A Time Series Model for Nonstationary Data Exhibiting Non-Integer Seasonal Periodicity

Thomas M. Menino Convention & Exhibition Center 

Time series data with non-integer seasonal periodicities (e.g., weekly series, which would have a seasonal period of approximately 52.18) present modelling challenges; standard seasonal differencing operators are designed for an even division of the year, and as such, may not adequately stabilize series that fall outside this paradigm.  We develop a non-integer differencing operator that can overcome this limitation.  We do so by first determining the specific frequencies for removal and then reverse engineering the differencing operator that can achieve this.  We examine the mathematical properties of our proposed differencing operator and demonstrate its effectiveness in signal extraction and seasonal adjustment for weekly time series via some practical applications.