Environmental drivers and forecast combinations for hospital stays: A hierarchical regional study for Spain

R. Morales Arsenal, Á. Rabadán Navarrete

We examine whether environmental information improves short-run forecasts of hospital utilisation. Using an annual regional panel for Spain (19 regions, 2005–2019), we forecast hospital stays for circulatory and respiratory conditions under a rolling-origin, one-step-ahead design, producing predictions for both regions and the national aggregate. We compare naïve, drift, exponential smoothing and ARIMA benchmarks with climate-based models using CO2 and NOx as drivers. Our main specification is a CO2 shock–exposure model linking out-of-sample driver forecasts to hospital-stay predictions through an exposure parameter. Results show that simple benchmarks are hard to beat, but climate-based models are competitive; the strongest standalone result uses a BCCG-based driver with regional exposure, while the lowest aggregate errors come from equal-weight forecast combinations.

Keywords: Forecasting environmental drivers hospital utilisation hierarchical time series ARIMAX

J. Contreras Reyes

P. Huidobro Fernández, A. Bouchet, S. Diaz-Vazquez, S. Montes

F. J. Martinez Peral, J. A. Ferez Rubio, C. Pérez Vidal, J. V. Segura Heras

M. Pérez Casany, A. Duarte-López, M. Gonzalez-Scotto

J. V. Segura Heras, F. J. Martinez Peral, M. A. Martínez Mayoral, J. Morales Socuéllamos, C. Pérez Vidal