Study guide
9.4 The PCPA Technical Report: Structure & Rubric
CAS Property & Casualty Predictive Analytics PCPA
Track your progress as you read
Finished sections get a check mark.
1 Chapter 1: PCPA Exam & Project Orientation
1.1 PCPA Exam Format, Scoring & Registration 1.2 The PCPA Project: Windows, Deliverables & Grading
2 Chapter 2: Dealing with Data — Sourcing, Importing & Assessing
2.1 Requirements Gathering & Assessing Data Relevance 2.2 Importing, Joining & Reshaping Data Sets 2.3 Exploratory Data Analysis & Evaluating a Data Set 2.4 Data Quality Under ASOP 23 & Modeling Under ASOP 56
3 Chapter 3: Dealing with Data — Conditioning P&C Data for a GLM
3.1 Exposure Bases, Weights & Data Grain 3.2 Missing Data Mechanisms & Imputation Strategies 3.3 Outlier Diagnostics, Trimming & Winsorization 3.4 Variable Transformation & Scaling
4 Chapter 4: Model Diagnostics & Selection — Building a GLM
4.1 Exponential Dispersion Family & Link Functions 4.2 Poisson Regression & Exposure Offsets 4.3 Overdispersion, Quasi-Poisson & Negative Binomial 4.4 Gamma & Inverse Gaussian Models for Claim Severity
5 Chapter 5: Model Diagnostics & Selection — Model Form & Predictors
5.1 Tweedie Compound Poisson-Gamma Modeling 5.2 Binomial & Logistic GLMs for Binary Outcomes 5.3 Categorical Predictors, Base Levels & High-Cardinality Grouping 5.4 Interaction Terms & Non-Linear Effects 5.5 Interpreting Coefficients & Multiplicative Relativities
6 Chapter 6: Model Diagnostics & Selection — Evaluating and Improving the Fit
6.1 Deviance, Likelihood Ratio Tests & Residual Diagnostics 6.2 Stepwise Selection & Information Criteria Penalties 6.3 Multicollinearity, Correlated Predictors & Aliasing 6.4 Under-fitting, Over-fitting & Cross-Validation
7 Chapter 7: Model Diagnostics & Selection — Validation & Model Choice
7.1 Actual vs. Predicted, Calibration & Loss Ratio Charts 7.2 Simple Quantile Plots & Measuring Lift 7.3 Double Lift Charts & Competing Model Evaluation 7.4 Gini Index, Lorenz Curves & ROC Curves
8 Chapter 8: Model Interpretation & Presentation — Tables, Charts & Exhibits
8.1 Designing Tables & Statistical Exhibits 8.2 Chart Selection & Graphical Perception 8.3 How Charts Mislead 8.4 One-Way, Two-Way & Partial Dependence Exhibits
9 Chapter 9: Model Interpretation & Presentation — Communicating the Analysis
9.1 Communicating Methodology & Modeling Decisions 9.2 Explaining Why a Variable Was or Was Not Included 9.3 Communicating to Non-Technical Audiences 9.4 The PCPA Technical Report: Structure & Rubric
9.4 The PCPA Technical Report: Structure & Rubric
Key Takeaways Explain This Topic AI
The PCPA Project technical report is capped at 1,250 words inclusive of the report and its appendices, and exceeding the cap is an automatic failed attempt.
Twelve published performance criteria map to the three domains; with roughly 100 words per criterion, every paragraph must be doing rubric work.
The three things the report must describe are exploratory data analysis and data issues, the GLM build and evaluation, and the interpretation of findings for the business decision.
Up to five tables or supporting graphics may be submitted as appendices, and they count toward the word total.
Code in R, Python or SAS must accompany the report and be able to reproduce the data preparation and modelling results, though it is not itself scored.
Last updated: September 2026
The Constraint That Shapes Everything
The PCPA Project technical report may be no longer than 1,250 words, inclusive of the report and its appendices . CAS states plainly that reports exceeding the word count will result in a failed attempt . There is no truncation, no revision cycle, and no discretion.
Two other limits carry the same consequence: submitting more supporting materials than allowed , and not following the file-format instructions . Each of these fails a submission regardless of how good the model is.
So the report is an exercise in compression. Twelve published performance criteria and 1,250 words gives roughly 100 words per criterion . Nothing that does not earn rubric credit can afford to be in the document.
What Must Be Described
CAS specifies three things the technical summary must cover:
How you conducted exploratory data analysis , identified and managed issues within the data — variables needing transformation, outliers, data errors, missing data — and identified appropriate target and predictor variables.
How you created a GLM (Gamma, link, polynomial, Poisson, binomial, log-normal, Tweedie), evaluated model performance , and used the evaluation results to improve it.
How you interpreted the findings from a technical perspective and for the business decision, clearly identifying the rationale for creating the model.
Alongside the report you submit code in R, Python or SAS capable of reproducing your data preparation and modelling results, up to five tables or supporting graphics , an attestation of independent work, and answers to questions about your analysis posed at submission. The code is not scored, but graders may run it and may check it for originality.
A Word Budget That Covers the Rubric
- Report section Words Rubric criteria served
- Business question and rationale for the model 80 C-3 (persuasive argument, rationale for creation)
- Data, basis, and EDA findings 150 A-3, A-4
- Data treatments: transformations, outliers, missing data 140 A-3, A-4
- Model specification: target, distribution, link, offset 90 B-1, C-2 (methods used and why)
- Iterative build and diagnostics 210 B-2 (iterative check, diagnostics, coefficient in/out)
- Variable inclusion and exclusion 120 C-2 (why a variable was or was not included)
- Model selection and out-of-sample performance 160 B-1, B-2 (assessment data set, model selected and why)
- Technical output and interpretation 110 B-2 (provides technical output), C-2
- Business implications and recommendation 100 C-3
- Limitations 60 ASOP 23-style disclosure; supports C-2
- Exhibit titles and table labels 30 C-1 (justified presentation, labelling)
- Total 1,250
The budget is a planning device, not a rule — but writing without one is how candidates arrive at 1,600 words with the interpretation section still unwritten.
Section-by-Section Guidance
Business question and rationale. Two sentences. What was asked, and why a model is the right instrument for answering it. The rubric explicitly wants the rationale for creating the model, so do not skip straight to the data.
Data and EDA. State the basis — period, loss basis, exposure definition, evaluation date — then report what the exploration found. Name specific anomalies with counts: "1,860 records carried a sentinel vehicle age of 999; these were recoded as missing." Specificity is cheaper in words than generality and scores better.
Data treatments. For each transformation and each anomaly, say what you did and why . Criterion A-3 asks how variables were transformed and why ; the second half is where candidates lose the mark.
Model specification. Target, distribution, link, offset or weight, and a one-line justification for the distribution choice tied to what EDA showed.
Iterative build. The largest block, because it serves the most criteria. Use the observation-diagnostic-action-result chain, three or four times. Name the diagnostics — AIC, BIC, deviance, VIF, cross-validation, partial residual plots — because the rubric asks which diagnostics were used and why.
Variable inclusion. A compact table is far more word-efficient than prose here.
Model selection and out-of-sample performance. State which candidate models were compared, on what evidence, and how the selected model performed on data it never saw. Criterion B-1 asks specifically that the model perform reasonably well on an assessment data set.
Technical output and interpretation. Coefficients or relativities, with a sentence saying what they mean.
Business implications. The finding in business units, the recommendation, and what it is worth.
Limitations. Two or three sentences. Thin segments, unresolved data issues, the range over which the model should be applied.
Failures That Have Nothing to Do With Modelling
Over the word count , usually because appendix text was not counted. Count the whole submission.
Too many exhibits. Five tables or graphics total.
Wrong file format or a missing required file.
Code that does not reproduce the reported results. Graders may run it.
Criteria never addressed. A report that never says why a variable was excluded forfeits that criterion even if the modelling was flawless.
Method described without justification. "Missing values were imputed with the median" earns less than "Median imputation was used for the 4% of records missing vehicle age, because the distribution is skewed and the affected records were spread evenly across territories."
Warning
Write the report against the rubric, not against the analysis. The natural instinct is to narrate what you did in the order you did it. The criteria are the grading instrument, so check off each of the twelve before submitting and cut anything that does not serve one of them.
Professional Standards in the Background
The project is a professional work product as well as a graded assessment. The habits the standards require map directly onto the rubric: ASOP 23 asks you to review data for appropriateness, reasonableness and comprehensiveness and to disclose limitations and reliance — which is the EDA and limitations sections. ASOP 56 asks you to understand a model's structure, assumptions and limitations and to validate it against its intended purpose — which is the specification, build and out-of-sample sections. A report that satisfies the rubric will generally satisfy the standards, and vice versa.