Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
Risk management is a lot like Warhammer 40K. Partly because it’s a nightmarish dystopia dominated by mutant cyborgs, but mainly because in order to play, you have to have a massive and surprisingly expensive army of models.
But how much more chaotic, grim and dark would the world of Warhammer be if you had to get official approval every time you touched up the paintwork on one of your Space Marines?
This might be one of the reasons why, although Warhammer is a massively lucrative global business, the recent updates to the European Central Bank’s Guide to Internal Models struggled to get coverage outside the ECB’s own newsletter. But for people of rarefied tastes, the world of model reviewing might be about to see some struggles that would put the Horus Heresy* to shame.
It’s all to do with model permissions. Pretty obviously, if you are going to use an internal model to calculate your regulatory capital requirements, then the supervisor needs to check that it’s adequate to the task. That’s a tricky and specialised task, and the ECB, like other supervisors, doesn’t have unlimited resources of the kind of skilled labour that’s needed to do it.
For a thorough “internal model investigation”, the ECB appears to have the capacity to do about a hundred a year:
The number of these investigations isn’t really under the ECB’s control. Only about 5 per cent of IMIs are launched on the supervisors’ initiative, usually as a result of concerns raised during the annual inspection process. The vast majority of them are carried out because the bank wants a new model approval, the extension of an existing model to a new asset class, or to make a “material change” to a model.
And that’s potentially a problem. Because a number of recent developments might come together in the not too distant future to significantly increase the demand on these limited resources:
The “Fundamental Review of the Trading Book” is what it says — a complete change of the rules for market risk models, requiring anyone who wants to use one to get it approved again.
The EU has, in response to the demise of the Fed’s “Endgame” proposals, delayed the implementation of the new market risk rules in Europe. This means that the official handbook now has to have two separate sections for market risk — one for the existing models (which need to be kept up to date as long as they are being used) and one for the new kind.
Some banks are going to want to start using machine learning in their models, which brings a new set of challenges because it isn’t always necessarily easy to know what a “material change” means in the context of some models of this kind.
The most recent version of the Capital Requirements Regulation shifted the basis of model approval for credit risk, from a “per institution” (you either use internal models for all your credit risk except sovereign bonds, or none of it) to “per asset class” (you can choose which categories to model, out of the ones where internal modelling is still permitted).
That last one is most important, because as the chart above shows, credit risk is the biggest source of model investigations. And as well as for new internal models, you need supervisory permission to stop using internal models and revert to the “standardised approach”. (This is to stop banks from “cherry picking” regulatory treatments).
Since the new rules were passed, a lot of the ECB’s model approval workload has resulted from banks taking advantage of the new rules to drop internal models for small or non-strategic asset classes, which were never really worth the expense of maintaining them but which had to be kept because of the “all or nothing” principle.
But potentially more importantly, there’s a significant risk that the new system will generate more work in and of itself.
Consider, for example, a change in group risk management policy which affects all asset classes, but which is implemented over time, with the same team spending a few weeks on one model then moving on to the next. Is this one “material change”, or half a dozen? If you extend the dataset of a model to cover more assets, at what point does this require a permission request? And so on.
Nobody benefits from creating a great big bottleneck of model approvals, so the ECB has been subtly and quietly lobbying the European Banking Authority to easy up a bit. The minutes of the last EBA Board of Supervisors’ meeting outline the problem, if you read between the acronyms:
On the RTS [Regulatory Technical Standard] on material model changes, the Head of RBM [Risk Based Metrics] reminded the Members of the discussion during the BoS [Board of Supervisors] meeting in October 2024.
He noted that there was a disagreement at the experts’ level, which was preventing a further simplification of the RTS. In particular, the ECB Banking supervision, which had around 80% of IRB [Internal Risk-Based] RWA [Risk-Weighted Assets] under their scope, was challenged with the number of material model changes, both due to the sheer number of material model changes, but also probably because the largest banks had more complex internal models.
Basically, the ECB (which does the majority of the work) wants to stop itself being overwhelmed with model changes, but it doesn’t make the rules on materiality. The EBA sets those rules, but its voting members represent national authorities, some of which aren’t under such pressure from approval requests. They can therefore afford to be risk-averse about the possibility of allowing non-compliant models to hang around for longer.
You can see this in the following paragraph from the EBA board meeting minutes, where “Option 1” is to do nothing, “Option 2” is to make marginal technical changes and “Option 3” is to really have a go at reducing the number of changes:
. . . a few Members supported option 1; one Member supported options 1 and 2 saying that option 3 could result in a risk of implementation of non-compliant models and that the IRB repair programme has advanced significantly and therefore, there would be fewer model changes in the future. One Member supported option 2 and asked for additional flexibility for supervisors, and several Members supported option 3.
The minutes seem to indicate that there’s majority support for helping the ECB out, but it’s all taking longer than it might have done.
In the meantime, the banking industry could help itself out a bit by trying to be sensible in bundling together requests, and making sure that all their documentation is in order.
The ECB has already had occasion to be a little bit grumpy about the quality of some of the requests they’ve been receiving, at least judging by its newsletter write-up of the situation:
Banks are also encouraged to submit high-quality and accurate applications up front. Better submissions make the assessment process more efficient. They also ensure timelines are more predictable and they enable swifter decision-making. All this is ultimately to the benefit of both the ECB and the banks themselves.
If inefficiencies are averted, banks will eventually benefit from faster implementation of material model changes. The tone from the top is crucial for achieving these objectives, and the ECB will continue to assess whether banks’ management teams are committed to the quality of their applications. If there are poor quality applications and/or persistent deficiencies, the ECB may look more closely at the institution’s underlying governance arrangements.
* The author respectfully requests that any and all quibbling over details of the WH40K fictional universe be readdressed to somebody who cares.
