CSPA
CSPA - iCAS Certified Specialist in Predictive Analytics
More study resources
Cheat sheet
Keep the key ideas within reach.
ExploreKey terms
Get comfortable with the vocabulary.
ExploreStudy plan
Give each week a clear focus.
ExploreCommon mistakes
Understand the traps and how to avoid them.
ExploreCase studies
Put your knowledge to work in context.
ExploreExam-day guide
Prepare for the day ahead.
ExploreAbout this exam
The Certified Specialist in Predictive Analytics (CSPA) is an iCAS credential involving multiple learning and assessment components, practical analytical work, and professional responsibilities. It should not be treated as one uniform multiple-choice exam, and it is not interchangeable with the separate CAS PCPA assessment. This original practice resource supports the shared foundations described in the iCAS curriculum; it does not replace the official component guides, project instructions, ethics requirements, or current credential rules.
The six editorial study domains connect property-casualty insurance quantities with data preparation, generalized linear models, nonlinear methods, honest validation, and responsible insurance applications. Calculations include frequency and severity, earned premium under supplied assumptions, log-link and logistic interpretation, offsets, tree impurity, ensemble variance, confusion-matrix metrics, calibration, and decision costs. Other questions examine data availability at prediction time, group and temporal separation, target maturity, privacy, and reproducible delivery.
Use the questions to explain why an answer follows the stated data and assumptions, not merely to memorize a key. A strong learner can reproduce a calculation, identify the unit of observation, and distinguish predictive evidence from causal or regulatory conclusions. The official credential also requires capabilities that a multiple-choice bank cannot demonstrate, including a defensible analysis, documented code and data work, and clear communication.
Administrative facts are intentionally null rather than presented as a universal fee, duration, passing score, or question count across components. The weights here are editorial study allocations, not an official iCAS blueprint. Confirm current official requirements before registration or project submission.
What you'll study
About CSPA preparation
Is CSPA one fixed multiple-choice examination?
No. The iCAS pathway combines component learning and assessments with practical project and professional requirements. This resource is a foundations practice bank, not a substitute for the current official component requirements or an assertion that every component has identical administration.
Is this the CAS PCPA assessment?
No. CSPA is the iCAS Certified Specialist in Predictive Analytics credential. The separate CAS PCPA assessment should not be treated as an interchangeable title, and candidates should verify which pathway their own requirement actually names.
Why are the profile administration fields null?
A single fee, duration, passing score, or question count would be misleading across a multi-component credential without specifically supported current facts. The practice-bank size is an editorial content quantity, not an official credential requirement.
What insurance background should I practice?
Be comfortable with exposure, claim frequency, positive-claim severity, pure premium, premiums, losses, and simple reserving quantities. Always state coverage, dates, expense treatment, and population definitions, because superficially similar ratios can represent different targets.
How should I approach the model calculations?
Write the link or probability formula, substitute the stated values, and check units and scale. Distinguish log effects from mean ratios, odds from probabilities, count offsets from free coefficients, and variance assumptions from observed certainty.
How do I avoid an unrealistically good validation result?
Split to match deployment, preserve related groups when appropriate, respect prediction-time availability, and learn transformations using training data only. Keep tuning separate from final evaluation and account for insurance outcomes that are not yet fully developed.
Does a good discrimination score establish an appropriate insurance decision?
No. Ranking, probability calibration, review capacity, expected costs, fairness concerns, legal permissions, and operational constraints answer different questions. A useful analysis connects the metric to the actual decision instead of treating one score as universal approval.
How should I prepare for practical analytical work?
Build a reproducible pipeline with documented data versions, target definitions, feature timing, splits, code, and assumptions. Communicate limitations and recommendations accurately, protect information appropriately, and follow the actual project and professional-conduct instructions.