What's New Table of Contents |
What's New in SAS High-Performance Forecasting 2.2
Overview
Note: This section describes the features of SAS High-Performance Forecasting that are new or enhanced since SAS 9.0.
SAS High-Performance Forecasting has a new release numbering scheme. SAS High-Performance Forecasting 2.2 provides new procedures, features, and functionality while maintaining all the capabilities of SAS 9.1.3 High-Performance Forecasting software.
New features, procedures, and functions have been added to SAS High-Performance Forecasting as follows:
- New features have been added to the HPF procedure.
- New procedures have been added for creating model specifications and model
selection lists:
- PROC HPFARIMASPEC
- PROC HPFESMSPEC
- PROC HPFEXMSPEC
- PROC HPFIDMSPEC
- PROC HPFSELECT
- PROC HPFUCMSPEC
- The new HPFEVENTS procedure has been added for creating calendar events.
- New procedures have been added for computations:
- PROC HPFDIAGNOSE
- PROC HPFENGINE
- The new HPFRECONCILE procedure has been added for reconciliation of hierarchical forecasts.
- New experimental functions have been added for forecast scoring:
- HPFSCSIG function
- HPFSCSUB function
Details
High-Performance Forecasting Release Numbering Scheme
SAS High-Performance Forecasting has a new release numbering scheme. SAS High-Performance Forecasting 2.2 provides the same features and functionality as SAS 9.1.3 High-Performance Forecasting software, and includes new procedures, functions, and features.
HPF Procedure
The HPF procedure has the following new features:
- New options related to forecast model selection:
- HOLDOUTPCT= option
- SEASONTEST= option
- SELECT= option
- The new NBACKCAST= option is related to forecast model initialization.
- The new IDM statement is related to intermittent demand models.
- New options related to out-of-sample forecast performance statistics:
- BACK= option
- PRINT=PERFORMANCE option
- PRINT=PERFORMANCESUMMARY option
- PRINT=PERFORMANCEOVERALL option
- New options related to the processing of the data:
- NOTSORTED option
- REPLACEBACK option
- SORTNAMES option
- New options related to printed and graphical output:
- PLOT= option
- PRINT=STATES option
- The new MAXERROR= option is related to the message log.
HPFARIMASPEC Procedure
The new HPFARIMASPEC procedure is used to create an Autoregressive Integrated Moving Average (ARIMA) model specification file. The output of the procedure is an XML file that stores the intended ARIMA model specification. This XML specification file can be used to populate the model repository used by the HPFENGINE procedure. (Likewise, the XML files generated by the other model specification procedures in this section can also be used to populate the model repository used by PROC HPFENGINE.)
HPFESMSPEC Procedure
The new HPFESMSPEC procedure is used to create an Exponential Smoothing Model (ESM) specification file. The output of the procedure is an XML file that stores the intended ESM model specification.
HPFEXMSPEC Procedure
The new HPFEXMSPEC procedure is used to create an External Model (EXM) specification file. The output of the procedure is an XML file that stores the intended EXM model specification.
HPFIDMSPEC Procedure
The HPFIDMSPEC procedure is used to create an Intermittent Demand Model (IDM) specification file. The output of the procedure is an XML file that stores the intended IDM model specification.
HPFSELECT Procedure
The new HPFSELECT procedure is used to create model selections lists. A model selection list contains references to candidate model specifications stored in the model repository. The output of the procedure is an XML file that stores the intended model selection list.
The HPFSELECT procedure has the following experimental features:
- external variable mapping (EXMMAP= option)
- external forecast function (EXMFUNC= option)
HPFUCMSPEC Procedure
The new HPFUCMSPEC procedure is used to create an Unobserved Component Model (UCM) specification file. The output of the procedure is an XML file that stores the intended UCM model specification.
HPFEVENTS Procedure
The HPFEVENTS procedure provides a way to create and manage events associated with time series. The procedure can create events, read events from an events data set, write events to an events data set, and create dummies based on those events, if date information is provided.
A SAS event is used to model any incident that disrupts the normal flow of the process that generated the time series. Examples of commonly used events include natural disasters, retail promotions, strikes, advertising campaigns, policy changes, and data recording errors.
An event has a reference name, a date or dates associated with the event, and a set of qualifiers. The event exists separately from any time series; however, the event may be applied to one or more time series. When the event is applied to a time series, a dummy variable is generated that may be used to analyze the impact of the event on the time series.
HPFDIAGNOSE Procedure
The new HPFDIAGNOSE procedure is an automatic modeling procedure to find the best model among ARIMA Models, Exponential Smoothing Models, and Unobserved Component Models.
The HPFDIAGNOSE procedure has the following functionality:
- intermittency test
- functional transformation test
- simple differencing and seasonal differencing test
- tentative simple ARMA order identification
- tentative seasonal ARMA order identification
- outlier detection
- significance test of events
- transfer functions identification
- intermittent demand model
- exponential smoothing model
- unobserved component model
HPFENGINE Procedure
The new HPFENGINE procedure provides large-scale automatic forecasting of transactional or time series data. The HPFENGINE procedure extends the foundation built by PROC HPF, enabling the user to determine the list of models over which automatic selection is performed.
The use of many forecast model families is supported when HPFENGINE is used in conjunction with new procedures that generate generic model specifications. Among these models are the following:
- ARIMA
- Unobserved Component Models (UCM)
- Exponential Smoothing Models (ESM)
- Intermittent Demand Models (IDM)
- External Models (EXM)
Furthermore, users can completely customize the operation by defining their own code to generate forecasts.
For models with inputs, the STOCHASTIC statement is especially helpful for automatically forecasting those inputs that have no future values.
Also supported is the generation of a portable forecast score. The output of the SCORE statement is a file or catalog entry that, when used with the new function HPFSCSUB, can be used to efficiently generate forecasts outside of the HPFENGINE procedure.
The new HPFDIAGNOSE procedure produces output that is compatible with HPFENGINE. As a result, the task of candidate model specification can be entirely automated.
HPFRECONCILE Procedure
The new HPFRECONCILE procedure provides the following functionality:
- large-scale reconciliation of hierarchical forecasts by using predictions, standard errors, and confidence limits reconciliation
- the following methods of reconciliation:
- top-down reconciliation
- bottom-up reconciliation
- constrained reconciliation using an interior-point quadratic programming technique that combines statistical forecasts reconciliation with judgmental constraints (overrides)
- replaces functionality of the %HPFRECON macro
HPFSCSIG Function
The experimental HPFSCSIG function generates a sample signature for subsequent use by the HPFSCSUB function.
HPFSCSUB Function
The experimental HPFSCSUB function uses score files to produce forecasts outside of the HPFENGINE procedure. Being a function, it is particularly well suited for use within other SAS programming contexts, such as the DATA step, or procedures that permit the specification of functions, such as the NLP procedure. The only input required is a reference to the score function, the horizon, and future values of any inputs.