The BIS just released the Regulatory Consistency Assessment Programme (RCAP) – Second report on risk-weighted assets for market risk in the trading book (December, 2013). It looks at the variability of RWA results across banks. It is useful to examine the BIS’ most recent thoughts on the topic.
Banks can tweak their internal models to generate the most advantageous RWA numbers – no surprises there. The BIS cited differences on the length of data periods, correlation assumptions and probability of default as among several of the drivers of RWA variability. But as a result, similar assets at one bank could absorb different RWA amounts at another. This makes meaningful comparisons impossible. We know that no one bank is identical to another, but shouldn’t the RWA results for similar businesses look the same across institutions?
We have written about this issue before. Back in July, we posted “Banca d’Italia paper looks at factors behind RWA calculations, why they differ so much”.
From the December 2013 report,
“…The focus of the analysis was to identify the design elements of banks’ internal models that contribute to the observed variability in trading book RWAs…”
and
“…Some variation in RWAs should therefore be expected across banks…it is undesirable for banks’ capital calculation inputs to generate excessive variation in risk measurement, as it would undermine the credibility of capital ratios, distort the international playing field and hamper the functioning of financial markets…”
Differences in internal modeling drive the variation in RWA. “…analysis confirms that differences in modelling choices are the most significant drivers of the amount of variation in market risk RWAs (mRWA) across banks…”
In the first report from the BIS in January 2013, the BIS recommended
“…(i) improving public disclosure and the collection of regulatory data to aid the understanding of mRWAs;
(ii) narrowing the range of modelling choices for banks; an
(iii) further harmonising of supervisory practices with regard to model approvals…”
This new report included more complex trading books, most notably correlation books. Those correlation books, most of which are legacy pre-2008, usually are in “shrink mode”. The internal risk models behind the correlation books (known as the comprehensive risk measure (CRM)) generally show more RWA variability across institutions than other models used for different parts of the trading book. In October 2013 the BIS, in an effort to reduce RWA variability, recommended that correlation books use a market-standardized approach.
The initial BIS exercise asked banks to simulate RWA for a set portfolio of “vanilla” trades. But in the new study, a more diversified set of trades was included, better simulating a “real” trading book. Adding to the sample size resulted in less RWA diversification – also not a surprising result but a very important issue to keep in mind when these studies are done. There might be a lesson in here about the lower risk of a larger, more diversified book but we will save that for another time.
Also as part of the new study, banks were asked run a permutation using consistent assumptions for a number of variables. By fixing those inputs, the BIS could see how much of the variability was driven by those variables. “…When differences in the choices of these drivers were removed, the observed variability across banks fell by approximately one third…”
The implication may be that regulators may be encouraged to manage model variables and standardize them across banks. But in the report the BIS noted “…it is desirable to have some diversity in risk modelling practices; if all banks modeled in the same way, they could create additional financial instability…” If nothing else was learned from the financial crisis, we know that no model is perfect and each has one or more Achilles heel. And those vulnerabilities change over time. Having everyone do the exact some thing creates correlations that may not be pretty in a stressed environment.
Tweaking the models that banks use to advocate a uniformity agenda may help fix the RWA variability problem – one of the main rubs driving the push for the simplicity of the SLR capital rules. At the end of the day, we wonder if standardizing model inputs will be emerge as the least worst solution, giving U.S. regulators some room to tone down the one-size-fits-all SLR rules?