24 IMA Default Risk Charge
400 An AI should have a separate internal model to measure the default risk of trading book positions. The general criteria in subsection 18.2 and the qualitative standards in subsection 18.3 also apply to the default risk model.
401 Default risk is the risk of direct loss due to an obligor’s default as well as the potential for indirect losses that may arise from a default event.
402 An AI should measure the IMA default risk charge (IMA-DRC) by using a VaR model.
The AI should use a default simulation model with two types of systematic risk factors.
Default correlations should be based on credit spreads or on listed equity prices.
Correlations should be based on data covering a period of 10 years that includes a period of stress as defined in paragraph 384 and based on a one-year liquidity horizon.
The AI should have clear policies and procedures that describe the correlation calibration process, documenting in particular in which cases credit spreads or equity prices are used.
The AI has the discretion to apply a minimum liquidity horizon of 60 days to the determination of default risk capital charge for equity sub-portfolios.
The VaR calculation should be conducted weekly and be based on a one-year time horizon at a one-tailed 99.9th percentile.
403 All positions subject to market risk capital charges that include default risk as defined in paragraph 401, with the exception of those positions subject to the Standardised Approach, are subject to the IMA-DRC model.
All sovereign exposures (independent of their denomination currency), equity positions and defaulted debt positions should be included in the model.
For equity positions, the default of an issuer should be modelled as resulting in the equity price dropping to zero.
404 The IMA-DRC is the greater of:
the average of the IMA-DRC measures over the previous 12 weeks; or
the most recent IMA-DRC measure.
405 An AI should assume constant positions over the one-year horizon, or 60 days in the context of designated equity sub-portfolios.
406 Default risk should be measured for each obligor.
Market-implied probabilities of default (PDs) are not acceptable unless they are corrected to obtain an objective PD.
PDs are subject to a floor of 0.03%.
407 An AI may reflect netting of long and short exposures to the same obligor in its IMA-DRC model. If such exposures span different instruments with exposure to the same obligor, the effect of the netting should account for different losses in different instruments (e.g. differences in seniority).
408 The basis risk between long and short exposures of different obligors should be modelled explicitly. The potential for offsetting default risk among long and short exposures across different obligors should be included through the modelling of defaults. The pre-netting of positions before input into the model other than as described in paragraph 407 is not allowed.
409 An AI’s IMA-DRC model should recognise the impact of correlations between defaults among obligors, including the effect on correlations of periods of stress as described below.
These correlations should be based on objective data and not chosen in an opportunistic way (depending on the mix of long and short exposures).
The AI should validate that its modelling approach for these correlations is appropriate for its portfolio, including the choice and weights of its systematic risk factors. The AI should document its modelling approach and the period of time used to calibrate the model.
These correlations should be measured over a liquidity horizon of one year.
These correlations should be calibrated over a period of at least 10 years.
The AI should reflect all significant basis risks in recognising these correlations, including, for example, maturity mismatches, internal or external ratings etc.
410 An AI’s IMA-DRC model should capture any material mismatch between a position and its hedge. With respect to default risk within the one-year capital horizon, the model should account for the risk in the timing of defaults to capture the relative risk from the maturity mismatch of long and short positions of less-than-one-year maturity.
411 The IMA-DRC model should reflect the effect of issuer and market concentrations, as well as concentrations that can arise within and across product classes during stressed conditions.
412 As part of the IMA-DRC model, an AI should calculate, for each and every position subjected to the model, an incremental loss amount relative to the current valuation that the AI would incur in the event that the obligor of the position defaults.
413 Loss estimates should reflect the economic cycle; for example, the model should incorporate the dependence of the recovery on the systemic risk factors.
414 The IMA-DRC model should reflect the non-linear impact of options and other positions with material non-linear behaviour with respect to default. In the case of equity derivatives positions with multiple underlyings, subject to approval by the HKMA, simplified modelling approaches (e.g. modelling approaches that rely solely on individual jump-to-default sensitivities to estimate losses when multiple underlyings default) may be applied.
415 Default risk should be assessed from the perspective of the incremental loss from default in excess of the mark-to-market losses already taken into account in the current valuation.
416 Owing to the high confidence level and long capital horizon of the IMA-DRC, robust direct validation of the IMA-DRC model through standard backtesting methods will not be possible.
Accordingly, validation of an IMA-DRC model necessarily should rely more heavily on indirect methods, including but not limited to stress tests, sensitivity analyses and scenario analyses, to assess its qualitative and quantitative reasonableness, particularly with regard to the treatment of concentrations.
Such tests should not be limited to the range of events experienced historically in order to ensure the soundness of the IMA-DRC model.
The validation of an IMA-DRC model represents an ongoing process in which an AI and the HKMA jointly determine the exact set of validation procedures to be employed.
417 An AI should strive to develop relevant internal modelling benchmarks to assess the overall accuracy of its IMA-DRC model.
418 Due to the unique relationship between credit spread and default risk, an AI should seek approval from the HKMA for each trading desk with exposure to these risks, both for credit spread risk and default risk. Trading desks which do not receive approval will be deemed ineligible for internal modelling standards and be subject to the Standardised Approach.
419 Where an AI has approved PD estimates as part of the internal ratings-based (IRB) approach, these data should be used. Where such estimates do not exist, PDs should be computed using a methodology consistent with the IRB methodology and satisfy the following conditions.
The AI should not use risk-neutral PDs as estimates of observed (historical) PDs.
The AI should measure PDs based on historical default data including both formal default events and price declines equivalent to default losses. Where possible, these data should be based on publicly traded securities over a complete economic cycle. The minimum historical observation period for calibration purposes is five years.
The AI should estimate PDs based on historical data of default frequency over a one-year period. The PD may also be calculated on a theoretical basis (e.g.
geometric scaling) provided that the AI is able to demonstrate that such theoretical derivations are in line with historical default experience (e.g. by using proxies).
The AI may also use PDs provided by external sources as long as they are relevant to its portfolio.
420 Where an AI has approved loss-given-default (LGD)58 estimates as part of its IRB approach, these data should be used. Where such estimates do not exist, LGDs should be computed using a methodology consistent with the IRB methodology and satisfy the following conditions.
The AI should determine LGDs from a market perspective, based on a position’s current market value minus the position’s expected market value subsequent to default. The LGD should reflect the type and seniority of the position and cannot be less than zero.
LGDs should be based on an amount of historical data that is sufficient to derive robust, accurate estimates.
An AI may also use LGDs provided by external sources as long as they are relevant to its portfolio.
421 An AI should establish a hierarchy ranking its preferred sources for PDs and LGDs, in order to avoid the cherry-picking of parameters.
58 LGD should be interpreted in this context as 1 – recovery rate.