Latent class models - correct answer ✔✔use a discrete mixture which models each distinct subgroup of
data with its own distribution. Note that the mixture of 2 Poisson variables with different lambdas does
not result in a Poisson distribution
Zero-inflated models - correct answer ✔✔discrete inflated model
- enables modeling the probability of 0 claims with greater flexibility
- response is always 0 for the first subgroup and could be zero or nonzero for the second (AKA - could be
two reasons why response is 0)
-ensures the variance of Y exceeds the mean of Y when (Y| alpha=2) is modeled by Poisson
- (Y | alpha = 2) can be modeled by Poisson or any distribution whose domain starts at 0
- requires modeling as a mixture of a point mass at zero and another distribution whose domain starts
with 0. The point mass at zero can be interpreted as the event of no loss experience. A claim of zero
comes from either no loss experience or a non-reporting when the policyholder experienced a loss.
Hurdle models - correct answer ✔✔- requires modeling as a mixture of a point mass at zero and another
distribution whose domain starts with 1.
- distinguish 0 claims from nonzero claims regardless of underlying reasons
- (Y| alpha = 2) must have a distribution whose domain starts at 1
- Var(Y) may not equal E(Y) ; notably can be either greater than or less than E(Y)
Heterogeneity models - correct answer ✔✔- requires modeling with a continuous mixture
Negative binomial model - correct answer ✔✔- use with log link
R chart - correct answer ✔✔-examines the stability of the variability of a time series
xbar chart - correct answer ✔✔examines the stability of the mean of a time series