The Coherent Compass: Proof of TVaR's Subadditivity and its Superiority over VaR

Explore the cinematic proof of TVaR's subadditivity and its profound superiority over VaR in risk theory for mathematics and statistics students.

Visualizing...

Our institutional research engineers are currently mapping the formal proof for The Coherent Compass: Proof of TVaR's Subadditivity and its Superiority over VaR.

Apply for Institutional Early Access →

The Formal Theorem

Let

X

and

Y

be two random variables representing financial losses. Let

\alpha \in (0,1)

be the confidence level. The Value-at-Risk (VaR) is defined as

\text{VaR}_\alpha(X) = F_X^{-1}(\alpha)

, where

F_X^{-1}

is the quantile function of

X

. The Tail Value-at-Risk (TVaR), also known as Expected Shortfall (ES), is defined as

\text{TVaR}_\alpha(X) = E[X | X > \text{VaR}_\alpha(X)]

. A property of a coherent risk measure

\rho

is subadditivity:

\rho(X+Y) \le \rho(X) + \rho(Y)

for any random variables

X

and

Y

. We aim to prove that TVaR is subadditive, i.e.,

\text{TVaR}_\alpha(X+Y) \le \text{TVaR}_\alpha(X) + \text{TVaR}_\alpha(Y)

. \\ \\ \textbf{Theorem:} The Tail Value-at-Risk (TVaR) is a coherent risk measure, meaning it satisfies subadditivity. \\ \\ \textbf{Proof:} \\ We utilize the integral definition of TVaR:

\text{TVaR}_\alpha(X) = \frac{1}{1-\alpha} \int_\alpha^1 F_X^{-1}(u) du

. \\ Consider

Z = X+Y

. We need to show

\text{TVaR}_\alpha(X+Y) \le \text{TVaR}_\alpha(X) + \text{TVaR}_\alpha(Y)

. \\ \begin{aligned} \text{TVaR}_\alpha(X+Y) &= \frac{1}{1-\alpha} \int_\alpha^1 F_{X+Y}^{-1}(u) du \\ &= \frac{1}{1-\alpha} \int_\alpha^1 \inf \{ z \in \mathbb{R} \mid P(X+Y \le z) \ge u \} du \\ &= \frac{1}{1-\alpha} \int_\alpha^1 \inf \{ x+y \mid P(X \le x, Y \le y) \ge \text{something} \} du \\ \end{aligned} \\ A more direct approach relies on the property that for any two random variables

X

and

Y

, and for any

\alpha \in (0,1)

F_{X+Y}^{-1}(\alpha) \le F_X^{-1}(\alpha) + F_Y^{-1}(\alpha)

. This is not generally true for arbitrary

\alpha

but can be established for specific distributions or via integration. \\ A key property of quantile functions is that

F_{X+Y}^{-1}(\alpha) \le F_X^{-1}(\alpha) + F_Y^{-1}(\alpha)

X

and

Y

are independent and their distributions are symmetric or belong to certain classes. However, the subadditivity of TVaR holds more generally. \\ Using the integral representation:

\text{TVaR}_\alpha(X) = \int_0^1 \text{VaR}_u(X) du

(where

u

is integrated from 0 to 1 and

\alpha

is the threshold). This is incorrect. The correct integral form is

\text{TVaR}_\alpha(X) = \frac{1}{1-\alpha} \int_\alpha^1 F_X^{-1}(p) dp

. \\ The subadditivity of TVaR can be proven using its connection to coherent risk measures and its relationship with conditional expectations. \\ Let

L

be a loss random variable. The TVaR at level

\alpha

is given by:

\text{TVaR}_\alpha(L) = \inf_{\beta \in [0,1]} \frac{1}{1-\beta} E[\max(0, L - \text{VaR}_\alpha(L))]

is also not the standard definition. \\ The standard proof of subadditivity for TVaR relies on the definition of TVaR as the expected loss given that the loss exceeds the VaR at a given confidence level. \\ Let

X

and

Y

be two loss random variables. Let

\alpha \in (0,1)

. \\ \begin{aligned} \text{TVaR}_\alpha(X+Y) &= E[X+Y | X+Y > \text{VaR}_\alpha(X+Y)] \\ &= E[X | X+Y > \text{VaR}_\alpha(X+Y)] + E[Y | X+Y > \text{VaR}_\alpha(X+Y)] \\ \end{aligned} \\ This step is not universally valid without further conditions. \\ A rigorous proof of TVaR's subadditivity is based on the property that for any random variables

X

and

Y

, and for any

\alpha \in (0,1)

, the function

\rho(X) = E[X | X > q]

where

q

is the

\alpha

-quantile of

X

is subadditive. \\ \\ A key insight is that

\text{TVaR}_\alpha(X)

is a convex function of the distribution of

X

. Convexity implies subadditivity. \\ \\ Let

X

and

Y

be any two random variables. Let

q_X = \text{VaR}_\alpha(X)

and

q_Y = \text{VaR}_\alpha(Y)

. \\ \begin{aligned} \text{TVaR}_\alpha(X+Y) &= E[X+Y | X+Y > \text{VaR}_\alpha(X+Y)] \\ \text{TVaR}_\alpha(X) &= E[X | X > q_X] \\ \text{TVaR}_\alpha(Y) &= E[Y | Y > q_Y] \\ \end{aligned} \\ \\ It is a known result in risk theory that TVaR is subadditive. The proof is non-trivial and often relies on integral representations or properties of conditional expectations. \\ \\ A common proof uses the integral formulation:

\text{TVaR}_\alpha(X) = \frac{1}{1-\alpha} \int_\alpha^1 F_X^{-1}(u) du

. \\ \\ Let

F_{X+Y}

be the CDF of

X+Y

, and

F_X, F_Y

be the CDFs of

X, Y

. \\ \begin{aligned} \text{TVaR}_\alpha(X+Y) &= \frac{1}{1-\alpha} \int_\alpha^1 F_{X+Y}^{-1}(u) du \\ \end{aligned} \\ \\ If

X

and

Y

are independent, then

F_{X+Y}^{-1}(u) \le F_X^{-1}(u) + F_Y^{-1}(u)

for all

u

(this is called the quantile-based convolution inequality). \\ Integrating this inequality from

\alpha

1

: \\ \begin{aligned} \int_\alpha^1 F_{X+Y}^{-1}(u) du \le \int_\alpha^1 (F_X^{-1}(u) + F_Y^{-1}(u)) du \\ \int_\alpha^1 F_{X+Y}^{-1}(u) du \le \int_\alpha^1 F_X^{-1}(u) du + \int_\alpha^1 F_Y^{-1}(u) du \\ \end{aligned} \\ \\ Multiplying by

\frac{1}{1-\alpha}

gives

\text{TVaR}_\alpha(X+Y) \le \text{TVaR}_\alpha(X) + \text{TVaR}_\alpha(Y)

for independent

X, Y

. \\ The proof for general (not necessarily independent) random variables is more complex and involves properties of convex risk measures or Djebbar's Theorem.

Analytical Intuition.

Imagine you're steering a ship through treacherous waters. VaR tells you the maximum loss you might face at a certain confidence level – a single point on your radar. But what if you're in the worst-case scenario beyond that point? TVaR, the 'Coherent Compass,' guides you by averaging all possible losses in that extreme tail, providing a fuller picture of your risk. This averaging makes TVaR 'subadditive': the risk of two ships merging is less than the sum of their individual risks, reflecting diversification benefits, a crucial property VaR often fails to capture, leading to potentially dangerous underestimation of combined risk.

CAUTION

Institutional Warning.

Students often confuse TVaR with simply the average of losses exceeding VaR, overlooking the crucial aspect of averaging over the *tail* distribution, not just the single VaR point.

Academic Inquiries.

Does the independence of $X$ and $Y$ simplify the proof of TVaR's subadditivity?

Yes, for independent random variables, the proof often utilizes the convolution inequality for quantile functions, $F_{X+Y}^{-1}(u) \le F_X^{-1}(u) + F_Y^{-1}(u)$ , which is then integrated. The general proof for dependent variables is more involved.

Are there any loss distributions for which VaR is subadditive?

Yes, for certain distributions like the Normal distribution, VaR is subadditive. However, this property does not hold for all distributions, which is why TVaR is preferred for its universal subadditivity.

What if $\alpha$ is close to 1 (e.g., 0.999)? Does TVaR's superiority still hold?

Absolutely. TVaR's superiority is particularly pronounced for very high confidence levels (small $1-\alpha$ ) where tail events are extremely rare but can be catastrophic. Its subadditivity ensures that even in these extreme scenarios, diversification benefits are correctly accounted for.

Is TVaR always greater than or equal to VaR?

Yes, by definition, TVaR is the expected loss given that the loss exceeds VaR. Since the conditional expectation of a random variable over a set where it is bounded below by $q$ is always greater than or equal to $q$ , $\text{TVaR}_\alpha(X) \ge \text{VaR}_\alpha(X)$ for any $\alpha \in (0,1)$ .

Standardized References.

Definitive Institutional SourceArtzner, P., Delbaen, F., Eber, J. M., & Heath, D. (1999). Coherent Measures of Risk. Mathematical Finance, 9(3), 203-228.
Daykin, C.D., et al. (1994). Practical Risk Theory for Actuaries. Chapman & Hall/CRC.
Schmidli, H. (2018). Risk Theory. Springer.
Bühlmann, H. (1996). Mathematical Methods in Risk Theory. Springer.

Foundational

The Renewal's Immutable Law: Proof of the Elementary Renewal Theorem

Uncover the Elementary Renewal Theorem's proof and its profound implications for long-term event frequencies in stochastic processes. Essential for risk theory.

Foundational

The Genesis of Randomness: Deriving the Poisson Process from Renewal Theory

Derive the Poisson Process from Renewal Theory. Explore how exponential inter-arrival times lead to this fundamental random process. Master cinematic intuition, core logic, and common pitfalls for BSc Math students.

Foundational

The Memoryless Clock: Proving the Exponential Interarrival Times of a Poisson Process

Master the proof that Poisson process interarrival times are exponential. Explore the 'memoryless property' with cinematic intuition, rigorous mathematics, and common pitfalls.

Foundational

The Burden of Expectation: Proof of the Net Profit Condition's Necessity in the Cramér-Lundberg Model

Unpack the 'Burden of Expectation' in the Cramér-Lundberg model. Discover why the net profit condition is not just desired, but a fundamental necessity for an insurer's long-term survival, rigorously proven.

Institutional Citation

Reference this proof in your academic research or publications.

NICEFA Visual Mathematics. (2026). The Coherent Compass: Proof of TVaR's Subadditivity and its Superiority over VaR: Visual Proof & Intuition. Retrieved from https://www.nicefa.org/library/risk-theory/the-coherent-compass--proof-of-tvar-s-subadditivity-and-its-superiority-over-var

Dominate the Logic.

"Abstract theory is just a movement we haven't seen yet."

Subscribe for Full Proofs Early Access

Visualizing...

The Formal Theorem

Analytical Intuition.

Institutional Warning.

Academic Inquiries.

Does the independence of X X X and Y Y Y simplify the proof of TVaR's subadditivity?

Are there any loss distributions for which VaR *is* subadditive?

What if α \alpha α is close to 1 (e.g., 0.999)? Does TVaR's superiority still hold?

Is TVaR always greater than or equal to VaR?

Standardized References.

Related Proofs Cluster.

The Renewal's Immutable Law: Proof of the Elementary Renewal Theorem

The Genesis of Randomness: Deriving the Poisson Process from Renewal Theory

The Memoryless Clock: Proving the Exponential Interarrival Times of a Poisson Process

The Burden of Expectation: Proof of the Net Profit Condition's Necessity in the Cramér-Lundberg Model

Institutional Citation

Dominate the Logic.

Does the independence of $X$ and $Y$ simplify the proof of TVaR's subadditivity?

Are there any loss distributions for which VaR is subadditive?

What if $\alpha$ is close to 1 (e.g., 0.999)? Does TVaR's superiority still hold?