CSSM Procedure

PROC CSSM Statement

PROC CSSM <options>;

The PROC CSSM statement is required. You can specify the following options in the PROC CSSM statement:

BREAKPEAKS

prints an alternate form of the break summary tables when the CHECKBREAK option is used in the STATE or TREND statement or when the MAXSHOCK option is used in the OUTPUT statement. In this alternate form, the summary tables report the significant peaks of the shock statistics curves; see Example 14.8: Diagnostic Plots and Structural Break Analysis for examples of these curves.

DATA=libref.data-table

names the input data table for PROC CSSM to use. libref.data-table is a two-level name, where

libref: refers to a collection of information that is defined in the LIBNAME statement and includes the library, which includes a path to the data, and a session identifier, which defaults to the active session but which can be explicitly defined in the LIBNAME statement. For more information about libref, see the section Using CAS Sessions and CAS Engine Librefs.
data-table: specifies the name of the input data table.

LIKE=DIFFUSE | MARGINAL

specifies the type of likelihood to use for parameter estimation. You can specify the following values:

DIFFUSE: specifies diffuse likelihood.
MARGINAL: specifies marginal likelihood.

By default, LIKE=DIFFUSE. For more information about different likelihood types, see the section Likelihood Computation and Model-Fitting Phase.

NOPRINT

turns off all the printing and plotting for the procedure. Any subsequent print options are ignored.

OPTIMIZER(<TECHNIQUE=technique> <HESSTYPE=Hessian-type> <MAXITER=integer>)

specifies options that are associated with the optimizer used in the maximum likelihood parameter estimation. The default settings of the optimization process are adequate for most problems. However, in some cases it might be useful to change the optimization technique, the Hessian type, or the maximum number of iterations. You can specify one of the following techniques:

ACTIVESET: uses the active-set method.
DBLDOG: uses the double-dogleg method.
INTERIORPOINT: uses the primal-dual interior point method.
IPDIRECT: uses the primal-dual interior point augmented Lagrangian method.
NEWRAP: uses the Newton-Raphson method.
QUANEW: uses the (dual) quasi-Newton method.
TRUREG: uses the trust region method.

By default, TECHNIQUE=TRUREG. If the technique is ACTIVESET, INTERIORPOINT, or IPDIRECT, you can specify one of the following Hessian-types:

BFGS: uses the quasi-Newton Broyden-Fletcher-Goldfarb-Shanno (BFGS) Hessian approximation.
FULL: uses the full Hessian.
SR1: uses the dense quasi-Newton symmetric rank 1 (SR1) Hessian approximation.

By default, HESSTYPE=FULL. The ACTIVESET, INTERIORPOINT, and IPDIRECT techniques and the Hessian types are documented in Chapter 10, The Nonlinear Programming Solver (SAS/OR User's Guide: Mathematical Programming). The remaining techniques are documented in Chapter 6, Nonlinear Optimization Methods (SAS/ETS User's Guide). You can alter the maximum number of iterations in the nonlinear optimization search by specifying a nonnegative integer as the MAXITER= value.

PLOTS <(global-plot-options)> = plot-request <(options)> PLOTS<(global-plot-options)> = ( plot-request <(options)> <…plot-request <(options)> > )

controls the plots produced with ODS Graphics. When you specify only one plot-request, you can omit the parentheses around it. Here are some examples:

plots=none
plots=all
plots=residual
plots=residual(normal)
plots=(maxshock residual(normal))
plots(unpack)=residual

If you do not specify any specific plot-request, then by default PROC CSSM produces the plot of standardized residuals against time. For general information about ODS Graphics, see Chapter 24, Statistical Graphics Using ODS (SAS/STAT User's Guide).

Global Plot OptionsThe global-plot-options apply to all relevant plots generated by the CSSM procedure. The following global-plot-option is supported:

UNPACK: displays each graph separately. (By default, some graphs can appear together in a single panel.)

Specific Plot OptionsThe following list describes the specific plot-requests and their options:

ALL

produces all plots appropriate for the particular analysis.

AO< (prediction-error-plot-options)>

produces the prediction error plots—one for each response variable. You can specify the following prediction-error-plot-options:

NORMAL

produces a summary panel of the prediction error diagnostics, which consist of the following:

histogram of prediction errors
normal quantile plot of prediction errors

STD

produces a scatter plot of standardized prediction errors against time.

MAXSHOCK

produces a scatter plot of maximal state shock statistics against time.

NONE

suppresses all plots.

RESIDUAL <(residual-plot-options)>

produces the residuals plots—one for each response variable. You can specify the following residual-plot-options:

NORMAL

produces a summary panel of the residual diagnostics, which consist of the following:

histogram of residuals
normal quantile plot of residuals

STD

produces a scatter plot of standardized residuals against time.

For more information about the precise meaning of the terms maximal state shock statistics and prediction errors, see the section Delete-One Cross Validation and Structural Breaks.

PRINTALL

turns on all the printing options for the procedure. All subsequent NOPRINT options in the procedure are ignored.

STATEINFO

prints two tables that provide information about the composition of the state vector in terms of the components specified in the model. One table describes the composition of state , and the other table describes the diffuse vector and the regressors, which are part of the initial condition specification . For more information about the state space model notation, see the section State Space Model and Notation.

ZSPARSE

enables the exploitation of the sparsity of the matrices in the observation equation during the modeling calculations (see the section State Space Model and Notation for further information). The use of this option can improve the computational efficiency of models that have a large state dimension and sparse matrices—that is, many of their elements are zero. You should use the ZSPARSE option only when the state dimension is sufficiently large (at least 30) and a good percentage (at least 50%) of entries are zero; otherwise, the computational efficiency can in fact degrade. For example, the illustration that is discussed in the section Getting Started: CSSM Procedure is a good candidate for the use of the ZSPARSE option:

 proc cssm data=mylib.Cigar plots=residual zsparse;

Last updated: November 24, 2025