The HPCDM Procedure (Experimental)
A scenario represents a state of the world for which you want to estimate the distribution of aggregate losses. The state consists of one or more entities that generate the loss events. For example, an entity might be an individual who has an insurance policy or an organization that has a workers’ compensation policy. Each entity has some characteristics of its own and some external factors that affect the frequency with which it generates the losses and the severity of each loss. For example, characteristics of an individual with an automobile insurance policy can include various demographics of the individual and various features of the automobile. Characteristics of an organization with a workers’ compensation policy can be the number of employees, revenue, ratio of temporary to permanent employees, and so on. The organization can also be affected by external macroeconomic factors such as GDP and unemployment of the country where the organization operates and factors that affect its industry. You need to quantify and specify all these characteristics as external factors (regressors) when you fit severity and frequency models.
You should specify all the information about a scenario in the DATA= data set that you specify in the PROC HPCDM statement. Each observation in the DATA= data set encodes the characteristics of an entity. For proper simulation of counts and severities, you must specify in the DATA= data set all the characteristics that you use as regressors in the frequency and severity models. All the regressors are expected to have nonmissing values. If any of the regressors have a missing value in an observation, then that observation is ignored.
The information in the DATA= data set is interpreted as follows, based on whether you specify the EXTERNALCOUNTS statement:
-
If you do not specify the EXTERNALCOUNTS statement, then all the observations in the data set, or in one BY group if you specify the BY statement, form a scenario. The observations are used together to compute one random draw from the compound distribution. The total number of draws is equal to the value that you specify in the NREPLICATES= option. The simulation process is described in the section Simulation with Regressors and No External Counts and illustrated using an example in the section Illustration of Aggregate Loss Simulation Process.
-
If you specify the EXTERNALCOUNTS statement, then the DATA= data set is expected to contain multiple replications (draws) of the frequency counts that you simulate externally for a scenario. The DATA= data set must contain the COUNT= variable that you specify in the EXTERNALCOUNTS statement. The replications are identified by the observation number or the ID= variable that you specify in the EXTERNALCOUNTS statement. For each observation in a given replication, the COUNT= variable is expected to contain the count of losses that are generated by the entity associated with that observation. All the observations of a given replication are used together to compute one random draw from the compound distribution. The size of the compound distribution sample is equal to the number of distinct replications that you specify in the DATA= data set, multiplied by the value that you specify in the NREPLICATES= option. The simulation process is described in the section Simulation with External Counts and illustrated using an example in the section Illustration of the Simulation Process with External Counts.
In both cases, an observation can also contain severity adjustment variables that you can use to adjust the severity of the losses generated by that entity, based on some policy rules. For more information about simulating the adjusted compound distribution sample, see the section Simulation of Adjusted Compound Distribution Sample.