Insurance Research & Evidence-Based Practice 5 — Questions and Answers
Question 1: A property insurer wants to quantify the uncertainty in its catastrophe model output. Which measure best captures the range of possible outcomes at the tail of the loss distribution?
- Mean annual loss
- Probable Maximum Loss (PML) at the 99th percentile (Correct answer)
- Loss ratio
- Net premium written
Correct answer: Probable Maximum Loss (PML) at the 99th percentile
PML at a high percentile (e.g., 99th or 250-year return period) quantifies extreme tail losses that could threaten insurer solvency.
Question 2: When conducting a regression analysis on insurance claim data, multicollinearity is a problem because it:
- Increases the sample size requirements for the study
- Makes it difficult to isolate the independent effect of each predictor variable (Correct answer)
- Causes the model to underfit the training data
- Violates policyholder privacy regulations
Correct answer: Makes it difficult to isolate the independent effect of each predictor variable
Multicollinearity inflates standard errors and makes coefficient estimates unstable, preventing reliable identification of each variable's unique contribution.
Question 3: In evidence-based underwriting, which type of evidence is generally considered the weakest in the hierarchy of evidence?
- Randomized controlled trials with large samples
- Systematic reviews of multiple high-quality studies
- Expert opinion and anecdotal case reports (Correct answer)
- Prospective cohort studies with control groups
Correct answer: Expert opinion and anecdotal case reports
Expert opinion and anecdotal evidence sit at the bottom of the evidence hierarchy because they are most susceptible to bias and are not systematically derived.
Question 4: What is the 'base rate' problem in insurance risk communication, and why does it matter?
- Confusion between the base premium and the final charged rate
- People tend to ignore general probability information when given specific case details, leading to poor risk perception (Correct answer)
- Regulators require a base rate filing before any deviation is allowed
- Agents set rates based on their base territory without adjusting for individual risk
Correct answer: People tend to ignore general probability information when given specific case details, leading to poor risk perception
The base rate neglect bias causes individuals to underweight statistical probability when vivid or specific case information is available, distorting risk decisions.
Question 5: An insurer conducts an A/B test comparing two renewal premium increase messages. The result shows a statistically significant difference with p=0.04. What additional information is most critical before implementing the winning message?
- The number of agents who prefer the new message
- The practical effect size and business significance of the difference (Correct answer)
- Whether the test was approved by the state insurance department
- The font size used in the message
Correct answer: The practical effect size and business significance of the difference
Statistical significance alone does not indicate whether the effect is large enough to justify a business decision; effect size and practical significance must also be evaluated.
Question 6: What is 'triangulation' in the context of insurance research methods?
- Using three separate actuaries to independently price the same risk
- Combining multiple data sources or research methods to validate findings (Correct answer)
- A geometric approach to mapping catastrophe exposure zones
- Applying three deductible options to the same policy
Correct answer: Combining multiple data sources or research methods to validate findings
Triangulation strengthens research credibility by cross-checking findings from different sources, methods, or investigators to reduce the impact of any single source's limitations.
Question 7: Which statistical concept describes the minimum sample size an insurance researcher needs to detect a real effect with a given probability?
- Confidence level
- Statistical power (Correct answer)
- Standard error
- Coefficient of variation
Correct answer: Statistical power
Statistical power (commonly set at 80%) is the probability of correctly rejecting a false null hypothesis; it determines the sample size needed for reliable detection of true effects.
Insurance Research & Evidence-Based Practice 5
A property insurer wants to quantify the uncertainty in its catastrophe model output.
Which measure best captures the range of possible outcomes at the tail of the loss distribution?