Testing Rate Adequacy Under Catastrophe-Tail Changes

Why a Catastrophe-Tail Change Can Alter a Rate Indication

A catastrophe shift can affect a portfolio in two distinct ways. A higher event frequency or severity may raise expected losses, which directly affects the loss cost. It may also change tail risk, which affects the capital required under a tail-based risk measure. These effects need not move by the same amount: two scenarios can have equal expected losses but different capital needs, or the mean can rise modestly while the tail capital rises sharply.

This tutorial compares hypothetical annual aggregate losses before and after such shifts. It uses a specified premium rule so that the indicated premiums are comparable. That rule is an analytical convention for the examples, not a universal pricing formula. The methods are aligned with the SOA Exam ASTAM (Advanced Short-Term Actuarial Mathematics) syllabus.

Definition: Let $L$ be the aggregate loss for one policy period, and let $F_L$ be its distribution function. For confidence level $\alpha$, define $\operatorname{VaR}_{\alpha}(L)=\inf\{x:F_L(x)\geq\alpha\}$. Define $\operatorname{TVaR}_{\alpha}(L)$ as the average of the quantile function over probabilities from $\alpha$ to 1. In this tutorial, tail capital is $K_{\alpha}=\operatorname{TVaR}_{\alpha}(L)-E[L]$. It measures the tail amount above the mean; it is not the total capital measure $\operatorname{TVaR}_{\alpha}(L)$.

For a discrete loss distribution, the upper tail can include only part of the probability mass at the VaR. If $v=\operatorname{VaR}_{\alpha}(L)$, then:

\operatorname{TVaR}_{\alpha}(L) = \frac{E[L\,\mathbf{1}_{\{L>v\}}]+(F_L(v)-\alpha)v}{1-\alpha}.

The premium rule used below is $P_{\mathrm{ind}}=(E[L]+e+kK_{\alpha})/(1-r)$, where $e$ is a fixed expense allowance per exposure, $r$ is the variable expense proportion of gross premium, and $k$ is the charge per dollar of tail capital. All three are held constant across the scenarios being compared. The rule assumes the indicated gross premium covers expected loss, fixed expenses, and the capital charge, while variable expenses equal $rP_{\mathrm{ind}}$. Rearranging $P_{\mathrm{ind}}=E[L]+e+kK_{\alpha}+rP_{\mathrm{ind}}$ gives the formula. It does not apply unchanged if expenses, risk loads, or the capital definition differ between scenarios.

Calculating Discrete Tail Measures

The quantile-integral definition explains why discrete TVaR is not always the average of all outcomes strictly greater than VaR. The upper-tail probability has size $1-\alpha$; if the probability above VaR is smaller than that, the remaining tail probability must be taken from the mass at VaR.

Find the VaR.
Locate the smallest loss $v$ for which the cumulative probability is at least $\alpha$.

Form the upper tail of size $1-\alpha$.
Include all probability at losses strictly above $v$, then add probability $F_L(v)-\alpha$ at $v$.

Average that tail and subtract the mean for capital.
Divide the tail’s loss total by $1-\alpha$ to obtain TVaR, then calculate $K_{\alpha}=\operatorname{TVaR}_{\alpha}(L)-E[L]$.

The adjustment at VaR follows directly from the quantile integral. The interval from $\alpha$ to 1 has probability length $1-\alpha$. Losses above $v$ occupy probability $1-F_L(v)$, so the remaining length is $(1-\alpha)-(1-F_L(v))=F_L(v)-\alpha$. On that remaining interval, the quantile equals $v$. This yields the displayed formula, including when there is an atom at VaR.

Worked Example: A Higher Catastrophe Frequency Raises Both Mean and Capital

A hypothetical portfolio has a deterministic annual ordinary-loss component of $40. A catastrophe adds $100 if it occurs. In the original scenario, the catastrophe probability is 0.02; in the stressed scenario it is 0.08. Use $\alpha=0.95$, fixed expenses $e=\$5$, variable expense proportion $r=0.20$, and capital charge $k=0.10$. A current gross premium is $70. Compare the indicated premiums and assess the current premium under each scenario.

Method and original scenario. The annual loss is $40 with probability 0.98 and $140 with probability 0.02. Its mean is $0.98(40)+0.02(140)=39.20+2.80=\$42$. The cumulative probability at $40 is 0.98, so $\operatorname{VaR}_{0.95}=\$40$. The upper 5% consists of 2% at $140 and 3% at $40. Thus $\operatorname{TVaR}_{0.95}=[0.02(140)+0.03(40)]/0.05=(2.80+1.20)/0.05=\$80$, and $K_{0.95}=80-42=\$38$.

The indicated premium is $(42+5+0.10(38))/(1-0.20)=(42+5+3.80)/0.80=50.80/0.80=\$63.50$.

Stressed scenario. The loss is $40 with probability 0.92 and $140 with probability 0.08. Its mean is $0.92(40)+0.08(140)=36.80+11.20=\$48$. Since the cumulative probability at $40 is only 0.92, $\operatorname{VaR}_{0.95}=\$140$, and $\operatorname{TVaR}_{0.95}=\$140$. Therefore $K_{0.95}=140-48=\$92$. The indicated premium is $(48+5+0.10(92))/0.80=(48+5+9.20)/0.80=62.20/0.80=\$77.75$.

The mean rises by $\$48-\$42=\$6$, while tail capital rises by $\$92-\$38=\$54$. The indicated premium rises by $\$77.75-\$63.50=\$14.25$. Relative to the original indication, the current premium is $70/63.50-1=0.10236$, or 10.24% above indicated. Relative to the stressed indication, it is $70/77.75-1=-0.09968$, or 9.97% below indicated. Check: the stressed indication is higher even though most years still have only the ordinary loss, because both the probability of catastrophe and the capital charge increase.

Common mistake avoided: attributing the full premium increase to the $6 increase in expected losses. The capital-load contribution also rises: $0.10(54)/(0.80)=\$6.75$, while the expected-loss contribution is $6/0.80=\$7.50$; together they give the $14.25 change.

Same Mean, Different Catastrophe Severity Tail

A severity shift can alter capital even when the catastrophe probability and expected catastrophe severity stay fixed. Averages alone will not identify that change. The following example holds the expected loss constant and compares the upper tail directly.

Worked Example: Severity Dispersion Raises TVaR Without Raising the Mean

A hypothetical portfolio has a deterministic ordinary-loss component of $20. A catastrophe occurs with probability 0.05. In the original scenario, its severity is always $100. In the changed scenario, its severity is $50 or $150 with equal probabilities, independently of the catastrophe occurrence. Use $\alpha=0.975$, $e=\$5$, $r=0.20$, and $k=0.08$. Compare the indications.

Original scenario. Aggregate loss is $20 with probability 0.95 and $120 with probability 0.05. The mean is $0.95(20)+0.05(120)=19+6=\$25$. Since $F_L(20)=0.95<0.975$, VaR is $120. The upper 2.5% is entirely at $120, so TVaR is $120 and $K_{0.975}=120-25=\$95$. The indication is $(25+5+0.08(95))/0.80=(25+5+7.60)/0.80=\$47.00$.

Changed severity scenario. Aggregate loss is $20 with probability 0.95, $70 with probability $0.05(0.5)=0.025$, and $170 with probability 0.025. The mean is $0.95(20)+0.025(70)+0.025(170)=19+1.75+4.25=\$25$, unchanged. The cumulative probability at $70 is 0.975, so VaR is $70. The upper 2.5% is exactly the mass at $170, so TVaR is $170 and $K_{0.975}=170-25=\$145$. The indication is $(25+5+0.08(145))/0.80=(25+5+11.60)/0.80=\$52.00$.

The expected loss is unchanged, but capital rises by $\$145-\$95=\$50$, increasing the indicated premium by $0.08(50)/0.80=\$5.00$. Check: the changed severity has the same conditional mean, $(50+150)/2=\$100$, but a larger upper-tail loss.

Common mistake avoided: concluding that the rate indication cannot change because the expected loss is unchanged. Under the stated rule, the capital charge depends on TVaR-based capital, which does change.

When Frequency and Severity Both Shift

In an actual stress comparison, frequency and severity may change together. Recalculate the full aggregate-loss distribution in each scenario rather than applying a frequency multiplier to a tail measure or assuming the two effects add linearly. These operations can move probability across the chosen VaR threshold, changing which outcomes make up the upper tail.

Worked Example: Separating Expected-Loss and Capital Contributions

A hypothetical annual portfolio has an ordinary-loss component of $100. In the original scenario, a catastrophe adds $200 with probability 0.04. In the changed scenario, it adds $250 with probability 0.06. Use $\alpha=0.95$, fixed expenses $e=\$4$, variable expense proportion $r=0.20$, and capital charge $k=0.05$. Compare indicated premiums and calculate the increase before rounding.

Original scenario. Aggregate loss is $100 with probability 0.96 and $300 with probability 0.04. The mean is $0.96(100)+0.04(300)=96+12=\$108$. The cumulative probability at $100 is 0.96, so VaR is $100. The upper 5% contains 4% at $300 and 1% at $100. Hence TVaR is $[0.04(300)+0.01(100)]/0.05=(12+1)/0.05=\$260$, and capital is $260-108=\$152$. The indicated premium is $(108+4+0.05(152))/0.80=(108+4+7.60)/0.80=\$149.50$.

Changed scenario. Aggregate loss is $100 with probability 0.94 and $350 with probability 0.06. The mean is $0.94(100)+0.06(350)=94+21=\$115$. Since $F_L(100)=0.94<0.95$, VaR is $350, and TVaR is also $350. Capital is $350-115=\$235$. The indicated premium is $(115+4+0.05(235))/0.80=(115+4+11.75)/0.80=130.75/0.80=\$163.4375$, or $163.44 to the nearest cent.

The unrounded increase is $163.4375-149.50=\$13.9375$, which rounds to $13.94. It can also be checked by separating the contributions: the mean change contributes $(115-108)/0.80=\$8.75$, and the capital change contributes $0.05(235-152)/0.80=\$5.1875$. Their sum is $\$13.9375$. Check: the mean rises by $7 while capital rises by $83; the rate increase is therefore driven by both effects.

Common mistake avoided: subtracting rounded indications before calculating the requested unrounded difference. Here the displayed premiums are $163.44 and $149.50, but the difference must be calculated from $163.4375 and $149.50, giving $13.9375, or $13.94.

Interpretation and Exam Traps

A rate-adequacy comparison is meaningful only when the measurement basis is consistent. Use the same policy period, exposure basis, confidence level, capital definition, expense assumptions, and premium rule in both scenarios. If one scenario uses capital equal to TVaR and another uses TVaR less the mean, the premium comparison mixes different quantities.

Do not equate VaR and TVaR. VaR is a threshold; TVaR averages the upper tail and can be much larger.
Do not ignore probability mass at VaR. A discrete distribution may require only a fraction of the probability at that loss to complete the tail of size $1-\alpha$.
Do not infer tail change from mean change alone. A severity distribution can become more dispersed with no change in its mean, while TVaR and capital increase.
Do not confuse a capital amount with a capital load. Under the rule here, $K_{\alpha}$ is an amount and $kK_{\alpha}$ is the premium charge.
Do not round intermediate indications before subtracting. Keep full precision for the comparison and round only the final monetary result to cents.

The practical conclusion should identify both the indicated-premium change and its drivers. State whether expected loss rose, whether tail capital rose, and how the specified pricing rule converts those changes into a gross premium. A concise conclusion that reports only a stressed rate can conceal whether the action is responding to more expected claims, greater tail risk, or both.

Key takeaway: Rebuild the aggregate-loss distribution under each catastrophe scenario, calculate its mean and tail capital using the same definitions, and then apply the same premium rule. The difference between the indications can be decomposed into expected-loss and capital-load contributions.

Check Your Understanding

For each problem, use the stated discrete aggregate-loss distribution and definitions. Monetary answers are in dollars.

A portfolio has loss $30 with probability 0.90 and $100 with probability 0.10. At $\alpha=0.95$, calculate the mean, VaR, TVaR, and capital $K_{\alpha}$. (Answer: Mean $37; VaR $100; TVaR $100; capital $63.)
Aggregate loss is $10 with probability 0.96 and $90 with probability 0.04. At $\alpha=0.95$, calculate TVaR and capital. (Answer: TVaR $74; capital $60.80.)
A portfolio has ordinary loss $20 and catastrophe probability 0.05. Catastrophe severity is always $80 in scenario A, and is $40 or $120 with equal probabilities in scenario B. At $\alpha=0.975$, compare the means and tail capitals. (Answer: Both means are $24; scenario A capital is $76; scenario B capital is $116.)
An indicated premium rule is $P_{\mathrm{ind}}=(E[L]+e+kK_{\alpha})/(1-r)$. Given $E[L]=\$50$, $K_{\alpha}=\$110$, $e=\$6$, $k=0.04$, and $r=0.25$, calculate the indicated premium to the nearest cent. (Answer: $80.53.)
Under the same rule, an original scenario has mean $42 and capital $38; a stressed scenario has mean $48 and capital $92. Given $e=\$5$, $k=0.10$, and $r=0.20$, calculate each indicated premium and the increase. (Answer: Original $63.50; stressed $77.75; increase $14.25.)

What You'll Learn