Factor models with local factors — Determining the number of relevant factors

We extend the theory on factor models by incorporating “local” factors into model. Local affect only an unknown subset of observed variables. This implies a continuum eigenvalues covariance matrix, as is commonly in applications. derive which are pervasive enough to be economically important and estimable using common principal component estimator. then introduce new class estimators determine number those relevant factors. Unlike existing estimators, our use not but also its eigenvectors. find that partial sums eigenvectors leads significant gains performance simulations.