The Empire Strikes Back: How the Basel Proposal Would Shift Credit Allocation from the Private Sector to the Government

Most attention on the proposed implementation of the Basel agreement has focused on its significant capital increase as a whole and on particular asset classes. Something more fundamental is at stake, however: the relative role of the private sector and the government in allocating credit.

Banks compete for loans primarily based on price and terms, which are ultimately driven by each bank’s assessment of credit risk and risk tolerance. Thus, banks compete with each other most fundamentally on their ability to measure and manage risk. For a given loan, one bank might assess the risk differently from another, or may choose to take more risk in return for a higher price. This competition is the heart and soul of the banking industry and, one could argue, of the American market economy. Banks that manage risk well gain market share and earn profits; those that manage it poorly lose money and ultimately market share. American companies and consumers benefit because there is competition for their business, meaning a variety of prices and terms from which to choose.

The Basel proposal released by the U.S. banking agencies would significantly undermine that system by eliminating bank measures of credit risk from U.S. capital regulation altogether, with the result that credit allocation would be driven to a significant extent by a government framework of simplistic, one-size-fits-all rules and secret government models.


Currently, the largest banks operate under what is (aptly) called the advanced approach to credit risk. Each bank models the risk of its loans based on both external data and internal loss experience, subject to a range of parameters prescribed by regulation, and thus determines the appropriate credit risk weights for regulatory capital purposes. To ensure accuracy, the models used for that purpose are verified by internal compliance and audit teams and by agency examiners; they are constantly backtested, so any systematic underestimation (or overestimation) of risk can be noticed and corrected.[1]

The alternative to the advanced approach is a decidedly less advanced “standardized approach,” which was first conceived in the Basel I capital accord in 1988. By necessity and design, a standardized approach does not differentiate among loans or other exposures on a granular level; rather, it clumps them into very broad categories and assigns all the loans in each category the same risk weight. (So, for example, under the current standardized approach, all mortgages are assigned the same risk weight.) There is no backtesting – so, no course correction if the standardized charge is understating or overstating risk given problems with the initial standard or market changes subsequent to its adoption. Perhaps most importantly, there is no role for data or actual loss experience whatsoever; these broad categories and risk weights are not based on any quantitative assessment of risk but instead a one-time, unchanging decision by the Basel Committee. It is thus no surprise that the track record of these standardized risk weights is less than stellar; for example, at the time of the Global Financial Crisis, U.S. capital requirements were determined solely by this type of standardized approach.

Currently, the largest U.S. banks must calculate risk-weighted assets under both the standardized and advanced approach; all other banks (including regional banks) are subject only to the standardized approach. For the largest banks, the higher of risk-weighted assets under the standardized and advanced approaches binds, with the former most frequently functioning as the binding constraint.

In the aftermath of the Global Financial Crisis, the Basel Committee recognized that the advanced approach to credit risk was substantially more risk-sensitive — in other words, accurate — than the standardized approach, and set out to find a way to retain that approach yet better ensure that modeled credit risk would not vary inappropriately among banks. The grand compromise of the 2017 Basel agreement was to continue use of bank internal models for credit risk but ensure greater consistency by (i) introducing a range of new constraints and limits on particular modeling choices and parameters and (ii) providing that these could not produce an aggregate charge less than 72.5 percent of their standardized approach. European and UK regulators this year decided to follow that path. So, in other words, bank internal models will always determine the capital charge for loans, subject only to a limit on how low a charge they can produce. Furthermore, to ensure greater risk-sensitivity in the risk weights assigned to loans under the standardized approach, the 2017 Basel agreement also permits banks to use external credit ratings (Moody’s, S&P, Fitch) to determine credit risk weights for larger companies. Here, too, the UK and Europe have already followed suit.

The Basel Proposal

One might reasonably expect that, after negotiating these changes at the Basel Committee in 2017, the U.S. banking agencies would have promptly proposed to implement them here. But they have done no such thing.

Six years later, they have proposed to eliminate altogether the advanced approach to credit risk, making the United States the only major banking center in the world where a bank’s own assessment of credit risk would be irrelevant to its capital charge. Furthermore, under existing U.S. law, reliance on external credit ratings for capital purposes is prohibited. Only the standardized model devised in Basel in 2017 would be left standing, and imposed at 100 percent, in the U.S., not 72.5 percent like in the UK and Europe.

The Stress Test

In the United States (and only in the United States), the largest banks also receive a stress capital charge, calculated anew each year under the Federal Reserve’s annual stress test. (It is labeled by the Fed a “stress capital buffer,” but “buffer” is misleading because, since 2020, the stress test produces a binding charge that comes with penalties for failing to maintain it.) Akin to the advanced approach, the stress capital buffer is based on more than 20 relatively granular models that look at loan-level detail. While the models are more simplistic than bank models, they are still more risk-sensitive than any standardized approach, and unlike standardized risk weights, have the benefit of being based on actual data and loss experience.

Notably, however, those models are secret, and not subject to meaningful oversight. It is unclear if they are backtested; there is no evidence that they have changed over the years in the way that rigorous backtesting would certainly require. Furthermore, it is not even clear if the stress capital charge is really a product of those models. On the one hand, the Fed’s description of its stress test methodology states, “The Federal Reserve does not adjust supervisory projections for individual firms or implement firm-specific overlays to model results used in the stress test. This policy ensures that the stress test results are determined solely by supervisory models and firm-specific input data.” On the other hand, in response to a recent FOIA request, the Federal Reserve stated, “The setting of the stress capital buffer is not a rote application of the model[s], and the Board, in implementing the process, has the ability to exercise discretion to vary from the model outputs.” Only one of these statements can be true; which one is unclear.

There is evidence suggesting that the Federal Reserve’s models, either individually or when combined, may produce inaccurate outcomes. For instance, stress test results have varied significantly from year to year, even when the severity of the scenarios and the risk profiles of banks’ portfolios stayed similar. Moreover, there have been notable differences between outcomes from the Federal Reserve’s models and those produced by the banks using their own, more granular and risk-sensitive models. That variance has generally been left unexplained.

The Ramifications

The current proposal would mean that all banks with greater than $100 billion in U.S. assets would have their capital requirements determined by a combination of crude standardized models devised in 2017 and secret models operated by the central bank. There would be no room for the innovation and expertise that comes with private sector analysis of risk, nor would borrowers and the general public have any real insight into the government view of risk and underlying data that is reflected in the Fed’s secret models.

There are a range of problems with this outcome, but one is most obvious and significant – which is likely why no other major banking center has chosen it. Because the models are crude (standardized) and potentially inaccurate (stress test), capital will be misallocated in the United States. The effect will be subtle and intangible, but of course this is the essence of why, traditionally, markets have done a better job of allocating capital than governments. Also, banks will have less incentive to invest in risk management; they will compete less on their ability to measure and manage risk, because capital requirements are now so high as to drive decision-making, and those requirements would be set by the government.

The Solution

The best solution would be for U.S. regulators to follow the rest of the world and retain the advanced approach for credit risk, making it an option for any U.S. bank that is willing to develop and maintain the considerable infrastructure and internal controls necessary to implement it. If the advanced approach is eliminated, then it becomes even more imperative that the Federal Reserve provide public notice and seek comment on all the models it uses to assess bank performance for purposes of its stress capital charge, as well as the process by which those models are combined to produce the charge. Notably, those models would become the only granular, adaptable measures of credit risk at our largest banks for capital purposes. In addition to being good policy, transparency and public input on the models is also required by law.[2]

[1] The Advanced Approaches were first introduced as a regulatory capital measure by the Basel II Agreement in 2004 and incorporated into the U.S. capital framework in 2007. Basel III re-confirmed the commitment to Advanced Approaches in 2010, and by 2014 it was a fully integrated and operational component of the regulatory capital regime for large U.S. banks.