Modified seasonal factors in exponential smoothing
- Pennsylvania Univ., Philadelphia, PA (USA). Wharton School of Finance and Commerce
- Oak Ridge National Lab., TN (USA)
- Tennessee Univ., Knoxville, TN (USA)
Current practice uses statistical tests to determine whether seasonal factors should be applied in a given forecasting situation. Research suggests that an optimal policy might lie somewhere between using full seasonal factors and using no seasonal factors on series. This research proposes and tests use of a modified seasonality factor. Modified seasonal factors reduce the emphasis on the seasonal adjustments when forecasts are made. The adjustments account for errors in the estimation of the factors and for possible changes in the factors over the forecast horizon. An analysis of data from US Navy personnel inventories was conducted to test the use of a modified seasonality factor. Modified seasonal factors led to improved accuracy for predictions of inventories by paygrade using quarterly data from the Navy Personnel Research and Development Center (NPRDC). Under certain selections of factors, the mean absolute percent error (MAPE) was reduced by 4.4%. No gain was obtained, however, for the inventories by length of service. It is expected, but not shown here, that the modified seasonal factors will only be of value for series where the estimated seasonal factors show a substantial variation across the year. 3 refs., 6 tabs.
- Research Organization:
- Oak Ridge National Lab. (ORNL), Oak Ridge, TN (United States)
- Sponsoring Organization:
- USDOD; DOE/EH
- DOE Contract Number:
- AC05-84OR21400
- OSTI ID:
- 6357640
- Report Number(s):
- ORNL/M-1262; ON: DE91002802
- Country of Publication:
- United States
- Language:
- English
Similar Records
Proceedings on forecasting methodology for time-of-day and seasonal electric utility loads
Testing Systems Database interface operation manual