Graphs

Question 1

GLM - Steps to create a simple quantile plot

Answer 1

1. Sort the holdout dataset based on that model’s predicted loss cost.
2. Bucket the data into quantiles with each quantile having equal exposure
3. Calculated the average predicted loss cost and average actual loss cost for each bucket and plot them on a graph.

Question 2

GLM - How to pick winning model from quantile plot?

Answer 2

The winning model should be based on 3 criteria:
1. Predictive Accuracy - Difference between actual and predicted in each quantile.
2. Monotonicity - The actual pure premium should consistently increase across quantiles.
3. Vertical distance of actual loss cost between first and last quantile. Indicates how well model distinguishes best and worst risks.

Question 3

GLM - Graph of Quantile and Axis

Answer 3

Question 4

GLM - How to create double lift chart?

Answer 4

1. For each observation, calculate sort ratio = model 1 predicted loss cost / model 2 predicted loss cost.
2. Sort the data by sort ratio in ascending order.
3. Bucket the data into quantiles with equal exposures.
4. Calculate the average predicted loss cost for each model and average actual loss cost for each bucket, divide each by the overall average loss cost from that source, and plot the quantities on a graph.

Question 5

GLM - Explain the graph of a double lift chart

Answer 5

Question 6

GLM - Explain how to create a Loss Ratio Chart

Answer 6

1. Sort dataset based on that model’s predicted loss costs.
2. Bucket the data into quantiles with each have equal exposure.
3. Calculate the actual loss ratio (based on the current rating plan, not the model) for each bucket and plot them on a graph.

Question 7

GLM - How to determine which model performs better in a Loss Ratio Chart

Answer 7

The greater the vertical distance between the lowest and highest loss ratios, the greater the model does at identifying further segmentation opportunities not present in the current rating plan.

Question 8

GLM - Describe the graph of a loss ratio chart

Answer 8

Question 9

GLM - How to create Gini Index/Lorenz Curve?

Answer 9

1. Sort the holdout dataset based on that model’s predicted loss costs.
2. Plot a graph with the x-axis being the cumulative percent of exposures and the y-axis being the cumulative percent of actual losses.

Question 10

GLM - Describe Lorenze Curve and identify Gini?

Answer 10

Question 11

GLM - Describe ROC Curve.

Answer 11

Question 12

Robertson - Explain which graph is better?

Answer 12

The proposed method should be used since it results in well separated groupings, as there is
little overlap between the indicated deductible factors within each coverage A grouping. The
current method groupings have significant overlap, particularly for the 2 middle groups.

Question 13

Bahnemann - Excess Severity Curve Shapes - Pareto, Exponential, Weibull, Lognormal, Gamma

Answer 13

Question 14

Bahnemann - Answer the following:
a) Express G+H in Integral form with layer method
b) Express B+C+D in integral form with size method
c) In letters, what is ILF for S with base of R

Answer 14

Question 15

Fisher - Describe how to perform Quintile’s Test

Answer 15

1. Sort risks by mod from lowest to highest
2. Group into 5 buckets
3. Calculate manual and standard loss ratios using premium weighted average

Question 16

Fisher - how to interpret Quintile’s Test

Answer 16

1. Maximum dispersion in Manual Loss Ratios. Shows plan does good job of identifying risk differences.
2. Flat/Equal Standard LRs. Shows plan corrects for different risks

Question 17

Fisher - how to do Efficiency Test?

Answer 17

Efficiency Test Stat = Variance in SLRs / Variance in Manual Loss Ratios. Lower = better.

Question 18

Fisher - Table L, Graphically identify k.

Answer 18

Question 19

Fisher - Table L, Graphically identify k E(Ld)/E(A)

Answer 19