BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Local scale invariance and robustness of proper scoring rules - Da
 vid Bolin (King Abdullah University of Science and Technology (KAUST))
DTSTART:20250606T101500Z
DTEND:20250606T111500Z
UID:TALK230860@talks.cam.ac.uk
DESCRIPTION:Averages of proper scoring rules are often used to rank probab
 ilistic forecasts. In many cases\, the individual terms in these averages 
 are based on observations and forecasts from different distributions. We s
 how that some of the most popular proper scoring rules\, such as the conti
 nuous ranked probability score (CRPS)\, give more importance to observatio
 ns with large uncertainty\, which can lead to unintuitive rankings. To des
 cribe this issue\, we define the concept of local scale invariance for sco
 ring rules. A new class of generalized proper kernel scoring rules is deri
 ved and as a member of this class we propose the scaled CRPS (SCRPS). This
  new proper scoring rule is locally scale invariant and\, therefore\, work
 s in the case of varying uncertainty. Like the CRPS\, it is computationall
 y available for output from ensemble forecasts\, and does not require the 
 ability to evaluate densities of forecasts. We also extend the concept of 
 local scale invariance to proper scoring rules for extremes and introduce 
 a weighted version of the SCRPS. Finally\, we discuss robustness propertie
 s of proper scoring rules and introduce robust version of both the CRPS an
 d the SCRPS. &nbsp\;
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR