IDENTIFICATION OF SEASONALITY IN TIME-SERIES - A NOTE

Identification of patterns in time series data is critical to facilita te forecasting. One pattern that may be present is seasonality. A meth od is proposed which adds statistical tests of seasonal indexes to the usual autocorrelation analysis in order to identify seasonality with greater confidence. The methodology is tested with known time series. In some cases, a discrepancy arose between the original classification of the time series, which was reportedly based solely on autocorrelat ion analysis, and the classification which resulted from the statistic al tests. Further analysis, which included examination of time series plots, indicated that these discrepancies were likely to occur in seri es with possible structural changes. Out-of-sample forecasts were perf ormed with seasonal and deseasonalized data to determine the benefits of the ''corrected'' classification. Results show improvement, albeit small.