A REINTERPRETATION OF PRINCIPAL COMPONENT ANALYSIS CONNECTED WITH LINEAR MANIFOLDS IDENTIFYING RISKY ASSETS OF A PORTFOLIO

Abstract

We use the mean-variance model to study a portfolio problem characterized by an investment in two different types of asset. We consider m logically independent risky assets and a risk-free asset. We analyze m risky assets coinciding with m distributions of probability inside of a linear space. They generate a distribution of probability of a multivariate risky asset of order m. We show that an m-dimensional linear manifold is generated by m basic risky assets. They identify m finite partitions, where each of them is characterized by n incompatible and exhaustive elementary events. We suppose that it turns out to be n > m without loss of generality. Given m risky assets, we prove that all risky assets contained in an m-dimensional linear manifold are related. We prove that two any risky assets of them are conversely α-orthogonal, so their covariance is equal to 0. We reinterpret principal component analysis by showing that the principal components are basic risky assets of an m-dimensional linear manifold. We consider a Bayesian adjustment of differences between prior distributions to posterior distributions existing with respect to a probabilistic and economic hypothesis. AMS Subject Classification: 51F99, 60B05, 91B06, 91B30, 91B82