Configuring the Forecast Tree

This chapter describes how to configure the forecast tree. In the case of PE mode, it also describes how to configure the influence relationships, and competition among the combinations.

This chapter covers the following topics:

Configuring the Forecast Tree

Caution: The Promotion Optimization engine uses levels in the Oracle Demantra forecast tree, and uses their names rather than their internal identifiers. This means that if you change the name of a level, you must rebuild the forecast tree to make sure that Promotion Optimization can find the level (because the forecast tree is not automatically synchronized with the level definitions).

If you are not using Promotion Optimization, you would need to rebuild the forecast tree only if you remove a level or add a new level that you want to include in the forecast tree.

See also

"Basic Concepts"

"Guidelines for the Forecast Tree"

To configure the forecast tree

  1. Click Engine > Forecast Tree. Or click the Forecast Tree button.

    The Configure Forecast Tree Engine Profiles dialog box appears.

    the picture is described in the document text

  2. Select the engine profile for the forecast tree you want to modify and click OK.

    The Forecast Tree Editor displays lists of all the item and location levels that you have created in the system.

    Note: General levels can also be selected in the forecast tree instead of an item or location level. For example, in the case of service parts planning, the general level "Lowest spf Level" and "SPF Latest Revision" are selected as item levels. For general levels to be available in forecast tree configuration, the engine profile must refer to a general level data table. The engine parameter EngDimDef_ItemOrLoc determines whether the general levels will appear in the item or location dimension of the forecast tree.

    the picture is described in the document text

    You use this dialog box to select the item levels and location levels to include in the forecast tree.

    Note: As you select item and location levels in the following steps, add levels from the lowest level to the highest. Business Modeler automatically adds the highest fictive level to each list.

    You can have different number of elements in these two lists.

  3. Select the item levels to be included in the forecast tree. To do so, use the two lists at the top of the dialog box. Use any of the following techniques:

    • In the left list, double-click a row.

    • Click a row and then click Add.

    • Click Add All to transfer all items.

  4. Select the location levels to be included in the forecast tree. Use the two lists at the bottom of the forecast tree, and use any of the methods described in the previous step.

  5. When you have finished selecting levels, click Save.

  6. Click Next.

    The Forecast Tree Editor displays a dialog box that you use to build the forecast tree itself.

    the picture is described in the document text

    In this dialog box, each row corresponds to a level in the forecast tree. In turn, a level in the forecast tree consists of one item level and one location level.

    Note: As you build the forecast tree, add levels from the lowest level to the highest. Business Modeler automatically adds the HFL, if you do not do so explicitly.

  7. To create a level in the forecast tree, do the following:

    1. Click Add.

    2. In the drop down list in the Item Order column, select an item level.

    3. In the drop down list in the Location Order column, select a location level.

  8. Add more levels to the forecast tree as needed, and then click Save.

  9. Click Exit or click Next.

    If your system includes Promotion Effectiveness, the Forecast Tree Promotion Levels screen appears. This screen displays the forecast levels as created in the previous screen.

    the picture is described in the document text

  10. (PE mode only) On this screen, specify the following:

    • Level to use as the lowest promotional level (LPL). This is the lowest aggregation level the Analytical Engine will consider when evaluating the effects of promotions.

    • Level that defines the influence groups. This is the influence group level (IGL). This indirectly specifies the item groups and location groups.

    • Level that defines the influence ranges. This is the influence range level (IRL).

    • If the system includes modules AFDM, PTP or TPO, an additional screen is available. This screen controls whether the engine simply aggregates data when forecasting at higher levels or whether it groups aggregated data nodes into longer time series.

    • As a default, the Forecast Detail and Forecast Range should be set to the same levels as those in the dialog with title “Forecast Tree (page 2).” For additional information regarding modifications of this screen see Pooled Time Series Below.

    For example, in the row that should corresponds to the influence range level, select Influence Range from the drop down list in Promotion Level Type.

    Note: To establish the LPL and IGL at the same level, select the option Lowest Promotion Level & Influence Group.

  11. Do one of the following:

    • Click Next. Business Modeler next displays the Causal Factors dialog box; see "Configuring Global and Local Causal Factors".

    • Click Exit. You can return later to configure causal factors.

    See also

    "Guidelines for the Forecast Tree"

Pooled Time Series

When the forecast tree calculation encounters a node for which it cannot generate a forecast, it traverses up to the next highest forecast tree node to generate the forecast. The engine then allocates an appropriate value back down, according to a proportional algorithm. For example, if a particular product/store did not have enough historical data, then the forecast engine would aggregate data at a higher level, such as product/region, generate a forecast at this level, then allocate a proportional amount of that back down to the product/store node.

One issue with this approach is that the product/region contains information that has been aggregated (summed), and aggregated data is often smoothed, resulting in less granular information being available. This can cause granular historical behavior as well as causal factor information to be factored out of the forecast. For example, the product/region node may be an aggregate of a number of different stores, and if there is specific historical pattern for that store--say, that the location always has a sales spike at a particular time of year-- then that information would become smoothed over when forecasting at an aggregate level.

Pooling a times series is a way of supplying the engine with more information than what would have been available without summing several nodes together. Instead of aggregating the data at a higher level node, it concatenates, or “pools” the data, allowing it to evaluate all the more granular data points together. The figures below show an example of these different types of data:

Raw Data

the picture is described in the document text

Aggregated Data

the picture is described in the document text

Pooled Data

the picture is described in the document text

In addition to making up for limited sales histories, another key reason for using pooled time series analysis is to increase the chance for utilizing data relating to irregular occurrences, or “events.” Events can be simple actions, such as a change of a price at a store, or more rare events such as natural disasters. Each event may cause an increase or decrease in demand, and the pooled time series calculation is better able to reflect this when evaluating historical demand.

Note: For combinations with sufficient history it is still recommended the combination be forecasted independently and not in a pooled manner. Pooled information may average the behavior across several combinations and be less accurate than focusing on the data of a specific combination.

Configuring the Forecast Tree for Pooled Time Series

When configuring the forecast tree for Pooled Time Series, an additional configuration must be made in the forecast tree. For each forecast tree level, a Forecast Detail and a Forecast Range must be defined. Forecast Detail is the data aggregation to be used during forecast generation. For example, if a region has three stores, when setting detail to level region, the forecast node value will be the aggregation of the three stores.

The Forecast Range defines how Forecast Detail nodes are pooled together to form larger data sets with additional information. A Forecast Range can be set to the same value as the Forecast Detail. In this case, data is aggregated to the Detail node and a forecast is generated for that node independently. If Forecast Range is set higher than Forecast Detail, more than one Detail node will be pooled and concatenated together, and then forecasted together. If the above region detail level is associated with a range level of country, then all aggregated region information will be pooled together and forecasted rather than each region-based node independently.

In the example below, we see that the Item/Site detail level is associated with two range levels. The first forecast tree level (where Range and Detail are set the same: Item/Site -- Item/Site) means that when the forecast calculation is trying to generate a forecast for the Item/Site detail level, it will generate a forecast at each Item/Site node individually. However, if the forecast fails at a specific Item/Site node, then the next forecast tree level will be used. At this next level, Range is greater than Detail (Item/Site – Item/Customer). This tells the forecast calculation to generate a forecast for Item/Site using the all the Item/Sites in the range of Item/Customer.

The following replaces Step 8 in the procedure above.

8. Click Next.

  1. Click Add.

  2. In the drop down list under Detail Levels, select an item and location level.

  3. In the drop down list under Range Levels, select an item and location level.

the picture is described in the document text

Defining Influence and Competition (PE Mode Only)

To describe how the item-location combinations affect each other, you specify the following information:

For the first three tasks, see "Configuring the Forecast Tree". To define the CI and CL, do the following:

  1. For each level you create, Business Modeler creates a row in the group_tables table for each level. Make a note of the level ID of the levels that you want to use as the CI and CL.

  2. Navigate to the Parameters > System Parameters > Engine > Shell parameters. Each value should be a level ID as given in the group_tables table.

    Parameter Purpose
    COMPETITION_ITEM Specify the level whose members are the competitive item groups.
    COMPETITION_LOCATION Specify the level whose members are the competitive location groups.

    The CI and CL should be consistent with the item groups and location groups. Specifically, any lowest level items within a given item group must belong to the same competitive item group. The easiest way to follow this rule is to set the CI equal to an item level that is higher than I and that is within the same hierarchy. A similar rule applies for the locations.

    See also

    "Switching Effects"

    "Guidelines for the Forecast Tree"

Defining the Forecast Tree for Service Parts Planning Supersessions

Service Parts Forecasting supports the superseding of old parts with new parts, know as supersessions in EBS Service Parts Planning. The Forecast Spares Demand engine profile has been defined to use this functionality. In particular, the Forecast Spares Demand engine profile forecasts on the t_ep_spf_data table instead of SALES_DATA. In addition, service parts forecasting also refers to two engine parameters to configure the forecast. They are:

For more information about these engine parameters, see Analytical Engine Parameters.

For information about how the split at supersession is done as part of the proport, see Proport When Forecasting on General Levels.

Specifying Additional Parameters

Use the Business Modeler user interface to set the following additional engine parameters, if needed:

Parameter Purpose
max_fore_level The maximum level on the forecast tree at which a forecast may be produced. Upon failure at this level, the NAIVE model will be used, if enabled.
For PE mode:
  • This level is usually below the IRL.

  • Sometimes the natural top forecast level does not make a good choice of IRL, and a more aggregated level would be better for the IRL. This new level may be too high for forecasting, but it is useful for calculating indirect effects. In such a case, set max_fore_level to the highest level to use for forecasting, and the IRL to the higher level.

min_fore_level Minimum forecast level that the engine will forecast. From that level down, the engine will split the forecast using the precalculated proportions in the mdp_matrix table.
The engine does not necessarily create the forecast at this level. If the results are not good at this level (for a portion of the forecast tree), the Analytical Engine moves to a higher node of the forecast tree, creates a forecast there, and splits down to the minimum forecast level. As before, the engine splits using the precalculated proportions in the mdp_matrix table.
For PE mode, this level must be at or above the LPL.

For information on these parameters, see "Engine Parameters".

The Forecast Tree Editor displays a dialog box that you use to build the forecast tree itself. The Forecast Tree Editor displays lists of all the item and location levels that you have created in the system.