Resumen en español

Time series models are quantitative techniques commonly used to forecast the behavior of variables. These models include the exponential smoothing with trend or Holt model that requires the definition of the smoothing constants α and β and the initialization values, both required for the model upgrade. This paper proposes a different way to obtain the parameter values and initial conditions of the Holts model, optimizing the tracking signal range (TSR), in order to achieve a more robust model from the viewpoint of accuracy of the results and historical performance. Some comparisons between the proposed approach and the traditional methods based on the mean absolute deviation (MAD) and the mean square error (MSE) are provided. These are the measures traditionally used to determine the degree of accuracy of a model, and a better model performance is obtained.