Multivariate Analysis of Business Cycles

Ulrich Heilemann and Heinz Müench - RWI and Universität Duisburg

Resuming the work of Meyer/Weinberg (1975), in a recent study, Heilemann/Müench (1995) demonstrated that West German business cycles between 1963 and 1994 can be well described by a simple multivariate four-stage scheme: upswing, upper turning-point phase, downswing, and lower turning-point phase. Linear discriminant analysis (LDA) has been applied to a set of quarterly National Account (NA) data and of some monetary data to examine stability/change of West German business cycles.

While the impression over the 1963-1994 period is one of stability, the expansion to pre-1960 data reveals considerable changes in the cyclical behavior of the '50s and early '60s (Heilemann/Müench 1996). This paper has three goals: first, to supplement LDA with additional classification methods like class modeling procedures (Derde/Massart 1988), recursive partitioning procedures (Breiman et al. 1984), and neural networks (Ripley 1994); second, to broaden the data base by survey and/or indicator data/variables; and third, to identify different regimes of variables for the distinct phases of business cycles and to evaluate the explanatory power of variables over time.

The paper will be organized as follows: The first section describes the classification scheme for West German business cycles. Section II gives a short overview about the multivariate methods and the data base employed. The results are reported in section III. The final section summarizes the findings and makes suggestions for further research.

Scheduled for Session 1.1 Computation And Econometrics - I