Tail correlation

Tail correlation (or tail dependence) is a central concept in copula theory and extreme finance, denoting the fact that random variables may exhibit weak static dependence (linear correlation close to zero in normal conditions) but strong extreme dependence (effective correlation close to one in tail scenarios). Formally, upper tail dependence between two risks X and Y is defined as the probability that Y exceeds its extreme quantile conditional on X exceeding its own. Gaussian copulas (used in the Solvency II standard formula) exhibit zero tail dependence, that is, they assume extreme events remain independent across assets. Student and Archimedean copulas (Gumbel, Clayton) by contrast capture positive tail dependence. The 2008 financial crisis showed violently that structured credit products (CDOs) had been priced using Gaussian copulas assuming independence of defaults in extreme scenarios, whereas tail correlation in the underlying real estate was close to one. For systemic cyber risk, tail correlation is structurally close to one: an event like CrowdStrike simultaneously and identically affects all insureds sharing the same technology, regardless of the apparent decorrelation of their activities in normal conditions. This characteristic makes the Solvency II standard formula inadequate for cyber risk and cancels the diversification benefit normally granted.

A portfolio of 1,000 SME policyholders of a cyber insurer appears well diversified: they operate in different sectors and different countries. The linear correlation of their losses in normal conditions is low. But 800 of them use the same endpoint security tool. In a CrowdStrike scenario, the effective tail correlation is close to one for these 800 entities: the single event triggers all their losses simultaneously. Capital calculated on a moderate correlation assumption is massively insufficient.

tail correlation, tail dependence, dépendance de queue, corrélation extrême