Entropic value at risk

In financial mathematics and stochastic optimization, the concept of risk measure is used to quantify the risk involved in a random outcome or risk position. Many risk measures have hitherto been proposed, each having certain characteristics. The entropic value-at-risk (EVaR) is a coherent risk measure introduced by Ahmadi-Javid, which is an upper bound for the value at risk (VaR) and the conditional value-at-risk (CVaR), obtained from the Chernoff inequality. The EVaR can also be represented by using the concept of relative entropy. Because of its connection with the VaR and the relative entropy, this risk measure is called "entropic value-at-risk". The EVaR was developed to tackle some computational inefficiencies of the CVaR. Getting inspiration from the dual representation of the EVaR, Ahmadi-Javid developed a wide class of coherent risk measures, called g-entropic risk measures. Both the CVaR and the EVaR are members of this class.

Definition

Let $(\Omega,\mathcal{F},P)$ be a probability space with $\Omega$ a set of all simple events, $\mathcal{F}$ a $\sigma$-algebra of subsets of $\Omega$ and $P$ a probability measure on $\mathcal{F}$. Let $X$ be a random variable and $\mathbf{L}_{M^+}$ be the set of all Borel measurable functions $X:\Omega\rightarrow \R$ whose moment-generating function $M_X(z)$ exists for all $z\geq 0$. The entropic value-at-risk (EVaR) of $X\in \mathbf{L}_{M^+}$ with confidence level $1-\alpha$ is defined as follows:

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \text{EVaR}_{1-\alpha}(X):=\inf_{z>0}\{z^{-1}\ln(M_X(z)/\alpha)\}. \,

(1)

In finance, the random variable $X \in \mathbf{L}_{M^+}$, in the above equation, is used to model the losses of a portfolio.

Consider the Chernoff inequality

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \text{Pr}(X\geq a)\leq e^{-za}M_X(z),\quad \forall z>0.\,

(2)

Solving the equation $e^{-za}M_X(z)=\alpha$ for $a$, results in $a_X(\alpha,z):=z^{-1}\ln(M_X(z)/\alpha)$. By considering the equation (1), we see that Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \text{EVaR}_{1-\alpha}(X):=\inf_{z>0}\{a_X(\alpha,z)\} , which shows the relationship between the EVaR and the Chernoff inequality. It is worth noting that $a_X(1,z)$ is the entropic risk measure or exponential premium, which is a concept used in finance and insurance, respectively.

Let $\mathbf{L}_{M}$ be the set of all Borel measurable functions $X:\Omega\rightarrow \R$ whose moment-generating function $M_X(z)$ exists for all $z$. The dual representation (or robust representation) of the EVaR is as follows:

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \text{EVaR}_{1-\alpha}(X)=\sup_{Q\in \Im}(E_Q(X))\,

(3)

where $X\in \mathbf{L}_{M}$, and $\Im$ is a set of probability measures on $(\Omega,\mathcal{F})$ with Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \Im=\{Q\ll P:D_{KL}(Q||P)\leq-\ln\alpha\} . Note that $D_{KL}(Q||P):=\int\frac{dQ}{dP}(\ln\frac{dQ}{dP})dP$ is the relative entropy of $Q$ with respect to $P$, also called the Kullback–Leibler divergence. The dual representation of the EVaR discloses the reason behind its naming.

Properties

• The EVaR is a coherent risk measure.
• The moment-generating function $M_X(z)$ can be represented by the EVaR: for all $X\in \mathbf{L}_{M^+}$ and $z>0$

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): M_X(z)=\sup_{0<\alpha\leq 1}\{\alpha\exp(z\text{EVaR}_{1-\alpha}(X))\}.\,

(4)

• For $X,Y\in\mathbf{L}_M$, Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \text{EVaR}_{1-\alpha}(X)=\text{EVaR}_{1-\alpha}(Y)

for all $\alpha\in]0,1]$ if and only if $F_X(b)=F_Y(b)$ for all $b\in\R$.

• The entropic risk measure with parameter $\theta$, can be represented by means of the EVaR: for all $X\in \mathbf{L}_{M^+}$ and $\theta>0$

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \theta^{-1}\ln M_X(\theta)=a_X(1,\theta)=\sup_{0<\alpha\leq 1}\{\text{EVaR}_{1-\alpha}(X)+\theta^{-1}\ln\alpha\}.\,

(5)

• The EVaR with confidence level $1-\alpha$ is the tightest possible upper bound that can be obtained from the Chernoff inequality for the VaR and the CVaR with confidence level $1 - \alpha$;

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \text{VaR}(X)\leq \text{CVaR}(X)\leq\text{EVaR}(X).\,

(6)

• The following inequality holds for the EVaR:

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \text{E}(X)\leq\text{EVaR}_{1-\alpha}(X)\leq\text{esssup}(X)\,

(7)

where Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \text{E}(X)

is the expected value of $X$ and Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.):  \text{esssup}(X)
is the essential supremum of $X$, i.e., Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \inf_{t\in\R}\{t:\text{Pr}(X\leq t)=1\} 

. So do hold Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \text{EVaR}_0(X)=\text{E}(X)

and Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.):  \lim_{\alpha\rightarrow 0}\text{EVaR}_{1-\alpha}(X)=\text{esssup}(X) 

.

Examples

Comparing the VaR, CVaR and EVaR for the standard normal distribution

Comparing the VaR, CVaR and EVaR for the uniform distribution over the interval (0,1)

For $X\sim N(\mu,\sigma)$,

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \text{EVaR}_{1-\alpha}(X)=\mu+\sqrt{-2\ln\alpha}\sigma.\,

(8)

For $X\sim U(a,b)$,

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \text{EVaR}_{1-\alpha}(X)=\inf_{t>0}\left\lbrace t\ln\left(t\frac{e^{t^{-1}b}-e^{t^{-1}a}}{b-a}\right)-t\ln\alpha \right\rbrace. \,

(9)

Figures 1 and 2 show the comparing of the VaR, CVaR and EVaR for $N(0,1)$ and $U(0,1)$.

Optimization

Let $\rho$ be a risk measure. Consider the optimization problem

$\min_{\boldsymbol{w}\in \boldsymbol{W}}\rho(G(\boldsymbol{w},\boldsymbol{\psi}))\,$

(10)

where $\boldsymbol{w}\in\boldsymbol{W}\subseteq\R^n$ is an $n$-dimensional real decision vector, $\boldsymbol{\psi}$ is an $m$-dimensional real random vector with a known probability distribution and the function $G(\boldsymbol{w},.):\R^m\rightarrow\R$ is a Borel measurable function for all values $\boldsymbol{w}\in\boldsymbol{W}$. If $\rho$ is the Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \text{EVaR} , then the problem (10) becomes as follows:

$\min_{\boldsymbol{w}\in\boldsymbol{W}, t>0}\{t\ln M_{G(\boldsymbol{w},\boldsymbol{\psi})}(t^{-1})-t\ln\alpha\}.\,$

(11)

Let $\boldsymbol{S}_{\boldsymbol{\psi}}$ be the support of the random vector $\boldsymbol{\psi}$. If $G(.,\boldsymbol{s})$ is convex for all $\boldsymbol{s}\in\boldsymbol{S}_{\boldsymbol{\psi}}$, then the objective function of the problem (11) is also convex. If $G(\boldsymbol{w},\boldsymbol{\psi})$ has the form

$G(\boldsymbol{w},\boldsymbol{\psi})=g_0(\boldsymbol{w})+\sum_{i=1}^mg_i(\boldsymbol{w})\psi_i,\quad g_i:\R^n\rightarrow\R, i=0,1,\dots,m,\,$

(12)

and $\psi_1,\dots,\psi_m$ are independent random variables in $\mathbf{L}_M$, then (11) becomes

$\min_{\boldsymbol{w}\in\boldsymbol{W}, t>0}\left\lbrace g_0(\boldsymbol(w))+t\sum_{i=1}^m\ln M_{g_i(\boldsymbol(w))\psi_i}(t^{-1})-t\ln\alpha \right\rbrace.\,$

(13)

which is computationally tractable. But for this case, if one uses the CVaR in problem (10), then the resulting problem becomes as follows:

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \min_{\boldsymbol{w}\in\boldsymbol{W}, t\in\R}\left\lbrace t+\frac{1}{\alpha}\text{E}\left[ g_0(\boldsymbol{w})+\sum_{i=1}^{m}g_i(\boldsymbol{w}) \psi_i-t \right]_+ \right\rbrace.\,

(14)

It can be shown that by increasing the dimension of $\psi$, problem (14) is computationally intractable even for simple cases. For example, assume that $\psi_1,\dots,\psi_m$ are independent discrete random variables that take $k$ distinct values. For fixed values of $\boldsymbol{w}$ and $t$, the complexity of computing the objective function given in problem (13) is of order $mk$ while the computing time for the objective function of problem (14) is of order $k^m$. For illustration, assume that $k= 2$, $m= 100$ and the summation of two numbers takes $10^{-12}$ seconds. For computing the objective function of problem (14) one needs about $4\times 10^{10}$ years, whereas the evaluation of objective function of problem (13) takes about $10^{-10}$ seconds. This shows that formulation with the EVaR outperforms the formulation with the CVaR (see for more details).

Generalization (g-entropic risk measures)

Drawing inspiration from the dual representation of the EVaR given in (3), one can define a wide class of information-theoretic coherent risk measures, which are introduced in. Let $g$ be a convex proper function with $g(1)=0$ and $\beta$ be a non-negative number. The $g$-entropic risk measure with divergence level $\beta$ is defined as

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \text{ER}_{g,\beta}(X):=\sup_{Q\in\Im}\text{E}_Q(X)\,

(15)

where Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \Im=\{Q\ll P:H_g(P,Q)\leq\beta\}

in which $H_g(P,Q)$ is the generalized relative entropy of $Q$ with respect to $P$. A primal representation of the class of $g$-entropic risk measures can be obtained as follows:

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): \text{ER}_{g,\beta}(X)=\inf_{t>0,\mu\in\R}\left\lbrace t\left[ \mu+\text{E}_P\left( g^*\left( \frac{X}{t}-\mu+\beta \right) \right) \right] \right\rbrace\,

(16)

where $g^*$ is the conjugate of $g$. By considering

$g(x)=\begin{cases} x\ln x & x>0 \\ 0 & x=0 \\ +\infty & x<0, \end{cases} \,$

(17)

with $g^*(x)=e^{x-1}$ and $\beta=- \ln\alpha$, the EVaR formula can be deduced. The CVaR is also a $g$-entropic risk measure, which can be obtained from (16) by setting

Failed to parse (Missing <code>texvc</code> executable. Please see math/README to configure.): g(x)=\left\lbrace \begin{array}{lr} 0 & 0\leq x\leq \frac{1}{\alpha} \\ +\infty & \text{otherwise}, \end{array} \right. \,

(18)

with $g^*(x)=\frac{1}{\alpha}\max\{0,x\}$ and $\beta=0$ (see for more details).

For more results on $g$-entropic risk measures see.

See also

• Stochastic optimization
• Risk measure
• Coherent risk measure
• Value at risk
• Conditional value-at-risk
• Expected shortfall
• Entropic risk measure
• Kullback–Leibler divergence
• Generalized relative entropy