Functional time series forecasting of extreme values

Abstract

Abstract We consider forecasting functional time series of extreme values within a generalized extreme value distribution (GEV). The GEV distribution can be characterized using the three parameters (location, scale, and shape). As a result, the forecasts of the GEV density can be accomplished by forecasting these three latent parameters. Depending on the underlying data structure, some of the three parameters can either be modeled as scalars or functions. We provide two forecasting algorithms to model and forecast these parameters. To assess the forecast uncertainty, we apply a sieve bootstrap method to construct pointwise and simultaneous prediction intervals of the forecasted extreme values. Illustrated by a daily maximum temperature dataset, we demonstrate the advantages of modeling these parameters as functions. Further, the finite-sample performance of our methods is quantified using several Monte Carlo simulated data under a range of scenarios.