Calling the ATT forecast engine with Multiple Regression algorithm

Multiple Regression is used to predict the value of a variable based on two or more variables. The algorithm uses the defined independent variables to generate a regression analysis against the dependent variable (history and forecast), for the forecast engine.

The required conditions to call the ATT forecast engine and generate forecasts for the selected item and location, when the Default Algorithm = Multiple Regression:

• The history and mask data of each item or location is retrieved and transferred to the forecast engine based on the standard process.
• The measures mapped to these exceptions for each item and location that are transferred to the engine, based on the other algorithms, must be cleared:
  • Invalid Stat
  • Obsolescence
  • Outliers
  • Short History
  • Step Change
  • Tracking Signal
• Independent variables that are mapped to the specified forecast engine act as additional inputs. Therefore, the appropriate scenario values that are transferred to the engine (for each mapped independent variable measure) are retrieved.
  • The values of the historical horizon for all independent variables are transferred. The history length must be the same as the dependent data.
  • The values of the future horizon for all independent variables are transferred. The forecast length must be the same as the forecast horizon.
  Note: Additional parameters are transferred based on the standard process.

Based on forecast engine measure mapping settings, the appropriate results from the forecast engine result set are retrieved and applied to the application's scenario values. The forecast data is returned and processed based on the other algorithms. The engine also returns additional results specific to the Multiple Regression algorithm:

• Beta coefficients, t squared, and t squared P-values corresponding to each of the independent variables are generated. However, due to the intercept one additional element (regressors) is generated. The first value in each result array corresponds to the intercept and must be written to the associated mapped measure, if defined. Other values are written in the order of the transferred independent variables.
• Variance Inflation Factors corresponding to each of the independent variables are generated.
  • The number of factors and independent variables are the same. These factors are returned in the order of the transferred independent variables.
  • The intercept equivalent variance inflation factor is not applicable.
• Additional results specific to the multiple regression algorithm such as R Squared, Adjusted R Squared, F Statistic, and F Statistic P-Value are written to the corresponding mapped measures, if defined.
• A regressionStandardDeviation is also returned. This is written to the mapped 'standard deviation' measure, if defined.
• All results are static (time-independent) and are written to PCONST constant.

All the additional results that are based on the mapping to the engine, excluding those listed in this topic, are processed based on the other algorithms.