Understand the derivation and characteristics of the Generalized Extreme Value and the Generalized Pareto distributions.

Catastrophe reinsurance, operational loss, and large-claim XL pricing all live in the tail. Two distributions dominate that territory: GEV (block maxima) and GPD (peaks over threshold).

The extreme-value problem. Lognormal and gamma fit the body of losses well but underestimate the tail. Asymptotic extreme-value theory delivers two limit distributions for tail data, mirroring the way the CLT delivers the normal for averages.

GEV derivation (Fisher-Tippett-Gnedenko). Let $X_1, \dots, X_n$ be iid with CDF $F$ and let $M_n = \max(X_1, \dots, X_n)$. If sequences $a_n > 0$ and $b_n$ exist such that $(M_n - b_n)/a_n$ converges in distribution to a non-degenerate $H$, then $H$ is GEV.

Read the CDF as a survival mechanism: the outer $\exp(-\cdot)$ is the probability that no block member exceeds $x$, and the inner $(1 + \xi z)^{-1/\xi}$ governs how fast that no-exceedance probability decays as $x$ climbs. The shape $\xi$ is the dial controlling that decay.

Common mistakes

• Reading $\xi < 0$ as "lighter than Gumbel". It is bounded, with endpoint $\mu - \sigma/\xi$.
• Confusing GEV scale $\sigma$ with GPD scale $\beta$. They are different parameters of different fits.
• Quoting GPD mean when $\xi \ge 1$. Mean is $\beta/(1-\xi)$ and undefined for $\xi \ge 1$.

Bottom line

• GEV models the maximum of a block of losses; GPD models excesses above a high threshold.
• GEV unifies Fréchet, Gumbel, and reverse Weibull through one shape $\xi$: $\xi > 0$ heavy, $\xi = 0$ Gumbel (light), $\xi < 0$ bounded.
• A bounded tail ($\xi < 0$) has finite endpoint $\mu - \sigma/\xi$; it is not merely lighter than Gumbel.
• Fisher-Tippett-Gnedenko: normalized block maxima converge to GEV (analogue of CLT for maxima).

Exam shortcut

Read $\xi$ first. Sign of $\xi$ tells you tail behavior immediately (heavy, light, bounded). For GPD survival, memorize $(1 + \xi y/\beta)^{-1/\xi}$. At $\xi = 0$ it collapses to $\exp(-y/\beta)$. For GEV return level, set $H(x_T) = 1 - 1/T$ and invert: $x_T = \mu + (\sigma/\xi)[(-\ln(1-1/T))^{-\xi} - 1]$.

The full lesson (about 1,714 words, 11 min read) adds 2 worked examples, all 6 common mistakes, a self-check, free in the app.

Learning objectives

• 1e

Browse all free Exam ASTAM lessons or jump into free Exam ASTAM practice questions.