ASTAM Parametric Models

The Spring 2026 ASTAM syllabus gives parametric models a 14-24% weight and explicitly includes maximum likelihood estimation, estimator variance, normal and non-normal confidence intervals, delta method, Bayesian estimation, goodness-of-fit testing, likelihood ratio tests, and score-based criteria such as AIC, BIC, and SBC.

That makes this one of the most statistics-heavy parts of ASTAM and one of the most reusable outside the exam.

This block is about turning a family of models into a defendable choice. You estimate parameters, measure uncertainty, compare candidate fits, and decide whether the selected model is good enough for the actuarial task.

It is the cleanest place where ASTAM stops feeling like short-term insurance arithmetic and starts feeling like applied statistical inference.

Know how to set up a likelihood and why the log-likelihood is usually the easier object to optimize. Know what the delta method is doing conceptually. Know the difference between estimation, interval construction, goodness-of-fit, and model comparison.

You should also know when the exam is asking for a modeling judgment instead of a formula dump. AIC, BIC, graphical checks, Kolmogorov-Smirnov, chi-square, and LRT are not there to be recited mechanically. They are there to help justify a model choice.

ASTAM uses parametric models because short-term actuarial work is full of severity, frequency, and aggregate distributions that need estimating and defending. The syllabus explicitly extends beyond MLE into Bayesian estimation and model-selection logic, which means the exam expects more than one estimation vocabulary.

This is also one of the strongest bridges into SRM, PA, and broader data-science work because the core habits are the same: fit, check, compare, and explain.

Start with maximum likelihood estimation and interval ideas first, because the rest of the topic depends on understanding what the fitted model is. Then study the goodness-of-fit and comparison tools as answers to a practical question: why should anyone believe this distribution choice?

For written answers, practice saying what a statistic is telling you, not just calculating it. Candidates often know the formula for AIC or LRT and still lose points because they never convert the result into a modeling recommendation.