projects
Current
- Constraint Heterogeneity in Dynamic Choice ModelsA day is assembled out of what is possible, not only out of what is wanted, and what is possible differs between people. A model that cannot carry that difference reads constraint as preference, and the error propagates into accessibility and welfare; models that can carry it exactly have paid one value function per person. This work asks what per-person feasibility actually costs — classifying constraints by what they read, and so by how many distinct computations a population genuinely requires.
Completed
- A Scalable Computational Framework for Activity-Based Dynamic Discrete Choice ModelsExact estimation of activity-based DDCMs at city scale. Because the value of a day is a log-sum over every feasible day, nothing may be dropped for being unlikely — only for being impossible. A forward search under the space-time prism keeps just the reachable states, cutting the theoretical state space by more than ninety-nine percent without approximation; the pruned graph is acyclic, so a whole time layer solves at once; and one solve is shared across everyone facing the same constraints. Estimated on the Higashi-Hiroshima travel diary.
- Where Ride-Hailing Drivers WaitWhere online motorcycle-taxi drivers position themselves between orders in Bandung, combining platform transaction records with driver perception surveys to identify spatiotemporal idling hotspots and the preferences behind them.Asian Transport Studies EASTS 2025 — Best Paper