What Economic Capital Measures

Why Economic Capital Needs a Clear Definition

An insurer can report a capital figure without explaining what that amount is intended to cover. For risk management, the figure is meaningful only when its purpose and measurement basis are stated. Which losses are included? Over what period are they measured? How much confidence does management require that available resources will be sufficient? Economic capital provides an internal way to connect these choices to a resource requirement.

The central idea is not that there is one universally correct amount of capital. Rather, an insurer specifies a loss measure, a time horizon, and a confidence standard, then estimates the resources needed under that framework. A capital estimate without those specifications is difficult to interpret or compare with another estimate.

Definition: Economic capital is an insurer’s internally determined amount of resources required to support a specified set of modeled losses, over a specified time horizon, at a specified confidence level. The estimate depends on the loss basis, the model and assumptions, the horizon, and the chosen confidence level. It is an internal risk-management measure, not automatically the same as accounting equity or a regulatory capital requirement.

Resources, Losses, and the Confidence Standard

For a simple illustration, suppose an insurer begins with assets $A_0$ and obligations $B_0$, all measured on the same basis and at the same date. Its initial resources in excess of those obligations are $S_0=A_0-B_0$. Let $L_T$ be the modeled loss over horizon $T$. In this simplified setting, the insurer can meet the modeled loss from its initial resources if $L_T\leq S_0$.

An internal target might require that this condition hold with probability at least $\alpha$, where $\alpha$ is the chosen confidence level. Economic capital is then interpreted as the total amount of resources required to meet that target under the selected model and loss definition. If the insurer already has resources available for this purpose, those existing resources must be considered separately from the total requirement.

$$ S_0=A_0-B_0 $$

This simplified balance-sheet relationship assumes that assets and obligations are measured consistently, that the modeled loss reduces the resources available to meet obligations dollar for dollar, and that no other cash flows or balance-sheet changes occur during the horizon. Real internal models may include additional features, but those features must be specified rather than silently assumed.

Why the Resource Test Works

The basic relationship follows from the definition of surplus. After a loss $L_T$, the simplified remaining surplus is $A_0-B_0-L_T$. Resources are sufficient when the remaining surplus is nonnegative. Rearranging that condition gives the loss-coverage test.

A_0-B_0-L_T\geq 0 \quad\Longleftrightarrow\quad L_T\leq A_0-B_0=S_0

To impose a confidence target, require the loss-coverage test to hold with probability at least $\alpha$. In this basic setup, the total required resources are the smallest amount that passes that probability test. This is a justification for interpreting the amount as a resource requirement: whenever resources are below the required level, the model does not meet the stated confidence standard. The calculation is meaningful only for the loss distribution and assumptions actually chosen.

Key distinction: The required economic capital is a total resource target under a stated framework. Existing resources are what the insurer already has available toward that target. If those resources are below the target, the difference is a shortfall; it is not the definition of the total target itself.

Worked Applications

Worked Example: Reading a Capital Requirement From a Simple Balance Sheet

A hypothetical insurer has opening assets of $24 million and obligations of $21 million. Over the next year, its modeled loss $L_1$, in millions of dollars, is 0 with probability 0.70, 2 with probability 0.20, and 5 with probability 0.10. Management chooses a 90% confidence target. Assume the listed outcomes are the complete one-year loss distribution, the assets and obligations are measured on the same basis, losses reduce surplus dollar for dollar, and there are no other cash flows or discounting. There is no mortality assumption in this aggregate loss model. Determine the total resources required under this target and whether the opening resources meet it.

Method. Order the possible losses and add their probabilities until the cumulative probability reaches 90%. Then compare the resulting requirement with the opening resources.

The opening resources are $24-21=3$ million dollars. The cumulative probabilities for the ordered loss outcomes are:

P(L_1\leq 0)=0.70,\qquad P(L_1\leq 2)=0.70+0.20=0.90,\qquad P(L_1\leq 5)=1.00.

A resource amount of $2 million covers losses of $0 million and $2 million, which together have probability 0.90. The required total resources are therefore $2 million. The insurer has $3 million available, so its opening resources exceed the requirement by $3-2=1$ million dollars. Under the stated model and assumptions, the 90% target is met.

Check. The probability of a loss no greater than the available $3 million is 0.90, since only the $5 million outcome exceeds that amount. The opening resources therefore meet the 90% target. The result is consistent with the comparison above.

Common mistake avoided: Selecting the $5 million outcome because it is the worst case would impose a 100% coverage standard, not the specified 90% target. Economic capital reflects the chosen confidence standard, not automatically the maximum modeled loss.

Worked Example: Separating Total Required Resources From a Shortfall

A hypothetical specialty insurer has opening assets of $18 million and obligations of $15 million. For a one-year horizon, its modeled loss $L_1$, in millions of dollars, is 0 with probability 0.60, 3 with probability 0.25, and 7 with probability 0.15. Management requires a 95% confidence level. Assume this is the full loss distribution, opening resources are measured consistently with the loss amounts, losses reduce surplus dollar for dollar, and there are no other cash flows or discounting. There is no mortality assumption. Find the total required resources and any shortfall in existing resources.

Method. Accumulate probabilities from the smallest loss upward to identify the least resource amount that satisfies the confidence target; then compare that amount with current resources.

The opening resources are $18-15=3$ million dollars. The cumulative probabilities are:

P(L_1\leq 0)=0.60,\qquad P(L_1\leq 3)=0.60+0.25=0.85,\qquad P(L_1\leq 7)=0.60+0.25+0.15=1.00.

The cumulative probability is only 0.85 at $3 million, below the 0.95 target. It reaches 1.00 at $7 million, so $7 million is the least resource amount in this discrete set that satisfies the target. The total required resources are $7 million.

The insurer has $3 million available. Its shortfall relative to the requirement is $7-3=4$ million dollars. This comparison does not change the total requirement: $7 million remains the target, while $4 million is the additional amount needed to reach it from the stated opening position.

Check. At $7 million, all modeled outcomes are covered, giving probability 1.00, which is at least 0.95. At $3 million, only 0 and 3 are covered, giving probability 0.85, which is below 0.95.

Common mistake avoided: Reporting $4 million as the economic capital requirement confuses the total requirement with the shortfall after existing resources are considered. The total is $7 million; the shortfall is $4 million.

Worked Example: Changing the Horizon Changes the Capital Question

A hypothetical insurer models the loss in each of three consecutive years as follows: a loss of $0.5 million occurs with probability 0.80, and a loss of $1.5 million occurs with probability 0.20. Assume annual losses are independent and have the same distribution. For this example, the three-year loss is the undiscounted sum of the three annual losses; there are no other cash flows, no mortality assumption, and no adjustment to the annual loss distribution. Determine the total resources needed for a 95% confidence target over three years.

Method. Count the number $N$ of years in which the $1.5 million outcome occurs. The three-year total loss is $0.5 million in each year, plus an additional $1 million for each such higher-loss year. Independence gives a binomial distribution for $N$.

Here $N$ has three independent trials, each with probability 0.20 of a higher-loss year. Thus $P(N=k)=\binom{3}{k}(0.20)^k(0.80)^{3-k}$. The probabilities are:

\begin{aligned} P(N=0)&=(0.80)^3=0.512,\\ P(N=1)&=3(0.20)(0.80)^2=0.384,\\ P(N=2)&=3(0.20)^2(0.80)=0.096,\\ P(N=3)&=(0.20)^3=0.008. \end{aligned}

The total loss is $1.5+N$ million dollars, because the three annual base losses sum to $3(0.5)=1.5$ million dollars. Therefore, the possible three-year totals and cumulative probabilities are:

\begin{array}{c|c|c} \text{Three-year loss (millions)} & \text{Probability} & \text{Cumulative probability}\\ 1.5 & 0.512 & 0.512\\ 2.5 & 0.384 & 0.896\\ 3.5 & 0.096 & 0.992\\ 4.5 & 0.008 & 1.000 \end{array}

At $2.5 million, the cumulative probability is 0.896, below 0.95. At $3.5 million, it is 0.992, at least 0.95. The required resources for the stated three-year target are therefore $3.5 million.

Check. The probability of a three-year loss no greater than $3.5 million is $0.512+0.384+0.096=0.992$. The next lower possible loss amount, $2.5 million, covers only 0.896 of outcomes. Thus $3.5 million is the least listed amount that satisfies the target.

Common mistake avoided: Using a one-year loss distribution to answer a three-year question ignores the accumulation of losses across the specified horizon. The result also depends on the independence assumption; a different dependence structure could change the three-year loss distribution.

What Must Be Specified Alongside the Estimate

An economic capital figure is conditional on a modeling framework. A useful statement of the estimate identifies its principal choices rather than presenting the amount alone.

Loss basis: State which losses and cash flows are included and how they are measured. A model of underwriting losses, for example, does not automatically include asset losses.
Horizon: Specify the period over which losses accumulate. A one-year requirement and a three-year requirement answer different questions.
Confidence level: State the probability standard management selected. A higher standard generally requires resources that cover more of the modeled loss distribution.
Model assumptions: Identify material assumptions such as dependence between risks, event frequency, and whether losses are discounted. An estimate is only as informative as its assumptions.
Resource basis: Explain what assets or other resources are treated as available to meet losses, and compare existing resources with the total requirement without conflating them.

These points also explain why two capital estimates may differ without either calculation being arithmetically wrong. They may use different horizons, confidence levels, loss definitions, or assumptions. Comparisons are meaningful only after those choices have been aligned or their differences made explicit.

Common Interpretation Traps

Leaving out the horizon. A capital amount stated without its measurement period is incomplete. Losses accumulated over multiple years need not have the same distribution as losses over one year.
Treating a confidence level as a guarantee. A 95% target describes a probability under the chosen model. It does not guarantee that losses will never exceed the resources or that the model captures every possible outcome.
Confusing total requirement and shortfall. Existing resources affect how much additional capital may be needed, but they do not redefine the total amount required by the chosen target.
Assuming the estimate is universal. Economic capital is internal and framework-dependent. It is not automatically accounting equity, regulatory capital, or a rating agency requirement.
Using a different loss basis without noticing. A result for one modeled set of losses cannot be interpreted as a requirement for risks that were excluded from that model.

Key takeaway: Economic capital is an internally chosen resource requirement tied to a defined loss basis, time horizon, model, and confidence level. Always state those choices, then distinguish the total requirement from the resources already available to meet it. This treatment is aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus.

Check Your Understanding

For each problem, use the stated horizon and assumptions. Treat the listed loss outcomes as complete distributions, and identify total required resources separately from any shortfall.

A hypothetical insurer has assets of $12 million and obligations of $10 million. Its one-year loss is $0 million with probability 0.75, $1 million with probability 0.15, or $4 million with probability 0.10. At a 90% confidence target, what are its opening resources and the total requirement? (Answer: $2 million opening resources; $1 million total requirement.)
A hypothetical insurer has $5 million of opening resources. Its one-year loss is $0 million with probability 0.50, $2 million with probability 0.30, or $6 million with probability 0.20. What is the total requirement at a 90% confidence target, and what is the shortfall? (Answer: $6 million total requirement; $1 million shortfall.)
In a two-year model, annual losses are independent; each year the loss is $0 million with probability 0.80 or $2 million with probability 0.20. Losses are summed without discounting or other cash flows. What resources meet a 95% confidence target for the two-year total? (Answer: $2 million.)
A hypothetical insurer has assets of $16 million and obligations of $13 million. Its one-year losses are $0 million with probability 0.70, $2 million with probability 0.20, and $5 million with probability 0.10. At a 95% target, determine the total requirement and shortfall. (Answer: $5 million total requirement; $2 million shortfall.)
Two internal estimates use the same modeled loss outcomes, but one covers one year at a 90% confidence level and the other covers three years at a 95% confidence level. Can the amounts be compared directly as if they measured the same requirement? (Answer: No; the horizons and confidence levels differ, so they represent different targets.)

What You'll Learn