Applicazioni finanziarie di tecniche fattoriali. un'analisi di sensitività

Abstract

The identification of the exact dimension of a factor model is usually done using the likelihood ratio. In this paper some financial models are estimated by factor analysis and it is emphasized that the likelihood ratio (analysed in respect to alternative specifications of the asset returns number and time series periodicity) is not a robust measure of the number of factors underlying to the asset returns of 100 assets quoted at the Milan stock exchange. The research of a stable factor structure, continued using different methods, suggests the superiority of cross-validation. Cross-validation and Schwarz's criterion suggest a number of factors sensibly smaller than the one indicated by likelihood ratio.