What relationship do Bailey & Simon expect between credibilities and #yrs exp
Credibilities should increase in proportion to the number of years of experience, if the chance of accidents for individual risks remains constant and no risks enter or leave the class.
What conditions are necessary for Bailey & Simons assumptions to be valid?
- Risk must stay constant (chance of accident can’t increase or decrease over time)
- No new risks enter or leave the book of business
What is the key point that Bailey & simon are trying to highlight in their paper?
The experience for one car for one year has significant and measurable credibility for experince rating
What exposure base do Bailey & simon use and why
Earned premium at current class B rates. There is ‘maldistribution’ when we use exposure as a base
What causes the ‘maldistribution’ (exposure correlation) in the Bailey & Simon paper
IF higher frequency territories produce more X,Y and B risks, causing correlation between merit rating and territory factors
Maldistribution is caused by higher claim frequency territories being correlated with number of accident-free years. (i.e. higher freq on less acci-free years).
Since higher clm freq terr = higher premium terr, this can be addressed by setting EP as base
How does variance within a class affect the credibility for an individual risk
When variance within the class is higher, credibility for an individual risk is also higher ( easier to ‘learn’ from a single risk when it is more distinct?)
How does variance between classes affect the credibility for an individual risk (i.e. more refined plan >?)
more refined plan, less variation, less relative cred
Describe why a class with a higher volume of claims and more exposures may have less credibility than a class with fewer claims and exposures.
A smaller volume class may receive higher cred because it has higher variation within the class
exp rating credibility depends on not only data volumn, but also variance within a class.
so if more variance within the class, it will receive more credibility even tho it’s a small dataset
