IAI Actuarial Core Principles · Risk Modelling and Survival Analysis
Introduction to Extreme Value Theory: GEV and GPD Explained
Extreme value theory models the largest or smallest values of a random variable, not its typical values. You use block maxima with the GEV distribution, or exceedances over a high threshold with the generalised Pareto distribution. Then you fit the model and estimate tail measures such as return levels and high quantiles.
What this chapter covers
Extreme value theory (EVT) studies the far tail of a distribution. Ordinary fitting is driven by the bulk of the data. EVT focuses on rare, large losses such as a flood, a market crash or a very large insurance claim. The chapter gives two main routes. The first takes the maximum of each block of data and models it with the generalised extreme value (GEV) distribution. The second keeps only the observations above a high threshold and models the excesses with the generalised Pareto distribution (GPD).
The key idea is a limit result. Under suitable conditions, the maximum of n independent, identically distributed variables, after suitable scaling and shifting, converges to a GEV distribution as n grows. The shape parameter decides the tail type: heavy-tailed, light-tailed with no finite upper end, or bounded above. You must be able to read the tail behaviour from the sign of the shape parameter.
This chapter connects to the rest of CS2 through risk modelling distributions, where heavy-tailed claim sizes appear, and through parameter estimation and model checking. It also links to your R work in Paper B, where you fit models, draw diagnostic plots and read off return levels. In the written paper, expect a mix of short derivations, interpretation of parameters and a tail risk estimate with stated assumptions.
Risk modelling distributions carries a 20% weighting in the 2026 CS2 syllabus, and EVT sits in the tail-modelling part of that theme. The chapter is compact and rule-based, so marks are easier to secure than in longer topics. Questions reward clear definitions, correct notation, careful parameter interpretation and sensible comments on assumptions. It also suits Paper B, where a fitted model and a return level are a natural computer task. Students who learn the two approaches side by side, and when each fits, pick up marks that others lose through vague explanations.
Introduction to extreme value theory: topics in the order to study them
- 1Introduction to Extreme Value TheoryStart here to learn why the tail needs its own models and what heavy tails and extremes mean in practice.
- 2Distribution of the Maximum and Block MaximaYou need the distribution of the maximum of n independent variables, and the idea of blocks, before the limit result makes sense.
- 3Generalised Extreme Value DistributionThis is the limit family for block maxima, and the shape parameter links directly to tail behaviour.
- 4Peaks over Threshold and Generalised Pareto DistributionThis second approach uses more of the data and builds on the same tail ideas, so compare it with block maxima.
- 5Fitting Extreme Value Models and Tail Risk EstimationStudy this last because it uses everything: fit the model, check it, then estimate return levels and high quantiles.
How to prepare Introduction to extreme value theory
Treat the chapter as one story: define extremes, model them in two ways, then fit and use the model. Build the theory first, then practise numbers and interpretation.
- Write the distribution of the maximum of n independent, identically distributed variables from first principles: P(Mₙ ≤ x) = [F(x)]ⁿ. Practise this with simple distributions until it is automatic.
- Learn the GEV family in standard notation. Be able to state what the location, scale and shape parameters do, and what each sign of the shape parameter says about the tail.
- Learn the GPD in the same way. Be clear on what the threshold is, what an excess is, and how the GPD parameters relate to the GEV shape.
- Make a comparison note for block maxima and peaks over threshold: how each uses data, what choice you must make (block size or threshold), and the trade-off in each choice.
- Practise fitting in R. Fit a GEV or GPD to a data set, read the parameter estimates, look at the diagnostic plots and compute a return level or high quantile.
- Do past-paper style questions. For each written answer, state the assumptions (independence, identical distribution, a high enough threshold) and comment on whether the result is reliable.
- Finish with timed MCQs on definitions and tail types, since these are quick marks.
Common mistakes in Introduction to extreme value theory
Using the ordinary fitted distribution to estimate extreme probabilities.
Fix: Say that the tail needs its own model, then use block maxima with the GEV or excesses with the GPD.
Getting the tail type wrong from the shape parameter.
Fix: Write the three cases side by side with their tail behaviour and test yourself on reading a sign into a tail description.
Confusing the distribution of the maximum with the distribution of one observation.
Fix: Always start a maximum question by writing the event that all n values are at most x, then use independence.
Choosing a threshold or block size without comment.
Fix: State the trade-off: a low threshold or small block adds bias, a high one or large block leaves too little data and raises variance.
Ignoring assumptions when giving a tail risk estimate.
Fix: Add a line on independence, identical distribution and the reliability of extrapolating beyond the observed data.
Mixing up the GEV and GPD roles.
Fix: Remember: GEV is for block maxima, GPD is for excesses over a threshold.
Last-day revision: Introduction to extreme value theory
- Extreme value theory models the tail, not the centre, of a distribution.
- For independent, identically distributed variables, P(Mₙ ≤ x) = [F(x)]ⁿ.
- Block maxima: split the data into blocks and model the maximum of each block.
- Under suitable conditions, the scaled and shifted maximum converges to a GEV distribution.
- The GEV has location, scale and shape parameters.
- The shape parameter's sign gives the tail type: positive is heavy-tailed, zero is light-tailed with no finite upper end, negative is bounded above.
- Peaks over threshold: keep observations above a high threshold and model the excesses.
- Excesses over a high threshold are modelled by the generalised Pareto distribution.
- Block size and threshold are the key choices: too low a threshold adds bias, too high leaves little data.
- Check the fit with diagnostic plots before you trust any tail estimate.
- A return level is a high quantile expressed in terms of an event expected once in a stated period.
- Always state your assumptions, especially independence and a high enough threshold.
Introduction to extreme value theory practice questions
- A GEV distribution has location mu = 50, scale sigma = 10 and shape xi = -0.5. Its distribution function is F(x) = exp(-[1 + xi(x-mu)/sigma]…
- Losses are independent and uniformly distributed on (0, 1) (in Rs crore). For a block of n = 5 losses, what is the expected value of the blo…
- A GPD with shape ξ and scale β is fitted to excesses over a threshold. For which value of ξ do the excesses have an infinite mean?
- A GEV distribution has shape parameter ξ = 0.4. Which statement about the tail of this distribution is correct?
- Block maxima of daily losses are modelled by a GEV distribution with shape ξ > 0 after suitable normalisation. Which statement is correct?
- A GEV distribution has μ = 50, σ = 10 and ξ = -0.5. What is the upper endpoint of its support?
- Fire losses for a portfolio are modelled as independent with a common exponential distribution with mean Rs 2 lakh. In a block of 3 claims, …
- A reinsurer uses annual block maxima of fire losses. Which is a recognised drawback of the block maxima method compared with peaks over thre…
Introduction to extreme value theory in other exams
The same ground in other exams, if you are preparing for more than one or want another angle on it.
Introduction to extreme value theory: frequently asked questions
What is the difference between block maxima and peaks over threshold?
Block maxima keeps one value, the maximum, from each block of data and fits a GEV distribution. Peaks over threshold keeps every observation above a high threshold and fits a GPD to the excesses. The second approach often uses more of the data, but you must choose the threshold carefully.
What does the shape parameter tell me in the GEV distribution?
It controls the tail. A positive value means a heavy tail, a zero value means a light tail with no finite upper end, and a negative value means the distribution is bounded above. Be ready to state this in words and link it to a real example.
Will extreme value theory appear in both Paper A and Paper B?
It can appear in either. Paper A tests definitions, derivations and interpretation, while Paper B, the computer-based exam, suits fitting a model and estimating a return level or quantile in R. Prepare both the written reasoning and the R steps.
How should I answer a tail risk estimation question?
State the model and its assumptions, give the fitted parameters, show the formula for the quantile or return level, and compute it step by step. Then comment briefly on how reliable the estimate is, since you are extrapolating into the tail.