Print Email Facebook Twitter Tail characteristics of CRPS-based distributions Title Tail characteristics of CRPS-based distributions Author Roseboom, Jeroen (TU Delft Electrical Engineering, Mathematics and Computer Science) Contributor Chen, P. (mentor) Degree granting institution Delft University of Technology Programme Applied Mathematics Date 2022-01-14 Abstract In my thesis I researched the potential paths and pitfalls of the newly created ``Taillardat index''.This index uses the tail characteristics of several CRPS-based distributions to rank forecasters on how well they forecast, with a slight emphasis on extreme events.From my research I concluded that the ``Taillardat Index'' in its current form is unstable and should be avoided.Even with theoretical changes, such as moving away from using p-values as a ranking method, the ideas behind the ``Taillardat index'' have to be handled with precaution.I tried to construct a new index based on the ideas of the paper by Taillardat et al. (2019) and the general theory of forecaster validation, to no avail.The findings of my endeavours can be found after the reflection of the Taillardat index and in the discussion. Subject Continuous Ranked Probability ScoreCRPSP-valueHypothesis TestingExtreme EventsProbability Intregral TransformPITPIT-histogramTaillardat IndexCalibrationSharpnessMaximise Sharpness given Calibration To reference this document use: http://resolver.tudelft.nl/uuid:6385af3e-e801-41cc-878c-17299f01df3a Part of collection Student theses Document type master thesis Rights © 2022 Jeroen Roseboom Files PDF Thesis_Jeroen_Roseboom_Fi ... ersion.pdf 6.55 MB Close viewer /islandora/object/uuid:6385af3e-e801-41cc-878c-17299f01df3a/datastream/OBJ/view