Job Market Paper

Efficient Allocation of Freeway Capacity through Truthful Elicitation of Willingness to Pay: Using a Vickrey-Clarke-Groves Mechanism to Manage Express Lanes     

Abstract:  The proliferation of freeway express lanes throughout the country provides transportation agencies with the means to improve overall driver welfare by addressing heterogeneity in travel time preferences. There have been relatively few attempts to maximize driver welfare through pricing, however, due to the significant challenge associated with obtaining real-time, comprehensive driver preferences. This paper demonstrates the workability of a truth-telling mechanism for efficiently allocating this public good. I conduct a traffic experiment consisting of an interactive multi-user driving simulator, and attempt to allocate human subjects to lanes of a freeway using an optimal tolling scheme where users reveal their valuation of the road through a Vickrey-Clarke-Groves mechanism. I work from a framework where individuals have heterogenous values of time that are unknown to the regulator, preventing the optimal allocation of individuals across lanes. When implemented, I find significant deviation from truth-telling which is largely caused by difficulty in understanding the complexity of the mechanism as well as stochasticity in travel time outcomes. Nevertheless, I show that given parameters estimated from the experiment, the mechanism may still dominate alternatives. 

Working Paper

Learning equilibria in multi-state traffic networks 

Abstract: This paper investigates if the presence of multiple states in traffic networks adversely impacts the speed at which users learn route-choice equilibria. Discrete network states are prevalent in the real-world due to circumstances such as lane-closures and recurrent time-varying demand patterns, and in many cases the network equilibria correspond to social optima. Thus, delays in route-choice equilibrium convergence could result in significant welfare losses system-wide. To address this topic, data were generated from several sessions of a repeated binary route-choice experiment with human subjects. Exogenous random state changes were introduced as discrete, varied reductions in roadway capacity. The sessions were comprised of “simple” network treatments with only two states, and “complex” network treatments with five states. I estimated reinforcement learning models from the experimental data and found that learning is significantly impaired in the complex five-state treatment but not the simple two-state treatment, and that the learning errors are related to memory. I then simulated the estimated learning behavior from both treatments over an extended horizon in the same environment for an “apples-to-apples” comparison. The simulations show that the error-prone learning from the five-state treatment results in disproportionately slower and sometimes non-existent equilibrium convergence compared to learning from two-states. These findings suggest that network operators could increase user welfare by implementing measures that address the higher cognitive demands of learning over multiple states.