Yes, we believe it would be good to include writing similar to the one proposed on LGD in default. I.e., that same downturn component may be utilized to downturn adjust LGD in default estimates.
Additional paragraphs which highlight that LGD in-default appropriate for a downturn could be estimated based on the downturn methodology performed for the LGD estimates of non-defaulted exposures could be added.
The proposed policy in paragraph 15 could create undue burden if applied to every downturn period identified.
To reduce operational burden, a simplified approach would be to compare loss data with the relevant economic factors. If losses do not increase as a result of economic factors indicating a downturn period, even if introducing an appropriate time lag, that should be enough evidence to exempt the identified downturn period from the proposed policy in paragraph 15.
Evidence as structural breaks could be documented by the institution and approved by the supervisor.
If the economic factors analysis reveals different downturn periods, one could consider using the average of the different downturn periods instead of the worst of the worst.
It is not clear on which level DT should be identified and we believe that this may cause undue variation. It will be more beneficial from a capital planning point of view to have fewer grades or pools, although from a steering point of view this could potentially drive risk. If estimates shall be based on the worst observed crises, then the more granular you get in estimation the less you may benefit from diversification effects. I.e. the granularity level of estimates may have a significant impact on DT estimates.
More details could be provided i.e. the proposed level of the downturn LGD estimation and the relation to the downturn periods and economic factors could be more detailed.
There seems to be an issue with the granularity. The proposed policy in the GL allows to quantify downturn LGD estimates at a more granular level than the long-run average LGD estimates (in the case where this provides more appropriate downturn LGD estimates).
Yes. However, we see a potential driver of undue variation in adapting external time series to internal ones. In particular, variations may be significant due to how discounting (5 % add-on) and time in default is accounted for in external data. Competent authorities have an important responsibility in assuring there is a level playing field across banks.
It’s positive to harmonise the best practice and limit the approaches for quantification of downturn LGD estimates.
It is impossible to answer the question as it is unclear when supervisory authorities will find extrapolation methods non-applicable. Given a strict interpretation and reliance on data from 1990’s this may be a large share of non-retail portfolios.
The impact can be both material and immaterial. The 20% add-on seems conservative and LGDs might be overestimated. A better understanding of the chosen 20% as add-on is needed.
Corporate exposures not secured by real estate may fall within this category.
We would suggest a QIS to detect possible differences between different exposure classes, type of collaterals and geographical region (i.e. legislation). It should not be beneficial to use add-on but neither should you be heavily punished for having to use MoC due to the fact that it has not experienced a crisis during recent years.
As stated in the answer to Q6 above, the chosen 20% add-on needs to be justified. It seems conservative for specific portfolios. The 20% add-on could be replaced by the reference value approach.
We would suggest a QIS to see if 20% unit add-on is feasible, see answer to Q6 and Q8.
The reference value approach seems to be sufficiently described.
However, the calculation of the reference value might be problematic in some cases. In pools with a low number of facilities, the calculated reference value might be more impacted by the low number of facilities than the losses.
For structural models there might be correlation effects between the components.