The Latest GSE Stress Test Results: Showcasing the Need for Regulatory Capital Revision (Part 2)

August 30th 2022 | Donald H. Layton

Aerial view of suburban homes

Background

On August 11, the Federal Housing Finance Agency (FHFA), the regulator and conservator of Freddie Mac and Fannie Mae, the two government-sponsored enterprises (GSEs), released the latest annual regulatory stress test results for the two firms, based upon their yearend 2021 financial statements.1 This is the ninth year of the legally-mandated test, which has at its core the “severe adverse” economic scenario produced by the Federal Reserve. 

In this second of a two-part series, I show how the stress test result—i.e., a $4.5 billion loss for the two GSEs combined—is consistent with a level of required capital for the GSEs that is conservatively calculated to be in the $120-135 billion range.2 This is dramatically lower than the $312 billion of capital currently required, at a minimum, by the Enterprise Regulatory Capital Framework (ERCF3, a regulation established by the FHFA in 2020 with only a few targeted revisions since. As I will describe later, this leads me to conclude the ERCF should be significantly revised downward.    

Notably, the question of how much capital the two GSEs are required to maintain is not just an academic one.  If the regulatory requirement is too low, the safety and soundness4 of the GSEs will be inadequate, reducing the stability of the entire U.S. financial system.   If it is too high, the economic equivalent of an unnecessary tax on homeownership will be created, which is contrary to the long-standing and bipartisan public policy goal of supporting sustainable homeownershi5. It is therefore incumbent upon the FHFA to find the “just right” Goldilocks level of capital requirement, i.e., enough for fully adequate safety and soundness, but not materially more.

This post first recounts the origin and concept of the regulatory stress tests, noting some specific insights learned from that process that are particularly relevant. It then describes the framework for calculating a required capital level based upon stress test results, followed by applying it to the just-announced GSE results to calculate a reasonable and credible range for their required capital. Lastly, it explains why the stress-based capital calculation produces a more accurate and market-relevant result than the much higher ERCF level.

Origin of Regulatory Stress Tests6:  The Intersection of Capital and Confidence

In 2006, when the stresses that would become the global financial crisis (GFC) were just starting to appear on the horizon, the minimum capital required for the very largest banks as determined by their regulators was considered very advanced and credible by the marketplace.  It reflected almost two decades of increasing sophistication ever since the first Basel7 accord in 1988—later designated Basel I—established that bank regulators in the most advanced economies would use the same framework to distinguish capital requirements between assets based on their riskiness.  Previously, U.S. bank regulators mostly used a simple percentage of assets based on their informed judgment. 

By 2006, this range of capital requirements to reflect the riskiness of assets also incorporated statistics-based calculations to show how large a loss might be incurred in a statistically-remote adverse market, according to the available history of actual loss data.8; This trend towards such statistics-based capital calculations was regarded as producing far more accurate results than the previous approach of relying wholly or almost wholly upon informed regulator judgment. Thus, going into the GFC, the largest banks had required minimum capital levels based on a combination of informed regulator judgment and statistical analysis9. I will refer to this as the “Basel approach.”

Then, to borrow a phrase, the Basel approach was mugged by reality during the GFC. The losses on mortgages, central to the crisis, were quickly appearing to be far larger than either the history-based statistical analysis or their credit ratings indicated. As a result, market confidence first faltered for a small number of mortgage-intensive firms because there was just no way for their funders to know whether the capital levels of those firms were adequate to absorb the losses that seemed to be quickly emerging. As the loss of market confidence spread far more widely, it eventually became apparent that the markets10 wanted banks to have sufficient capital levels to cover more than just estimated stress losses.  Instead, it became clear that the market additionally needed what became called a “going-concern buffer,” i.e., enough extra capital for a bank to maintain market confidence even while absorbing large stress losses, lest such a stream of losses trigger a liquidity crisis well before its capital might be exhausted. This is particularly important for banks because they are, by design, rather susceptible to a loss of liquidity when confidence falters; as a result, the going-concern buffer wanted by the marketplace is quite large for banks.11

Of course, this realization of what it took to counter and restore the market confidence then collapsing at the height of the GFC in the second half of 2008 and into 2009 didn’t happen all at once.  Government officials worldwide tried all sorts of things to get that confidence back, usually by injecting capital into specific financial institutions or by guaranteeing certain types of assets, as well as running very loose monetary policy. Then, in early 2009, the U.S. government announced it would be doing stress tests on all its largest banks, adopting a technique used in various forms intermittently and infrequently throughout the banking system until then. These official stress tests were carefully structured to have very high market transparency:  (1) the scenario12 of the stress tests (which would be over nine quarters, i.e., 2-1/4 years) would be set by the Federal Reserve and publicly disclosed to show it was a realistic portrayal of a truly severe stress event; (2) there would be a defined going-concern buffer13  for all the banks to add to the modeled stress losses to create a capital requirement; and (3) the results of the stress tests for each large bank would be publicly disclosed in considerable detail.14 Then, if a large bank’s capital level was judged inadequate to maintain market confidence (as measured by adding together the stress test’s modeled loss plus the going-concern buffer) Treasury committed to make up the deficit by injecting the missing amount into the bank. 

It’s fair to say that this approach was considered a great success. The shaky confidence by the market in large American banks15 subsided relatively quickly after the test results were released, especially as they showed that there was no need for further injections of equity into the largest banks by Treasury.16  As a result, stress tests became a standard feature of American bank regulation, required by legislation not just for large banks but for the two GSEs as well.  

Two Lessons Learned From the 2009 Recovery of Market Confidence

The GFC years created a giant learning opportunity for bank regulators. Two of the lessons learned are particularly relevant to the issue of establishing how much capital is needed by the two GSEs.

The going-concern buffer relates to the susceptibility to a loss of liquidity. As explained above, the development of the need for a going-concern buffer shows that there is a linkage between how much a particular financial institution is susceptible to a loss of liquidity and how large a buffer it needs.  Banks, for which the official stress tests were first developed, are particularly vulnerable to such a liquidity loss for three main reasons. One, banks are structurally illiquid, specifically designed to engage in what regulators call “maturity transformation” by having assets with materially longer maturities than their liabilities. Two, banks rely on specific funding sources (e.g., demand deposits) that can easily “run,” i.e., be quickly withdrawn by customers. And three, banks have large amounts of commitments to their commercial customers (e.g., lines of credit, revolving credit agreements) that allow those customers to draw down loans upon demand—which they will do to protect themselves if they have concerns that their bank may be in trouble. Adding it all up, banks should absolutely be required to have a very substantial going-concern buffer.17 

The stress tests re-established market confidence in the U.S. banking system when the Basel approach to capital requirements did not. I see three main reasons for this. One, the stress tests were transparent to, and easily understood by, the marketplace. By comparison, while the Basel approach may have been transparent to the regulators, it was by contrast a “black box” to the marketplace. This made the market wholly dependent on the regulators’ word that capital was sufficient according to the Basel approach at a time when confidence in regulators was rather challenged. Two, the Basel approach was significantly based on statistical calculations using a history of losses incurred almost entirely during normal times, and it thus proved to be almost irrelevant, as mortgage losses had already ballooned far past what those statistics calculated.18 By comparison, the stress tests directly addressed the then-current situation far more accurately. And three, the core nature of the Basel approach —which calculated lifetime losses on assets using historic statistics, ignoring revenue altogether—just seemed overly theoretical. It simply seemed to be removed from the reality of what the markets needed to know given the high uncertainty that accompanies a major stress event: would a particular bank stay adequately healthy—all things considered—or not? By comparison, the stress test answered that exact question.   In short, the stress tests have both market-friendly transparency and stress-period relevancy, whereas the Basel approach had neither. 

Both of these lessons are important to determining how much capital the GSEs should require, as discussed further below.

Calculating the GSE Stress-Based Capital Requirement

Results of the stress test—in this case, a modeled income loss of $4.5 billion—provide only the first piece of information needed to calculate a capital requirement based upon it. The formula to do so is as follows:

Stress-based capital requirement = modeled stress loss + going-concern buffer + countercyclical buffer


Modeled stress loss

As reported by the FHFA, this was $4.5 billion as of 12/31/21, which I will round up to $5 billion.

Going-concern buffer

I have looked at three different approaches to calculating this buffer.

  1. The risk-weight approach. In 2009, the Federal Reserve determined, quite judgmentally19, that 4% of risk-weighted20 assets (RWA) was the right level of buffer for the banks then under severe stress21. For mortgages in the Basel era, the U.S. banking regulators assigned a risk weight of 50%, which equates to a going-concern buffer of 2% on the (not risk-weighted) amount of mortgage loans outstanding. This approach translates to $146 billion, or 2% of the $7.3 trillion of total GSE assets22.
  1. The bank stress loss comparison approach. By all evidence gathered in the last decade, the reality is that U.S. first-lien mortgages (which is what the GSEs exclusively deal with), seem to have materially less risk intensity than is implied by the 50% risk weighting used for decades by U.S. banking regulators.  For example, going back to the large bank stress test results of 2016, which seem to me to be a year neither cyclically high nor low23 in housing markets, the mortgage loss was just 2.2% versus the 100% risk-weighted commercial and industrial loan loss of 6.4%. This implies a risk-weighting of 34% (i.e., 2.2 divided by 6.4), rather than 50%, which when applied to the $7.3 trillion of GSE assets equals $99 billion.24
  1. The FHFA’s recent estimate. In developing and proposing what later became the ERCF, the FHFA proposed a going-concern buffer of 0.75% of assets during the last years of Democrat-appointed Mel Watt’ directorship. In the long tradition of regulatory capital requirements, the proposal was based upon informed regulator judgment. When the ERCF came out during the directorship of Republican-appointed Mark Calabria, the 0.75% calculation (considered a minimum that the FHFA could potentially raise as part of its supervisory process) was retained25. It equals $55 billion.  

My view is that the 50% risk-weighting approach is simply inapplicable, well out of line with the reality of the risk of U.S. first-lien mortgages (reflecting at least partly the successful reforms in U.S. mortgage markets implemented post-GFC). The other two estimates each have credibility in their own way, one based on loss statistics and the other on informed regulator judgment. I will, therefore, average those two results, i.e., $99 billion and $55 billion, to get $77 billion as a reasonable estimate of what the going-concern buffer should be. While this estimate has a significant band of uncertainty around it, I note that it is 15 times the severe adverse stress loss of $5 billion, which certainly looks more than adequate, making it a decidedly conservative estimate.   

Countercyclical buffer

When calculated at the height of the GFC in 2009, the modeled stress losses for the banks were quite large. As the economy recovered over the next several years, the modeled stress loss declined significantly. Thus, it became evident that a stress-based capital requirement calculation could be materially pro-cyclical, i.e., low in good times and high in bad times. It also was apparent that the regulators could not assume banks would be able to successfully move their capital up and down to match this cyclicality, as their ability to raise additional capital during tough times could be impaired as markets would likely be disrupted. This was the logic behind creating a second capital buffer, i.e., the countercyclical buffer, that would be high in good times and low or zero in bad times. Such an additional buffer would ensure that the extra capital needed in a possible future cyclical downturn was already in hand.   

While countercyclical buffers are conceptually common in regulatory circles, there is no standard and accepted approach to calculating one. Given this absence, I have chosen an approach to estimate what the countercyclical buffer should be. This calculation takes the drop in the modeled GSE stress loss from 2016 (which I consider a neutral year, neither cyclically high nor low) to 2021 (i.e., $100 billion in 2016 minus $5 billion in 2021, or $95 billion), and then judgmentally attributes two-thirds of it, or $63 billion, to the fall in house prices (which increased 56% during this period) back to their 2016 level. (I have attributed the other one-third of the decrease to the risk-reducing business model changes cited in Part 1.) I then adjust down the $63 billion for the general inflation that occurred during the same period, i.e., 16% (per the consumer price index), so the buffer is reduced to $45 billion.26 This, then is my estimate of how much the countercyclical buffer must be so GSE capital can readily absorb a reversal of today’s unduly high house prices back to 2016 levels (on an inflation-adjusted basis). Because it implies that, between the countercyclical buffer and the underlying stress scenario, house prices cumulatively decline by almost 70%27 from year-end 2021’s level (which is 53% on an inflation-adjusted basis), it is a very conservative, i.e., high, estimate for this second buffer.

Total

Given the above calculations, the total stress-based capital requirement for the two GSEs combined would be:

 

Stress-based capital requirement = $5 billion modeled stress loss + $77 billion going-concern buffer + $45 billion countercyclical buffer = $127 billion.

 

This $127 billion is my conservative estimate of the capital requirement for the two GSEs, combined, based upon the stress test approach as applied to the most recent results. Not wanting to overstate its precision, I estimate it falls into the $120 to 135 billion range. On a personal note, I had been expecting the stress-based capital requirement to be higher, in the $150 billion range. However, even while being conservative at every step in the calculation, the result is materially less, showing how much the changes in the business model of the GSEs (as described in Part 1) have lowered the fundamental riskiness of the two companies. 

Why the Large Difference Versus the ERCF? (Hint:  GSEs Are Not Banks)

Conceptually, the GSEs should have the same capital per unit of risk as the very largest banks, which are currently designated as “systemically important financial institutions” (SIFIs).  In fact, anything less looks like an attempt to deliberately undercapitalize the two companies28  However, I would argue that the $120-135 billion range of capital requirement as calculated herein meets the SIFI-equivalency requirement to have the same capital per unit of risk. This leaves the large variance from the ERCF’s much higher calculation of $312 billion to be explained.  First, the large variance stems partly from the floors and discretionary buffers that bias the requirement higher than the Basel approach alone would produce.  And second, it stems from the ERCF treating the GSEs too much as if they were banks, which they absolutely are not. 

Consider two major differences between banks and the GSEs:

  • Much less susceptibility to a loss of liquidity requires a much smaller going-concern buffer.  As I discussed earlier, the going-concern buffer for banks must be very substantial, given their by-design maturity transformation role and reliance on easy-to-run deposits, which makes them particularly prone to a loss of liquidity. By contrast, the GSEs today have little such structural illiquidity and are minimally susceptible to a loss of liquidity: about 90% of their assets are funded by pass-through MBS29, where there is no maturity transformation whatsoever and whose investors have zero ability to “run.” Also, they are not in the business of giving their customers lines of credit that can be suddenly drawn upon in a time of stress. Thus, the going-concern buffer for the GSEs should be much lower than the 4% of risk-weighted assets developed for banks in 2009 (i.e., $146 billion). The $77 billion buffer used herein, lower by about half, is therefore readily justifiable, and may in fact still be too high given the very low GSE susceptibility to a loss of liquidity.30
  • Less risk-intensity per dollar of assets and risk-weighted assets (RWA). For each dollar of assets on a bank’s balance sheet, a bank takes considerable risks, with credit risk, interest rate risk and liquidity risk being the largest and most well-known.31 The GSEs have structurally only a small fraction of the interest rate and liquidity risk found in banks, as about 90% of their assets are financed via pass-through agency MBS, which substantively lay both such risks off onto the MBS’s institutional investors. In terms of credit risk, the GSE exposure to those risks has been rapidly declining as, beginning in 2013, the two firms began to increasingly lay off this risk via transactions known as credit risk transfers (CRTs32) to a broad range of global institutional investors. This means that the GSEs have extraordinarily low retention of risk per dollar of accounting assets (and continually becoming even lower as more CRT is transacted) and RWA. This business model is unique to the GSEs33; it is virtually never found in the banking system.34 Unfortunately, the ERCF ties many of its capital requirements to assets and RWA just as is does for banks, which leads to the overstatement of the GSEs’ need for capital, as GSE assets and RWA have such low-risk intensity as compared to banks.35 

As I indicated earlier, the GSEs are not banks in critical ways, and this explains why it makes sense and is reasonable that my stress loss-based estimation of the amount of capital required for the GSEs, i.e., $120-135 billion, should be well under the $312 billion bank-style requirement established by the ERFC. In addition, the lower figure just makes common sense—after all, does a stress loss calculated at $5 billion really require a level of capital equal to $312 billion, i.e., 62 times the loss? I think not. 

Conclusion and Implications

This brief estimated that a stress-test approach to calculating a minimum regulatory capital requirement would generate in the range of $120-135 billion. In addition, at each step of developing that range for a stress-based capital requirement, I made conservative choices that will possibly overstate it.36 This means there is indeed a very large gap between what the ERCF requires of GSEs—$312 billion in capital—and what a stress test approach generates.    

Moreover, a lesson from the GFC as described above was that the stress test approach to determining a capital requirement proved effective in re-establishing market confidence during the very tough times of 2008 and 2009.  By contrast, the Basel approach proved to be inadequately relevant and insufficiently transparent to market participants to get the job done. Thus, it is my view that the level of capital required by the stress-test approach should be the driver of the general level of regulatory capital, which in turn leads me to conclude that the ERCF needs to be revised downward.

As stated above, the capital requirement level is not just of academic or technical regulatory interest. Setting such a high ERCF-based capital requirement—an estimated $185 billion more than the stress-based requirement— acts as an unnecessary tax upon American homeowners37. I have roughly estimated that the $185 billion difference increases the mortgage rate to the average homeowner whose house is financed via the GSEs by about 0.21%, which for a typical loan of $350,000 translates into a “tax” of $735 per year, or $61 per month.38 This tax would be paid by roughly one of every two homeowners in the U.S., as the GSEs finance about half of all first-lien residential mortgages in the country.   

Because the variance between the ERCF and the stress-based capital calculation developed herein is so large, it seems incumbent upon the FHFA to revisit the ERCF and the stress tests, together, and subsequently make the official changes needed to get them to produce reasonably comparable capital requirements. If this process were to validate my estimate of required capital to be in the $120-135 billion range, I note that this is still a sizeable amount of capital when compared to historic standards. For comparison purposes, it’s worth mentioning that, upon entering the GFC, the two GSEs combined had only $71 billion of stockholder’s equity (as of year-end 2007). This means that the $120–135 billion of capital required developed above is almost double what existed back then, despite the two firms having balance sheets that, while smaller than today’s, had in comparison much riskier assets and with considerably more susceptibility to a loss of liquidity.

Footnotes

[1]; See the FHFA’s Dodd-Frank Act Stress Test Results, August 11, 2022.  https://www.fhfa.gov/AboutUs/Reports/ReportDocuments/Final_2022-Public-Disclosures-FHFA_SA.pdf.

[2] The GSEs, beginning in September 2019, were permitted to build capital by retaining their earnings. As of 12/31/21, i.e., the date of the stress test, they had a combined net worth of $70.9 billion. Thus, the calculated $4.5 billion loss could easily be absorbed. 

[3] The ERCF’s required capital is technically only applicable if and when the two GSEs leave conservatorship. In the meantime, it sets the target for how much capital they need, at a minimum, as the most important—but not the only—requirement to leave conservatorship. During conservatorship, the ERCF, at the discretion of the FHFA, can be employed by the GSEs for certain pricing and other risk-versus-reward decisions. Please note that this figure is as of 3/31/22, rather than 12/31/21, because of certain disclosure issues; this will have no meaningful impact on any of the analysis or recommendations in this brief.

[4] “Safety and soundness” is a financial institution regulatory term. Colloquially, it means that should a financial institution be subject to some stress, it would be highly resistant to collapsing from it. Given the large size of the GSEs, any instability on their part will impact the entire U.S. financial system.

[5] Homeownership is regarded as the method by which the typical family builds wealth for retirement or to help the next generation, as well as supporting the family stability that comes from not being exposed to undue rent increases or the elevated likelihood of needing to relocate that accompanies being a renter.  

[6] Much of this section is based upon recollections from my own career in banking and finance, which began in 1975, during which this was all developed, as many of the specific jobs I held were heavily involved in the development and implementation of regulatory and internal capital requirement systems. The discussion is also very simplified to suit the purpose of this article. The description of the development of the stress tests during the toughest period of the financial crisis is significantly based on “Stress Test: Reflection on Financial Crises,” by Timothy F. Geithner (2014), who was Secretary of the Treasury at the time; it is a great insider account of how policymaking happened amidst all the uncertainties, politics and media frenzy at the time.

[7] Basel, a city in Switzerland, is the location of the Bank for International Settlements (BIS), which is more colloquially known as the “central bank for central banks.” It is under the auspices of the BIS that such multi-country bank regulation issues are addressed. 

[8] This statistical approach was usually based on a technique most commonly known as value-at-risk (VAR). 

[9] Credit ratings were also incorporated into the calculated required capital. This is a proxy for using statistical analysis, which significantly underlies credit ratings. 

[10] In this case, “markets” refers to all the collective actions of the individuals and organizations that provide funds to banks and other large financial institutions in the form of deposits, loans, bonds, commercial paper, etc. This term also includes others who do business with those firms, such as derivatives counterparties. 

[11] In current regulatory parlance, the going-concern buffer is now called a “stress capital buffer.” 

[12] There were originally two, and later three, scenarios produced by the Federal Reserve each year, with the “severe adverse” scenario being the most extreme. Over the years, the results of scenarios other than the severe adverse one had little impact, and thus mostly dropped from public discussion. 

[13] It was set equal to 4% of risk-weighted assets, as discussed further below.

[14] This was considered highly unusual at the time. In my recollection, it was necessary because market confidence was so low then, including in the regulators.  So, instead of expecting the market to take a regulator's word about whether a specific bank passed the test or not, showing the bank's underlying numbers was considered necessary to help the market regain adequate confidence.

[15] As a reminder, the largest securities houses became banks in the crisis, which meant they could get the financial support legally available only to banks. Thus, the stress tests were applied to them as well.

[16] Interestingly, European bank regulators tried to do the same thing afterwards. It was not received well by the market because the scenario used to calculate stress losses was regarded as insufficiently severe. As a result, it took considerably longer for large European banks to regain full market confidence. 

[17] Banks need to keep significant liquid assets to help temper this high susceptibility to a loss of liquidity. The Basel multi-country process had, however, focused exclusively on capital, a glaring defect made very obvious by the GFC. Today, the regulatory specification of minimum liquidity requirements is much more sophisticated and much higher than in pre-GFC times. 

[18] This should have been expected, as two targeted stress events a decade earlier [(1) the 1997 Asia debt crisis and (2) the related 1998 near-simultaneous collapse of Long Term Capital Management, a sizeable hedge fund, and the default by Russia on its international debt] had already shown how stress-market losses were far greater than what was predicted by the Basel approach’s statistical calculations. It is unclear why this lesson was only learned later by regulators during the GFC rather than at this time.

[19] The Geithner “Stress Test” book describes how judgmental this was. 

[20] Risk-weights are a concept developed for the original Basel I capital regime. The typical bank loan, known as a commercial and industrial loan, was given a 100% risk weight. Mortgages—reflecting all types of mortgages across all the countries involved in developing Basel I—were given a risk-weight of 50% by U.S. regulators. 

[21] The 4% requirement was developed for the first bank stress tests in 2009. By 2021, bank regulators had made it all very complicated by comparison to that straightforward calculation, including distinguishing between large and very large banks, along with the nature of their business, and heavily using judgement-determined buffers. To avoid overly complicating the issue, I am using the original and uncomplicated 4% calculation throughout. 

[22] The GSEs have some additional off-balance sheet risks that this calculation would not capture, but they also have low-risk liquid assets where the 2% figure is too high. In this brief, I have assumed these roughly net out in the interest of simplicity. 

[23] As background for my selection of 2016 as such a “normal” year, nominal GDP and Personal Consumption Expenditures (PCE) grow at a rate of between 4% and 5% over the long-term (respectively, 4.53% and 4.79% in the 25 years ending Q4 of 2016. Thus, house prices growing in that range are in rough equilibrium, not causing housing to require a greater or lesser share of the typical household budget or GDP. I arrived at Freddie Mac in 2012, the year after house prices bottomed out.  Prices proceeded to grow in excess of that range, which was expected given how much they had to recover after falling by nearly one quarter as the bubble burst. By 2016, I recall thinking that house prices had recovered fully, as the level of house prices had recently surpassed the bubble’s peak back in 2006/2007 (but of course, it was a full decade later, so in inflation-adjusted terms, they were still modestly lower). After that, house prices grew faster than the 4% to 5% range: 6.21%, 5.64%, and 5.45%, respectively, in 2017 through 2019, even before the pandemic supercharged the increases.  I note that the shortfall of house production during this whole period is now recognized as a source of undue house price increases. [All calculations based on the FHFA’s index of U.S. house prices on purchase mortgages, seasonally adjusted, from Q4 of each year cited.]  https://www.fhfa.gov/DataTools/Downloads/Pages/House-Price-Index-Datasets.aspx#mpo]

[24] The latest stress test results, reflecting that we are now at the height of the mortgage cycle given how high house prices are today, would produce an even lower risk weighting. Specifically, as of 12/31/21, the typical commercial and industrial loan had an 8% projected loss, while first-lien U.S. home mortgages had only 1.3%, indicating mortgages deserved a 16% risk weighting (i.e., 1.3 divided by 8). It would be inappropriate to use that figure.

[25] Additional discretionary buffers approved by former FHFA director Mark Calabria, which are not required by the usual stress test approach to capital calculation as determined back in early 2009, muddied the picture. I have, therefore, not considered them in estimating what the GSE going-concern buffer should be. 

[26] The calculation to adjust the buffer for inflation is:  $63 billion x (1 - .16/.56), which equals $45 billion.

[27] House prices went up 56% from 2006 to 2021, and the stress scenario assumes a further 29% decline. The cumulative percentage impact is to reduce prices to 31% [(1 - .56) x (1 - .29)] of their starting level, or a decline of 69%. The stress-based capital calculation, therefore, requires capital first for a reduction of house prices back to their 2016 level, and then for a further 29% decline. (All data from the FHFA index of house prices.) This is then adjusted for general inflation of 16% during the same time period. I note that an alternative to using the CPI as a measure of inflation (which was 3.01% per annum over the five years) would be to use a general long-term increase in house prices in the range of 4% to 5%; that would reduce the needed countercyclical buffer below $45 billion. 

[28] Prior to the GFC, there was a long history of mortgage-related financial institutions being undercapitalized to help reduce the cost of mortgage credit.  This significantly contributed to the failure of thousands of thrift institutions in the 1980s as well as the GSEs themselves being placed into conservatorship in 2008. 

[29] The GSEs fund their ownership of mortgage loans by issuing “pass-through mortgage-backed securities” (i.e., pass-through MBS). They are called “pass-through” because the investors in the MBS only receive interest and principal payments, with rare exceptions, as the borrowers of the mortgages packaged into each issuance of MBS make those payments into the GSEs. The GSEs, in turn, guarantee the MBS investors against any loss from the borrowers defaulting on their loans. 

[30] I note that this is true today but it was not in 2007 and 2008. The GSEs, at that time, borrowed from short and medium-term unsecured debt markets to fund their substantial largely-discretionary investment portfolios that had, on average, longer maturities, thus exposing them to a bank-style loss of liquidity. Specifically, as of December 31, 2007, they had a combined outstanding $1.5 trillion of unsecured debt representing roughly one-third of their combined balance sheet, which became hard to fund as the stresses of the GFC grew. Today, by contrast, there is less than $.4 trillion outstanding, representing just 5% of their assets as various post-2008 government requirements forced a massive reduction in the size of their investment portfolios, in turn similarly reducing the unsecured debt issued to fund them.

[31] Credit risk: the risk of non-payment by the borrower; interest rate risk: how much the cost of funding an asset will squeeze net interest income as interest rates change up or down; and liquidity risk: the risk that obtaining the funds to carry an asset will become unduly expensive or difficult to find. 

[32] CRT transactions do not usually remove mortgage assets from the GSEs’ balance sheets.  This distinguishes them from various attempts by banks to construct “off-balance sheet” transactions to reduce their accounting assets. In other words, CRT is about reducing real risks, not accounting assets. 

[33] The mortgage insurers who are adjuncts to the GSEs businesses by insuring the credit risk of amounts over 80% LTV on GSE-funded mortgage loans, have partially copied this business model. 

[34 Increasingly, GSE assets are mainly accounting entries with large (and growing) percentages of their associated risks passed-through to institutional investors and not requiring the GSEs’ own capital to support.

[35] There is definitely an attempt in the ERCF to adjust down how assets and RWA translate into a capital requirement as compared to what bank regulators use, but it is in my view nowhere near enough to reflect the actual and declining risk intensity of GSE assets and RWA. 

[36] For example, I assumed a full DTA write-down when that is rather unlikely. I used a $77 billion going-concern buffer when I easily could have used $55 billion by applying the official FHFA calculation; and the countercyclical buffer at $45 billion could, based on a 53% after-inflation reduction in house prices, be reduced by assuming a natural growth rate of 4% to 5% per year in house prices for inflation adjusting, instead of using the smaller CPI.   

[37] The “tax” works through how the GSEs charge, for their securitization and guarantee function, guarantee fees (or G-fees). The largest cost that goes into calculating the G-fee is a market return on the capital required to support the risk, meaning that unduly high capital requirements will produce unduly high G-fees. Current average G-fees are about 0.50% per annum and apply to about half of all first-lien mortgages in the U.S. 

[38] I have assumed $185 billion of unnecessary extra capital in the ERCF and that it requires an 11% pre-tax return, a figure used by the FHFA. Currently, the 10-year Treasury rate is approximately 3%, and the combined guarantee books for the GSEs are $7.1B per the 12/31/21 10-Ks. This translates into a required increase in G-fees of about 0.21% per annum. 

Donald H. Layton

Donald H. Layton is a Senior Visiting Fellow from Practice. Prior to joining the NYU Furman Center, he served as a Senior Industry Fellow at Harvard’s Joint Center for Housing Studies, where he wrote extensively about the Government Sponsored Enterprises (GSE) of Freddie Mac and Fannie Mae and more broadly on housing finance. Before his stint in academia, Layton was the CEO of Freddie Mac from May 2012 until June 2019, where he championed the development of Credit Risk Transfers, one of the most significant reforms to the housing finance system in decades.

« Previous | The Stoop | Next »