Plain-vanilla contracts can create complex counterparty exposure. Future value depends on market paths; collateral depends on contractual timing and thresholds; loss depends on default and recovery; and valuation adjustments combine these elements across many future dates. A simple product description does not imply a simple risk calculation.
REFERENCE · NO CALL
This report consolidates counterparty-risk teaching, a July 2021 XVA model-risk study, a May 2017 capital-cost presentation and a small credit-loss workbook. It explains their methods in original prose. Historical study parameters and teaching numbers are retained with their source status; names, institutional affiliations and private contact details are excluded.
Future exposure is different from notional
Notional determines contractual payments but is not generally the amount lost if a derivative counterparty defaults. Let denote the portfolio’s value to the institution at future time , and the collateral available under the relevant close-out assumptions. A simplified positive exposure is
Expected exposure averages this quantity across scenarios. Potential future exposure is a specified percentile of the future-exposure distribution. They answer different questions and vary over the trade’s life.
Netting changes the object being measured. The exposure of an enforceable netting set can be materially smaller than the sum of positive values of individual trades. That benefit depends on the legal agreement, product coverage and close-out mechanics; it is not obtained merely by adding a netting-set label in the system.
Interest-rate swaps, foreign-exchange contracts, credit protection and options produce different exposure profiles. A purchased option can have positive replacement value without a symmetric negative exposure before exercise. A forward-starting trade can concentrate risk later. Credit protection can create jumps and strong dependence between the counterparty’s health and the reference exposure.
Default and exposure can move together
A simple unilateral credit valuation adjustment can be written schematically as
where is a discount factor, expected positive exposure, loss given default and marginal default probability over interval . This form is useful for understanding the components but assumes enough separation between exposure and default to use the product of the terms as shown.
Wrong-way risk arises when exposure tends to rise as the counterparty becomes more likely to default. In that setting, an unconditional exposure average multiplied by an independent default curve can understate risk. The joint mechanism needs to be represented or otherwise controlled.
Different valuation adjustments answer different questions. CVA concerns counterparty credit loss; a debit adjustment concerns the institution’s own default effect under the valuation convention; funding and margin adjustments address funding-related costs. A capital valuation adjustment is an economic pricing construct rather than a universal accounting treatment. Their assumptions and boundaries should be explicit to avoid double counting.
Figure 01 · Conceptual framework
From future scenarios to counterparty valuation
Collateral reduces exposure without making timing irrelevant
Variation margin seeks to cover changes in current value. Initial margin provides protection against potential movement during the period needed to close out or replace positions after default. Thresholds, minimum transfer amounts, settlement delay, eligible collateral and disputes all affect the residual exposure.
The July 2021 study deliberately uses a near-frictionless collateral setup for part of its analysis: zero thresholds, zero minimum transfer amounts and a two-day margin period of risk. It still finds residual exposure because collateral tracks portfolio value with a delay and contractual cash flows can create sharp changes.
The two-day assumption is a study parameter, not a general statement about an appropriate regulatory margin period. Its purpose is to isolate the effects being studied. Applying it to another counterparty without contractual and operational evidence would create false precision.
A simulation grid can miss the most important exposure
The study’s central numerical lesson is that collateralised exposures can spike around payment dates and during the margin period of risk. A coarse regular time grid can miss those spikes, even if the overall exposure curve looks smooth.
Increasing the number of simulated paths does not fix a time grid that never evaluates the relevant event. The model can converge very accurately to the wrong discretised answer. Grid design and path-count convergence should therefore be tested separately.
The study’s adopted calculation settings include:
Historical study settings and their interpretation
| Parameter | Historical study setting | Interpretation |
|---|---|---|
| Regular time step | One month | Base grid, supplemented by event-aware treatment |
| Simulation paths | 5,000 | Chosen study accuracy/performance compromise |
| Margin period of risk | Two days | Collateral timing assumption |
| Threshold | Zero | Deliberately low-friction case |
| Minimum transfer amount | Zero | Deliberately low-friction case |
Those settings are not a recommended universal production configuration. A different portfolio, payoff or collateral agreement may require another grid and more paths. The useful method is to compare a parsimonious event-aware grid with a denser benchmark and quantify the error in exposure and XVA, rather than selecting resolution solely from runtime.
Figure 02 · Qualitative comparison
Three quantities that should not be confused
| Approach or concept | What it contributes | What still matters |
|---|---|---|
| Notional | Measures contractual scale | Does not measure future credit exposure |
| Future exposure | Measures possible replacement value through time | Depends on markets, netting and collateral |
| XVA | Values specified costs and risks | Model assumptions and funding conventions matter |
Separate economic parameters from numerical parameters
Economic, numerical and model changes answer different questions
| Aspect | Retained observation |
|---|---|
| Contractual threshold | Changing a contractual threshold changes the economic exposure. |
| Monte Carlo paths | Changing the number of Monte Carlo paths changes estimator precision. |
| Time grid | Changing the time grid can change approximation bias. |
| Interest-rate dynamics | Changing the interest-rate dynamics changes the model itself. |
These are different sources of variation.
A validation study should alter them in controlled groups. Otherwise the effect of a new collateral assumption can be mistaken for numerical convergence, or a faster calculation can appear accurate because an offsetting model error happens to cancel its discretisation error.
The historical research uses a two-factor Gaussian interest-rate model in a multi-curve setting and compares simulation with analytical results where a suitable benchmark exists. Agreement is meaningful only when both calculations use the same dynamics and contractual assumptions. A discrepancy between different models is not necessarily a coding defect.
Dynamic initial margin adds another layer of approximation
Future initial margin can depend on future sensitivities. Calculating those sensitivities at many future dates and along thousands of paths is computationally demanding. A bump-and-revalue calculation inside a simulation can multiply the workload substantially.
The study examines approximations to future vega and checks them against alternative constructions at the valuation date. That is a useful control principle: compare approximation routes where a reliable common benchmark is available, then investigate how the approximation behaves away from that point.
An acceptable approximation should have a stated purpose, tested range and error assessment. A fast method should not be accepted merely because an exact calculation is inconvenient. Conversely, an expensive method is not automatically preferable if an independently validated approximation achieves the accuracy the decision requires.
Figure 03 · Conceptual framework
Can the time grid miss the important exposure?
How the parts connect
- Exposure simulation designPath AGrid captures critical events
- Exposure simulation designPath BEvents fall between grid points
- Grid captures critical eventsReviewQualified exposure result
- Events fall between grid pointsReviewQualified exposure result
A small credit workbook shows the danger of mixed assumptions
The teaching workbook lists three exposures, default probabilities and loss-given-default inputs:
Credit-workbook exposures, default probabilities and listed losses given default — Exposure: A
| Measure | Value |
|---|---|
| Amount | 25 |
| Default probability | 5% |
| Listed LGD | 80% |
Credit-workbook exposures, default probabilities and listed losses given default — Exposure: B
| Measure | Value |
|---|---|
| Amount | 30 |
| Default probability | 10% |
| Listed LGD | 50% |
Credit-workbook exposures, default probabilities and listed losses given default — Exposure: C
| Measure | Value |
|---|---|
| Amount | 45 |
| Default probability | 20% |
| Listed LGD | 20% |
Its enumerated loss distribution nevertheless uses the full exposure amount on default. Under independent defaults and 100% LGD, expected loss is
matching the workbook’s reported expected loss. Its 95% quantile is 45, and the difference between that quantile and expected loss is 31.75.
If the listed LGDs are applied instead, single-default losses become 20, 15 and 9. Expected loss becomes
Enumerating the same independent default states gives a 95% loss quantile of 20. The two results belong to different assumptions. The original 13.25 should not be presented as incorporating the listed LGDs, and the recalculated 4.30 should not silently replace the historical workbook output.
The exercise also uses the term unexpected loss for quantile minus expected loss. Elsewhere, unexpected loss can refer to a loss standard deviation. A report must define which meaning it uses rather than assuming the label is self-explanatory.
Capital costs change product economics
The May 2017 presentation argued that higher capital and collateral costs would raise the economic cost of OTC derivatives and increase the value of optimisation. That was a historical directional assessment. It is retained as such, not declared a verified current forecast.
The underlying decision framework remains useful. Compare contracts on their full economics: credit exposure, collateral liquidity, funding, capital, flexibility and execution price. A lower upfront spread can be offset by less favourable collateral or optionality. Restructuring can reduce one cost while increasing another, so the comparison needs consistent valuation and cost assumptions.
Model uncertainty needs an explicit response
The 2021 research assesses valuation uncertainty across plausible modelling choices and relates it to additional valuation adjustment methodology. The durable principle is to quantify sensitivity to defensible alternatives rather than hiding it behind one production configuration.
The resulting control framework should contain exposure-profile benchmarks, event-aware grid tests, simulation convergence, collateral-contract checks, sensitivity validation, wrong-way-risk assessment and an explicit uncertainty treatment. XVA is reliable when these pieces agree about the same future portfolio and collateral process—not merely when the final spreadsheet produces a number.
