7. UNIVARIATE MODEL FORECASTING › 7.1 Functional Description

7.1 Functional Description

Univariate Model Forecasting allows you to develop forecasts
based on regression using linear, quadratic, or cubic models
of data elements contained in the Capacity Planning database
files.  These files can be either resource element files
based on data from one or more CA MICS files, meta files or
business element files based on data obtained from any other
source (see Chapter 3).

These forecasts are based on "least squares" regression
models.  Univariate modeling uses the SAS REG procedure to
fit a line of least squares to the historical data.  The
model developed for the historical data is then used to
estimate future observations for the number of periods which
you specify.  Optionally, it can store the forecast values in
special Capacity Planning forecast files.

Univariate Model Forecasting offers several additional
features to assist you in fitting a least squares line to the
data.  Computer system data often exhibits a high degree of
variability, and is sometimes a non-linear function.  Both
high variability and functional form make fitting a least
squares line very difficult.

To overcome problems with variability and functional form,
Univariate Model Forecasting offers you smoothing functions
based on geometric moving averages, logarithmic
transformation, and user-defined transformations.  You can
use geometric moving averages to dampen a series which
exhibits a high degree of variability (both positive and
negative) around a central trend.  Logarithmic
transformations allow you to analyze exponential
relationships using linear models.  You can also code a user
exit to define any other type of transformation which may
apply to your installation.  An example of a user-defined
transformation is the normalization of CPU time to the speed
of a base CPU.