Reasons to use a premium base for frequency
-higher frequency territories have higher avg premium due to territory differentials of class rating
-however, higher frequency territories also
produce more X, Y and B risks => this would increase the avg premium due to experience rating
-therefore using a premium base for frequency cancels the territory differentials and avoid the maldistribution/overlap between class rating and experience rating
**is use car-years as the denominator of frequency, the credibility calculation would account for both “within territory differences” and “between territory differences”. However, territory relativities already account for the between territory differences. Compounding territory relativities with credibility would doublecount the between territory differences, and therefore the credibility would be overstated.
Therefore, claims free drivers would be undercharged while other drivers would be overcharged.
Conditions to use a premium base for frequency
Conditions:
**if data is split by other variables, need to aggregate by territory before testing the 2 conditions
Conclusions of paper
if variance in loss experience between risks is largely explained by classification rating variables, then
experience rating wouldnt add much predictive power
Experience rating distinguishes the individual risk from the class average risk. So individual risk experience is more credible when there is more variance in loss experience
in homogeneous class , do no want to assign a lot of credbility to an individual since their experiencce is most similar to other risks in class
Premium adjustements for relative frequency calculation
Bailey and Simon assumptions for calculating R
Hazam conclusions
Credibility increases in proportion to years only for low credibility values
For larger risk : Simple Buhlmann credibility shows that double or triple the data doesn’t result in exactly double or triple credibility (since k does not change for different samples of n sizes taken from the X variable)
Z = n / (n + k)