I am performing Hierarchical Forecasting where I have set the reconciliation level at somewhere mid of the Hierarchy. So, while dis-aggregation for top down approach, which method should be better to use: DIFFERENCE or PROPORTION?
Thanks in advance! Looking forward to your response.
The reconciliation level is essentially a reference level over which all other levels are adjusted to match. Ideally you want to find a level that optimizes your lowest level forecasts. However, typically, when you select a different level some series will improve some other will get worse. There is no single recipe to select the reconciliation level. It all depends on your data, like Yue said. It's a bit of an art mixed with business knowledge. Typically you want to select a level for which time series models would fit the data. For example, you do not want many intermittent series. Often the lowest levels are too intermittent and "noisy" to be modeled with time series models.
I suggest you export your data to VA (Visual Analytics) and explore it hierarchically to get a feeling of how each level looks like. Unfortunately, currently VF does not allow the explorations of different level of a hierarchy. It is planned for future releases.
Similarly, the choice between proportion and difference should be driven by performance of the forecasts at the level of your interest. In practice, most people use proportions because it is more intuitive and easier to explain. Note that you can use proportions even with mixed-sign data.
Registration is open! SAS is returning to Vegas for an AI and analytics experience like no other! Whether you're an executive, manager, end user or SAS partner, SAS Innovate is designed for everyone on your team. Register for just $495 by 12/31/2023.
If you are interested in speaking, there is still time to submit a session idea. More details are posted on the website.