Reconciling Hybrid Model Components

Why Hybrid Models Need a Reconciliation

A hybrid capital model combines detailed component models with an enterprise-level view. That design can use granular information where it is available and still recognize risks or adjustments that do not fit neatly within a component. But the outputs cannot simply be added together. A component and an enterprise adjustment may cover the same exposure, use different valuation dates, or represent different loss bases. Unless those differences are addressed, the reported total may be internally inconsistent.

Reconciliation is the disciplined process of putting model outputs on a common basis, identifying duplicated or omitted amounts, and documenting the bridge to the enterprise result. It does not, by itself, resolve how dependence between risks should be modeled or establish that a chosen capital measure is appropriate. It makes the inputs to the hybrid result interpretable and comparable.

Definition: A hybrid-model reconciliation is a documented bridge from component outputs to an enterprise-level result. The bridge identifies the scope, loss basis, valuation date, time horizon, and calibration of each input; removes identified duplication; adds justified enterprise adjustments; and reports any remaining difference. Inputs should be combined only when their meanings and measurement bases are compatible.

Establish the Common Basis Before Combining Amounts

A reconciliation starts by specifying what the enterprise result is intended to measure. For example, the basis might be a one-year loss measure, net of specified reinsurance recoveries, after tax, valued at the reporting date, and evaluated under a stated calibration. The precise basis depends on the model’s purpose. The important point is that every component and adjustment must either already match it or have an explicit, supported bridge to it.

At a minimum, record these features for every input:

Scope: which entities, portfolios, perils, and types of loss are included?
Loss basis: is the amount gross or net of recoveries, and does it include expenses or other cash flows?
Valuation: is it before or after tax, and what valuation date and discounting convention apply?
Timing: what is the measurement horizon, and when do the cash flows occur?
Calibration: what scenario, confidence level, or other severity basis produced the output?

These are not interchangeable labels. For example, discounting a payment from year two to the reporting date changes its present value; it does not convert a two-year risk measure into a one-year risk measure. Similarly, a gross loss cannot be compared directly with a net loss without accounting for recoveries on a consistent basis.

Reconciliation identity: When all amounts have been expressed on one compatible basis, a useful bridge is

R = \sum_{i=1}^{n} C_i - D + A

Here, $C_i$ are the component contributions, $D$ is the amount of identified duplication, $A$ is the net enterprise adjustment, and $R$ is the reconciled result. Each term must represent the same measurement basis. The identity is an accounting bridge, not a formula for aggregating standalone risk measures that have different dependence or calibration assumptions.

Why the Overlap Adjustment Works

Suppose each component amount is a loss under the same defined scenario and valuation basis. If a particular loss item is included in $m$ components, adding those components counts that item $m$ times. The enterprise total should count it once. Therefore, the duplicate amount to remove is $(m-1)$ times the item amount.

\text{duplicate amount for item } j = (m_j-1)O_j

Here, $O_j$ is the item’s amount and $m_j$ is the number of times it appears in the component outputs. Subtracting the duplicates leaves one copy: $m_j O_j-(m_j-1)O_j=O_j$. This reasoning applies to identified shared items measured on the same basis. It does not justify subtracting an estimated overlap between unrelated capital quantiles or between amounts from incompatible scenarios.

Worked Example: Removing a Shared Catastrophe Loss

A hypothetical insurer’s common enterprise stress produces these component losses, all in millions of dollars at the same valuation date and on the same pre-tax, net-of-reinsurance basis: underwriting, $9.4; reserve deterioration, $4.6; catastrophe, $3.1; and operational risk, $1.8. The underwriting output includes $1.2 of the same catastrophe loss represented in the catastrophe component. A separately assessed enterprise expense under the stress is $0.7 and is not included in any component. Reconcile the total.

Method. Use the overlap bridge on the common scenario basis. The shared $1.2 loss appears twice, so one copy must be removed. Add the enterprise expense because it is in scope but absent from the components.

First sum the four component amounts:

$$ 9.4+4.6+3.1+1.8=18.9 $$

Then subtract the duplicate catastrophe amount and add the omitted expense:

$$ R=18.9-1.2+0.7=18.4 $$

The reconciled stressed loss is $18.4 million. As a check, the unique component losses total $9.4+4.6+3.1+1.8-1.2=17.7$ million, and the separate $0.7 million expense brings the result to $18.4 million. The common mistake is to add all components and the expense without removing the overlap, which gives $19.6 million and counts the shared loss twice.

Reconciling Loss Basis and Tax Treatment

Gross and net amounts differ when recoveries, such as reinsurance, are recognized. Pre-tax and after-tax amounts also differ when the model includes a tax effect. Reconciliation requires knowing whether an adjustment applies to gross losses, recoveries, or the resulting net amount. Tax treatment may depend on the model’s assumptions about tax capacity, timing, and whether tax relief is realizable. A simple tax factor is appropriate only when those assumptions are explicitly made.

The following example assumes all stated losses are eligible for an immediate tax benefit, the benefit is fully realizable, and the tax rate is 25%. Under those assumptions, an eligible net loss $L$ has after-tax amount $L(1-0.25)$. This conversion would not be appropriate if the tax benefit were delayed, limited, or unavailable.

Worked Example: Converting Gross Component Losses to an After-Tax Basis

Two hypothetical components report gross losses of $8.0 million and $5.0 million. Recognized reinsurance recoveries are $1.2 million and $0.8 million, respectively. A separate eligible enterprise expense is $2.0 million gross and has no recovery. The target result is net of recoveries and after tax, at a 25% tax rate. Assume all net losses receive an immediate, fully realizable tax benefit. Find the reconciled amount.

Method. Convert each amount to the target basis in order: subtract recoveries where applicable, then apply the stated tax factor. Do not tax the gross amounts and then subtract untaxed recoveries.

For the first component, the net loss is $8.0-1.2=6.8$ million. Its after-tax amount is $6.8(0.75)=5.10$ million. For the second, the net loss is $5.0-0.8=4.2$ million, giving $4.2(0.75)=3.15$ million after tax. The enterprise expense has no recovery, so its after-tax amount is $2.0(0.75)=1.50$ million.

$$ R=5.10+3.15+1.50=9.75 $$

The reconciled after-tax, net amount is $9.75 million. As a check, total gross losses are $8.0+5.0+2.0=15.0$ million, and total recoveries are $1.2+0.8=2.0$ million. The net amount before tax is $15.0-2.0=13.0$ million; applying the 75% after-tax factor gives $13.0(0.75)=9.75$ million. The check agrees. The common mistake is applying tax only to the two components and leaving the enterprise expense on a pre-tax basis.

Aligning Valuation Dates and Calibration

A component may report cash flows at future dates while another reports present values. If the target is a present value at time zero, each cash flow must be discounted to time zero using the stated rate and timing convention. This is a valuation-date conversion, not an arbitrary adjustment to the severity of the risk. The discount rate and cash-flow dates must be appropriate to the model’s stated basis.

Calibration also needs an explicit check. A component result from one severity scenario is not automatically comparable with an enterprise adjustment derived under another. In particular, standalone quantiles cannot generally be added to obtain an enterprise quantile: the joint loss distribution and dependence matter. If a component must be expressed under a different calibration, use a documented rerun or a supported mapping. Do not assume that proportional scaling is valid unless the model justifies it.

Worked Example: Bringing Cash Flows to a Common Valuation Date

A hypothetical model targets present value at time zero. Under its common stress, one component pays $4.20 million at the end of year one and another pays $3.3075 million at the end of year two. An enterprise adjustment of $1.50 million is payable at the end of year one. Use an effective annual discount rate of $i=5\%$, with $v=1/(1+i)$. Assume the cash flows are deterministic amounts under the stated stress, and round the final amount to two decimal places in millions.

Method. Discount each payment separately to time zero. The discount factors are $v=1/1.05$ for year one and $v^2=1/(1.05)^2=1/1.1025$ for year two.

The year-one component has present value $4.20/1.05=4.00$ million. The year-two component has present value $3.3075/1.1025=3.00$ million, since $3.00(1.1025)=3.3075$. The enterprise adjustment has present value $1.50/1.05=1.428571\ldots$ million.

R=4.00+3.00+1.428571\ldots=8.428571\ldots

Rounded to two decimal places, the present value is $8.43 million. The check is that each present value, accumulated at 5% for its own term, returns to its stated cash flow. Simply adding the nominal payments gives $9.0075 million, which is not a time-zero value. That incorrect sum also fails to distinguish valuation-date alignment from any change in the risk measurement horizon.

Worked Example: Reconciling a Calibration Difference

A hypothetical enterprise stress is defined at calibration $C^*$. A component model’s original $7.0 million output used a different calibration, $C_1$. A documented rerun of that component under $C^*$ gives $8.4 million. A second component already reports $4.6 million under $C^*$. The enterprise adjustment under $C^*$ is $1.1 million, and $0.5 million of that adjustment is already included in the second component. All amounts use the same scope, valuation date, and loss basis. Reconcile the result at $C^*$.

Method. Replace the incompatible $7.0 million result with the rerun at the target calibration. Then remove the identified overlap once. Do not add the old and rerun outputs or use a percentage scaling unsupported by the stated rerun.

$$ R=8.4+4.6+1.1-0.5=13.6 $$

The reconciled amount at $C^*$ is $13.6 million. The component subtotal is $8.4+4.6=13.0$ million; adding the $1.1 million adjustment gives $14.1 million, and removing the $0.5 million already counted gives $13.6 million. This bridge is valid because the rerun provides a component amount on the target calibration and the overlap is explicitly identified. It does not establish that the original $7.0 million could be converted by a general scaling rule.

Common Reconciliation Traps

A well-presented total can still be misleading if the bridge hides incompatible inputs. The following checks help keep the reconciliation interpretable:

Double counting a shared exposure. Identify the shared item, its amount on the common basis, and how many times it appears. Subtract only the duplicate copies, not the entire item.
Mixing gross and net amounts. State which recoveries are recognized and apply them consistently. A gross component and net enterprise adjustment cannot be compared without a bridge.
Mixing tax bases. Do not combine pre-tax and after-tax amounts. State the tax assumptions and apply them only where supported.
Confusing present value with horizon conversion. Discount cash flows to a common date when required, but do not treat discounting as a change in the risk horizon.
Adding mismatched calibrations. A standalone quantile or stress at one calibration is not interchangeable with an amount at another. Use a rerun or a documented mapping, and retain its limitations.
Hiding a residual. If the component bridge does not equal the enterprise result, report the residual explicitly and investigate whether it reflects scope, rounding, an omitted adjustment, or an implementation difference.

A useful reconciliation schedule records, for each input, its source, scope, loss basis, valuation date, horizon, and calibration; the amount before and after each adjustment; and the rationale and owner for that adjustment. This makes the bridge reviewable and helps distinguish a genuine enterprise overlay from an unexplained plug. Reconciliation supports model governance, but it does not replace a separate assessment of model limitations or the appropriateness of the capital measure.

Key takeaway: Reconcile meaning before arithmetic. Align scope, loss basis, valuation, timing, and calibration; remove only identified duplication; add only supported enterprise adjustments; and make any residual visible. This structured approach is aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus.

Check Your Understanding

For each problem, identify the relevant basis adjustment before combining the amounts.

A hypothetical common stress gives component losses of $6.2 million and $3.8 million. The first includes $0.7 million also reported by the second. An omitted enterprise expense is $0.4 million. What is the reconciled amount on this shared basis? (Answer: $9.7 million.)
Two gross component losses are $7.0 million and $4.0 million, with recoveries of $1.0 million and $0.5 million. Assume the net amount is eligible for a fully realizable immediate tax benefit at 20%. What is the after-tax net total? (Answer: $7.6 million.)
At an effective annual rate of 4%, a hypothetical stress has payments of $5.20 million at the end of year one and $5.408 million at the end of year two. What is their total time-zero present value, rounded to two decimal places in millions? (Answer: $10.00 million.)
A component reports $5.5 million under calibration $C_1$, but a documented rerun gives $6.3 million under target calibration $C^*$. Another component reports $2.8 million under $C^*$. An enterprise adjustment at $C^*$ is $0.9 million, of which $0.2 million is already in the second component. What is the reconciled amount at $C^*$? (Answer: $9.8 million.)
Three component amounts on the same scenario and basis each include a shared $0.6 million item. The component amounts are $4.0 million, $3.0 million, and $2.0 million, and there is a separate omitted adjustment of $0.5 million. What is the reconciled result? (Answer: $8.3 million.)

What You'll Learn