SOA PA-Predictive Analytics Exam | Practice Questions
And Answers 2026/2027 | Complete Questions And
Answers (Verified Answers) | Exam Prep |
Comprehensive Exam Guide 2026&2027
1. An insurer wants to estimate which policyholders are most likely to cancel their policies
during the next six months so that retention resources can be targeted. Which type of
analytics is most directly applicable?
A. Descriptive analytics
B. Predictive analytics
C. Prescriptive accounting
D. Exploratory reporting
Answer: B
Predictive analytics is appropriate because the objective is to estimate the likelihood of a
future event, policy cancellation, using available historical and current information.
2. An actuarial team reports that claim frequency increased from 8% to 10% last year.
Management asks what factors contributed to the increase, without requesting a
prediction. Which analytical objective is most appropriate initially?
A. Descriptive analytics
B. Prescriptive analytics
C. Predictive analytics
D. Reinforcement learning
Answer: A
Descriptive analytics summarizes what has happened. Since management is initially asking
about the observed historical change rather than a future outcome or recommended action,
descriptive analysis is the appropriate starting point.
3. A model performs extremely well on its training data but substantially worse on new
observations. Which problem is most likely present?
A. Underfitting
B. Measurement invariance
C. Overfitting
D. Random sampling
Answer: C
,Overfitting occurs when a model captures noise or idiosyncrasies in the training data,
producing excellent training performance but weaker performance on unseen observations.
4. An analyst increases the complexity of a model by adding many interaction terms.
Training error decreases, but validation error begins to increase. What does this pattern
most strongly indicate?
A. Reduced variance
B. Improved generalization
C. Elimination of bias
D. A worsening bias-variance trade-off
Answer: D
Adding complexity can reduce bias while increasing variance. When validation error rises as
complexity increases, the model is moving toward excessive variance and poorer
generalization.
5. A company asks, “Why are our customers unhappy?” Before selecting a predictive
model, what is the most important next step?
A. Immediately fit a random forest
B. Translate the question into a measurable analytical objective
C. Remove all categorical variables
D. Standardize every variable
Answer: B
A predictive analytics project should begin by defining a specific, measurable business
problem. A vague question must be translated into an outcome that can be analyzed with
appropriate data and methods.
6. A health insurer wants to predict whether a member will submit at least one claim
during the next policy year. What is the natural response variable type?
A. Binary
B. Continuous
C. Ordinal with unlimited levels
D. Textual
Answer: A
The outcome has two possible states: at least one claim or no claim. This makes it a binary
response suitable for classification methods such as logistic regression.
,7. An insurer wants to estimate the exact dollar amount of a future claim for each
policyholder. Which response type is most appropriate?
A. Binary
B. Nominal
C. Continuous
D. Count-only categorical
Answer: C
Claim amount is measured numerically and can take many possible values, making it a
continuous response for modeling purposes.
8. A data scientist is asked to predict customer churn. The available data contains
information collected only after customers have already cancelled. What is the primary
concern?
A. Excessive regularization
B. Data leakage
C. Underdispersion
D. Multicollinearity only
Answer: B
Using information that would not have been available at prediction time introduces data
leakage. It can make a model appear highly accurate during development while failing in
actual deployment.
9. A model is intended to identify high-risk policyholders before renewal decisions are
made. Which data-design principle is particularly important?
A. Use only information available before the prediction point
B. Use all information regardless of collection date
C. Exclude the target variable from historical records
D. Randomly change the prediction date for each observation
Answer: A
Predictive features must be available at the time the prediction would actually be made.
Otherwise, the model can rely on future information that would not be available
operationally.
10. A model's objective is to predict whether an applicant will default. The analyst
evaluates it using the same observations used to estimate all model parameters. What is the
main weakness of this approach?
, A. It guarantees unbiased estimates
B. It measures only deployment performance
C. It may provide an overly optimistic assessment
D. It eliminates sampling variability
Answer: C
Evaluating a model on the same data used for fitting can underestimate its true prediction
error. A separate validation or test procedure provides a more realistic estimate of
performance on new data.
11. A company has millions of transaction records but only wants to estimate average
monthly revenue by region. What consideration should be made before building a complex
predictive model?
A. Whether predictive modeling is necessary for the business objective
B. Whether every record can be converted into text
C. Whether clustering must be performed
D. Whether the response should be randomized
Answer: A
Model complexity should be justified by the business problem. If a straightforward descriptive
analysis answers the business question, a complex predictive model may add unnecessary
cost and implementation risk.
12. An analyst notices that a variable has a long right tail. Which visualization is
particularly useful for examining its distribution and potential extreme values?
A. Boxplot
B. Pie chart
C. Stacked bar chart only
D. Confusion matrix
Answer: A
A boxplot summarizes the center, spread, and potential outliers of a numerical variable and is
particularly useful for detecting skewness and extreme observations.
13. Which variable is categorical rather than numerical?
A. Annual premium
B. Number of claims
C. Vehicle age
D. Vehicle type
And Answers 2026/2027 | Complete Questions And
Answers (Verified Answers) | Exam Prep |
Comprehensive Exam Guide 2026&2027
1. An insurer wants to estimate which policyholders are most likely to cancel their policies
during the next six months so that retention resources can be targeted. Which type of
analytics is most directly applicable?
A. Descriptive analytics
B. Predictive analytics
C. Prescriptive accounting
D. Exploratory reporting
Answer: B
Predictive analytics is appropriate because the objective is to estimate the likelihood of a
future event, policy cancellation, using available historical and current information.
2. An actuarial team reports that claim frequency increased from 8% to 10% last year.
Management asks what factors contributed to the increase, without requesting a
prediction. Which analytical objective is most appropriate initially?
A. Descriptive analytics
B. Prescriptive analytics
C. Predictive analytics
D. Reinforcement learning
Answer: A
Descriptive analytics summarizes what has happened. Since management is initially asking
about the observed historical change rather than a future outcome or recommended action,
descriptive analysis is the appropriate starting point.
3. A model performs extremely well on its training data but substantially worse on new
observations. Which problem is most likely present?
A. Underfitting
B. Measurement invariance
C. Overfitting
D. Random sampling
Answer: C
,Overfitting occurs when a model captures noise or idiosyncrasies in the training data,
producing excellent training performance but weaker performance on unseen observations.
4. An analyst increases the complexity of a model by adding many interaction terms.
Training error decreases, but validation error begins to increase. What does this pattern
most strongly indicate?
A. Reduced variance
B. Improved generalization
C. Elimination of bias
D. A worsening bias-variance trade-off
Answer: D
Adding complexity can reduce bias while increasing variance. When validation error rises as
complexity increases, the model is moving toward excessive variance and poorer
generalization.
5. A company asks, “Why are our customers unhappy?” Before selecting a predictive
model, what is the most important next step?
A. Immediately fit a random forest
B. Translate the question into a measurable analytical objective
C. Remove all categorical variables
D. Standardize every variable
Answer: B
A predictive analytics project should begin by defining a specific, measurable business
problem. A vague question must be translated into an outcome that can be analyzed with
appropriate data and methods.
6. A health insurer wants to predict whether a member will submit at least one claim
during the next policy year. What is the natural response variable type?
A. Binary
B. Continuous
C. Ordinal with unlimited levels
D. Textual
Answer: A
The outcome has two possible states: at least one claim or no claim. This makes it a binary
response suitable for classification methods such as logistic regression.
,7. An insurer wants to estimate the exact dollar amount of a future claim for each
policyholder. Which response type is most appropriate?
A. Binary
B. Nominal
C. Continuous
D. Count-only categorical
Answer: C
Claim amount is measured numerically and can take many possible values, making it a
continuous response for modeling purposes.
8. A data scientist is asked to predict customer churn. The available data contains
information collected only after customers have already cancelled. What is the primary
concern?
A. Excessive regularization
B. Data leakage
C. Underdispersion
D. Multicollinearity only
Answer: B
Using information that would not have been available at prediction time introduces data
leakage. It can make a model appear highly accurate during development while failing in
actual deployment.
9. A model is intended to identify high-risk policyholders before renewal decisions are
made. Which data-design principle is particularly important?
A. Use only information available before the prediction point
B. Use all information regardless of collection date
C. Exclude the target variable from historical records
D. Randomly change the prediction date for each observation
Answer: A
Predictive features must be available at the time the prediction would actually be made.
Otherwise, the model can rely on future information that would not be available
operationally.
10. A model's objective is to predict whether an applicant will default. The analyst
evaluates it using the same observations used to estimate all model parameters. What is the
main weakness of this approach?
, A. It guarantees unbiased estimates
B. It measures only deployment performance
C. It may provide an overly optimistic assessment
D. It eliminates sampling variability
Answer: C
Evaluating a model on the same data used for fitting can underestimate its true prediction
error. A separate validation or test procedure provides a more realistic estimate of
performance on new data.
11. A company has millions of transaction records but only wants to estimate average
monthly revenue by region. What consideration should be made before building a complex
predictive model?
A. Whether predictive modeling is necessary for the business objective
B. Whether every record can be converted into text
C. Whether clustering must be performed
D. Whether the response should be randomized
Answer: A
Model complexity should be justified by the business problem. If a straightforward descriptive
analysis answers the business question, a complex predictive model may add unnecessary
cost and implementation risk.
12. An analyst notices that a variable has a long right tail. Which visualization is
particularly useful for examining its distribution and potential extreme values?
A. Boxplot
B. Pie chart
C. Stacked bar chart only
D. Confusion matrix
Answer: A
A boxplot summarizes the center, spread, and potential outliers of a numerical variable and is
particularly useful for detecting skewness and extreme observations.
13. Which variable is categorical rather than numerical?
A. Annual premium
B. Number of claims
C. Vehicle age
D. Vehicle type
