pricing

Hi,

I'm a bit confused about the solution given in pricing questions.

1) When the solution says that we fit a model (stoch\stoch, stoch\det etc), does it refer to "statistical models" as opposed to burning cost techniques?

Can we always use this approach, or when pricing certain classes we should not refer to statistical models?

2) When the solution says that we fit a model (stoch\stoch, stoch\det etc) does it refer to GLMs or statistical distributions? It then mentions that we try both additive and multiplicative ones which must refer to GLMS?

3) The solution then says that we can use GLMs to identify the rating factors to use...Does that mean that the models fitted above are not GLMs ?

 

1)
Burning cost models are based on past claims and exposure data. We do not model stochastically past data. We need to project the past data, either using simple grossing up methods or using statistical models .

2)
A GLM is a one type of model, which can be additive or multiplicative (both are statistical models) looking at claim freq and sev either individually or combined.
Other models can be based on freq /sev distributions.

3)
GLM's can identify rating factors. I've not used any GLM's in practice but i've seen software like Emblem which test correlations between factors and can identify which ones are best to use.

 

2) when it refers to additive and multiplicative models it refers to GLMs right?

Also, when it says that e.g a suitable distribution for frequency is Poisson, does it refer to the actual statistical distribution or to the GLM error structure?

 

hmm, good question. i guess we could have add/mult models that are not GLMs, but can't think of any off the top of my head.
anyone else know for sure?

poisson is the error structure.

 

It could also refer to fitting an actual probability distribution though, since Poisson is frequently used for frequency...