文档介绍：Dynamic Ba y w orks for Classi
cation of
Business Cycles
Ursula Sondhauss and Claus W eihs
Departmen t of Statistics
Univ ersit y of Dortm und
German y
Marc h
revised June
Abstract
W e use Dynamic Ba y w orks to classify business cycle phases
W pare classi
ers generated b y learning the Dynamic Ba y
w ork structure on di
eren t sets of w ork structures
Included
are sets w ork structures of the T ree Augmen ted Naiv eBa y es
T AN
classi
ers of F riedman
Geiger
and Goldszmidt
adapted for dy
namic domains
The p erformance of the dev elop ed classi
ers on the giv en
data w as mo dest
In tro duction
Business cycle researc h attained new atten tion in the last t w o decades
Causes
w ere on the one hand that cyclical c hanges b ecame more imp ortan t relativ e to
gro wthindev elop ed economies
and on the other hand new theories corresp ond
ing to the course of and the reasons for cyclical dev elopmen ts as w ell as impro v ed
facilities for the analysis of empirical dev elopmen ts and for plex sys
tems of h yp otheses
The new actualit y of business cycle researc h can also b e rec
ognized b y means of n umerous publications
see e
g
Filardo
Symp osium
on Dev elopmen ts in Business Cycle Researc h
and also b y the immediate
discussion of new ideas in empirical business cycle analysis
see e
g
Council of
Economic Advisers
This w ork has b een supp orted b y the Deutsc he F orsc h ungsgemeinsc haft
Sonderforsc h ungs
b ereic h
In the literature
business cycles are mainly treated as a univ ariate phe
nomenon
Often business cycles are analyzed b y univ ariate time series mo dels
ignoring the in terpla y of di
eren t economic v ariables th us missing the p ossible
adv an tages of m ultiv ariate analysis
Diagnoses and predictions based on m ulti
stage link ed mo dels and not only on unlink ed individual predictions can lead to