Credit Risk

Might also be worth taking a look at the detail in sections 2.5.5.1 - 2.5.5.2 dealing with customer migrations and migration matrix stability respectively, in the ECB's Instructions for reporting the validation results of internal models, pgs. 23 - 24

A number of UK banks have their wholesale PD models produce both “TTC” and “PIT” ratings – with the latter typically being driven by some sort of adjustment based on Moody’s KMV EDF’s or equivalents – you also have a number of other banks that for IFRS 9 will apply macro-economic model outputs on top of a largely TTC rating model to produce a PIT PD for IFRS 9 provisioning/ allowances purposes. The dual rating approach at big UK banks was heavily driven by a couple of modellers by the names of Scott Aguais and Lawrence Forest, and they have published a few papers describing their approach.

On the question of splitting by “investment Grade” vs “non-investment grade”, I’ve never seen this, although a split between leveraged and non-leveraged is common historically (when we built the model for one of the large German banks, we were able to reintegrate them). But size r whether a customer is rated/ quoted/ listed is a common basis for segmentation in the commercial space.

For EU banks, qualitative questions still tend to be included, although people are working to make them objective where possible, but not sure many have entirely removed them (again Scott Aguais was wanting to do this for one of the big UK banks – not sure whether he succeeded) – the issue with qualitatives is whether you can find the sweet-spot – at bottom end SME, often I don’t think credit officers have any real insights into their customers given how many they cover; at top-end, question is whether the credit officer has real insight beyond what can be captured by tools such as Factiva Sentiment Signals

Historically, it has been difficult to remove size segmentation from the entire corporate customer base, in part because you tend to see some factors have different relationships for small vs large firms (e.g. you might want a fair bit of cash on balance sheet of an SME, but for a large firm, this would be inefficient and you’d perhaps worry if management wasn’t trying to make their cash work hard )

Airplanes are ships of the sky from an LGD perspective, but not from a PD perspective. Railcars are also ships of the land by the same token

From a PD perspective, they are generally separate. The maritime companies and airlines have very different obligor dynamics

From an LGD perspective, the main drivers are downtime (i.e. how long the asset sits idle after a default) and shortfall (i.e. the decrease in the new lease after a default since usually periods of default coincides with pressure on asset prices and lease rates), but not the value on the plane. It is seldom that you’d actually lose the asset in any meaningful way, so that does not affect the LGD. There are international treaties that lessors would seek the jurisdiction of the lessee be a part of, before sending an expensive plane over with a long lifetime left on the asset. The other jurisdictions get the older planes where the lessor does not care all that much, whether they get it back or not

To be clear, you’d parameterize the LGD model differently for different assets as the downtime and shortfall dynamics are different. But the model structure is the same. So, it becomes a bit of an optical choice on whether you call that a single model or not

This may just be a rumour and so worth trying to Google to verify it, but I recall a long time ago that the MKMV model didn’t work well for banks, this was a known shortcoming of the model

It’s not at all clear what other type of structural model (other than something like KMV) one could even attempt for a bank…

Hi there,

Based on previous experience, for PD this is often not relevant: PDs are 12-month and the seasoning tends to be generally captured by the scoring model itself. A qualitative explanation of each scoring model and which characteristics it is considering that relate to seasoning may be enough, especially if complemented with quantitative analyses on the seasoning effect.

For a more quantitative approach, suggest testing time since origination and time until maturity as potential risk drivers using the general risk driver assessment framework during PD calibration - in the past I've observed this not to be significant but again, this is anecdotal evidence.

On LGD it may be relevant. However it should be understood that seasoning actually correlates with other significant risk drivers, particularly LTV and outstanding exposure amount. Here a deeper analysis of these parameters' significance should help "paint the broader picture".

Regards

A couple of thoughts on this subject, from one of our experts:

The discrepancy is caused by the adjustment implicitly assuming that a Bank would have had more defaults and lower scores (and so a worse average score) – while applying the theorem to a population which still has the same set of defaulted cases. This means the average scores are not worse, and hence you predicted PD will be lower.

There are at least two approaches to deal with this effect:

Adjust the constant term in the logistic until it hits the 2% target Run a “goal seek analysis" so that the average PD after mapping scores to the Bank grades, and applying the appropriate post-rating adjustments so the PD reaches 2%

Especially for European banks IRB models are actually required to be quite conservative unless Banks have "perfect" data, so the long-run average can become a moot point to a certain extent

On the topic of perfect data: if the Bank has enough data and the PD model is really powerful, it should find that there is no straight-line relationship between PD from logistic model vs. observed default rate. This is actually caused by the fact that whilst the errors are broadly normally distributed in logOdds space, when the distribution is converted to PD/default rate space the expectation will be closer to the mean than the original prediction.