Two-part models based o
n ge
neralized li
near models are widely used i
n i
nsura
nce rate-maki
ng for predicti
ng the expected loss. This paper explores a
n alter
native method based o
n qua
ntile regressio
n which provides more i
nformatio
n about the loss distributio
n a
nd ca
n be also used for i
nsura
nce u
nderwriti
ng. Qua
ntile regressio
n allows estimati
ng the aggregate claim cost qua
ntiles of a policy give
n a
number of covariates. To do so, a first stage is required, which i
nvolves fitti
ng a logistic regressio
n to estimate, for every policy, the probability of submitti
ng at least o
ne claim. The proposed methodology is illustrated usi
ng a portfolio of car i
nsura
nce policies. This applicatio
n shows that the results of the qua
ntile regressio
n are highly depe
nde
nt o
n the claim probability estimates. The paper also exami
nes a
n applicatio
n of qua
ntile regressio
n to premium safety loadi
ng calculatio
n, the so-called Qua
ntile Premium Pri
nciple (QPP). We propose a premium calculatio
n based o
n qua
ntile regressio
n which i
nherits the good properties of the qua
ntiles. Usi
ng the same i
nsura
nce portfolio data-set, we fi
nd that the QPP captures the riski
ness of the policies better tha
n the expected value premium pri
nciple.
相似文献