Statistical modeling and forecasting climate-driven respiratory pharmaceutical demand: a spatio-temporal analysis from a Greek panel dataset

Climate change is reshaping health risks and the timing and intensity of medical needs, yet evidence that rigorously quantifies and forecasts climate-driven variation in pharmaceutical demand remains limited. This thesis develops a statistical framework to measure and predict climate-sensitive demand for respiratory medications, using a multi-regional weekly panel for Greece spanning January 2016–June 2023 and covering 20 pharmaceutical regions. Methodologically, the thesis (i) studies the univariate demand dynamics of respiratory pharmaceutical demand and builds a time-series forecasting benchmark; (ii) explores climate–demand linkages through frequency-domain Granger causality, spatialspillover specifications, and variable and lag selection via LASSO and moving-blockbootstrap Random Forest screening (MBB–RF); (iii) benchmarks forecasting models under a unified and reproducible expanding-window protocol, comparing a Vector Autoregression with exogenous inputs (VARX), a MBB–RF, and Long Short-Term Memory networks (LSTM); and (iv) incorporates regional information through a bottom-up two-way fixed-effects panel model (FE–OLS) with region and week effects to produce geographically resolved projections coherent with national aggregates. Empirically, temperature emerges as the dominant driver of climate–demand comovement. Despite strong autoregressive and seasonal structure in the series, incorporating climate covariates systematically improves forecasting relative to climate-agnostic baselines. Temperature exhibits a positive spillover effect, whereas wind speed exhibits a negative spillover effect across neighboring regions. In head-to-head evaluations, VARX remains competitive and transparent, while the two-way FE panel performs similarly and is preferred for projections because it does not require recursive use of the outcome variable, thereby avoiding multi-step error propagation. National projections to mid-2028 show increases of 26.7% under SSP2 (the “middle-of-the-road” Shared Socioeconomic Pathway) and 31.9% under SSP5 (a high-emissions, fossil-fueled development pathway) relative to the 2022–2023 post-pandemic baseline.