Through the Cycle – PD Scalar

hmm, not quite agreeing with your article I’m afraid.

Point in time (PIT) and Through the Cycle (TTC) should refer to a rating system. This is the whole concept that a bank uses to set its regulatory capital. In principle, a PIT system will be procyclical with the economy (ie estimate higher levels of capital in the bad times) and a TTC system will estimate constant levels of capital during the economic cycle. In simple terms a TTC system is better as there is concerns that a PIT system could cause banks to fuel an economic downturn (increased capital requirement leading to reduced lending activity). However, for a rating system to be effective and compliant it must react to changes in a business and the level of risk it is taking. For a TTC system it is very difficult to separate out the effects in changes in the economy and non-cyclical changes in risk (ie due to lending rules or borrower behavioural changes).

In either case, the capital calculated by the rating system must use a long run average (LRA) PD. As you say, there aren’t any banks which have LRA PDs for their retail business particularly when divided into their current risk grades (using their current scorecards). To estimate the LRA PD using a short PD history, an excepted technique is to apply a scalar to the short run of data. This is usually through external data and longer run portfolio level internal data. The ongoing monitoring of this adjustment is key to IRB compliance.

I have to say the basel 2 jargon everyone is facing is pretty confusing. I fell into the same misunderstanding and had to edit a very large pile of documents to be consistent with the regulators view.

So, for PDs to be compliant for calculating capital, they must be a true long run average by ratings grade. If this is not available (over a economic cycle, say 15-25 yrs) then it is acceptable to estimate this using a shorter run of data and a scalar adjustment. A short run average PD could be refered to as a PIT PD (I do, to be honest). However the capital model requires a “long run average” PD, either measured directly (unlikely for retail) or be estimation from a short run average PD (or PIT PD if you like). No matter the method, the result is a LRA PD.

The “through the cycle” term refers to the calculation of capital.
In reality this has more to do with ratings migration. A TTC rating system sets a constant level of capital during economic cycles (where other portfolio risks are fixed). A PIT rating system varies the amount of capital through a economic cycle, largely due the movement of exposures between ratings grades through time. For example, in a period of economic downturn it would be reasonable to expect that a proportion of exposures would shift into higher risk categories if up to date information is available (ie current behavioural credit scores, common in retail lending). This would increase the level of regulatory capital. This could have some undesirable effects on a macro economic level. The pillar 1 stress test is an important process for understanding the extent of this variability of capital for an individual portfolio and to form the basis of the pillar 2 capital plan (whch covers the way these capital obligations would be met by the business).

I hope that makes sence, it took me a while.

I think Bruce has nailed the key concept behind PIT/ TTC which is that this has more to do with the ability of the rating system to respond to changes in credit quality – i.e. portfolio migration. PIT and TTC represent different extremes on the spectrum. Correspondingly, the need to apply a scalar differs between a TTC and a PIT system. For a TTC system that scalar would change with differing economic conditions requiring constant monitoring (adjustment) whereas a PIT system is conditioned on current economic events and thus does not require an adjustment. The difficulty for most rating systems is that they are neither, rather they tend to be a ‘hybrid’ requiring an estimation of the portfolio migration contribution in order to estimate a lesser scalar than a completely TTC system. I don’t think that the requirement for a long run PD actually implies that a system has to be one or the other as the average from both systems will be the same through an economic cycle. However, given the implication for procyclicality for a PIT system, it is likely that a capital requirement will need to be held in Pillar 2 to compensate.

Regarding retail pools – to be honest i tend to think in terms of non-retail grades, but i’m sure the concept is the same. A behaviour score seems fairly PIT to me.