---
title: Negative Simplex Dependence (NSD)
url: https://www.emergentmind.com/topics/negative-simplex-dependence-nsd
type: topic
---

# Negative Simplex Dependence (NSD)

Negative Simplex Dependence (NSD) is a form of multivariate negative dependence designed to capture the weakest structural requirement ensuring that the lower tail probability of a sum of random variables does not exceed that under independence. NSD is characterized by a simple yet robust probabilistic inequality involving the sum and marginal distributions. It holds a fundamental position in the classification of extremal negative dependence concepts, acting as a minimal requirement in applications such as risk aggregation, where additivity properties of risk measures—especially Value-at-Risk (VaR)—are of primary interest. NSD links probabilistic orderings, convex geometry of distribution spaces, and supermodular function theory, serving as a bridge between extremal negative dependence, negative association, and independence.

## 1. Formal Definition

Let $X_1,\ldots,X_n$ be real-valued random variables defined on a probability space, all with prescribed continuous marginal distributions $F_{X_i}$. Denote the vector $\mathbf{X} = (X_1,\dots,X_n)$, the sum $S = \sum_{i=1}^n X_i$, and the joint CDF $F_{\mathbf{X}}(x_1,\dots,x_n) = P(X_1 \le x_1, \ldots, X_n \le x_n)$. 

$\mathbf{X}$ is said to satisfy Negative Simplex Dependence (NSD) if, for all $t \in \mathbb{R}$,

$$
F_S(t) \le \prod_{i=1}^n F_{X_i}(t),
$$

where $F_S(t) = P(S \le t)$. In the case where $X_i \ge 0$, the domain is typically restricted to $t \ge 0$ [2512.07787].

This inequality says that, for any threshold $t$, the probability that the aggregated sum falls below $t$ is less than or equal to the probability that all marginals simultaneously fall below $t$ when independent. NSD may be equivalently described via integration over the $n$-simplex:

$$
F_S(t) = \int_{\Delta_t} dF_{\mathbf{X}}(x) \le \prod_{i=1}^n F_{X_i}(t),
$$

where $\Delta_t = \{(x_1,\dots,x_n): x_1 + \cdots + x_n \le t\}$ [2512.07787]. NSD is strictly weaker than negative lower-orthant dependence (NLOD), which requires the joint CDF at all points to be no greater than the product of marginals: $F_{\mathbf{X}}(x_1,\dots,x_n)\le\prod_{i=1}^n F_{X_i}(x_i)$.

## 2. Geometric and Polyhedral Structure

In the case of discrete random vectors, especially for $d$-variate Bernoulli laws with marginal means $p = (p_1, \dots, p_d)$, the set of all joint laws with fixed marginals forms a convex polytope—the Fréchet class $B_d(p)$. NSD corresponds to requiring that the law is smaller than independence in the supermodular order, i.e., $I \preceq_{\text{sm}} I^\perp$ for $I$ a Bernoulli vector and $I^\perp$ independent Bernoullis with means $p_j$ [2504.17679].

The extremal NSD structures—those saturating this inequality—form a lower-dimensional face of $B_d(p)$. Specifically, for Bernoulli vectors, the extreme NSD laws are the $\Sigma$-countermonotonic (Σ-ctm) distributions: these concentrate all their mass on those $i \in \{0,1\}^d$ whose sum $i_\bullet$ is either the integer part $m = \lfloor \sum_j p_j \rfloor$ or $m+1$. The polytope $B_d^\Sigma(p)$ formed by these distributions is an antichain in the supermodular order and is weakly minimal within the Fréchet class [2504.17679].

## 3. Key Properties and Hierarchies of Dependence

NSD establishes an ordering among dependency structures that is weaker than NLOD and negative association, but sufficiently strong to enforce critical risk-aggregation monotonicities:

- **Negative Lower-Orthant Dependence (NLOD):** $F_{\mathbf{X}}(x) \le \prod_i F_{X_i}(x_i)$ for all $x$; implies NSD but not conversely (NSD might fail outside the diagonal).
- **Negative Association:** Requires that for disjoint subsets $A,B$ and componentwise increasing functions $f,g$, $\mathrm{Cov}(f(X_A),g(X_B)) \le 0$. Not directly implied by or implying NSD [2512.07787].
- **Co-Monotonicity (Positive Extremal):** The opposite extreme; the joint CDF is $\min_i F_{X_i}(x_i)$, and $F_S(t)$ is maximized.
- **$\Sigma$-Countermonotonicity:** For every nontrivial partition $I$ of $\{1,\dots,d\}$, the partial sums $(\sum_{j\in I} X_j, \sum_{j\notin I} X_j)$ are bivariate countermonotonic (i.e., minimal in convex sum order).

In the Bernoulli setting, the class $B_d^\Sigma(p)$ represents precisely the set of extreme NSD laws, and is characterized by supporting distributions only on sums $m$ and $m+1$ [2504.17679]. 

## 4. Consequences for Risk Aggregation and Value-at-Risk

A major application of NSD is in establishing bounds for aggregated risk measures. For random variables $X_1,\ldots,X_n \ge 0$ with continuous margins and joint law exhibiting NSD, the following holds for all $p \in (0,1)$:

$$
\mathrm{VaR}_p\left(\sum_{i=1}^n X_i\right) \ge \sum_{i=1}^n \mathrm{VaR}_p(X_i),
$$

where $\mathrm{VaR}_p(Z) = \inf\{z: F_Z(z) \ge p\}$ [2512.07787]. Thus, NSD is both necessary and sufficient for strict super-additivity of VaR for all $p$: the sum’s quantile at any level never falls below the sum of marginals’ quantiles. This property sharply distinguishes NSD from both independence (exactly additive) and positive dependence (sub-additive or even strictly so). 

Examples constructed using countermonotonic continuous (e.g., Pareto/reciprocal) or discrete mutually exclusive distributions provide concrete illustrations [2512.07787].

## 5. Connections with Extremal Negative Dependence Structures

NSD is intimately related to several classical and modern negative-dependence concepts:

- **Countermonotonicity:** The classical bivariate case (d=2); equivalent to NSD along the diagonal.
- **Pairwise Countermonotonicity:** Demands all pairs are countermonotonic; rarely possible in $d>2$ except with restrictive marginals.
- **Joint Mixability:** All mass concentrated on constant-sum configurations; a special sharper case subsumed by NSD.
- **$\Sigma$-Countermonotonicity:** Generalizes countermonotonicity to $d>2$; always exists, coincides with pairwise countermonotonicity and joint mixability whenever those are possible [1512.03232].

Extremal NSD laws correspond to the smallest possible (in convex order) aggregate sums, and provide tight lower bounds in Fréchet classes with fixed marginals.

## 6. Strongly Rayleigh Property and Entropic Structure

Within the NSD/Σ-ctm face of the Bernoulli Fréchet polytope, certain distributions satisfy the strongly Rayleigh property: their multivariate generating polynomial is real stable, implying negative association and, hence, NSD [2504.17679]. The unique maximum-entropy point in $B_d^\Sigma(p)$ is constructed by conditioning independent Bernoulli random variables on the sum being $m$ or $m+1$. This conditional law is both strongly Rayleigh and maximally random at fixed marginals, providing an explicit representative of the strongest negative dependence compatible with NSD.

A canonical mixture (convex combination) of this strongly Rayleigh law and the independence law yields a one-parameter family of distributions interpolating monotonically in the supermodular order from extremal negative dependence to independence.

## 7. Summary and Comparative Table

Negative Simplex Dependence (NSD) offers a natural, minimal, and geometrically tractable framework for quantifying negative dependence in multivariate aggregation problems, especially in risk aggregation and optimization under sum constraints. The table below summarizes NSD relative to related concepts:

| Concept                      | Definition/Characterization                       | Existence/Extremality            |
|------------------------------|---------------------------------------------------|-----------------------------------|
| NSD                          | $F_S(t) \le \prod_i F_{X_i}(t)$                  | Always exists (weakest)           |
| $\Sigma$-countermonotonicity | All partitions’ sums countermonotonic             | Always exists; extremal NSD face  |
| Negative Association         | All increasing set functions negatively correlated| Not directly comparable           |
| Co-monotonicity              | $F_{\mathbf{X}}(x)=\min_i F_{X_i}(x_i)$           | Always exists (positive extremal) |

Each framework fulfills a sharply distinguished role in both theoretical and applied contexts, delineating the principal boundary cases for sum distributions under fixed marginals [1512.03232, 2504.17679, 2512.07787].

Source: https://www.emergentmind.com/topics/negative-simplex-dependence-nsd