Frequency vs Severity Trending

[tatos]

For the frequency-severity approach, the notes state that we apply severity trending at the ground-up individual loss level whereas we apply frequency trending to the claim frequencies for each historical policy year - why is there this difference?

And then later on the notes go on to say that we split by size of claim: "we may consider losses in aggregate or banded into two or more size-based groupings. If we allow for this approach we will need to split the frequency." I don't understand this. Why do the notes first say we apply severity trending to individual losses and then say we split the losses into groups or just consider the total loss? And then it says we will need to split frequency? But we were just told that frequency gets applied to whole policy years?

I'm really confused - I don't know if it's just that I've only worked in personal lines and that this would otherwise be clearer or obvious, but I'd be really glad if someone could help me understand this. Thanks ..

 

[Katherine Young]

For the frequency-severity approach, the notes state that we apply severity trending at the ground-up individual loss level whereas we apply frequency trending to the claim frequencies for each historical policy year - why is there this difference?

This is just saying that you can trend each individual loss amount, but that it doesn't make sense to calculate a "claim frequency for individual losses", that's nonsense. So you have to calculate claim frequency for a group of data, eg by policy year.

we may consider losses in aggregate or banded into two or more size-based groupings. If we allow for this approach we will need to split the frequency

Small claims and large claims may be subject to different trends, so you might want to analyse these separately. If so, you would have to calculate the frequency of large losses, and the frequency of small claims separately too. That way, we would calculate a "risk premium for large losses", and a "risk premium for small losses", and the total risk premium would be the sum of the two.

Why do the notes first say we apply severity trending to individual losses and then say we split the losses into groups or just consider the total loss?

See above explanation. You only need to group data if the claims characteristics / trends are different for each group - if this is the case, you need to analyse the different claim groups separately.

 

[tatos]

Okay that makes sense - thanks Katherine