|
Abstract:
|
Businesses often require accurate forecasts of time series data that are highly seasonal with low periodic volumes. The sales of highly seasonal retail items frequently show this pattern. During times of high activity, sales are high; during the off-season, sales might be deterministically zero or on the level of white noise with respect to the period of high activity. In addition, the periods of high activity might have irregularities. The number of active periods might not be consistent from year to year, or the active period might not begin and end in the same period each year. Also, these series often have regression components related to the calendar, so the timestamp of each observational period is important. This paper examines methods that are practical for a large number of series to perform accurate statistical seasonal adjustment and forecasting of these series.
|
