In a recent speech, Federal Reserve Governor Daniel Tarullo criticized the use of banks’ internal models for determining capital adequacy. There are several reasons to be dissatisfied with the internal ratings-based (IRB) approach, starting with the complexity and opacity that Governor Tarullo highlights. Our uppermost concern is the lack of consistent results across banks. That is, given the same portfolio of assets, different banks’ models yield very different estimates of required capital. These model-driven differences undermine both the trust in banks’ reported capital ratios and their usefulness.
As a part of the implementation of the Basel III capital standards for internationally active banks, supervisors have been reviewing the IRB approach for computing each bank’s risk-weighted assets. Last July, the Basel Committee published results of an extremely important “hypothetical portfolio exercise” (HPE). The Committee asked banks to use their internal models to estimate the capital requirements for holding a specific portfolio consisting of a wide range of public and private assets. Because each bank assessed the same portfolio, any difference in their estimated capital requirements must reflect differences in their IRB models.
The key HPE result is in the chart below, which shows that the required ratios of capital to risk-weighted assets for the same portfolio differ by up to four percentage points! Given that these banks have capital requirements averaging around 10 percent, this gap represents a difference of something like 50% from the highest to the lowest ratio.
As Governor Tarullo notes, banks’ internal models are complex and opaque, so it is difficult to know the source of these big differences. Yet, if we are to have credible global regulatory standards, capital requirements across large, internationally active banks must be consistent.
We see three approaches to meeting this goal. First, there is Governor Tarullo’s solution of discarding the IRB method in favor of one in which supervisors do the computations using their own model. He notes that this would be lots of work, but we have other concerns. As banks learn the supervisors’ model (either directly by having the supervisor share it or indirectly by seeing it applied repeatedly across many banks), regulatory arbitrage – gaming the model – would become rampant. Remember what happened when the credit rating agencies made their models available to the people who do securitization (and re-securitization)? In addition, a faulty supervisory model would encourage herding and risk concentration that amplify, rather than diminish, systemic vulnerability. Since nobody is perfect, we see this as an inevitable outcome of the move to a single model.
As a second approach, we could give up on risk weighting and move to a simple leverage ratio. In an earlier post, we discussed why we believe that relying solely on a leverage ratio is neither simple nor wise.
Finally, we could exploit what we have learned from the Basel Committee’s recent work and administer hypothetical portfolio exercises on a regular basis to make the differences between banks’ risk models transparent. Imagine that every quarter you could look at a set of charts that compare each of the largest banks’ evaluations of the risk-weighted assets associated with a set of 50 or 100 portfolios? Bank boards, senior managers and supervisors could all compare their results with those of other banks and ask that discrepancies be explained. Investors would surely ask questions, too, putting pressure on the banks that could force a race to the top.
If we are to continue to have risk-based capital regulation for banks, we need models to assess what is risky and what is not. Models are subjective and error-prone, so having a variety of them out there (rather than one supervisory version) is essential, as is competition to improve the models.
But we should tolerate differences only up to a point. To ensure that discrepancies remain small – without relying on a single model constructed by supervisors – we can require large banks to disclose standardized HPE computations that allow us to evaluate their models and form a better view of their true capital needs.