CCDM Procedure
COUNTMODEL Statement
COUNTMODEL distribution-name(parameter-value <(value-option)><,> …parameter-value <(value-option)>);
The COUNTMODEL statement enables you to specify a parametric count model that has fixed parameter values. You must specify this statement if you do not specify the COUNTSTORE= option or the EXTERNALCOUNTS statement. On the other hand, PROC CCDM ignores this statement if you specify either the COUNTSTORE= option or the EXTERNALCOUNTS statement.
You must specify the following:
- distribution-name
identifies your desired distribution. For a list of supported distributions and their acceptable names, see Table 2 or Table 3 in the section Fixed Parameter-Value Distributions. Each distribution accepts a fixed number of parameters in a specific order.
- parameter-values
-
specifies the values of the distribution parameters. You must specify as many parameter-values as the expected number of parameters for the distribution that corresponds to distribution-name. PROC CCDM associates your specified values to parameters in the order the parameters are listed from top to bottom in the Parameters column of Table 2 or Table 3. The Parameters column also specifies the acceptable bounds, if any, for each parameter. Your specified parameter-value must be within those bounds.
You can optionally specify a standard error for each parameter by enclosing the following value-option in parentheses immediately after a parameter’s value:
-
STDERROR=number
SE=number
number specifies the standard error of the parameter. If you specify a value v and standard error s for a parameter, then PROC CCDM assumes that the parameter follows a normal distribution
.
You can separate the parameter value specifications by using either a comma or a space.
-
STDERROR=number
The following example COUNTMODEL statement specifies that the counts follow the standard negative binomial distribution with parameters 
and 
:
countmodel negbin(0.7 5);
If you request perturbation analysis by specifying a positive value for the NPERTURBEDSAMPLES= option and the parameters of your count distribution are stochastic, then it is recommended that you specify the standard error for each parameter. The following COUNTMODEL statement specifies that the counts follow the Weibull distribution with parameters 
and 
such that the standard errors for the 
and 
parameters are 3.5 and 0.25, respectively:
countmodel weibull(50(se=3.5), 0.75(0.25));
With this specification, when PROC CCDM conducts the perturbation analysis, it sets the perturbed values of and to randomly drawn quantiles from the normal distributions 
and 
, respectively.
PROC CCDM permits a continuous count distribution (such as the Weibull distribution in the preceding example) only if you use the SIMULATIONMODE= option to specify a simulation mode other than the collective risk mode. For example, in the pure premium simulation mode (SIMULATIONMODE=PP), the count might represent an average claim frequency that an insurance company observes for a type of policy in a particular time period. Such an average frequency can have fractional values that are best modeled by a continuous distribution.