Ratemaking - Risk Classification Flashcards

1
Q

What are the pros and cons of univariate approaches?

A
  • They do not take into consideration other variables
    • They are distorted by distributionnal bias and the results can also be heavily distorted by unsystematic effects (noise).
  • The result is a set of answers with no additional information about the certainty of the results.
  • Interactions can be incorporeted but only with two-way or three-way analysis
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

What is the primary advantages of univariate approaches?

A

Easy to understand and transparency results.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

What are the pros and cons of the mimimum bias methods?

A

Pros

  • Account for eneven mix of business
  • High transparency

Cons

  • Computationally inefficient
  • No diagnostics are included
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

What are the pros and cons of the multivariate methods?

A

Pros:

  • Consider all rating variable -> adjust for exposure correlation
  • Remove noise and capture signal
  • Produce model diagnostics
  • Allow interaction

cons:

  • Lack of transparency (glm can still be good)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

Why the balancing of the fundamental insurance equation is necessary at a individual level?

A

Because otherwise, the compagnies would subject to adverse selection, resulting in financial deterioration.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

Explain adverse selection.

A

Adverse selection is the process that happen when a compagny have bad segmentation, resulting in all low-risk leaving and high-risk coming.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

What a compagny can do to break the adverse selection circule?

A
  1. Improve segmentation
  2. Become insolvant
  3. Specialise in high risk only and raise rate accordingly
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

Balancing the insurance equation at the aggregate level brings __?

Balancing the insurance equation at the individual level brings __?

A

Aggregate = profitability

Individual = competivity

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

What is the favorable selection? What are the two option that a compagny can do when they are in a favorable selection possition?

A

It when a compagny identify a new rating variables that other don’t have, given a competitif advantage.

  1. Implement the new variable in rating -> better segmentation -> better low-risks
  2. Risk Selection -> better risk will increase profitability -> long-term rate will decrease.
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

What are the 4 criteria for evaluating rating variables?

A
  1. Statistical
  2. Operational
  3. Social
  4. Legal
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

In variable selection, explain what is the statistical criteria?

A

The rating variables should reflect the variation in expected costs among different groups of insureds.

  • Statistical signifiance
  • Homogeneity -> group of similar risk
  • Credibility -> group large enough and/or stable enough
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

In variable selection, explain what is the operational criteria?

A

The variable should be practical to use:

  • Objective -> objective definition
  • Inexpensive to administer
  • Verifiable -> should not be manipulate by the inssured
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

Complete; The goal of classification is to balance ___

A

Grouping into sufficient level to be homogeneous while having stability (credibility) in these groups.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

In variable selection, explain what is the Social criteria?

A

Insurance cie are affected by social perception

  • Affordability
  • Causality -> direct impact to the amount of expected loss
  • Controllability -> the inssured have some control to his class
  • Private concerns
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

In variable selection, explain what is the Legal criteria?

A

Each state/province have law and regularisation to be sure that rate are “not excessive, not inanequate, and not unfairly discriminatory”.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

What is self-selection?

A

Self-selection is when a product is on volunter base, only the risk that have advantage to take it will take it -> biased data.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
17
Q

What are the distortion of the Pure Premium Approach in Risk Classification.

A
  • The method assume a uniform distribution of exposure across all other rating variable (assume no exposure correlation).
    • Can create a double count effect if exposure are correlated
  • Can be hard to allocate ULAE to the differents class, so the pure premium used may only include ALAE.
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
18
Q

What are the distortion of the Loss Ratio Approach in Risk Classification.

A
  • Can be hard to allocate ULAE to the differents class, so the pure premium used may only include ALAE.
  • Adjuste for distributional bias but will yield true relativities only if other variables relativities reflect the true value.
  • The method will adjust for inequity present in other variable.
19
Q

What are the distortion of the adjusted pure premium Approach in Risk Classification.

A

Same results as with the Loss Ratio method.

  • Can be hard to allocate ULAE to the differents class, so the pure premium used may only include ALAE.
  • Adjuste for distributional bias but will yield true relativities only if other variables relativities reflect the true value.
  • The method will adjust for inequity present in other variable.
20
Q

Why the multivariate method were adopted?

A
  1. Computer power increased -> data no longuer need to be aggregated
  2. More granularity and accessibility of the data
  3. Competitive pressure -> Not to be in adverse selection
21
Q

What are the different type of Minimum Bias Procedure.

A
  1. Rating structure
    • Additive
    • Multiplicative
    • Combined
  2. Bias function
    • Balance principle
    • Least square
    • Chi-deux
    • Miximum likelihood
22
Q

What is the Sequential Analysis.

A

It is a Minimum Bias Procedure with one iteration. The order of variable will affect the results.

23
Q

Why GLMs are the most used multivariate method?

A
  • Each iteration can be analysed
  • Easy to understand
  • Produce a serie of multiplier - which is what the insustry is used.
24
Q

What is the difference between the GLM and LM models?

A

The GLM remove the normality and constant variance assumption.

25
Q

What are the step to fit a GLM

A
  1. Have suitable number of observation
  2. Select the link function - define relationship between Y and eta (linear predictor)
  3. Specify the distribution of the random process - Mean and variance
26
Q

Why the Loss Ratio is not used in GLM modelisation?

A
  • Hard to on-level premium at grannular level
  • Experience Actuary have priori expectation on frequency and severity
  • Loss Ratio model would became useless after a rate change
  • There is no commonly accepted distribution for modeling Loss Ratio
27
Q

Explain on which type of data GLM modelisation are done?

A
  • One GLM for frequency and severity
  • Using homogeneous claims data (e.g., by coverage)
28
Q

Reminder: GLM’s results for a given variables is only meaningful if all other variable are considered at the same time -> Removing variable will change results

A
29
Q

Explain GLM’s standard errors diagnostic.

A

Standard errors show the range within which the modeler is sure at 95% that the true value is between.

  • Small standard error = variable is signifiant
  • Wide standard error = variable is detecting mostly noise and should be removed.
30
Q

Explain what is the signal and the noise in a GLM?

A

The signal (systematic effect) is what lets the model generalize to new situations. The noise (unsystematic effect) is everything else that gets in the way of that

31
Q

What is deviance in GLM diagnostics?

A

Deviance is a measure of how much the fitted value differ from the observation.

Used to compare nested model between them

32
Q

How can we compare non-nested GLM models?

A

Using the AIC and BIC

33
Q

What the Chi-square and F-test GLM diagnostic measure?

A

They measure the trade off between the gain in accuracy by adding more variable, while increasing the complexity.

34
Q

Explain two practical in nature diagnostic for GLM.

A
  1. Comparing GLM results for individual years expecting them to be consistant.
  2. Cross-validation or test validation data.
35
Q

What are the practival considerations when modeling using a GLM?

A
  1. Ensuring data is adequate
  2. Identifying anomalous results
  3. Reviewing model with consideration of statistical theory and business application
  4. Developing appropriate methods of communication for the results.
  5. Take consideration of large loss, cat, premium development and premium trend
  6. Take consideration of geopgraphies in the data (state/province)
  7. Take consideration IT constraint, marketing objective and regulation rules.
36
Q

Explain the factor analysis.

A

The factor analysis is a method that create new uncoreled variables that are a linear combinaison of the other. This can be used to reduce the number of predictor. Example is principle components technique.

37
Q

Explain the Cluster Analysis.

A

Method that combine small groups of similar risk into large homogeneous categories (cluster). It aim to minimises within variance and maximize between variance.

Often use to create geographical levels.

38
Q

Explain the CART method?

A

Create tree that can help the actuary to identify a strong list of initial variable. It also help detecting interaction between variable.

39
Q

Explain the MARS (Multivariate Adaptive Regression Spline) method.

A

Operate as multiple piecewise linear regression where each breackpoint defines a region for a particular linear regression equation.

This is uselly used to select breackpoint for categorizing continuous variable. Can also help detecting interactions.

40
Q

Explain the Neural Nerwork.

A

Sophisticated modeling techniques. It is descrive as a recursion applied to a GLM.

41
Q

In general, how can the data mining methods can help?

A
  • Reducing variable of the GLM
  • Providing guidance on how to categorize discrete variable
  • Reduce level of discrete variable
  • Identifying the interaction between variable
42
Q

Name 4 external sources of data that can be used in GLM modelisation?

A
  • Geo-demographics data (population density)
  • Weather
  • Property characteristics
  • Other information about individual insured (credit score)
43
Q
A