Read out

Research Seminar Series in Statistics and Mathematics

Wirtschaftsuniversität Wien, Departments 4 D4.4.00809:00 - 10:30

Type Lecture / discussion
SpeakerRodney Strachan (School of Economics, University of Queensland, Australia)
Organizer Institut für Statistik und Mathematik

Rod­ney Strachan (School of Eco­nom­ics, Uni­versity of Queensland, Aus­tralia) about "Re­du­cing Di­men­sions in a Large TVP-VAR"

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://­math/ressem­inar


This pa­per pro­poses a new ap­proach to es­tim­at­ing high di­men­sional time vary­ing para­meter struc­tural vector autore­gress­ive mod­els (TVP-S­VARs) by tak­ing ad­vant­age of an em­pir­ical fea­ture of TVP-(S)VARs. TVP-(S)VAR mod­els are rarely used with more than 4-5 vari­ables. However re­cent work has shown the ad­vant­ages of mod­el­ling VARs with large num­bers of vari­ables and in­terest has nat­ur­ally in­creased in mod­el­ling large di­men­sional TVP-VARs. A fea­ture that has not yet been util­ized is that the co­v­ari­ance mat­rix for the state equa­tion, when es­tim­ated freely, is often near sin­gu­lar. We pro­pose a spe­cific­a­tion that uses this sin­gu­lar­ity to develop a factor-­like struc­ture to es­tim­ate a TVP-S­VAR for 15 vari­ables. Us­ing a gen­er­al­iz­a­tion of the re­cen­ter­ing ap­proach, a rank re­duced state co­v­ari­ance mat­rix and ju­di­cious para­meter ex­pan­sions, we ob­tain ef­fi­cient and sim­ple com­pu­ta­tion of a high di­men­sional TVP- SVAR. An ad­vant­age of our ap­proach is that we re­tain a formal in­fer­en­tial frame­work such that we can pro­pose formal in­fer­ence on im­pulse re­sponses, vari­ance de­com­pos­i­tions and, im­port­ant for our model, the rank of the state equa­tion co­v­ari­ance mat­rix. We show clear em­pir­ical evid­ence in fa­vour of our model and im­prove­ments in es­tim­ates of im­pulse re­sponses.

Back to overview