VaR at Its Extremes: Impossibilities and Conditions for One-Sided Random Variables

This paper establishes that Value-at-Risk (VaR) for non-negative random variables is strictly sub-additive only in the degenerate co-monotonic case, while providing a unified framework based on negative simplex dependence and simplex dominance to characterize conditions under which VaR becomes fully super-additive across all probability levels.

The Gist

Imagine you are the captain of a ship, and your job is to predict how much damage a storm might do to your cargo. In the world of finance and insurance, this "damage prediction" is called Value-at-Risk (VaR). It answers the question: "What is the worst loss I can expect 99% of the time?"

For decades, experts have argued about a specific rule called Sub-additivity. This is the idea that "the whole should be less than the sum of its parts." In plain English: If you mix two risky things together, the total risk should go down because they might cancel each other out (diversification).

This paper, written by Nawaf Mohammed, pulls back the curtain on when this "diversification magic" actually works and when it completely fails. Here is the breakdown in everyday language.

1. The "Magic Trick" That Doesn't Exist (Sub-additivity)

The Scenario: Imagine you have a bag of money that can only go up (like a stock price or an insurance claim). It can't be negative.
The Question: If I combine two of these bags, will the total risk be lower than adding their individual risks?
The Paper's Verdict: No. Never.

The author proves a "No-Go" theorem. For any risks that start at zero and go up (like insurance losses), the only time the total risk is not higher than the sum of the parts is when the two risks are perfectly synchronized.

• The Analogy: Imagine two dancers.
  • If they dance randomly, sometimes one trips while the other is fine. But if they are co-monotonic (perfectly synchronized), they trip at the exact same time.
  • The paper says: For these types of risks, diversification is a myth. You can only get the "safe" result (where risks add up perfectly) if the risks are already doing the exact same thing at the exact same time. If they aren't perfectly synchronized, the total risk is actually worse than just adding them up.

2. The "Explosion" Effect (Super-additivity)

The Scenario: Now, imagine the risks are "heavy-tailed." This means they usually stay small, but occasionally, they have a massive, catastrophic explosion (like a once-in-a-century hurricane or a massive market crash).
The Question: Can combining these risks make the total danger much worse than just adding them up?
The Paper's Verdict: Yes. Absolutely.

This is called Super-additivity. It's the opposite of diversification. Instead of canceling out, the risks amplify each other.

• The Analogy: Think of two fireworks.
  • If you light them separately, they make a nice pop.
  • But if they are linked in a specific, negative way (like a chain reaction), lighting one might trigger the other to explode with double the force.
  • The paper shows that if you have "heavy" risks (infinite average loss) and they are linked in a specific "negative" way, the combined risk can be terrifyingly high.

3. The Two Rules for the "Explosion"

The author doesn't just say "it happens." He gives you a checklist to know when it will happen. You need two things to happen at the same time:

1. The "Negative Simplex" Rule (The Dance): The risks must be linked in a specific way where they tend to avoid each other's worst moments, but when they do hit, they hit hard together. It's a complex dance where they are "negatively dependent."
2. The "Simplex Dominant" Rule (The Shape): The individual risks must have a specific "shape" to their distribution. They must be heavy-tailed enough (like the Pareto or Fréchet distributions mentioned in the paper) that their "tail" is heavy enough to cause an explosion.

The Takeaway: You can't just look at the risks individually, and you can't just look at how they are linked. You have to look at the interaction between the shape of the risks and how they are linked.

4. The "Moving Target" (Generalizing the Results)

The paper also asks: "What if the risks don't start at zero? What if they start at a higher number (like a deductible) or stop at a maximum number (like a capped payout)?"

• The Verdict: The rules flip.
  • If your risks have a floor (they can't go below a certain point), you can't get the "safe" diversification benefit (Sub-additivity).
  • If your risks have a ceiling (they can't go above a certain point), you can't get the "explosive" super-additivity.
  • It's like a seesaw: if the floor is fixed, the ceiling is free to go wild, and vice versa.

Summary: Why Should You Care?

This paper is a wake-up call for risk managers, bankers, and insurance companies.

1. Don't trust diversification blindly: If you are dealing with heavy-tailed risks (like natural disasters or crypto crashes), simply buying different types of insurance or stocks might not save you. In fact, it might make the worst-case scenario worse.
2. Know your math: The paper gives you a new "detector" (the NSD and SD conditions) to check if your portfolio is safe or if it's about to blow up.
3. The "Impossible" Dream: It proves that for many real-world risks, the dream of "perfect diversification" (where risk always goes down when you mix things) is mathematically impossible unless everything moves in perfect lockstep.

In a nutshell: When dealing with extreme risks, mixing them doesn't always calm the storm. Sometimes, it creates a bigger one. This paper tells you exactly how to spot that danger before it hits.

Technical Summary

Here is a detailed technical summary of the paper "VaR at Its Extremes: Impossibilities and Conditions for One-Sided Random Variables" by Nawaf Mohammed.

1. Problem Statement

The paper investigates the extremal aggregation behavior of Value-at-Risk (VaR) for sums of random variables, specifically focusing on one-sided risks (variables supported on $[0, \infty)$ or bounded intervals).

• Context: VaR is a standard risk measure in finance and actuarial science, defined as the left-quantile of a loss distribution. A desirable property for risk measures is sub-additivity ($VaR(S) \leq \sum VaR(X_i)$), which implies that diversification reduces risk. However, VaR is known to violate sub-additivity, particularly with heavy-tailed distributions.
• Gap: While previous literature (e.g., Imamura & Kato, 2025; Chen & Shneer, 2026) has explored VaR additivity and super-additivity, existing frameworks often rely on restrictive assumptions such as:
  • Identical marginal distributions.
  • Finite expectations (integrability).
  • Specific dependence structures (e.g., Negative Lower Orthant Dependence - NLOD).
  • Focus on asymptotic tail behavior rather than full-range probability levels.
• Objective: The author aims to characterize the conditions under which VaR is strictly sub-additive, strictly super-additive, or additive across all probability levels $p \in (0, 1)$ for non-identical and potentially non-integrable (infinite mean) random variables.

2. Methodology

The paper employs a rigorous probabilistic approach combining:

• Truncation and Monotone Convergence: To extend results from integrable variables to those with infinite means, the author constructs truncated versions of random vectors and applies limit arguments.
• Structural Decomposition: The analysis separates the problem into two distinct components:
  1. Dependence Structure: How the joint distribution behaves relative to the marginals.
  2. Marginal Behavior: The functional properties of the individual distribution functions.
• New Definitions: The author introduces two novel structural concepts to unify the analysis:
  • Negative Simplex Dependence (NSD): A condition on the joint Cumulative Distribution Function (CDF).
  • Simplex Dominance (SD): A condition on a margin-dependent functional $\Phi$.
• Duality Arguments: The paper utilizes the duality between sub-additivity and super-additivity via reflection (transforming $X$ to $-X$ or shifting endpoints) to extend results to variables with arbitrary finite lower or upper bounds.

3. Key Contributions and Results

A. Impossibility of VaR Sub-additivity (Theorem 2.2)

• Result: For random variables supported on $[0, \infty)$, VaR sub-additivity is impossible unless the variables are co-monotonic (perfectly positively dependent).
• Implication: If $X$ is VaR sub-additive, it must be exactly additive ($VaR(S) = \sum VaR(X_i)$), and the joint CDF must be the Fréchet upper bound ($F_X = \min(F_{X_1}, \dots, F_{X_n})$).
• Significance: This extends the findings of Imamura and Kato (2025) to the general case of infinite means. It proves that for non-negative risks, diversification never reduces VaR unless the risks move in perfect lockstep (in which case there is no diversification benefit).

B. Characterization of Full VaR Super-additivity (Theorem 3.5)

The paper provides a unified framework for when VaR is super-additive ($VaR(S) \geq \sum VaR(X_i)$) for all $p$. This requires two simultaneous conditions:

1. Negative Simplex Dependence (NSD): The joint CDF satisfies $F_S(t) \leq \prod_{i=1}^n F_{X_i}(t)$ for all $t \geq 0$. This is a weaker condition than NLOD (which requires the inequality to hold for all vectors, not just on the diagonal).
2. Simplex Dominance (SD): The function $\Phi(x_1, \dots, x_n) = \sum_{i=1}^n x_i \log F_{X_i}(x_i)$ must satisfy $\Phi(x) \geq \Phi(s, \dots, s)$ where $s = \sum x_i$.

Key Theoretical Insights:

• Non-Integrability Requirement: The paper proves (Proposition 3.12) that for $\Phi$ to be non-increasing (a sufficient condition for SD), the random variables must have infinite means ($E[X_i] = \infty$). Thus, genuine full-range super-additivity requires heavy tails.
• Generalization: Unlike previous works (e.g., Chen & Shneer, 2026) that required identical margins and independence, this framework accommodates non-identical margins and general negative dependence structures.
• Verification: The author provides a practical sufficient condition: if the variables are NLOD and the function $\phi_i(x) = x \log F_{X_i}(x)$ is non-increasing for each margin, then VaR is super-additive. This condition is verified for various heavy-tailed distributions (Fréchet, Pareto II, Lévy, etc.).

C. Extensions to Arbitrary Endpoints (Section 4)

• The results are extended to random variables with arbitrary finite lower bounds ($a_i$) and upper bounds ($b_i$).
• Duality:
  • For variables with finite lower bounds, strict sub-additivity is impossible (only additivity under co-monotonicity is possible).
  • For variables with finite upper bounds, strict super-additivity is impossible (only additivity under co-monotonicity is possible).
• This establishes a sharp dichotomy: VaR behaves as a "diversification-averse" measure in lower-bounded settings (super-additive) and "diversification-seeking" in upper-bounded settings (sub-additive), but only under specific structural constraints.

D. Transformation Invariance

The paper demonstrates that the property of VaR super-additivity is preserved under strictly increasing, convex transformations $\xi_i$ (with $\xi_i(0)=0$) applied to the components, provided the original vector satisfies the NSD and SD conditions.

4. Significance and Practical Implications

• Theoretical Clarity: The paper resolves the ambiguity surrounding VaR aggregation by proving that sub-additivity is structurally impossible for non-negative risks without perfect dependence. It shifts the focus to understanding when super-additivity occurs.
• Risk Management:
  • Diversification Failure: It confirms that for heavy-tailed, non-negative risks (common in insurance and catastrophe modeling), diversification often fails to reduce VaR; in fact, it can increase the aggregate risk (super-additivity).
  • Modeling Guidelines: The introduction of NSD and SD provides a verifiable checklist for practitioners. If a portfolio exhibits negative dependence (NSD) and heavy tails (SD condition), the aggregate VaR will likely exceed the sum of individual VaRs, warning against naive diversification strategies.
• Broader Applicability: By removing the requirement for identical distributions and finite means, the framework is directly applicable to heterogeneous portfolios (e.g., combining operational risk, liability, and catastrophe risks) where margins differ significantly.

Summary Conclusion

Nawaf Mohammed's paper establishes that for one-sided random variables, VaR sub-additivity is an impossibility except in the trivial case of co-monotonicity. Conversely, VaR super-additivity is a natural phenomenon for heavy-tailed, negatively dependent risks, provided the joint distribution satisfies Negative Simplex Dependence and the marginals satisfy Simplex Dominance. This work provides the first comprehensive, non-parametric framework for analyzing VaR aggregation across all probability levels, bridging the gap between theoretical risk measures and the realities of heavy-tailed financial and actuarial data.