Which univariate time series model predicts quicker a crisis? The Iberia case

Abstract

In this paper four univariate models are fitted to monthly observations of the number of passengers in the Spanish airline IBERIA from January 1985 to October 1994. During the first part of the sample, the series shows an upward trend which has a rupture during 1990 with the slope changing to be negative. The series is also characterized by having seasonal variations. We fit a deterministic components model, the Holt-Winters algorithm, an ARIMA model and a structural time series model to the observations up to December 1992. Then we predict with each ofthe models and compare predicted with observed values. As expected, the results show that the detenninistic model is too rigid in this situation even if the within-sample fit is even better than for any of the other models considered. With respect to Holt-Winters predictions, they faH because they are not able to accornmodate outliers. Finally, ARIMA and structural models are shown to have very similar prediction performance, being flexible enough to predict reasonably well when there are changes in trend.