A new approach to monitoring stochastic features of earthquake series was applied to track the preparatory process of the 2017 Tehuantepec, Mexico, Mw8.4 earthquake. The seismicity was parameterized by elapsed times and epicentral distances from the directly preceding earthquakes and by earthquake magnitudes. The transformation to equivalent dimensions ensured the comparability of these parameters. We were calculating the average distance between earthquakes in the transformed parameter space in moving in time data windows, each consisting of 100 events. The average distance exhibited a consistent upward trend from ten to two years before the main shock. Then, it declined until the main shock. This precursory up-down signal was highly significant statistically. We showed that the detected time changes of the average distance resulted from the evolution of the earthquake clustering in the considered parameters' equivalent dimensions space.