Forecast Decomposition

Forecast decomposition provides insights on how a forecast that's based on Bayesian machine learning was generated for your demand or demand and supply plan. It details various components that come together to form a forecast.

These are the two types of forecast decomposition:

• Method forecast decomposition

• Causal forecast decomposition

When you select the Include details of forecast methods and Include details of causal factors check boxes on the Parameters tab in the Run Plan dialog box, the duration of the plan run increases significantly. For this reason, decompose the forecast by forecasting methods and causal factors preferably when you simulate the demand from a table for a subset of the plan that's based on the table selections. By doing so, you can perform a focused analysis of the forecasting methods and causal factors. Include the details of the forecasting methods and causal factors only when you want to analyze the results and not every time you run the plan.

Method Forecast Decomposition

Use method forecast decomposition to see results for individual forecasting methods.

To enable method forecast decomposition, select the Include details of forecast methods check box on the Parameters tab in the Run Plan dialog box. After you run the plan, you can view the combined forecast with each individual forecasting method's result.

You can then determine the forecasting methods that meet your expectations and disable forecasting methods that you don't find appropriate. In the same view, you can see the weights assigned to each forecasting method and understand how the combined forecast was generated.

You can use the Demand Forecast Method dimension in tables and graphs to analyze forecasts by forecasting methods.

Note: Method forecast decomposition isn't supported for the Croston for Intermittent (F), Multiplicative Monte Carlo Intermittent (K), Regression for Intermittent (J), Naive (N), Moving Average Naive (O), and Holt Naive (T) forecasting methods. While there is method decomposition output for these forecasting methods, the values should be ignored because they are approximations and won't add up to the forecasted values for items that are forecast with these forecasting methods.

Causal Forecast Decomposition

Use causal forecast decomposition to see the results for individual decomposition groups. This capability breaks down the forecast into subcomponents that together add up to the total forecast.

To enable causal forecast decomposition, select the Include details of causal factors check box on the Parameters tab in the Run Plan dialog box.

You can use the Decomposition Group dimension in tables and graphs to analyze forecasts by decomposition groups.

Note: Causal decomposition isn't supported for the Causal Winters (B), Croston for Intermittent (F), Multiplicative Monte Carlo Intermittent (K), Regression for Intermittent (J), Naive (N), Moving Average Naive (O), and Holt Naive (T) forecasting methods. No decomposed values are generated for any items that are forecast with these forecasting methods.