Overview (FIT command)
FIT
displays
a variety of descriptive statistics computed from the residual series
as an aid in evaluating the goodness of fit of one or more models.
Options
Statistical Output. You can produce statistics for a
particular residual series by specifying the names of the series after FIT
. You can also obtain percent error statistics
for specified residual series by specifying observed series on the OBS
subcommand.
Degrees of Freedom. You
can specify the degrees of freedom for the residual series using the DFE
or DFH
subcommands.
Basic Specification
The basic specification
is simply the command keyword FIT
. All other specifications are optional.
- By
default,
FIT
calculates the mean error, mean percent error, mean absolute error, mean absolute percent error, sum of squared errors, mean square error, root mean square error, and the Durbin-Watson statistic for the last ERR_n (residual) series generated and the corresponding observed series in the active dataset. - If neither residual nor observed series are specified, percent error statistics for the default residual and observed series are included.
Syntax Rules
- If
OBS
is specified, theERRORS
subcommand naming the residual series is required.
Operations
- Observed series and degrees of freedom are matched with residual series according to the order in which they are specified.
- If residual series are explicitly specified but observed series are not, percent error statistics are not included in the output. If neither residual nor observed series are specified, percent error statistics for the default residual and observed series are included.
- If subcommand
DFH
is specified,FIT
calculates theDFE
(error degrees of freedom) by subtracting theDFH
(hypothesis degrees of freedom) from the number of valid cases in the series. - If a
PREDICT
period (validation period) starts before the end of the observed series, statistics are reported separately for theUSE
period (historical period) and thePREDICT
period.
Limitations
- There is no limit on the number of residual series specified. However, the number of observed series must equal the number of residual series.