Reconcile the imported portfolio, classify direct, proxy, and unmodeled positions, and report coverage on several meaningful bases. Keep missing exposures in the denominator where appropriate and label partial losses explicitly. Use clearly hypothetical sensitivity overlays to explore uncertainty, without presenting them as validated bounds or replacing absent data with zero risk.
Define the perimeter before measuring completeness
Identify the accounts, legal entities, valuation timestamp, and types of positions intended to be included. Reconcile that perimeter with a source inventory containing assets, liabilities, unsettled transactions, cash, and derivatives as applicable. An imported file may be internally consistent but incomplete relative to the intended portfolio. Completeness cannot be established by checking only the rows that successfully entered the model.
BCBS 239 identifies completeness alongside other risk-data principles. For an educational portfolio workflow, the practical lesson is to compare the model's inventory with an independent source count and balance. This is not a claim of regulatory compliance. It is an accounting and analytical safeguard against allowing rejected records or unsupported instruments to disappear before the stress calculation begins.
Use several coverage measures, not one percentage
Row coverage measures the proportion of records represented, but it treats a small cash line like a large derivative. Market-value coverage gives more weight to economically large holdings, yet can understate the importance of near-zero-value derivatives. Gross notional and factor-sensitivity coverage provide other perspectives, although notionals are not directly comparable across all instruments. Name each denominator and explain its limitations.
Risk-channel coverage is different again. A bond may be included for benchmark-rate risk but lack credit, option, or currency treatment. Calling that position fully covered would overstate the model's scope. Maintain an instrument-by-risk matrix where useful, distinguishing included, held constant, proxied, and unsupported channels. A valuation model's successful execution is not evidence that every material risk has been modeled.
Classify missing information by its consequence
Separate unresolved identity, absent price, stale price, missing contract terms, incomplete look-through, and unsupported pricing. Each problem requires a different remedy and carries a different uncertainty. An old price may allow a sensitivity calculation with a qualified starting value; an unknown option strike can prevent meaningful repricing altogether. Do not hide these distinctions inside one generic data-quality score.
The ECB's 2024 risk-data guide emphasizes clear definitions, ownership, and data lineage. A portfolio exception record can adopt those ideas by naming the missing field, its source, the affected calculation, and the person or process responsible for resolution. Preserve the original row and any proxy decision. Replacing an unavailable value with zero without a warning turns missing information into an unsupported economic assumption.
Interpret a hypothetical partially modeled portfolio
Suppose a synthetic USD 1,000,000 portfolio has 100 position records, of which 90 can be modeled. Direct models cover USD 600,000 and documented proxies cover USD 200,000; USD 200,000 remains unmodeled. Row coverage is 90%, while modeled market-value coverage is 80%. Direct-model coverage is only 60%. All three statements are correct, but they describe different properties of the same inventory.
Assume the modeled subset loses USD 48,000 under a hypothetical scenario. That is 6% of the modeled USD 800,000, and a known loss contribution equal to 4.8% of the full starting value. It is not a verified 4.8% full-portfolio return because the missing portion has no estimated outcome. Keep the unresolved amount visible rather than rescaling the modeled subset to pretend it represents everything.
| Synthetic coverage measure | Numerator | Denominator | Coverage |
|---|---|---|---|
| Modeled rows | 90 records | 100 records | 90% |
| Direct-model value | USD 600,000 | USD 1,000,000 | 60% |
| Direct plus proxy value | USD 800,000 | USD 1,000,000 | 80% |
| Unmodeled value | USD 200,000 | USD 1,000,000 | 20% |
Explore uncertainty with explicit overlays
For the same hypothetical example, assigning the missing USD 200,000 a 10% loss adds USD 20,000, producing USD 68,000 total loss, or 6.8%. Assigning it a 30% loss adds USD 60,000, producing USD 108,000, or 10.8%. These are alternative judgmental overlays. They are not estimates established by the available data and are not guaranteed lower or upper bounds.
A leverage-sensitive or nonlinear missing position could fall outside either overlay, and an offsetting exposure could gain instead. Explain why a chosen challenger is useful for the question and which features it ignores. Keep the original partial calculation separate from the overlay results so later users cannot mistake a provisional assumption for repaired instrument coverage or validated portfolio valuation.
| Hypothetical missing-sleeve assumption | Additional loss USD | Combined loss USD | Loss as share of full value |
|---|---|---|---|
| 10% loss on USD 200,000 | 20,000 | 68,000 | 6.8% |
| 30% loss on USD 200,000 | 60,000 | 108,000 | 10.8% |
Do not let netting conceal missing gross exposure
Long and short positions can have offsetting marked values while carrying different factors, maturities, or counterparties. A net-value denominator can therefore hide substantial unsupported gross risk. Report relevant gross measures alongside net value, especially for derivatives and leveraged strategies. Avoid summing heterogeneous notionals into a supposedly universal risk total without explaining what that number can and cannot indicate.
Check for missing hedge legs and missing financed assets independently. An incomplete hedge can exaggerate loss, while an omitted asset can understate it. Neither direction should be assumed in advance. Reconcile corporate actions, trade settlements, and duplicate identifiers before interpreting changes in coverage. A sudden improvement may reflect a duplicated file or changed classification rather than genuinely better data.
Make incomplete results visibly incomplete
Place coverage, proxy share, valuation dates, and unresolved material items next to the headline result. Use wording such as modeled-subset loss where the full portfolio is not represented. Define any publication or internal-use thresholds before reviewing the result, and distinguish an analyst's chosen threshold from an official requirement. A high coverage percentage does not automatically make the remaining gap immaterial.
When missing data arrive, rerun the same scenario and explain changes attributable to repaired coverage separately from changes in holdings or market assumptions. Retain the earlier exception history so the improvement is auditable. The objective is an honest account of what the calculation can establish. Partial analysis can still be useful, but it cannot justify a complete-portfolio assurance or a personalized investment decision.
Review checkpoints
- Row, value, notional, and risk-channel coverage answer different questions.
- Unmodeled positions are unknown risk, not automatically zero risk.
- Keep partial results and hypothetical overlays separate until coverage is repaired.
Source notes
Continue the analysis
Mapping Portfolio Holdings to Risk Factors
Create an auditable bridge from securities and contracts to signed factor exposures, consistent units, and reconciled scenario results.
Finding Concentration Risk Behind Portfolio Labels
Look through funds, aggregate connected exposures, and distinguish capital weights from conditional loss concentrations.
Read the stress report before accepting its headline
Audit scope, coverage, sign conventions, denominators, loss bridges, and limitations before interpreting a portfolio stress total.
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.