Ch 18: GLM Flashcards
1
Q
One-way analysis
A
- Looks at the frequency and severity of each rating factor separately.
- Ignores correlations and interaction effects between variables, as a results the model may underestimate or double count the effects of variables.
2
Q
When should an interaction term be used in the GLM?
A
- Should be used where the pattern in the response variable is better modelled by including extra parameters for each combination of two or more factors
- Interaction exists when an effact of one factor depends on the value of another factor.
3
Q
Drawbacks of simple linear regression
A
- Assumes the response variable is normally distributed, which may not be appropriate
- The normal distribution has constant variance which may not be appropriate for the variable beoing modelled (for example variance of claims increase as claim numbers increase - poisson distribution has this property)
- Normal model ‘adds’ together the effects of different explanatory variables, but is seldom what is observed in practice, effects may be multiplicative rather than additive
- More difficult to find solution where there is more thna 2 explanatory variables
4
Q
Advantages of GLM over simple regression
A
- Response can take any distribution from the exponential family
- Link function is introduced, acts to remove the assumption that affects of different variables must be added together
- Additionaly allow for an offset term
5
Q
Properties of exponential family of distributions
A
- Distribution is completely specified in terms of its mean and variance
- Variance of the response is a function of its mean
6
Q
Why tweedie distribution is nice for modelling claims experience
A
- Variance proportional to mu^p (p is additional parameter)
- If p is selected between 1 and 2, the distribution has a point mass at zero
- Distribution corresponds to the compound distribution of a Poisson claim number process and a gamma claim size distribution
7
Q
What does chi-squared test measure?
A
Measures whether the inclusion of one or more additional explanatory variables in the model improves the model fit significantly