Hierarchical Bayesian modeling of hotel availability using high-resolution panel data: Application to post-earthquake accommodation market dynamics following the 2024 Noto Peninsula Earthquake

This study proposes a hierarchical Bayesian model to estimate regional accommodation availability using high-resolution booking data from an online travel agency. The model employs a binomial structure to separate facility scale from vacancy probability, capturing facility-level heterogeneity beyond aggregated data. It incorporates booking lead-time information up to one month before check-in. Validation using pre-disaster data demonstrates high forecasting accuracy in regions with sufficient facility coverage, with competitive performance relative to conventional time-series models at longer booking lead times. As a case study, we apply the framework to examine post-earthquake accommodation market dynamics relative to a reference scenario representing conditions without the earthquake. The results reveal heterogeneous recovery trajectories across accommodation facilities: smaller facilities tend to exhibit larger and more variable deviations from the reference scenario and slower convergence toward the reference trajectory, whereas larger facilities tend to show earlier convergence.