Expected Loss and Unexpected Loss

Why Expected Loss and Capital Must Be Distinguished

A loss distribution describes more than the amount an organization expects to pay. It also describes how far actual outcomes may depart from that expectation. These are different planning questions: what amount is expected to be incurred, and what additional buffer is needed for adverse outcomes? Confusing the two can overstate required resources by counting expected losses twice, or understate them by omitting an expected-loss allowance that is not funded elsewhere.

The distinction is aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus. Throughout this tutorial, losses are measured over one stated period, such as one year, and every quantity is on the same loss basis, in the same currency, and before any unmentioned discounting.

The term “total loss” can refer to the realized aggregate loss \(L\), or, in a capital calculation, to a risk-measure amount such as a selected quantile. The context should make clear which meaning is intended. In this tutorial, “total loss at the selected confidence level” means the quantile of \(L\), while “capital” means the quantile less expected loss.

Expected loss is an average across possible outcomes, not a promise that the realized loss will equal that average. Unexpected-loss deviation describes the difference in a particular outcome. Unexpected-loss capital, by contrast, is a fixed amount calculated from a risk measure and a chosen loss distribution. It is not itself a forecast of the next realized deviation.

Quantile-Based Capital and Its Interpretation

For a loss \(L\) with cumulative distribution function \(F_L(x)=P(L\le x)\), use the left-quantile convention

This convention selects the smallest loss amount whose cumulative probability is at least \(p\). Given a confidence level \(p\), a common quantile-based economic-capital convention is

The VaR is the total loss amount at the selected confidence level. Subtracting expected loss expresses the additional amount above the mean. The amount is not defined independently of its inputs: it depends on the loss variable, the period, the probability model, the confidence level, the quantile convention, and whether expected loss is deducted.

Why Subtracting Expected Loss Centers the Quantile

The subtraction has a direct mathematical justification. Let \(m=E[L]\), a constant, and define the centered loss \(Z=L-m\). Its distribution function is

The left \(p\)-quantile of \(Z\) is therefore the smallest \(x\) for which \(F_L(x+m)\ge p\). Writing \(y=x+m\), the smallest such \(x\) is the smallest such \(y\), less \(m\). Thus

This identity relies on subtracting a constant from the loss. It does not say that the quantile of an arbitrary transformed or rescaled loss is found by simply subtracting the mean. It also does not specify whether a particular reporting framework calls the resulting amount “capital”; that convention must be stated.

Expected Loss May Be Funded Separately

An organization may plan for expected losses through pricing, an allowance, or another funding mechanism, and use capital for losses beyond that amount. In that presentation, the capital figure is an increment above expected loss. If a report instead asks for the total resources needed to meet a selected loss quantile, the expected-loss allowance and incremental capital together make up that total. The arithmetic is simple, but the labels matter: an amount described as “capital” may be incremental or may be reported on a gross basis that includes expected losses.

The following example makes the convention explicit. The expected-loss amount is assumed to be funded separately; this is a hypothetical accounting and planning assumption, not a claim about a particular insurer or product.

A Second Discrete Example: Track the Cumulative Probabilities

For a discrete model, the quantile need not be an outcome whose cumulative probability equals the selected confidence level. It is the first outcome at which the cumulative probability reaches or exceeds that level. This distinction can change the capital calculation, particularly when an outcome carries a substantial probability mass.

Common Mistakes and Reporting Checks

• Adding the expected loss incorrectly. Multiply each outcome by its probability and add every contribution. For example, the weighted terms $350, $300, and $500 sum to $1,150, not $1,050.
• Calling the total-loss quantile capital. Under the convention used here, capital is the quantile less expected loss. State both amounts when the distinction matters.
• Subtracting expected loss twice. If capital is already calculated as \(\operatorname{VaR}_p(L)-E[L]\), adding an expected-loss allowance once gives total resources equal to the VaR. Do not add expected loss again to a total-resource figure that already represents the VaR.
• Using the wrong discrete quantile. Apply the stated quantile convention to the cumulative probabilities; do not interpolate between discrete loss outcomes unless a different model or convention specifically calls for it.
• Mixing units or periods. A one-year expected loss in dollars cannot be subtracted from a multi-year quantile in thousands of dollars without first aligning both the period and units.