4.6 Operational Risk

4.6 Operational Risk

Question 1

PE2018Q58 / PE2019Q58 / PE2020Q58 / PE2021Q58 / PE2022Q58
An operational risk analyst is attempting to estimate a bank’s loss severity distribution. However, there is a limited amount of historical data on operational risk losses. Which of the following is the best way to address this issue?

A. Generate additional data using Monte Carlo simulation and merge it with the bank’s internal historical data.
B. Estimate the parameters of a Poisson distribution to model the loss severity of operational losses.
C. Estimate relevant probabilities using loss information that is published by credit rating agencies.
D. Merge external data from other banks with the bank’s internal data after making appropriate scale adjustments.

Answer: D
Learning Objective: Describe the common data issues that can introduce inaccuracies and biases in the estimation of loss frequency and severity distributions.

D is correct. Using external data obtained from other banks is one good way to increase the data set of historical operational losses. Data from other banks needs to be adjusted for size before being merged with the bank’s internal data.

A is incorrect. Using distributions does not help resolve the issue of incomplete underlying data.

B is incorrect. Lognormal distributions, not Poisson distributions, are generally used for modeling loss severity. Also, using distributions does not help resolve the issue of incomplete underlying data.

C is incorrect. Credit losses are generally much better documented than operational losses inside the bank. External credit ratings publish probability of default and expected loss data that provides additional data. Operational loss is generally documented much less rigorously and regulatory initiatives are now pushing banks to document operational losses data.

---

Question 2

PE2019Q16 / PE2020Q16 / PE2021Q16 / PE2022Q16
The CRO of a large bank is interviewing a candidate for an operational risk analyst position. Which of the following statements made by the candidate concerning the measurement of operational risk is correct?

A. Economic capital of a bank should be sufficient to cover both the expected and the worst-case operational risk losses of the bank.
B. Loss severity and loss frequency are often modeled with lognormal and Poisson distributions, respectively.
C. Operational loss data available from data vendors tend to be biased toward small losses but are particularly useful in determining loss frequency.
D. The standardized approach used by banks in calculating operational risk capital requires the calculation of unexpected as well as expected losses.

Answer: B
Learning Objective: Describe the allocation of operational risk capital to business units.

Economic capital covers the difference between the worst-case loss and the expected loss.

It is true that loss frequency is typically modeled using a Poisson distribution and loss severity tends to be modeled with a lognormal distribution.

Operational loss data available from data vendors tends to be biased towards large losses and are most useful for determining relative loss severity.

In the standardized approach to calculating operational risk, a bank’s activities are divided up into several different business lines, and a beta factor is calculated for each line of business. The bank does not have to estimate unexpected losses under the standardized approach.

---

Question 3

PE2022PSQ20
An operational risk manager is presenting to a group of risk analysts about different techniques to model operational risk. An analyst asks the manager about the appropriate use of the power law in estimating operational losses. Which of the following would be a correct statement for the manager to make about the use of the power law?

A. It implies that operational losses tend to follow a normal distribution.
B. It is more effective in modeling some types of operational risk, such as losses from fraud, than others, such as losses from natural disasters.
C. It is generally used to estimate routine operational losses which occur at a relatively high frequency.
D. It is suitable for modeling the tail of the operational loss distribution, but not for modeling the body of the distribution.

Answer: D
Learning Objective: Explain how to use the power law to measure operational risk.

D is correct. If $v$ is the value of a random variable and $x$ is a high value of $v$, then the power law holds it is approximately true that:

$$Pr(v>x)=K{x}^{-\alpha }$$

where $Pr$ denotes probability and $K$ and $\alpha$ are parameters. The power law only describes the right tail of the distribution (not the whole distribution). That is why the equation above is approximately true only for high values of $x$.

A is incorrect. Operational losses do not follow a normal distribution, nor does the use of the power law in connection with a random variable suggest that the random variable is normally distributed.

B is incorrect. The power law has been shown to be applicable to a wide range of distributions, relating to events including natural disasters (i.e., the magnitude of earthquakes). The power law is no less appropriate for modeling operational losses from natural disasters than those from fraud or other sources.

C is incorrect. These losses occur in the body of the distribution so are unlikely to be modeled using the power law. These losses are not considered extreme events.

---