Analytics and Statistical Forecasting

Analytics and Statistical
Forecasting
Overview
Copyright Oracle, 2007. All rights reserved.
Analytics and Statistical Forecasting

Effective mm/dd/yy
Page 1 of 27
338445155.doc
Rev 1

System References
None
Distribution
Oracle Demantra Fundamentals
Job Title*
Ownership
The Job Title [list@YourCompany.com?Subject=338445155.doc] is responsible for ensuring that
this document is necessary and that it reflects actual practice.

Effective mm/dd/yy
Page 2 of 27
338445155.doc
Rev 1
Schedule:
Timing
90 minutes
45 minutes
105 minutes
Topic
Lecture
Practice
Total

Effective mm/dd/yy
Page 3 of 27
338445155.doc
Rev 1
Objectives
Objectives

Effective mm/dd/yy
Page 4 of 27
338445155.doc
Rev 1
Agenda
Agenda

Effective mm/dd/yy
Page 5 of 27
338445155.doc
Rev 1
Overview
Overview

Effective mm/dd/yy
Page 6 of 27
338445155.doc
Rev 1
Understanding Analytics
The Analytical Engine is designed for large-scale installations handling analysis and modeling of
tens to hundreds of thousands of different demand patterns.
Instructor Note
Refer to Guided Demo Using the Analytical Engine [LAB404BY]

Effective mm/dd/yy
Page 7 of 27
338445155.doc
Rev 1
Within an Oracle Demantra solution, the Analytical Engine runs periodically (in the
background), reading data from the database and generating forecast data. The forecaster uses a
worksheet to view the forecast and make adjustments, saving those changes to the database.
The updated forecast is available to all users with the appropriate authorization.

Effective mm/dd/yy
Page 8 of 27
338445155.doc
Rev 1
Demantra Forecasting Ideology
Demantra Forecasting Ideology

Demantra uses a Best of Breed approach to forecasting. When you run the Analytical Engine, a
series of forecasting models run on a series. The Analytical Engine chooses the model that best
meets the chosen selection criteria.
Note: Any information not captured by selected model is lost.
The Analytical Engine can also use a Bayesian Estimator, in which a library of models is run and
based on the goodness of fit the models combined into a single result set. Unlike with best-ofbreed forecasting, using the Bayesian Estimator maintains the nuances of each model.
The Analytical Engine uses a set of theoretical models, each trying to explain history using
different methods and algorithms:
Regression:
Regression

Effective mm/dd/yy
Page 9 of 27
338445155.doc
Rev 1
Log (log transformation before regression)
CMReg (Markov chain selection of subset of causal factors)
Elog (uses Markov chain after log transformation)
Exponential smoothing:
Holt
Bwint
Intermittent Models:
CMReg for Intermittent
Regression for Intermittent
Croston
Time Series Models:
ARX and ARIX
Logistic and AR Logistic
Other Models
BWint (a mixture of regression and exponential smoothing)
DMULT (Multiplicative)

Effective mm/dd/yy
Page 10 of 27
338445155.doc
Rev 1
Analytic Engine Modes
Analytics Engine Modes

Depending on your license, the Analytical Engine runs in one of two modes:
Demand Management (DM):
Generates base-only causals
Generates base forecast
Promotions Effectiveness (PE):
Generates Base and Promotional casuals
Forecast decomposed to base and lift
Causal factors generalized and coefficients generated

Effective mm/dd/yy
Page 11 of 27
338445155.doc
Rev 1
Batch vs. Simulation Modes
Batch vs. Simulation Modes

The Analytical Engine runs in two modes:
Batch mode:
Runs periodically when system is offline
Refreshes the entire forecast set
Simulation mode:
Only runs on a designated subset of combinations
Faster than batch run
Used for what if analysis
Useful for quickly evaluating changes in history, causal factors or promotions


Effective mm/dd/yy
Page 12 of 27
338445155.doc
Rev 1
Simulation mode is launched by users on an as-needed basis

Effective mm/dd/yy
Page 13 of 27
338445155.doc
Rev 1
Demand and Causal Factors
Demand and Causal Factors

Historical demand data is imported from external systems. Causal factors provide information
about historical events that are expected to recur in the future.
Typically, the data that is actually imported needs to be adjusted by the forecasters, as they apply
their own knowledge to better describe the history.
Causal factors cause demand to deviate from a trend. More specifically, a causal factor is a timevarying quantity (such as promotions, price, season, or day of the week) that affects demand.
Demantra requires historical data for causal factors, as well as future data that describes expected
occurrences that will affect demand.
In the case of Promotions Effectiveness, you also configure promotional causal factors, influence
ranges, and influence groups, all of which control how the Analytical Engine determines the
effects of promotions on the forecast.

Effective mm/dd/yy
Page 14 of 27
338445155.doc
Rev 1
Causal Factors
Causal Factors
Global causal factors: Apply to all item-location combinations. For example, a season is a global
causal factor. Most Demantra implementations use global factors.
Local causal factors: Apply to specific item-location combinations. For example, a discount
applied to a specific item in a specific sales region is a local causal factor.

Effective mm/dd/yy
Page 15 of 27
338445155.doc
Rev 1
Understanding Forecasting Models
Understanding Forecasting Models

The Analytical Engine uses a set of theoretical models, each trying to explain history using
different methods and algorithms. Example forecast models include:
Regression
Arx and Arix
Monte-Carlo Regression
Holt
Moving Average
Intermittent

Effective mm/dd/yy
Page 16 of 27
338445155.doc
Rev 1
When generating a forecast, the Analytical Engine uses your history determines which model is
most likely to produce a quality forecast. Alternately, you can manually configure which
forecasting models to use from within the Business Modeler.
For a complete list of models, see Theoretical Engine Models in the Consultant help.

Effective mm/dd/yy
Page 17 of 27
338445155.doc
Rev 1
Optimizing the Forecast Level
Optimizing the Forecast Level

Users specify a forecast level based on their business needs. When Demantra generates a
forecast, all valid nodes are attempted at this level. At each of these nodes all appropriate models
are attempted.
Sometimes a forecast may fail at the specified level for several reasons:
Not enough data
Forecast generated does not have sufficient accuracy
Other validation rules not passed
If forecast for all models in the node fails, Demantra attempts to generate a forecast at a higher
level of aggregation.
The mechanism used to generate this forecast is the Forecast Tree.

Effective mm/dd/yy
Page 18 of 27
338445155.doc
Rev 1
Forecast Trees
Forecast Trees
How your data is aggregated is configured using a Forecast Tree. Forecast trees define initial
forecast level, and enables contingencies if:
Data unavailable
Too much noise
Models not finding the appropriate solution
The forecast tree determines which item/location aggregation combination the engine will work
on. The highest level (highest number) is ALWAYS all items/locations not actually used. The
second highest level is the FIRST level actually examined for the forecast.
In general, forecasting is most accurate when it can be performed at the lowest possible allowed
aggregation level. However, sometimes there is not enough data at that level for all
combinations. For those combinations, the Analytical Engine aggregates the data to a higher

Effective mm/dd/yy
Page 19 of 27
338445155.doc
Rev 1
level and tries to generate a forecast there. The purpose of the forecast tree is to organize data for
this process.
The forecast tree determines which item/location aggregation combination the engine will work
on:
The highest level (highest number) is ALWAYS all items/locations not actually used.
The second highest level is the FIRST level actually examined for the forecast.

Effective mm/dd/yy
Page 20 of 27
338445155.doc
Rev 1
Forecast Tree Example
Forecast Tree Example

In this example, there is enough data at the SKU-store level for SKU 001 and SKU 002; the
Analytical Engine can generate a forecast at that level for those SKUs.
On the other hand, there is less data for SKU 003, so the Analytical Engine aggregates data for
that SKU across all the stores in Region AA, generates the forecast for those SKUs at the SKUregion level, and then splits to the store level.
The Analytical Engine uses proportional calculation series to determine weights for aggregated
forecasts.

Effective mm/dd/yy
Page 21 of 27
338445155.doc
Rev 1
Shape Modeling
Shape Modeling
Shape modeling reapplies a demand profile that has a distinct shape over time. Shape modeling
allows you to:
Identify a historic demand curve as a reusable curve
Create another instance of that curve starting at some future date, creating a new activity
Some common uses:
New product Introduction
Manage product life-cycle
Learn from historical behavior not easily represented by a causal factor
For example, the Analytical Engine uses historical shapes to predict the new release shape for a
product launch:

Effective mm/dd/yy
Page 22 of 27
338445155.doc
Rev 1
1.
Find comparable products, and pull shapes.
2.
Tells the Analytical Engine when the launch is.
3.
The Analytical Engine de-seasonalizes and removes outliers.
4.
The Analytical Engine produces a new shape.

Effective mm/dd/yy
Page 23 of 27
338445155.doc
Rev 1
Shape Modeling Example
Shape Modeling
The Analytical Engine uses historical shapes to predict the new release shape.
For example, the Analytical Engine uses historical shapes to predict the new release shape for a
product launch:
1.
Find comparable products that are similar to the one youre launching, and pull the
demand curve shape.
2.
The user designates a series as beginning, ongoing and ending during the life of the
shape.
3.
The Analytical Engine de-seasonalizes and removes outliers.
4.
The Analytical Engine produces a new shape.

Effective mm/dd/yy
Page 24 of 27
338445155.doc
Rev 1
Advanced Analytics and Nodal Tuning
Advanced Analytics and Nodal Tuning

You can make these adjustments in the Analytics window:
Fine-tune which engine models to use on this node (this combination).
Specify how many data points are needed to use each model.
To access the Analytics window, choose Open Analytics from the Collaborator Workbenchs File
menu.

Effective mm/dd/yy
Page 25 of 27
338445155.doc
Rev 1
Summary
Summary

Effective mm/dd/yy
Page 26 of 27
338445155.doc
Rev 1
Practice 10 Overview: Creating a Forecast Tree
Practice 10 Overview: Creating a Forecast Tree

Refer to Creating a Forecast Tree [LAB404CY]

Effective mm/dd/yy
Page 27 of 27
338445155.doc
Rev 1

Analytics and Statistical Forecasting

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Analytics and Statistical Forecasting

Uploaded by

Copyright:

Available Formats

Analytics and Statistical

Copyright Oracle, 2007. All rights reserved.