Reserves and Capital for Losses

Two Different Amounts for Two Different Purposes

An insurer estimating future claims faces two related but distinct questions. What amount of claims should it expect to pay? And what additional resources would help it withstand claims that are higher than expected? The first question concerns a provision for expected claims; the second concerns capital for adverse deviations. Treating these amounts as interchangeable can lead to an incorrect view of how much protection is available.

This tutorial uses a simplified one-period model to isolate the distinction. It is aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus. In an actual reporting or risk-management framework, the required claims provision can also depend on the applicable measurement rules, expenses, discounting, risk adjustments, and the definition of the claims being measured. Those features are excluded here so that the roles of expected losses and capital can be seen separately.

Definition: Let $X$ be aggregate claims over a specified period, measured on a specified loss basis. In the simplified model, the expected-claims provision is $R=E[X]$. For a selected confidence level $\alpha$, define $q_\alpha(X)=\inf\{x:F_X(x)\ge\alpha\}$. The quantile-based capital buffer for adverse deviations is $K_\alpha=q_\alpha(X)-E[X]$, when this difference is nonnegative. Thus $R$ covers the modeled mean claim amount, while $K_\alpha$ is the additional amount from the mean to the selected loss quantile.

The model is deliberately specific: $X$ is the aggregate claim loss on the chosen basis, $R$ is set equal to its expected value, and the capital measure is defined as a loss quantile less that expected value. This is one convention for expressing economic capital; some settings instead report the gross quantile itself as a required amount. Always check which convention is being used before comparing figures.

The period and loss basis must match. For example, an expected-claims provision for one year should not be combined with a capital amount based on a different horizon or a different definition of loss. The earlier material on expected loss and unexpected loss distinguishes the mean from deviations around it; here, that distinction is translated into two resource amounts.

Why the Provision and Buffer Add to a Loss Threshold

Under this model, the amount $R+K_\alpha$ is intended to reach the selected loss quantile. This follows directly from the definitions:

R+K_\alpha =E[X]+\left(q_\alpha(X)-E[X]\right) =q_\alpha(X).

The justification is a simple decomposition. Start with resources equal to the expected claims, then add the amount by which the selected quantile exceeds that expectation. The expected amount cancels in the sum, leaving the quantile. Because $q_\alpha(X)$ is defined as the smallest loss value at which the cumulative distribution reaches at least $\alpha$, the resulting amount satisfies $P(X\le R+K_\alpha)\ge\alpha$. For a discrete loss distribution, this probability can be greater than $\alpha$ because the cumulative distribution can jump at the quantile.

This relationship depends on the stated convention. If a reported capital figure already equals $q_\alpha(X)$, adding a separate expected-claims provision to it would not reproduce this decomposition; it would count the expected amount twice. The labels alone are not enough: determine whether capital means a buffer above expected loss or a gross loss threshold.

Worked Example: Separating Expected Claims From a 90% Buffer

A hypothetical insurer models one-year aggregate claims, in thousands of dollars, as follows. There are no expenses, investment effects, or recoveries in this example.

Aggregate claims $X$	Probability
0	0.60
100	0.30
400	0.10

Calculate the expected-claims provision, the 90% quantile-based capital buffer, and their combined amount.

Step 1: Calculate expected claims. Multiply each possible loss by its probability and add:

$$ R=E[X]=(0)(0.60)+(100)(0.30)+(400)(0.10) =0+30+40=70. $$

The provision is therefore $70,000. The amounts in this calculation are in thousands of dollars.

Step 2: Find the 90% quantile. The cumulative probabilities are $F_X(0)=0.60$, $F_X(100)=0.60+0.30=0.90$, and $F_X(400)=1.00$. The smallest loss with cumulative probability at least 0.90 is $q_{0.90}(X)=100$, or $100,000.

Step 3: Calculate capital above the provision.

K_{0.90}=q_{0.90}(X)-E[X]=100-70=30.

The buffer is $30,000. Combined, the provision and buffer are $70+30=100$ thousand dollars, exactly the 90% quantile. As a check, $P(X\le100)=0.60+0.30=0.90$.

Method and mistake avoided. The method is to calculate the mean first, then subtract it from the selected quantile. A common mistake is to call the entire $100,000 quantile “capital” under this buffer convention and then add the $70,000 provision. That would produce $170,000, counting the expected claims twice.

Reading the Two Amounts in a Claims Model

A provision and capital respond differently when the claims distribution changes. The provision reflects the model’s central expected amount. The capital buffer reflects the distance between that expectation and the selected adverse-loss threshold. A change in either the expected amount or the shape of the distribution can affect the buffer; the two figures should therefore be calculated from the same modeled loss distribution rather than estimated independently without reconciliation.

The distribution also matters more than any single “average claim” figure. Two portfolios can have the same expected claims but different upper-tail outcomes, and therefore different quantile-based buffers. Conversely, a change in expected claims does not automatically imply the same change in capital. The calculation depends on the selected threshold relative to the revised mean.

Worked Example: A Different Tail Produces a Different Buffer

A hypothetical insurer has a one-year aggregate-claims model, in thousands of dollars, with the following outcomes. Assume the distribution is the complete model for this example and ignore discounting, expenses, and recoveries.

Aggregate claims $X$	Probability
50	0.50
110	0.30
230	0.20

Find the expected-claims provision and the 80% quantile-based capital buffer.

Step 1: Calculate the expected claims.

$$ R=E[X]=(50)(0.50)+(110)(0.30)+(230)(0.20) =25+33+46=104. $$

The provision is $104,000.

Step 2: Determine the 80% quantile. The cumulative probability at 50 is 0.50. At 110 it is $0.50+0.30=0.80$. Thus $q_{0.80}(X)=110$, or $110,000.

Step 3: Subtract the provision from the quantile.

K_{0.80}=110-104=6.

The capital buffer is $6,000, and the combined provision and buffer equal $104+6=110$ thousand dollars. The check is $P(X\le110)=0.50+0.30=0.80$.

Method and mistake avoided. This calculation uses the cumulative probabilities to locate the quantile before finding the difference from the mean. A tempting but incorrect shortcut is to treat the 20% probability on the largest outcome as the capital amount. Probability is not a dollar buffer: the buffer is the difference between $110,000 and the $104,000 expected amount.

Model Changes Can Move Both Amounts

A change in a claims assumption can alter both the central estimate and the adverse-deviation buffer. The following example applies a deterministic severity uplift to each positive claim outcome while holding the probabilities fixed. It does not assume that every real-world change affects all claim sizes in this uniform way; that is simply the hypothetical model specified here.

Worked Example: Recalculating After a Severity Uplift

A hypothetical one-year claims model has losses, in thousands of dollars, of 0 with probability 0.70, 100 with probability 0.20, and 300 with probability 0.10. A revised assumption increases every positive loss by 10% and leaves the probabilities unchanged. Calculate the original and revised expected-claims provisions and 90% buffers. Ignore all other changes.

Step 1: Calculate the original provision.

R_{\text{old}}=(0)(0.70)+(100)(0.20)+(300)(0.10) =0+20+30=50.

The original provision is $50,000. The original cumulative probabilities are 0.70 at zero and $0.70+0.20=0.90$ at 100, so $q_{0.90,\text{old}}=100$. Therefore,

K_{0.90,\text{old}}=100-50=50.

The original buffer is $50,000, and the combined amount is $100,000.

Step 2: Apply the stated uplift to positive outcomes. The revised losses are 0, $100(1.10)=110$, and $300(1.10)=330$ thousand dollars. The probabilities remain 0.70, 0.20, and 0.10.

Step 3: Calculate the revised provision.

R_{\text{new}}=(0)(0.70)+(110)(0.20)+(330)(0.10) =0+22+33=55.

The revised provision is $55,000. Since the cumulative probability at 110 is $0.70+0.20=0.90$, the revised 90% quantile is $110,000. Hence,

K_{0.90,\text{new}}=110-55=55.

The revised buffer is $55,000, and the combined amount is $55+55=110$ thousand dollars. The check is that the selected revised loss threshold is $110,000 and the probability of claims at or below it is 0.90.

Method and mistake avoided. Rebuild the loss outcomes under the revised assumption, then recalculate both the mean and the quantile. Applying the 10% uplift only to the provision would be incomplete: the positive loss outcomes, including the quantile, also change under the stated model.

Common Interpretation Traps

Adding a gross threshold to an expected-claims provision. If capital is defined as $q_\alpha(X)-E[X]$, then $R+K_\alpha=q_\alpha(X)$. Adding $q_\alpha(X)$ itself to $R$ double-counts the expected amount.
Assuming every reserve is exactly the modeled mean. That equality is an assumption of this simplified tutorial model, not a universal statement about every claims liability measurement.
Mixing loss bases or periods. A one-year buffer cannot be reconciled to a provision for another period without an appropriate model connecting them.
Confusing a probability with a dollar amount. The probability identifies where the quantile lies in the cumulative distribution. The capital buffer is a difference in loss amounts.
Ignoring discrete jumps. For a discrete model, the probability at or below a selected quantile can exceed the specified confidence level. Use the defined quantile rather than interpolating without a stated convention.

Key takeaway: On a consistent loss basis, the simplified expected-claims provision is $E[X]$, while quantile-based capital for adverse deviations is the selected loss quantile less $E[X]$. The provision and buffer have distinct roles, and their sum equals the selected loss threshold under this convention.

Check Your Understanding

For each hypothetical one-period model, use the stated convention: provision $R=E[X]$, buffer $K_\alpha=q_\alpha(X)-E[X]$, and combined amount $R+K_\alpha$. Amounts are in thousands of dollars unless stated otherwise.

A claims model has losses 0, 40, and 120 with probabilities 0.50, 0.30, and 0.20. Find the expected-claims provision, the 80% quantile, and the buffer. (Answer: $R=36$, $q_{0.80}=40$, $K_{0.80}=4$ thousand dollars.)
A model has losses 20, 60, and 180 with probabilities 0.25, 0.50, and 0.25. Find the provision and the 90% buffer. (Answer: $R=80$, $q_{0.90}=180$, $K_{0.90}=100$ thousand dollars.)
A hypothetical portfolio has expected one-year claims of $90,000 and a 95% loss quantile of $140,000. Under the buffer convention, calculate capital and the combined amount. (Answer: capital $=$50,000$; provision plus capital $=$140,000$.)
A model has losses 0, 50, and 150 with probabilities 0.60, 0.30, and 0.10. Every positive outcome is then increased by 20%, with probabilities unchanged. Find the revised expected provision and 90% buffer. (Answer: revised outcomes are 0, 60, and 180; $R=36$, $q_{0.90}=60$, and $K_{0.90}=24$ thousand dollars.)
A hypothetical model has expected claims of $75,000 and a 90% quantile of $105,000. A report instead lists $105,000 as gross capital and proposes adding the $75,000 provision. Under the buffer convention, identify the capital amount and the error in that proposal. (Answer: the buffer is $30,000 and the combined amount is $105,000; adding $75,000 to the gross quantile incorrectly double-counts expected claims.)

What You'll Learn