CSSM Procedure

Suppose that a score-store, such as mylib.inscore, contains context information based on the data up to time T (), and suppose you are interested in predicting and/or interpolating the response sequence for the next p time points. You can easily accomplish this in PROC CSSM by using a suitable data packet (or when multiple observations can share the same time index) as the input data table and by using mylib.inscore as the input score store in the SCORE statement (score in=mylib.inscore;). This is called the scoring of data in . Because the missing values in the response variables are an indication to PROC CSSM to compute model-based predictions/interpolations, you must set the values of to missing in appropriate rows of (of course, you must provide appropriate nonmissing values for the predictor variables and ).

In what-if analysis, you are interested in predicting the future course of response values for different settings of the predictors and . You can achieve this by repeatedly scoring for different settings of and , while keeping the input score-store the same. See Example 14.19: Scenario Analysis Using Scoring for an illustration of this type of what-if analysis.

In process control, the goal is to find suitable predictor settings so that the future response values are close to some desirable levels. In an informal, quick-and-dirty approach, you can do this by repeatedly scoring with different trial values of predictors. In a formal approach, you can use the scoring functionality to define a suitable objective function, which you can optimize by using sophisticated optimization methods to find the optimal predictor settings.