CDO Meltdown (3): Summary of Results

(Part 1, 2)

My previous post summarized a few major aspects of that were well-known about CDOs--mostly notes from the first 27 pages of Anna Katherine Barnett-Hart's thesis, "The Story of the CDO Market Meltdown." However, this is just the first quarter of a 115 pp. paper. What follows is attempt to summarize more of the actual findings peculiar to the paper.

EXAMINATION OF THE CREDIT RATING AGENCIES

The credit rating agencies (CRAs) consist of Moody's, Standard & Poor's, and Fitch. Fitch was a relatively minor player in the CDO sector for reasons that will be investigated in the "Findings" part of this post. Letter ratings by the CRAs have had statutory significance for decades, since pension funds were required by law to invest only in AAA securities.¹ In theory, institutional investors or banks were obligated to practice due diligence; complete abdication of responsibility to CRAs for such an important function seems rather extraordinary.

The ratings supplied by CRAs are supposed to measure the likelihood of fulfillment of the obligations of the underlying security. The securities we are talking about are securitized debt and bonds, or derivatives thereof, and the question that a rating is supposed to answer is, How likely is it that the investor will receive principal and interest on time? In some cases, interest payments and principal are rated separately; in other cases, the only thing being rated is the likelihood of the of the investor getting the initial investment back.²

After 2002, when the CDO became a major investment vehicle, CRA business soared.³ A curious aspect of this boom in ratings demand was that CRAs were now the most important customers for their own product. CRAs rated residential mortgage backed securities (RMBS) or other constituents of CDOs, and then used these same ratings (plus formulae for pooling and diversifying risk) . For purposes of pooling, it was useful to have multiple CRAs rate the same security, and then use it in multiple CDOs. If a CRA was assessing securities already rated by another agency, it would typically "notch" the rating (downgrade it one grade).⁴ So "duplication of effort" was a major source of business.

Of course, "duplication of effort" did not occur as far as due diligence was concerned. Instead, a few data inputs were used to assess the base assets. Ratings did not correlate well with rates of default or with subsequent downgrades, even by the limited metrics imposed by the agencies themselves. But CDOs were supposed to incorporate the broader, systemic risk of loss; this, the CRAs were unable to do.⁵

In the event, the CRAs' own staffs were never able to agree on a single coherent notion of what risk was.

EXAMINATION OF THE UNDERWRITERS

The underwriters were investment banks that created CDOs . By far the largest ABS CDO underwriter was Merrill Lynch, followed at a distance by Citigroup.⁶ Soon the underwriters ran out of asset-backed securities and began to incorporate assets consisting of tranches from earlier CDOs. Merrill Lynch, for example, bought 32% of its own RMBS and CDO issues for CDO repackaging.⁷

Underwriters made the situation worse by carrying the super-senior tranches in off-book special purpose vehicles (SPVs) , justified to regulators by hedging--i.e., purchases of credit default swaps (CDS). Merrill Lynch was thus obligated to write down USD 51.2 billion in the wake of the crisis.⁸

However, there was immense difference in performance of CDOs depending on the originator. Goldman Sachs performed comparatively well, with "only" 10% default compared to 40% for JP Morgan.

Hypotheses

Barnett-Hart lists several hypotheses she tests using a combination of regression analysis and probit analysis (p.36).

• Hypothesis 1A ("The Housing Effect"): Increasing exposure to residential mortgages, specifically subprime and Alt-A RMBS, is associated with worse CDO performance as measured by defaults.
• Hypothesis 1B ("The Vintage Effect"): Increasing exposure to 2006 and 2007 vintage collateral, particularly assets with floating interest rates, is associated with worse CDO performance as measured by defaults.
• Hypothesis 1C ("The Complexity Effect"): Increasing the amount of synthetic collateral, the amount of pre-securitized CDO collateral, and the overall number of collateral assets is associated with worse CDO performance as measured by defaults.
• Hypothesis 2A ("The Underwriter Effect"): Holding constant general CDO characteristics, CDO performance varies based on the underwriting bank.
• Hypothesis 2B ("The Size [of the underwriter's CDO business] Effect"): The performance of an underwriter’s CDOs varies according to the size of their CDO business, with overly-aggressive or very inexperienced banks issuing worse CDOs, as measured by their ex-post defaults and rating downgrades.
• Hypothesis 2C ("The Originator [of the underlying collateral] Effect"): Controlling for the type of mortgages issued, as measured by average FICO, CLTV, and DTI scores, the performance of a CDO depends on the specific entities that originated its collateral assets; in other words, was CDO performance affected by the emergence of banks that acted as both CDO underwriters and collateral originators?
• Hypothesis 2D ("The Asymmetric Information Effect"): CDO performance will be affected if it contains collateral originated by its underwriter, although the performance might improve or decline, depending on the importance of reputation vs. adverse selection and moral hazard.
• Hypothesis 3A ("Recycled Ratings Effect"): The most important factor in explaining initial levels of AAA given to a CDO are the credit ratings of their collateral pool.
• Hypothesis 3B ("The Peer Pressure Effect"): The % of AAA given to a CDO will depend on the number of rating agencies rating the deal.
• Hypothesis 3C ("The Seniority Effect"): Controlling for the default rate of the CDO collateral, senior tranches have experienced more severe downgrades.
• Hypothesis 3D ("The Asset-Class Effect"): The realized defaults associated with a given credit grade varies based on the asset type. Similar to Hypothesis 1A, except that here we're interested in the effect on default likelihood for a given CRA rating applied to CMBS versus RMBS or HEL.
  
• Hypothesis 3E ("The Super-Senior Effect" : Rating agencies were overly optimistic in giving AAA ratings. CDOs given more initial AAA ratings, in terms of number of AAA tranches and percent of the transaction rated AAA, are now exposed to larger losses. The name for this hypothesis was apparently inspired by Janet Tavakoli (2005), in which she cites the impact on the actual risk of a waterfalled security of a super-tranche. CRAs did not specifically account for the additional risk posed to the smaller AAA tranche by having a larger share of the CDO's liabilities be senior to it.
  
• Hypothesis 3F ("Conflict of Interest"): Conflicts of interest caused by the fee system of credit ratings would result in more aggressive initial ratings, subsequently more downgrades, and worse accuracy in prediction for the CDOs of large underwriters. Barnett-Hart tested this hypothesis by comparing the amount of business the CRAs did with each underwriter with the number of tranche-rating downgrades. A very large downgrade meant a larger favorable bias toward the underwriter, which was (in turn) tested against the volume of business the CRA did for that particular underwriter.
  

In total, thirteen hypotheses about the factors that most severely stimulated the crisis. These hypotheses seem well-chosen because they test market failure under conditions that most closely match perfect competition.

RESULTS OF THE STUDY

About half of the variation in ABS CDO performance was the result of CDO asset and liability properties (Hypotheses 1A-1C). In particular, regressions of asset and liability properties explained almost 60% of the downgrades in ratings for individual tranches of CDOs. The results are summarized on p.91.

Most hypotheses were validated, although it takes a close reading of the result tables to say how validated each one was. CDOs had a strong likelihood of failing if they were created in 2006-2007, were issued backed mostly by residential mortgage securities (as opposed to commercial mortgages , etc.),⁹ and were complex.¹⁰ This is to be expected; narratives of financial crises should rely on failed business models or regulatory systems. Narratives that "expose" a single person or institution as wickedly inflicting the disaster on a hapless financial system are unsatisfying.

Hypotheses 2A-2D dealt with the underwriter contributions to CDO performance, and explained less of the variance in CDO performance. Nonetheless, they were still interesting to economists. Hypothesis 2A involved a simple ranking of underwriters by CDO performance; effect of [collateral] origination and assymetric information (i.e., hypotheses 2C & 2D) was ambiguous. Some collateral originators performed worse than the rest, but mostly the asset composition was a stronger explanation (2C). Likewise, in some cases it helped that the originator of CDOs was a large player, but aggressiveness in growing structured finance departments was not a good sign (2B); and there was considerable variation from underwriter to underwriter as to the effect on using one's own collateral in a CDO tranche (2D).¹¹ These results arguably suggest that underwriters were eclipsed in importance by the fundamental business model applied to CDOs and the strategic position of the particular underwriters. For example, Goldman Sachs was in a peculiarly advantageous position with respect to the underwriters and was not obligated to pursue CDO business aggressively.

(It needs to be added that table 7 summarizing the ranking of CDO underwriters based on (a) CDO default and (b) CDO tranche rating downgrade yielded entirely different results. Goldman's CDOs had the lowest frequency of default, followed by Lehman; but Goldman was ranked 11th for ratings downgrades.)

The hypotheses 3A through 3F pertained to corruption of CRA ratings per se. Here, the hypotheses behaved very differently indeed from what was expected. While attempting to confirm Hypothesis 3F ("Conflict of Interest"), for example, Barnett-Hart ran up against the problem that the ranking of underwriters by ratings demand was the same for all three CRAs. Moreover, the largest underwriters often were the worst. To make matters more ambiguous, the real problem was that there was too little variation in ratings, not to much. Using jargon from behavioral economics, the problem was signals compression: the difference in real quality of securities, from T-bills to Countrywide CES RMBS was huge, and the ratings ought to have reflected this. Instead, the ratings reflected very small differences across the range of underlying reality.

One regression, for example, ranks underwriters by the accuracy of their ratings (table 14, panel B, p.89). The twist, here, is that the ranking reflect the success of the CRAs to make accurate judgments of the underwriters being ranked. It happens to closely match the ranking of underwriters by performance of CDO (where higher ranking means fewer defaults). In effect, the CRAs were "righter" about the best underwriters, and "wronger" about the worst.

Barnett-Hart concludes:

The errors of the rating agencies stemmed from neither conflicts of interest nor preferential treatment given to certain banks. The true culprit behind the rating agencies’ failure was the outsourcing of credit analysis to computer models and the low level of human input used to rate CDOs (p.94).

In a later post I want to address broader conclusions regarding the CDO market meltdown.

(Part4)

---

Notes

1. The term "AAA" is used by Standard & Poor's and Fitch; Moody's uses "Aaa." Here et alibi, "AAA" = "Aaa."


2. Tavakoli (2005), p.9; via Barnett-Hart (2009). Barnett-Hart uses the page numbers from the PDF file; here, page numbers are from the issue of the original periodical.
   


3. From exhibit , Senate Permanent Subcommittee on Investigations, April 2010, p.18.
   

   Moody’s gross revenues from RMBS and CDOs increased from just over $61 million in 2002 to over $208 million in 2006. S&P's net annual revenues from ratings nearly doubled from $517 million in 2002, to $1.16 billion in 2007. During that same period, the structured finance group's revenues tripled from $184 million in 2002, to $561 million in 2007. In 2002, structured finance contributed 36 percent to S&P’s bottom line; in 2007, it contributed 48 percent – nearly half of all S&P revenues. In addition, from 2000 to 2007, operating margins at the CRAs averaged 53 percent, far outpacing companies like Exxon and Microsoft, which had margins of 17 and 36 percent respectively in 2007.

   This information added in an update to the original post.


4. Barnett-Hart (2009), p.20


5. Tavakoli (2005), p.10.

   One would think that rating agencies would at least be inter- nally consistent. But that isn’t necessarily true. Even within the same rating agency, portfolio tests and restrictions may vary by deal, and some deals are better protected than others. Different structurers within a rating agency may choose different stress scenarios when evaluating cash flows for an ultimate rating.
   
   [....]
   
   In 2004, Fitch’s model showed such unreliable results for structurers using Fitch’s fre- quently changing correlation matrix that industry observers dubbed it the "Fitch Random Ratings Model."

   With respect to the broader measure of systemic risk: " Conventional general market risk to a portfolio is not captured by ratings." (p.9)


6. Barnett-Hart (2009), p.26. Here is the complete list.
   
   

   This table presents the number of ABS CDO deals underwritten by the top 10 underwriters between 2002-2007. The data were obtained from S&P’s CDO Interface.
   Underwriter	2002	2003	2004	2005	2006	2007	TOTAL
   Merrill Lynch	0	3	20	22	33	18	107
   Citigroup	3	7	13	14	27	14	80
   Credit Suisse	10	7	8	9	14	6	64
   Goldman Sachs	3	2	6	17	24	7	62
   Bear Stearns	5	2	5	13	11	15	60
   Wachovia	5	6	9	16	11	5	52
   Deutsche Bank	6	3	7	10	16	5	50
   UBS	5	2	5	10	16	6	46
   Lehman Brothers	3	4	3	6	5	6	35
   Bank of America	2	2	4	9	10	2	32
   TOTAL DEALS	47	44	101	153	217	135	697



7. Ibid., p.27. Repackaging of CDO securities into a new CDO resulted in a "CDO-squared." The Royal Bank of Scotland did this an average of 5.32 x for its CDO securities. Merrill Lynch did it 4.79x. These were extremes, but Merrill Lynch was the biggest underwriter.


8. Ibid., p.32. CDS is supposed to be a form of insurance for default (rather than a proper swap), but almost immediately became very popular as a way for non-holders of CDO securities, like John Paulson, to bet against the housing market overall. The counterparties in the CDS transactions received a premium when defaults did not occur, but when they did, suffered enormous losses that required government bailouts to prevent a general run of CDO securities. This was problematic since so much of the money used to bail out the financial system went to pay off speculators like Paulson.


9. This hypothesis is somewhat problematic because such a large share of ABS CDOs were based on housing collateral. Remember, this category excludes the larger pool of synthetic CDOs as well as a smaller group of CDOs based on corporate bonds. RBMS accounted for about 14% of all ABS deals in 2006-2007 (Barnett-Hart, p.9, table 2) , but home equity loans (HEL) were more radioactive still and accounted for another 34% of deals during that period. Of the remaining 52%, another 11% were pre-existing CDOs (which incorporated a slightly different share of HEL & RMBS; see next footnote below). Only about 40% of the total were commercial mortgages or "other": credit card debt, car loans, and so on.


10. Complexity is measured by the coefficients on Number of Assets, % Synthetic, and % CDO. Please note that CDOs were a non-housing asset and "% synthetic" impinges on the 40% of ABS CDOs that were neither housing nor CDO. As it happens, synthetic CDOs performed relatively better than ABS CDOs--they failed at about a fifth the rate of ABS CDOs (See Part 2, Ftnt 4 & 5).
11. For a description of the findings:
    1. Hypothesis 2A, p. 56 & table 7 (p.63, panel A);
    2. 2B p.57 & table 7 (p.64, panels B.1 & B.2)
    3. 2C: p.58 & table 8 (p.63)
    4. 2D: p.59 & table 8 (p.65)

*

---

Sources & Additional Reading

Efraim Benmelech & Jennifer Dlugosz, "The Credit Rating Crisis" National Bureau of Economic Research (2009)

Anna Katherine Barnett-Hart, "The Story of the CDO Market Meltdown: an Empirical Analysis" Thesis, Harvard University (2009)

Janet Tavakoli, "Structured Finance: Rating the Rating Agencies " , Global Association of Risk Professionals, Issue 22 (Jan/Feb 2005), p.9; via Barnett-Hart (2009), p.20. Barnett-Hart uses the page numbers from the PDF file; here, page numbers are from the issue of the original periodical.

Dr. Michael Wang, Shwn Meei Lee, & Dr. John Ku, "Risks and Risk Management of Collateralized Debt Obligations" (February 2009)

Yves Smith, "The Role of CDOs in Merrill’s Losses (Updated and Expanded Version)," Naked Capitalism (24 October 2007)

Global CDO Issuance, SIFMA (Excel spreadsheet). Outstanding source on CDO statistics.

CDO Meltdown (2)

(Part 1)

The single most valuable resource on theCDO meltdown is Barnett-Hart (2009); this was a thesis paper submitted for a BA at Harvard University, which received a lot of attention thanks to a favorable cite by Michael Lewis.¹ Fittingly, the thesis has won many honors, and it's pretty impressive to read.

A few points introductory to this essay; first, I'm not an expert and this post is my collection of notes on what I regard as helpful or trustworthy sources. The reason why I am writing about this topic is that it's very important to understanding the financial crisis and its concomitant economic catastrophe. Readers are almost certainly not going to share my political views of the matter, but I think some grasp of the mechanics of the crisis will help out in forming their own.

Second, readers interested in an explanation of CDOs are advised to read my first installment or, better still, the first 27 pages of Barnett-Hart's paper.

Additional Background from Barnett-Hart

Most of the CDO market growth took place between 2002 and 2007; a major driving force was the fact that real interest rates were abnormally low, and institutional investors like pension fund managers were required to invest only in AAA-bond assets. CDOs used the waterfall structure to create putatively AAA tranches from pools of assets of far lower rating. In fact, this would have worked except for the fact that the asset pools were (a) not successfully or adequately diversified (i.e.,the assets in the pool were prone to defaulting all at once), and (b) composed of loans of unprecedentedly poor grade (i.e., the loans were made to borrowers who were almost certain to default).

Another reason was that bank holding companies in the USA were now underwriting securities as well as re-lending deposits as loans. Banks could increase their leverage-to-equity ratio by unloading assets to a special purpose entity (SPE; also known as a structured investment vehicle, or SIV), then harvest the returns from the SPE. In a few years, an enormous share of assets and liabilities of the US banking system (including US subsidiaries of foreign banks) were parked in this shadow banking system.

The cash "freed up" by "selling" loans to off-balance sheet zombies (Barnett-Hart called them "brain dead," p.5) was then lent out again. This ruse allowed a large increase in the supply of loanable funds available to the banking system, and spurred a huge boom in housing prices.

About 13% of CDOs issued were created to vacate the balance sheet of a bank.²

The CDOs were exceptionally complex, large entities; they tended to be just under a billion dollars in size, were divided into 7-8 tranches, which usually included one unrated tranche.³ As they became more popular, structured finance came to be the preferred form of collateral (synthetic CDOs).⁴


CDO performance in the Crisis was highly varied. Quarter of issue, asset type, and underwriter (e.g., Goldman Sachs versus J.P. Morgan) each had an effect that was statistically significant. Bonds incorporated in CDOs generally performed much worse than those that were not (Barnett-Hart, p.12, figure 3). After 2005, synthetic CDOs became the main part of the story, but as of this writing, they remain a comparatively small part of the CDO crisis.⁵

A Digression on Subordination

A very important part of the CDO boom was the role of ratings and the enhanced return relative to ratings that CDOs offered to investors. CDOs were popular with investors precisely because they were rated much higher than the underlying securities taken separately would have been, had they been rated accurately. Credit rating agencies (CRAs) such as Fitch, Moody's, and Standard & Poor's assumed that the underlying securities were pooled and subordinated in a way that prevented serious losses to investors even if the CRA ratings were excessively optimistic.

One of the indices of "safety" (or unlikelihood of failure) was loan subordination. A typical bank borrows money short-term from depositors and lends it long-term to borrowers. In order to make a return on its capital, it must lend the money at a higher rate than it pays out to depositors. If this spread is large, then the bank can increase its reserves and its capital. In either case, the bank's excess of loan revenues over expenditures (i.e., interest paid in on loans less interest paid out on accounts) protects the depositors from default by borrowers. "Credit enhancement" includes this and other ways of reducing the likely cost of default: [putatively] excessive collateral, and credit default swaps (CDS), or "wrapped securities" (securities insured against loss by a third party) are other ways.

Ratings agencies use a complex formula to pool their estimate of the value of these guarantees; the subordination is supposed to reflect the excess of cash flow over obligations as a percentage, with each tranche in the CDO possessing a different subordination value.

The Effect of Ratings

Earlier, I mentioned that super-senior tranches were supposedly better than AAA, meaning that the waterfall agreement used in CDOs ensured that even market anomalies capable of hitting a AAA security were unlikely to impact a super-senior tranche. This was reflected in the subordination levels for super-seniors: as late as 2007, they were estimated at 22%, compared to <15% for AAA, <10% for AA, and <5% for BBB (Barnett-Hart, p.15, fig.6).⁶

The ratings agencies did not know how to cope with the huge demand for data that investors (and regulators were now putting on them). The financial system was trained to measure performance by returns relative to risk. High risk investments were supposed to have high rates of return; it wasn't especially impressive if they did. With CDOs, investment houses could churn out high rates of return on low risk investments, with AAA rating. This was the essence of "alpha," or performance of a portfolio adjusted for risk.⁷

The great majority of CDOs were issued in order to arbitrage the favorable interest and risk imparted by the CDO structure itself. Hence, the CRAs were major instigators of the CDO mania. Barnett-Hart goes so far as refer to the CRAs as "manufacturing AAA CDO securities from collateral with much lower ratings" (p.23), which certainly fits my understanding of the situation.

---

Sources & Additional Reading

1. Peter Lattman "Michael Lewis’s The Big Short? Read the Harvard Thesis Instead!" Deal Journal [blog], The Wall Street Journal (15 March 2010)


2. SIFMA spreadsheet (accessed for this post), in worksheet "CDO Purpose." Refers to period 2005-2010, during which 13% of the USD 1.3 trillion in CDO par value issued was for shadow banking. The remaining USSD 1.1 trillion was used for arbitraging the interest premium offered by putatively AAA/AA securities.


3. Barnett-Hart's paper examines 735 CDO transactions with an average value of $829 million each (pp.7-8). This accounts for about half the USD-denominated CDO transactions issued between 1999 and 2007. During this period, USD-denominated transactions accounted for about three-quarters of the total (SIFMA spreadsheet, "Denomination" tab). Unfortunately, the SIFMA data does not cover CDOs before 2000, but this was a minor year in terms of the total volume. After 2008, total CDO issuance plummeted to a tiny fraction of its 2005-2007 rate, especially in the USA.


4. According to the SIFMA spreadsheet ("Collateral" tab), structured finance was the collateral in 1% of CDO issuance for 2001; it shot up to 63% in 2005, accounting for 91% of the growth of issuance during this period. The reason, naturally, was a shortage of the other types of collateral. Still, the CDOs included as collateral about a half-trillion USD in "high-yield loans."
   
   Barnett-Hart's paper focuses on asset-backed security (ABS) CDOs, rather than structured finance. By February 2009, about half of these deals had been written off ("Half of all CDOs of ABS failed," Financial Times, 11 February 2009, via Naked Capitalism). This accounted for about USD 105 billion in defaults, out of a total of USD 815 billion (as of 9 Feb 2009; see "Banks’Subprime Market-Related Losses Top $815 Billion," Bloomberg). So basically about a third of ABS CDO failed, and these accounted for 13% of losses in the 2007-2008 Financial Crisis.


5. An additional USD 24 billion in losses was attributed to synthetic CDOs ("Warning over CDO losses if CIT defaults," Financial Times-14 July 2009). Since our estimate for synthetics comes five months later than the one for total losses and ABS-based CDOs, it's reasonable to assume the figures for the latter two were higher and synthetics performed much better. If not, ABS-based CDOs ($105 B) and synthetics ($24 B) lost $129 B of the total $815 B lost by the top 100 financial institutions worldwide in the Crisis, or about a sixth of the total. Moreover, ABS CDOs losses accounted for 81% of CDO losses overall.
   
   In view of the fact that over half of CDOs issued during the problem period (2005-2007) were synthetic CDOs, this suggests that part of the design principle of the CDO actually worked.


6. For a prescient critique of the methods used by the CRAs to compute debt subordination, see Xudong An, Yongheng Deng, & Tony Sanders, "Subordination Level as a Predictor of Credit Risk" Real Estate & Urban Analysis, University of Cambridge (April 2006).


7. "Alpha" refers to the rate of return on an investment (or the performance of its manager) relative to the rest of the market. If the manager invests in high risk securities, the return on the portfolio will be likely be higher than otherwise, and such investors will be compensated for greater risk; but higher risk also includes a higher risk of unpleasant surprise. For this reason, people who rate portfolio managers measure "alpha" to compensate for things like the manager's risk preference and the overall behavior of the securities market.
   
   The effect of high ratings on CDO tranches was to permit managers to achieve the illusion of high alphas because they were getting rates of return that exceeded what could have been expected, given the low risk of their portfolios.
   

---

Sources & Additional Reading

Efraim Benmelech & Jennifer Dlugosz, "The Credit Rating Crisis" National Bureau of Economic Research (2009)

Anna Katherine Barnett-Hart, "The Story of the CDO Market Meltdown: an Empirical Analysis" Thesis, Harvard University (2009)

Dr. Michael Wang, Shwn Meei Lee, & Dr. John Ku, "Risks and Risk Management of Collateralized Debt Obligations" (February 2009)
Yves Smith, "The Role of CDOs in Merrill’s Losses (Updated and Expanded Version)," Naked Capitalism (24 October 2007)

Global CDO Issuance, SIFMA (Excel spreadsheet). Outstanding source on CDO statistics.

CDO Meltdown (1)

By now, the collateralized debt obligation is surely a household word. The technical nuances of CDOs remain obscure, but the essential problem here was that financial engineers believed that, by using past data they could quantify risk of default, and then pool that risk. Pooling risk, under certain circumstances--explained below--reduces risk to manageable levels.

Those circumstances include two crucial conditions: the risk being pooled must absolutely have zero correlation. Very small correlation is all right--it's not the "zeroness" that's important, it's the certitude that it is low under all circumstances. The other condition, which is really just a reiteration of the first, is that the actions of the financial engineer must assuredly have no impact whatever on the risk itself.

The first condition, in other words, means that an event that affects all the things at risk (e.g., an earthquake hitting a large ratio of properties insured by a particular firm) must never happen. The second condition means that the act of creating an insurance pool, or financial instrument, can never actually influence the risk of the parties themselves (taken as individuals). If the invention of insurance sharply alters the behavior of the insured, then actuarial data is inherently inaccurate.¹

As is now well established, the CDO made credit extremely cheap. This was done by producing an estimate of risk that could be prepackaged into a tradeable security, bundled using a structured investment vehicle (SIV), and sold with a mathematically-inferred projection of return. While credit extended to high-risk borrowers has a high probability of default, this probability was supposedly offset by pooling (with other, lower-risk borrowers) to achieve a suitable investment-grade bond.

THE CDO: A SHORT DESCRIPTION

A collateralized debt obligation (CDO) is a type of structured finance product, in which the originator creates an entity like a special purpose vehicle (SIV) to own loans and distribute payments on certificates. The CDO represents a claim many different potential kinds of assets:

• investment grade and high-yield corporate bonds;
• emerging market bonds;
• residential mortgage-backed securities (RMBS);
• commercial mortgage-backed securities (CMBS);
• real-estate investment trusts (REIT) debt;
• bank loans;
• special-situation loans and distressed debt.

The first CDOs were used for repackaging high-risk corporate bonds and convertible bonds, but later developed into an entire parallel banking system.^2,3

CDOs are generally classified as "cash" or "synthetic." A cash CDO consists of a pool of bonds or loans. Their asset value is tied to the cash flow of the underlying assets, which consists of payments on the principle and interest.

Synthetic CDOs sell credit protection via credit default swaps (CDS) rather than debt-based assets. The CDO's asset value is pegged to specific tradeable assets, but it emulates the capital gains of the assets with derivatives (rather than ownership); hence, it is more highly leveraged. The collateral comes from a cash deposit by the depositor/investor.

The CDO pays out returns at maturity to investors at different levels of priority. In the event that there is widespread default, some of the tranches are guaranteed first, second, or third priority of repayment. The highest-risk [equity] tranches, with a high nominal rate of return, take any hits in the event of default. This is called a "waterfall structure."

CDO asset pools were quite large; typically, they involved nominal values of about USD 500 million to >USD 5 billion.⁴

RATINGS

Click for larger image

A crucial aspect of the CDO boom was the role of the ratings agencies, Moody's, Fitch, and Standard & Poor's. These agencies are quite old, but their business model has evolved over the last 40 years. Originally they were paid to rate securities by investors, who paid for their ratings. This business model dried up and was superseded by a new one in which the issuers of securities paid for the ratings.

There was an obvious conflict of interest in this arrangement, but structured finance and tranching made it worse. That's because the ratings for CDO issues were extremely important to the creation of a shadow banking system, in which liabilities could be parked off bank balance sheets, without the usual reserves or capital adequacy; and because ratings were so important to the design of CDOs.

Because of what a CDO is, it requires the ratings agencies to collude in the design and ultimate strategy of CDO-style structured finance. The tranches are rated separately; in theory, the superior senior tranch of 2006-vintage synthetic CDOs were actually better than AAA, since they had AAA-assets as collateral.⁵

COLLATERALIZATION

A critical part of the narrative was collateralization. A debenture is a debt instrument (like a bond) that is not backed by any collateral. Most bonds are debentures; CDOs supposedly would be safer than debentures because the creditors could, in the event of default, auction off the collateral. As we now understand, this collateralization was to allow the crisis to spill over more readily into the real economy.⁶ This was because CDOs created a bubble in housing prices (millions of people could borrow more than before, and bid up the price of housing--see Warren & Tyagi, 2004)

The effect of collateralization was most obviously to create not only a huge gap between the bubble price and the bust price of houses; it also flooded the rest of the credit markets with loanable funds for consumption. Arguably, this is a strategy that is not readily available for economies whose currency is not the universal reserve currency. That's because rapid expansion of the money supply is liable to lead to capital flight (interest rates fall below a competitive level). But it has a disproportionate effect on consumption, since entrepreneurs are guided by other considerations besides cheap loanable funds. This may explain why the US has such a large and persistent trade surplus. In any event, the CDO meltdown has led policymakers to wonder how the financial system could remain both innovative and stable.

(Part 2)

---

Notes

1. There is evidence it has a modest impact on the insured, but over time this impact has been statistically neutralized. See my post on "The Curmudgeon's Fallacy." Actuarial data, whether for insurance companies or financial instruments, seeks a stable estimate of risk so that the potential for loss is accurately priced. If the methods of assessing loss potential produce results that change frequently over time, then those methods are not useful for managing the probable costs of loss.


2. List of assets from Fabozzi, Davis, & Choudhry (2006), p.119


3. For a summary of the "parallel banking system," see Tobias Adrian & Hyun Song Shin, "The Shadow Banking System: Implications for Financial Regulation" , Federal Reserve Bank of New York Staff Reports, no. 382 (July 2009)


4. The largest CDO I was able to locate was MAX 2008-1 A1, with a notional value of USD 5.4 billion; it was underwritten by Deutsche Bank, which retained 94% equity in it. See Yves Smith, "Debunking Some AIG/Fed/CDO Theories," Naked Capitalism (4 Feb 2010). Smith cites disclosures of Maiden Lane III LLC transactions by Federal Reserve.


5. "A 'Rational' Explanation of the Financial Crisis" Macroeconomic Resilience (November 2009)
   


6. The real economy is distinguished from the financial sector. Usually there is some insulation between the real economy of goods and services, and the "not-so-real" economy of stocks and derivatives.
   

---

Additional Reading

Anna Katherine Barnett-Hart, "The Story of the CDO Market Meltdown: an Empirical Analysis" Thesis, Harvard Kennedy School of Government (March 2009)

Frank J. Fabozzi, Henry A. Davis, & Moorad Choudhry, Introduction to Structured Finance, John Wiley and Sons (2006)

Michael S. Gibson, "Understanding the Risk of Synthetic CDOs" Trading Risk Analysis Section, Division of Research and Statistics, Federal Reserve Board (July 2004)

Elizabeth Warren & Amelia Warren Tyagi, The Two-Income Trap: Why Middle Class Parents are Going Broke, Basic Books (2004). This book provides an invaluable explanation of how financial institutions and the two-income family combined to create a bidding war for houses in attractive neighborhoods. Warren & Tyagi wrote before the 2008 Financial Crisis but described a growing emergency in household finance that had burgeoned during the 1990's.

Credit Write-downs: Creditflux's list of credit write-downs announced since the start of the subprime crisis. Creditflux write-downs (Microsoft Excel Spreadsheet, 78 kb)

"Global Cash Flow and Synthetic CDO Criteria" Standard & Poor's Structured Finance

Structured Finance

Disclaimer: the author is not an expert on this topic and is merely taking notes.

"Structured Finance" refers to a class of financial products that involve structured investment vehicles (SIVs).

A SHORT DESCRIPTION OF STRUCTURED INVESTMENT VEHICLES

Usually a structured investment vehicle (SIV) is a type of legal entity called an "express trust."¹ Trusts are famously used by wealthy people to manage assets that they want to pass on to their children, but they were also used in the latter 19th century to pool the interests of shareholders in multiple corporations--allowing many competing firms to become more profitable by acting as a monopoly. Hence, the efforts of Pres. Theodore Roosevelt to restore competitive markets were known as "antitrust actions" or "trust-busting."

Express trusts are also used to "own" securities and apportion payouts according to a legal formula. In the example in the footnote, the trust was created to own loans. It would then use the proceeds to make payments on the loan certificates (which it gave to ACE Securities Corp in exchange for loans it already had).

Not all SIVs are trusts; some are more complex business entities, such as corporations or limited partnerships.²

SIVs are used to own financial assets with different risk levels, pooling the returns as a hedge against the risk of any single asset class. They are also used to make the trustees' actions very difficult to track, such as when the top investment banks in the USA unloaded toxic assets on their own customers, or when the trustees want to avoid taxes/regulations.

The SIVs exist to own long-term commercial paper, including residential mortgage-backed securities (RMBS), and finance the long-term lending with short-term borrowing (such as repurchase agreements). Short-term borrowing is cheaper than long-term borrowing, except for the stress associated with rolling over debt every few days.

Typically SIVs are organized in tax havens such as the Cayman Islands, Liechtenstein, or Channel Islands. They can issue debt or equity (shares of ownership); they are effectively regulated by the ratings agencies (Fitch, Moody's, Standard & Poor), since they are otherwise beyond the reach of any direct regulation.

STRUCTURED FINANCE

The follow attributes are common to structured finance:

1. a complex financial transaction that may involve actual or synthetic transfer of assets or risk exposure, aimed at achieving certain accounting, regulatory, and/or tax objectives;
2. a transaction ring-fenced in its own SIV;
3. a bond issue that is asset-backed and/or external reference index-linked;
4. a combination of interest rate and credit derivatives;
5. a transaction employed by banks, other financial institutions, and corporations as a source of funding and/or favorable capital, tax, and accounting treatment;
6. disintermediation between banks and other corporate entities.

Financial transactions employing several of these features are generally characterized as structured finance.³

A typical form of complex security that does not involve structured finance would be an indexed fund. In this case, the fund is actually an account in a financial center that takes the money of investors and uses it to buy shares of companies listed in an index (say, the S&P 500) at a fixed ratio. Movement of individual shares in the companies listed are "passed through" to the investors in the proportion that the shares have in the index.

Click for larger image

Tranching refers to the creation of different tiers, or ranks of priority of repayment in the event of default. When a borrower can no longer meet mortgage payments, the pool to which that mortgage belongs loses value by some amount. Holders of super-senior tranches are first in line to get reimbursed in the event of default, followed by senior, mezzanine, and equity. In exchange for accepting a lower priority of repayment, the lower tranches receive a higher rate of return in the event of non-default.

The benefits of tranching were that they permitted originators to generate securities with risk-return values that met what the market demanded. Even if the average mortgage in a pool had an effective rate of return of 6%, an equity tranche could offer a far higher rate in exchange for the normally-low likelihood of default. In an historic period of exceptionally low real interest rates (see figure) and reduced fears of debt default (because of rising housing prices), tranching made housing finance cheap and attractive.

BIS (2005, p.5) explains the significance of tranching to the whole deal:

Tranching, in turn, contributes to both the complexity and risk properties of structured finance products. Beyond the challenges posed by estimation of the asset pool’s loss distribution, tranching requires detailed, deal-specific documentation to ensure that the desired characteristics, such as the seniority ordering of the various tranches, will be delivered under all plausible scenarios. In addition, complexity may be further increased by the need to account for the involvement of asset managers and other third parties, whose own incentives to act in the interests of some investor classes at the expense of others may need to be balanced.

Unfortunately for the rest of the planet, there was a poor grasp of what "all plausible scenarios" meant. The BIS article goes on to compare tranches with conventional bond portfolios, where the risk of loss in the event of massive nonperformance was comparatively small. In other words, for investors in a bond portfolio, default would lead to loss of some capital, but not a wiping out of "junior" investors. This may have marginally impacted the way the crisis unfolded, as investors had an extremely strong incentive to bail out at any cost.

Structured finance products are generally rated by the major agencies. To the best of my knowledge, ratings agencies are typically based in the USA and subject to the US government's regulatory regime (or lack thereof). According to BIS (2005, p.6), structured finance accounted for 40% of rating agency revenues.⁴

Another feature of structured finance, finally--from which the term "structured" comes--is the use of a carefully chosen pool of assets to achieve the desired rating. Usually, the notion that a security has any value whatever arises from the issuer's ability to pay its obligations. A share of stock's value is determined, ultimately, from the future income stream of the issuing firm: if investors believe the future business prospects of the firm are poor, then investors will not want to own a share of stock, and the price of the share will be lower. Likewise, bond ratings reflect the ability of the borrow to earn enough money to pay off the interest and principle. Structured finance, in contrast, usually features an SIV which doesn't earn income because its sole function is to own assets and distribute capital gains. The allure of ownership arises from the quality of the assets included in the SIV. So parties creating SIVs select assets with a carefully-chosen blend of high-risk, high-yield assets that are in some way hedged against failure.


In one sense, the phenomenon of collateralization was extremely old for lenders. From the very beginning of the business of lending wealth, lenders asked for collateral: if the borrower had some other asset, such as a house, then the lender could seize the collateral if the borrower failed to pay. But the SIVs made collateralization a perpetual motion machine: borrowers possessed as collateral the asset they were buying with the loan, which in turn created the market for the asset.

Moreover, the role of collateral was dramatically different. Whereas before, a lien on the borrower's collateral was a deterrent against default; borrowers reliably hoped that their collateral would always be irrelevant to the actual borrowing and repayment process. Here, the income stream of the originating firm was legally and economically detached from the SIV, so that the ability of the SIV to meet its financial duties was dependent entirely on the asset pool. If those assets stopped performing or lost value (as, for example in the event of a price bubble popping), then the lending involved not only became unsecured, it became non-performing.

GENERAL OBSERVATIONS

While structured finance was an innovation of the 1990s that led rather obviously to the Financial Crisis of 2008, it seems hard to take seriously the idea that it can be legislated out of existence. One reason is that the Financial Crisis did not destroy the power of its principals, the way the Banking Crisis of 1931-1933 did.⁵

However, structured finance remains an insidious force. If the industrialized nations of the world were not going through a political implosion, and if the business managers who had instigated this crisis were subject to some accountability, then I think it is fair to say that they would not be allowed in the future to flout banking regulations though the shadow banking system. Instead, financial intermediaries would be obligated to deal with all of their liabilities on their regular books.

---

Notes

1. For example, see this pooling and servicing agreement (1 Sept 2007) involving ACE Securities Corp, as depositor, Wells Fargo Bank, N.A., as master servicer/securities administrator, Clayton Fixed Income Services Inc., as credit risk manager, and HSBC Bank USA, National Association, as trustee. I use the full legal term "express trust" to distinguish it from far more common uses of the word "trust."


2. Peter Levine (2009) describes SIVs as trusts, and the sample SIV linked in note 1 is an express trust. However, Moorad Choudhry (2007, p.1147) describes SIVs as emerging in 1988 as fully-formed corporations, with a perpetual existence. Indeed, this perpetuity distinguishes CDOs from SIVs, according to Choudhry.


3. Fabozzi, Davis, & Choudhry (2006), p.2. Another definition, from Stefano Gatti, Structured Finance: Techniques, Products and Market, Springer (2005), insists on the importance of the SIV and its near-synonym, the special purpose vehicle (SPV). According to Gatti, the SPV is paramount for fulfilling the aims of the originator ("sponsor").
   
   An alternative definition is from the Bank of International Settlements (BIS):

   Structured finance instruments can be defined through three key characteristics: (1) pooling of assets (either cash-based or synthetically created); (2) tranching of liabilities that are backed by the asset pool (this property differentiates structured finance from traditional "pass-through" securitizations); (3) de-linking of the credit risk of the originator, usually through the use of a finite-lived, standalone special purpose vehicle (SPV).
   BIS (2005), p.5

   
   The above is cited in Fabozzi, Davis, & Choudhry (2006), p.2. Note that the BIS definition emphasizes the permanence of the SPV/SIV.


4. A well-nigh universal observation is that the ratings agencies had a perverse incentive structure; they were paid by the entities seeking a rating. Evidence suggests that there was massive conflict of interest in which Moody's, Fitch, and S&P sought to accommodate companies marketing financial products with favorable ratings. See, for example, Daniel M. Covitz & Paul Harrison, "Testing Conflicts of Interest at Bond Ratings Agencies with Market Anticipation: Evidence that Reputation Incentives Dominate" Federal Reserve Board (2003). As is now well established, the Fed was wrong and the naysayers were right.


5. For a summary of the events of the 1931-33 Banking Crisis, see Charles Poor Kindleberger, The world in Depression, 1929-1939University of California Press (1986). It was actually quite unusual that the Great Depression, unlike the many other financial and economic crises in US history, resulted in a severe blow to the status of financiers and industrialists. One reason was that the crisis of the '30s pitted rival capitalists against each other.
   

---

Additional Reading

Moorad Choudhry, Bank Asset and Liability Management: Strategy, Trading, Analysis, John Wiley and Sons (2007)

Frank J. Fabozzi, Henry A. Davis, & Moorad Choudhry, Introduction to Structured Finance, John Wiley and Sons (2006)

Peter Levine, "Abstract finance: Structured Investment Vehicles and other Tricks," Rethinking Markets Economic Sociology from the Ground Up (blog) (8 Jan 2009)

The Role of Ratings in Structured Finance: Issues and Implications" Committee on the Global Financial System, BIS (2005), p.5