On information in static and dynamic factor models

This paper employs concepts from information theory in factor models. We show that in the exact factor model the whole distribution of eigenvalues of the covariance matrix contributes to the information and not only the largest ones. In addition, we derive the condition that the first q say eigenvalues diverge whereas the rest remain bounded in the static model rather than having to assume it. Finally, we calculate information in static and dynamic factor models, which can be used to find the dimensions of the factor space. We illustrate the concepts with simulation experiments.