Modelos no lineales de prevision para series temporales de interes farmaceutico.Aplicacion al caso de la escarlatina

摘要

This paper presents the fit of both a nonlinear and a robust model to a time series data about the incidence of scarlet fever in Catalunya from 2000 to 2005.Data were published by the Departament de Salut of the Generalitat de Catalunya in Butlleti Epidemiologic de Catalunya(BEC).Fitting was performed both by ordinary least squares and least median of residuals method.Since residuals are non-normally distributed,the confidence intervals were evaluated by a bootstrap procedure.The model was validated looking at the forecast for the last 47 weeks and the published values.Except for acute incidence periods,acceptable forecast are obtained.The sinusoidal model fitted has a period of 52 weeks,maxima are located at the second half of march and minima at the second half of September.Inflexion points are located about summer and winter solstice.