Research Seminar Series in Statistics and Mathematics

Wirtschaftsuniversität Wien, Departments 4 D4.4.00811:00 - 12:10

Art Vortrag/Diskussion
Vortragende/rTobias Fissler (Department of Mathematics, Imperial College London)
Veranstalter Institut für Statistik und Mathematik
Kontakt katrin.artner@wu.ac.at

Tobias Fissler (Depart­ment of Mathe­ma­tics, Impe­rial College London) about "The Elici­ta­tion Problem or The Quest of Compa­ring Fore­casts in a Meaningful Way"

The Insti­tute for Statis­tics and Mathe­ma­tics (Depart­ment of Finance, Accoun­ting and Statis­tics) cordi­ally invites ever­yone inte­rested to attend the talks in our Rese­arch Seminar Series, where inter­na­tio­nally renowned scho­lars from leading univer­si­ties present and discuss their (working) papers.

The list of talks for the winter term 2018/19 is avail­able via the follo­wing link: https://www.wu.ac.at/en/stat­math/resse­minar

A proven stra­tegy in deci­si­on-­ma­king to cope with unknown or uncer­tain future events is to rely on fore­casts for these events. Exam­ples range from weather fore­casts for agri­cul­ture, airlines or a conve­nient ever­yday life, to fore­casts for supply and demand in a busi­ness context, to risk-as­sess­ment in finance or predic­tions for GDP growth and infla­tion for pruden­tial economic policy. In the presence of multiple diffe­rent fore­casts, a core chal­lenge is to assess their rela­tive quality and to even­tually rank them in terms of their historic perfor­mance. This calls for an accu­racy measure which is commonly given in terms of a loss func­tion speci­fying the discre­pancy between a fore­cast and the actual obser­va­tion. Exam­ples include the zero-one loss, the abso­lute loss or the squared loss. If the ulti­mate goal of the fore­casts is speci­fied in terms of a statis­tical func­tional such as the mean, a quan­tile, or a certain risk measure, the loss should incen­ti­vise trut­hful fore­casts in that the expected loss is strictly mini­mised by the correctly speci­fied fore­cast. If a func­tional possesses such an incen­tive compa­tible loss func­tion, it is called elici­table. Besides enab­ling meaningful fore­cast compa­rison, the elici­ta­bi­lity of a func­tional allows for M-esti­ma­tions and regres­sion. Acknow­led­ging that there is a wealth of elici­table func­tio­nals (mean, quan­tiles, expec­tiles) and non-e­li­ci­table func­tio­nals (vari­ance, Expected Short­fall), this talk addresses aspects of the follo­wing Elici­ta­tion Problem:
1) When is a func­tional elici­table?
2) What is the class of incen­tive compa­tible loss func­tions?
3) What are distin­gu­ished loss func­tions to use in prac­tice?
4) How to cope with the non-e­li­ci­ta­bi­lity of a func­tional?
The emphasis will lie on main achie­ve­ments for multi­va­riate func­tio­nals such as the pair of risk measures (Valu­e-a­t-­Risk, Expected Short­fall). It will also give an outlook to modern and very recent achie­ve­ments in the realm of set-va­lued func­tio­nals which are suited to consider set-va­lued measures of systemic risk or confi­dence inter­vals and regions.

Kindly note that on November 9 two talks are sche­duled at our insti­tute:
9:00 – 10:10  Nestor Parolya (Leibniz Univer­sity Hannover)
11:00 – 12:10  Tobias Fissler (Impe­rial College London)

zurück zur Übersicht