Why the Risk Horizon Matters
A capital estimate is always tied to a period of measurement. A loss that occurs just after the endpoint may be outside the measured period, while a claim that occurred before the endpoint may still affect the balance sheet even if its payment comes later. The risk horizon makes this boundary explicit: it determines when the measurement begins and ends, and therefore which changes in financial position the capital estimate is intended to capture.
The horizon is not, by itself, a statement about how severe a loss can be or what confidence level should be used. Those are separate choices. A capital measure must specify the horizon alongside its loss basis and risk measure so that the resulting number has an interpretable meaning.
Define the Loss at the Horizon Endpoint
A convenient balance-sheet definition uses the economic value of assets and liabilities at the beginning and end of the horizon. Let \(A_0\) and \(V_0\) be the values of assets and liabilities at the start, and let \(A_h\) and \(V_h\) be their values at the endpoint \(h\). Economic surplus is assets less liabilities. Under the assumption that there are no unadjusted shareholder distributions or capital contributions during the period, the horizon loss is the decrease in economic surplus:
A positive \(L_h\) represents deterioration: assets have fallen, liabilities have risen, or both. A negative value represents an improvement in surplus. If distributions or capital contributions occur, the measurement must adjust for them; otherwise a distribution could be mistaken for a loss, or a contribution could disguise one. All values must also use a consistent valuation basis at the two dates.
The formula is a balance-sheet identity, not a claim that every cash payment is a new economic loss. A payment made to settle an obligation can reduce assets while also reducing the corresponding liability. Its effect on surplus may therefore be small or zero. Conversely, an obligation incurred before the endpoint can affect \(V_h\) even if it has not yet been paid or reported.
How the Endpoint Determines What Is Included
The endpoint is not simply a cutoff for cash transactions. The relevant question is what has changed in economic position between the two measurement dates. An event occurring within the horizon may create an obligation whose payment falls after the endpoint; if that obligation is recognized in the endpoint liability value, it affects the horizon loss. An event occurring after the endpoint is generally outside the horizon, unless information available at the endpoint changes the value of an obligation already within the measurement basis.
The analyst should distinguish at least three dates: when an event occurs, when it is discovered or reported, and when cash is paid. Treating these as interchangeable can omit incurred obligations or count the same loss twice. The model should also state how it handles premiums, expenses, investment income, revaluation, and any changes in assumptions during the period.
State the dates or duration and the values measured at each date.
Identify the assets, liabilities, cash flows, and valuation rules included.
Determine whether an event occurred within the period and whether it changes endpoint values, regardless of when cash is paid.
Compare beginning and endpoint positions, adjusting for external capital flows if necessary.
Use the distribution of the loss defined for this horizon, not a distribution for a different period.
Horizon-Specific Risk Measures
The horizon determines the random variable whose distribution is measured. Write \(L_h\) for the loss over horizon \(h\), and let \(F_h(x)=P(L_h\le x)\) be its cumulative distribution function. For confidence level \(q\), the quantile definition of Value at Risk is:
This definition is especially important for discrete loss distributions: VaR is the smallest loss amount at which cumulative probability reaches or exceeds \(q\). Changing the horizon can change the probabilities and amounts in the loss distribution, so it can change VaR or TVaR even when the confidence level and the chosen risk measure stay fixed. The result should always be labelled with its horizon and loss basis.
Worked Applications
Worked Example: A Discrete Loss Distribution at Two Horizons
A hypothetical technology insurer measures losses from the start of a quarter. There is a 3% probability of a \$600,000 covered event in the first three months and a 6% probability of a \$2,400,000 event in months 4 through 12. The outcomes are mutually exclusive, and there are no other losses. Find the 95% VaR for a three-month horizon and for a twelve-month horizon.
Assumptions and method. Treat the stated event probabilities and loss amounts as the complete discrete loss model. Losses are stated in horizon-date dollars, so no separate interest adjustment is needed; mortality is not relevant. Use the cumulative-distribution definition of VaR, checking the probability at each possible loss amount.
Three-month horizon. Only the first event can occur within this horizon. The loss is \$600,000 with probability 0.03 and zero otherwise. Thus \(P(L_3=0)=1-0.03=0.97\). Since \(F_3(0)=0.97\ge0.95\), the smallest loss at which the cumulative probability reaches 0.95 is zero:
Twelve-month horizon. The probability of no event is \(1-0.03-0.06=0.91\). The cumulative probability at \$600,000 is \(0.91+0.03=0.94\), which is below 0.95. At \$2,400,000 it is \(0.94+0.06=1.00\), which reaches the target. Therefore:
Check and common mistake. The 12-month distribution has probabilities 0.91, 0.03, and 0.06 at losses of \$0, \$600,000, and \$2,400,000, respectively; these sum to 1.00. A common error is to stop at the low loss without checking its cumulative probability: 0.94 is still below 0.95. The longer horizon includes the later event, so the 95th percentile reaches the larger loss.
Worked Example: Measuring a Change in Economic Surplus
A hypothetical insurer measures its economic position over six months. At the start, assets are valued at \$12.00 million and liabilities at \$10.50 million. At the endpoint, assets are valued at \$11.72 million and liabilities at \$11.18 million. The liability increase consists of a \$180,000 revaluation of existing obligations and a \$500,000 present value of newly incurred obligations. Calculate the six-month loss and reconcile it to the balance-sheet values.
Assumptions and method. All figures are endpoint-date values in dollars, with liabilities valued consistently at a 4% effective annual discount rate at both dates. There are no mortality-contingent cash flows, distributions, or capital contributions. The stated endpoint asset value already reflects cash flows and investment changes, so do not add those again. Apply the change-in-surplus identity.
At the start, surplus is:
At the endpoint, surplus is:
The decrease in surplus, and hence the six-month loss, is:
The same result follows by components. Assets declined by \(\$12.00-\$11.72=\$0.28\) million. Liabilities increased by \(\$11.18-\$10.50=\$0.68\) million, which agrees with \(\$0.18+\$0.50=\$0.68\) million. Therefore, total loss is \(\$0.28+\$0.68=\$0.96\) million. The two methods reconcile exactly.
Common mistake avoided. Do not add a claim payment as a separate loss if its effect is already reflected in endpoint assets and liabilities. Doing so would double count a cash flow already captured by the balance-sheet comparison.
Worked Example: An Incurred Claim Paid After the Endpoint
A hypothetical insurer measures a twelve-month horizon. At time zero, assets are \$50.0 million and liabilities are \$45.0 million. At month 12, assets are \$49.0 million and liabilities are \$45.8 million. The endpoint liability includes a \$0.8 million estimate for a covered event that occurred in month 11 but was not reported until month 14. A separate event occurs in month 13; at month 12 there was no information indicating that event. Calculate the measured loss and determine which event enters it.
Assumptions and method. The stated liability values are already present values on a consistent basis using a 3% effective annual discount rate. There are no mortality-contingent obligations, external capital flows, or other adjustments. Use the twelve-month endpoint comparison; the event dates, not the eventual payment dates, determine whether an event occurred within the period.
Beginning surplus is \(\$50.0-\$45.0=\$5.0\) million. Endpoint surplus is \(\$49.0-\$45.8=\$3.2\) million. Thus the measured loss is \(\$5.0-\$3.2=\$1.8\) million. Equivalently, the asset decline is \(\$50.0-\$49.0=\$1.0\) million and the liability increase is \(\$45.8-\$45.0=\$0.8\) million, giving \(\$1.0+\$0.8=\$1.8\) million.
The month-11 event is included: it occurred within the horizon and its estimated obligation is in the month-12 liability, even though it was reported later. The month-13 event is outside this horizon and, by assumption, does not affect the month-12 values. It is not included in this measured loss. The reconciliation confirms the result: the \$0.8 million incurred obligation is already part of the endpoint liability, so it must not be added to the \$1.8 million loss a second time.
Common mistake avoided. Excluding the month-11 event because it was reported after month 12 confuses reporting date with occurrence date and ignores the endpoint reserve. Including the month-13 event would instead move the measurement boundary beyond the stated horizon.
Choosing and Communicating a Horizon
A useful horizon is the one that matches the purpose of the capital estimate and the financial changes the estimate is intended to capture. A short horizon may emphasize changes that can emerge quickly; a longer horizon can include additional events and changes in the value of obligations. Neither is automatically appropriate for every use. The selection should be made before interpreting the resulting capital figure, and the model should be capable of representing the relevant changes over that period.
For a clear specification, document:
- the measurement date and the exact endpoint or duration;
- whether the loss is based on accounting values, economic values, or another stated basis;
- which cash flows and balance-sheet revaluations are reflected in endpoint values;
- how incurred but unpaid or unreported obligations are treated; and
- whether external capital flows are excluded or explicitly adjusted for.
A horizon change is not merely a relabelling of the same estimate. It may alter which events are in scope, how long financial positions can change, and what obligations are recognized at the endpoint. Consequently, results for different horizons should not be compared as though only the units of time had changed. Any conversion or scaling between periods requires assumptions that are not supplied by the horizon definition itself.
Common Horizon Traps
- Using a payment-date cutoff. A claim incurred during the horizon can affect endpoint liabilities even if its cash payment occurs later.
- Ignoring the endpoint valuation. The horizon loss includes changes in the value of obligations, not only cash losses already paid.
- Double counting cash flows. If a payment or receipt is already reflected in endpoint assets and liabilities, do not add it again as a separate loss.
- Comparing risk measures with different horizons as if they were identical. The underlying loss distributions can differ, even if the confidence level is unchanged.
- Leaving the basis implicit. State what is being measured and how values at the two dates are determined.
Check Your Understanding
For each problem, identify the stated horizon and use the loss basis given. Check discrete quantiles by cumulative probability rather than selecting the largest loss automatically.
- A hypothetical lender has a 4% probability of a \$300,000 loss during the first month and no other loss in a one-month horizon. What is its 95% VaR? (Answer: \$0)
- A covered event occurs in month 5 of a six-month horizon and is paid in month 8. The endpoint liability includes its full estimated value. Is the obligation reflected in the six-month loss? (Answer: Yes, through the endpoint liability value)
- At the start of a horizon, assets are \$8.0 million and liabilities are \$7.1 million. At the endpoint, they are \$7.8 million and \$7.25 million. With no external capital flows, what is the loss? (Answer: \$350,000)
- A one-year loss is \$0 with probability 0.90, \$400,000 with probability 0.04, and \$1,500,000 with probability 0.06. What is the 95% VaR? (Answer: \$1,500,000)
- For a six-month horizon, assets decline by \$120,000 and liabilities decline by \$50,000. Assume no external capital flows and consistent endpoint values. What is the economic loss, and is it an improvement or a deterioration in surplus? (Answer: \$70,000 loss, a deterioration in surplus)