PRONÓSTICOS CON RESTRICCIONES EN SERIES DE TIEMPO UNIVARIADAS: APLICACIÓN AL SEGUIMIENTO DEL PIB DE MEXICO EN 2001

Abstract

This paper presents, in a unified way, the restricted forecasting methodology that serves to incorporate additional information into the forecasts arising from a linear model built from historical data of a univariate time series. If the additional information is in the form of linear constraints about future values of the series, it can be optimally and formally incorporated into the model forecasts. The linear model considered is an ARIMA type. As a complement of the method, a statistic is introduced to decide whether the additional data are compatible with the historical series or not. Such a statistic is the basis for wider possibilities of analysis in terms of the certainty associated with the two sources of information: historical and additional. Finally, in order to illustrate the methodology, the annual target of Mexico's real GDP, announced by the government authorities for year 2001, is examined. In this example, the results that are obtained while more data of the series are observed justify changing the target with the course of the year.