Previous Topic: 7.6.1.3.1 Model Analysis ReportNext Topic: 7.6.2 Performance Group/Period Workload Tracking


7.6.1.3.2 Residual Analysis Report

Figure 7-18 shows a sample Residual Analysis Report for the
channel I/O study.  The regression model that was developed
using the smoothed data has a maximum of approximately 10 to
12% difference from the actual historical observations.  You
must expect this magnitude of residuals, while non-trivial,
from a model that uses smoothing to eliminate apparent random
variations from the historical data series.

When viewing a Residual Analysis Report, look for obvious
patterns in the unexplained data.  For example, in Figure
7-18, you can see large positive residuals in the oldest and
newest historical observations and large negative residuals
in the middle observations.  This pattern indicates that the
data forms an arc around the regression line.  The data is
above the line initially, slides down below the line in the
winter months, then rises again in April.

If a pattern is identified in the residuals, you should
reconsider your choice of model.  In this example, a
quadratic model may provide more predictive power.  When more
data is available, try the quadratic model and compare it to
the linear model by contrasting the r-squared, F, and p
values, as illustrated in Sections 7.4.2.2 through 7.4.2.4,
to determine which model is better for a limited number of
data points.

A quadratic model may fit the data quite well but may fail to
provide useful predictions for future elements, especially if
relatively few historical observations are available.
Therefore, it is not recommended that you use quadratic
models when less than 30 historical observations are
available.

CA MICS Capacity Planner UNIVARIATE MODELING RESIDUAL ANALYSIS UNIVARIATE FORECAST OF: CHAN5CNT BASED ON TIME ELEMENTS: LINEAR TRANSFORMED* PREDICTED RESIDUAL -------------PLOT OF RESIDUALS------------- CONFIDENCE DATE CHAN5CNT CHAN5CNT CHAN5CNT CHAN5CNT -195E4 0.0 +195E4 LIMITS +/- ------- ---------- ------------ ---------- ---------- ------------------------------------------- ---------- 30SEP97 11882418 . 10324734 1557683.74 | 0+++++++++++++++ | 1624240.54 30OCT97 12860633 12371525.50 10912672 1947961.05 | 0++++++++++++++++++++| 1416789.90 30NOV97 11743581 12302107.00 11500610 242971.35 | 0++ | 1221119.32 31DEC97 12510012 12126796.50 12088547 421464.66 | 0++++ | 1043874.36 29JAN98 12071368 12290690.00 12676485 -605117.03 | ------0 | 896056.61 26FEB98 12076524 12073946.00 13264423 -1187899 | ------------0 | 794269.29 31MAR98 12186976 12131750.00 13852360 -1665384 | -----------------0 | 757306.06 30APR96 14355185 13271080.50 14440298 -85113.10 | 0 | 794269.29 28MAY98 15011392 14683288.50 15028236 -16843.79 | 0 | 896056.61 30JUN98 17176448 16093920.00 15616173 1560274.52 | 0++++++++++++++++ | 1043874.36 30JUL98 17421158 17298803.00 16204111 1217046.83 | 0++++++++++++ | 1221119.32 31AUG98 18042957 17732057.50 16792049 1250908.14 | 0++++++++++++ | 1416789.90 31SEP98 . . 17379987 . | 0 | 1624240.54 31OCT98 . . 17967924 . | 0 | 1839490.02 31NOV98 . . 18555862 . | 0 | 2060095.20 31DEC98 . . 19143800 . | 0 | 2284505.07 31JAN99 . . 19731737 . | 0 | 2511700.04 31FEB99 . . 20319675 . | 0 | 2740987.65 31MAR99 . . 20907613 . | 0 | 2971883.57 31APR99 . . 21495550 . | 0 | 3204040.12 * - DATA SMOOTHED USING A GEOMETRIC MOVING AVERAGE. ALPHA = 0.50


 Figure 7-18.  Extended Example Residual Analysis Report