MODEL SELECTION IN DESCRIPTIVE TIME SERIES: A PARAMETRIC AND NON-PARAMETRIC PERSPECTIVE

In this paper, we investigated the choice between additive and multiplicative models using statistical test in time series decomposition. The Buys-Ballot procedure was ad opted and two parametric and one non parametric test for constant variance were applied to the column variance of the Buys-Ballot table. The results of the illustrative examples using the monthly demand data of Abeokuta base Danico food limited show ed that the parametric (Bartlett and Hartley) and non-parametric ( Square Rank) were significant at 5% level indicating that the appropriate model for decomposition is the additive model. Also the application of the sa me test to quarterly data of the price of square meter of housing in Spain from January 1987 until October 2003 show that the respective statistic were less than their corresponding critical level at 1% significant level indicating that the appropriated model for decomposition of the series is the multiplicative model.