Abstract:
Time series may contain multiple seasonal cycles of different lengths. There are several
notable features in Figure 1, reference to the hourly electricity demand in Sri Lanka, data
are given in the Table lFirst, we observe that the daily cycles are not all the same,
although it may reasonably be claimed that the cycles for Monday through Sunday are
similar. A second feature of the data is that the underlying levels of the daily cycles may
change from one week to the next, yet be highly correlated with the levels for the days
immediately preceding. Thus, an effective time series model must be sufficiently flexible
to capture these principal features without imposing too heavy computational or inferential
burdens. The goal of this paper is to introduce a new procedure that uses innovation
ARIMA models to forecast time series with multiple seasonal patterns.
