Reconciling Top-Down and Bottom-Up Allocations

Why Reconciliation Requires More Than Matching Totals

A top-down capital allocation starts with a portfolio-level capital measure and distributes that amount to components. A bottom-up estimate starts with component calculations and builds toward a portfolio total. The two approaches can disagree even when both calculations are internally consistent: they may cover different risks, use different measurement bases, or treat diversification and dependencies differently.

The reconciliation task is therefore not simply to force two numbers to match. It is to identify what each number measures, explain the difference, and document how any adjustment is assigned. This matters because an allocation can sum exactly to the aggregate capital and still be misleading if its components do not have comparable scope or meaning.

Definition: A top-down capital amount is an aggregate capital target allocated among components. A bottom-up estimate is a set of component estimates developed separately and combined to produce or approximate a total. A reconciliation is a documented bridge between these amounts and, where relevant, between the component values themselves.

Let \(T\) be the aggregate target and \(b_i\) the bottom-up estimate for component \(i\). A useful total-level bridge is

T=\sum_{i=1}^{n} b_i+\Delta_{\text{scope}}+\Delta_{\text{basis}}+\Delta_{\text{aggregation}}+\Delta_{\text{data}}+R.

Each adjustment is signed: a positive value increases the bottom-up total toward the target, and a negative value decreases it. The residual \(R\) is the amount not explained by the identified adjustments. A sound reconciliation makes the residual visible rather than silently assigning it to components.

Reconcile the Measurement Before the Allocation

Before comparing component numbers, establish a common measurement definition. In practice, this means checking the valuation date, risk horizon, confidence level or other risk measure, included entities and exposures, treatment of taxes or management actions, and model assumptions. These checks do not require the two methods to be identical; they make differences interpretable.

Difference	Question to investigate	Typical bridge treatment
Scope	Are the same entities, exposures, and risk types included?	Add or remove the omitted or extra exposures, with their effects measured on a stated basis.
Measurement basis	Do the calculations use the same valuation date, horizon, risk measure, and assumptions?	Quantify the effect of changing from one basis to the other.
Aggregation	Are dependencies, diversification, and interactions treated consistently?	Show the aggregation effect separately; do not call it diversification unless its sign and definition support that interpretation.
Data and implementation	Do inputs, approximations, or model versions differ?	Recalculate or quantify the effect, and leave any unresolved amount as a residual.

A further distinction is essential: reconciling the total does not automatically reconcile the components. If \(a_i\) are top-down allocations and \(q_i\) are adjusted bottom-up values on the same scope and basis, compare \(a_i-q_i\) by component. When both sets fully allocate the same target, their differences must sum to zero. If they do not, either the totals still differ or the component mapping is incomplete.

\sum_{i=1}^{n}(a_i-q_i)=\sum_{i=1}^{n}a_i-\sum_{i=1}^{n}q_i.

This identity is simple but useful: a component-level comparison is meaningful only after the totals and component definitions have been aligned. A mismatch between, for example, a risk-type allocation and a business-unit estimate may reflect a mapping problem rather than a capital-model disagreement.

Top-Down Euler Allocations and Full Allocation

For a differentiable, degree-one homogeneous aggregate capital function \(C(x_1,\ldots,x_n)\), an Euler allocation assigns component \(i\) the amount \(a_i=x_i\,\partial C/\partial x_i\). Here \(x_i\) is the component’s exposure or scale variable. The formula is relevant only when the stated function and differentiability conditions hold; it does not make arbitrary bottom-up estimates comparable by itself.

C(t x_1,\ldots,t x_n)=tC(x_1,\ldots,x_n),\qquad a_i=x_i\frac{\partial C}{\partial x_i},\qquad \sum_{i=1}^{n}a_i=C.

The last equality follows from Euler’s theorem for homogeneous functions. Differentiate \(C(t x_1,\ldots,t x_n)=tC(x_1,\ldots,x_n)\) with respect to \(t\), then set \(t=1\). The chain rule gives \(\sum_i x_i\,\partial C/\partial x_i=C(x)\), which is precisely full allocation. This establishes why the Euler components sum to the aggregate under these conditions; it does not establish that a separate bottom-up method must produce the same component values.

Worked Example: Reconciling Standalone Capital With a Covariance Aggregate

Method: Calculate the covariance aggregate, derive its Euler allocations, then compare those values with standalone capital. A hypothetical portfolio has two risk components with standard deviations \(100\) and \(80\), in thousands of dollars, and correlation \(0.25\). Assume the aggregate measure is the standard deviation of the sum, with this covariance model and no additional capital adjustment. Find the aggregate, the Euler allocations, and the diversification benefit.

Aggregate capital. Under the stated covariance model,

C=\sqrt{100^2+80^2+2(0.25)(100)(80)} =\sqrt{10{,}000+6{,}400+4{,}000} =\sqrt{20{,}400} =142.8286.

Thus the aggregate is \(142.8286\) thousand dollars. The standalone values sum to \(100+80=180\) thousand dollars.

Euler allocation. For this covariance aggregate, the first component contribution is its variance plus its covariance with the second component, divided by \(C\). The second contribution is calculated in the same way:

a_1=\frac{100^2+(0.25)(100)(80)}{142.8286} =\frac{12{,}000}{142.8286} =84.0168, $$ $$ a_2=\frac{80^2+(0.25)(100)(80)}{142.8286} =\frac{8{,}400}{142.8286} =58.8118.

The check is \(84.0168+58.8118=142.8286\) thousand dollars, matching the aggregate to the displayed precision. The diversification benefit relative to the standalone sum is \(180-142.8286=37.1714\) thousand dollars. The Euler attribution assigns \(100-84.0168=15.9832\) thousand dollars of reduction to component 1 and \(80-58.8118=21.1882\) thousand dollars to component 2. These reductions sum to \(37.1714\).

Common mistake and correction. Adding the standalone values and reporting \(180\) thousand dollars as the covariance aggregate ignores the positive covariance term and the square root. The correction is to aggregate using the stated covariance model first, then compare the standalone sum with \(C\). All monetary values are in thousands; rounding is to four decimal places for the reconciliation check.

Build a Bridge With Identified Drivers

A bridge should distinguish a known adjustment from an unexplained residual. Where a difference has a measurable cause, quantify it on a consistent basis and identify its destination. If several components are affected, use an explicit allocation rule or a model recalculation. Do not infer the cause from the desired final total.

Fix the target definition.
Record what the aggregate includes and the risk measure, valuation date, horizon, and material assumptions used.

Map bottom-up components.
Check that each estimate represents the same business or risk category as the corresponding top-down allocation.

Quantify the bridge items.
Measure scope, basis, aggregation, and data effects separately where possible; record the sign and method for each.

Assign adjustments transparently.
Use a defensible component mapping or allocation rule. If the evidence does not support an assignment, retain a labelled residual.

Check totals and components.
Confirm that the bridge reaches the target and that any component-level comparison uses common scope and basis.

Worked Example: Separating Scope, Basis, and Aggregation Effects

Method: Create a signed bridge from the bottom-up total to the aggregate target, then assign each identified adjustment to the component it affects. A hypothetical insurer reports bottom-up estimates of \(240\), \(150\), and \(90\) thousand dollars for underwriting, investment, and operational risk. The top-down target is \(540\) thousand dollars. Investigation finds that the bottom-up calculation omitted a cyber exposure contributing \(20\) thousand dollars to underwriting risk; an updated investment valuation basis increases investment capital by \(25\) thousand dollars; and a concentration treatment in the aggregate model adds \(15\) thousand dollars to operational risk. Assume these are the only differences.

Bottom-up total. The initial total is \(240+150+90=480\) thousand dollars. Its difference from the target is \(540-480=60\) thousand dollars.

Bridge. The identified adjustments total \(20+25+15=60\) thousand dollars. Therefore \(480+20+25+15=540\) thousand dollars, and the residual is zero.

Adjusted components. Assigning the scope effect to underwriting gives \(240+20=260\). Assigning the basis effect to investment gives \(150+25=175\). Assigning the concentration effect to operational risk gives \(90+15=105\). The adjusted component sum is \(260+175+105=540\) thousand dollars.

Common mistake and correction. One might distribute the \(60\) thousand dollar gap pro rata across the original estimates. The shares \(240/480\), \(150/480\), and \(90/480\) would produce additions of \(30\), \(18.75\), and \(11.25\), respectively. This balances to \(540\), but it contradicts the known investigation: the documented drivers are \(20\), \(25\), and \(15\) by component. Use the measured drivers rather than a convenient balancing rule. Amounts are in thousands of dollars; all sums are exact.

Reconcile Hierarchies Without Hiding Local Differences

Capital is often reported at several levels, such as enterprise, business, and product. The allocation at each level should add to its parent amount. A bottom-up estimate for a child unit may not equal its parent’s top-down allocation, even when the child estimates sum correctly within the local model. In that case, the difference should be traced at the parent level and assigned to children only under a stated rule.

Worked Example: Reconciling Child Units to Parent Allocations

Method: First reconcile at the parent level, then allocate the parent adjustment using the stated exposure weights. A hypothetical group has total capital of \(500\) thousand dollars. Its top-down allocation is \(280\) thousand dollars to Group A and \(220\) thousand dollars to Group B. Bottom-up child estimates are \(140\) and \(120\) for A’s two units, and \(125\) and \(95\) for B’s two units. A review confirms the B estimates are on the parent basis. For Group A, the approved local exposure weights for distributing a parent-level difference are \(60\%\) and \(40\%\).

Compare child totals with parent allocations. Group A’s children sum to \(140+120=260\), which is \(280-260=20\) thousand dollars below its parent allocation. Group B’s children sum to \(125+95=220\), exactly matching its parent allocation.

Assign A’s difference. The first child receives \(0.60(20)=12\) thousand dollars and the second receives \(0.40(20)=8\) thousand dollars. The adjusted Group A values are \(140+12=152\) and \(120+8=128\).

Check the hierarchy. Group A’s children sum to \(152+128=280\); Group B’s children sum to \(125+95=220\). The four adjusted child allocations total \(280+220=500\), matching the enterprise total.

Common mistake and correction. Applying the \(20\) thousand dollar adjustment to all four child units would cross the parent boundary and could alter Group B despite its exact reconciliation. Keep the adjustment within Group A and apply the approved \(60\%/40\%\) rule. Amounts are in thousands of dollars; no rounding is needed.

Interpretation and Practical Checks

A zero residual is evidence that the bridge adds correctly, not proof that every adjustment is well founded. Likewise, a nonzero residual is not necessarily an error: it may represent unresolved data, model limitations, or a difference that cannot yet be quantified. It should be investigated, reported, and monitored rather than hidden in a component.

A defensible reconciliation should let a reviewer answer three questions: what is the common target, what explains the difference, and how were any adjustments assigned? It is aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus to recognize that allocation is not only an arithmetic exercise: the interpretation depends on scope, model treatment, and the chosen allocation basis.

Check that an aggregate and its component allocations use the same currency units and reporting date.
Separate a diversification effect from a scope change or a model-basis change.
Confirm that parent allocations equal the sum of their child allocations after documented adjustments.
Label any balancing amount as a residual until its cause and assignment are supported.

Key takeaway: Reconcile the measurement first, the total second, and the components third. An explained bridge preserves the meaning of each estimate; an unexplained balancing adjustment can make unlike quantities appear comparable.

Check Your Understanding

For each problem, state the reconciliation and check the final total.

A covariance model has standard deviations \(60\) and \(50\), with correlation \(0.20\). Calculate the aggregate standard deviation and both Euler contributions. Report amounts in thousands to the nearest cent. (Answer: aggregate \(85.44\); Euler contributions \(49.16\) and \(36.28\) thousand.)
Three bottom-up estimates total \(390\) thousand dollars. The target is \(410\) thousand dollars. A scope addition is \(12\) thousand and a basis adjustment is \(8\) thousand. What residual remains? (Answer: \(0\) thousand dollars.)
A parent allocation is \(360\) thousand dollars. Its three child estimates are \(150\), \(130\), and \(70\) thousand dollars. Allocate the difference using weights of \(30\%\), \(70\%\), and \(0\%\), respectively. What are the adjusted child amounts? (Answer: \(153\), \(137\), and \(70\) thousand dollars; total \(360\).)
A bottom-up total is \(630\) thousand dollars and the aggregate target is \(600\) thousand dollars. Investigation identifies an overlap adjustment of \(-18\) thousand and a basis adjustment of \(-12\) thousand. What residual remains, and what check should be reported? (Answer: residual \(0\); \(630-18-12=600\) thousand dollars.)
Bottom-up estimates for three risks are \(250\), \(210\), and \(190\) thousand dollars. The top-down target is \(720\) thousand dollars. The review identifies a scope addition of \(25\), a model-basis increase of \(15\), and a concentration adjustment of \(30\) thousand dollars. Does the bridge close? (Answer: yes; \(650+25+15+30=720\) thousand dollars, with zero residual.)

What You'll Learn