Research Reports

Cities face persistent welfare trade-offs in transport policy. Congestion pricing, emissions controls, transit investment for ageing populations, and the integration of autonomous mobility all alter who can travel, when, and at what cost — yet the tools most widely used by planners cannot quantify these effects in welfare terms. The dominant four-step traffic model predicts aggregate vehicle volumes but treats mode and departure time as exogenous inputs, leaving the distributional consequences of policy invisible.

Dynamic Discrete Choice Models (DDCMs) for activity-based travel demand address both the behavioral and the welfare measurement problem within a single unified framework. By modeling each time step of the day as a sequential choice over destination, mode, and activity type, DDCMs generate activity-travel patterns as the outcome of forward-looking utility maximization under time-space constraints. The log-sum of the value function at the start of the day constitutes a welfare-consistent consumer surplus measure — policy prediction and welfare measurement emerge from the same model, without additional assumptions.

The central practical obstacle is computational. Computing the welfare measure exactly requires backward induction over every reachable state; for a city-scale network this runs to the order of 10⁸ theoretical states. Every approximation that reduces this cost modifies the value function and therefore corrupts the welfare measure. This thesis demonstrates that exact computation is tractable: reachability-based state pruning and GPU-accelerated backward induction together reduce the effective state space by more than 99% while preserving the value function exactly.