|
Autobox
Autobox is AFS's flagship product, providing cutting edge forecasting
for the PC desktop for 28 years. Autobox provides a start to finish
environment designed to make forecasting easy whether you have one
series or one thousand.
Features
Autobox has
a complete set of forecasting features that will appeal to both
novice and expert forecasters. Autobox's automatic features are
unparalleled in breadth and depth of implementation. Autobox is
truely the power forecasters dream tool with a pallette of tools
that allows the forecaster to build models that work. Read on for
a summary of Autobox's features.
General
Modeling Environment
Autobox is
a program with an easy to use design built around its powerful forecast
engine. You can specifiy your own model or run in a batch environment.
Optional
Automatic Modeling
AFS was the
first company to automate the BJ model building process. Our approach
is to program the model identification, estimation and diagnostic
feedback loop as originally described by Box and Jenkins. This is
implemented for both ARIMA (univariate) modeling and Transfer Function
(multivariate or regression) modeling.
What this means is that the user, from novice to expert, can feed
Autobox any number of series and the programs powerful modeling
heuristic can do the work for you. This option is implemented in
a such that it can be turned on at any stage of the modeling process.
There is complete control over the statistical sensitivities for
the inclusion/exclusion of model parameters and structures. These
features allow the user complete control over the modeling process.
The user can let Autobox do as much or as little of the model building
process as you or the complexity of the problem dictates.
Complete set of Box-Jenkins modeling Tools
Autobox comes with a complete set of indentification
and modeling tools for use in the Box-Jenkins framework. This means
that you have the ability to transform or prewhiten the chosen series
for identification purposes. Autobox handles both ARIMA (univariate)
modeling and Transfer Function (multivariate) modeling allowing
for the inclusion of interventions (see below for more information).
Tests for interventions, need for transformations, need to add or
delete model parameters are all available.
Autocorrelation (both traditional and robust), partial autocorrelation
and cross-correlation functions and their respectives tests of significance
are calulated as needed. Model fit statistics, including R²,
SSE, variance of errors, adjusted variance of errors all reported.
Information criteria statistics for alternate model indentification
approaches are provided.
Intervention Detection
One
of the most powerful features of Autobox is the inclusion of Automatic
Intervention detection capabilities in both ARIMA and Transfer Function
models. Almost all forecasting packages allow for interventions
to be included in a regression model. What these packages don't
tell you is how sensitive all forecasting methodoligies are to the
impact of interventions or missing variables. These packages don't
tell you if your series may be influenced by missing variables or
changes that are outside the current model.
If a data series
is impacted by changes in the underlying process at discrete points
in time both ARIMA models and Transfer Function models will produce
poor results. For example a competitors price change changes the
level of demand for your product. Without a variable to account
for this change you forecast model will perform poorly. Autobox
implements ground breaking techniques which quickly and accurately
indentify potential interventions (level shifts, season pulses,
single point outliers and cha nges in the variance of the series).
These variables can then be included in your model at your discretion.
The result is more robust models and greater forecast accuracy.
Graphical
Analysis Tools
Autobox
has a set of graphing tools that help present complex statistical
information in a way that is easy and clear at every stage of the
forecasting process. For example graphs of autocorrelation, partial-autocorrelation
and cross correlation functions are all available. Even more incredibly
these can be compared to theoretical values for various models forms.
Plotting of any combination of variables, included fit versus actual
and forecasts with confidence limits is simply a few mouse clicks
away. Standardization of the variables is always an option before
plotting.
Forecasting
and Diagnostics
All
forecast packages allow for you to produce forecasts using the models
you have constructed. Autobox presents the critical information
you need to determine of those forecasts are acceptable. Autobox
has options that allow you to analyze the stability and forecasting
ability of your forecast model. This is achieved through a series
of ex-poste forecast analyses.
You can automatically withhold any number of observations, reestimate
the model form and forecast. Observations are then added back one
at a time and the model is reestimated and reforecast. Forecast
accuracy statistics, including Mean Absolute Percent Error (MAPE)
and Bias, are calculated at each forecast end point. Thus the stability
of the model and its ability to forecast from various end points
can be analyzed.
Finally,
you can optionally allow Autobox to actually re-identify the model
form at each level of withheld data to see if the model form is
unduly influenced by recent observations.
What-If
Modeling
After developing a model
the user can evaluate alternative strategies for user-specified
future values e.g. price points, promotions, special offers and
get a quick impact report. The user can scan both tabular and graphical
presentations to assess the best strategy.
|
