Clark

Question 1

Three advantages to using parameterized curves to describe loss emergence patterns

Answer 1

Only have to estimate 2 parameters
Can use data from triangles without evenly spaced evaluations
Final pattern is smooth

Question 2

Why CC is preferred over CL when few points exist

Answer 2

Requires estimation of fewer parameters; CL requires parameter for each AY + parameters for growth curve, tends to be overparameterized

Question 3

Two components of variance of actual loss emergence

Answer 3

Process - random variation in actual loss emergence

Parameter - uncertainty in estimator

Question 4

Two advantages of using ODP to model actual loss experience

Answer 4

1) High flexibility (can match first/second moments of any distribution)
2) MLE produces CC and CL estimates

Question 5

Key assumptions underlying Clark model

Answer 5

Incremental losses are iid
Variance/mean scale parameter is fixed and known
Variance estimates based on Rao-Cramer lower bound

Question 6

Three graphical tests used to validate Clark’s assumptions

Answer 6

Plot normalized residuals against:

1) increment age (scattered around zero - model appropriate)
2) Expected loss in each increment (scattered around zero, scale is constant)
3) CY (scattered around zero, no CY effects)

Question 7

Why it might be necessary to truncate LDFs

Answer 7

Reduce reliance on extrapolation for curves with heavy tails

Question 8

Compare/contrast process and parameter variances of CC and CL

Answer 8

CC can produce higher or lower process variance

CC produces lower parameter variance since it requires fewer parameters and has information from exposure base