I’ve noted several times in recent months that the Basel Committee on Banking Supervision is slowing turning its weighty turret towards the key secret in Australian banking: its use of internal risk weightings to allocate capital to mortgages.
To cut a long story short, the opaque internal risk models of the banks enable them to heavily discount the amount of capital they are required by Basel to reserve in the event that there’s a problem with defaults. This factor ultimately determines the bank’s leverage and thus the amount it can lend.
Australian banks run opaque models that nobody knows (except maybe APRA but who knows?). And the circumstantial evidence, such as fabulously over-valued housing, rather suggests that the banks are being a tad optimistic in their modelling assumptions, allowing them to lend more and create said over-valuation.
Well, that’s going to change. At least, maybe it is. Basel wants to bring such models into the light and last night it released a new study into how. Firstly, it lays blame for the problem mostly at the feet of regulators:
- A sizeable portion of the variation is due to supervisory decisions applied either to all banks in a jurisdiction, or to individual banks. An example of the former would be policy decisions to restrict modelling options (eg to disallow any diversification benefit between types of risk). An example of the latter would be the application of supervisory multipliers: around one-quarter of the total variation in the hypothetical diversified portfolio could be attributed to this single factor. These supervisory actions typically result in higher capital requirements than would otherwise be the case but can also increase the variation in mRWA between banks, particularly across jurisdictions. These supervisory actions, particularly at an individual bank level, are often not disclosed.
- Another important source of variation is due to modelling choices made by banks. The exercise found that a small number of key modelling choices are the main drivers of the remaining model-driven variability.
Next, it looks at how to make such models transparent:
Common standards for the frequency of reporting – less than half of the banks in the sample reported information on a quarterly basis;
2. Common standards for explanations of the drivers of the change in mRWAs from period to period;
3. A more granular and consistent segmentation of the components of mRWAs to facilitate a deeper recognition of a bank’s market risks;
4. Disclosure of key modelling choices, particularly those highlighted by the hypothetical test portfolio exercise as driving the greatest variation in the results of models; and
5. Disclosure of key differences in models used for internal risk management and those used for regulatory capital calculations. It was found that banks seldom directly report the 10-day 99% VaR used in regulatory capital calculations
Closely defining the modelling approach for the IRC model, including the assumptions used for migration and default probabilities and the correlation structure;
• Reducing the flexibility in choosing the length of historical data to calibrate VaR models; and
• Defining a single scaling approach to obtain a 10-day VaR and sVaR measure.
These areas can be considered in addition to policy options which already form part of the fundamental review of the trading book, namely:
• Strengthening the relationship between standardised approaches and internal models to be able to benchmark internal model results;
• Moving from separate VaR and sVaR based measures to a single Expected Shortfall based measure; and
• Enhancing regulatory oversight through a more granular approval process of internal models.
This is a policy discussion document and is not yet in train for actual rules. But there is every chance it will be.
Which makes me wonder why APRA does not have a representative on the panel doing the study. I don’t know why but it seems a bad idea.