Abstract
We examine the role of asset-backed security collateralized debt obligations (ABS CDOs) as a primary catalyst for the financial crisis. We show how ABS CDOs became the main investment vehicle for the riskiest investment-grade securities in the private-label mortgage market. We estimate a final tally of writedowns on ABS CDOs, $410 billion in total, with $325 billion assumed by AAA and “super-senior” securities, which had minimal capital, margin, or liquidity requirements. Pre-crisis regulations allowed excessive leverage at some firms investing in these securities, imperiling their solvency and placing them at the center of the financial crisis.
TOPICS: Asset-backed securities (ABS), credit risk management, financial crises and financial market history
It is well understood that the financial crisis emanated from overleveraged financial firms’ investments in risky US mortgages following a severe downturn in US housing markets (Mian and Sufi 2014; Duffie 2019). In this study, we show how many of the losses on US mortgages were concentrated in unrated “super-senior” and triple-A-rated (AAA) securities issued by so-called “asset-backed security collateralized debt obligations” (ABS CDOs).1 We also show how the primary investor base for those ABS CDO classes was large, undercapitalized financial firms and how losses on these ABS CDOs—often in combination with direct losses on mortgage portfolios—imperiled their solvency, placing them at the center of the financial crisis.
The article proceeds as follows. Despite some $635 billion of losses on nonprime mortgage loans, recent research shows that realized losses on AAA private-label mortgage-backed securities (PLMBS) were surprisingly low (Ospina and Uhlig 2018). Rather, catastrophic losses were absorbed mainly by lower-rated PLMBS tranches.
We then extend the work of Cordell, Huang, and Williams (CHW 2011) and show that most subprime PLMBS tranches rated below AAA were placed into ABS CDOs. During the critical period from mid-2005 to 2007, these bonds came to dominate the collateral of ABS CDOs, both in cash form and with credit default swaps (CDS) referencing these bonds. Around three-quarters of the non-AAA investment-grade tranches of ABS CDOs were cross-sold into other ABS CDOs, retaining the risks within ABS CDOs. Furthermore, many PLMBS and ABS CDO equity holders hedged their risk by taking short positions through credit default swaps, with ABS CDOs often taking the long position of those transactions.
The result of these activities was to concentrate nonprime mortgage losses in the AAA and super-senior tranches of ABS CDOs. We update loss figures from CHW (2011) for what is a near-final tally of all ABS CDO losses, $410 billion in total, with $325 billion (79%) realized by the holders of the AAA bonds and super-senior securities.
We then describe how ABS CDO originators bought substantial amounts of PLMBS tranches from each other to place into ABS CDOs that they originated. These ABS CDO originators were among the world’s largest financial firms and suffered some of the largest ABS CDO losses. We extend the work of the existing literature to point out the modeling flaws that led these firms to believe these AAA and super-senior securities were ultra-safe.2 When losses hit, a combination of minimal capital requirements on these securities and enormous leverage at these firms would ultimately imperil their solvency. We conclude with some lessons learned.
NONPRIME RISK TRANSMITTED THROUGH PLMBS
We start with an examination of losses on mortgage loans placed into nonprime PLMBS. These losses were historically very large, with around 90% concentrated in 2005–07 originations. Losses on subprime loans placed in PLMBS will ultimately total close to $400 billion; losses on Alt-A mortgages—to borrowers with strong credit scores but other risky characteristics—will be about $250 billion (Exhibit 1).3 While losses of this size were unprecedented, prospects that they could grow to these magnitudes were contemplated back in 2007 by Federal Reserve Chairman Ben Bernanke and Treasury Secretary Hank Paulson, although both expected the economic and financial impacts of the subprime market turmoil would be “contained” (Bernanke 2007; Bernanke 2010, pp. 53–54).
Subprime and Alt A Losses in PLMBS Cumulative by Vintage through October 2018 ($ billions)
Source: CoreLogic Solutions.
Given the magnitude of the housing crisis and the losses suffered on poorly underwritten mortgages, the AAA rated tranches of subprime PLMBS would ultimately generate surprisingly small writedowns. Exhibit 2 is a stylized model showing how the PLMBS securitization process provides protection for holders of AAA PLMBS securities. Subprime mortgage loans are pooled into security “trusts” and serve as the assets, or collateral, of the securities. The trust issues securities to fund these pools. These securities are constructed of varying risk grades, referred to as “tranches,” and are paid from the principal and interest of the underlying mortgage loans. The AAA securities receive the highest rating due to the protections provided by the trust. When the underlying mortgage loans default, losses are absorbed first by the equity holders and lower-rated securities.4 Prepayments are mostly allocated first to AAA securities. In the case of subprime PLMBS, the protection that lower-rated securities afforded to the AAA securities, referred to as the “subordination,” averaged 23%, meaning losses would need to reach that amount in the mortgage pool before the AAA PLMBS securities suffer their first loss. (Subordination levels for all securities are provided in parentheses.) Because defaulting mortgages can recover as much as 50% or more of their value, this meant that roughly half of the mortgages would need to default before AAA securities would bear any permanent loss.
Subprime Private-Label MBS Trust
Notes: Figures on subordination levels for each class of liabilities are taken from CHW (2011) and are averages over 1998–2007 subprime PLMBS from Intex.
As it turned out, losses did not often reach these levels, meaning that subprime losses were largely absorbed by the non-AAA subordinate PLMBS. While AAA subprime PLMBS prices fell sharply during the crisis with many ratings downgrades, they ultimately generated surprisingly few losses. By one recent estimate, losses on AAA subprime PLMBS originated between 1987 and 2007 averaged less than 1% (Ospina and Uhlig 2018)—see Exhibit 3. The reason for this is PLMBS were structured to protect the AAA tranches from catastrophic losses. According to S&P, AAA subprime PLMBS could sustain 30% house price declines on the coasts and 10% on the interior before they suffered their first dollar loss (S&P 2007, p. 49). Moody’s stresses were similar.
MBS Losses by Credit Rating and Mortgage Type, as of December 2013 ($ billions)
Note: Figures taken from Ospina and Uhlig (2018, Table 5).
Losses on AAA PLMBS backed by prime-quality loans or Alt-A loans were ultimately higher than those of subprime securities, 1.3% and 6.5%, respectively. They experienced larger losses than AAA subprime PLMBS because their subordination levels were much lower, about 2% and 7%.5
The main reason losses on AAA PLMBS were relatively modest was the loss-absorbing capacity of the lower-rated securities. Losses on securities originally rated BBB- to AA, commonly referred to as sub-AAA investment-grade securities, were catastrophic—averaging 51% for subprime PLMBS, 57% for Alt-A, and 34% for prime. What was unprecedented about securitizations in the 2000s was that the principle investor vehicle for these lower-rated investment-grade PLMBS was ABS CDOs, which we examine next.
HOW THE SECURITIZATION PROCESS CONCENTRATED AND MAGNIFIED PLMBS LOSSES IN ABS CDOs
As shown in Exhibit 4, nonprime PLMBS were concentrated in ABS CDOs (Goodman et al. 2008; FCIC 2011), and mortgage losses were concentrated in these ABS CDOs. Those losses would ultimately be assumed by holders of the AAA and super-senior securities. To understand how this happened requires us to describe in detail the securitization process for ABS CDOs.
Transformation of Mortgage Loans to CDO2s, 1998–2007
Notes: Figure taken from CHW (2011). This figure shows the total dollar amounts and counts for the various sources of mortgages, MBS, CDOs, and CDO2s that made up the mortgage market from 1998 to 2007. RMBS = residential mortgage-backed securities.
CHW (2011) provided a comprehensive analysis of the entire ABS CDO market.6 Exhibit 4 shows how various PLMBS tranches made their way into ABS CDOs and how tranches of ABS CDOs were sold into CDO2s. It shows the average subordination levels of each class of securities in parentheses. Over time, the investment-grade securities of nonprime PLMBS became the primary collateral for ABS CDOs. BBB-rated PLMBS securities were concentrated in so-called “mezzanine” ABS CDOs; A-rated securities were concentrated in “high-grade” ABS CDOs. Unlike PLMBS, however, the ABS CDOs were not collateralized by underlying mortgage loans with substantial potential recovery values. Rather, they were collateralized by lower-rated PLMBS securities vulnerable to catastrophic loss because their low position in the payment waterfall meant they would absorb losses at low mortgage default rates. For example, BBB subprime PLMBS on average covered losses on their underlying mortgage pools starting at 4% and became worthless when pool losses reached 8%;7 on average, A rated subprime PLMBS became worthless when losses reached 13%.
Over time, these subprime PLMBS securities became the primary collateral in ABS CDOs. Exhibit 5 shows the collateral composition of ABS CDOs. When first issued in 1999, ABS CDOs were correctly classified as “multisector” ABS CDOs, investing in a broad array of PLMBS and other asset-backed securities (ABS). But when non-mortgage ABS underperformed during the 2001 recession, ABS CDOs became increasingly concentrated in PLMBS, which performed far better at the time.
Original Asset (collateral) Composition of ABS CDOs by Issuance Year
Notes: Figure taken from CHW (2011, Figure 2) and Intex. This exhibit summarizes shares of assets for structured finance ABS CDOs by five asset classes in the bar charts and also includes shares of synthetic collateral in the line plot. CMBS = commercial mortgage-backed securities; Other = all remaining asset classes, including non-ABS CDOs.
The introduction of CDS on PLMBS made the ABS CDO business grow substantially, starting in the second half of 2005. With this innovation, ABS CDO managers could take “long” positions in CDS that referenced subprime PLMBS, instead of investing in actual PLMBS notes. CDS positions in ABS CDOs are often referred to as “synthetic” because they are not real cash securities; they are referencing the cash securities. As shown by the line in Exhibit 5, by 2007 nearly half of all collateral in ABS CDOs were CDS. BBB PLMBS securities and CDS referencing these securities increasingly became the securities of choice for ABS CDOs. Although $77 billion of BBB subprime PLMBS were issued between 1998 and 2007, there was a total of $140 billion of BBB subprime PLMBS placed into ABS CDOs, including both cash and CDS exposures (Exhibit 6).8 That was nearly twice the dollar amount of actual BBB tranches issued in that period.
Subprime PLMBS Placed into ABS CDOs, by Original Rating (1999–2007)
Notes: This exhibit summarizes the numbers and dollar amounts of subprime bonds by original rating issued between 1998 and 2007 and placed into ABS CDOs.
Source: Summarized from CHW (2011, Table 3) and Intex.
CDS are “side bets” between two parties. “Protection buyers” want to hedge their mortgage risk or speculate on a downturn, and are referred to as “short” investors. “Protection sellers” insure against loss and are considered “long” investors. While the exact size of the CDS market is unknown, ABS CDOs became a means through which short sellers could place their short bets on subprime PLMBS using credit default swaps: all told, $201 billion of CDS were placed into ABS CDOs (Exhibit 7). These short sellers (i.e., protection buyers) became what Lewis (2010) famously dubbed “The Big Short.”
ABS CDO Cash and CDS Collateral, by Vintage, 1999–2007 ($ millions)
Notes: Figures summarized from CHW (2011, Table 2) and Intex.
This exhibit breaks out ABS CDO issuance between cash and CDS for 1999–2005H1, then semi-annually thereafter.
What led to the enormous growth of CDS in ABS CDOs is that in July 2005 the International Swaps and Derivatives Association (ISDA) introduced standardized contracts that allowed CDS referencing PLMBS to replicate the payment characteristics of cash bonds. This greatly facilitated the issuance of CDS referencing PLMBS and of ABS CDOs in general.9 As shown in Exhibit 7, of all CDS placed into ABS CDOs, 93% were issued starting in the second half of 2005.
The entrance of CDS short sellers into the ABS CDO market had two important consequences. According to Moody’s analysts, CDS facilitated the “time to market” for an ABS CDO. What took a couple of months for cash ABS CDOs to be issued got reduced to as short as a few weeks with the ability to use CDS. From mid-2005 to 2007, $489 billion of ABS CDOs were issued, over three-quarters of the total ever issued. The second effect was that it introduced a major new class of investors to the ABS CDO market, those betting on them to underperform. Moody’s analysts acknowledged their models could not account for the risks this brought to the ABS CDOs (Kolchinsky 2010).
A feature of ABS CDOs that further increased the risk for AAA and super-senior tranche investors was that most of the lower-rated tranches of ABS CDOs were “cross-sold” to other ABS CDOs, undermining the subordination that the AAA tranches should have had. Cross-selling occurs when ABS CDO securities are sold to other ABS CDOs, retaining risks within the ABS CDO market. As shown in Exhibit 8, of the lower-rated investment-grade tranches of ABS CDOs, 75% were sold into or referenced as CDS in other ABS CDOs.
ABS CDO Bonds Placed into ABS CDOs by Original Ratings, 1999–2007
Note: $ Placed includes CDS references.
Source: Summarized from CHW (2011, Table 5) and Intex.
What proved damaging to holders of super-senior and AAA tranches of ABS CDOs is that cross-selling reduced the subordination provided to senior bondholders. In Appendix A, we present a hypothetical case of how two ABS CDO trusts cross-selling their subordinated bonds to each other eliminates subordination altogether. While extreme, we observed in ABS CDOs 249 bonds that were directly cross-sold between two CDO trusts. More fundamentally, the cross-selling of three-quarters of subordinated ABS CDO bonds into other ABS CDOs retained these risks within the ABS CDOs and reduced the protection that subordination provided for holders of ABS CDO super-senior and AAA tranches (unlike their AAA PLMBS counterparts).10
A final issue relates to the equity investors in ABS CDOs. Equity investors typically serve the role of sponsor in a securitization and have input in the selection of collateral. By the second half of 2006, as housing prices declined, it became popular to short PLMBS or ABS CDOs through CDS while financing the short position with equity tranches in ABS CDOs (FCIC 2011, pp. 192–195). While there was nothing wrong with this strategy, when an equity position is hedged or used to support short positions, it alters the incentives of the equity holders who, in the absence of hedges, are in a long, first-loss position. Furthermore, there were documented cases in which hedge funds bought the equity tranches of ABS CDOs while shorting other tranches of the same ABS CDOs—and then used their role as the equity investor to recommend particularly low-quality assets, unbeknownst to senior bondholders who assumed they were both on the same side of the trade.11
Hedge fund activities are very confidential. But in 2010, the FCIC was able to crudely measure the extent of this long–short activity by conducting an anonymous survey of 170 hedge funds with $1.1 trillion in assets under management. As of June 2007, the largest hedge funds that responded to the survey held $25 billion in equity and other lower-rated tranches of PLMBS, offset by $45 billion in short positions. Similarly, a handful of hedge funds accumulated positions totaling more than $1.4 billion in ABS CDO equity tranches and almost $3 billion in short positions.
HOW BIG WERE ABS CDO LOSSES AND WHO HELD THE ABS CDO RISK?
Given that three-quarters of ABS CDOs were issued between 2005 and 2007, where around 90% of PLMBS losses were concentrated (Exhibit 1), it is not surprising that ABS CDOs would suffer catastrophic losses. What has not been documented is just how concentrated these losses would become in the AAA and super-senior tranches of ABS CDOs—and in the firms originating them.
To estimate losses, CHW (2011) conducted a full “bottom up” valuation of all ABS CDOs in the market. As shown in Exhibit 9, we updated these numbers through December 2018, giving us a near-final tally on total losses in the ABS CDO market: $410 billion, a loss rate of 65%, with increasingly higher loss rates for later vintages.12 Nearly 90% of those losses were concentrated in the mid-2005 through 2007 vintages, which had the preponderance of the CDS, most of the cross-sold ABS CDO tranches, and the riskiest subprime PLMBS. Most importantly, $325 billion of these losses were assumed by AAA and super-senior investors, with loss rates of 70% and 84% for the 2006 and 2007 vintages.
Summary Expected Losses for ABS CDOs by Issuance Year and Original Rating
Notes: Super-senior are securities often not rated but senior to AAA in their trusts. Because rating agencies do not have ratings above AAA, they are combined with AAA. Losses include unpaid original principal plus $2.9 billion of unpaid capitalized portions of accumulated interest shortfalls, so losses can exceed 100%.
Sources: Intex for valuations and Bloomberg and IDC for fair values on ABS and PLMBS collateral bonds.
An equally important question is, who held ABS CDO risk and why? As shown in Exhibit 10, a major investor base was the ABS CDO originators themselves, as ABS CDO writedowns at these firms totaled $127 billion in all.13
Subprime PLMBS and ABS CDO Issuance Volumes and Losses ($ millions) for Top 18 ABS CDO Originators
Notes: This exhibit details subprime activities of the top 18 ABS CDO originators from 1998 to 2007, which issued 89% of all ABS CDOs. It includes their subprime PLMBS issuance and their ABS CDO issuance along with the volumes of senior AAA ABS CDO issuance that many retained on or off balance sheet. Finally, it includes all reported writedowns as reported on their financial statements from 2007 through January 26, 2009, from the CreditFlux Writedown Report.
Sources: Intex for issuance volumes, Creditflux for writedowns.
How did these ABS CDO originators end up holding ABS CDO securities when their business models were originate-to-distribute? We already described the cross-selling of lower-rated ABS CDO bonds between originators. Beltran, Cordell, and Thomas (2017) also showed that most ABS CDO originators were fully vertically integrated through the entire mortgage market, from originations through PLMBS and ABS CDO issuance, and mortgage servicing. Exhibit 10 confirms that 15 of 18 major ABS CDO originators also issued subprime PLMBS.14 Rating agency issuer-diversification rules dictated that each originator could only use a small portion of their own subprime PLMBS in their ABS CDO trusts. These ABS CDO originators became important buyers of each other’s BBB PLMBS bonds.15
As for why ABS CDO originators held their own ABS CDOs, some ABS CDO originators were unable to sell their securities, particularly after the subprime market shut down in 2007. But many of these firms were willing to hold the super-senior tranches of ABS CDOs because they believed they were ultra-safe.16 The confidence of many investors and financial institution risk managers was driven by a new and innovative modeling technology employed at the rating agencies and more broadly in the market, which we examine next.
MODEL RISK: WHAT WENT SO WRONG WITH ABS CDO MODELS?
As discussed, subprime PLMBS were structured to protect AAA investors against catastrophic loss, and loan-level data provided a straightforward conceptual framework to model losses and easily handle stress scenarios, such as a 30% drop in house prices, that were used to size the AAA subprime PLMBS subordination levels. For ABS CDOs, it was infeasible to separately estimate expected losses on each of the underlying ABS collateral securities, as we were able to do with the full set of historical data from the market.
To model credit risk in an ABS CDO market increasingly dominated by securities tied to mortgages, starting around 2005, rating agencies adopted a variant of the Gaussian copula model; the broader market generally followed this same approach.17 The modeling framework adopted by Moody’s (2005) is described in Appendix B.18 To show how the copula model was able to transform BBB subprime bonds into AAA and super-senior securities, we walk through the valuation of a very simple ABS CDO structure, Abacus 2007-AC1, a Goldman Sachs deal composed of $2 billion of CDS referencing $1.2 billion of BBB subprime bonds. Exhibit 11 shows the final structure with the bond balances and ratings, initial support levels, and fees.
Abacus 2007-AC1 Ratings—Initial Support and Fees
Source: Intex and Abacus Prospectus.
For the probability of default and recovery assumptions, Moody’s and S&P had long histories of corporate bond downgrades stretching back to before the Great Depression. Moody’s used these to proxy for expected downgrades on PLMBS bonds in their ABS CDOs, even though PLMBS risk is arguably very unlike corporate bond risk. From these assumptions, Moody’s constructed “Idealized Probability of Default” tables based on the weighted average life and a composite score based on the weighted average ratings of the assets in each ABS CDO (see Exhibit B1 in Appendix B).19
Most critical was the correlation assumption across different assets within an ABS CDO. Their correlation assumptions were based on “historical correlations” of bond downgrades. But because most PLMBS issuance occurred during a period of rising house prices, there were few PLMBS downgrades over their history, leading modelers to conclude that correlations among PLMBS bonds were very low; Moody’s assumed a 12% correlation, S&P’s a much lower 6%.20 As we show in Appendix B, these very low correlations allowed 79% of the Abacus deal to be rated AAA. And Abacus was typical of the market as a whole. As a result of these very low assumed correlations, as much as 76% of mezzanine ABS CDOs and 89% of high-grade ABS CDOs were rated AAA, as we show in Exhibit 4.21
After the financial crisis, S&P raised its correlation assumption for PLMBS to 70% (Adelson 2016, p. 42). At that level, implied default rates would not have allowed any portion of the Abacus deal to be rated AAA.
What went wrong with these models? The easy answer would be that the gaussian copula models did not directly incorporate house prices as a variable, so the low correlations in their models were based on recent history during a booming housing market. Another, more fundamental problem was that significant diversification was taking place within plmbs pools, so the resecuritization into abs cdos did not generally increase diversification. Essentially, the abacus deal—like all abs cdos of this period—was composed of plmbs bonds with the same essential characteristics.22
Ultimately, all 90 BBB bonds in Abacus 2007-AC1 would suffer 100% loss. What proved damaging for financial markets was that the losses were borne by highly levered financial institutions that had not held sufficient capital against their positions. The Dutch bank ABN AMRO had guaranteed the super-senior CDS tranche, suffering over $900 million in losses, while the German bank IKB acquired $150 million of the two AAA CDS, suffering a total loss. Both eventually were rescued by their governments.
Abacus 2007-AC1 was symptomatic of the broader market. While CDS losses are a zero-sum game between protection buyers and sellers, what made ABS CDO and CDS losses a primary catalyst for the financial crisis was that the AAA ABS CDOs and CDS referencing them had little to no capital, margin, or liquidity requirements and were concentrated among the world’s largest financial firms, exposed to this risk by massive leverage at the firms, which we explore next.
MORTGAGE RISK TRANSMITTED THROUGH THE FINANCIAL SYSTEM BY LEVERAGE
In this section, we focus on the Basel risk-based capital rules and gaps in laws and regulations that would leave these firms dangerously exposed. First, favorable capital rules on AAA rated securities created strong incentives for banks to buy and hold these securities. The Basel Committee on Banking Supervision had set a basic 8% capital charge relative to “risk-weighted assets,” with weights assigned under Basel I and II regulations. As shown in Exhibit 12, commercial loans received a 100% risk weight and an 8% capital charge. Whole mortgage loans carried a 50% risk weight, meaning they required 4% capital.23 After revisions by US regulators in 2001, securities rated AA or AAA carried a 20% risk weight, for a capital charge of just 1.6%, a leverage ratio of 62.5 to 1 (OCC et al. 2001b). With Basel II, international regulators in 2004 further eased standards for AAA rated securities. For “granular pools” like those of private-label PLMBS and CDOs, capital on super-senior securities was set at 0.56%—a leverage ratio of 179 to 1!24
Risk-Weighted Assets, Capital, and Leverage for Select Assets
Notes: Tier 1 capital is core capital, consisting primarily of common equity and disclosed reserves. “Super Senior” are securities with subordinate AAA bonds.
Source: Federal Reserve Regulatory Y-9C Reports.
A second issue is that gaps in regulations resulted in increased leverage at the firms. CDS referencing super-senior ABS CDOs escaped regulation altogether because over-the-counter (OTC) derivatives markets were banned from regulation by federal and state regulators by the Commodity Futures Modernization Act of 2000.25 The CDS written on ABS CDOs had effectively no capital, margin, mark-to-market, or liquidity requirements.26 The decision to write CDS on ABS CDOs proved disastrous for many firms, particularly AIG. AIG’s Financial Products Group wrote CDS on some $62 billion of ABS CDOs. After bailing out AIG, the Federal Reserve exercised the CDS, making all of AIG’s counterparties whole, and then placed the underlying assets of the ABS CDOs into the Maiden Lane III trust for $29 billion—representing a $33 billion writedown for AIG (McDonald and Paulson 2015).
Another significant gap was in the supervision of the five large independent investment banks—Bear Stearns, Goldman Sachs, Lehman Brothers, Merrill Lynch, and Morgan Stanley. In 2004, the SEC placed these firms under its Consolidated Supervised Entity (CSE) program, subject to Basel II. This allowed them to use their own internal models to determine their capital charges and increase their leverage.27 By the end of 2007, these firms had leverage ratios as high as 40 to 1 (FCIC 2011, p. xix).
US Commercial Banks: The Case of Citigroup
While US bank regulators never fully adopted Basel II, the risk weightings they implemented for securitized assets were similar. With mortgage assets and securities, this meant that AAA MBS and ABS CDOs carried a 20% risk weight; liquidity puts on off-balance-sheet conduits were half of that (Exhibit 12).
Because of the variety of exposures Citigroup had to ABS CDOs, the case is a good example of the problems with the pre-crisis use of risk weightings in capital standards. In June 2007, Citigroup had $2.2 trillion of total book assets, and $1.3 trillion of risk-weighted assets, driven partly by low-risk weightings on their AAA ABS CDO exposures. This meant that Citigroup’s regulatory capital ratios nearly doubled when using risk-weighted assets as the denominator. As shown in Exhibit 13, Citigroup’s regulatory capital ratio was 8%–9% of risk-weighted assets clear through the financial crisis, which made it well capitalized by regulatory capital standards (OCC 2010).28 But its leverage ratio (total book assets to capital) increased from 15 to 1 in 1999 to 25 to 1 by 2007.
Citigroup: Leverage vs. Risk-Based Capital
Source: Federal Reserve’s Regulatory Y-9C reports.
Citigroup issued $25 billion of super-senior ABS CDOs in the form of commercial paper through off-balance-sheet conduits. To ensure an A-1 rating from rating agencies, Citigroup provided liquidity protection in the form of liquidity puts on the commercial paper. Off-balance sheet, these liquidity puts carried a 10% risk weight, which meant $200 million in required capital. After the commercial paper market shut down with the onset of the financial crisis in August 2007, Citigroup bought the commercial paper to avoid exercising the liquidity puts, bringing the underlying exposures on balance sheet. Over the next three years, based on its 10-K filings, Citigroup would take a cumulative loss of $13 billion on those exposures. Citigroup also held $18 billion in super-senior AAA tranches of synthetic ABS CDOs it had underwritten. With a 20% risk weight, this implied $300 million in capital.29 Over the next three years, it would mark that down by $14 billion. In total, perhaps a half billion dollars in capital was held against exposures that ultimately generated $27 billion in losses.
What is most remarkable is that because these ABS CDOs were “super-senior,” Citigroup did not consider them part of their subprime exposures on their 10-K filing as late as October 15, 2007. But on November 4, Citigroup revised its 10-K to add these $43 billion to their total subprime exposures, increasing it to $55 billion (FCIC 2011, p. 265). According to CEO Charles Prince, these losses were the leading source of Citigroup’s eventual rescue with $45 billion of funds from the federal Troubled Asset Relief Program in 2008.
Citigroup was a microcosm for how ABS CDO losses were transmitted through the financial system in 2007 and 2008. Super-senior securities were also held in large amounts on balance sheet at UBS and Merrill Lynch (Exhibit 10). ABS CDO writedowns were recorded at 16 of the 18 ABS CDO issuers, many of which would receive government support in various forms or would be merged out of existence. In addition to AIG’s $33 billion of writedowns, several other insurers suffered large writedowns. All told, among the world’s major financial firms, $220 billion of ABS CDO losses were recorded, 42% of all writedowns recorded between July 2007 and January 2009 by Creditflux (Exhibit 14), making these ABS CDO losses a primary catalyst for the financial crisis.
Crisis-Related Writedowns through January 26, 2009 ($ millions)
Source: Creditflux. All information based on publicly disclosed company information or on reliable press reports of public disclosures, compiled by Creditflux Ltd.
CONCLUSIONS AND LESSONS LEARNED
In this article, we describe how the securitization process for PLMBS in the mid-2000s concentrated some $325 billion of losses in ABS CDO AAA and super-senior securities, placing them at the center of the financial crisis. At one level, it is easy to blame model risk and flawed assumptions used by the rating agencies and many investors. But we have shown that catastrophic losses suffered by ABS CDOs go far beyond model error. The catastrophic losses suffered by these ABS CDOs were the result of ABS CDOs being the primary investment vehicle for virtually all non-AAA investment-grade subprime PLMBS that got placed and referenced in ABS CDOs many times; being in the long position for some $200 billion of CDS short bets; and being the primary investor for most of the lower-rated subprime bonds in the ABS CDOs themselves. For these reasons, Goodman et al. (2008) called ABS CDOs “the greatest ratings and risk management failure ever.”
But while ABS CDOs were a primary catalyst for the financial crisis, the primary cause of the financial crisis remains the enormous leverage built up in the pre-crisis financial system (Duffie 2019). Little or no capital, margin, or liquidity requirements were placed on ABS CDO AAA and super-senior investments or on related commercial paper conduits. Pre-crisis regulations that led to excessive leverage at many of these financial firms is what made ABS CDO losses so damaging.
Have we learned our lessons? To be sure, many changes were made to directly address the risks exposed by ABS CDOs and off-balance-sheet investment vehicles. In 2009, the US Financial Accounting Standards Board, a private standard-setting body regulated by the SEC, required bank sponsors to bring back on-balance-sheet commercial paper programs and other special purpose entities. In July 2009, international financial regulators substantially increased capital requirements for resecuritizations, removing much of the capital incentive that had helped drive the creation of ABS CDOs in 2006 and 2007 (BCBS 2009). The Dodd–Frank Act promoted the use of central clearing in the derivatives market, improving margin standards and reducing bilateral counterparty risk (Duffie 2019). In an effort to better align incentives in structured products, “risk retention” rules were set to better align incentives for investors with a controlling interest in the trusts.
We have supplemented the new Basel capital regime with bank stress tests. While this is a clear improvement from a supervisory perspective, in the United States it exposes the Federal Reserve System to some of the same model risks that rating agencies and financial institutions faced in the run-up to the crisis. A recent study by Aikman et al. (2019) points out that it would have been very difficult for the post-crisis Dodd–Frank Act regulatory apparatus to have halted the damaging practices in the pre-crisis mortgage market. It is humbling to contemplate that few market participants fully understood the catastrophic risks built into ABS CDOs, and it’s very unlikely anyone could have uncovered in real time what we describe in this study with the benefit of hindsight.
The new capital rules have increased capital requirements, and so far, our largest banks are significantly less leveraged. (Citigroup’s leverage ratio has been as low as 10 to 1 in recent years.) Still, two former FDIC officials note that “America’s biggest banks are among the most leveraged institutions operating in the country today” (Hoenig and Bair 2018). Risk-weighted assets remain central to capital adequacy requirements, but they are imperfect. It remains essential for bank regulators to retain non-risk-weighted leverage standards as a crucial backstop to prevent a recurrence of overleveraged firms in the run-up to the financial crisis.
ADDITIONAL READING
Inflated Ratings on Pre-Crisis CDOs: A Deeper Look
Mark Adelson
The Journal of Structured Finance
https://jsf.pm-research.com/content/22/2/37
ABSTRACT: Methodologies for analyzing and rating collateralized debt obligations (CDOs) backed by structured finance securities (SF-CDOs) were significantly flawed before the 2008 financial crisis. Two key issues were model risk and the absence of calibration to historical benchmarks. Although financial literature highlighted model risk as a potential problem starting in the mid-1990s, market participants— including the rating agencies—failed to properly address the issue. That partly explains why SF-CDOs that performed horribly during and after the financial crisis received high credit ratings before the crisis. With the 2009 update to its corporate CDO rating methodology, Standard & Poor’s was the first rating agency to embrace measures designed to mitigate model risk. Those measures included calibrating the methodology’s simulation model against external benchmarks and adding outside-the-model tests. Today, with the benefit of experience from the crisis, CDO investors should be wary of relying on any analytic methodology that does not specifically address model risk.
CDOs in the Financial Crisis
Eric S. Adams
The Journal of Structured Finance
https://jsf.pm-research.com/content/15/4/11
ABSTRACT: Assets underlying collateralized debt obligations (CDOs) have been hit particularly hard by the credit crisis. Newissuance activity for CDOs has dropped dramatically, and liquidation, restructuring, and litigation scenarios have become increasingly commonplace. The author describes common issues that troubled CDOs are facing and offers an overview of some of the legal issues and provisions in indentures that are being discussed in the CDO marketplace. The article also points out that distressed asset sales undertaken by liquidating CDOs may offer opportunities for savvy investors.
Commercial Real Estate CDOs
Douglas J. Lucas, Laurie S. Goodman, Frank J. Fabozzi, and Rebecca J. Manning
The Journal of Portfolio Management
https://jpm.pm-research.com/content/33/5/158
ABSTRACT: Collateralized debt obligations backed by commercial real estate (CRE CDOs) have quickly become an important and accepted part of commercial real estate finance. Issuers are drawn to CRE CDOs for efficient financing, and investors are drawn to the diversification benefits. The authors describe CRE CDOs, their evolution, current market trends, and historical performance. CRE CDOs give traditional real estate asset managers the flexibility to take advantage of relative value opportunities within the real estate market, while providing non-recourse financing that matches the average life and interest profile of the underlying asset pool. And as managers are drawn to CRE CDOs for structural flexibility and term financing, investors benefit from exposure to a diverse portfolio of managed CRE assets. As the CDO market continues to develop and evolve, there is an increasing number of managed CRE CDOs and CRE CDOs backed by commercial real estate loans. While this evolution brings new risks to CRE CDOs, it also provides new opportunities for investors. The authors also provide a framework for evaluating CRE CDO investments and discuss how rating agencies rate CRE CDOs.
ACKNOWLEDGMENTS
The views expressed are solely those of the authors and do not necessarily reflect the views of the Federal Reserve Bank of Philadelphia, the Federal Reserve System, Yale University, or the University of Illinois. Special thanks go to Alan Huang, Liang Geng, Xudong An, and Jason Keegan for support and work beyond what was provided by Yilin Huang and Meredith Williams from a prior research paper. The authors would also like to thank Benjamin Kay, Suleman Khan, Andrew Metrick, and Sriram Rajan for valuable comments on earlier drafts.
APPENDIX A
In this appendix, we create a stylized world of two ABS CDO trusts to illustrate how cross-selling of subordinated ABS CDO bonds eliminates their support. Assume a world with two banks each holding $40 million of MBS assets, $80 million in all. Assume each sets up an ABS CDO trust by issuing $40 million of AAA securities and a $10 million BBB security. They sell the AAA security into the market and cross-sell the $10 million BBB securities to each other. (For simplicity, we ignore the equity tranche.) Exhibit A1 depicts the two trusts. As a result, this stylized world is composed of two ABS CDO trusts with total assets of $100 million, made up of $80 million of MBS and $20 million of each other’s BBB securities.
Two Trusts
Moody’s Idealized Probability of Default Table
Notes: WARF = weighted average rating factor. The WARF is a numerical score assigned to each rating class and is a way to create a numerical composite rating for valuation.
Source: Moody’s.
Now assume $10 million of losses are suffered on the MBS of ABS CDO A. In the following, we describe in five steps how losses are transmitted through these two trusts, wiping out the BBB bonds in both trusts and generating $10 million of losses to the AAA bond of ABS CDO A.
1. $10 million in losses are realized for ABS CDO A, reducing assets to $40 million. ABS CDO A’s liabilities are reduced by $10 million, making ABS CDO A’s BBB bond worthless.
2. ABS CDO A’s writedown of its BBB bond results in a loss of $10 million to the assets of ABS CDO B, reducing ABS CDO B’s assets to $40 million.
3. ABS CDO B’s liabilities get reduced by $10 million, wiping out its BBB bond.
4. ABS CDO B’s writedown of its BBB bond results in a $10 million loss to the assets of ABS CDO A, reducing ABS CDO A’s assets to $30 million.
5. Now ABS CDO A must write down its own AAA bond by $10 million (to $30 million), resulting in a $10 million loss to the AAA bond of ABS CDO A, showing how the cross-selling of subordinated bonds to each other eliminates the protection from subordination for senior ABS CDO bondholders.
The key is that no new assets were actually created as a result of the cross-selling. Had the BBB bonds been sold to outside investors, the BBB bonds would have provided protection to the ABS CDO bondholders. By retaining the BBB bonds within their world, the risks were retained within the ABS CDOs; no subordination was created. Thus, after the $10 million loss, the two CDOs have total assets of $80 million − $10 million = $70 million, with the full $10 million loss absorbed by ABS CDO A’s AAA investors.
Draycott (2012) provided an example showing that even a small loss to one BBB bond “ricochets” losses back and forth, making both bonds worthless. Thus, because cross-selling does not add true assets or subordination, losses of any size will wipe out the BBB bonds. While this is an extreme case, direct cross-selling was common among ABS CDOs. We identified 249 cases of direct cross-selling in the population of ABS CDOs. More broadly, this illustrates that cross-sold subordinate bonds don’t create additional loss-bearing capacity for the ABS CDO market.
APPENDIX B
THE GAUSSIAN COPULA MODEL ADOPTED BY MOODY’S (2005)
Moody’s formally adopted the Gaussian copula model in May of 2005. In this model the copula determines the dependency structure between assets in the portfolio. To create a closed-form solution the one-factor Gaussian copula model introduced by Vasicek (2002) and implemented by Li (2000) was used. This widely used one-factor copula model offered a closed-form solution that could be easily implemented in spreadsheets, as was done with Moody’s CDOROM© software.
In this model, the Copula determines the dependency structure between assets in the portfolio. Simplifying assumptions are made such that the portfolio is homogenous where
n = number of assets,
t = weighted average life (WAL) of the portfolio,
p = probability of default from Moody’s Idealized Probability of Default tables,
ρ = pairwise asset correlations,
r = recovery rate, and
Asset Value = (Original notional amount)/n.
The components we need in order to calculate tranche ratings and subordination levels are the amount of loss at a given default, the probability at which that would occur, and a recovery rate on defaulting assets. The final model used to calculate the probability of k defaults is
The loss that would occur at k defaults is derived as: Loss given k defaults = k × Asset value × (1 − r).
Based on this information, the following assumptions were used for the Abacus deal:
Balance = $2 billion
n = 90
Homogeneous asset amount = $2 billion/90 = $22,222,222
t = 4.2 years
p = 1.276%
ρ = 12%
r = 30%
As shown in Exhibit 11, the support level for the Aaa bonds was $420 million. In order for the Aaa bonds to suffer a loss, 28 of the homogeneous bonds would have to default.30
Based on the 2005 model, P(k ≥ 28) ≈ 0. If we update the parameter ρ = 70%, then P(k ≥ 28) ≈ 0.61%. In this case, the rating will be Baa1, BBB+ in S&P vernacular.
We can calculate the support needed in order to achieve a Aaa rating when ρ = 70%:
Because 0.0029% > 0.0020%, based on our probability of default alone we would not be able to rate any portion of this portfolio Aaa. However, because we assumed a recovery rate of 30%, this allows 30% of the portfolio to be rated AAA. In the Abacus deal, the subordination level for the AAA bonds starts at 21%, meaning that even if all 90 assets failed, the AAA bonds would still pay off. This assumption also proved wrong, as all 90 PLMBS were fully written down.
ENDNOTES
↵1 “Super-senior” securities are most often credit default swap (CDS) contracts referencing senior AAA ABS CDO securities. While not rated, they are senior to AAA bonds in a deal. Because rating agencies do not have a rating above AAA, we classify them alongside AAA in the analysis.
↵2 See Benmelech and Dlugosz (2009), Griffin and Tang (2012), Adelson (2016), and FCIC (2011).
↵3 We do not consider here the $30 billion of prime PLMBS losses or losses by Freddie Mac, Fannie Mae, or Ginnie Mae because they had no or tiny shares of securities placed in ABS CDOs.
↵4 For subprime PLMBS, overcollateralization (OC) was common, where additional loans were placed into securities, providing an additional layer of protection against loss. Equity holders receive any “excess spread” not provided to bondholders, which is the first to absorb losses after the OC.
↵5 What would prove catastrophic for AAA nonprime PLMBS were mark-to-market writedowns, which resulted in large reported losses for many firms. According to Creditflux, PLMBS writedowns totaled $109 billion at large financial firms (see Exhibit 14), even though ultimate writedowns were smaller.
↵6 Intex classified ABS CDOs as CDOs that included residential mortgage securities or other securities that required active management because of paydown options.
↵7 In Lewis (2010) and the accompanying movie, they referred to “Armageddon” as occurring when subprime PLMBS losses reached 8%, the loss rate when BBB bonds became worthless, which meant catastrophic losses for even the AAA and super-senior securities of mezzanine ABS CDOs. We will show later how this happened with Abacus 2007-AC1.
↵8 For this analysis, we focus on subprime PLMBS placement into CDOs because it represented about half of all the collateral. Alt-A PLMBS collateral was a much smaller 15%.
↵9 This structure became referred to as “pay as you go” (PAUG) because PLMBS have prepayment and default options that cause balances to fluctuate over time. The PAUG rules were set up to match the cash characteristics of the underlying reference securities.
↵10 We thank Miles Draycott (2012) for valuable discussions on this. When AIG Financial Products Group CEO Joseph Cassano was asked why AIG did not hedge its CDS exposures on ABS CDOs, he stated, “The subordination is a structural short against the long.” See http://fcic.law.stanford.edu/interviews/view/318.
↵11 This phenomenon was first reported on by ProPublica in a Pulitzer Prize winning series of articles. See http://www.propublica.org/article/banks-self-dealing-super-charged-financial-crisis.
↵12 CHW (2011) estimated losses at $420 billion. Our process follows theirs, with most losses fully realized on the underlying bonds. We used price information from Bloomberg and IDC for the remaining bonds.
↵13 Our source for writedowns is the newsletter Creditflux, which tallied writedowns at the world’s largest insurers and commercial and investment banks between July 2007 and January 2009, which we verified independently. All initial writedowns on ABS CDOs were realized over this time. We use these figures for firm losses throughout.
↵14 These 18 ABS CDO originators issued 90% of all ABS CDOs.
↵15 In our online supplement, we provide a large 18 × 18 table that shows the share of BBB PLMBS placed by each CDO originator with each other. Shares generally range from 2% to 6% for each originator.
↵16 For example, the FCIC reported that Citigroup retained most of the super-senior and AAA tranches of most of the CDOs that it created in 2006 and 2007. In many cases, it would hedge the credit risk from these transactions by buying protection from a monoline insurance company, making a spread between the return on the CDS and the cost of the protection. This is referred to as a “negative basis trade.”
↵17 Prior to 2005, rating agency models determined AAA subordination levels by the extent to which firms were diversified across different industries, or asset types in the case of ABS CDOs. See Hu (2007) for a discussion.
↵18 While rating agency models generally needed to converge in their assessments, we use the model from Moody’s (2005) because of its tractability for our example. See Adelson (2016) for details about the S&P model.
↵19 The composite score was called the “weighted average rating factor” (WARF) that gave a composite number to the overall collateral pool to match with the ratings. The number assigned to each bond rating is listed in the first column of Exhibit B1 in Appendix B.
↵20 See Moody’s (2005, p. 4) and Adelson (2016, p. 41).
↵21 A pioneering study by Griffin and Tang (2012) also showed how a rating agency made subjective adjustments to subordination levels of ABS CDOs to increase the share assigned the AAA rating.
↵22 Rating agencies required PLMBS trusts to be diversified across geographies. Combining PLMBS bonds in an ABS CDO did not diversify the collateral pool. Coval, Jurek, and Stafford (2009, p. 16) said, “The overlap in geographic locations and within mortgage pools raised the prospect of higher-than-expected default correlations.”
↵23 US guidelines called for subprime loans to carry a risk weight of 150% to 300% of similar non-subprime loans, which meant the capital charge for a bank holding a subprime mortgage on-balance sheet would have been 6% to 12%. See OCC et al. (2001a).
↵24 By contrast, investment-grade BBB rated securities received the full 100% risk weight, and non-investment-grade and equity tranches received a 1,250% risk weight, meaning dollar-for-dollar capital charges.
↵25 See Shapiro (2010) and FCIC (2011, p. xxiv).
↵26 Had these CDS been regulated as insurance contracts, which they effectively were, they could not have escaped those disciplines. See FCIC testimony by Michael Greenberger at https://www.c-span.org/video/?294322-2/2008-financial-crisis-derivatives-day-1-academics-panel.
↵27 According to Shapiro, “these models were thought to more accurately reflect the risks posed by these activities, but were expected to reduce the capital charges and therefore permit greater leverage by the broker-dealer subsidiaries” (Shapiro 2010, p. 9). See also Duffie (2019).
↵28 Lehman CEO Dick Fuld told the FCIC that when it declared bankruptcy in September 2008 its Tier 1 capital ratio was 11% (Fuld, 2010, p. 8).
↵29 Citigroup also bought protection from monoline insurers on its CDOs. The monolines’ protection would prove insufficient. Citigroup took $8 billion of losses on these exposures. It would not have held any capital against this portion of its super-senior CDO exposure. Total figures we report here match Creditflux’s reported combined writedowns of $34 billion (Exhibit 9).
↵30 We arrived at 28 defaults because the loss on 28 bonds will generate a loss of 28 × 22,222,222 × (1−r) = $622,222,216 × (1 – 30%) = $435,555,551, piercing the $420 million attachment point for the A2 bond in Exhibit 11.
- © 2019 Pageant Media Ltd
Jump to section
- Article
- Abstract
- NONPRIME RISK TRANSMITTED THROUGH PLMBS
- HOW THE SECURITIZATION PROCESS CONCENTRATED AND MAGNIFIED PLMBS LOSSES IN ABS CDOs
- HOW BIG WERE ABS CDO LOSSES AND WHO HELD THE ABS CDO RISK?
- MODEL RISK: WHAT WENT SO WRONG WITH ABS CDO MODELS?
- MORTGAGE RISK TRANSMITTED THROUGH THE FINANCIAL SYSTEM BY LEVERAGE
- CONCLUSIONS AND LESSONS LEARNED
- ADDITIONAL READING
- ACKNOWLEDGMENTS
- APPENDIX A
- APPENDIX B
- ENDNOTES
- REFERENCES
- Supplemental
- Info & Metrics
- PDF (Subscribers Only)
- PDF (Subscribers Only)