Forecasting in SAP IBP with correlation factors

[former_member645099]

Introduction

Moving holidays like Ramadan, Easter & Chinese New Year can cause planners an increase in manual effort to get the demand right in the right place, at the right time.

Traditional seasonal forecasting models handle the seasonal patterns for fixed holiday periods like Christmas & New Year, however no out-of-box forecasting models are available in any demand planning system for moving holidays.

This blog post intends to deep dive into the functionality within SAP IBP for Demand module to determine what system features are available to handling this solution.

Understanding the mechanism behind the tool

Within SAP IBP for Demand, there is an option to setup statistical forecast to consider correlation factors - these are independent factors which can impact the statistical forecast output. The pre-requirement for using correlation factors within the statistical forecast models are as follows:

• The future correlation factors are known and available


• Should be a relationship between the correlation factors and the historical data


• Should be no interdependency between different correlation factors

There are a number of different integration methods to upload/maintain data in to the correlation factors defined, including; CPI-DS, data integration jobs & loading via excel planning views.

Once the correlation factors have been defined and populated, the correct statistical forecasting method needs to be chosen and the related correlation factor selected. Currently within SAP IBP for Demand, the use of correlation factors within statistical forecast models are limited to:

• Gradient Boasting


• Multiple Linear Regression


• Auto-ARIMAX


• Future release - Integration with R-Studio which will open the door to many more forecast algorithms with correlation factors

Using the correlation factors key figures, patterns for each moving holidays can be incorporated.



Once the statistical forecast has been regenerated, the impact of the correlation factors will be visible against the demand quantity. In the picture below we can see an increase in demand for months Apr 2020 & May 2020. Both of these months have a value for the key figure ‘Holiday Calendar’. If we were to analysis the historical data set, we would determine values in this key figure yield a positive impact on the demand.



Using the show messages functionality within IBP for Demand, you can determine the importance of each correlation factors defined in the forecasting model.



The key benefit of using correlation factors as opposed to seasonal models include:

• Flexibility to map multiple moving holidays


• Can be applied to multiple levels of the product/sales hierarchy


• Using correlation factors requires less data point than conventional seasonal models

For example, within the Gulf States during Ramadan, sales of liquid drinks with high levels of sugar often sky rocket as these types of drinks are a popular choice to be used to open the fasting. At the same time, the same drinks in Western countries can dip, as the summer holiday period starts where consumers are more likely to be on holiday.

Olivehorse Value Add

The use of correlation factors is not limited to mapping out moving holidays. Other case studies for using correlation factors include:

• Using average monthly temperature for beverage sales

Predicting future average temperatures can be tricky especially with periods of unseasonable warm, dry weather in autumn as part of an ‘Indian Summer’ or unusually low temperatures and heavy snowfall from the ‘Beast from the East’. Using the temperature as a correlation factors not only allows for easy ‘What-if’ scenario planning, but also allows changes to the demand plan to be made based on predicated changes in the temperature forecast.

• Key sporting events (England winning the football world cup perhaps?) for predicting purchase of mobile data bundles with the telecommunication industry

In the example below, key sporting events like the Abu Dhabi GP in November, FA Cup Final in May and FIFA World Cup in June/July have been mapped out using a correlation key figure called ‘Sporting Events’.  The impact of the Sporting Events produces a higher demand for mobile data bundles, as consumers are more likely to watch the events on mobile devices and spend more time on social media discussing results from the events.

• 3^rd party services

A number of 3^rd party services are available for API feeds, including weather forecasts, house price indexes and population growth rate. These 3^rd party services can be integrated with IBP in order to update correlation factors based on these 3^rd party services. This allows a hands off approach to updating the correlation factors with the most recent figures.

Conclusion

Forecasting with correlation factors allows the influence of external factors to be recognised by the demand plan. This allows planners to broaden the amount of system functionality available to create a more accurate demand plan.

Within IBP for demand, there is a strong push for forecast automation using machine learning. Enabling correlation factors and integrating the 3^rd party services only reinforces the ways of working with an automation approach in mind.

The original blog post is posted here:

https://blog.olivehorse.com/forecasting-with-correlation-factors