CSSM Procedure
Initial Score-Store Creation
After you fit a state space model by using a sufficiently long series of data and you determine that the model is satisfactory, it is ready to be used for scoring. The first step in this process is the creation of a score store. You can use the SCORE statement to create the initial score store, such as mylib.inscore, by fitting the model to a suitable series of input data 
(score out=mylib.inscore;). From this stage forward, the fitted model is assumed to hold for all future time points (
). In particular, the estimated parameter vector, 
, is treated as the true and known parameter vector in all subsequent steps. The score store mylib.inscore stores the following information:
the model specification and the estimated parameter vector,
. As part of the model specification, the structural break detection settings in the STATE and TREND statements are also stored. Similarly, the settings of the BREAK and AO options in the OUTPUT statement are also saved. If necessary, the options in the OUTPUT statement can be modified in later scoring calls. 
, 
, 
, 
, and their covariances, which denote the full-sample (smoothed) estimates of these quantities on the basis of the data (for more information, see the section Smoothing Phase). If the score store is created using OUT(k)= syntax (
score out(k)=mylib.inscore;), then
, 
, 
, 
, and their covariances are stored; these are full-sample estimates of these quantities based on the data 
. In this case, the data rows in 
are also stored.
Because of the Markovian property of state space models, the information that the score store contains is sufficient for you to forecast subsequent observations of the process, without having access to its past history.