Change Points Considered by Multiple Linear Regression and Auto-ARIMAX/SARIMAX
You can now set the multiple linear regression and auto-ARIMAX/SARIMAX algorithms in the Manage Forecast Models app to consider change points that were identified in the time series by the last forecast automation job. When you do so, the system generates independent variables for each change point. These variables are added to the input data along with the values of the slope dummy variable to help the system produce a more accurate forecast.
Technical Details
| Types |
New |
| Available as of | 2005 |
Additional Details
Connected to this change, the Slope Dummy system-generated feature is also available now for the auto-ARIMAX/SARIMAX algorithm. It is a dummy independent variable whose values are multiplied with a regression coefficient for each period. As a result, the algorithm can predict trend values for the future and calculate a forecast that takes the trend into consideration.
The Consider Change Points option is enabled by default for both algorithms. Until you change this, the Slope Dummy option is automatically enabled as well because all three variables are needed for the calculations. You can select Slope Dummy by itself but not the Consider Change Points option.
Effects on Configuration
You can use the new NUM_OF_CHANGE_POINTS_CONSIDERED global configuration parameter to specify how many of the most recent change points should be considered by the multiple linear regression and auto-ARIMAX/SARIMAX algorithms. This can prevent the system from creating too many key figures.