Modellability of Risk Factors Passing the RFET

may be considered representative for a systematic risk factor as long as they share the same attributes as the systematic risk factor.

327 In addition to the approach set out in paragraph 326, where systematic risk factors of credit or equity risk factors include a maturity dimension (e.g. a credit spread curve), one of the bucketing approaches set out above must be used for this maturity dimension to count real price observations for the RFET.

328 Once a risk factor has passed the RFET, an AI should choose the most appropriate data to calibrate its model. The data used for calibration of the model does not need to be the same data used to pass the RFET.

329 Once a risk factor has passed the RFET, an AI should demonstrate that the data used to calibrate its ES model are appropriate based on the principles set out in subsection 20.2. Where the AI has not met these principles to the satisfaction of the HKMA for a particular risk factor, the HKMA may choose to deem the data unsuitable to calibrate the model and, in such case, the risk factor should be excluded from the ES model and subject to capital charges as an NMRF.

330 There may, on very rare occasions, be a valid reason why a significant number of modellable risk factors across different AIs may become non-modellable due to a widespread reduction in trading activities (for instance, during periods of significant cross-border financial market stress affecting several AIs or when financial markets are subjected to a major regime shift). One possible supervisory response in this instance could be to consider a risk factor that no longer passes the RFET as modellable. However, such a response should not facilitate a decrease in capital charges. The HKMA will only pursue such a response under the most extraordinary, systemic circumstances.

20.2 Modellability of Risk Factors Passing the RFET

331 An AI may use various types of models to determine the risks resulting from its trading book positions. The data requirements for each model may be different. For any given model, the AI may use different sources or types of data for risk factors.

The AI should not rely solely on the number of real price observations to determine whether a risk factor is modellable. The accuracy of the source of real price observations should also be considered.

332 In addition to the requirements specified above, an AI should follow the principles set out in paragraphs 333 to 339 to determine whether a risk factor that passed the RFET can be modelled using the ES model or should be subject to capital charges as

an NMRF. The AI is required to demonstrate to the HKMA that these principles are being followed. The HKMA may determine a given risk factor to be non-modellable in the event the AI does not follow these principles for the risk factor.

333 Principle one: The data used to price instruments may include combinations of modellable risk factors. In general, risk factors derived solely from a combination of modellable risk factors are modellable. For example, risk factors derived through multifactor beta models for which inputs and calibrations are based solely on modellable risk factors can be classified as modellable. However, a risk factor derived from a combination of modellable risk factors that are mapped to distinct buckets of a given curve/surface is modellable only if this risk factor also passes the RFET.

334 Principle two: The data should allow the internal models to capture both general market and idiosyncratic risk. If the data used in the model do not reflect either idiosyncratic or general market risk, an AI should apply an NMRF charge for those aspects that are not adequately captured in its models.

335 Principle three: The data should allow the model to reflect volatility and correlation of the risk positions. An AI should ensure that it does not understate the volatility of an asset (e.g. by using inappropriate averaging of data or proxies) and accurately reflects the correlation arising between risk factors.

336 Principle four: The data should be reflective of prices observed and/or quoted in the market. Where data are not derived from real price observations, an AI should demonstrate that the data are reasonably representative of real price observations.

The AI should periodically reconcile price data used in its internal model with front and back office prices. The AI should also document its approaches to deriving risk factors from market prices.

337 Principle five: The data should be updated at a sufficient frequency (at a minimum on a monthly basis but preferably daily) in order to account for frequent turnover of positions in the trading portfolio and changing market conditions. Furthermore, where an AI uses parametric functions, e.g. regressions, to estimate risk factor parameters, these should be regularly re-estimated at a minimum on a bi-weekly basis. Calibration of pricing models in the internal models should also not be less frequent than the calibration of front office pricing models. The AI should have clear policies and sound processes for backfilling and/or gap-filling missing data.

338 Principle six: The data used to determine stressed expected shortfall (ESR,S) should be reflective of market prices observed and/or quoted in the period of stress. The data

for the ESR,S model should be sourced directly from the relevant historical period whenever possible. There may be cases where the characteristics of current instruments in the market differ from those in the stress period. Nevertheless, an AI should empirically justify any instances where the market prices used for the stress period in its internal models are different from the market prices actually observed during that period. Further, in cases where instruments that are currently traded did not exist during a period of significant financial stress, the AI should demonstrate that the prices used match changes in prices or spreads of similar instruments during the stress period.

339 Principle seven: An AI should limit the use of proxies, and proxies used should have sufficiently similar characteristics to the transactions they represent. Proxies should be appropriate for the region, quality and type of instrument they are intended to represent. The HKMA will assess whether methods for combining risk factors are conceptually and empirically sound.