Comonotonic Losses and Tail Value-at-Risk Additivity

Why Comonotonicity Matters for Portfolio Tail Risk

Dependence determines whether large losses in different parts of a portfolio tend to occur together. In the strongest form of positive alignment, each component is driven by the same underlying risk rank: when that rank is high, every component is at its corresponding high quantile. Such losses are called comonotonic. This structure makes it possible to add the components’ tail value-at-risk (TVaR) directly.

The result is useful when assessing a portfolio exposed to one common severity driver, such as a hypothetical group of coverages whose losses all increase with the same regional hazard index. It is a special dependence result, not a claim that TVaR is additive for every portfolio. The quantile convention used here is $Q_X(p)=\inf\{x:F_X(x)\ge p\}$, for $0<p<1$.

Definition: A collection of losses $X_1,\ldots,X_n$ is comonotonic if there is a single random variable $U\sim\operatorname{Uniform}(0,1)$ such that, for each component, $X_i=Q_{X_i}(U)$ almost surely. Equivalently, all components are nondecreasing functions of the same risk driver. The shared $U$ aligns their ranks; it does not mean their loss amounts or marginal distributions are identical.

A high value of $U$ places every loss high in its own distribution, though not necessarily at the same dollar amount or percentile transformation. For example, if $X_1=Q_{X_1}(U)$ and $X_2=Q_{X_2}(U)$, then the portfolio loss is the nondecreasing function $Q_{X_1}(U)+Q_{X_2}(U)$. This common ordering is the key fact behind the result.

Quantiles and TVaR for a Comonotonic Portfolio

Write $S=\sum_{i=1}^n X_i$ for the aggregate portfolio loss. Under the common-$U$ representation, its value at a given rank $U$ is the sum of the component values at that rank. Consequently, the portfolio quantile is the sum of the component quantiles, apart from any immaterial endpoint or discontinuity details: the equality holds for almost every percentile $p$. “Almost every” is sufficient for TVaR because TVaR averages quantiles over an interval.

Theorem: If $X_1,\ldots,X_n$ are comonotonic and have finite tail expectations, then for $0\le\alpha<1$, $$ \operatorname{TVaR}_{\alpha}(S) =\sum_{i=1}^n\operatorname{TVaR}_{\alpha}(X_i), \qquad S=\sum_{i=1}^n X_i. $$ Here $\operatorname{TVaR}_{\alpha}(X)=\frac{1}{1-\alpha}\int_\alpha^1 Q_X(p)\,dp$. This integral-quantile definition applies to discrete as well as continuous loss distributions.

The assumption of comonotonicity is essential to this theorem. The components must be aligned through a shared increasing risk driver; knowing only their marginal distributions is not enough to establish additivity.

Why the Additivity Formula Holds

The argument follows directly from the common risk rank and the integral definition of TVaR. Let $Q_i$ denote $Q_{X_i}$ and define $g(u)=\sum_{i=1}^n Q_i(u)$. Each $Q_i$ is nondecreasing, so $g$ is nondecreasing. The aggregate loss is $S=g(U)$.

Represent every component with the same rank.
Comonotonicity gives $X_i=Q_i(U)$ almost surely, so $S=\sum_i Q_i(U)=g(U)$.

Identify the aggregate quantile.
Because $g$ is nondecreasing, the distribution of $g(U)$ has quantile $g(p)$ for almost every $p$. Thus $Q_S(p)=\sum_i Q_i(p)$ almost everywhere.

Integrate over the upper tail.
Substitute the aggregate quantile into the integral definition and use linearity of integration.

In equations, the final step is

\operatorname{TVaR}_{\alpha}(S) =\frac{1}{1-\alpha}\int_\alpha^1 Q_S(p)\,dp =\frac{1}{1-\alpha}\int_\alpha^1\sum_{i=1}^n Q_i(p)\,dp =\sum_{i=1}^n\operatorname{TVaR}_{\alpha}(X_i).

A nondecreasing function can have jumps or flat portions, which is why quantile equality is stated almost everywhere rather than relying on endpoint values. Those exceptional percentile points do not change the integral. For discrete distributions, the integral naturally accounts for the appropriate fraction of probability mass at the tail boundary; no special continuous-distribution assumption is needed.

Worked Example: Two Continuous Losses Driven by One Risk Rank

A hypothetical portfolio has annual losses $X=100+200U$ dollars and $Y=50+100U$ dollars, where $U\sim\operatorname{Uniform}(0,1)$. Calculate each loss’s 80% VaR and TVaR, then calculate the portfolio’s 80% TVaR and verify additivity. Use the common-$U$ method.

Component quantiles. Since $U$ is uniform, $Q_X(p)=100+200p$ and $Q_Y(p)=50+100p$. Therefore,

\operatorname{VaR}_{0.80}(X)=100+200(0.80)=260, \qquad \operatorname{VaR}_{0.80}(Y)=50+100(0.80)=130.

For a uniform risk rank, the mean of $U$ over the percentile interval from $\alpha$ to 1 is $(1+\alpha)/2$. At $\alpha=0.80$, this is $(1+0.80)/2=0.90$. Thus,

\operatorname{TVaR}_{0.80}(X)=100+200(0.90)=280, \qquad \operatorname{TVaR}_{0.80}(Y)=50+100(0.90)=140.

Portfolio calculation. The losses share $U$, so they are comonotonic. Their sum is $S=150+300U$, with $Q_S(p)=150+300p$. Its 80% TVaR is

\operatorname{TVaR}_{0.80}(S) =150+300(0.90)=420.

The component total is $280+140=420$, matching the portfolio calculation. The 80% VaR also matches in this example: $260+130=390$, and $Q_S(0.80)=150+300(0.80)=390$. The method avoids treating the two high losses as if they arose from unrelated ranks. All monetary results are exact to the cent: $260.00, $130.00, $280.00, $140.00, and $420.00.

Worked Example: Discrete Losses with Different Thresholds

A hypothetical insurer models two coverages using the same $U\sim\operatorname{Uniform}(0,1)$. Coverage A pays $X=0$ for $U\le0.40$, $X=100$ for $0.40<U\le0.80$, and $X=300$ for $U>0.80$. Coverage B pays $Y=0$ for $U\le0.60$, $Y=50$ for $0.60<U\le0.90$, and $Y=200$ for $U>0.90$. Calculate each 75% TVaR and the 75% TVaR of $S=X+Y$. Apply the integral-quantile definition.

Check the dependence. Both payment functions are nondecreasing functions of the same $U$. The losses are therefore comonotonic even though their payment thresholds differ.

For $X$, the quantile is 100 from percentile 0.40 to 0.80 and 300 from 0.80 to 1. Over the upper tail from 0.75 to 1, the respective percentile widths are $0.80-0.75=0.05$ and $1-0.80=0.20$. Therefore,

\operatorname{TVaR}_{0.75}(X) =\frac{100(0.05)+300(0.20)}{0.25} =\frac{5+60}{0.25}=260.

For $Y$, the quantile is 50 from percentile 0.60 to 0.90 and 200 from 0.90 to 1. The widths within the upper tail are $0.90-0.75=0.15$ and $1-0.90=0.10$. Hence,

\operatorname{TVaR}_{0.75}(Y) =\frac{50(0.15)+200(0.10)}{0.25} =\frac{7.5+20}{0.25}=110.

Aggregate quantiles. Combining the payment levels over the intervals defined by all thresholds gives the following portfolio values: $S=0$ up to $U=0.40$, $S=100$ from 0.40 to 0.60, $S=150$ from 0.60 to 0.80, $S=350$ from 0.80 to 0.90, and $S=500$ from 0.90 to 1. In the 75% tail, the widths are 0.05 at 150, 0.10 at 350, and 0.10 at 500. Thus,

\operatorname{TVaR}_{0.75}(S) =\frac{150(0.05)+350(0.10)+500(0.10)}{0.25} =\frac{7.5+35+50}{0.25}=370.

Additivity checks: $260+110=370$. The calculation also shows why the quantile integral is practical for discrete losses: it weights each tail level by the part of the percentile interval occupied by that level. The results are $260.00, $110.00, and $370.00.

Worked Example: Nonlinear Losses with a Shared Driver

A hypothetical portfolio has $X=120U$ dollars and $Y=80U^2$ dollars, where $U\sim\operatorname{Uniform}(0,1)$. Both losses increase with $U$, so use comonotonicity to calculate their individual and aggregate 50% TVaRs. Round monetary answers to two decimal places only at the end.

The component quantiles are $Q_X(p)=120p$ and $Q_Y(p)=80p^2$. The common rank ensures $Q_S(p)=120p+80p^2$ almost everywhere. Integrate each quantile from 0.50 to 1 and divide by the tail width $1-0.50=0.50$:

\begin{aligned} \operatorname{TVaR}_{0.50}(X) &=\frac{1}{0.50}\int_{0.50}^{1}120p\,dp =2\left[60p^2\right]_{0.50}^{1} =120(1-0.25)=90,\\ \operatorname{TVaR}_{0.50}(Y) &=\frac{1}{0.50}\int_{0.50}^{1}80p^2\,dp =2\left[\frac{80}{3}p^3\right]_{0.50}^{1} =\frac{160}{3}(1-0.125)=\frac{140}{3}. \end{aligned}

The aggregate TVaR is therefore $90+140/3=410/3=136.666\ldots$. Directly integrating the aggregate quantile gives the same check:

\operatorname{TVaR}_{0.50}(S) =2\left(60(1-0.25)+\frac{80}{3}(1-0.125)\right) =2\left(45+\frac{70}{3}\right) =\frac{410}{3}.

Rounded to two decimal places, the component TVaRs are $90.00 and $46.67, and the portfolio TVaR is $136.67. This example uses nonlinear quantile functions; the additivity comes from their shared increasing driver, not from linear loss amounts.

Common Mistake: Applying Additivity Without Comonotonicity

A frequent error is to add component TVaRs merely because the component losses are in the same portfolio. Dependence can instead reduce or increase the portfolio tail relative to that sum. For a simple contrast, let $U\sim\operatorname{Uniform}(0,1)$, $X=U$, and $Y=1-U$. Both marginal losses are uniform on $[0,1]$, but they move in opposite directions and are not comonotonic. For either component, the 90% TVaR is the average of the quantiles from 0.90 to 1:

\operatorname{TVaR}_{0.90}(X) =\operatorname{TVaR}_{0.90}(Y) =\frac{1}{0.10}\int_{0.90}^{1}p\,dp =0.95.

Adding these component values gives 1.90. But the portfolio loss is $X+Y=1$ for every outcome, so $\operatorname{TVaR}_{0.90}(X+Y)=1$, not 1.90. The incorrect calculation is $0.95+0.95=1.90$ as the portfolio TVaR; the correction is to recognize that the losses are oppositely ordered, not comonotonic, and calculate the portfolio loss distribution before applying TVaR. This example is hypothetical and illustrates dependence, not an estimate for a real portfolio.

Key takeaway: Comonotonicity means losses share a common percentile rank. That alignment makes the aggregate quantile the sum of component quantiles almost everywhere, and integrating those quantiles proves TVaR additivity. Do not use the result unless the required common increasing dependence structure is justified.

Check Your Understanding

For each problem, use the stated quantile convention and the integral definition of TVaR where needed.

A hypothetical pair of losses is $X=40+60U$ and $Y=20+30U$, where $U\sim\operatorname{Uniform}(0,1)$. Find their 90% portfolio TVaR. (Answer: $144.00)
Two comonotonic losses have quantiles $Q_X(p)=200p$ and $Q_Y(p)=100p^2$. Find $\operatorname{TVaR}_{0.50}(X)$, $\operatorname{TVaR}_{0.50}(Y)$, and the portfolio TVaR. (Answer: $150.00, $58.33, and $208.33)
A loss $X$ is 0 for $U\le0.70$ and 100 for $U>0.70$; a loss $Y$, driven by the same uniform $U$, is 0 for $U\le0.40$ and 60 for $U>0.40$. Find both 80% TVaRs and the 80% TVaR of their sum. (Answer: $100.00, $60.00, and $160.00)
Let $U\sim\operatorname{Uniform}(0,1)$, $X=U$, and $Y=1-U$. Find the sum of their 80% TVaRs and compare it with the 80% TVaR of their portfolio. (Answer: Component sum $1.80; portfolio TVaR $1.00)
Two losses are $X=50+100U$ and $Y=80+40U$, with the same $U\sim\operatorname{Uniform}(0,1)$. Find the 75% VaR and TVaR for the aggregate loss, and verify the TVaR using the component values. (Answer: VaR $300.00; TVaR $315.00, equal to component TVaRs $225.00 + $90.00)

What You'll Learn