9. MULTIVARIATE REGRESSION FORECASTING › 9.1 Functional Description

9.1 Functional Description

Multivariate Regression Forecasting produces regression
models that relate different resource consumption data
elements or business-related data elements from one or more
capacity planning database files.  These forecasts are based
on stepwise multiple regression models using any of the
resource element or business element files produced by the
CA MICS Capacity Planner. You can use the same or different
files as the source for dependent and independent variables
in the model.

The process implements the multivariate forecasting
methodology in a two-step procedure.  The first step uses the
SAS REG procedure with the stepwise option to investigate the
relationships among the historical observations of the
variables.  This step determines which of several historical
variables have the most predictive power on the variable to
be forecast.  In the second step, the SAS REG procedure
builds a predictive model using the variables selected in the
previous step, estimates future values of the forecast
variable, and generates confidence intervals on the
forecasts.

The operation and usage of Multivariate Regression
Forecasting is nearly identical to that of the Business
Element Forecasting, described in Chapter 10. In Business
Element Forecasting, the control screens ensure you are
forecasting a resource consumption variable based on its
possible relationship to one or more business-based
variables.

Multivariate Regression Forecasting allows you to investigate
relationships among variables from any two capacity planning
database files, or among variables from a single file.  Any
of these files can be resource element files or business
element files.  Thus, although Multivariate Regression
Forecasting and Business Element Forecasting use the same
statistical techniques, the first type of forecasting is
potentially much more general.

To illustrate the use of Multivariate Regression Forecasting,
we have developed a multivariate regression model that
estimates long-range capture ratios.  This model relates the
historical use of specified processor workload categories to
overall processor use.  A complete description and
interpretation of the capture ratio model is presented in
Section 9.6.1.