Introduction to SCA WorkBench 5.1
Automatic Time Series Modeling in the SCA Statistical System

- ARIMA Forecasting -

SCA WorkBench provides a graphical user interface (GUI) for the SCA Statistical System.
Under this GUI, users can

v Employ easy-to-use menus and dialogs supporting all SCA commands

v Type SCA commands directly at the SCA Input Console

v Organize and execute SCA macro procedures

v Import data from Excel and other spreadsheet programs

The working directory can be typed directly in the text box or it can be selected by clicking on the Browse button. For this illustrative example, the TSDATA directory is selected as the working directory. This directory contains a variety of time series data and is automatically installed by SCA WorkBench under the SCAWORKB directory.

Once the working directory is specified, an interactive SCA System session is started by selecting Run SCA System Interactively from the System menu.

The initial environment of an interactive SCA session is displayed below.

The items under the SCA Session and Windows topics are displayed below.
  

SCA Session Windows

From the SCA Session menu, users can select a preferred interface mode, view a sorted list of SCA commands, and display information about variables currently stored in the SCA System workspace. The Windows menu provides control over the input console window and output window. From this menu, users can change fonts, add comments to the output, or print/save the output and SCA command history.

If a user is familiar with the SCA command syntax, SCA commands can be typed directly at the input console. This is often an efficient mode of operation when executing simple SCA commands.

SCA WorkBench provides a graphical user interface that consists of menus and dialogs for all SCA commands. Currently, SCA provides more than 100 individual commands that cover a wide range of topics from data and workspace management to statistical modeling and analysis. To view the dialogs by topic, select Menu Mode from the SCA Session menu. This is the recommended mode of operation for new users.

The items under each topic correspond to the SCA command name with few exceptions. A longer description of the command is displayed by clicking once on the item. The dialog box is launched by double-clicking the item.

When an SCA session is started, the first action is typically reading data into the workspace. The SCA System can read data into its workspace using the INPUT, FINPUT or BINPUT commands. Alternatively, the data can be read from an SCA data macro. In this example, we select the CALL Data Macro dialog which builds the command to read data into the workspace via an SCA data macro.

The items in red signify required elements of the command. The items in blue or black indicate that the element is optional.

The Browse button is used to select the SCA data macro file to read. Here, we select the “VSTORES.mad” file located under the TSData sub-directory of SCAWORKB. After the file is selected, the procedures contained in the data macro are automatically loaded into the Procedure drop-down box. If the SCA data macro file is typed directly into the File Name text box, the Retrieve button must be executed to retrieve the individual data macro procedures into the Procedure drop-down box. Click on Submit to execute the command immediately in the SCA System.

After the data resides in the SCA workspace, click on the ARIMA Forecasting item under the SCA Menu to launch the command dialog.

After the ARIMA Forecasting dialog loads, select the monthly variety stores (VSALES) series as the dependent variable. Before doing anything, it is a good idea to view a time plot of the series. Select the Original Y only option under Graphics Display to generate the graph shown below in the SCA Graphics Manager.

The time series plot reveals a strong monthly seasonal pattern. We may also investigate if a transformation will stabilize the variance of the series. From the SCA Graphics Manager, select Transformation from the TimeSeries topic.

The VSALES variable is selected from the Box-Cox Transformation dialog and we investigate how the series would look using no transformation (1.0), a square root transformation (0.5), logarithmic transformation (0.0), and an inverse square root transformation (-0.5).

After selecting the appropriate choices, click on OK to generate the four transformation plots shown below.

Upon visual inspection, it seems that a logarithmic transformation does an adequate job of stabilizing variance over time. It is recommended that the SCA Graphics Manager session be terminated before returning to the SCA session if the user intends to employ the SCAGRAF dialog again.

The SCA System allows only one SCAGRAF session to be opened. If an SCAGRAF session is already open, the variables can not be passed from the SCA System to the SCA Graphics Manager. This is applicable to Windows NT, ME and XP operating systems. Under Windows 95/98 operating systems, this restriction does not apply.

We now return to the ARIMA Forecasting dialog, knowing that VSALES has strong monthly periodicity and that a natural logarithmic transformation is warranted.

The automatic ARIMA modeling command (IARIMA) requires information about the potential seasonality (periodicity) of the time series. Type “12” in the seasonality text box to indicate that the series is organized as a monthly time series. The automatic modeling procedure will then determine if seasonal parameters are required in the model.

To model the natural logarithmic transformed series, set the lambda value to “0” in the power transform text box. Lastly, specify the number of forecasts you wish to generate (set to 24 here) and how you would like to display the forecast results. The choices would be to retransform the forecasts back into original scale of VSALES (Original Y + Forecasts), or to leave the forecasts in its transformed state (Transformed Y + Forecasts). Click on Submit to execute. The forecast plot is displayed in the SCA Graphics Manager and the modeling results are provided in the output window.

Click on refresh output to update the screen.

By default, the forecast origin is set to the last available observation and the forecasts are generated from the end of the series. If we would like to examine how well the forecasts compare to a holdout sample, the forecast origin can be modified.

For example, the forecast origin is now set to 129 which holds out the last 24 observations for forecast comparison. By doing so the first 129 observations are used for modeling/estimation and 24 forecasts are generated starting at observation 130. Because we elected to employ automatic ARIMA modeling, it is possible that a different model may be identified for the reduced span of data considered. If we wanted to employ the same model, the model may be spcified directly by the user.

Upon submitting the command dialog for exaecution, the following forecast graph is displayed and model results obtained.