Primary tabs


When comparing the two options, we think that the more appropriate is option 1b: Y%=60% and Z%=30%. This is because, we think that 30% threshold should be sufficient to capture the materiality of risk drivers.
The quantitative approaches suggested would require both the calculation of sensitivities and a relevant shock to be applied for all derivatives products, even if to only assess a threshold. Operationally, this would require significant development to source risks for every derivative transaction. For linear products the sensitivity is not readily available for every transaction as under the market risk framework, risks are not stored at that level of granularity. Risks are considered at the level of books and portfolios where derivative positions are aggregated with hedges to get the correct representation of risk. Where a product has optionality primary trading systems are more likely to have sensitivities stored, but this would be a subset of the total derivatives portfolio. Furthermore, this would put a significant dependency between SA-CCR and FRTB as resources will be primarily focused on FRTB implementation challenges.
We do not have any views on the appropriateness, for smaller institutions, of the alternative SA-CCR add-ons approach (Article 3(2)).
A normal model (as opposed to lognormal) will be more suitable. This is because, it will naturally solve the negative rates concern and it does not require any further recalibration or change of the formula.
However, if we have to use the shifted lognormal distribution, we prefer option 3 a - currency level, as this provides more transparency on the used model (having the same model for all transactions within the same currency). This is also consistent with our risk management framework.
For the extreme case of very negative strikes (below lambda), where the formula will not work, we propose to put a conservative delta at 1.
We believe that Option 4c: 1% is more appropriate as a threshold. This is because, it is the option which generates a reasonable probability distribution.
We recommend a shift between 1% and 3%. This is because, if the shift is too small, the mean of the log normal distribution will be too close to zero, generating an inappropriate distribution with a big density around zero. Thus, in order to have a reasonable probability distribution and meaningful prices, the shift should be large enough.
provide some benefits:
Having a dynamic lambda is a good way to design a lambda that works in all cases (or almost, apart for very negative strikes in case of option 3a).
It becomes unnecessary to set up a process to monitor the rates, and change the lambda when needed, which will raise more questions around when to trigger the change, and what should be the level of the new lambda.
Thus, the proposed methodology seems more practical and simpler to put in place.

raise some concerns:
There are no concerns with setting the shift λ according to the prevalent market conditions.
We do consider it necessary to make an adjustment to the supervisory volatility parameter σ as defined in Article 5. This is because, when the shift changes, this will impact the ATM price. Hence, the volatility needs to be changed in order to reflect that. The adjustment should be carried out in order to fit the same ATM price.
The ATM price is given by the Black-Scholes (“BS”) formula:
ATM price = BS(σ, Fwd + λ, Fwd + λ, t)
Where σ is the volatility parameter, Fwd is the forward rate, λ is the shift parameter
If λ changes to λ’, then σ needs to be changed to σ’, where:
BS(σ, Fwd + λ, Fwd + λ, t) = BS(σ’, Fwd + λ’, Fwd + λ’, t)
As a first order approximation, this equation can be simplified to: σ * (fwd + λ) = σ’ * (fwd + λ’)
There are instances, such as basis swaps where it is not immediately clear whether there is a long or short position in the risk driver. However, as long as a consistent convention is adopted by each individual bank for all instruments then, the outcome remains the same whether the position is a long position or a short position, as SA-CCR does not differentiate between a negative and a positive value.
Katherine Wolicki