Resolving the Mobility-Equity-Emissions Trilemma: A Causal Machine Learning Framework for Dynamic Tolling Policy

Christian Mulomba MukendiarXiv 2026 · DOI · arXiv

TL;DR

This paper introduces a Causal Machine Learning (CML) framework designed to solve the "trilemma" of urban tolling: balancing traffic efficiency, social equity, and carbon emissions. By leveraging Double Machine Learning (DML) and a novel adaptive quota mechanism, the authors demonstrate that cities can achieve significant emission reductions and efficiency gains without unfairly marginalizing residential areas.

What Problem Does This Paper Address?

Urban planners face a persistent challenge when implementing dynamic congestion pricing. While varying tolls in real-time can effectively reduce traffic delays, optimizing purely for efficiency often leads to two major side effects:

  1. Equity Gaps: Tolls tend to cluster in high-traffic commercial zones, leaving residential areas underserved or creating unfair economic burdens.
  2. Environmental Externalities: Emission impacts are usually calculated after the fact rather than being integrated into the pricing logic itself.

Existing systems often rely on "cordon-based" pricing, which is too coarse to handle the fine-grained spatial and temporal realities of a city. The authors aim to move beyond these static paradigms by creating a framework that treats a city as a dynamic network, optimizing for mobility, equity, and emissions simultaneously.

Methods

The core of the framework is Double Machine Learning (DML), a technique used to estimate Conditional Average Treatment Effects (CATEs). DML is particularly useful here because it allows researchers to control for high-dimensional "confounders"—variables like road quality, weather (rain/events), and time-of-day—that might bias the estimated impact of a toll.

The authors implemented a two-stage "nuisance model" approach:

  • Stage 1: Predicting the outcome (traffic delay) and the treatment (toll activation) using independent ML models (LassoCV and XGBoost).
  • Stage 2: Residualization, where the influence of the confounders is removed to isolate the true causal effect of the toll.

To address the equity portion of the trilemma, they introduced an Adaptive Quota Mechanism. Unlike standard models that only activate tolls where they are most profitable, this mechanism enforces minimum activation shares for residential (≥25%) and mixed-use (≥15%) zones while capping commercial zones (≤60%) to prevent over-concentration.

Datasets

The framework was developed and validated using a high-resolution synthetic dataset simulating 90 days of urban traffic:

  • Spatial Grid: A 10x10 grid representing 100 distinct zones (60% residential, 25% commercial, 15% mixed-use).
  • Features: 14 base features (vehicle counts, road quality, peak hour indicators) expanded to 89 features through preprocessing, including temporal lags and spatial interaction terms.
  • Validation: The synthetic results were validated against real-world data from Singapore’s Electronic Road Pricing (ERP) system, achieving a delay reduction Root Mean Square Error (RMSE) of 12.3 minutes.

The simulation parameters and their justifications are detailed in the paper's policy simulation setup:

Policy Simulation Parameters and Justification

Key Insights and Ideas

  • Commercial vs. Residential Sensitivity: The models revealed that commercial zones yield nearly twice the delay reduction of residential zones. This explains why purely profit-driven algorithms ignore residential areas.
  • The Power of Spillovers: A major "aha moment" was discovering that 3rd-order spatial effects (delays occurring three zones away) are dominant predictors of current traffic. This suggests that even "suboptimal" tolling in residential zones can create network-wide benefits by preventing delay propagation.
  • Policy-Invariant Toll Range: The authors identified a robust toll range ($0–$5) where performance remains stable across different demand scenarios. This is a critical insight for policymakers, as it suggests implementation can be simplified without losing efficacy.
  • The Quota "Sweet Spot": The Quota Hybrid policy emerged as the most practical solution. It expanded residential coverage 34-fold compared to standard hybrid models while still preserving 93% of the total delay savings.

Key Results

The framework demonstrated impressive quantitative gains across all three pillars of the trilemma:

  • Mobility: Achieved 380.6 delay-hour reductions while respecting budget constraints.
  • Emissions: Reduced CO2 emissions by 743.5 metric tons (valued at approximately $141,265 in climate benefits).
  • Equity & ROI: The Quota Hybrid policy maintained a strong Return on Investment (ROI) of 1.31, proving that equity-driven constraints do not have to lead to economic failure.

A comparison of the different policy thresholds and their performance is summarized below:

Policy Threshold and ROI Comparison

Limitations and Future Work

While the framework is robust, the authors acknowledge several areas for further exploration:

  • Real-World Testing: The study relies heavily on synthetic data. While validated against Singapore's system, the authors emphasize the need for testing in live, diverse urban environments.
  • Toll Level Sensitivity: The current study did not extensively vary the dollar amount of the tolls to test sensitivity; future work should explore how different price points affect driver behavior.
  • Demand Suppression: The model assumes "near-unitary elasticity," meaning vehicle counts stay relatively stable while flow improves. In reality, high tolls might suppress demand entirely, a factor that needs deeper integration into the causal models.

Summary updated

Abstract

We propose a causal machine learning framework to integrate dynamic tolling for mobility, equity, and climate objectives, resolving a long-standing policy trilemma. Using Double Machine Learning, we estimate zone-type-specific treatment effects, showing that commercial zones yield nearly twice the delay reduction of residential zones. Our adaptive quota mechanism expands residential coverage 34-fold while preserving 93% of total delay savings and sustaining high return on investment. The framework delivers climate co-benefits, reducing emissions by 743.5 metric tons of CO2 (valued at 141 USD) via congestion-sensitive pricing under moderate traffic conditions with sensitivity bounds from 557.6 to 929.3 tons. Crucially, we uncover a policy-invariant toll range where performance remains robust across demand scenarios, simplifying implementation. This study reveals critical, previously unidentified insights, such as a policy-invariant toll range, that can significantly simplify real-world implementation. Validated on synthetic data mirroring real-world networks, our methodological advances lay the foundation for future public deployment.
BibTeX
@misc{mukendi2026resolving,
  author = {Christian Mulomba Mukendi},
  title = {{Resolving the Mobility-Equity-Emissions Trilemma: A Causal Machine Learning Framework for Dynamic Tolling Policy}},
  year = {2026},
  eprint = {2610.33994},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.33994},
}
Figure