within a narrow margin of the absolute minimum.

What Problem Does This Paper Address?

Economists have long argued for "Pigouvian tolls," where drivers are charged the exact marginal cost they impose on others. In a perfect world, this means tolls should fluctuate every minute based on real-time traffic. However, implementing this is a logistical nightmare: it requires massive sensor infrastructure and places a heavy "cognitive burden" on drivers who just want to know how much their commute costs before they leave the house.

Because of these hurdles, every major city with congestion pricing (NYC, London, Stockholm, Singapore) uses much simpler, often static, tolling structures. The central question this paper tackles is: How much are we actually leaving on the table? By choosing simplicity over mathematical optimality, are cities losing out on billions in revenue or causing massive unnecessary delays?

Methods

The authors bridge the gap between theory and practice by analyzing two foundational frameworks in transportation economics:

  1. Vickrey’s Bottleneck Model: A classic model that tracks how queues form and dissipate at a single point (like a bridge). Crucially, the authors add a "public transit outside option" (e.g., a subway) to make the model reflect modern cities.
  2. Macroscopic Fundamental Diagram (MFD): A city-scale model where traffic flow is treated as a function of overall vehicle density across an entire network, rather than just one road.

The researchers performed the mathematical heavy lifting to derive closed-form solutions for both revenue-optimal and system-cost-optimal tolls. This allowed them to prove "worst-case" performance guarantees—mathematical bounds that hold true regardless of specific traffic patterns.

Datasets

To move beyond pure theory, the authors calibrated their models using high-fidelity, real-world data from two iconic transit hubs:

  • San Francisco–Oakland Bay Bridge: Used to test the bottleneck model. Data included PeMS traffic sensor data and BART ridership figures.
  • New York City Congestion Relief Zone (CRZ): Used for the city-scale MFD model. This involved a massive synthesis of NYC Subway ridership, taxi/for-hire vehicle trip records (TLC), and OpenStreetMap data to calculate total roadway capacity in Lower Manhattan.

Key Insights and Ideas

  • The Power of the "Outside Option": The presence of a viable subway or bus system is the "secret sauce" that makes static tolls work. If transit is a decent alternative, a static toll effectively nudges "on-the-fence" drivers toward the train, naturally smoothing out the traffic peak without needing the toll to change every five minutes.
  • Revenue vs. Welfare: Interestingly, tolls designed to maximize revenue (money for the city) often end up being very close to those designed to maximize social welfare (minimizing total travel time). For policymakers, this means they don't necessarily have to choose between a balanced budget and a fast commute.
  • Robustness of Simplicity: The paper proves a "half-revenue guarantee." In the bottleneck model, even the worst possible static toll will still collect at least 50% of the revenue of a perfect, infinitely complex dynamic toll.
  • The "Discomfort Multiplier": The performance of these tolls is highly sensitive to how much people hate taking the train (the "discomfort multiplier"). As transit becomes less attractive, the gap between simple and optimal pricing tends to widen.

Key Results

The study’s findings are a major win for "keep it simple" advocates:

  • Revenue Capture: In real-world simulations for NYC and the Bay Area, static tolls achieved 80-90% of the revenue generated by an "ideal" dynamic system.
  • System Efficiency: Total system costs (time lost + vehicle costs) were only 8-20% higher under static tolls compared to the theoretical minimum.
  • Current Policy Validation: When looking specifically at the $9 NYC toll and the $8.50 Bay Bridge toll, the models suggest these "simple" choices are currently capturing roughly 98% of the potential benefits of dynamic tolling.

Limitations and Future Work

The authors acknowledge a few areas where the models could be further refined:

  • Homogeneous Users: The current model assumes every driver has the same "value of time." In reality, a CEO and a delivery driver value an hour in traffic differently.
  • Deterministic Demand: The models assume we know exactly when people want to travel. They don't yet account for "stochastic" demand—the random spikes caused by accidents or weather.
  • Transit Crowding: The "outside option" (the subway) is assumed to have a fixed cost. Future work could incorporate "hypercongestion" on trains, where a crowded subway becomes less attractive as more people switch from cars.

Overall, the paper provides a powerful theoretical backbone for the simple tolling systems already in use, suggesting that the "loss" from avoiding complex dynamic pricing is much smaller than previously feared.

Summary updated

Abstract

Congestion pricing has emerged as an effective tool for mitigating traffic congestion, yet implementing welfare or revenue-optimal dynamic tolls is often impractical. Most real-world congestion pricing deployments, including New York City's recent program, rely on significantly simpler, often static, tolls. This discrepancy motivates the question of how much revenue and welfare loss there is when real-world traffic systems use static rather than optimal dynamic pricing. We address this question by analyzing the performance gap between static (simple) and dynamic (optimal) congestion pricing schemes in two canonical frameworks: Vickrey's bottleneck model with a public transit outside option and its city-scale extension based on the Macroscopic Fundamental Diagram (MFD). In both models, we first characterize the revenue-optimal static and dynamic tolling policies, which have received limited attention in prior work. In the worst-case, revenue-optimal static tolls achieve at least half of the dynamic optimal revenue and at most twice the minimum achievable system cost across a wide range of practically relevant parameter regimes, with stronger and more general guarantees in the bottleneck model than in the MFD model. We further corroborate our theoretical guarantees with numerical results based on real-world datasets from the San Francisco Bay Area and New York City, which demonstrate that static tolls achieve roughly 80-90% of the dynamic optimal revenue while incurring at most a 8-20% higher total system cost than the minimum achievable system cost.
BibTeX
@misc{jalota2026simple,
  author = {Devansh Jalota and Sharon Di and Adam N. Elmachtoub},
  title = {{Simple vs. Optimal Congestion Pricing}},
  year = {2026},
  eprint = {2602.21495},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2602.21495},
}

Loading related content...

Figure