These requirements and best practices are important to follow for any SAS Forecast Server project that you create or update.

Requirements for SAS Forecast Studio (failure to heed most of the following requirements might result in a corrupted project):

• Never manipulate any output data set that SAS Forecast Server produces.
• Never open any output data set that SAS Forecast Server produces while SAS Forecast Studio is running on that project.
• Do not terminate SAS Forecast Studio in the middle of running project.

• Structural changes to an existing SAS Forecast Studio project are not supported. Example: adding a new BY variable to the project hierarchy. Although you can add new *levels* to the existing BY variables, as well as new time series observations, you cannot add a new BY variable to an existing project.

• The BY-variable length must stay the same for the lifetime of the project. BY-variable attributes should not be changed when updating the data for an existing SAS Forecast Studio project. To learn more about character variable BY-variables, see SAS Note 51478: "Guidelines for using BY variables in a SAS Forecast Studio^® project"

• Do not add data to a child of an inactive series or add a new series to an inactive parent. See SAS Note 54512: "Inactive parent series in SAS^® Forecast Server"

• Never delete (or rename) a variable from the updated data set. Exception: even though deleting is not recommended, you can delete an independent, reporting, or adjustment variable. You must then run METHOD=DIAGNOSE when updating the project).

• It is mandatory that all series in the project have forecasts so that reconciliation succeeds for all series. If one series fails to produce statistical forecasts, then reconciliation fails for all the series whose reconciliation computation depends on the failed series.

• Never assign a forecast environment to the SASMeta workspace server. Example: select SASApp instead of SASMeta.

• Do not use data compression with SAS Forecast Server.

• Do not make any changes to the libraries_declaration.sas file in the SAS Forecast Studio projects directory. Additional library declarations can be made by including LIBNAME statements in the start-up code for an environment. Or you can make the declaration in your SAS or server configuration files. NOTE: Do not use the project start-up code to assign the input libraries for a project. The timing of this assignment is not appropriate for input data since the workspace server connection has already been made. For information about libraries see SAS Note 56423: "Troubleshooting libraries that are not displayed in SAS^® Forecast Studio."

• Often your data is not in the appropriate format for SAS Forecast Studio. To avoid misleading or incorrect analyses from your time series data, you should conduct data preprocessing. Your time series data must meet the following requirements:

  • The data is equally spaced. Equally spaced means that successive observations are a fixed time interval apart, and that the data can be described by a single interval. Examples: hourly, daily, or monthly. Preprocess this data so that there is a row in the data set for each datetime value.

  • Remove duplicates from your data.

See SAS Note 56928: "Data pre-processing requirements for SAS^® Forecast Studio" for more information about cleaning your data.

Best practices for SAS Forecast Studio:

• Do not delete (or rename) an independent, reporting, or adjustment variable from an updated input data set. If you choose to delete an independent, reporting or adjustment variable, then you must run METHOD=DIAGNOSE when updating the project. However, deleting these types of variables from an updated data set is not recommended. Also, you are not allowed to delete other types of variables. See the bullet point in the Requirements section.
• Save backup copies of each important project in case the project becomes corrupted. Create backups before every data update that includes new observations. To create backups, use the archive facility.
• If server crashing or maintenance is common, it is important to implement clustered environments to achieve high availability.
• When you update data or reforecast a project, use METHOD=DIAGNOSE (or at least METHOD=SELECT) if possible.
• Do not include independent variables that contain many missing values in the historic region. There is a general check at the diagnose stage for missings in independent variables. Only observations with nonmissing data are used as a starting point for subsequent model identification.
• Do not include an independent variable that is very highly contemporaneously correlated with the dependent variable. You can make an exception when you have future values of the independent variable.
• Do not include independent variables that do not vary, or vary very little in the historic region.
• The reconciliation level is the most important level – choose it well.
• Set up the hierarchy in a way that makes sense for forecasting. Do not build the hierarchy based on reporting requirements. See Mistakes in the Forecasting Hierarchy.
• Avoid using the TSFSSELECT model selection list. The SAS Forecast Server diagnostics provide a superior method of choosing models. The TSFSSELECT model selection list is shipped with SAS Forecast Server so that former SAS Time Series Forecasting System users can compare results.
• Use the suggested JRE version in JAVA web start. See the Java Runtime Environments (JRE).

• Make sure that you have applied the latest SAS Forecast Server Hot Fix. See SAS Hot Fix Analysis, Download and Deployment Tool.

• Due to the output data sets that SAS Forecast Server generates, it is recommended that you have sufficient disk space in the project directory. Sufficient disk space is 30 times the size of the input data set.

• Consider using the SAS Forecast Server batch macros, instead of the user interface, to create and update projects. Examples: %FSCREATE and %FSRUNPRJ. A benefit of using these macros is that you can write simple code one time and continually reuse it. Also, the code serves to document what was done to the project. These macros are documented in the SAS Forecast Server administrator's guide.

For information about settings that might influence performance, see SAS Note 57621: "Slow Performance in SAS^® Forecast Server might be caused by project settings or input data."