WP01-26
Score-driven long memory dynamics of worldwide regional temperature anomalies
Resumen
We develop a score-driven fractionally integrated quasi-autoregressive model with Student’s t innovations (t-FI-QAR), in which the degree of fractional integration dt, the conditional mean, and the conditional scale evolve via likelihood-based updates. Using monthly U.S. National Oceanic and Atmospheric Administration (NOAA) land-and-ocean temperature anomalies for six regions (Arctic, Antarctic, Atlantic Ocean, Northeast Pacific Ocean, Northern and Southern Hemispheres) from January 1850 to October 2025, we show that dt rises over time across all regions, implying increasingly persistent temperature dynamics. The increase is strongest in the Arctic and Northern Hemisphere, where we detect a struc tural shift over the past three decades consistent with a move toward non-stationary behavior. Significant but more moderate increases occur in the Atlantic, Northeast Pacific, and Southern Hemisphere, while the Antarctic exhibits a slight upward trend. Models that allow heavy tails and time-varying scale outperform homoscedastic specifications in likelihood-based criteria and diagnostics. The findings indicate that climate shocks now propagate more durably than in earlier periods, suggesting that stronger and more sustained mitigation and adaptation policies are needed to counter long-lived deviations in regional temperatures.
Autores: Szabolcs Blazsek, Raven Amina Dupree, Luis Alberiko Gil-Alana.