200+ Free SOA Exam PA Practice Questions

Pass your SOA Exam PA Predictive Analytics exam on the first try — instant access, no signup required.

✓ No registration✓ No credit card✓ No hidden fees✓ Start practicing immediately

200+ Questions

100% Free

1 / 200

Question 1

Score: 0/0

A life insurer wants to identify policyholders most likely to lapse in the next 90 days so the retention team can contact them. Which type of analytics project is this?

Predictive analytics

Prescriptive analytics

Descriptive analytics

Diagnostic analytics

to track

2026 Statistics

Key Facts: SOA Exam PA Exam

3.5 hrs

Time Limit

SOA syllabus

$1,234

Standard Fee

SOA fee table

6-10

Passing Grades

SOA 0-10 scale

Weighted Domains

SOA syllabus

12 tasks

Recent Public Form

October 2025 project statement

~6 weeks

Grade Turnaround Pilot

SOA exam page

As of March 11, 2026, SOA lists Exam PA at $1,234 and a 3.5-hour Prometric-delivered written-answer format. The current syllabus weights Generalized Linear Models most heavily at 25-35%, followed by Data Exploration and Visualization at 20-30%, with Problem Definition, Data Transformations and Unsupervised Learning, and Tree-Based Models each at 10-20%. SOA reports grades on a 0-10 scale with 6-10 passing, does not publish a fixed raw percentage cut score, and continues the faster-grading pilot that shortens grade turnaround from 11 weeks to about 6 weeks.

Sample SOA Exam PA Practice Questions

Try these sample questions to test your SOA Exam PA exam readiness. Each question includes a detailed explanation. Start the interactive quiz above for the full 200+ question experience with AI tutoring.

1A life insurer wants to identify policyholders most likely to lapse in the next 90 days so the retention team can contact them. Which type of analytics project is this?

A.Predictive analytics

B.Prescriptive analytics

C.Descriptive analytics

D.Diagnostic analytics

Explanation: The insurer is trying to estimate a future outcome, which makes this a predictive analytics task. Descriptive and diagnostic analytics explain what happened or why it happened, while prescriptive analytics recommends specific actions after prediction.

2Which problem statement is best framed for an Exam PA renewal model?

A.Use machine learning to improve insights from policy data.

B.Predict whether each auto policy will renew at its next expiration date using information available 30 days before renewal.

C.Build the most accurate model possible for underwriting.

D.Analyze all company data to find profitable customers.

Explanation: A strong business framing states the target, unit of observation, and time at which predictors are available. The other options are vague and do not define a precise prediction problem.

3An auto insurer wants to model annual claim frequency for pricing. Data are available at the policy, vehicle, and payment transaction levels. Which unit of observation is most appropriate?

A.One row per agency office

B.One row per reported claim

C.One row per policy-year

D.One row per payment transaction

Explanation: For annual claim frequency, the natural unit is the exposure period being priced, such as a policy-year. Transactions and claims do not represent the full exposure base needed to model claim counts.

4In a lapse model, which predictor should generally be treated as categorical rather than numeric?

A.Issue age in years

B.Policy tenure in months

C.Annual premium amount

D.Payment plan with levels monthly, quarterly, and annual

Explanation: Payment plan consists of unordered or nominal categories rather than a continuous numeric scale. Treating it as numeric would impose an artificial distance between categories that may not reflect the business meaning.

5A workers compensation insurer wants to predict claim severity at first notice of loss. Which variable should be excluded because it is likely data leakage?

A.Adjuster closing note sentiment

B.Injured worker age

C.Occupation class

D.Jurisdiction state

Explanation: Closing notes are created well after first notice of loss, so they contain future information unavailable when the prediction is made. Including them would overstate model performance and violate the intended decision timing.

6A claims department wants to predict the size of individual claims above the deductible. Which dataset structure is most appropriate?

A.All policies at renewal with one row per policy

B.Only reported claims with one row per claim

C.One row per insured household

D.Monthly aggregate totals by state

Explanation: If the response is the size of an individual claim, the unit should be the individual claim. Policy-level or monthly aggregate data would mix together multiple processes and make severity modeling less coherent.

7A very deep boosted model has much lower training error than a simple GLM, but its validation performance varies sharply across resamples. What is the main concern?

A.High bias from using a flexible model

B.The holdout set must contain leakage

C.High variance and likely overfitting

D.Underdispersion in the response

Explanation: Large swings in validation performance across samples indicate instability, which is a variance problem. The deep model is fitting training data well but may not generalize reliably to new business data.

8A health insurer wants to group employer groups into similar risk profiles without any predefined target variable. Which analytic approach is most appropriate?

A.Randomized causal inference

B.Time-series forecasting

C.Supervised classification

D.Unsupervised clustering

Explanation: Clustering is used when the goal is to discover natural groupings without a labeled response. Supervised methods require a known target, and forecasting or causal studies answer different business questions.

9An actuary wants to compare claim severity distributions across four territories, and the data are skewed with several extreme losses. Which visualization is most appropriate?

A.Side-by-side boxplots

B.Stacked area chart

C.Line chart of territory means

D.Pie chart of territory share

Explanation: Boxplots are well suited for comparing centers, spread, and outliers across multiple groups. They are more informative than pie or line charts for skewed continuous distributions.

10Homeowners claim severity is strongly right-skewed because a small number of catastrophe losses are very large. Which summary best describes a typical claim?

A.Mean and standard deviation

B.Median and interquartile range

C.Minimum and maximum only

D.Correlation and covariance

Explanation: For highly skewed data, the median and interquartile range are more resistant to extreme observations. The mean can be pulled upward substantially by a few very large claims.

About the SOA Exam PA Exam

The SOA Exam PA Predictive Analytics exam is a core ASA-pathway assessment that tests whether candidates can frame a business problem, explore and transform data, select and interpret predictive models, and communicate analytics recommendations under exam conditions.

Assessment

Written-answer task exam; the October 14, 2025 public project statement used 12 tasks totaling 70 points.

Time Limit

3 hours 30 minutes

Passing Score

Grade of 6 or higher on SOA's 0-10 scale

Exam Fee

$1,234 (Society of Actuaries (SOA) / Prometric)

SOA Exam PA Exam Content Outline

10-20%

Predictive Analytics Problem Definition

Translate business questions into predictive-analytics tasks, define scope, and weigh business impact, data availability, and implementation constraints.

20-30%

Data Exploration and Visualization

Work with variable types, data design, granularity, and effective visualizations for univariate and bivariate exploratory analysis.

10-20%

Data Transformations and Unsupervised Learning Techniques

Create useful features, apply data transformations, and use PCA, K-means, and hierarchical clustering where appropriate.

25-35%

Generalized Linear Models

Select and validate GLMs, use offsets and weights correctly, interpret coefficients and interactions, and manage diagnostics or regularization.

10-20%

Tree-Based Models

Build, prune, validate, and interpret classification or regression trees, bagging, random forests, and boosting approaches.

How to Pass the SOA Exam PA Exam

What You Need to Know

Passing score: Grade of 6 or higher on SOA's 0-10 scale
Assessment: Written-answer task exam; the October 14, 2025 public project statement used 12 tasks totaling 70 points.
Time limit: 3 hours 30 minutes
Exam fee: $1,234

Keys to Passing

Complete 500+ practice questions
Score 80%+ consistently before scheduling
Focus on highest-weighted sections
Use our AI tutor for tough concepts

SOA Exam PA Study Tips from Top Performers

1Practice turning vague business requests into a clear analytics objective, target variable, success metric, and implementation recommendation.

2Get fast at criticizing graphs: axis choices, aggregation mistakes, overplotting, misleading scales, and mismatched chart types are common PA-style tasks.

3Know when a transformation changes the modeling problem materially, especially logs, feature engineering, PCA, clustering, and encoding choices.

4Spend extra time on GLMs because that domain is the largest part of the blueprint and often combines selection, interpretation, diagnostics, and classification tradeoffs.

5Treat tree models as a workflow rather than isolated facts: splitting, cp or pruning, overfitting control, bagging vs. random forest, boosting, and interpretation should all connect.

Frequently Asked Questions

How many questions are on SOA Exam PA?

SOA does not present Exam PA as a fixed multiple-choice question count. It is a written-answer task exam, and the public October 14, 2025 project statement used 12 tasks worth 70 total points. The exact task mix can vary by administration, so it is better to prepare for a task-based case exam than for a fixed item count.

What is the passing score for SOA Exam PA?

SOA reports exam results on a 0-10 scale, and grades from 6 to 10 are passing. SOA does not publish one fixed raw percentage cut score for Exam PA because pass marks are set using its content-based grading process for each sitting.

Which topics matter most on Exam PA?

Generalized Linear Models is the heaviest section at 25-35% of the syllabus. Data Exploration and Visualization is next at 20-30%, while Problem Definition, Data Transformations and Unsupervised Learning, and Tree-Based Models each carry 10-20% and still matter materially.

What changed for SOA predictive analytics content in 2026?

The main 2026 program change is on the e-learning side: SOA moved the Predictive Analytics modules into PD Edge+ in January 2026 and removed the normal certificate final assessment path from the new format. SOA's FAQ says the last legacy certificate assessment will be offered in July 2026 for certain earlier purchasers, while the Exam PA syllabus content itself remains aligned to the same five topic groups.

Do I need SRM before I can take Exam PA?

Yes. SOA's Guide to SOA Exams states that Statistics for Risk Modeling (SRM) is a formal prerequisite for Exam PA. The syllabus also assumes prior knowledge from Exam P, VEE Mathematical Statistics, and Exam SRM.

What software is available during Exam PA?

SOA's syllabus says candidates test on Prometric computers with Microsoft Word and Microsoft Excel available. R and RStudio have not been available on the live exam since the April 2023 administration, so you should practice interpreting provided code and output rather than expecting to run your own code on exam day.