Gains:
- Ability to design the entire actuarial workflow as an end-to-end process supported by artificial intelligence, from data entry to the final signed report
- Ability to establish version control, audit trail and repeatability discipline when integrating artificial intelligence tools into the workflow
- Ability to understand how the actuarial profession is transforming in the age of artificial intelligence and plan one's own skill development and professional responsibilities.
Module Exam
1. An actuary asks the AI "what was the loss frequency of our portfolio last year?" without providing policy data. he asks, and the model gives a confident number. Which risk does this situation most exemplify?
- A) Hallucination: the model produces a made-up number without access to data ✔
- B) Overfitting
- C) Data leakage
- D) Mortality improvement
Explanation: When no data is given to the model, the number it produces is a hallucination: a fluent but made-up output that is not based on real data. Numerical results such as frequency should always be calculated from the institution's own data.
2. What is the main reason for the distinction between "frequency" and "severity" in damage modeling?
- A) Because the legislation prohibits modeling frequency and intensity together
- B) Because it is only valid for life insurance
- C) Frequency and intensity follow different distributions; Modeling separately and combining them predicts total damage more accurately ✔
- D) Since frequency and intensity always take the same value, it is sufficient to calculate one of them.
Explanation: The number of damages (frequency) and the amount per damage (severity) have different behaviors and different probability distributions; The total damage (risk premium) is obtained by modeling and multiplying separately. This distinction allows more accurate prediction of both.
3. What does IBNR equivalent mean?
- A) Legal fund allocated for premium refund
- B) Provision for damages incurred but not yet reported ✔
- C) Commission to be received from the reinsurer
- D) Total of paid and closed damages
Explanation: IBNR (Incurred But Not Reported) is the provision set aside for the future cost of damages that have occurred but have not yet been reported to the company. It is estimated by methods such as chain ladder.
4. What is the use of the "development factor" in the chain-ladder method?
- A) Updates premiums according to inflation
- B) Divides the number of policies by years
- C) Calculates the capital requirement directly
- D) It is used to predict the final damage by giving the growth rate of cumulative damage from one developmental period to the next. ✔
Explanation: The growth factor indicates how many times the cumulative damage of a damage period grows from one year of development to the next; Using these factors, the final damage of the immature stages is estimated.
5. Why is GLM (generalized linear model) used in pricing widely preferred in actuaries?
- A) Because it always gives the highest prediction accuracy
- B) Since the coefficients are interpretable, the effect of risk factors on premium can be explained transparently ✔
- C) Because it works without requiring data
- D) Because the legislation only allows GLM
Description: GLM transparently shows the impact of each risk factor on the premium because its coefficients are interpretable; This is a critical advantage in terms of legislation, auditing and job description.
6. What does the qx value on a mortality table indicate?
- A) Probability of a person aged x dying within a year ✔
- B) Total premium paid up to age x
- C) Present value of x-year annuity
- D) Technical interest rate
Explanation: qx is the probability that a person who has reached age x will die within a year. It is the basic input of life insurance and retirement accounts; The probability of survival is found by px = 1 - qx.
7. How does increasing the technical interest (discount) rate in an annuity account affect the present value of future payments?
- A) Increases present value
- B) It does not affect at all
- C) Decreases the present value ✔
- D) It only changes mortality
Explanation: As the discount rate increases, the present value of future payments decreases because the time value of money is more strongly applied. Therefore, the interest assumption directly affects the liability value and is chosen carefully.
8. What does "data leakage" mean in feature engineering?
- A) Storing data without encryption
- B) Filling in missing values with the average
- C) Not backing up data
- D) Information that is not actually available at the time of prediction is mixed into the model and misleadingly inflates the performance ✔
Explanation: Data leakage is the introduction of information into the model that does not actually exist at the time of prediction (for example, a variable containing the future). This artificially inflates the test performance and causes the model to fail in the real world.
9. What does a Monte Carlo simulation essentially provide to an actuary?
- A) Reduces the result to a single and precise number
- B) It allows to see the probability distribution and tail risk of the outcome by generating a large number of random scenarios ✔
- C) It gives accurate predictions without the need for historical data.
- D) Writes the legislation text automatically
Explanation: Monte Carlo repeats random scenarios a large number of times so that the outcome is viewed as a probability distribution (e.g. tail risk, 99.5% level) rather than a single point. It is a powerful way to quantify uncertainty.
10. What does SCR (Solvency Capital Requirement) mean within the framework of Solvency II?
- A) Total annual premium income
- B) Total of paid damages
- C) Capital requirement to be held to cover one-year losses at a 99.5% confidence level ✔
- D) Reinsurance commission rate
Explanation: SCR is the capital requirement that the company must maintain in order to meet its obligations within one year at a 99.5% confidence level (in a worst case scenario that will occur approximately once in 200 years). It is a key measure of an insurer's financial resilience.
11. How does "indirect discrimination" occur in a pricing model produced by artificial intelligence?
- A) A proxy variable associated with a prohibited variable indirectly carries the same discriminatory effect ✔
- B) The model uses too little data
- C) Setting premiums high
- D) Choosing low technical interest
Explanation: Even if a forbidden variable (e.g. ethnicity) is removed from the model, a strongly related surrogate variable (e.g. neighbourhood/zip code) may indirectly carry the same discriminatory effect. Therefore, the model needs to be inspected not only for explicit but also for indirect discrimination.
12. What is the basic privacy rule when analyzing policy and claims data with a publicly available artificial intelligence tool?
- A) If the data is small, there is no need for anonymization
- B) Just changing the names is sufficient in all cases
- C) If the data is encrypted, it can be uploaded to any vehicle
- D) Anonymize personal/private data, do not share unnecessary data and choose a secure/institutional tool ✔
Explanation: Personal and special data (TR ID, name, diagnosis, protocol) should be anonymized before sharing; If possible, corporate tools with data processing contracts that do not use data in model training should be preferred. KVKK requires this.
13. Why does an actuary comparing the chain ladder with the Bornhuetter-Ferguson (BF) method often prefer BF for immature (new) damage periods?
- A) Because BF is mandatory by law
- B) Because BF does not require data
- C) In new periods, chain ladder factors become very volatile; BF gives more stable prediction using expected damage rate ✔
- D) Because the chain ladder only works in life insurance
Explanation: Since very little damage occurs in new periods, chain ladder development factors become extremely sensitive and volatile to small fluctuations. BF smoothes this volatility using expected loss rate (a priori) information and gives a more stable forecast.
14. What is the main problem if an actuary directly signs the AI-generated reserve account and sends it to the regulator?
- A) Responsibility and final approval lies with the actuary; Signing unverified printout does not eliminate professional and legal risk ✔
- B) Artificial intelligence output is always inaccurate and should not be used
- C) The regulator has completely banned the use of artificial intelligence
- D) Artificial intelligence output does not require signature
Description: The AI output is a draft; Responsibility for method selection, assumptions and final figure lies with the signing actuary. Signing unverified output is as risky as publishing an unsigned report and does not eliminate professional/legal liability.