Primary Submission Category: Dynamic Treatment Regimes
Evaluating Dynamic System-Level Congestion Pricing Regimes with Endogenous Interference
Authors: Aleksander Holleran,
Presenting Author: Aleksander Holleran*
Vehicle tolling operates as a dynamic treatment regime applied at the system level, where interventions affect shared congestion states that feed back into future behavior. Such settings violate standard causal assumptions through interference and state dependence.
We study dynamic treatment regimes in a transportation setting using an agent-based simulation with endogenous state evolution. The simulation is embedded in the physical road geometry and resolves vehicle movement at one-second intervals, allowing congestion to emerge from spatially constrained driving behavior and mode choice. Time-varying tolls are determined by rules that map observed conditions to prices. Under dynamic regimes, tolls are updated continuously. Congestion functions as a shared, time-varying state through which policy effects propagate, generating interference across agents and over time.
We compare policy regimes across synthetic seasons of varying demand, crash, and closure conditions. Seasons allow agents to learn expected travel times from experience, generating behavioral responses to policy-driven congestion relief. Regime performance is assessed using aggregate welfare measures that combine travel time and toll payments, evaluated with both individual-specific values of time (VOT) and median VOT to support distributionally neutral comparisons. This framework supports regime-level policy evaluation in settings where interference, learning, and feedback are intrinsic features.