This paper has as main objective to present and to test a tool of multivariate statistics in financial models. This methodology, known as clusters analysis, separates the observations in groups through its determined characteristic, in contrast of the traditional methodology, which is only the order through quantiles. This tool was applied in 213 shares negotiated in the São Paulo Stock Exchange (Bovespa), separating to the groups for size and book-to-market. Later, the new portfolios were applied in the Fama and French Model (1996), comparing the results in a portfolio formation for quintiles and for cluster analysis. Better results were found in the second methodology. The authors conclude that the cluster analysis can be more adequate, because tends to form more homogeneous groups, being useful its application for portfolio formation, and for financial theory.
Quantiles; Cluster analysis; Data Mining; Anomalies; Fama and French model
