5

Question 1

Question 1
Which of the following statements about multiclass classification is CORRECT?
a) Cumulative logistic regression estimates a separate intercept per rating class, which results in parallel logit functions.
b) Decision trees cannot be used for multiclass targets.
c) In the cumulative logistic regression model, the cumulative probability of a corporate C having a rating higher than R is modelled.
d) Cumulative logistic regression is typically used for nominal target variables.

Answer 1

a) Cumulative logistic regression estimates a separate intercept per rating class, which results in parallel logit functions.

b) decision tree possible for multiclass targets
c) does not model C > R, rather just the cimulative probabaility of C and R seperatly
d) not nomianal, but ordinal target variables.

Question 2

Question 2
Application scoring models typically have
a) higher AUC and higher amount of variables than behavioural scoring models.
b) lower AUC and higher amount of variables than behavioural scoring models.
c) higher AUC and lower amount of variables than behavioural scoring models.
d) lower AUC and lower amount of variables than behavioural scoring models.

Answer 2

d) lower AUC and lower amount of variables than behavioural scoring models.

Question 3

Question 3
When defining ratings, the difference between the AUC of the original scorecard and the AUC of the rating system should be
a) maximized.
b) minimized.

Answer 3

b) minimized.

Question 4

Question 4
When using decision trees for rating definition, it is recommended to use
a) classification trees.
b) regression trees.

Answer 4

b) regression trees.

Unlike classification trees that are suitable for categorical targets, regression trees are designed for predicting continuous numerical values, making them more appropriate for rating definitions where the output is often a numerical score or value.

Question 5

Question 5
Consider the following decision tree for defining PD ratings
This decision tree contains
a) a monotonicity violation.
b) no monotonicity violation.

Answer 5

a) a monotonicity violation.

Question 6

Question 6
When using decision trees to define PD ratings, the monotonicity constraint can be enforced during
a) the splitting decision only.
b) the stopping decision only.
c) both the splitting and the stopping decision.

Answer 6

c) both the splitting and the stopping decision.

Question 7

Question 7
Point in time (PIT) ratings are typically
a) stable.
b) volatile.

Answer 7

b) volatile.

– Rating changes rapidly with macroeconomic situation (lots of rating mobility!)

Question 8

Question 8
Through the Cycle (TTC) ratings are typically
a) stable.
b) volatile.

Answer 8

a) stable.

– Rating robust with respect to macro-economic situation (not much rating mobility)

Question 9

Question 9
The Markov assumption in rating migration analysis states that
a) the rating at time t+1 only depends upon the rating at time t.
b) the rating at time t+1 depends upon the ratings at time t, time t-1, etc.

Answer 9

a) the rating at time t+1 only depends upon the rating at time t.

Question 10

Question 10
A higher value for the mobility metric implies that the rating philosophy is
a) TTC.
b) PIT.

Answer 10

b) PIT.

Question 11

Question 11
Calibration tries to deal with the following sources of default rate variability:
a) Sampling uncertainty
b) Economic conditions
c) Internal effects
d) All of the above

Answer 11

d) All of the above

Question 12

Question 12
Assuming Nt is the number of customers at risk at the beginning of year t, Dt the number that defaulted during year t and Nw the number of withdrawals during year t, the withdrawal corrected cohort DR calibrates is calculated as follows:
a) DR = Dt / Nt
b) DR = Dt / (Nt – 0.5Nw)
c) DR = Dt / (Nt + 0.5Nw)
d) DR = Dt / (Nt + Nw)

Answer 12

b) DR = Dt / (Nt – 0.5Nw)

(2) Withdrawal corrected cohort DR
– DR = Dt / (Nt – 0.5Nw)

(1) Uncorrected cohort DR
– DR = Dt / Nt

Question 13

Question 13
Which statement is NOT CORRECT?
a) The reason for defining ratings, is that the credit scores are too fine granular, or in other words too much detailed. When working directly with the credit scores, a slight change in customer characteristics might give a different score and hence different capital. This would make the capital very volatile which is clearly undesirable.
b) A key problem when using classification trees that optimize the misclassification rate to define default ratings, is that all goods are assigned into only a few ratings without much diversity.
c) An n-period migration matrix can be easily obtained by adding the migration matrix M, n times which results in n x M.
d) Many migration matrices are typically diagonally dominant which means that many obligors tend to keep their rating and only a minority migrates.

Answer 13

c) An n-period migration matrix can be easily obtained by adding the migration matrix M, n times which results in n x M.

*correct operation is M^n,

Question 14

Question 14
Which statement is CORRECT?
a) A PIT rating philosophy combined with a PIT calibration philosophy is considered to be Basel compliant.
b) A TTC rating philosophy combined with a TTC calibration philosophy is considered to be IFRS 9 compliant.
c) A TTC rating philosophy combined with a PIT calibration philosophy is considered to be IFRS 9 compliant.
d) A PIT rating philosophy combined with a TTC calibration philosophy is considered to be non-Basel compliant.

Answer 14

c) A TTC rating philosophy combined with a PIT calibration philosophy is considered to be IFRS 9 compliant.