CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: A forecasting-EVT method Authors:  Clara Cordeiro - CEAUL and FCT, UALg (Portugal) [presenting]
Manuela Neves - University of Lisbon and CEAUL (Portugal)
Abstract: Statistical analysis of extreme values has been in the spotlight in studies of flooding events, heatwaves, hurricanes, sea level rise, among many others. This type of events has promoted the development of special statistical methodologies for its study, understanding and control, whenever possible. These events present unique statistical challenges and require characterizing adequately the tail of the distribution of the quantity of interest. Extreme value theory (EVT) is the branch of the statistics that has been used to model such type of data. Well-known forecasting methods do not capture so well these events what is reflected in its extrapolation. A forecasting-EVT method, which applies a forecasting procedure to the time series and models the residuals through an adequate EVT distribution is proposed. Moreover, the extreme distribution of residuals is estimated and bootstrap estimators of the shape parameter will be used to improve the modelling of the tail behaviour of the residuals distribution. This is a preliminary work presenting a contribution to model extremes of a time series.