Setting the Model Options

To use the Modeling and Forecasting task, you must select a forecasting model type. You can choose from six model types: random walk, moving average, exponential smoothing, ARIMA, ARIMAX, and unobserved components.
Random Walk
Moving Average
Exponential Smoothing
ARIMA
ARIMAX
Unobserved Components

Random Walk

To create a random walk model:
  1. From the Forecasting model type drop-down list, select Random walk.
  2. Select one of these types of random walk models:
    • Drift creates a Random Walk model with Drift, or in ARIMA notation ARIMA(0, 1, 0).
    • Trend .
    • Seasonal creates a Seasonal Random Walk model, or ARIMA(0, 1, 0)(0, 1, 0)s with no intercept.
  3. Under the Plots heading, select the plots to include in the results. You can choose from a variety of series plots, residual plots, and forecast plots.

Moving Average

The formula for the moving average model with width k is y sub t , equals . fraction left bracket . y sub t minus 1 end sub . plus dot dot dot plus . y sub t minus k end sub . right bracket , over k end fraction . plus e r r o r. Click image for alternative formats..
In ARIMA notation, this model is ARIMA(k, 0, 0) with no intercept and with the autoregressive parameters (AR) fixed: eh r equals , 1 over k , comma , 1 over k , comma dot dot dot comma , 1 over k. Click image for alternative formats..
To create a moving average model:
  1. From the Forecasting model type drop-down list, select Moving average.
  2. In the Window (periods) box, specify the number of periods for the moving average. This value must be an integer greater than 0 and less than 14.
  3. Under the Plots heading, select the plots to include in the results. You can choose from a variety of series plots, residual plots, and forecast plots.

Exponential Smoothing

Exponential smoothing is a forecasting technique that uses exponentially declining weights to produce a weighted moving average of time series values. You can choose from several forecasting models.
To create an exponential smoothing model:
  1. From the Forecasting model type drop-down list, select Exponential smoothing.
  2. From the Forecasting model drop-down list, select the model that you want to use. You can choose from these models.
    • Simple (single) exponential smoothing, which is the default
    • Double (Brown) exponential smoothing
    • Linear (Holt) exponential smoothing
    • Damped trend exponential smoothing
    • Additive seasonal exponential smoothing
    • Multiplicative seasonal exponential smoothing
    • Winters multiplicative model
    • Winter additive model
  3. From the Transformation drop-down list, select the transformation to apply to the time series. By default, no transformation is applied. If you select the Box-Cox transformation, then you must specify a parameter value between -5 and 5 in the Box-Cox transformation parameter box.
  4. From the Forecast type drop-down list, specify whether the model uses the mean forecasts or the median forecasts.
  5. Under the Plots heading, select the plots to include in the results. You can choose from a variety of model plots, error plots, and forecast plots.

ARIMA

When you create an Autoregressive Integrated Moving Average (ARIMA) model, you can specify the autoregressive and moving average polynomials of an ARIMA model.
To create an ARIMA model:
  1. From the Forecasting model type drop-down list, select ARIMA.
  2. Under the ARIMA heading, specify the autoregressive, differencing, and moving average orders for the ARIMA model.
    Here are the options for the simple ARIMA:
    • Autoregressive order (p) specifies the simple autoregressive order. You can specify an integer from 0 to 13. The default value is 0.
    • Differencing order (d) specifies the simple differencing order. You can specify an integer from 0 to 13. The default value is 0.
    • Moving average order (q) specifies the simple moving average. You can specify an integer from 0 to 13. The default value is 0.
    Here are the options for the seasonal ARIMA:
    • Autoregressive order (P) specifies the seasonal autoregressive order. You can specify an integer from 0 to 5. The default value is 0.
    • Differencing order (D) specifies the simple differencing order. You can specify an integer from 0 to 3. The default value is 0.
    • Moving average order (Q) specifies the simple moving average. You can specify an integer from 0 to 5. The default value is 0.
  3. Specify whether to include the intercept in the model. The intercept is included by default.
  4. Under the Plots heading, select the plots to include in the results. You can choose from a variety of series plots, residual plots, and forecast plots.

ARIMAX

When you create an Autoregressive Integrated Moving Average (ARIMA) model, you can specify the autoregressive and moving average polynomials of an ARIMA model. In an ARIMAX model, you can also include independent variables in the model.
To create an ARIMAX model:
  1. From the Forecasting model type drop-down list, select ARIMAX.
  2. Under the ARIMA heading, specify the autoregressive, differencing, and moving average orders for the ARIMA model.
    Here are the options for the simple ARIMA:
    • Autoregressive order (p) specifies the simple autoregressive order. You can specify an integer from 0 to 13. The default value is 0.
    • Differencing order (d) specifies the simple differencing order. You can specify an integer from 0 to 13. The default value is 0.
    • Moving average order (q) specifies the simple moving average. You can specify an integer from 0 to 13. The default value is 0.
    Here are the options for the seasonal ARIMA:
    • Autoregressive order (P) specifies the seasonal autoregressive order. You can specify an integer from 0 to 5. The default value is 0.
    • Differencing order (D) specifies the simple differencing order. You can specify an integer from 0 to 3. The default value is 0.
    • Moving average order (Q) specifies the simple moving average. You can specify an integer from 0 to 5. The default value is 0.
  3. In the Independent variables role, assign the variables from the input data set that you want to include in the model.
  4. Specify whether to include the intercept in the model. The intercept is included by default.
  5. Under the Plots heading, select the plots to include in the results. You can choose from a variety of series plots, residual plots, and forecast plots.

Unobserved Components

To create an unobserved components model:
  1. From the Forecasting model type drop-down list, select Unobserved components.
  2. (Optional) To include independent variables in the model, expand the Regression Effects heading and select the Include independent variables check box. Assign the variables that you want to include in the model to the Independent variables role.
  3. To include an irregular component, expand the Irregular Component heading and select the Include an irregular component check box. An irregular component is included by default.
    The irregular component corresponds to the overall random error in the model. The initial variance is the value used as the initial value during the parameter estimation process. To change this value, select Specify variance and enter a different value. To keep this value as your initial variance, select Fix variance value.
  4. To include a trend component, expand the Trend Component heading. The level component and the slope component combine to define the trend component for the model. If you specify both a level and slope component, then a locally linear trend is obtained. If you omit the slope component, then a local level is used.
    1. To include a level component in the model, select the Include a level component check box. (The level component is included by default.) Then you can specify whether to change the initial variance (which is 0 by default) and whether to check for level breaks.
    2. To include a slope component in the model, select the Include a slope component check box. Then you can specify whether to change the initial variance (which is 0 by default).
  5. (Optional) To include a seasonal component, the season length must be greater than one. Expand the Seasonal Component heading and select the Include a seasonal component check box. Specify the type of seasonal component. A seasonal component can be one of two types: dummy or trigonometric. You can also specify whether to change the initial variance (which is 0 by default).
  6. (Optional) To include a cycle component, expand the Cycle Component heading and select the Include a cycle component check box. You can specify these options:
    • To specify an initial cycle period to use during the parameter estimation process, select the Specify cycle period check box. Then specify the initial value in the box. This value must be an integer greater than 2. By default, the initial value is 3.
    • To specify an initial damping factor to use during the parameter estimation process, select the Specify damping factor check box, and then specify the initial value in the box. You can specify any value between 0 and 1 (excluding 0 but including 1). By default, the initial value is 0.01.
    • To specify an initial value for the disturbance variance parameter that the task uses during the parameter estimation process, select the Specify variance check box. Then specify the initial value in the box. This value must be greater than or equal to 0. By default, the initial value is 0.
  7. Under the Plots heading, select the plots to include in the results. You can choose from a variety of residual plots, smoothed component estimates, filtered component estimates, and series decomposition and forecast plots.
Copyright © SAS Institute Inc. All rights reserved.