Insurance Research & Evidence-Based Practice 3 — Questions and Answers
Question 1: In predictive modeling for insurance, what is 'overfitting'?
- Setting premiums too high relative to actual risk
- A model that performs well on training data but poorly on new data (Correct answer)
- Including too many perils in a single policy
- A reinsurance structure that exceeds treaty limits
Correct answer: A model that performs well on training data but poorly on new data
Overfitting occurs when a model captures noise in training data rather than true patterns, reducing its predictive power on out-of-sample data.
Question 2: Which data source would an actuary most likely use for credible industry-wide loss frequency benchmarks in the US property market?
- Individual agent production reports
- ISO (Insurance Services Office) statistical databases (Correct answer)
- State DMV driving records only
- Social media sentiment analysis
Correct answer: ISO (Insurance Services Office) statistical databases
ISO collects and analyzes loss data from many insurers to produce industry-standard actuarial benchmarks used in ratemaking.
Question 3: What is the difference between 'incurred losses' and 'paid losses' in insurance accounting?
- Incurred losses include reserves for unpaid claims; paid losses are only amounts disbursed to date (Correct answer)
- Paid losses are larger because they include interest penalties
- Incurred losses apply only to liability lines; paid losses apply to property
- There is no practical difference in GAAP reporting
Correct answer: Incurred losses include reserves for unpaid claims; paid losses are only amounts disbursed to date
Incurred losses equal paid losses plus the outstanding reserve for claims not yet fully settled, giving a complete picture of total expected cost.
Question 4: A researcher uses a Generalized Linear Model (GLM) for auto insurance ratemaking. What is the primary advantage of GLMs over ordinary linear regression?
- GLMs are faster to compute on legacy systems
- GLMs allow non-normal error distributions and link functions appropriate for insurance data (Correct answer)
- GLMs eliminate the need for credibility weighting
- GLMs automatically adjust for inflation trends
Correct answer: GLMs allow non-normal error distributions and link functions appropriate for insurance data
GLMs extend linear regression by accommodating non-normal distributions (e.g., Poisson for claim counts, Gamma for severity) that better fit insurance data.
Question 5: What does 'external validity' mean in the context of an insurance research study?
- The study was reviewed by an external audit firm
- The extent to which findings can be generalized beyond the study sample (Correct answer)
- The accuracy of financial statements certified by an outside actuary
- The validity of claims submitted by third parties
Correct answer: The extent to which findings can be generalized beyond the study sample
External validity refers to whether a study's results can be applied to other populations, settings, or time periods outside the original research context.
Question 6: In insurance fraud research, a 'propensity score' is used to:
- Estimate the likelihood that a claim is fraudulent based on historical patterns (Correct answer)
- Calculate the premium propensity of new applicants
- Measure the score assigned to an agent's book of business
- Determine the probability of a catastrophe event
Correct answer: Estimate the likelihood that a claim is fraudulent based on historical patterns
Propensity scores in fraud analytics estimate the probability that a claim has characteristics associated with fraudulent activity, enabling targeted investigation.
Question 7: Which type of validity ensures that a survey used to measure policyholder satisfaction actually measures satisfaction and not something else?
- Concurrent validity
- Construct validity (Correct answer)
- Predictive validity
- Face validity
Correct answer: Construct validity
Construct validity confirms that a measurement instrument truly captures the theoretical construct it is designed to measure, such as satisfaction.
Insurance Research & Evidence-Based Practice 3
In predictive modeling for insurance, what is 'overfitting'?