The Latest GSE Stress Test Results: Confirming a Major De-risking Success (Part 1)

Introduction

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 stress test results for the two firms. This is the ninth year for the GSEs to be subject to the legally-mandated test, which reflects the “severe adverse” economic scenario produced by the Federal Reserve¹being applied to the companies’ year-end 2021 financial statements.

This article, the first of a two-part series, will address how the latest stress test results confirm that the GSEs, combined, have undergone a surprisingly large de-risking during their years in conservatorship. More specifically, the stress tests from 2013, the first year they were administered, showed almost $200 billion of loss for the companies, a very elevated level.  These losses reflected in particular that there were still many assets on the GSEs’ books that had been purchased before the mortgage bubble burst, kicking off the global financial crisis (GFC). Since then, the modeled stress losses steadily declined to where they are now down to just $4.5 billion², a drop of 98% and also an unusually small 0.06% of the $7.3 trillion of GSE assets³.  Even the average modeled loss of the last three years, $14.9 billion, is down by a rather large 92% since 2013.   This very significantly reduced stress loss, in turn, confirms (1) a major improvement in the stability of the country’s entire financial system, and (2) a large reduction in taxpayer exposure to GSE risks¹, both of which are important policy successes. 

The government developed stress tests mainly during the first half of 2009, at the height of the GFC then underway,⁵  when market confidence in large U.S. financial institutions, especially banks, was quite shaky.  At the time, the government faced the possible need to put more capital into some very large banks to keep them from failing, which might even have turned into de facto nationalization⁶ of one or more of them.  A core reason for this lack of confidence by market participants was that they could not determine to their satisfaction whether the largest financial firms had sufficient capital to absorb the full extent of the losses the GFC was causing them to generate, much less if the weak economy at the time turned further down to produce even more losses.  Designed to be a highly transparent, credible estimation of any such capital sufficiency shortfall, stress tests came with a promise by Treasury that it would plug any such shortfall (plus, as explained in Part 2, a “going concern buffer”) if private market capital failed to quickly do so. In other words, one way or another, the capital shortfall ―as calculated by the stress test― would be eliminated so that depositors and other liability holders would become confident that they were not at any real risk.  

Those first stress tests turned out to be a great success, as market confidence quickly improved without any additional Treasury funds being injected into the large banks.  Naturally, this series of events cemented stress tests as a valuable tool to determine the capital requirements that a large bank, or indeed any large financial institution, would need to keep market confidence, especially during difficult times.

The De-risking and Its Sources

The 2022 stress test (based upon year-end 2021 financials) adds just one more year of evidence demonstrating how the GSEs in conservatorship have undergone an incredibly successful reduction of their risk.  The table below illustrates this reduction beginning  with 2013, the first year of combined GSE results.  I can confidently say that the level of de-risking in subsequent years went well beyond what the FHFA, Treasury, or other policymakers ever expected when the first test results appeared in the summer of 2014.

History of GSE Stress Test Results 2013 - 2021

Stress Test7 for Year Ending 12/318	Combined GSE Modeled Comprehensive Income (Loss)9
	Without DTA Write-down10	With DTA Write-down
2013	($90.3 billion)	($195.8 billion)
2014	($73.4 billion)	($162.1 billion)
2015	($51.3 billion)	($127.6 billion)
2016	($35.1 billion)	($99.5 billion)
2017	($41.9 billion)	($77.4 billion)
2018	($18.0 billion)	($43.3 billion)
2019	($7.1 billion)	($29.1 billion)
2020	$10.8 billion	($11.0 billion)
2021	$ 15.6 billion	($4.5 billion)
Percent Loss Reduction (2013 to 2021)	100%+	98%

The obvious next questions are, what caused such an large risk reduction and is it durable over time?  My view is that there are five critical factors in the decline, only the last of which is not a permanent source of reduction.

1. Working through elevated credit losses on an accelerated basis. Going into 2013, the GSEs still had  many already impaired (e.g. in default) or weak (i.e. with an elevated likelihood of default) mortgages on their books that had been made before the mortgage bubble burst. A stress test applied to those already-troubled mortgages produced further outsized losses, as such assets are especially prone to show high additional modeled stress losses. The amount of such impaired or weak mortgages outstanding then declined steeply during the next few years, reflecting the traditional process of taking defaulted mortgages through foreclosure. However, the GSEs accelerated the decline in their exposure to those assets by (1) developing robust programs to modify loans, and also expanding the conditions under which such loans could be refinanced at much lower interest rates, both of which reduce risk and thus the test’s modeled stress losses; and (2) developing the ability to sell, starting in 2014-2015, packages of non-performing, modified or other types of impaired loans to investors¹¹.  Together, the result that followed the mortgage bubble bursting was an accelerated return to a more normal level of potential stress losses. 
2. Banning non-QM products and generally better credit policy. A result of the GFC, as embodied in the Dodd-Frank Act¹², was that the government defined certain mortgage product characteristics as being problematic. Evidence gathered after the bubble burst showed that these mortgage structures were too prone to abuse and thus inherently likely to lead to elevated credit losses. Among these products were low-doc/no-doc mortgages¹³, loans with teaser rates, and others. All such types of structures became known as “non-QM” (qualified mortgage) products.  On orders from the FHFA, the GSEs then banned these products and have never purchased them since.¹⁴  This action was complemented by the GSEs generally tightening up their credit risk appetites, which returned approximately back to a pre-bubble level, circa 2000 to 2002. As a result, new loans purchased by the GSEs for the last decade have been, on average, considerably better quality than those made previously during the bubble years.  So, the level of exposure to stress losses by each GSE steadily declined as old loans rolled off and new ones came on.
3. Developing and implementing credit risk transfer. Before 2013, the GSE business model consisted of three broad steps: (1) purchase mortgages from primary¹⁵ market lenders, (2) package and finance those mortgages by issuing mortgage-backed securities (MBS), and (3) guarantee to the MBS investors the credit risk of the underlying loans. The result was that the GSEs had little residual interest rate or liquidity risks on the purchased mortgages, which were instead passed through to the investors in their MBS¹⁶. By contrast, their credit guarantees to investors resulted in the two companies retaining trillions of dollars of exposure to mortgage credit risk. (This concentrated mortgage credit exposure was the core reason the GSEs lost market confidence in 2008, leading to conservatorship by the government.)  Most of the losses generated by the GSEs after the bubble burst came from this large and concentrated risk exposure. Then, in 2013, the GSEs began to do credit risk transfer (CRT) transactions by which institutional investors, through different types of financial transactions, absorbed certain credit losses that previously would have fallen onto the GSEs. The core business model of the GSEs was thus changed, becoming materially less intensive of credit risk. As new mortgages, which have linked CRT transactions, replaced old mortgages which had no such CRT, the fundamental riskiness of the business has thus steadily declined. While public disclosure of how much risk has been transferred from the GSEs to investors by such transactions is limited, I roughly calculate that between one-fourth and one-third of the credit losses projected by the stress tests were transferred via CRT¹⁷.  Taking CRT to its reasonable maximum usage over the next several years, I estimate that between one-half and three-quarters of all modeled stress test credit losses could be eliminated in this fashion, permanently lowering such losses in the annual stress tests¹⁸. 
4. The mandated investment portfolio reduction. Going into the last mortgage bubble years of  2006- 2007, the two GSEs had immense investment portfolios that topped out, combined, at over $1.5 trillion¹⁹ This reflected the two companies exploiting an unintended consequence loophole in their design through which they could buy, on a discretionary basis, an unlimited amount of mortgage and mortgage-related securities and then finance them with debt that paid only near-Treasury rates.  That inordinately low financing cost reflected the implied guarantee of the two companies by the government.²⁰  While some of the investment securities were high quality, others created significant losses during the GFC. As a result, the PSPA, instituted when the two companies entered conservatorship, required the GSEs to reduce these portfolios to much lower levels over the following decade (this extended timeframe allowing the reduction to occur in an orderly manner), a task completed by 2018. Since then, the FHFA has directed the two companies to reduce the size of the portfolios even further; they are now down by over 80% from the peak. Thus, modeled stress test losses generated by the investment portfolios have been very greatly reduced as their outstanding balance has so dramatically declined.²¹
5. The good fortune of rising house prices. GSE credit losses are very much tied to the level of house prices, for with lots of equity in a home (i.e., its current market value minus its mortgage amount outstanding), even a homeowner with a compromised ability to make the required monthly mortgage payment does not pose a great deal of risk of loss to a mortgage lender. As a result, when house prices rise the stress tests will show reduced losses, and vice versa when they decline.  In fact, house prices did not bottom out after the GFC until 2011²² and then recovered at a modest pace. To the degree that house prices were, in those first years after 2011, just returning from a depressed level, the resulting shrinkage of the GSE exposure to stress losses was natural and reasonable. In 2016, the general level of house price increases was harder to view as “recovering” and seemed to start rising above normal levels of annual appreciation²³  Then, in the pandemic years of 2020 and 2021, the annual increase in prices exploded to levels never seen before. (By the end of 2021, national house prices on average had increased by an astounding 36% over their pre-pandemic level.²⁴) The annual stress test implications of such rising housing prices have been very positive, translating into much-reduced losses even under the Federal Reserve’s severe adverse scenario.²⁵; This reduced exposure is very real, but it is not permanent, as over time the average loan-to-value (LTV) ratios in the portfolios of the two GSEs - now unusually low due to the recent rapid increase in house prices - should settle back to a more normal level.

To summarize, the reduction of the GSEs’ stress losses from elevated post-financial crisis levels to their unusually low levels today has many causes. But only one of them―rapidly rising house prices producing an unusually low estimated average LTV²⁶ ― is transitory. The rest of the improvement has a permanent nature, reflecting the very significant changes in the underlying business model of the GSEs while they have been in conservatorship.²⁷

Implications and Conclusion

There are two major policy implications of this dramatic reduction of the modeled stress losses under the severe adverse scenario from the Federal Reserve.

The stability of the financial system is much improved.  The events of 2008 revealed that the design of the GSEs was almost exactly what is bad for financial stability, i.e. extremely large institutions that not only took on immense amounts of credit risk but were, by law, required to be “monolines” (i.e., the opposite of diversified) so that all their credit risk was in just the one asset class of residential mortgage²⁸.  Thus, when that asset class had its severe problems in the GFC, the GSEs became especially destabilized - and given their extreme size had the knock-on impact of making the entire financial system more unstable. Making CRT a core part of their business model so those mortgage credit risks are instead increasingly being put into the hands of a large number of diversified institutional investors, along with the other overall de-risking business model changes discussed above, has thereby substantially improved the stability of the country’s financial system.

Taxpayers’ exposure to the risks of the GSEs is now much reduced. Taxpayers have always stood behind the GSEs, and that backing continues to be necessary for their business model to work. Before 2008, this was in the form of an “implied guarantee”; since then, a legal agreement (the PSPA) fulfills that role.  In both cases, the taxpayer is exposed to the risks of the two companies, as the events of the GFC amply demonstrated, when taxpayers injected over $190 billion into the two companies as the government made good on the implied guarantee.  But with the dramatically lowered modeled stress loss, this exposure is smaller than it has been in decades.  Given the existing level of capital at the two companies,  the probability of taxpayers having to inject more funds into the GSEs is approaching levels that I believe are so small they cannot be statistically measured. Removing this burden from taxpayers is significant and will free up room in the federal budget²⁹, allowing spending on other priorities. 

In conclusion, stress testing as a core part of the regulation of banks and other financial institutions came about in response to the GFC. In my view, this was a significant advancement in regulatory sophistication, as stress testing proved its value during a period of great financial instability by providing transparency and clarity to the markets on the risks of an individual financial firm versus its level of capital.  As described above, the official government stress test results reveal an incredible de-risking of Freddie Mac and Fannie Mae.  My analysis concludes that, except for today’s cyclically high house prices, all the factors driving that de-risking are a permanent part of their new post-2008 business model. This confirms how that business model has much less potential systemic risk and also much less taxpayer exposure to the risks of the two companies.

It’s a great success story. 

Part 2 of this series will examine how the stress test results are incompatible with the formal regulatory requirement for GSE capital, leading to the conclusion that the regulatory requirement is far too high and needs to be significantly revised lower.

Note: understanding the impact of DTA write-downs in stress testing

The two GSEs have substantial assets on their books categorized as deferred tax asset (DTA), a complex result of how the financial statements of the two companies are maintained (as required by regulations) on a generally accepted accounting principles basis (known as GAAP) versus how the IRS taxes companies. (As of yearend 2021, the combined DTA of the two companies was $19 billion.)

In the case of the GSEs, the most important source of revenue consists of their guarantee fees (G-fees). A significant portion of those G-fees, which are usually talked about as being wholly on a per annum basis (e.g., 50 basis points per annum), are actually received upfront (known as delivery fees). The DTA arises primarily from this fact.³⁰ Specifically, the IRS requires that companies calculate most of their revenues and profits on a cash basis, meaning when cash is received and paid out. So, to the IRS, the delivery fee creates a big upfront profit on the typical mortgage purchased by the GSEs, with only more modest per annum revenue and profits extending during the life of the associated mortgage. Taxes are paid on this basis, i.e., with a hefty upfront payment.

However, public companies like the GSEs must maintain their books and report to the public and their regulators using GAAP accounting. Under GAAP, delivery fees must be amortized over the life of the associated mortgage, as the obligation of the GSEs to make good on their credit guarantees to MBS investors continues for that entire period. This creates what is called a “timing difference”, as the taxes generated by delivery fee revenues and profits have been paid to the IRS up front versus the amortized, per annum recognition of those same delivery fees (and the profits they generate, triggering a GAAP-basis tax expense) annually on the GAAP books of the GSEs. As a result, on a GAAP basis, the taxes paid up front are considered a pre-payment of taxes, creating a “deferred tax asset” (i.e., the amount prepaid in cash to the IRS before it is slowly amortized into GAAP-basis tax expenses over future years).

It is worth noting that if the prospects for profits of the companies in the future severely deteriorate, due to prospective loan losses for example, then GAAP requires that the existing DTA on the balance sheet be “written down”³¹, i.e. taken as an immediate charge against GAAP net income as there is unlikely to be future GAAP-basis profits to generate tax expense to absorb the continuing amortization of the DTA. This is exactly what happened in 2008, resulting in a considerable loss to the GSEs on their publicly reported financial statements.

In the stress test “severe adverse” scenario, the deterioration of profits may or may not be large enough to trigger such a DTA write-down. (That determination is significantly judgmental, as it requires estimates of future profits.) If such a write-down was determined to be required, then the written down DTA would generate a big upfront loss in the stress test. The FHFA, instead of making a judgment about whether such a DTA write-down would or would not be required, has therefore always reported the stress test results both on a “without” and a “with” DTA write-down basis.

In my view, examining the history of the stress test losses from 2013 to 2021, it is evident that a DTA write-down would have been highly likely in 2013, and would have been highly unlikely in 2021. Nevertheless, in the interest of conservatism and simplicity, I have exclusively utilized the “with DTA write-down” figures in this article.

----------

Footnotes

[1] The FHFA supplements the Federal Reserve scenario with several consistent mortgage-specific inputs, which is necessary as the two GSEs are so concentrated in such assets. 

[2][2] In this post, I always use the stress loss reported “with DTA write-down” as the larger of the two versions reported in an effort to be conservative.  See below for an explanation about DTA, i.e., deferred tax asset.

[3] All figures in this article are as of December 31, 2021, unless otherwise specified.

[4] The government has always supported the creditworthiness of the GSEs, without which their fundamental business model would not work.  Prior to conservatorship commencing in September 2008, this was done via an “implied guarantee” by which the government, through statements and other actions, signaled to the marketplace that it would not let the two GSEs fail.  Upon being placed in conservatorship, this support was upgraded by implementing a formal legal agreement, called the Preferred Stock Purchase Agreement (PSPA), between each GSE and the U.S. Treasury.   

[5] See “Stress Test:  Reflections on Financial Crises” by Timothy R. Geithner, who was Secretary of the Treasury at the time of their development, for more background on this history.

[6] In this instance, such de facto nationalization meant that the government, by injecting funds into a bank in order for it to avoid failing, as a by-product would end up with majority ownership of the shares of the bank, and thus gain operating control of the institution.  This is different than de jure nationalization where the government, through legislation, simply takes ownership of a company away from its private sector owners.

[7] The comparability of the modeled comprehensive income (loss) for each year would be enhanced if the adverse scenario never changed during the near-decade of the tests. At its core, the severe adverse scenario looked like a variation on what transpired during the GFC in 2008 and afterward; as that stress event was so concentrated in mortgage assets, it makes an excellent scenario when analyzing a GSE. While the test scenarios did vary somewhat over time, as the Federal Reserve tried to have a scenario most reflective of the risks at the time of each test, they really did not vary that much over the years with the reduction in house prices being at or near 25% each year, and - given the extreme recent increase in house prices - only increasing to just under 29% for 2021.

[8] The first few years of tests did not use 12/31 financials, but those of 9/30.   This is because the first stress tests, under the time pressure of very shaky markets, were conducted before the release of yearend financials. This inconsistency has no meaningful impact on the conclusions drawn.

[9] This is the calculated cumulative comprehensive income (a broader measure of profits than net income, and used by the FHFA during conservatorship for various reasons) over the nine quarters of the stress test calculation period.  This was judged by the Federal Reserve, in its original stress tests for large banks in early 2009, to represent a period during which modeled losses (which are based upon front-loaded stress movements, e.g. house prices dropping 25% very quickly) would be incurred, with the banks returning to profitability by the end of the period. 

[10] See the sidebar for the explanation of “with DTA write-down” vs. “without DTA write-down.”

[11] Treasury during the Obama administration was particularly interested in such sales. It was concerned with potentially large further losses if another economic downturn occured (thus requiring more funding from the taxpayer through Treasury) and wanted this risk to be reduced as quickly as possible. 

[12] Formally, the Dodd-Frank Wall Street Reform and Consumer Protection Act, passed in 2010.

[13] This means the mortgages had something less than the full documentation, which includes the lack of procedures such as verification of borrower income levels, etc. 

[14] Such non-QM loan products are still being done by private market firms and banks, but are a very small portion of the markets.

[15] As a reminder, the GSEs are restricted by law to being “secondary market” firms, i.e., they do not deal directly with underlying homeowners in making mortgages but only with lenders to the homeowners (known as “primary market lenders”), such as banks or non-bank mortgage companies.   

[16] Those interest rates and liquidity risks on mortgages are quite high. They were, for example, the core cause of the collapse of the U.S. thrift industry (mainly savings & loan associations) in 1989.  Such MBS has the impact of transferring (i.e. passing through) those risks to the investors in the MBS. Hence, they are often called “pass-through MBS.”

[17] Risk transfers were also done on an economically efficient basis, i.e. not requiring undue payment to the investors for their taking on the risk. As a result, CRT is a win-win: reduced risk to the GSEs and the taxpayers supporting them, and no higher cost of mortgages.

[18] The GSEs purchase mortgages on  single-family and multi-family properties (meaning rental apartment houses). CRT is done on both categories of assets, but the discussion herein focuses on single-family only; I note that certain CRT activities on multifamily predate 2013.

[19] Interestingly, this was almost double the size of the balance sheet of the Federal Reserve at that time.

[20] The cost of the funding was well below what a regular private sector company, even the best rated, could achieve, reflecting the implied guarantee. This then was a major covert subsidy to the two GSEs, and was the largest source of the profits they earned at the time. 

[21] The portfolios, after being reduced by over 80%, now are comprised of assets directly needed by the underlying mortgage purchase-securitize-guarantee business model rather than being discretionary.

[22] There are multiple and competing indices of house prices.  In 2011, although at slightly different times, all the major indices bottomed out.

[23] See Part 2 for an explanation of why I view 2016 as a year that was neither cyclically high nor low. 

[24] Using the FHFA’s index of house prices, the specific increase was 36.3% from Q1 2019 to Q4 2021.  See https://www.fhfa.gov/DataTools/Downloads/Pages/House-Price-Index-Datasets.aspx#qpo, Quarterly Data, Purchase Only, U.S. Summary.

[25]The impact of house prices on GSE credit risk is best followed by watching the average loan-to-value (LTV) ratio.  For the Freddie Mac book of single-family mortgages, the LTV was 61% in 2016, declined to 59% by the end of 2019, and then very quickly improved by declining further to 55% as of yearend 2021.  (All data from Freddie Mac 10-K annual reports.)   I view the 61% level in 2016 as a relatively “normal” year, and thus consistent with long-term stability, whereas the 55% level seems to me to be  very much cyclically favorable and inevitably will increase back towards 61%.

[26] This is known as the CLTV, (or current LTV), where the market value of each home is estimated for the calculation as of the latest reporting date. This is a complement to the OLTV, (or the original LTV), calculated at the time a mortgage was made where an appraisal has often established the market value.

[27] In one sense, the GSE balance sheets reflect assets and liabilities where the major risks of mortgage lending are passing through the GSEs to institutional investors, rather than being, as is the usual tradition, held by the two firms themselves. Thus, the large nominal assets of the GSEs – now over $7 trillion combined – reflect more accounting entries than risks retained by the GSEs. Naturally, then, the capital required per dollar of accounting asset is going to get to levels that are very low in comparison to those associated with institutions such as banks, which overwhelmingly hold onto their risks rather than pass them through to investors. 

[28] The mortgages are, mostly, on single-family homes but the GSEs also make considerable amounts of loans on apartment houses. 

[29] The obligation of Treasury to support the GSEs requires, by government accounting, a cost to be included in the federal budget. To the degree that this cost is reduced in line with the exposure Treasury has as calculated via the stress test, it therefore should reduce the associated federal budget expenditure, allowing Congress to possibly fund other programs with the money no longer required. 

[30]This is true today. Back in 2013, at the time of the first stress test, the DTA was extremely large (i.e. about $100 billion), then mostly being caused by tax-related issues associated with the heavy losses incurred in the GFC years.

[31] While colloquially such a event is called a write-down, in formal GAAP accounting terminology it is called “establishing a valuation allowance” against the DTA.