White Paper

Looking Inside SAS® Forecast Studio

Contents
Abstract................................................................................ 1

Introduction......................................................................... 1

Overview of SAS® Forecast Studio ................................ 2

Creating Forecast Projects............................................... 2

Selecting the Input Data Set.................................................2

Defining the Forecast Hierarchy..........................................3

Assigning Variable Roles.......................................................3

Defining Events.......................................................................4

Setting Additional Forecast Options..................................5

Forecast Generation.......................................................... 5
Time Series Diagnosis and Model Construction.............5

Model Selection......................................................................6

Forecast Summary..................................................................6

Forecasting View................................................................ 6
Navigating the Forecast Hierarchy......................................6

Creating Forecast Filters........................................................7

Entering Forecast Overrides................................................8

Series and Modeling Views............................................. 8

Evaluating Series Properties.................................................8

Analyzing and Comparing Models.....................................9

Editing and Creating Models...............................................9

Scenario Analysis View....................................................10

Exception-Based Forecasting........................................11

Forecast Reports...............................................................12

Conclusion.........................................................................12

References.........................................................................12
SAS White Papers................................................................ 12

Recommended Reading.................................................... 12
 1

Abstract Introduction
Organizations depend on statistical forecasting to provide a Generating forecasts is a crucial step in many strategic and
solid foundation for many important planning and decision- tactical planning and decision-making processes. For example,
making processes. SAS® Forecast Server, which comprises a future demand for products and services is forecast to support
graphical user interface (SAS Forecast Studio) and the statistical production planning, marketing activities, resource scheduling
forecasting engine (SAS Forecast Server Procedures), facilitates and financial planning. Much time and effort is spent in the
and speeds the forecasting process by providing a convenient, planning process, and thus improving the reliability of statistical
user-friendly interface to the large-scale automatic forecasting, forecasts that feed these processes can result in huge rewards
model building and time series exploration capabilities avail- such as greater operational efficiency, reduced expenses and
able in SAS. increased profits.

SAS Forecast Server also includes the SAS Time Series Studio Although forecasting is a key business function, many organiza-
graphical user interface, providing extensive time series explo- tions rely on a small forecasting team to generate large numbers
ration capabilities. The new SAS Forecast Server Client (avail- of statistical forecasts. Therefore, a large degree of automation is
able in release 14.1) makes certain functions of SAS Forecast often required to complete the forecasting process in the time
Studio and SAS Time Series Studio available from anywhere available during each forecasting and planning cycle. Forecasting
through a web browser. Find more information about SAS process tasks typically include generating forecasts based on
Forecast Server Client in the white paper “Looking Inside SAS® recent data, reconciling forecasts in a hierarchical manner, identi-
Forecast Server Client” (forthcoming). fying and correcting problematic forecasts, adding overrides to
the forecasts based on business knowledge or scenario analysis
Through SAS Forecast Studio, SAS Forecast Server addresses and publishing the forecasts to other systems or as reports.
the needs of novice forecasters by being largely automated, yet
still meets the needs of more experienced analysts by providing SAS Forecast Server, which consists of a graphical user interface
layers of sophistication that can be accessed as needed. (SAS Forecast Studio) as well as the statistical forecasting engine
SAS Forecast Server Procedures, facilitates and speeds the fore-
SAS Forecast Studio enables users to set up forecasting casting process by providing a convenient, user-friendly inter-
projects, perform large-scale automatic forecasting, identify face to the large-scale automatic forecasting, model building
exceptions, override forecasts and construct their own models if and time series exploration capabilities available in SAS. In
desired. Given the scale of many forecasting problems, addition, the SAS Time Series Studio graphical user interface
manually customizing many statistical models may not be provides extensive time series exploration capabilities. And SAS
feasible. SAS Forecast Studio provides an automated system Forecast Server Client makes certain functions of SAS Forecast
that selects appropriate models and intelligently chooses influ- Studio and SAS Time Series Studio available from anywhere
ential variables that improve the model. Forecasts that violate through a web browser.
business rules can be flagged for further attention. The system
supports hierarchical forecasting processes by providing The wizard-driven graphical user interface of SAS Forecast Studio
top-down, middle-out and bottom-up forecast reconciliation allows the novice forecaster to move quickly through the statis-
and generates SAS code that can be run in a batch tical forecasting process to generate forecasts using state-of-the-
environment. art forecasting methods. Forecasters do not need to know any
SAS programming or how to build time series models to use SAS
Forecast Server – the system can be entirely automated.

Although the SAS Forecast Studio user interface is streamlined

to enable less-experienced forecasters to easily generate fore-
casts, SAS Forecast Studio has all the power and sophistication
that more advanced analysts expect. SAS Forecast Studio
provides a wide array of diagnostic and model-building tools,
enabling more experienced analysts to benefit from the
productivity of the analyst interface and to utilize even more of
the SAS Forecast Server Procedures computational engine.
 2

Overview of SAS® Forecast Creating Forecast Projects

Studio SAS Forecast Server organizes forecasts as projects. A fore-
casting project may be small, or it may contain thousands of
As the graphical user interface for SAS Forecast Server, SAS
time series. Organization of forecasts and all the information
Forecast Studio has been designed to speed the work of both
supporting them in projects has many advantages:
novice and advanced analysts:
• Each member of a forecasting team may have his or her own
• A New Project Wizard walks the analyst through the steps of
projects, allowing him or her to work independently.
setting up and running forecast projects.
• Forecasters may maintain separate projects for the different
• An Events dialog box simplifies the addition of events – for
accounts or product lines for which they are responsible.
example, holidays and promotions – that might influence the
forecast. • Forecasters may maintain separate projects for each fore-
casting cycle – the Q1 forecast, the Q2 forecast and so on.
• A Model Builder dialog box guides the analyst through the
development of custom forecasting models, built from
The SAS Forecast Studio user interface includes an eight-step
scratch or based on models in the model repository.
Project Wizard to guide analysts quickly through setting up a
• Forecast, Series and Model views give the analyst multiple new project or working with existing projects.
ways to work with the forecast results.
• A Scenario Analysis view allows the analyst to create what-if Project folders – consisting mostly of SAS programs, SAS data
scenarios based on the models generated by SAS Forecast sets and XML files – contain information about the input time
Studio. series, event data, analyst-specified preferences, statistical
• Default and custom reports are readily generated in SAS models, parameter estimates, scenarios, SAS forecasting code
Forecast Studio. and ultimately, the forecasted values. The input data is not
stored within the project itself and does not need to be moved
• Forecasts created with SAS Forecast Studio can be exported
or copied to where projects are stored.
to a variety of other applications.

Selecting the Input Data Set

SAS® Forecast Server Fundamentally, forecasting is the process of detecting patterns
over time, and then using those patterns to project into the
future. Therefore, all data sets must contain a time variable, such
SAS Forecast Studio
®

Graphical User Interface

Windows Client

SAS Forecast
®

Server Client*
Segmentation and Tracking
Web Client

SAS Management
®
SAS Forecast Server
®
SAS Add-In for
®

Console Procedures Microsoft Office

Project Administration Forecast Engine SAS Enterprise Guide
® ®

Tasks SAS code

Forecasting Tasks

SAS/ETS ®

*SAS Forecast Server Client comes with

SAS Foundation SAS Forecast Server (v14.1 and later)
Technologies on WX6 and LAX.

Figure 1: An overview of the components of SAS® Forecast Server.

 3

as month or day. For example, the data set may contain data on (SKUs) at the bottom level, and then aggregated upward by
units sold per week, or electricity used per hour. If the data is product line, region and finally, an overall forecast. In this case,
recorded at regular intervals such as month or day, then it is the forecast hierarchy has four levels: product, product line,
time series data and is ready to be forecast. On the other hand, region and overall. “Product” is the fundamental unit to be
if data is collected at irregular intervals, such as raw transaction forecast; “Product Line” and “Region” are SAS BY variables.
data from a retail store or call center, then it must be aggre-
gated by time period to create a proper time series. If needed, SAS Forecast Studio produces separate statistical forecasts at
SAS Forecast Studio performs this aggregation automatically. each level of a user-specified hierarchy, based on the history at
each level. To generate the forecasts at upper levels of the hier-
The Project Wizard begins by prompting the analyst to navigate archy, the historical values from the lowest level are summed up
through SAS libraries and select the input data set. There is no to create the required input data time series.
predefined limit on the size of the data set or the number of
series to be forecast in a project. In practice, the limits on project Since each level of a hierarchy is forecast independently, it is
size will depend on the complexity of the data to be analyzed highly unlikely that the statistical forecast at the second level of
and the amount of history available for each series to be the hierarchy will equal the sum of the statistical forecasts at the
forecast. first level, and so on. In this case, the forecasts must be recon-
ciled within the hierarchy, so that the forecasts across the hier-
Defining the Forecast Hierarchy archy levels are consistent (i.e., lowest level forecasts add up to
the next level, and so on). The Project Wizard guides the user
Many forecasts are hierarchical; for example, the forecast for
through the selection of an appropriate reconciliation tech-
units sold might include the number of individual products
nique: top down, bottom up or middle out.

Figure 2: Defining a forecasting hierarchy.

Figure 3: Example of a forecasting hierarchy.

 4

Assigning Variable Roles Events can be added to regression, ARIMAX and unobserved
components models (UCM) as event variables: variables that
Variable roles determine how input variables are used within a
indicate when something out of the ordinary occurred in the
project. The required time variable is assigned a role of “Time
past or will occur again in the future. Once specified, SAS
ID.” As noted above, other input variables may be assigned the
Forecast Server statistically estimates the impact of the event in
BY variable role and will determine the forecast’s hierarchical
the past and uses the estimated impact to calculate future fore-
structure.
casts where the event recurs.

The Project Wizard steps the analyst through the assignment of

The Project Wizard’s Event Manager guides the user through
variable roles. The variable to be forecast – “units” in the
selecting, creating, editing, combining or deleting events. SAS
preceding example – is assigned the role of dependent
Forecast Studio includes a large collection of predefined events
variable. In hierarchical forecasts, there can be only one depen-
(mostly public holidays). In addition, externally generated tables
dent variable. Other possible variable roles include:
of events can be imported easily into SAS Forecast Studio. To
• Independent variables – used as potential explanatory inputs define a custom event within SAS Forecast Studio, the user
in the forecast models. specifies when the event occurred or will occur, the frequency
• Reporting variables – used in presenting the results of the of recurrence and the appropriate shape. For example, an event
forecast. may be very short-lived (such as a power outage or the
Olympics) and modeled as a pulse of limited duration, or an
• Adjustment variables – used to transform variables for fore-
event may represent an ongoing change (such as a new law or
casting (for example, conversion to a common currency).
regulation, or permanent price change) and thus modeled as a
permanent level shift. Once defined, events
may be saved externally for use by other
forecasters or in other forecast projects.

Figure 4: Defining variable roles.

Defining Events
Some models may be improved by the inclusion of events,
which are occurrences out of the ordinary that may disturb the
underlying time series. Some events are unplanned (such as
strikes or storms); their impact should be isolated so that it is not
propagated in future forecasts. Other events are planned
(holidays, promotions, price changes) and may be recurring;
their impact should be assessed and included in future fore-
casts if appropriate. Figure 5: Defining events.
 5

Setting Additional Forecast Options Forecast Generation

When the basic components of a forecasting project are in
Upon closing the Project Wizard, SAS Forecast Studio automati-
place – the hierarchy defined, variable roles assigned and
cally generates statistical forecasts for all series in the project. If a
events created – the Project Wizard guides the analyst through
hierarchy has been specified, the system will reconcile the statis-
the final, optional process of setting additional forecast param-
tical forecasts according to the options specified by the user in
eters, including:
the Project Set-Up wizard.
• How missing values in the data are to be interpreted.
• Which fit statistic should be used for model selection. During processing, SAS Forecast Studio will:

• How sensitive automatic outlier detection and other diag- • Diagnose the characteristics of each series to be forecast, at
nostic tests should be. each level of the hierarchy.
• How leading and trailing zeros in the time series are to be • Estimate parameters for appropriate candidate models.
treated. • Evaluate the performance of the candidate models.
• Whether or not forecasts are allowed to go below zero. • Select the best-performing model for each series, as well as
• Whether or not to use a holdout sample, and if so, its length alternative runner-up models.
definition. • Produce forecasts.
• Reconcile the forecasts in the hierarchy.
SAS Forecast Studio offers standard default settings for all of
these options, simplifying the forecast preparation task for
novice analysts and streamlining the process for advanced Time Series Diagnosis and Model
forecasters. Construction
When generating forecasts, SAS Forecast Server must first
determine an appropriate set of models for each time series
being forecast. Several diagnostic procedures are automatically
run for each time series in the forecast project. For example, in
addition to the standard time series decomposition analysis,
SAS Forecast Server tests for intermittency of the dependent
variable, as well as the presence of outliers and structural shifts.
If a time series is determined to be intermittent – that is, very few
nonzero observations in the data – SAS Forecast Server will
attempt to fit a specialized intermittent demand model.
Otherwise, SAS Forecast Server constructs candidate exponen-
tial smoothing, ARIMAX and UCM models.

In addition to testing for intermittent dependent variables, SAS

Forecast Server also examines each time series for outliers and
shifts, and takes outliers and level shifts into consideration when
constructing the forecast models.

Figure 6: Project settings can be easily customized.

SAS Forecast Studio performs many of the same diagnostic
tests on independent and event variables as are performed on
dependent variables. An important feature of SAS Forecast
In addition, many businesses develop custom forecasting
Studio is the ability to determine whether any of the indepen-
models that are used by many analysts throughout the
dent or events variables are dynamically related to the depen-
company. The Project Wizard allows analysts to specify custom
dent variable, and to specify the model accordingly. For
models to import into the project for use in the forecasting
example, when forecasting the demand for a product that
process, thus providing a means of further customizing the fore-
exhibits a relationship with both product price and special
casting environment to fit the needs of both the forecast analyst
promotional campaigns, SAS Forecast Server would detect and
and the business.
specify these separate dynamic relationships in an ARIMAX or
UCM forecasting model.
 6

Model Selection Forecast Summary

When the properties of the time series have been evaluated, When the forecast process is complete, the Forecast Summary
SAS Forecast Studio next evaluates a wide range of models, window provides an overview of important properties of the
including unobserved components models (UCM), exponential forecasting project by hierarchy level. The number of series
smoothing models (ESM) and autoregressive integrated moving forecast at each level of the hierarchy is given, as is the number
average models (ARIMA and ARIMAX), to determine which of forecast failures (series that could not be statistically forecast,
model best fits the time series data. usually because the series contained too few observations). A
bar chart displaying the distribution of fit statistics is shown, as
SAS Forecast Studio features more than four dozen model-fit are bar charts giving the distribution of model family and model
statistics, including scale dependent measures such as MSE characteristics (seasonal model, input variables present and
(Mean Square Error), measures based on percentage error such outliers detected.)
as MAPE (Mean Absolute Percentage Error), conventional statis-
tical measures such as AIC (Akaike’s Information Criterion) and
many others. Forecasting View
The detailed output from a SAS Forecast Studio project is
Holdout samples can be specified so that forecasting models
accessed through a set of tabbed views. The first of these, the
are selected not only by how well they fit the past data, but by
Forecasting view, displays the results of each time series
how well they are likely to predict the future. If a holdout sample
forecast in a set of panels. The forecast graph panel shows the
is specified – the preferred practice when enough data is avail-
statistical forecast with 95 percent prediction intervals for each
able – the model will be selected based on the goodness-of-fit
forecast; historical values are plotted as are the fitted values for
statistic in the holdout region (out-of-sample fit); otherwise
the historical time period. The data table shows the historical
in-sample fit is used to select the model. If a holdout sample
data, the output of the statistical forecast and the fit statistic. If
was specified, the model parameters are estimated using the
the model is part of a reconciled hierarchy, the reconciled
full range of data, including the holdout sample.
(adjusted) forecast and fit statistic are displayed. Manual
forecast overrides are also shown in the data table. The model
When selecting the best-fitting model, SAS Forecast Studio
details – type of model as well as parameter estimates – are
automatically tests candidate independent variables and events
revealed in a pop-up window.
­– identified during the forecast setup process – and determines
how they should be used in the forecast models. In addition to
examining the contemporaneous relationships between inde- Navigating the Forecast Hierarchy
pendent and dependent variables, lagged and dynamic rela- Output is generated for every time series in the forecast project,
tionships are explored. If appropriate, variable transformations, at every hierarchy level. To facilitate navigating through the
lags and transfer function definitions are calculated. forecast hierarchy, SAS Forecast Studio provides two navigation
modes: tree and table. In the Tree view, an expandable view of
the forecast hierarchy is show in the navigation panel; the high-
lighted series is “active” and the corre-
sponding output is shown in the graph and
data table. Clicking up and down the hier-
archy tree changes the active series and the
displayed output.

Figure 7: Forecast model statistics.

 7

Figure 8: Forecasting View: Tree form.

Creating Forecast Filters
In the Table Navigation view, the project series are listed in
tabular form; columns for each BY variable are shown, as are the Filters are automatically created for each hierarchy level and a
statistical and reconciled fit statistics. The table can be sorted by Filter dialog box guides the user through creating additional
clicking on the column headers. In addition, a drop-down filter custom filters. For example, a filter can be created that isolates
selector can be used to subset the displayed forecasts; filters all the forecasts with a fit statistic value above a specified level,
are automatically generated for each BY variable in the forecast such as MAPE values greater than 5 percent. Filters are the basis
hierarchy. of exception-based forecasting. Using filters, the forecast analyst
can quickly identify the forecasts that are
performing poorly and focus additional
modeling attention on those time series.
Other types of filters include filters on model
property (for example, model family or inclu-
sion of independent variables), series proper-
ties (for example, number of missing values,
maximum value or date of first and last obser-
vation) and if applicable, reconciled fit statistic.

Figure 9: Forecasting View: Table form.

 8

Series and Modeling Views

By using forecast filters to identify forecasts warranting further
investigation because, perhaps, they exceed a forecast fit crite-
rion threshold, the analyst can focus additional diagnostic and
modeling efforts where the value to the forecasting process is
greatest. Two SAS Forecast Studio views – the Series view and
the Modeling view – help the analyst delve deeper into the fore-
casts generated by SAS Forecast Server.

Evaluating Series Properties

Opening the Series view in SAS Forecast Studio reveals a
number of plots displaying properties of the active time series,
including:

• Seasonal decomposition plots.

• Component plots.
Figure 10: Forecast filtering. • Autocorrelation and white noise probability plots.
• Cross-series plots.

Entering Forecast Overrides In addition, in the Series view the analyst can explore the prop-
Statistical forecast values can be manually overridden by erties of both dependent and independent variables by experi-
entering new values in the data table; the numbers can be menting with variable transforms using drop-down selection
typed in, or the override calculator can be used to calculate new boxes. As in the Forecasting view, navigation among the time
values by adjusting the statistical value by a fixed amount or a series is done with the Tree view or Table view in the navigation
percentage. The override values are shown in the forecast panel.
graph, and, after forecast reconciliation, the impact of the over-
rides is shown at each level of the forecast hierarchy.

Figure 11: Series View.

 9

Analyzing and Comparing Models Editing and Creating Models

The heart of SAS Forecast Studio’s advanced modeling capabili- Within the Modeling view, analysts can copy and edit models
ties lies in the Modeling view. that were automatically generated by SAS Forecast Studio, or
develop new models from scratch using the Modeling dialog
For the active forecast series, a list of the three most appropriate box. From within the Modeling dialog box, users can select a
candidate models, based on model diagnostics and the chosen model family and edit model parameters, add independent
fit statistic, is shown. Clicking on a model activates it and variables and events to a model and so on. By using SAS
displays details of the model as well as several diagnostic plots, Forecast Studio’s powerful Modeling dialog box, users create
including plots of: virtually any form of ARIMA, subset and factored ARIMA, unob-
served components, exponential smoothing, multiple linear
• Prediction errors.
regression or intermittent demand model. The options offered
• Prediction error distribution. in the Modeling dialog box vary dynamically with the type of
• Autocorrelations of the prediction errors (and standardized model chosen.
autocorrelations).
• Partial autocorrelations of the prediction errors (and stan-
dardized partial autocorrelations).
• Inverse autocorrelations of the prediction errors (and stan-
dardized inverse autocorrelations).
• White noise probabilities for the prediction errors (also on
log scale).
• Components of the series (smoothed trend, season and level
states).

In addition, tables of model parameter estimates and forecast

values are readily accessible. The Modeling view also provides
tools for comparing multiple models, including tables showing
the values of a large number of fit statistics as well as model
comparison graphs.

Figure 13: Sample modeling dialog box for

ARIMA model.

Figure 12: Model comparison.

 10

Figure 15: Cross-series plot.

Figure 14: Model repository.
SAS Forecast Studio scenarios are based on models that use
When the model specification is complete, the parameters of independent variables, such as regression, ARIMAX and unob-
the new model are estimated and the model is added to the served components models. Using the Scenario Analysis dialog
comparison table of models. box, a scenario is built based on a relationship (model) already
estimated by the forecasting engine. New values for the inde-
When a model is edited or created, the plots in the Modeling pendent variables are entered and the model is rerun, gener-
view are automatically updated, enabling users to easily ating new forecast values. Multiple scenarios can be run and
evaluate the new model and compare it to existing models. In saved and the results compared. In addition, a scenario’s results
addition, models constructed or edited by the user can be can be used as forecast overrides and incorporated in the final
added to the project’s custom model repository for future use. reconciliation of the forecasting project.
In this way, the model repository can become increasingly
tailored to the business or forecasting problem over time.

Scenario Analysis View

Forecasters and business analysts are frequently asked to
conjecture about the future: “How many more units will be sold
if the price is lowered by 5 percent?” or “What would the impact
on production be of another late-winter blizzard?” The answers
to questions like these can have a direct impact on the final
forecast. One approach is to simply guess at the forecast impact
by manually entering overrides into the forecast data table in
the Forecasting view. A more sophisticated approach, however,
is used in SAS Forecast Studio’s Scenario Analysis view, in which
the dynamic relationship between input variables and the
forecast values, diagnosed and modeled by SAS Forecast
Server, forms the crux of the scenario analysis. Figure 16: Overriding forecast based on scenario analysis.
 11

Exception-Based Forecasting By following the principles of exception-based forecasting, an

analyst can quickly identify the forecasts which are performing
When faced with forecasts for thousands of time series, poorly and focus additional attention on those time series. The
manually reviewing each forecast is simply not feasible – the net result of the focused effort stemming from exception-based
effort and time required would not be commensurate with the forecasting is a significant increase in efficiency of the entire
reward. Instead, forecasters and planners turn to exception- forecast process and a shortening of forecast-cycle time.
based forecasting, in which business rules and forecasting
criteria are established, forecasts that meet all the rules and
criteria are automatically passed on, and the analysts focus their Forecast Reports
attention on those forecasts that fail to meet a rule or pass a
In addition to the powerful forecasting and scenario analysis
forecasting criterion. These are the forecast exceptions.
capabilities of SAS Forecast Studio, several reporting tools are
available. Graphs can easily be copied and pasted into presen-
tations, and users can choose among several predefined graph
styles as well as control the format of the data table.

The predefined reports in SAS Forecast Studio are based on

SAS Stored Processes, which comprise an open and extensible
system. Users can write custom reporting stored processes and
add them to the reports catalog. Predefined reports include
reports of the final forecast (forecast values and plots) as well as
detailed reports of series and model diagnostics and hierarchy
reconciliation.

Figure 17: Forecast exceptions.

All the tools needed to set up business rules and flag potentially
problematic forecasts are readily available in SAS Forecast
Studio:

• Business rules and forecasting criteria can easily be set up

using the Filter Generation dialog boxes. For example, the
analyst may be happy with all forecasts with a MAPE less
than 5 percent but may want to examine forecasts with a
MAPE value that exceeds this threshold.
• Using the filter drop-down box in the List view, the list of fore-
casts requiring additional scrutiny can be generated quickly.
• Using the tools in the Forecasting view, forecasts can be
manually adjusted using the forecast override function. Figure 18: Available reports and stored processes.
• In the Series view, the properties of the time series under-
lying a forecast exception can be evaluated, giving insight
into the behavior of the forecast.
• Tools in the Model view allow the analyst to create or edit
forecast models to generate, if possible, forecasts that
comply with business rules and forecast exceptions.
 12

Conclusion References
SAS Forecast Studio delivers the sophisticated time series SAS White Papers
exploration, model building and large-scale automatic fore-
Turbo-Charging Spreadsheets: Accessing SAS® Forecast Server
casting capabilities of SAS to novice forecasters through the
from Microsoft Excel.
SAS Forecast Studio interface. Advanced users can access SAS
Forecast Server Procedures (writing SAS code) if they so desire.
Large-Scale Automatic Forecasting Using Inputs and
Thus, as an extensible and flexible system, SAS Forecast Studio
Calendar Events.
meets the needs of both novice forecasters who need to move
through the production forecasting process quickly as well as
Structured, Large-scale Statistical Forecasting Using
more experienced forecast analysts who wish to delve deeply
SAS® Forecast Server
into the forecast model building process.

SAS Forecast Studio supports exception-based forecasting; Recommended Reading

forecasts that violate business rules or forecasting criteria can be Box, G. E. P, Jenkins, G. M., and Reinsel, G. C. (1994), Time Series
easily identified, allowing the analysts to focus their time and Analysis: Forecasting and Control, Englewood Cliffs, NJ:
energy where it is most needed. Prentice Hall Inc.

The scenario analysis capability in SAS Forecast Studio provides Brocklebank, John C. and Dickey, David A. (2003),
tools for what-if modeling, using the models and parameter SAS® for Forecasting Time Series, Second Edition.
estimates generated by the forecasting system or provided by
the user. This means that SAS Forecast Studio scenarios Chatfield, C. (2000). Time Series Models, Boca Raton, FL:
leverage the underlying, potentially complex, dynamic relation- Chapman & Hall/CRC.
ship between the forecast variable and the independent vari-
ables, adding significantly to their value in the business Gilliland, M. (2010). The Business Forecasting Deal, Hoboken,
planning process. NJ: John Wiley & Sons Inc.

Makridakis, S. G., Wheelwright, S. C., and Hyndman, R. J. (1997),

Forecasting: Methods and Applications, New York:
John Wiley & Sons Inc.
To contact your local SAS office, please visit: sas.com/offices

SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of
SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product
names are trademarks of their respective companies. Copyright © 2015, SAS Institute Inc. All rights reserved.
102418_S144585.0915