Chapter 15 Quiz: Credit Risk Modelling and Model Risk Management

Instructions: Select the single best answer for each question. Each question is worth one point. A score of 13 or higher (81%) reflects strong command of the material.

Question 1

A bank's SME portfolio has the following aggregate characteristics: Probability of Default = 3.5%, Loss Given Default = 45%, Exposure at Default = £180 million. What is the Expected Loss for this portfolio?

A) £2.83 million B) £28.35 million C) £2.835 million D) £56.70 million

Question 2

Under the Basel III Standardised Approach (SA), which of the following risk weights would apply to an unrated corporate exposure?

A) 20% B) 50% C) 75% D) 100%

Question 3

What is the primary distinction between Foundation IRB (F-IRB) and Advanced IRB (A-IRB) under the Basel framework?

A) F-IRB requires external credit ratings; A-IRB does not. B) Under F-IRB, banks estimate PD only; under A-IRB, banks also estimate LGD and EAD. C) F-IRB is available only for retail portfolios; A-IRB covers all exposure classes. D) A-IRB is mandatory for banks with assets above £50 billion.

Question 4

A credit scorecard's Gini coefficient is calculated as 0.52 on the hold-out test set. How should this result be interpreted?

A) The model is likely to have been built with an error — a Gini above 0.5 is impossible. B) The model shows good discrimination ability, well above the minimum acceptable threshold of 0.30. C) The model's calibration is acceptable, but its discrimination requires improvement. D) The Gini of 0.52 indicates the model is slightly worse than a random classifier.

Question 5

A Population Stability Index (PSI) calculation for a retail mortgage scorecard returns a value of 0.31, comparing the current scoring population against the development sample from three years ago. What is the most appropriate regulatory and risk management response?

A) No action required — PSI below 0.50 is within acceptable limits. B) Increase the monitoring frequency from quarterly to monthly and reassess in six months. C) The PSI indicates a major population shift; the model likely no longer reflects the current population and should be reviewed for redevelopment or overlay. D) Recalibrate the model's intercept to reduce the PSI below 0.25, then continue using it without further review.

Question 6

SR 11-7, issued by the Federal Reserve and OCC, addresses:

A) The capital treatment of model risk under Pillar 1 of Basel III. B) The supervisory framework for managing risks arising from the use of quantitative models in decision-making. C) The accounting treatment of provisions under IFRS 9 Expected Credit Loss. D) The specific validation requirements for IFRS 9 models only.

Question 7

Which of the following most accurately describes the SR 11-7 requirement for model validation independence?

A) Validation may be performed by the model developer, provided they have not used the model in production for at least six months. B) Validation must be performed by a third-party consultant external to the bank. C) Validation must be performed by staff who are independent of the model's development and approval process; internal staff may perform this role. D) Independence requires that the validator works in a different business line from the developer, but the same risk function is acceptable.

Question 8

Which of the following PD estimation approaches is most consistent with IFRS 9 Expected Credit Loss modelling requirements?

A) Through-the-cycle (TTC) PD, averaged across a full economic cycle. B) Point-in-time (PIT) PD, reflecting current creditworthiness and incorporating forward-looking macroeconomic information. C) Agency-implied PD derived from Moody's published historical default tables. D) The regulatory PD floor of 0.03% as specified in CRR2 Article 160.

Question 9

Under IFRS 9, a loan is classified as Stage 2 when:

A) The borrower is more than 90 days past due on any credit obligation. B) There has been a significant increase in credit risk (SICR) since initial recognition, even though the borrower has not yet defaulted. C) The expected credit loss provision exceeds the loan's carrying amount. D) The loan has been restructured and the borrower's payment plan has changed.

Question 10

A credit model has an AUC-ROC of 0.72 on the hold-out test set. What is the corresponding Gini coefficient?

A) 0.72 B) 0.44 C) 0.36 D) 0.28

Question 11

The Weight of Evidence (WoE) for a bin containing only non-defaulters (zero bad borrowers) would theoretically be:

A) Zero — the bin has perfect good performance. B) Negative infinity — the log of zero is undefined, requiring adjustment. C) 1.0 — representing 100% good borrowers. D) Positive infinity — indicating maximum WoE for good performance.

Question 12

A bank is developing a credit scoring model for an SME portfolio. Information Value calculations for candidate variables produce the following results:

Variable IV
Years in business 0.18
Industry sector code 0.09
Annual revenue 0.24
Number of employees 0.03
Postcode area 0.62

Which variable should prompt the most urgent additional investigation before inclusion in the model?

A) Industry sector code (IV = 0.09), because it is below the threshold for a strong predictor. B) Number of employees (IV = 0.03), because it barely exceeds the minimum IV threshold. C) Postcode area (IV = 0.62), because a very high IV may indicate data leakage, look-ahead bias, or a proxy for a protected characteristic. D) Annual revenue (IV = 0.24), because high-IV financial variables are typically multicollinear with other financial ratios.

Question 13

Under the Basel III capital floor introduced in the 2017 Basel 3.1 revisions, IRB-based Risk-Weighted Assets (RWAs):

A) Cannot fall below 80% of the standardised approach RWA for the same portfolio. B) Cannot fall below 72.5% of the standardised approach RWA for the same portfolio. C) Cannot fall below 50% of the standardised approach RWA for any single exposure class. D) Are unconstrained as long as the model has received regulatory approval.

Question 14

A retail mortgage origination scorecard was developed in 2019 using data from 2014–2018. In 2024, a model validator calculates the following metrics on recent data:

  • Gini: 0.38 (vs 0.51 at development)
  • KS: 0.22 (vs 0.34 at development)
  • PSI: 0.28 (development population vs 2024 scoring population)

Which conclusion is most appropriate?

A) The model remains acceptable; a Gini of 0.38 is above the minimum threshold of 0.30. B) The model shows moderate deterioration in discrimination but PSI is within normal limits. C) Both discrimination deterioration (Gini down 13 points) and a PSI indicating major population shift suggest the model is no longer fit for purpose; redevelopment or material overlay is required. D) Only the PSI requires action; the Gini and KS are immaterial concerns.

Question 15

Machine learning models (such as gradient boosting) present additional model risk compared to traditional logistic regression scorecards primarily because:

A) ML models are computationally expensive and slow to run in production. B) ML models cannot produce probability estimates — only binary accept/decline outputs. C) ML models are more complex, harder to explain to regulators and customers, and may produce non-monotonic relationships that are conceptually unsound. D) ML models cannot achieve Gini coefficients above 0.50 on credit datasets.

Question 16

Under EBA/GL/2017/16, which of the following minimum data requirements applies to banks seeking F-IRB (Foundation IRB) approval for PD estimation?

A) At least two years of default data, including data from the most recent quarterly period. B) At least five years of historical default data, ideally spanning at least one economic downturn. C) At least ten years of historical data, covering two complete economic cycles. D) Historical data requirements are not specified — they are at the discretion of the national competent authority.

Answer Key

Q Answer Explanation
1 C EL = 3.5% × 45% × £180m = 0.035 × 0.45 × 180 = £2.835m
2 D Unrated corporates carry a 100% risk weight under the SA.
3 B F-IRB: bank estimates PD only; supervisory LGD/EAD. A-IRB: bank estimates all three components.
4 B Gini of 0.52 is good discrimination — above the minimum threshold of 0.30 and within the typical "good" range for retail credit models.
5 C PSI > 0.25 signals a major population shift; the model should be reviewed and likely redeveloped or supplemented with an expert overlay.
6 B SR 11-7 is the supervisory guidance on model risk management — the risk that model outputs cause poor decisions.
7 C SR 11-7 requires independence from development and approval — internal validation functions satisfy this if structurally independent. External consultants are not mandatory.
8 B IFRS 9 explicitly requires forward-looking PIT estimates incorporating macro scenarios. TTC PDs would understate provisions in downturns.
9 B Stage 2 is triggered by SICR since origination — the borrower has not yet defaulted but credit risk has increased significantly.
10 B Gini = 2 × AUC − 1 = 2 × 0.72 − 1 = 0.44.
11 D A bin with only goods has zero bads: WoE = ln(dist_good / dist_bad) → ln(x/0) → +∞. In practice, a small adjustment (0.5 pseudocount) is applied. The theoretical answer is positive infinity.
12 C IV > 0.50 is a warning sign. Postcode may be acting as a proxy for protected characteristics (race, ethnicity), and must be reviewed for fairness and potential data leakage before use.
13 B The Basel 3.1 output floor is 72.5% of the SA RWA equivalent.
14 C Combined signal: Gini down significantly AND PSI above 0.25. Both discrimination and population stability concerns together indicate the model is no longer reliable; action beyond monitoring is required.
15 C The primary model risk of ML is complexity leading to inexplicability, potential non-monotonic outputs, and difficulty satisfying the SR 11-7 conceptual soundness requirement.
16 B EBA/GL/2017/16 specifies a minimum of five years of historical default data for PD estimation, with preference for data spanning at least one economic downturn.

Scoring Guide

Score Level
14–16 Distinction — ready to apply these concepts in professional practice
11–13 Pass — solid understanding; review any missed questions
8–10 Borderline — revisit Sections 15.2, 15.5, and 15.6
Below 8 Needs revision — re-read the chapter before proceeding

Chapter 15 of Regulatory Technology (RegTech): A Practitioner's Guide