Translating Forecasts Into Internal Capital Needs

From a Business Forecast to a Capital Requirement

A business plan forecasts more than premiums, policies, or assets. It also implies future exposure to loss. To translate that forecast into an internal capital need, an insurer must specify what losses it expects, how uncertain those losses are, the horizon and risk measure being used, and whether expected losses are already funded outside capital. The result is a required amount under a stated measurement basis; it is not the same as the financial resources the insurer actually has available.

This tutorial uses a one-year loss horizon and, unless stated otherwise, defines internal capital as unexpected loss at a selected confidence level: the loss quantile less expected loss. A company could use a different internal risk measure or define capital on a different accounting basis. Comparisons are meaningful only when the measure, horizon, valuation basis, and treatment of expected losses are consistent.

Definition: Let $L_t$ be aggregate loss over the chosen horizon under the projected exposures and loss assumptions at forecast date $t$. At confidence level $q$, define value at risk by $\operatorname{VaR}_q(L_t)=\inf\{x:P(L_t\le x)\ge q\}$. Under the unexpected-loss convention used here, required internal capital is $$ C_t=\operatorname{VaR}_q(L_t)-E[L_t]. $$ Expected losses are therefore not counted again as capital. If an organization instead defines its required amount as the full loss quantile, it must use that convention consistently and explain how expected losses are funded.

Scaling Exposure and Loss Variability

A forecast provides projected exposure measures, such as policy counts, insured values, or transaction volumes. A loss model translates those measures into a distribution of aggregate losses. For a transparent starting model, suppose there are $n$ homogeneous exposures, each with a one-year loss having mean $\mu$ and standard deviation $\sigma$. Assume the losses are independent and have the same distribution. If $L$ is their aggregate, then

E[L]=n\mu,\qquad \operatorname{Var}(L)=n\sigma^2,\qquad \operatorname{SD}(L)=\sigma\sqrt{n}.

The mean grows in proportion to exposure count, while the standard deviation grows in proportion to the square root of that count. These formulas require independence and homogeneous per-exposure assumptions. If risks are dependent or materially different from one another, the aggregate distribution must reflect that; the simple square-root scaling is not automatically valid.

Normal-model result: If aggregate loss is modeled as normal with mean $n\mu$ and standard deviation $\sigma\sqrt{n}$, then for $z_q=\Phi^{-1}(q)$, $$ \operatorname{VaR}_q(L)=n\mu+z_q\sigma\sqrt{n},\qquad C=z_q\sigma\sqrt{n}. $$ This result depends on the normal model (or a stated normal approximation) and on using the unexpected-loss convention. It does not imply that every individual loss is normally distributed.

The reason for the capital formula is direct. Under the normal model, a loss quantile at confidence level $q$ is the mean plus $z_q$ standard deviations. Subtracting the expected loss removes the mean:

\operatorname{VaR}_q(L)-E[L] =\bigl(n\mu+z_q\sigma\sqrt{n}\bigr)-n\mu =z_q\sigma\sqrt{n}.

This derivation also shows why projected expected losses and required capital should be tracked separately. A change in expected loss can be material to pricing, earnings, or provisions without changing unexpected-loss capital by the same amount.

Worked Example: Estimating Capital for a Projected Portfolio

A hypothetical insurer projects 625 homogeneous contracts for the next year. For each contract, the one-year loss has mean $1,200 and standard deviation $3,000. Assume losses are independent, aggregate loss is modeled as normal, and the internal target is the 97.725% quantile, for which $z_q=2$. Expected losses are funded separately. Estimate aggregate expected loss, the loss quantile, and required internal capital.

Method and assumptions. Use the independent, homogeneous exposure model and the normal quantile formula. The exposure count is $n=625$, the per-contract mean is $\mu=\$1{,}200$, and the per-contract standard deviation is $\sigma=\$3{,}000$.

Step 1: Calculate the aggregate mean. $$ E[L]=625(\$1{,}200)=\$750{,}000. $$

Step 2: Calculate aggregate standard deviation. Since $\sqrt{625}=25$, $$ \operatorname{SD}(L)=\$3{,}000(25)=\$75{,}000. $$

Step 3: Calculate the loss quantile. $$ \operatorname{VaR}_{0.97725}(L) =\$750{,}000+2(\$75{,}000) =\$900{,}000. $$

Step 4: Remove expected loss to obtain capital. $$ C=\$900{,}000-\$750{,}000=\$150{,}000. $$ The projected unexpected-loss capital need is $150,000, while the expected loss is $750,000. As a check, the loss quantile equals expected loss plus capital: $750,000 + $150,000 = $900,000.

Common mistake avoided: Multiplying the per-contract standard deviation by 625 would give $1,875,000, which is not the aggregate standard deviation under independence. Variances, not standard deviations, add; the correct standard deviation is $3{,}000\sqrt{625}=\$75{,}000$.

Separating Exposure Changes from Assumption Changes

When a forecast changes, the resulting movement in required capital can have more than one cause. The projected number of exposures may change; the assumed loss distribution per exposure may change; or both may change. A useful way to understand the forecast is to recalculate capital in stages, changing one input group at a time. This is a decomposition of the model result, not a claim that every underlying business change has only one cause.

Worked Example: Bridging a Change in Projected Capital

A hypothetical insurer previously modeled 400 independent, homogeneous exposures. The assumed loss per exposure had mean $1,000 and standard deviation $2,500. The new forecast has 625 exposures, with per-exposure mean $1,200 and standard deviation $3,000. Aggregate loss is modeled as normal, the target is the 97.725% quantile ($z_q=2$), and expected losses are funded separately. Decompose the change in required capital into an exposure-count step and a loss-assumption step.

Method. First calculate capital under the earlier inputs. Next change only the exposure count. Finally apply the new per-exposure loss assumptions. At each step, use $C=z_q\sigma\sqrt{n}$.

Baseline. With $n=400$, $\sqrt{n}=20$, and $\sigma=\$2{,}500$, $$ C_{\text{baseline}}=2(\$2{,}500)(20)=\$100{,}000. $$ The baseline expected loss is $400(\$1{,}000)=\$400{,}000$, so its modeled loss quantile is $500,000.

Change exposure count only. Keep $\sigma=\$2{,}500$, but set $n=625$, for which $\sqrt{n}=25$: $$ C_{\text{exposure step}}=2(\$2{,}500)(25)=\$125{,}000. $$ The exposure-count step adds $\$125{,}000-\$100{,}000=\$25{,}000$ of required capital.

Apply the new loss assumptions. At $n=625$, use $\sigma=\$3{,}000$: $$ C_{\text{projected}}=2(\$3{,}000)(25)=\$150{,}000. $$ The change in per-exposure variability adds $\$150{,}000-\$125{,}000=\$25{,}000$. Total required capital rises by $50,000, from $100,000 to $150,000.

The projected expected loss is $625(\$1{,}200)=\$750{,}000$, so the projected loss quantile is $750,000 + $150,000 = $900,000. This also checks the final calculation. The rise in expected loss from $400,000 to $750,000 is separate from the $50,000 increase in unexpected-loss capital.

Common mistake avoided: Treating the $350,000 increase in expected loss as an equal increase in capital confuses the loss mean with the uncertainty margin. Under the stated convention and model, capital is determined by the aggregate standard deviation and confidence multiplier.

Required Capital and Available Capital Are Different Quantities

Required capital is derived from the chosen risk measure applied to projected losses. Available capital is derived from the projected financial position: the resources that the organization counts as eligible to absorb losses under its stated basis. For example, an internal projection might define available capital as eligible assets less recognized liabilities and other specified adjustments. That calculation does not change the loss distribution or the required-capital formula.

A forecast can therefore show an increase in both amounts, an increase in one while the other falls, or movements of different sizes. The difference between available and required capital at a forecast point can be calculated, but it should not be mistaken for a change in the requirement itself. The selected risk measure and the method used to value available resources should be stated separately.

Worked Example: Calculating the Need Separately from Available Resources

At a forecast date, a hypothetical insurer models one-year aggregate loss as normal with mean $2,400,000 and standard deviation $550,000. Its target is the 99% quantile, using $z_{0.99}=2.326$. Expected losses are funded separately. The projected balance sheet has eligible assets of $8,600,000 and recognized liabilities of $7,150,000, with no further adjustments. Calculate required capital and available capital, then compare the two amounts.

Method. Calculate unexpected-loss capital from the loss model. Calculate available capital independently as eligible assets less recognized liabilities.

Step 1: Calculate required capital. $$ C=2.326(\$550{,}000)=\$1{,}279{,}300. $$ The corresponding loss quantile is the mean plus this capital: $$ \operatorname{VaR}_{0.99}(L) =\$2{,}400{,}000+\$1{,}279{,}300 =\$3{,}679{,}300. $$

Step 2: Calculate available capital. $$ \text{Available capital} =\$8{,}600{,}000-\$7{,}150{,}000 =\$1{,}450{,}000. $$

Step 3: Compare the amounts. $$ \$1{,}450{,}000-\$1{,}279{,}300=\$170{,}700. $$ On these stated assumptions, available capital exceeds the modeled requirement by $170,700. The required amount remains $1,279,300; it is not $3,679,300 because this example defines expected losses as separately funded.

Common mistake avoided: Subtracting recognized liabilities twice, or adding the expected loss to required capital after using the unexpected-loss convention, would mix measurement bases. Keep the loss-model calculation and the financial-resource calculation distinct, then compare amounts measured on compatible bases.

Practical Checks Before Reporting a Forecast

A capital estimate is only as interpretable as the assumptions behind it. For each forecast result, document the exposure measure and projection date, the loss horizon, the loss distribution and parameters, the dependence assumptions, the confidence level or other risk measure, and the treatment of expected losses. If the model uses a normal approximation, identify it as an approximation and avoid presenting its output as exact.

A staged calculation can help explain what changed, but the decomposition depends on the order in which inputs are changed. If exposure count and loss assumptions interact, a different ordering can allocate the total movement differently between steps, even though the final projected capital is unchanged. State the bridge convention used rather than implying that the component attribution is unique.

Check the unit of exposure. A count of contracts, insured locations, or transactions must match the unit used to calibrate the per-exposure loss assumptions.
Check aggregation. Under independence, variances add. With dependence, the aggregation model must include the relevant relationships.
Check the capital convention. Confirm whether expected losses are funded separately or included in the stated required amount.
Check the confidence level and horizon. A quantile from one horizon or confidence level cannot be compared directly with a requirement on another basis without adjustment.
Check available resources separately. A change in balance-sheet resources is not itself a change in the modeled loss requirement.

Key takeaway: Forecast exposures and loss assumptions determine a prospective distribution of aggregate loss; the selected risk measure converts that distribution into a required amount. Available capital comes from projected financial resources. Report both calculations and their assumptions distinctly.

Check Your Understanding

For each problem, use the stated unexpected-loss convention and show the intermediate steps before checking the final answer.

A hypothetical insurer projects 900 independent contracts. Each contract has one-year loss mean $500 and standard deviation $4,000. Aggregate loss is modeled as normal and the target uses $z_q=2$. Find expected loss, required capital, and the loss quantile. (Answer: expected loss $450,000; required capital $240,000; loss quantile $690,000.)
A portfolio has 256 independent, homogeneous exposures with per-exposure mean loss $750 and standard deviation $2,000. Use a normal aggregate model and $z_q=2.326$. Calculate the required capital and loss quantile. (Answer: required capital $74,432; loss quantile $266,432.)
A baseline model has 400 independent exposures and per-exposure standard deviation $1,800. The forecast has 625 exposures and standard deviation $2,000. Use $z_q=2$. First change exposure count alone, then change the standard deviation. Find the baseline capital, each bridge amount, and the total increase. (Answer: baseline $72,000; exposure-only capital $90,000, an increase of $18,000; projected capital $100,000, a further increase of $10,000; total increase $28,000.)
At a forecast date, aggregate loss is normal with mean $1,800,000 and standard deviation $400,000. Use $z_q=2$, with expected loss funded separately. Available capital is $2,550,000. Calculate required capital, the loss quantile, and the excess of available over required capital. (Answer: required capital $800,000; loss quantile $2,600,000; excess available capital $1,750,000.)
Two hypothetical business segments have independent aggregate losses. Segment A has mean $300,000 and standard deviation $100,000; segment B has mean $200,000 and standard deviation $150,000. The combined loss is modeled as normal and $z_q=2$. If available capital is $900,000, calculate combined required capital and the excess over required capital. (Answer: combined standard deviation approximately $180,277.56; required capital approximately $360,555.13; excess approximately $539,444.87.)

What You'll Learn