T-Test in MLR Forecasting
MLR forecasting may run a t-test before calculating the actual forecast to determine which MLR independent variables specified in the MLR forecast profile are really relevant, in other words, which variables have an effect on the demand.
Note
The t-test, also known as the t-statistic, indicates whether or not an independent variable is correlated to the dependent variable; that is, it tells you whether or not the independent variable helps to explain the dependent variable and, therefore, whether or not you should leave the independent variable in the model. The t-statistic says nothing about the significance of an explanatory variable's magnitude or impact. In other words, a t-statistic of say 4.6 is no more significant than a 2.4; it only means the independent variables associated with those t-statistics are significant in explaining the variation in the dependent variable. The magnitude of that relationship is measured by the coefficient of the independent variable and its unit of measure. The rule of thumb for a t-statistic for determining whether a coefficient of an independent variable is significantly correlated to the dependent variable at a 95% confidence level is a +/- 2.0. However, empirical testing has proven that a t-statistic of +/- 1.4 or greater is structurally significant at the 90% confidence level. We therefore recommend that you leave explanatory variables with a t-statistic of +/- 1.4 or greater in the model. If some of your independent variables have a t-statistic of less than +/- 1.4, run an ex-post forecast first with and then without them. If the forecast error is lower with the explanatory variables, leave them in. However, if the t-statistic for these independent variables is below +/- 1.4 your structural analysis will no longer be valid.
Prerequisites
If you want the system to consider the result of the t-test for a variable, you must select the Allow T-Test to Disregard
checkbox in the Past and Future for Independent Variables
table in the MLR forecast profile maintenance. In
case you run univariate forecasting in the combined mode, and you have selected the Allow T-Test to Disregard
checkbox, the system also considers if the independent variable has any future values until the end of the forecast horizon. If a variable has no values at all for
the future, the system disregards it.
To use this function, you must activate the business function SCM-APO-FCS, Events and Outliers, Causals, ABC/XYZ Classification
(SCM_APO_FORECASTING_1
). For more information, see SCM-APO-FCS,
Events and Outliers, Causals, ABC/XYZ Classification.
Activities
To use this function, on the SAP Easy Access
screen, choose 
Advanced Planning and Optimization
Demand Planning
Environment
Maintain Forecast
Profiles
. On the MLR Profile
tab page, in the Past and Future for Independent Variables
table, select the Allow T-Test to Disregard
checkbox.
If you have selected the Allow T-Test to Disregard
checkbox, the system proceeds as follows:
It performs preparation steps for MLR forecasting. It reads the history.
It runs MLR calculation for all MLR independent variables specified in the MLR forecast profile.
In case some MLR independent variables specified in the MLR forecast profile have the
Allow T-Test to Disregard
checkbox selected, it continues with step 4. Otherwise, it jumps to step 6.It performs the t-test.
It runs MLR calculation for only those MLR independent variables that the t-test found relevant.
It performs the finishing steps, such as applying phase in and phase out profiles, calculating forecast errors and saving results.