E
entact
Member
In working on a technical analysis (GLIM) at the moment and I have a couple of questions...
I've created frequency/severity models for TPI/TPD etc. using all potential rating factors. I will be presenting the results to underwriting and then requesting feedback based on whether each significant factor can be implemented, any restrictions on rating (e.g. max loads/discounts etc.)
In order to get the necessary feedback I need to proceed I presumably need to create a combined model and present the results of that to the underwriters? Since I will be changing all the models once I know which factors are not implementable, do I need to bother with adjustments like
profit margin
expenses
reinsurance loading
etc.
at this stage or can this wait until I am coming up with the technical rates (rates with implementable factors)?
Say the underwriters feedback is that 6 factors are not implementable. Do I need to adjust all of my individual freq/sev models and re-model all of the factors or is that overkill.
I presume an impact analysis, testing model prediction via an out-of-sample is done after this stage
Many thanks
I've created frequency/severity models for TPI/TPD etc. using all potential rating factors. I will be presenting the results to underwriting and then requesting feedback based on whether each significant factor can be implemented, any restrictions on rating (e.g. max loads/discounts etc.)
In order to get the necessary feedback I need to proceed I presumably need to create a combined model and present the results of that to the underwriters? Since I will be changing all the models once I know which factors are not implementable, do I need to bother with adjustments like
profit margin
expenses
reinsurance loading
etc.
at this stage or can this wait until I am coming up with the technical rates (rates with implementable factors)?
Say the underwriters feedback is that 6 factors are not implementable. Do I need to adjust all of my individual freq/sev models and re-model all of the factors or is that overkill.
I presume an impact analysis, testing model prediction via an out-of-sample is done after this stage
Many thanks