Baud, C., and Bierlaire, M. (2026)

A synthetic longitudinal individual generator

26th Swiss Transport Research Conference, Ascona, Switzerland

Most synthetic population methods rely on cross-sectional snapshots or pseudo-panels, which do not consistently track individuals over time. This paper proposes a model-agnostic framework in which the panel structure is defined by design. Individuals are represented by life-based trajectories, specified independently of calendar time, and a deterministic mapping recovers their state at any time (t). This allows coherent panel data and population distributions to be reconstructed at arbitrary points in time. The framework also includes a Bayesian updating mechanism to incorporate observed cross-sectional data. When such data are available, synthetic populations are sampled from the posterior distribution, combining prior knowledge with observed evidence to generate coherent longitudinal populations.