A MODEL statement specifies an observation equation that describes a response variable as a sum of regression effects and components that are defined in the program. The response variable must be a numeric variable from the input data set. The variables used in the right-hand side of the model expression can be numeric variables from the input data set, numeric variables defined by using programming statements, or names of components that are specified in the COMPONENT, DEPLAG, TREND, or IRREGULAR statements.
For a multivariate model, a separate MODEL statement is needed for each of the response variables. In this case, the observation errors, which are specified in an IRREGULAR statement, must be different in each MODEL statement.
The components that are specified in a given MODEL statement must correspond to distinct state subsections. This requirement is imposed only to simplify the overall readability of the program and does not limit the type of models that can be specified; if two components on the right-hand side of a MODEL statement share the same state subsection, a new component that combines the effect of these two components can always be defined.
You can specify the following options in the MODEL statement; they must be separated from the list of terms in the right-hand side of the model equation by a slash (/):
-
AGGREGATE(START=startFlag)
SUM(START=startFlag)
produces a table of full-sample predictions of the temporally aggregated values of the response variable that is specified in the MODEL statement. The variable that you specify in the START= option, startFlag, must be a zero-one variable that flags the start of an aggregation interval—equal to 1 at the start of an interval and 0 otherwise. For example, you can use this option to obtain the forecasts of weekly (or monthly) totals from a daily series. In this case, the value of startFlag is 1 at the start of the week (or month) and 0 otherwise. For more information, see the section Temporal Aggregation and Temporal Distribution. This option is valid only if the data form a time series (either univariate or multivariate). If you use the AGGREGATE option in a MODEL statement, you cannot use the DISTRIBUTE option in the same statement or in another MODEL statement.
-
DISTRIBUTE(START=startFlag)
indicates that the response variable that is specified in the MODEL statement is a temporally aggregated version of an unobserved variable. The variable that you specify in the START= option, startFlag, must be a zero-one variable that flags the start of an aggregation interval—equal to 1 at the start of an interval and 0 otherwise. This option can be used only when the data form a time series (either univariate or multivariate) and when the overall model specification does not contain terms that involve lagged response variables (that is, the model specification does not involve the use of DEPLAG statements). If you use the DISTRIBUTE option in a MODEL statement, you cannot use the AGGREGATE option in the same statement or in another MODEL statement. For more information, see the section Temporal Aggregation and Temporal Distribution.
-
PRINT=FILTER | SMOOTH
PRINT=(<FILTER> <SMOOTH>)
requests printing of the filtered or smoothed estimate of the specified response variable. The filtered estimate is produced during the filtering phase, and the smoothed estimate is produced by the smoothing phase of the Kalman filter and smoother algorithm. The filtered estimate is also called the one-step-ahead forecast of the response variable. The smoothed estimate corresponds to the full-sample prediction of the response variable. Since the full-sample prediction of a nonmissing response value is that value itself, full-sample predictions are printed only for the missing response values.