Extending Nonstationary Extreme Models
Background: Extreme value modeling often assumes stationary event processes, but real-world phenomena frequently exhibit nonstationarity in both the occurrence rates and the magnitudes of events. Future risk assessment in such contexts requires flexible models that can capture temporal dynamics and mark distribution variations without over-fitting or requiring strong parametric assumptions.
Question / Future Work: Expanding the Bayesian framework to incorporate time-varying background intensities or exogenous covariate dependency in the event rate and magnitude distributions is essential for realistic risk forecasting, as is moving toward multivariate modeling of interconnected risk phenomena.
Why It Matters: These extensions are critical for moving from idealized, self-contained data analysis toward robust, real-world forecasting systems that can account for exogenous drivers and interconnected risk events.
Metadata & Links
- created_at
- 2026-05-08T05:44:52Z