Clustering of business cycles in optimal directions found by SIR and DAME

We investigate the combination of the dimension reduction methods SIR (Li, 1991) and DAME (Gather et al., 2001) with fuzzy-clustering to validate a given classification. We consider certain economic variables which are assumed to contain the information relevant to determine the current phase of economics. For a period of about 40 years, observations of these variables are available, together with an experts' judgement about the corresponding business cycle phase. We show that the combination of dimension reduction and fuzzy-clustering leads to a classification reflecting the experts' opinion better than other classification methods. Moreover, the proposed method can be used also in high-dimensional situations where other procedures are no longer applicable.