Extreme Value Theory


Standard estimators of the dependence between assets are the correlation coefficient or the Spearman’s rank correlation for instance. However, as stressed by [Embrechts et al. ], these kind of dependence measures suffer from many deficiencies. Moreoever, their values are mostly controlled by relatively small moves of the asset prices around their mean. To cure this problem, it has been proposed to use the correlation coefficients conditioned on large movements of the assets. But [Boyer et al.] have emphasized that this approach suffers also from a severe systematic bias leading to spurious strategies: the conditional correlation in general evolves with time even when the true non-conditional correlation remains constant. In fact, [Malevergne and Sornette] have shown that any approach based on conditional dependence measures implies a spurious change of the intrinsic value of the dependence, measured for instance by copulas. Recall that the copula of several random variables is the (unique) function which completely embodies the dependence between these variables, irrespective of their marginal behavior (see [Nelsen] for a mathematical description of the notion of copula).

In view of these limitations of the standard statistical tools, it is natural to turn to extreme value theory. In the univariate case, extreme value theory is very useful and provides many tools for investigating the extreme tails of distributions of assets returns. These new developments rest on the existence of a few fundamental results on extremes, such as the Gnedenko-Pickands-Balkema-de Haan theorem which gives a general expression for the distribution of exceedence over a large threshold. In this framework, the study of large and extreme co-movements requires the multivariate extreme values theory, which unfortunately does not provide strong results. Indeed, in constrast with the univariate case, the class of limiting extreme-value distributions is too broad and cannot be used to constrain accurately the distribution of large co-movements.

In the spirit of the mean-variance portfolio or of utility theory which establish an investment decision on a unique risk measure, we use the coefficient of tail dependence, which, to our knowledge, was first introduced in the financial context by [Embrechts et al.]. The coefficient of tail dependence between assets Xi and Xj is a very natural and easy to understand measure of extreme co-movements. It is defined as the probability that the asset Xi incurs a large loss (or gain) assuming that the asset Xj also undergoes a large loss (or gain) at the same probability level, in the limit where this probability level explores the extreme tails of the distribution of returns of the two assets. Mathematically speaking, the coefficient of lower tail dependence between the two assets Xi and Xj , denoted by λ−ij is defined by

λ−ij = limu→0 Pr{Xi<Fi−1(u)|Xj < Fj−1(u)} —– (1)

where Fi−1(u) and Fj−1(u) represent the quantiles of assets Xand Xj at level u. Similarly the coefficient of the upper tail dependence is

λ+ij = limu→1 Pr{Xi > Fi−1(u)|Xj > Fj−1(u)} —– (2)

λ−ij and λ+ij are of concern to investors with long (respectively short) positions. We refer to [Coles et al.] and references therein for a survey of the properties of the coefficient of tail dependence. Let us stress that the use of quantiles in the definition of λ−ij and λ+ij makes them independent of the marginal distribution of the asset returns: as a consequence, the tail dependence parameters are intrinsic dependence measures. The obvious gain is an “orthogonal” decomposition of the risks into (1) individual risks carried by the marginal distribution of each asset and (2) their collective risk described by their dependence structure or copula.

Being a probability, the coefficient of tail dependence varies between 0 and 1. A large value of λ−ij means that large losses occur almost surely together. Then, large risks can not be diversified away and the assets crash together. This investor and portfolio manager nightmare is further amplified in real life situations by the limited liquidity of markets. When λ−ij vanishes, these assets are said to be asymptotically independent, but this term hides the subtlety that the assets can still present a non-zero dependence in their tails. For instance, two normally distributed assets can be shown to have a vanishing coefficient of tail dependence. Nevertheless, unless their correlation coefficient is identically zero, these assets are never independent. Thus, asymptotic independence must be understood as the weakest dependence which can be quantified by the coefficient of tail dependence.

For practical implementations, a direct application of the definitions (1) and (2) fails to provide reasonable estimations due to the double curse of dimensionality and undersampling of extreme values, so that a fully non-parametric approach is not reliable. It turns out to be possible to circumvent this fundamental difficulty by considering the general class of factor models, which are among the most widespread and versatile models in finance. They come in two classes: multiplicative and additive factor models respectively. The multiplicative factor models are generally used to model asset fluctuations due to an underlying stochastic volatility for a survey of the properties of these models). The additive factor models are made to relate asset fluctuations to market fluctuations, as in the Capital Asset Pricing Model (CAPM) and its generalizations, or to any set of common factors as in Arbitrage Pricing Theory. The coefficient of tail dependence is known in close form for both classes of factor models, which allows for an efficient empirical estimation.

Yield Curve Dynamics or Fluctuating Multi-Factor Rate Curves


The actual dynamics (as opposed to the risk-neutral dynamics) of the forward rate curve cannot be reduced to that of the short rate: the statistical evidence points out to the necessity of taking into account more degrees of freedom in order to represent in an adequate fashion the complicated deformations of the term structure. In particular, the imperfect correlation between maturities and the rich variety of term structure deformations shows that a one factor model is too rigid to describe yield curve dynamics.

Furthermore, in practice the value of the short rate is either fixed or at least strongly influenced by an authority exterior to the market (the central banks), through a mechanism different in nature from that which determines rates of higher maturities which are negotiated on the market. The short rate can therefore be viewed as an exogenous stochastic input which then gives rise to a deformation of the term structure as the market adjusts to its variations.

Traditional term structure models define – implicitly or explicitly – the random motion of an infinite number of forward rates as diffusions driven by a finite number of independent Brownian motions. This choice may appear surprising, since it introduces a lot of constraints on the type of evolution one can ascribe to each point of the forward rate curve and greatly reduces the dimensionality i.e. the number of degrees of freedom of the model, such that the resulting model is not able to reproduce any more the complex dynamics of the term structure. Multifactor models are usually justified by refering to the results of principal component analysis of term structure fluctuations. However, one should note that the quantities of interest when dealing with the term structure of interest rates are not the first two moments of the forward rates but typically involve expectations of non-linear functions of the forward rate curve: caps and floors are typical examples from this point of view. Hence, although a multifactor model might explain the variance of the forward rate itself, the same model may not be able to explain correctly the variability of portfolio positions involving non-linear combinations of the same forward rates. In other words, a principal component whose associated eigenvalue is small may have a non-negligible effect on the fluctuations of a non-linear function of forward rates. This question is especially relevant when calculating quantiles and Value-at-Risk measures.

In a multifactor model with k sources of randomness, one can use any k + 1 instruments to hedge a given risky payoff. However, this is not what traders do in real markets: a given interest-rate contingent payoff is hedged with bonds of the same maturity. These practices reflect the existence of a risk specific to instruments of a given maturity. The representation of a maturity-specific risk means that, in a continuous-maturity limit, one must also allow the number of sources of randomness to grow with the number of maturities; otherwise one loses the localization in maturity of the source of randomness in the model.

An important ingredient for the tractability of a model is its Markovian character. Non-Markov processes are difficult to simulate and even harder to manipulate analytically. Of course, any process can be transformed into a Markov process if it is imbedded into a space of sufficiently high dimension; this amounts to injecting a sufficient number of “state variables” into the model. These state variables may or may not be observable quantities; for example one such state variable may be the short rate itself but another one could be an economic variable whose value is not deducible from knowledge of the forward rate curve. If the state variables are not directly observed, they are obtainable in principle from the observed interest rates by a filtering process. Nevertheless the presence of unobserved state variables makes the model more difficult to handle both in terms of interpretation and statistical estimation. This drawback has motivated the development of so-called affine curve models models where one imposes that the state variables be affine functions of the observed yield curve. While the affine hypothesis is not necessarily realistic from an empirical point of view, it has the property of directly relating state variables to the observed term structure.

Another feature of term structure movements is that, as a curve, the forward rate curve displays a continuous deformation: configurations of the forward rate curve at dates not too far from each other tend to be similar. Most applications require the yield curve to have some degree of smoothness e.g. differentiability with respect to the maturity. This is not only a purely mathematical requirement but is reflected in market practices of hedging and arbitrage on fixed income instruments. Market practitioners tend to hedge an interest rate risk of a given maturity with instruments of the same maturity or close to it. This important observation means that the maturity is not simply a way of indexing the family of forward rates: market operators expect forward rates whose maturities are close to behave similarly. Moreover, the model should account for the observation that the volatility term structure displays a hump but that multiple humps are never observed.