Journal rankings of specific research fields are often used for evaluation purposes, both of authors and institutions. These rankings can be defined by means of several methods, as expert assessment, scholarly-based agreements, or by the ordering induced by a numeric index associated to the prestige of the journals. In order to be efficient and accepted by the research community, it must preserve the ordering over time, at least up to a point. Otherwise, the procedure for defining the ranking must be revised to assure that it reflects the presumably stable characteristic "prestige" that it claims to be quantifying. A mathematical model based on fractional p-variations of the values of the order number of each journal in a time series of journal rankings is explained, and its main properties are shown. As an example, we study the evolution of two given ordered lists of journals through an eleven-year series. These journal ranks are defined by using the 2-year Impact Factor of Thomson-Reuters (nowadays Clarivate Analytics) lists for MATHEMATICS and PHYSICS, APPLIED from 2002 to 2013. As an application of our model, we define an index that precludes the use of journal ranks for evaluation purposes when some minimal requirements on the associated fractional p-variations are not satisfied. The final conclusion is that the list of mathematics does not satisfy the requirements on the p-variations, while the list of applied physics does.