9. MULTIVARIATE REGRESSION FORECASTING › 9.2 Usage Guidelines › 9.2.4 Forecasting Future Values from the Model

9.2.4 Forecasting Future Values from the Model

The final step in developing the multivariate regression
model is to use the model to predict future values of the
dependent resource variable.  This is easily accomplished
using Multivariate Regression Forecasting.

To predict future values for the dependent resource variable,
you must assume that forecast values for all independent
resource variables used in the model are stored in the
forecast file, which corresponds to the resource file
containing these independent resource variables.  You can
supply these predicted values for the independent resource
variables by using any of the forecasting methods available
with the CA MICS Capacity Planning Component: Univariate
Model Forecasting, Business Element Forecasting, or by using
a different model generated by Multivariate Regression
Forecasting.

It is possible to keep multiple forecasts for different time
periods in a forecast file.  The future data observations of
the independent variables must come from a forecast period
that corresponds to the period for which you are forecasting
the dependent variable via Multivariate Regression
Forecasting.  For example, assume that a resource file named
ABC contains data representing 24 months of resource use
extracted from the CA MICS database.  You intend to use
variables from this file as independent variables for
Multivariate Regression Forecasting.  You want to forecast
ahead 12 months.  You have another resource file named XYZ
containing data for the same time period.  You intend to use
one of its variables as the dependent variable in this
forecast process:

    ABC file:  historical data for Jan97-Dec98
    XYZ file:  historical data for Jan97-Dec98

The first step is to use some forecasting process, perhaps
Univariate Model Forecasting, to produce a forecast for
Jan99-Dec99 for the independent variables in the ABC file and
save this forecast in its corresponding ABC forecast file:

    ABC forecast file: forecast data for Jan99-Dec99

Next, use Multivariate Regression Forecasting to produce a
forecast of the dependent variable in the XYZ historical
file, based on your forecasts of the independent variables in
the ABC file.  You might direct Multivariate Regression
Forecasting to use all the data from the ABC file and the XYZ
file to calculate the parameters for the model.  These
parameters and the data from the ABC forecast file are used
to produce a forecast for the dependent variable, optionally
saving it in an XYZ forecast file:

    XYZ forecast file: forecast data for Jan99-Dec99

Make one test upon the data in the ABC forecast file before
using it to produce the forecast for the XYZ file.  Any
forecast file can contain multiple forecasts.  Thus, you
might also have saved a forecast in the ABC file several
months ago, perhaps based only on actual historical data
through July, 1998.  Each forecast observation is stamped
with the year and month or week of the latest historical data
used to produce it.  Multivariate Regression Forecasting
guards against using any forecast data as input except data
which has a forecast stamp equal to the latest historical
data used during the multivariate regression analysis.  This
ensures that you do not use forecast data which does not
correspond to the historical data used in the multivariate
regression analysis.