A coherent forecast is the optimum starting point to make well-informed business decisions.
By reconciling your forecasts using hierarchical forecasting, you get a single source of truth. One that delivers highly accurate results on all levels of aggregation.
.gif)
We've built an econometric model (VAR) to identify leading indicators. Vector autoregression is a workhorse model in macroeconomics that defines each indicator as a function of other indicators. This way, instead of treating each indicator’s impact separately, the model captures interactions between them and their influence on your sales.
By using a lasso penalty through cross-validation, we ensure that only the relevant indicators are represented, delivering the most accurate results.
“It's been in our pipeline!
To be able to build multi-level, separate models
and aggregate them is a great way to help us achieve coherent results.”
A manufacturing company forecasting sales per region and total sales
Active in three regions, the EU, the US and ROW, they are interested in predicting sales per region, and also the total sales. However, the two traditional approaches typically used have both their advantages and disadvantages.
Misaligned decision making due to incoherent forecast results
Pitfalls with bottom-up forecasting
Bottom-up forecasting multiplies the error of every forecast, creating a top-level forecast that is very uncertain on higher levels of aggregation
Pitfalls with top-down forecasting
Top-down forecasting is unable to derive which products the forecasted demand stems from. This makes it hard to know which products to produce.
Pitfalls with
bottom-up forecasting
Each region is predicted individually, and for each time period in the forecast, the sum of all regions forms the total sales. This means that three overly optimistic regional forecasts will add up to a total forecast which will have aggregated the errors of the regional ones. The worldwide sales will usually be more stable, and thus easier to make an accurate forecast of. This advantage is lost when using the bottom-up approach.
Bottom-up forecasting multiplies the error of every forecast, creating a top-level forecast that is very uncertain on higher levels of aggregation
Pitfalls with
top-down forecasting
Here, the worldwide forecast would be created, and the regional predictions are calculated using a fixed percentage of the total.
This does not take into account the fact that the regional sales may move in different directions, sales in the US may be on an upward trend whereas ROW sales are declining, something the top-down method will not reflect.
Top-down forecasting is unable to derive which products the forecasted demand stems from. This makes it hard to know which products to produce.
Having forecasts that are contradicting in different levels of aggregation, it’s a source of confusion on how to plan ahead and holds the organization back making well-informed decisions.
With a coherent forecast, different functions will share the same view about the future. This will create a stronger foundation for alignment across the different levels of the organization.
By using optimal forecast reconciliation, forecasting every level in the hierarchy is possible.
This means that we will use the accuracy level of the top-level forecast to minimize the uncertainty seen in the lower-level forecasts.
Having forecasts that are contradicting in different levels of aggregation, it’s a source of confusion on how to plan ahead and holds the organization back making well-informed decisions.
With a coherent forecast, different functions will share the same view about the future. This will create a stronger foundation for alignment across the different levels of the organization.
.gif)
By using optimal forecast reconciliation, forecasting every level in the hierarchy is possible.
This means that we will use the accuracy level of the top-level forecast to minimize the uncertainty seen in the lower-level forecasts.
.gif)
What-if series
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Suspendisse varius enim in eros elementum tristique. Duis cursus, mi quis viverra ornare, eros dolor interdum nulla, ut commodo diam libero vitae erat. Aenean faucibus nibh et justo cursus id rutrum lorem imperdiet. Nunc ut sem vitae risus tristique posuere.
.png)
The modern approach is to use something called optimal forecast reconciliation.
It works by forecasting every possible level in the hierarchy, in this example, it would be four forecasts, one for each sales region and a worldwide total.
These forecasts are then adjusted using a method which weights them based on their historical accuracy, a step called reconciliation.
This means that an accurate worldwide forecast will cause less precise regional forecasts to be adjusted more than the other way around. This means that forecasts at every level of the hierarchy will benefit from the reconciliation in terms of accuracy. At the same time, the forecasts are guaranteed to respect the hierarchy, meaning that predicted regional sales will add up to predicted total sales.
The modern approach is to use something called optimal forecast reconciliation.
It works by forecasting every possible level in the hierarchy, in this example, it would be four forecasts, one for each sales region and a worldwide total.
These forecasts are then adjusted using a method which weights them based on their historical accuracy, a step called reconciliation.
This means that an accurate worldwide forecast will cause less precise regional forecasts to be adjusted more than the other way around. This means that forecasts at every level of the hierarchy will benefit from the reconciliation in terms of accuracy. At the same time, the forecasts are guaranteed to respect the hierarchy, meaning that predicted regional sales will add up to predicted total sales.
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Suspendisse tincidunt sagittis eros. Quisque quis euismod lorem. Etiam sodales ac felis id interdum.
Download PDF
Your reconciled forecast is only as good as the results of your individual forecasts. Having independent forecasts is an advantage - at each node or level, forecasts can be produced separately, based on different information.
With the ability to update the individual or sub-category forecasts, you will be able to identify and select the most relevant leading indicators and apply multivariate models.
When these forecasts are reconciled, this will improve the overall forecast accuracy of the final adjusted forecast.
.gif)
Article
One step closer to a unified view of demand with hierarchical forecasting
Imagine having a single source of (data) truth with which you can make better decisions.
.gif)
.png)
We are an automated forecasting software that is optimized to identify your relevant leading indicators and generate accurate forecasts using advanced statistical models. Our mission is to help you achieve the highest forecast accuracy and detect market trend shifts months ahead.
© 2024 Indicio Technologies All Rights Reserved