Separate position attribution from factor attribution, reprice controlled counterfactuals, and disclose how interactions are handled. Reconcile allocated contributions to the joint scenario total and avoid presenting bookkeeping allocations as proof of economic causation.
Choose the question before the decomposition
Position attribution asks which holdings contribute to the scenario result. Factor attribution asks which modeled changes explain those holdings' revaluation. These are different views of the same portfolio, not additional losses to add together. Begin with the position ledger because it provides a concrete reconciliation to ownership, quantities, and starting values. Then decide which factor view will help explain the important positions.
CFA Institute's Portfolio Performance Evaluation reading emphasizes reconciliation and the limitations of attribution methods. Applied to a hypothetical stress report, that means defining what is being allocated before choosing a chart. An allocation of stressed valuation change is not realized performance attribution, manager skill measurement, or evidence that one factor caused a historical event. Label the analytical object accurately throughout the report.
Reprice controlled counterfactuals
For a factor decomposition, retain a common starting portfolio and valuation setup. Calculate the joint scenario, then calculate selected counterfactuals with one factor or group of factors changed. Keep holdings, market-data versions, and pricing conventions fixed. Otherwise, the difference between runs can include unintended changes, and an apparently precise factor contribution becomes a mixture of several unrelated revisions to the experiment.
Group factors at a level the model can support. Separating every curve node and volatility point may create a detailed but unstable explanation. Combining too broadly can hide a concentrated exposure. Document the grouping rule before inspecting which allocation looks most favorable. Retain the exact shocked variables behind labels such as rates or credit so another reviewer can reproduce the decomposition.
Hypothetical calculation: find the interaction
Suppose an invented valuation model produces a loss of USD 80,000 when only factor A changes and USD 50,000 when only factor B changes. When both change together, the loss is USD 150,000. Using signed changes, the interaction is minus 150,000 minus negative 80,000 minus negative 50,000, equal to minus USD 20,000. The standalone contributions do not explain the joint result by themselves.
If A is applied first, its sequential contribution is minus USD 80,000 and B's is minus USD 70,000. Reversing the order assigns minus USD 50,000 to B and minus USD 100,000 to A. Averaging the two orders allocates minus USD 90,000 to A and minus USD 60,000 to B. All reconcile, but they answer the allocation question under different stated conventions.
| Hypothetical allocation convention | Factor A | Factor B | Separate interaction | Total |
|---|---|---|---|---|
| Standalone plus interaction | -USD 80,000 | -USD 50,000 | -USD 20,000 | -USD 150,000 |
| A then B | -USD 80,000 | -USD 70,000 | Allocated | -USD 150,000 |
| B then A | -USD 100,000 | -USD 50,000 | Allocated | -USD 150,000 |
| Average of both orders | -USD 90,000 | -USD 60,000 | Allocated | -USD 150,000 |
Recognize interaction without calling it an error
Nonlinear prices and multiplicative currency translation naturally produce interactions. CFA Institute's currency-management reading provides a simple example of the latter: local asset and exchange-rate changes combine multiplicatively. A residual created by adding standalone effects is therefore not automatically a broken model. First determine whether the residual represents a genuine joint response, a deliberate allocation convention, rounding, or an implementation inconsistency.
Separate known interactions from unexplained residuals. An unexplained difference should not be renamed diversification merely because it makes the total smaller. Reconcile at full calculation precision and show display rounding separately. If the model applies a manual overlay, place it on its own row with a reason and owner. This preserves a meaningful distinction between modeled economics and adjustments outside the pricing method.
Handle offsets and percentages carefully
A position with a positive scenario contribution can offset another holding's loss without being universally protective. Identify the particular shock and valuation assumptions that produce the gain. Review whether the offset survives another coherent scenario or a change in timing. A hedge label is not evidence that the same contribution will appear under all market configurations or be available as cash when needed.
Percentages of net loss become unstable when gains nearly offset losses. A large negative contribution and a large positive contribution can yield small net loss and apparently enormous percentage shares. Present currency contributions and gross losses alongside the net result. If percentage attribution is included, state its denominator and avoid interpreting a contribution above one hundred percent as a calculation error by itself.
Separate scenario drivers from report changes
Explaining a single scenario is different from explaining why this week's report differs from last week's. The latter may involve new holdings, revised prices, changed factor mappings, and a new model version. Build a separate change bridge with controlled reruns. Do not use a rates contribution within today's scenario to explain the entire week-to-week change when other elements of the calculation also moved.
Counterfactual differences can also depend on the order in which report revisions are introduced. State the chosen sequence or use a documented averaging method where appropriate. The objective is not to hide this dependence but to make the explanation reproducible. Keep the original reports intact so a later analyst can distinguish a genuine exposure change from a correction to previously incomplete data.
Use attribution to prioritize evidence
The final commentary should identify the largest reconciled contributions, the material offsets, and any interaction whose treatment affects the ranking. Discuss uncertainty in the factor mapping and valuation rather than treating the chart as an unquestionable answer. When allocation methods change the ranking but not the total, say so explicitly. That is useful information about the limits of the explanation, not a reason to conceal it.
Turn the decomposition into specific review questions: which holding needs a better factor mapping, which offset requires contract confirmation, and which residual needs investigation? Attribution can direct analytical effort without prescribing a trade. These hypothetical allocations illustrate report interpretation; they do not establish historical causality, forecast returns, or certify that the portfolio has passed a regulatory stress-testing process.
Review checkpoints
- Position and factor attribution are alternative views of one total.
- Joint effects can make standalone factor contributions non-additive.
- Disclose interaction allocation, ordering, and percentage denominators.
Source notes
Continue the analysis
Read the stress report before accepting its headline
Audit scope, coverage, sign conventions, denominators, loss bridges, and limitations before interpreting a portfolio stress total.
Reconcile local performance with reporting-currency loss
Translate foreign holdings consistently, preserve the asset-currency interaction, and distinguish economic hedges from settlement cash.
Keep stress-report revisions explainable and reproducible
Assign ownership, preserve report versions, bridge changed results, and prevent corrections or assumption changes from rewriting the record.
First edition, 16 October 2026. Hypothetical values are teaching inputs, not observed portfolio or provider results.