HP 3000 Manuals HP 3000 Manuals
Understanding the Stats Report
You can use the Stats command to generate a textual report on any
displayed forecast. This section explains how to read this report in
order to validate your forecast.
The Information Block
The first part of the Stats report is the information block, which
summarizes the selections made on the Forecasts command and RXForecast
Options commands dialog boxes.
Regression Analysis for Trapper. Metric = Sess. CPU
DATE : 01/29/90 TIME : 15:03:25 METHOD is Linear.
Seasonality: DAY WEEK MONTH QUARTER Auto Season ON.
Time Window Start : 08:00 Stop : 17:00
Date Window Start : 05/02/88 Stop : 06/30/88 Validate : 08/25/88
Summarization : DAY Ignore Weekends : OFF.
Forecast Calculations
The second part of the Stats report summarizes the calculations that
produce the forecast.
Model : FC = INT + SLOPE * TIME + DOW + WOM + MOQ + QOY
FC=Forecast INT=Intercept TIME=# of time periods since start of forecast.
DOW=Day of Week Seasonality WOM=Week of Month Seasonality
MOQ=Month of Quarter Seasonality QOY=Quarter of Year Seasonality
For more detailed information on the calculations that produce the
forecasts, see Appendix B .
For an explanation of how to produce forecasts manually using these
calculations, see "Producing Forecasts Manually" .
Statistical Measures
The third section of the Stats report displays the statistical measures
that indicate the validity of the forecast.
Trend line parameters:
Intercept = 4.6717e+000; T-Stat. = 7.7 T-Prob. = 100.0
Slope = 1.2302e-001; T-Stat. = 6.9 T-Prob. = 100.0
MSE = 5.39; Std. Err. = 2.32; R-Squared = 0.80; N = 54; P = 8;
The following table briefly defines each of these measures and how to
read them. More-detailed explanations follow the table.
Table 6-1. Statistical Measures Definitions
-----------------------------------------------------------------------------------------------
- Measure - Definition - How to Read -
-----------------------------------------------------------------------------------------------
| T-Stat. | T-Statistic. Indicates the | A value that is large--either |
| | significance of each trend line | positive or negative--indicates a |
| | parameter in the model. | significant parameter. A value |
| | | greater than 3 or 4 is |
| | | significant. |
-----------------------------------------------------------------------------------------------
| T-Prob. | T-Probability. The probability | A value close to 100 indicates |
| | that if the parameter in question | statistical significance. |
| | is really zero, you would see a | |
| | T-Statistic no larger than that | |
| | actually observed. | |
-----------------------------------------------------------------------------------------------
- MSE (1) - Mean Squared Error. - A value close to 0 is optimal. -
-----------------------------------------------------------------------------------------------
| Std. Err. (1) | Standard Error. The square root | A value close to 0 is optimal. |
| | of MSE. | |
-----------------------------------------------------------------------------------------------
| R-Squared | Coefficient of determination. | All values will be between 0 and |
| | | 1. A value close to 1 indicates a |
| | | very good fit. A value close to 0 |
| | | means a very poor fit. |
-----------------------------------------------------------------------------------------------
- N - Number of data points. - -
-----------------------------------------------------------------------------------------------
- P - Number of parameters in the model. - -
-----------------------------------------------------------------------------------------------
(1) MSE and Std. Err. do not appear on Stats reports generated from
Exponential or S-Curve forecasts.
* T-Statistic.
T-Statistic indicates the significance of each trend line
parameter in the model. A parameter is significant if the
T-Statistic is large, which indicates that the parameter is not
zero.
For example, if a straight line trend without any seasonality is
used, then the two parameters in the model are the intercept and
the slope. A T-Statistic that is greater than 3 or 4, either
positive or negative, indicates a significant parameter.
* T-Probability.
For those users familiar with statistical tests of hypotheses,
T-Probability is:
100*(1-Pvalue)
when the test is as follows:
H0: parameter = 0
HA: parameter != 0
* Mean Squared Error (MSE) and Standard Error.
The Mean Squared Error (MSE) and the square root of the MSE, the
Standard Error (Std. Err.), determine the amount of variability
between the forecast and the actual data. They measure the
variability in the forecasting technique.
A high MSE value indicates a poor forecast, while an MSE near 0
indicates a good forecast. You need experience to determine
whether the MSE should be considered a good or bad indicator for a
particular case. The MSE is valuable for comparing two trending
methods. Everything else being the same, you should choose the
forecast with the smallest MSE.
It is easier to interpret the Standard Error since its units are
the same as the metric that is being forecast. Interpret the
Standard Error as the average amount of deviation between what the
model states and the actual data value for a given date.
Forecasts can typically be expected to deviate from actual data
values by an amount equal to the Standard Error.
* R-Squared.
The R-Squared measure, also referred to as the coefficient of
determination, is a summarized measure of the forecasting method's
validity. That is, it is a measure of the fit of the forecast to
the data.
All values for R-Squared will be between 0 and 1. A value close
to 1 indicates a very good fit for the method, while a value close
to 0 means a very poor fit.
Seasonality Estimates
The last part of the Stats report explains the seasonality parameter
estimates.
Day of Week Seasonality Parameter Estimates:
Mon = 2.2947e+000; T-Stat. = 3.0 T-Prob. = 99.6
Tue = 2.7979e+000; T-Stat. = 3.7 T-Prob. = 99.9
Wed = 2.4862e+000; T-Stat. = 3.3 T-Prob. = 99.8
Thu = 2.5381e+000; T-Stat. = 3.3 T-Prob. = 99.8
Fri = 1.9204e+000; T-Stat. = 2.5 T-Prob. = 98.5
Sat = -5.5722e+000; T-Stat. = -6.9 T-Prob. = 100.0
Sun = -6.4652e+000;
Week of Month Seasonality Parameter Estimates:
Excluded: Not Statistically Significant
Month of Quarter Seasonality Parameter Estimates:
Excluded: Not Statistically Significant
Quarter of Year Seasonality Parameter Estimates: (Calendar Quarter)
Excluded: Not Statistically Significant
Seasonality is statistically significant if the T-Probability is greater
than 90 percent.
NOTE There is no T-Statistic or T-Probability for Sunday in the
preceding example because of the way its seasonality is calculated.
No T-Statistic or T-Probability exist for the last day in
Day-of-Week seasonality, the last week in Week-of-Month
seasonality, or the fourth quarter in Quarter-of-Year seasonality.