RISIKO MANAGER 10.2019

16

16 RISIKO MANAGER 10|2019 economic indicators are not reached simultaneously but are nonetheless a product of one isolated downturn.. The specified downturn conditions should be evidenced by elevated levels of realized LGD, including treatment for incomplete recovery processes driven by stressed levels of the relevant economic indicators. If high levels of realized LGD are not experienced simultaneously with the stress in economic indicators (but are still the result of such stress), they will be considered as the LGD estimates appropriate for the economic downturn. The most recent 20 years can be replaced with the most recent 10 years for estimations made during 2018. After 2018, the 10 year period should be increased each year by one year, until the 20 year period is re-established, provided representativeness requirements are met. Banks should be able to provide evidence that the period considered contains years which include adverse economic conditions. To identify the two individual years referred to above, B are expected (i) to group all defaults within the RDS and corresponding exposures and losses by the year in which the default occurred, in addition to obtaining the ratio of total losses to total exposure, and (ii) selecting the two individual years with the highest ratio of total losses to total exposure. This analysis considers years for which the maximum length of the recovery process has been observed. Reference values as the average realized LGD from two individual years, for each facility grade or pool which they use, are calculated by the consultant. Exceptions are only permitted when the LGD estimates result from combining different components (e.g. secured and unsecured). If downturn LGD estimates (by facility grade or pool) or estimates of model components (including margin of conservatism) obtained by bank are lower than those resulting from the reference value described above, the bank must provide evidence that its downturn LGD methodology is aligned with the target of elevated LGDs driven by economic conditions. The reference value should not be considered as a valid methodological option. B are expected to develop internal methodologies. If the bank does not have data series with the length described above, or cannot provide evidence that the available data include adverse economic conditions, the approach described shall be used with the available data series, an add-on or margin of conservatism MoC shall be applied. The economic environment observed for the data available (i.e. the better the observed economic environment, the higher the add-on or MoC should be) must be considered. Calculation of Expected Loss Best Estimate ELBE and LGD in-default Banks must take into consideration all relevant post-default information in their ELBE and LGD in-default estimates promptly. Where events from the recovery process invalidate the recovery expected by the most recent ELBE estimation, banks should update their ELBE and LGD in-default estimates immediately. To estimate ELBE and LGD in-default, banks use the same reference data set as used for LGD estimation, complemented to account for any relevant information observed during the recovery process, and at each reference date. For ELBE and LGD in-default estimation, banks set the reference dates which can be used to group defaulted exposures in a significant manner in terms of the recovery pattern observed. Clarification of ELBE In line with the applicable prudential requirements, ELBE represents the best estimate of expected loss, given the current economic circumstances and exposure status. Therefore, it should be sensitive to the current economic conditions, not fixed to long-run average. Nevertheless, the ELBE may be based on historical data (e.g. on a recoveries reference data set), and adjusted for conditions expected throughout the recovery process. The computation of the ELBE should reflect economic losses and should account for information on the time in default and recoveries already realized. Throughout the time in default, both the occurrence and non-occurrence of specific recovery events could influence the ELBE estimation. It is expected that recent operational information from the ongoing recovery process, including time since the default, is input into the ELBE estimation of each exposure. Banks should take into consideration all relevant post-default information for ELBE estimation. Where an bank considers time in default as a material risk driver, the ELBE estimation should take into account the outstanding exposure as at that reference date (i.e. time in default) rather than as at the time of default, as direct and indirect costs assigned to the credit obligation after that point, and expected recoveries, including appropriate discounting effects up to the reference date. Banks may use specific credit risk adjustments as ELBE estimates where the accounting model used to determine credit risk adjustments satisfies the requirements for own LGD estimates. In any case, banks are expected to demonstrate that ELBE estimates are adequately back-tested. The computation of the ELBE should reflect economic losses, and should account for information on the time in default and recoveries already realized. Throughout the time in default, both the occurrence and non-occurrence of specific recovery events could influence the ELBE estimation. Recent operational information from the ongoing recovery process, including time since the default, must be input for the ELBE estimation of each exposure. B should take into consideration all relevant post-default information for ELBE estimation. If the bank considers time in default as a material risk driver, the ELBE estimation will take into account the outstanding exposure at this reference date (i.e. time in default), rather than at the time of default. Besides direct and indirect costs assigned to the credit obligation after that point, expected recoveries – including appropriate discounting effects up to the reference date – must be added. B may use specific credit risk adjustments as ELBE estimates, where the accounting model used to determine credit risk adjustments satisfies the requirements for own LGD estimates. In any case, B are expected to demonstrate that ELBE estimates are adequately back-tested.

Kreditrisiko 17 Clarification of LGD in-default LGD in-default can be estimated directly, or as the sum of ELBE and an add-on capturing the unexpected loss, related to the exposures in default, that may occur during the recovery period. The use of a constant charge for unexpected losses for all defaulted exposures is not risk-sensitive and therefore is generally not acceptable. It should demonstrate that the specific constant charge is an adequate estimate of possible additional unexpected losses during the remaining recovery period, i.e. between the date for which estimates are being applied and the final closure of the recovery process. LGD in-default estimates are generally expected to be higher than ELBE. In exceptional cases, the LGD in-default equals the ELBE the relevant bank needs to justify. Banks should think through the deviations between LGD in-default and LGD estimates. Banks assess and duly justify situations where there are systematic deviations of the LGD in-default estimates just after the date of default from the LGD estimates, but before the date of default at the facility grade or pool which are not stemming from the use of risk drivers applicable from the date of default and onwards. Credit Conversion Factor CCF Banks which do not use their estimates of conversion factors for non-retail exposures are permitted to apply a conversion factor in certain situations. Realized conversion factors are calculated at the facility level. Banks are expected to demonstrate detailed understanding of the impact of changes in customer product mix on conversion factor estimates. The impact must effectively be addressed within the bank's estimation process. This is because changes in exposure characteristics or product profile transformations (e.g. a revolving loan that has been converted into a term loan or vice versa) which commonly occurs between reference and default dates that have a high potential for introducing substantial arbitrariness and downward bias into the estimates of CFs. The EAD for undrawn commitments is calculated as the committed but undrawn amount multiplied by a CCF. CCFs can also be derived from direct estimates of total facility EAD. The RDS should not be capped at the principal outstanding amount or facility limits. The RDS should include accrued interest, other due payments, and limit excesses. B should analyse the risk drivers not only twelve months before default, but also their volatility. Specified downturn conditions are evidenced by elevated levels of realized CCFs. Any lag between the beginning of the downturn period and the date of impact on the realized CCFs must be considered. Summary and conclusion Various inspection techniques and quality measures are explained in detail throughout this article. The technical tools are essential for successful quality improvement. Quality management systems alone do not reduce variability or improve quality. An assessment for the Internal Rating Based IT infrastructure must be created, presenting the main sources and databases involved. The various methods for getting data-flows for the IRB process of selected models are presented. The IRB data quality framework and processes in place should relate to the conclusions and findings on the level of the banks. Specifically the selected credit risk model should be reviewed. The review must contain related findings of the technical implementation, the bank's internal definition of default, and the potential implication of these on the historical data and subsequent PD/ LGD model estimates. The qualitative assessment performed, together with the interaction with the CR assessment team, and the available resources for the data quality review, serve as input for the selection of additional tests to be performed. For the future, more sophisticated measures and methods (may be taken from other scientific disciplines like physics, medicine, computer science with AI and machine learning, etc.) need to be found and applied.