This paper discusses a novel methodology for dynamic modeling of writing process. Sequent sub-documents of a given document are described through occurrences of the suitably selected N-grams. The Mean Dependence similarity measures the association between a present sub-document and numerous preceding ones and transforms a document into a time series, which is supposed to be weak stationary if the document is created using the same writing style. A periodogram of this signal estimates its Power Spectral Density providing a spectral attribute of the style. Numerical experiments demonstrate high ability of the proposed method in authorship identification and the reveal of writing style evolution.