Position size is not the same as factor exposure
A portfolio can be evenly split across instruments and still depend heavily on one risk driver. Market values describe how capital is allocated; factor exposures describe how valuation might respond to specified changes. Neither a position label nor a gross notional total is sufficient to explain cross-asset risk.
Start with verified holdings, choose risk factors suited to the instruments, and express sensitivities in explicit units. Reconcile gross and net measures, distinguish estimated exposures from observed positions, and test residual risks that a convenient factor model leaves outside its boundary.
Maintain a holdings view and a sensitivity view
The holdings view records quantities, market values, currencies, counterparties, maturities, and financing. The sensitivity view asks how those holdings respond to defined drivers such as interest rates, credit spreads, equity markets, or commodity prices. Keep both. A modeled exposure should never replace the legal and operational identity of an instrument, because settlement, collateral, and issuer risks can differ despite similar market sensitivity.
A position's market value may be small while its sensitivity is substantial, particularly for derivatives. Conversely, a large notional can include offsetting exposures or a structure with limited sensitivity to the selected shock. The ECB's leveraged-fund analysis discusses why gross notional alone can misrepresent risk. Use it as one descriptive measure, not a universal economic scale across unrelated products.
Select factors that match the instrument's economics
A government bond can require yield-curve sensitivities; a corporate bond adds spread and issuer effects. An equity portfolio may respond to market, industry, currency, and style influences. Commodity futures depend on maturity-specific price movements as well as the level of a commodity benchmark. Do not force every instrument into one broad growth factor merely because that narrative is easy to communicate.
Define the unit of each exposure. Dollar change per basis point, percentage return per one-percent market move, and sensitivity to an index-level change are not directly additive. Set sign conventions before aggregation. CME's basis-point-value material provides a concrete interest-rate example: sensitivity concerns the value effect of a specified yield change, not simply the face amount of the position.
Worked hypothetical example: equal capital, unequal rate exposure
Assume an invented USD 100 million portfolio holds two bond sleeves worth USD 50 million each. Sleeve A has modified duration two; sleeve B has modified duration eight. Using the first-order approximation, their dollar sensitivities per basis point are USD 10,000 and USD 40,000 respectively. Capital weights are equal, but sleeve B supplies 80% of the combined parallel-yield sensitivity.
For a hypothetical 25-basis-point parallel rise, estimated losses are USD 250,000 and USD 1 million, totaling USD 1.25 million. The calculation holds spreads and positions fixed and ignores convexity and income. It is not a precise repricing for a large or nonparallel move. The 80% figure refers only to this defined rate sensitivity, not to 80% of every possible portfolio risk.
| Hypothetical sleeve | Capital share | DV01 magnitude | Estimated loss for +25 basis points |
|---|---|---|---|
| A: duration two | 50% | USD 10,000 | USD 250,000 |
| B: duration eight | 50% | USD 40,000 | USD 1,000,000 |
| Combined | 100% | USD 50,000 | USD 1,250,000 |
Distinguish structural sensitivities from estimated betas
Some exposures come from instrument pricing models and current contract terms. Others are estimated statistically from historical returns. A regression beta depends on the factor definition, sample, frequency, and other included variables. It is not a directly observed holding. Report estimation uncertainty and investigate whether a beta remains meaningful after the portfolio or the economic environment changes.
Correlated explanatory factors can make individual coefficients unstable even when overall fit appears strong. Omitting a relevant factor can assign its effect to another variable. Use an economically motivated specification and compare it with holdings information where available. A high model fit does not prove that the factors caused the returns, and a residual does not automatically represent manager skill or instrument-specific alpha.
Preserve basis and nonlinear risks after netting
Two positions can offset under a parallel-rate measure while retaining curve, spread, currency, or maturity-basis exposure. Aggregate at a useful level of detail before netting everything into one number. A near-zero net sensitivity can coexist with large gross positions, collateral demands, and exposure to a breakdown in the assumed relationship. Identify the conditions under which the apparent hedge works.
For options, delta changes as the underlying price, volatility, and time change. A small initial delta does not make the full payoff harmless under a large move. Consider gamma, volatility sensitivity, or full scenario repricing as the question requires. Likewise, a duration estimate does not cover every bond risk. State which dimensions are omitted rather than attach a universal low-risk label to a small selected sensitivity.
Separate exposure, contribution, and funding requirements
Exposure describes sensitivity; realized contribution describes what an instrument added to return over a period. Risk contribution additionally depends on the chosen risk measure and relationships among factors. A position with large standalone volatility may reduce modeled portfolio volatility if its covariance offsets other holdings. That result remains conditional on the covariance model and should not be confused with guaranteed stress protection.
Keep a separate liquidity view for variation payments, financing maturities, and collateral eligibility. A market hedge can reduce valuation sensitivity while increasing immediate cash needs. ECB research makes this distinction relevant for leveraged portfolios. Do not infer that netting economic exposure permits netting cash obligations across different counterparties or legal entities without verifying the actual arrangements.
Explain concentration in the factor's own units
Report the factors driving the largest sensitivities, the holdings behind them, and the assumptions used in aggregation. Provide selected scenarios that reveal residual risks and the limits of local approximations. When holdings are unavailable, label proxy-based exposures and show the coverage gap. A model should make missing transparency more visible, not convert it into a precise-looking complete portfolio map.
Common failures include comparing derivative notionals directly with cash market values, interpreting a historical beta as a permanent characteristic, and calling a duration-neutral portfolio rate-neutral under every curve move. A useful cross-asset report shows both where capital sits and which specified shocks matter. That distinction supports economic understanding without prescribing a current hedge or suggesting that factor modeling eliminates uncertainty.
What to retain
- Capital weights and factor sensitivities describe different dimensions of a portfolio.
- State exposure units and distinguish pricing sensitivities from estimated historical betas.
- Net market exposure can conceal basis, nonlinear, collateral, and funding risks.
Source notes
Related research
Duration and the Mechanics of Rate Shocks
Calculate bond-price sensitivity, convexity, and curve exposure while keeping approximations separate from realized returns.
Constructing coherent cross-asset scenarios
Build conditional macro paths, translate shocks into instrument effects, and distinguish valuation losses from funding requirements.
Credit Spreads, Financing Conditions, and Growth
Disentangle default exposure, risk appetite, benchmark rates, and index composition when interpreting credit-market information.
Version 1, 15 October 2026. First publication. Illustrative calculations are identified in context; source observations retain their stated dates.