SAS'S Time Series Forecasting System ASSUMES that a pre-set or
"pick-best" approach is good enough. It isn't !
The greatest strength of the Time Series Forecasting system is
the wide range of forecasting models it provides.
Using the system, you can construct an appropriate forecasting model
for almost any time series.
( NOTE : the key word is "almost" )
Forecasting Models
- exponential smoothing
- simple exponential
- double exponential
- linear exponential
- damped-trend linear exponential
- seasonal exponential
- Winters smoothing, additive and multiplicative
- Box-Jenkins ARIMA models, including seasonal ARIMA models
- predictor variables
- simple regressors
- seasonal dummy variable regressors
- intervention (dummy) variables to model exceptional events,
level shifts, or trend shifts
- adjustment variables
to adjust the forecasts by fixed amounts at each period
- transfer functions or dynamic regression:
use transformations, lags, or time series filters
to model the impact of predictor variables
- automatic and user specified forecasting models for predictor variables
- time trend models
- linear
- quadratic
- cubic
- logistic
- logarithmic
- exponential
- hyperbolic
- power function
- exp(A+B/time)
- data transformations
- logarithmic
- logistic
- square root
- Box-Cox
- combining or average the predictions of other forecasting models
- external (judgmental) forecasts
- customized models