Earthquake analysis for prediction purposes is a delicate and still open problem largely debated among scientists. In this work, we want to show that a successful time-predictable model is possible if based on large instrumental data from dense monitoring networks. To this aim, we propose a new simple data-driven and quantitative methodology which takes into account the accumulated geodetic strain and the seismically-released strain to calculate a balance of energies. The proposed index quantifies the state of energy of the selected area and allows us to evaluate better the ingoing potential seismic risk, giving a new tool to read recurrence of small-scale and shallow earthquakes. In spite of its intrinsic simple formulation, the application of the methodology has been successfully simulated in the Eastern flank of Mt. Etna (Italy) by tuning it in the period 2007-2011 and testing it in the period 2012-2013, allowing us to predict, within days, the earthquakes with highest magnitude.