The capital rule proposed by the federal banking agencies establishes risk weights for residential mortgages that exceed both those negotiated by agency staff at the Basel Committee on Banking Supervision and, more importantly, what empirical historical analysis demonstrates is appropriate for the U.S. mortgage market. Furthermore, while the Basel agreement allows banks to use their own internal loss data to assign risk weights to mortgages, the proposed rule would make the United States the only country to reject that option and rely solely on a one-size-fits-all government formula to assess mortgage risk.
U.S. mortgage loans would also incur an additional charge for operational risk and continue to incur a stress capital charge determined by the Federal Reserve’s stress test. Both charges would be unique to the United States, as the operational risk charge would be in large part driven by sales of mortgage loans to the U.S. government-sponsored enterprises (GSEs) that have no foreign parallel, and the stress test add-on has not been adopted in any other jurisdiction.
The likely consequences of overstating risk for residential mortgages would be disturbing. As described in detail below:
Currently, residential mortgages are subject to a standardized 50-percent risk weight, a relic of the original Basel I international capital agreement dating back to 1988. A goal of the 2017 Basel agreement was to introduce greater risk sensitivity to the standardized approach and recognize that for many mortgage loans, the risk of loss is significantly lower than a 50-percent risk weight would imply.
The U.S. agencies’ proposed rule sets capital requirements for mortgage loans using a standardized approach that is empirically unsupported and demonstrably punitive to U.S. mortgage borrowers. For credit risk, the agencies propose to use the same risk segments as Basel but would arbitrarily add 20 percentage points to each risk segment. This means the proposal would increase the risk weight for high loan-to-value (LTV), first-lien mortgages (with LTVs over 80 percent) well beyond the 50-percent risk weight applicable under the current standardized approach.
Banks would also encounter another capital charge for operational risk. This would significantly elevate the total risk weight for mortgages sold to the GSEs, as we outline below.
Moreover, banks would face a substantial stress capital buffer charge for all mortgage exposures, given the large fall in home prices assumed in the stress tests and the inclusion of the stress tests’ operational risk losses on top of the existing standard operational risk charge.
Finally, banks operating in the United States are not allowed to use internally modeled approaches for determining risk-weighted assets, which are permitted in other jurisdictions subject to a floor of 72.5 percent of risk-weighted assets under the standardized approach. So, uniquely for U.S. banks, there would be no escape from punitive standardized charges, regardless of whether these are demonstrably excessive in relation to historical losses.
Figure 1 shows the aggregate risk weight for credit risk, under the agencies’ proposal, for first-lien conventional home purchase loans originated in 2022 by banks subject to the proposed rule. (This is based on a data sample described below.) Figure 1 compares that with (1) the aggregate risk weight under the 2017 BCBS standardized approach; and (2) the average aggregate risk weight for first-lien residential mortgages under the U.S. Advanced Approaches, based on the quarterly FFIEC 101 reports of U.S. Advanced Approaches banks since 2014. The Advanced Approaches calculation is empirically based: banks use their own empirically estimated segment-level risk parameters (expected default and loss rates under downturn conditions) in calculating a portfolio risk weight, overseen by agency examiners.
These comparisons demonstrate the proposal’s punitive aspect. Under the U.S. agencies’ proposal, aggregate risk weights for first-lien conventional home purchase loans would be 53 percent. This is significantly higher than the 33-percent risk weight under the BCBS standardized approach being adopted in other countries. In other jurisdictions, banks are allowed to use the advanced approaches. They only use the standardized approach to set a floor for the advanced approaches RWA, such that the latter cannot go below 72.5 percent of standardized approach RWA. If the United States were to follow the same rules, the 33-percent risk weight under the standardized approach would be equivalent to a 25-percent floor (i.e., 33% × 72.5%) for the advanced approaches. From 2014 to 2022, the average risk weight for U.S. banks using the advanced approaches was about 25 percent, according to the FFIEC 101 reports. [2]
The study recently released by the Urban Institute yields similar results. [3] Those researchers used a different data set: losses on loans originated by the GSEs from 2005 to 2008 (those associated with the mortgage and foreclosure crisis) by credit score and LTV range, to extrapolate the losses that would occur in a similar stress environment, given the data they have on the current credit score and LTV distribution of bank portfolios. They conclude that the proposed risk weights are extremely excessive relative to the capital a bank should have to hold against these losses. Applying the same extrapolative approach to our sample of home purchase loans originated in 2022 by banks affected by the proposed rule, we calculate a portfolio loss rate of 2.9 percent. [4] This is close to the 2.7-percent rate that the Urban Institute researchers infer using their data, confirming that the proposed risk weights are excessive.
Moreover, this historical benchmarking itself is conservative, understating the degree to which the proposed loss rates are excessive. As pointed out by the Urban Institute researchers, the historical loss rates used as benchmarks “overestimate the potential loss rate for the current bank book of business.” Mortgage lending “has become more prudent in ways not directly reflected in FICO scores and LTV ratios.” [5]
Furthermore, the largest banks are subject to capital add-ons as needed based on the outcome of stress tests, which typically assumes a more than 35-percent decline nationwide in home prices.
In other words, the agencies have proposed risk weights for U.S. mortgages more than double those that would apply under the advanced approach being adopted by other Basel members. These risk weights far exceed what can be justified based on experience in the U.S. mortgage market. This is particularly true given the extra layer of protection afforded by stress testing, and they have offered no data or analysis to explain why.
Our analysis finds that banks subject to the proposed rule originate a large percentage of high-LTV mortgages. Low- or moderate-income (LMI) or minority borrowers are disproportionately the people who rely on those mortgages. These households tend to be more wealth- and credit-constrained and less able to afford the increase in mortgage costs from increased capital requirements.
Under the agencies’ proposal, 38 percent of the loans to LMI borrowers would receive a risk weight of 70 percent or greater, compared with 17 percent of the loans to non-LMI borrowers. In addition, more than half (52 percent) of loans to Black borrowers would receive a risk weight of 70 percent or more, compared with just 22 percent among white borrowers.
The analysis also finds that mortgage loans to LMI borrowers are a significant share of those originated by large banks for sale to the GSEs. Because of the agencies’ operational risk charge, the increase in risk weights under the standardized approach is most pronounced for these loans, as described below.
In summary, our analysis of HMDA data validates concerns that the relatively high capital charges in the proposed U.S. implementation of Basel III will make affordable mortgages less available to households already facing significant barriers to homeownership, even while evidence strongly suggests that the charges are excessive given the risk they present.
The U.S. banking agencies’ proposed rule applies to institutions with at least $100 billion in total assets and assigns risk weights to mortgages held on the balance sheet based on their LTV as a proxy for their credit risk, which translates into a risk-weighted capital charge. The agencies’ proposal also imposes an operational risk capital charge against all mortgage loans originated by the bank. This extra charge would be applicable even if the loans were intended for sale and only temporarily held on a bank’s balance sheet. Finally, bank mortgage loans would continue to accrue capital charges as components of the Federal Reserve’s stress capital buffer derived from its annual stress test. These charges would be added to the charge for credit risk based on risk-weighted assets and the charge for operational risk.
To understand the potential impact of the proposal on banks’ mortgage loan costs, we will examine two scenarios outlined in the following example boxes: (1) a mortgage loan held on a bank’s balance sheet; and (2) a mortgage loan sold to a GSE. For both scenarios, we assume that the mortgage loan has an LTV of between 80 and 90 percent.
Mortgage loans made to first-time homebuyers and to lower-income and minority borrowers and retained by banks often feature higher LTVs in the range assumed by our scenario. Under the U.S. proposal, these would be assigned a 60-percent risk weight. This is higher than both the current U.S. standardized risk weight of 50 percent and the 40-percent risk weight in the Basel agreement. In addition, U.S. banks subject to the proposal would also have to include a substantial stress capital buffer charge, based on the large nationwide home price decline (typically 35 percent) assumed in the stress tests and the stress tests’ large operational risk losses (driven in large part by mortgage litigation losses). Our estimate is that the effective risk weight for such a buffer charge would be about 20 percentage points, raising the total risk weight to approximately 80 percent. Finally, the U.S. proposal includes an operational risk charge that would add 5 percentage points to the risk weight, on average. [6] Consequently, the total risk weight for a mortgage with an LTV in the greater than 80- to 90-percent LTV range would be 85 percent.
Again, the risk weight under the 2017 BCBS standardized approach is 40 percent. The additional capital charge under the U.S. proposed implementation would likely be passed on to U.S. consumers through higher mortgage rates—whether by banks charging significantly more, or by nonbanks charging significantly more if banks leave this market. This would make it even more challenging for first-time homebuyers and people from disadvantaged communities to achieve homeownership. And it would do so based on no analysis, and contrary to considerable evidence that those challenges are unwarranted by an objective assessment of risk.
The impact of the U.S.-proposed rule would be even greater for mortgages sold to the GSEs, which include most loans targeted to first-time homebuyers. The proposal includes an uncapped operational risk capital charge against fee income. Typically, banks sell newly underwritten mortgages to GSEs within 30 to 60 days of funding and receive a fee known as mortgage banking income. These mortgage sales to the GSEs would draw high operational risk capital charges under the proposal.
Those charges would include not only a charge for the interest income generated by the mortgages but also from the fees earned by selling them to a GSE. Because the fee income component is based on an annual average, each new batch of loans sold to GSEs each quarter generates incrementally higher operational risk capital charges. The operational risk line in the example boxes shows the risk weight of mortgage loans sold to the GSEs increasing from 70 percent under the current standardized approach to 140 percent. [7] In addition, the stress tests include their own operational risk capital charge. Banks would be charged twice on the revenue side for selling a mortgage, both through the stress capital buffer and the operational risk charge. Our best estimate is that this charge would double the total effective risk weight of loans sold to the GSEs.
Finally, if a bank services the GSE mortgage, another operational risk charge applies to the mortgage servicing income. This acts as another charge based on fee income and comes on top of an already high starting risk weight of 250 percent for mortgage servicing rights. It is more challenging to estimate the operational risk charge for servicing income, because the regulatory reports only include servicing income net of expenses and the proposal setting the operational risk is calculated using gross revenues. Therefore, this note doesn’t try to gauge the impact of the proposed rule on servicing, but we anticipate that it would significantly affect that business.
Most banks will likely allocate part of the operational risk charge from loan sales and servicing to retained loans as well. In practice, this means the higher risk weight applied to loan sales will also affect the business line that retains mortgage loans.
Clearly, higher capital requirements for loans both retained on the balance sheet and sold to the GSEs will lead to higher interest rates on bank-originated mortgages and a reduction in bank mortgage originations. [8] This will further shift market share to nonbanks. Quantifying these effects is complex, and outside the scope of our discussion, as the effects depend on multiple factors. One such factor is the cost to a bank of raising the additional required capital. Others include the extent to which the required increase in capital requirements affects types of loans and borrowers that cannot be as well served by nonbanks, and the degree to which banks are currently able to offer lower rates or fees to their borrowers compared to nonbanks.
For our analysis of the potential effects of the proposed increases in required capital for the credit risk of balance sheet mortgages by borrower income and demographic segment, we rely on recently released 2022 HMDA data. HMDA is an annual regulatory data submission that is mandatory for mortgage lending institutions exceeding specified minimum volume requirements.
Our data sample is restricted to conventional first-lien mortgages financing home purchases originated by banking institutions that would be subject to the new rules—those with more than $100 billion in assets. [9] We include both mortgages that banks originate and hold on their balance sheets and those originated and then sold or securitized. Although only the former are subject to capital requirements for credit risk, the proposed new requirement for operational risk could lead banks to expand the share of originations that they retain, so that restricting attention to what banks currently retain does not seem appropriate. Moreover, HMDA data do not provide a fully reliable indicator of which loans are sold; in particular, the data only record loans as sold if the transaction is completed before the end of the year, and even loans originated for sale are subject to credit risk capital requirements while awaiting sale.[10] That said, very similar results are obtained if the sample is restricted to loans identified in HMDA as retained.
Another important distinction, in addition to sold versus retained, is loan size. Mortgages are eligible for sale to the GSEs only if the loan balance is within the conforming loan size limit set by the Federal Housing Finance Agency. In 2022, the applicable limit for most of the United States was $647,200.
Because there is not much of a secondary market for jumbo mortgages (those above the dollar threshold at which the GSEs can purchase), depository institutions are the dominant originators of these loans, and they rarely securitize them. [11] Naturally, most conforming-size loans originated by banks are sold to the GSEs. [12] Nevertheless, a material share of retained loans are within the conforming limit, reflecting the fact that banks sometimes offer better terms to borrowers when originating loans for their own portfolio rather than for sale. [13] The conforming-size loans that banks choose to retain may have unique circumstances affecting the borrower or property, or may be associated with targeted community reinvestment initiatives.
Table 1 summarizes the number and dollar volume of conventional first-lien home purchase loans in our data sample, separately for each of the three categories just described (jumbo, conforming-size and sold, and conforming-size and retained). The banks subject to the proposed increases in capital requirements originated more than 320,000 home-purchase loans totaling nearly $200 billion in 2022.
Table 1: 2022 Home-Purchase Lending by Banks with ≥$100 Billion in Assets
Source: 2022 HMDA data.
About half of the loans originated by these banks were reported as sold, while the remaining half were retained. However, because about half of the retained loans were jumbos and nearly all of the sold loans were of conforming size, the dollar volume of retained loans was roughly twice that of sold loans.
The actual percentage of loans retained among those originated in a given year will be lower than what is reported in the HMDA data. This is because, as noted above, the data only record loans as sold if the transaction is completed before the end of the year. [14]
The proposed rule’s treatment of high-LTV mortgages could be expected to make home purchase loans less affordable for households with more limited incomes and less wealth. In this section, we conduct a set of risk-weight calculations that demonstrate that the burden of higher risk weights for the credit risk of loans on bank balance sheets would fall disproportionately on lower-income and minority borrowers.
We begin by calculating balance-weighted average risk weights by borrower segment in our sample based on the proposed schedule of risk weights for the credit risk of balance sheet loans. This simulates the risk weights that would have been applicable in 2022 had the proposal been in place.
The calculated segment-level risk weights, shown in Figure 2, demonstrate that, while the proposal would increase risk weights across all income and demographic segments of borrowers, groups that, on average, will be less able to afford the resulting increases in borrowing costs will be most affected. For instance, the average risk weight for an LMI borrower is 57.5 percent, whereas for non-LMI borrowers it is 52.6 percent. Black borrowers as a group would be the most severely affected, with an average risk weight of 59.6 percent. As will be made clear shortly, these increased average risk weights reflect increases as large as 20 percentage points for significant segments of individual borrowers.
Average risk weights tell only part of the story. We get a fuller picture by observing the degree to which individual households would be placed into higher risk-weight categories based on LTV at origination. We next calculate by borrower segment the share of loans slotted to the 70-percent-or-greater category (Figure 3).
The differences across borrower segments are quite striking. Figure 3 shows that the share of LMI borrowers receiving a risk weight of 70 percent or greater is more than double the share among non-LMI borrowers (38.1 versus 17.4 percent). More than half of loans to Black borrowers would receive a risk weight of 70 percent or greater, compared with less than a quarter of those to white borrowers.
As the Urban Institute researchers observed, “There is a lot at stake here. Bank portfolios provide a home for loans that do not fit neatly into the credit boxes underwritten by the government-sponsored enterprises, the Federal Housing Administration, or the Veterans Administration. High-LTV mortgages are particularly important for first-time buyers, especially LMI borrowers and borrowers of color. Raising the capital charges on high-LTV loans raises the mortgage interest rates for the remaining borrowers least able to afford the increases.” [15]
The increase in the operational risk capital charge for sold loans would affect a bank’s costs of originating mortgages for sale to the GSEs. A large share of these sold loans can be expected to be modest, financing home purchases by LMI and minority households or in LMI and predominantly minority neighborhoods.
Figure 4 summarizes the household composition for the sold loan category in our analysis sample, confirming these characteristics. As shown in Panel A, most of the borrowers are low- to middle-income (defined as having incomes not exceeding 120 percent of the area median income), and nearly a third are low- to moderate-income. Half of the borrowers are either low- or moderate-income or are purchasing a home in an LMI or predominantly minority U.S. Census tract, while 40 percent are either low- or moderate-income or purchasing a home in an LMI tract. Panel B reveals that median income among borrowers is $100,000, and their median property value is $375,000. A quarter of the borrowers have incomes not exceeding $66,000, and a quarter of the financed home purchases are of properties priced at $245,000 or less.
As a result, the proposed new capital charge on large banks’ originations of loans sold to the GSEs will likely make mortgages less affordable and harder to obtain for low- to middle-income households buying modestly priced homes. In turn, homeownership will become even more out of reach for many American households.
The proposed rule would significantly increase a bank’s cost of originating, holding and servicing mortgages, with adverse effects on all homebuyers but particularly LMI and minority borrowers. Most importantly, the significant increase in cost has no foundation—no supporting data or analysis—in the proposed rule and runs counter to historical experience and the calibration planned in all other countries. Moreover, these increased capital charges are likely to further shift market share of mortgage origination and servicing to nonbanks (compounding the shift that has already occurred due to regulatory burdens imposed on banks by the Dodd-Frank Act). This shift is widely considered problematic, not only because many borrowers may prefer bank-offered products and services and because the market becomes less competitive, but also because nonbank lending tends to be more procyclical (since nonbanks are less able to obtain funding during an economic downturn). [16] Evidence also strongly suggests that nonbanks are less willing or able to offer foreclosure alternatives to borrowers facing financial difficulties. [17]
[2] We have adjusted the advanced approaches risk weight using the 1.06 scalar as required under U.S. regulatory capital rules.
[3] See op. cit. Goodman and J Zhu.
[4] For the purpose of calculating this loss rate, we assume that the dollar share of loans held on balance sheet while awaiting sale to the GSEs is 0.1233, corresponding to an average life of 45 days on the balance sheet, and the effective expected loss on these loans while on the balance sheet is zero.
[6] This charge is driven by the interest, lease and dividend component of the business indicator and assumes an internal loss multiplier of 1.2 (since the United States has not followed other jurisdictions in proposing to set this multiplier to 1).
[7] For a clearer understanding of the standardized approach for operational risk, refer to pages 64082-64089 in the U.S. Basel proposal. Alternatively, see BPI’s previous note on this topic, available here. Furthermore, line item 5 in Schedule HC-P of the FR Y-9C report includes “noninterest income for the quarter for the sale, securitization, and servicing of 1-4 family residential mortgage loans.” It also provides information on loan originations intended for sale to the GSEs. We utilized this data to calibrate the figures presented in our illustrative example.
[8] Goodman and Zhu, op. cit. page 7, offer an illustrative calculation, assuming a direct passthrough of a 12 percent cost of capital, demonstrating that the effect on cost of borrowing could be hundreds of dollars per year on a modest, $200,000 mortgage.
[9] The restriction to the conventional mortgage segment excludes loans that are government-insured—that is, FHA and VA loan products. Nonbank mortgage lenders originate up to 90 percent of government-insured home purchase loans, and a comparatively large share of these loans are made to Black and Hispanic households (see The Urban Institute, Housing Finance Chartbook (urban.org), August 2023, p. 19). We also restrict the sample to loans financing single- or two-family properties; very few owner-occupied homes incorporate more than two housing units. We focus on home purchase because rising interest rates in 2022 were not conducive to refinancing of existing mortgages.
[10] In other words, HMDA data overstate the percentage of retained loans if loans originated in the last quarter that are intended for sale are not actually sold before year’s end.
[11] According to an FDIC study, as of 2017, more than 80 percent of jumbo mortgages were originated by banking institutions. “Trends in Mortgage Origination and Servicing: Non-Banks in the Post-Crisis Period,” FDIC Quarterly, 2019, 13(4): 51–69.
[12] Nonbank firms that specialize in origination and servicing of mortgages dominate the market for mortgages sold to the GSEs, although banks and other depository institutions still originate about 25 percent of conventional home purchase loans sold to the GSEs. See The Urban Institute, Housing Finance Chartbook (urban.org), August 2023, page 12.
[13] For instance, banks sometimes may have other information enabling them to offer borrowers better terms than those implied by secondary market underwriting and pricing standards.
[14] In other words, HMDA data overstate the percentage of retained loans if loans originated in the last quarter that are intended for sale are not actually sold before year’s end.
[15] Goodman and Zhu, page 1.
[16] See Aldasoro, Iñaki, Sebastian Doerr, and Haonan Zhou; “Non-Bank Lending during Crises,” BIS Working Papers No. 1074, 16 February 2023. Using data from the global syndicated loan market, the authors find that during crises, nonbanks cut their syndicated credit significantly more than banks. Also see Fleckenstein, Quirin, Manasa Gopal, German Gallardo, and Sebastian Hillenbrand, “Nonbank Lending and Credit Cyclicality,” NYU Stern School of Business, June 2021. The authors find that nonbank lending is more than twice as cyclical as bank lending. Declines in nonbank lending explain most of the declines in syndicated lending during the Great Recession and COVID-19 crisis. Regarding the role of nonbank mortgage lenders and servicers in particular, see Kim, You Suk, Steven M. Laufer, Richard Stanton, Nancy Wallace, and Karen Pence, “Mapping the boom in nonbank mortgage lending—and understanding the risks | Brookings.” Brookings Institution Commentary, 10 September 2018; Kim, You Suk, Steven M. Laufer, Karen Pence, Richard Stanton, and Nancy Wallace, “Liquidity Crises in the Mortgage Market,” Brookings Papers on Economic Activity, 2018(1): 347–428. The authors argue that the short-term credit they depend on for financing can become more expensive or dry up entirely when financial market conditions tighten. Revenue from mortgage origination activity may drop sharply when interest rates rise, and they can face liquidity shortfalls when defaults rise during housing market downturns. For multiple reasons that will not necessarily recur in future downturns, nonbank mortgage servicers maintained their financial strength during the COVID-19 downturn; see Loewenstein, Lara, “Why Wasn’t there a Nonbank Mortgage Servicer Liquidity Crisis?” Federal Reserve Bank of Cleveland, Economic Commentary 2021-15.
