Article 363 of the EU CRR (Permission to use internal models) states that permission for an institution to use internal models to calculate own funds requirements is subject to competent authorities verifying compliance with:
A firm should be able to demonstrate that it meets the risk management standards in article 368 of the EU CRR (Qualitative requirements) on a legal entity and business-line basis where appropriate. This is particularly important for a subsidiary in a group subject to matrix management where the business lines cut across legal entity boundaries.
A VaR model permission will generally set out the broad classes of position within each risk category in its scope. It may also specify how individual products within one of those broad classes may be brought into or taken out of scope of the VaR model permission. These broad classes of permission are as follows:
linear products, which comprise securities with linear pay-offs (such as bonds and equities) and derivative products which have linear pay-offs in the underlying risk factor (such as interest rate swaps, FRAs, and total return swaps);
European, American and Bermudan put and call options (including caps, floors, and swaptions) and investments with these features;
Asian options, digital options, single barrier options, double barrier options, look-back options, forward-starting options, compound options and investments with these features; and
all other option-based products (such as basket options, quantos, outperformance options, timing options, and correlation-based products) and investments with these features.
A firm should ensure that the data series used by its VaR model is reliable. Where a reliable data series is not available, proxies or any other reasonable value-at-risk measurement may be used when the firm demonstrates that the requirements of article 367(2)(e) of the EU CRR (Requirements for risk measurement) are met. A firm should demonstrate that the technique is appropriate and does not materially understate the modelled risks
Data may be deemed insufficient if, for example, it contains missing data points, or data points which contain stale data. With regard to less liquid risk factors or positions, the FCA expects the firm to make a conservative assessment of those risks, using a combination of prudent valuation techniques and alternative VaR estimation techniques to ensure there is a sufficient cushion against risk over the close-out period, which takes account of the illiquidity of the risk factor or position.
A firm is expected to update data sets to ensure standards of reliability are maintained in accordance with the frequency set out in its VaR model permission, or more frequently if volatility in market prices or rates necessitates more frequent updating. This is in order to ensure a prudent calculation of the VaR measure.
In determining whether it is appropriate for a firm to use empirical correlations within risk categories and across risk categories within a model, the FCA expects certain features to be observed in assessing whether such an approach is sound and implemented with integrity. In general, the FCA expects a firm to determine the aggregate VaR measure by adding the relevant VaR measure for each category, unless the firm's permission provides for a different method of aggregating VaR measures which is empirically sound.
The FCA does not expect a firm to use the square root of the sum of the squares approach when aggregating measures across risk categories unless the assumption of zero correlation between these categories is empirically justified. If correlations between risk categories are not empirically justified, the VaR measures for each category should simply be added to determine its aggregate VaR measure. However, to the extent that a firm's VaR model permission provides for a different way of aggregating VaR measures:
A firm is expected to provide evidence of its ability to comply with the requirements for a VaR model permission. In general, it will be required to demonstrate this by having a back-testing programme in place and should provide three months of back-testing history.
If the day on which a loss is made is day n, the value-at-risk measure for that day will be calculated on day n-1, or overnight between day n-1 and day n. Profit and loss figures are produced on day n+1, and back-testing also takes place on day n+1. The firm's supervisor should be notified of any overshootings by close of business on day n+2.
Any overshooting initially counts for the purpose of the calculation of the plus factor, even if subsequently the FCA agrees to exclude it. Thus, where the firm experiences an overshooting and already has four or more overshootings for the previous 250 business days, changes to the multiplication factor arising from changes to the plus factor become effective at day n+3.
The FCA, will review as part of a firm's VaR model permission application, the processes and documentation relating to the derivation of profit and loss used for back-testing. A firm's documentation should clearly set out the basis for cleaning profit and loss. To the extent that certain profit and loss elements are not updated every day (for example, certain reserve calculations) the documentation should clearly set out how such elements are included in the profit and loss series.
In accordance with article 363(3) of the EU CRR (Permission to use internal models), the FCA expects a firm to provide and discuss with us details of any significant planned changes to the VaR model before those changes are implemented. These details must include detailed information about the nature of the change, including an estimate of the impact on VaR numbers and the incremental risk charge.
The use of overlapping intervals of 10-day holding periods for article 365 of the EU CRR (VaR and stressed VaR calculation) introduces an autocorrelation into the data that would not exist should truly independent 10-day periods be used. This may give rise to an under-estimation of the volatility and the VaR at the 99% confidence level. To obtain clarity on the materiality of the bias, a firm should measure the bias arising from the use of overlapping intervals for 10-day VaR and stressed VaR when compared to using independent intervals. A report on the analysis, including a proposal for a multiplier on VaR and stressed VaR to adjust for the bias, should be submitted to the FCA for review and approval.
Article 365 of the EU CRR requires a firm that uses an internal model for calculating its own funds requirement to calculate, at least weekly, a stressed VaR (sVaR) of their current portfolio. When the FCA considers a firm's application to use a sVaR internal model it would expect the features in IFPRU 6.3.20 G to IFPRU 6.3.24 G to be present prior to permission being granted, as indicative that the conditions for granting permission have been met.
The firm should calculate the sVaR measure to be greater than or equal to the average of the 2nd and 3rd worst loss in a 12-month time series comprising of 250 observations. The FCA expects, as a minimum, that a corresponding linear weighting scheme should be applied if the firm uses a larger number of observations.
The firm should ensure that the sVaR period chosen is equivalent to the period that would maximise VaR, given the firm's portfolio. There is an expectation that a stressed period should be identified at each legal entity level at which capital is reported. Therefore, group level sVaR measures should be based on a period that maximises the group level VaR, whereas entity level sVaR should be based on a period that maximises VaR for that entity.
The firm should explain the rationale for the choice of absolute or relative shifts for both VaR and sVaR methodologies. In particular, statistical processes driving the risk factor changes need to be evidenced for both VaR and sVaR.
The following information is expected to be submitted quarterly:
analysis to support the equivalence of the firm's current approach to a VaR-maximising approach on an ongoing basis;
the rationale behind the selection of key major risk factors used to find the period of significant financial stress;
summary of ongoing internal monitoring of stressed period selection with respect to current portfolio;
analysis to support capital equivalence of upscaled 1-day VaR and sVaR measures to corresponding full 10-day VaR and sVaR measures;
graphed history of sVaR/VaR ratio;
analysis to demonstrate accuracy of partial revaluation approaches specifically for sVaR purposes (for firms using revaluation ladders or spot/vol-matrices), which should include a review of the ladders/matrices or spot/vol-matrices, ensuring that they are extended to include wider shocks to risk factors that incur in stress scenarios; and
minutes of risk committee meeting or other form of evidence to reflect governance and senior management oversight of stressed VaR methodology.
Article 372 of the EU CRR (Requirement to have an internal IRC model) requires a firm that use an internal model for calculating own funds requirements for specific risk of traded debt instruments to also have an internal incremental default and migration risk (IRC) model in place to capture the default and migration risk of its trading book positions that are incremental to the risks captured by its VaR model. When the FCA considers a firm's application to use an IRC internal model, it expects that the matters in IFPRU 6.3.26 G to IFPRU 6.3.28 G will be included as demonstrating compliance with the standards in article 372.
The FCA expects the IRC model to capitalise pre-default basis risk. In this respect, the model should reflect that in periods of stress the basis could widen substantially. The firm should disclose to the FCA its material basis risks that are incremental to those already captured in existing market risk capital measures (VaR-based and others). This must take actual close-out periods during periods of illiquidity into account.
The price/spread change model used to capture the profit and loss impact of migration should calibrate spread changes to long-term averages of differences between spreads for relevant ratings. These should either be conditioned on actual rating events, or using the entire history of spreads regardless of migration. Point-in-time estimates are not considered acceptable, unless they can be shown to be as conservative as using long-term averages.
To achieve a soundness standard comparable to those under the IRB approach, LGD estimates should reflect the economic cycle. Therefore, the FCA expects a firm to incorporate dependence of the recovery rate on the economic cycle into the IRC model. Should the firm use a conservative parameterisation to comply with the IRB standard of the use of downturn estimates, evidence of this should be submitted in quarterly reporting to the FCA, bearing in mind that for trading portfolios, which contain long and short positions, downturn estimates would not in all cases be a conservative choice.