Getting Started with Time Series Forecasting

Outline of the Forecasting Process

The examples shown in the following sections illustrate the basic process you will use with the Forecasting System.

Specify the Input Data Set

Suppose you have a number of time series, variables recorded over time, for which you want to forecast future values. The past values of these time series are stored as variables in a SAS data set or data view. The observations of this data set correspond to regular time periods, such as days, weeks, or months. The first step in the forecasting process is to tell the system to use this data set by setting the Data Set field.

If your time series are not in a SAS data set, you must provide a way for the SAS System to access the data. You can use SAS features to read your data into a SAS data set; refer to SAS Language Reference. You can use a SAS/ACCESS product to establish a view of data in a database management system; refer to SAS/ACCESS documentation. You can use PROC SQL to create a SAS data view. You can use PROC DATASOURCE to read data from files supplied by supported data vendors; refer to Chapter 10, "The DATASOURCE Procedure," for more details.

Provide a Valid Time ID Variable

To use the Forecasting System, your data set must be dated: the data set must contain a time ID variable that gives the date of each observation. The time ID variable must represent the observation dates with SAS date values or with SAS datetime values (for hourly data or other frequencies less than a day), or you can use a simple time index.

When SAS date values are used, the ID variable contains dates within the time periods corresponding to the observations. For example, for monthly data, the values for the time ID variable may be the date of the first day of the month corresponding to each observation, or the time ID variable may contain the date of the last day in the month. (Any date within the period will serve as the time ID for the observation.)

If your data set already contains a valid time ID variable with SAS date or datetime values, the next step is to specify this time ID variable in the Time ID field. If the time ID variable is named DATE, the system fills in the Time ID field automatically.

If your data set does not contain a time ID, you must add a valid time ID variable before beginning the forecasting process. The Forecasting System provides features that make this easy to do. See Chapter 24, "Creating Time ID Variables," for details.

Select and Fit a Forecasting Model for each Series

If you are using the automated model selection feature, the system performs this step for you and chooses a forecasting model for each series automatically. All you need to do is select the Fit Models Automatically button and then select the variables to fit models for.

If you want more control over forecasting model selection, you can select the Develop Models button, select the series you want to forecast, and use the Develop Models window to specify a forecasting model. As part of this process, you may use the Time Series Viewer and Model Viewer graphical tools. Once you have selected a model for the first series, you can select a different series to work with and repeat the model development process until you have created forecasting models for all the series you want to forecast.

The system provides many features to help you choose the best forecasting model for each series. The features of the Develop Models window and graphical viewer tools are introduced in later sections.

Produce the Forecasts

Once a forecasting model has been fit for each series, select the Produce Forecasts button and use the Produce Forecasts window to compute forecast values and store them in a SAS data set.

Save Your Work

If you want only a single forecast, your task is now complete. But you may want to produce updated forecasts later, as more data becomes available. In this case, you want to save the forecasting models you have created, so that you will not need to repeat the model selection and fitting process.

To save your work, fill in the Project field with the name of a SAS catalog member in which the system will store the model information when you exit the system. Later, you will select the same catalog member name when you first enter the Forecasting System, and the model information will be reloaded.

Note that any number of people can work with the same project file. If you are working on a forecasting project as part of a team, you should take care to avoid conflicting updates to the project file by different team members.

Summary

This is the basic outline of how the Forecasting System works. The system offers many other features and options that you may need to use (for example, the time range of the data used to fit models and how far into the future to forecast). These options will become apparent as you work with the Forecasting System.

As an introductory example, the following sections use the Automatic Model Fitting and Produce Forecasts windows to perform automated forecasting of the series in an example data set.

Chapter Contents
Previous
Next
Top