Robust Statistics and Fuzzy Industrial Clustering

The search for groups of important sectors in an economy has been and still is one of the more recurrent themes in input-output analysis. But a sector can probably be important for some questions at the same time, to a different degree. In this direction, a multidimensional fuzzy clustering analysis gives as a result a classification of sectors illustrating the different roles that each one plays in the economy. But multivariate outliers, witch have not been studied in the literature in previous applications of clustering techniques to input-output studies, have even worst effects in clustering than the univariate ones, due to their influence on the correlation matrices, and because their presence can mask the real clusters. We will show how fuzzy clustering and robust statistics should work together in this kind of studies, so the clustering can benefit from the use of robust statistics in data preparation, identification and computation of dissimilarities or deciding the best number of clusters and specially avoiding the dangerous effects coming from the presence of multivariate outliers.