The Design of Subrpime Mortgage Backed Securities and Information Insensitivity Sun Young Parky Yale University

Abstract What is the purpose of securitization and what does its process entail? What exactly transpired in the subprime mortgage-backed securities (MBSs) market during the housing boom until the global …nancial crisis? To answer these questions, I use a novel hand-collected dataset of subprime MBSs issued between 2004 and 2007, which includes a detailed description of underlying mortgage characteristics and deal structures. This paper studies the purpose of the securitization process in the context of the subprime MBS market as a trigger of the global …nancial crisis. First, I …nd that credit enhancements re‡ect the credit risks of underlying collateral, thus ex-ante qualities of AAA subprime MBS tranches had not deteriorated. Second, AAA tranche spreads are largely explained by bond market conditions and are uncorrelated with credit risk of collateral. These results suggest that market participants designed and priced subprime MBSs in a way that the AAA tranche could be information-insensitive. In other words, the securitization process makes it possible that market participants have less incentive to learn about underlying collateral information. (JEL G21, G24, G28)

I am indebted to Gary B. Gorton for his insightful guidance and encouragement. I thank Frank J. Fabozzi and Edward Vytlacil for thoughtful suggestions and kind support. I thank Trivi Dang and Lei Xie for their comments. Any remaining errors are my own. y Department of Economics, Yale University, Box 208268, New Haven, CT 06520-8268. E-mail Address: [email protected]

I.

Introduction

The lending standards and performance of the subprime mortgage sector before the 2007 global …nancial crisis have been widely studied at the loan level (see Keys, Mukherjee, Seru and Vig (2010), Demyanyk and Van Hemert (2010), Dell’Ariccia, Igan, and Laeven (2008)). These studies focus on whether securitization decreases lending standards, especially in the subprime mortgage segment.1 Gorton (2010) noted that the evidence of a decline in lending standards is only a piece of the puzzle, and the argument must be that if this occurs, it was not re‡ected in the structure of the mortgage-backed security (MBS). Lending behavior is the starting point of the securitization process. The lending behavior of subprime originators in the primary mortgage market2 cannot fully explain why one segment of the …xed income market was the trigger of the worst …nancial crisis since the Great Depression.3 For example, suppose more protection for the senior tranche was inserted in the design of subprime MBS. Then it is di¢ cult to conclude that the subprime MBS market would have deteriorated as much as the subprime mortgage market. Therefore, it is important to investigate subprime MBSs themselves, which are the next link from the subprime mortgage to the …nal investor, for understanding the e¤ect of securitization on the …nancial crisis.4 However, with this paper as the exception, I have been unable to …nd any work which systematically and empirically analyzes the evolution of subprime mortgage backed securities from the housing boom to the collapse of the non-agency mortgage-backed security market. This paper poses two questions: (i) what is the securitization process for?, and (ii) what 1

See Gorton (2010), Brunnermeier (2009), and Calomiris (2009) for the complete survey on the global …nancial crisis of 2007-2009. 2 The primary mortgage market activity refers to the activity between the originating bank and the mortgage borrower. And secondary mortgage market activity pertains to the activity related with mortgage backed securities, collateralized debt obligations and other credit derivatives. 3 In the testimony in front of the Financial Crisis Inquiry Commission on September 2, 2010, Bernanke noted: “judged in relation to the size of global …nancial markets, prospective subprime losses were clearly not large enough on their own to account for the magnitude of the crisis.” 4 The next chain is a asset-backed security collateralized debt obligations (ABS CDOs) and I will explore this subject in “Anatomy of ABS CDO: why do ABS CDOs exist?”.

1

happened in the subprime MBS market between the housing boom and the global …nancial crisis? To answer these questions, I use a new hand-collected dataset of subprime MBSs issued between 2004 and 2007, which includes a detailed description of underlying mortgage characteristics and deal structures. In other words, this paper studies the purpose of the securitization process in the context of the subprime MBS market being a trigger of the global …nancial crisis. According to the Bank for International Settlements (2005), securitization is a formation of a structured …nancial instrument which has three key characteristics: pooling of assets, tranching of liabilities, and de-linking the credit risk of assets from the credit risk of an originator by using a special purpose vehicle (SPV). The asset securitization process transforms a pool of assets into one or more securities called asset-backed securities (ABSs). For instance, if the asset pool consists of mortgages, then it is called a mortgage-backed security (MBS). The economic rationale of pooling assets is clear: diversi…cation. Pooling of assets allows ABSs to achieve lower credit risk as a whole by canceling out idiosyncratic shocks among assets. Correspondingly, the motivation for using SPVs is that an issuer of SPVs wants to reduce bankruptcy cost, according to Gorton and Souleles (2005). They argue that a riskier issuer has more incentive to use SPVs since SPVs are bankruptcy remote entities. However, the economic motivation for tranching is not clear. Tranching is a technique to modify or redistribute the risk of the collateral among di¤erent bond classes (tranches) by the use of a structure. As a result, bond classes have di¤erent degrees of priority with respect to both cash in‡ows and loss write-o¤s, and thus it is said that the security has a subordinated structure. According to Gorton and Pennacchi (1990), tranching allows a …nancial intermediary to divide liabilities into two parts: one that is near-riskless, and another that contains most of the default risk. Hence, tranching reduces the overall adverse selection problem, and allows the …nancial intermediary to maximize its proceeds. More generally, it is worthwhile to investigate the purpose of securitization in the con-

2

text of the nature of current …nancial intermediation and the …nancial crisis. According to Nomura (2006), the …rst securitization activity involves mortgage securitization in 1971, by Federal Home Loan Mortgage Corporation (FHLMC), widely known as Freddie Mac. And the Bank of America National Trust & Saving Association became the …rst private sector issuer of MBS in 1977. In the ensuing 30-plus years, the securitization market has experienced exponential growth. At the end of 2006, a year before the credit market interruption, the total amount of structured …nance products outstanding excluding CDOs was $11.86 trillion, while outstanding Treasury and corporate debt totaled $4.87 and $5.34 trillion, respectively. The rise of securitization was accompanied by the growth of a shadow banking system and a sale and repurchase market (repo market). Gorton and Metrick (2009a) refer to the combination of securitization and repo …nance as securitized banking. Adrian and Shin (2010) also refer to the integration of banking with capital market developments as a marketbased …nancial system and note the importance of the shadow banking system, which grew out of securitizing assets. Therefore, an explanation about the purpose of securitization has to encompass the changing nature of …nancial intermediation and the …nancial crisis. According to Dang, Gorton and Holmstrom (2010), a motive for security design is to minimize the incentive of private information acquisition. They argue that the purpose of securitization is to create information-insensitive securities. information-insensitivity means that the securities are immune from adverse selection when trading. Thus, the values of these securities do not depend on the information known only to informed agents. This property makes the information-insensitive security liquid, which is particularly important since securitized products were commonly used as collateral in repo …nancing. The concept of information-insensitivity allows us to understand the expansion of the overall debt market, especially in the securitized product market, and the global …nancial crisis. In his testimony before the Financial Crisis Inquiry Commission (FCIC) on September 2, 2010, Ben Bernanke, Chairman of the US Federal Reserve, noted that the extraordinary withdrawal of funds from the money market by investors around the world during summer of

3

2007 could be linked to a traditional bank run as explained by the study of Gorton (2008b). Gorton argues that the run was triggered by the fact that information-insensitive securities became information sensitive due to the aggregate shock from the declining values in the U.S. housing market. Motivated by their study, I approach securitization as the creation of informationinsensitive securities. And in turn, I reformulate the concept of information-insensitivity into two testable implications for a seller and buyer at the issuance stage of a subprime MBS. First, if an issuer (seller) designs an information-insensitive security from an asset pool, then the credit enhancement mechanisms must re‡ect the credit risk of the asset pool. The idea is that the credit enhancement mechanisms provide su¢ cient protection for the security from expected loss of the underlying mortgages, and thus the issuer can delink the ex-ante credit risk of the asset pool from the security. From this perspective, the security is independent from the credit risk of its collateral to some extent. As mentioned above, one way of achieving credit enhancements is to use a subordinated structure with prioritized claims, which as a byproduct makes the remaining unprotected part of the liabilities information-sensitive. The second implication is the following: when an investor (buyer) prices an informationinsensitive security, he does not take into account the credit risk of collateral. This is possible because the value of an information-insensitive security is independent of the information about the asset pool to some extent. The inverse of this implication leads to another testable hypothesis. That is, when an investor prices an information-sensitive security, he considers the credit risk of collateral. For example, a BBB-rated tranche is exposed to losses ahead of other higher-rated tranches, and thus cash ‡ow to the BBB-rated tranche is closely related to the performance of the underlying asset pool. In this sense, the value of a BBB-rated tranche has to be sensitive to information about the collateral. Since this paper tests these implications in the context of the subprime MBS market, I use the credit rating of each tranche as a measure of information-insensitivity. I classify

4

AAA-rated tranches as information-insensitive and BBB tranches as information-sensitive. AA-rated and A-rated tranches which lie between AAA-rated and BBB-rated, are relatively classi…ed. As the tranche is more junior, it is more information-sensitive. Using these testable implications, my analysis yields three key results. First, I focus on the designing of information-insensitive securities by examining the relationship between the credit enhancement mechanisms in the structure of subprime MBSs and the collateral characteristics as measures of credit risk. And I …nd that credit enhancement mechanisms which include subordination levels to AAA tranches with initial overcollateralization levels and a delinquency test threshold, do re‡ect the credit risk of the collateral. In other words, if the collateral becomes riskier, then there is more protection for the AAA-rated tranche in the deal structure. In addition, I …nd that among other risk measures, the average credit score and LTV (loan-to-value) ratio are the most important determinants of credit enhancements. The above …ndings suggest that the market participants involved in issuing subprime MBSs have made AAA-rated tranche information-insensitive by inserting more protection from the credit risk of collateral. Furthermore, this implies that it is di¢ cult to judge whether the ex-ante qualities of AAA tranches of subprime MBSs deteriorated as much as the subprime mortgage market. Certainly the ex-ante qualities of BBB-rated tranches were exacerbated because the risk measures of their mortgage pools worsened. Next, I focus on the pricing of information-insensitive securities by analyzing the relationship between issue spreads of AAA-rated tranches and their collateral characteristics using a sample of 2,985 tranches issued before August 2007. I show that issue spreads of AAA-rated tranches over 1-month LIBOR (London Interbank O¤ered Rate) are 60% explained by bond market conditions and the weighted average life of the tranche. However, the subordination level of the tranche and the collateral characteristics have no marginal explanatory power of the spread at issuance. In other words, the issue spread is not sensitive to information about the collateral. For a robustness check, I repeat the same analysis with a sample issued after August 2007, which includes 110 tranches. The results show that the

5

spread started re‡ecting the collateral information, but the estimates are not robust, and the overall explanatory power of the spread decreases. This is evidence of the fact that information-insensitive securities became information-sensitive, and the market experienced di¢ culties with pricing. In summary, the second …nding implies that investors did not take account of the collateral information when they priced AAA tranches before August 2007. Note that this is di¤erent from investors’over-reliance on credit ratings when they decide to invest. Finally, related to the …ndings above, I focus on the pricing of information-sensitive securities by analyzing the issue spread of BBB-rated tranches using 1,704 observations. In contrast to AAA-rated tranches, the subordination level of the tranche and collateral characteristics have explanatory power on the issue spread of BBB-rated tranches. Issuer e¤ects also have explanatory power. In other words, the price of a BBB-rated tranche is sensitive to information about its collateral, subordination level and issuer. This implies that market participants acknowledge the riskiness of these securities, and thus there is an incentive to learn about them. However, with the independent variables at hand, I can explain the BBB issue spread by 35-40%. This suggests it is more di¢ cult to price a BBB tranche than an AAA tranche, since there are more uncertainties about cash in‡ows and write-o¤s. This is the main reason that most BBB subprime MBS tranches went through a re-securitization process in order to create information-insensitive securities. The rest of the paper is organized as follows. Section II provides a brief overview of institutional features of the subprime securitization market. Section III describes the data and presents summary statistics. Section IV presents the empirical results. Section VI concludes.

II.

Subprime MBS, Securitization and Information Insensitivity A.

Subprime MBS marekt and Securitization

According to the Bank for International Settlements (2005), 6

Structured …nance instruments can be de…ned 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; (3) de-linking of the credit risk of the collateral asset pool from the credit risk of the originator, usually through use of a …nited-lived, standalone special purpose vehicle (SPV). Securitization is the process that transforms an asset pool into a structured …nance instrument that has the common features described above. Depending on the collateral type, we can divide structured …nance instruments into two groups: mortgage-backed securities (MBS) and asset-backed securities (ABS), although ABS are often used as a broader category that includes MBS.5 And MBS are classi…ed into residential mortgage-backed securities (RMBS) and commercial mortgage-backed securities (CMBS). Asset-backed securities are backed by di¤erent asset classes: home equity loans, manufactured housing, credit card receivables, student loans, and auto loans. And if ABSs are backed by other ABSs, then they are called collateralized debt obligations (CDOs). Moreover, asset-backed commercial paper(ABCP) applies the securitization technique to issue liabilities.6 Securitization started from the mortgage securitization by government sponsored entities (GSEs) in the 1970s. In 1971, Federal Home Loan Mortgage Corporation (FHLMC), known as Freddie Mac, introduced the …rst conventional pass-through certi…cate. From 1971 to 1977, virtually all MBSs were either guaranteed by Government National Mortgage Association (GNMA), known as Ginnie Mae, or issued by FHLMC. Federal National Mortgage Association (FNMA), known as Fannie Mae, purchased mortgage loans to hold in its portfolio but did not actively issue MBS until later. In 1977, Bank of America National Trust & Savings Association became the …rst private sector issuer of MBS (Nomura, 2006). The mortgage-backed security market among GSEs is called an agency market, while the private5

The security backed by subprime mortgages is classi…ed as ABS but I will use subprime MBS since the latter is used more often. 6 At the end of 2006, non-agency MBS outstanding was $2.92 trillions. Agency MBS and collateralized mortgage obligation (CMO) outstanding was $5.71 trillions, ABS outstanding, $2.13 trillion, and ABCP outstanding, $1.11 trillions according to Securities Industry and Financial Markets Association (SIFMA).

7

label MBS market is called a non-agency market. Since the history of securitization started from the MBS market, understanding MBSs allow us to understand securitized products in general, as well as product process. Figure 1 shows the market shares of non-agency and agency securities between 2000 and 2010. Non-agency issues controlled 20% of the MBS market between 2001 and 2003, and then the proportion jumped to 55% by the end of 2006. Non-agency securitization is distinct from agency activities since agency activities have implicit (in the case of Fannie Mae and Freddie Mac) or explicit (in the case of Ginnie Mae) guarantees by the U.S. government. In other words, agency MBSs are treated almost as riskless as Treasuries, thus they do not need any credit enhancements. Non-agency MBSs are classi…ed into three groups depending on the nature of the underlying mortgages: Prime-jumbo, Alt-A, and subprime. A prime jumbo mortgage is a prime mortgage whose size exceeds the limit of GSEs’purchasing requirements. A subprime mortgage is a type of mortgage taken out by individuals with impaired credit histories such as a credit score below 660, or those with no/limited documentation, and/or those making a low down payment; a uniform de…nition of subprime mortgages does not exist. An AltA mortgage, which is short for Alternative-A, is a type of mortgage that is riskier than a prime mortgage, but considered safer than subprime mortgage. As will be seen, there is no clear distinction between Alt-A and subprime mortgages at the mortgage level. However, the MBS level classi…cation is self-reported by the issuer and hence there is no ambiguity. Note that prime, Alt-A, and subprime MBSs adopt di¤erent credit enhancement mechanisms depending on the nature of their underlying mortgages. As can be seen in Figure 2, the subprime MBS market dramatically expanded from 2004. While prime MBS issuance stagnated between 2003 and 2004 at $233.3 billion, Alt-A and subprime MBS issuances nearly increased twofold, from $74.2 billion to $158.6 billion and from $195 billion to $362.5 billion, respectively. As a result, the increase in non-agency MBS market share was lead by the subprime and Alt-A MBS markets. And for the …rst time

8

in the history of securitization, subprime MBS issuance far exceeded prime MBS issuance beginning in 2004 to just prior to the …nancial crisis, and then such issuances completely disappeared.

B.

Securitization, Information Insensitivity and the Financial Crisis

According to Financial Crisis Inquiry Commission (2010b), the shadow banking system refers to bank-like …nancial activities that are conducted outside the traditional banking system, many of which are unregulated or lightly regulated. The institutions in the shadow banking system are investment banks, …nance companies, money market funds, GSEs, hedge funds, and SPVs. These entities are critical players in the markets for securitized products, commercial paper, ABCP, repurchase agreements, and derivatives. Also it is important to note that the institutions in the shadow banking system are those who were the most severely damaged during the …nancial crisis. As mentioned earlier, the trigger of the global …nancial crisis was the deterioration in the subprime mortgage sector due to housing prices having stopped appreciating and information-insensitive securities becoming information-sensitive. I examine these events with my dataset from three perspectives: (i) price movements of existing subprime MBSs, (ii) behavior changes of issuers in 2007, and (iii) evolution of issue spreads over time. First, Figure 3 shows the prices of AAA-rated tranches that are components of ABX.HE.AAA 2006-1.7 I plot seven tranches from January 3, 2006, to November 3, 2009. As can be seen, AAA-rated subprime tranches were traded or priced at par, but started declining after the market disruption in August 2007. This suggests that the value of existing subprime MBSs 7 Trading in the …rst ABX index series started in January 2006. The ABX.HE indices are a liquid, tradable tool allowing investors to take positions on subprime mortgage-backed securities via credit default swap (CDS) contracts. According to Markit (2008), “The ABX.HE Index shall be constituted from reference obligations issued by twenty issuers of residential mortgage-backed securities that meet the criteria speci…ed in these ABX.HE Index Rules. There are six separate indices at benchmark rating levels (AAA, AA, A, BBB, and BBB-). Index prices re‡ect the willingness of investors to buy or sell protection on the basis of their views about the risk of the underlying subprime loans, and are quoted as a percentage of par.” See Gorton (2008a) more on the role of ABX.HE index in the crisis.

9

decreased due to growing uncertainty about subprime mortgage performance. Note the heterogeneous recoveries of price among the di¤erent AAA-rated tranches. The top line represents the price movement of Goldman Sachs Alternative Mortgage Product (GSAMP) 2005-HE4 A2C, and the bottom line presents the price of Argent Securities Inc. (ARSI) 2005-W2 A2C. Table 1 ABOUT HERE Second, the sudden change of market sentiment is visible in the designs and issue spreads of new subprime MBSs. Table 1 represents the issuing behavior of Bear Stearns Asset Backed Securities Trust during 2007. According to the 2009 Mortgage Market Statistical Annual, Bear Stearns placed 6th among the top subprime MBS issuers with 4.3% market share in 2007. There were …ve deals before the market disruption and two afterwards. As can be seen, the spread of AAA-rated tranches rose from 12 basis points to above 100 basis points. Note that before the market interruption, the issue spread of A-rated tranches was 100 basis points. And the subordination level of AAA tranches rose from an average of 25%, to above 35%. Lastly, Figure 4 illustrates the …rst AAA-rated subprime issue spreads during the entire sample period. Issue spreads continuously declined from 2004 to 2006, and variance is relatively small. However, issue spreads jumped after August 2007 which implies that the cost of issuing subprime MBSs started rising.

III.

Data and Summary Statistics A.

Sample Construction

An important contribution of this paper is to present a new dataset for reviewing the evolution of the subprime MBS market between 2004 and 2007. I use the classi…cation of the 2009 Mortgage Market Statistical Annual for collecting the subprime deal list. The Mortgage Bankers Association publishes the Mortgage Market Statistical Annual and reports primary 10

and secondary mortgage market activities. The data in the Annual includes a list of nonagency MBS activity each year and indicates whether the deal is subprime, alt-A, or prime. Thus, the observation unit in this paper is the deal and the tranche level. Using the subprime MBS deal list, I collect deal, tranche, and collateral information from Bloomberg. Bloomberg provides detailed information on each deal from the prospectuses submitted to Securities and Exchange Commission (SEC) and from monthly trustee reports submitted to the trustee of the deal. The information in my dataset can be classi…ed into three categories: (i) underlying collateral characteristics on the issue date, (ii) deal structure, especially credit enhancement mechanisms, and tranche characteristics, and (iii) performance of collateral after issuance. First, the collateral characteristics are the average risk measures of the mortgage pool backing the security. The risk characteristics include the weighted averages of: credit score, LTV ratio, weighted average coupon rate (%), age of loan (months), loan size ($1,000s), percentage of mortgages with limited documentation, percentage of mortgages originated in Southern/Northern California, percentage of …xed rate mortgages (FRMs) and adjustable rate mortgages (ARMs), percentage of owner-occupied or investment properties, percentage of single-family or multi-unit homes, and percentage of loans for the purpose of purchase, re…nance, or equity take-out. See Appendix for de…nitions of each variable. The second category of my dataset includes information about deal structure and tranche level. Deal structure involves the number of tranches in the deal, deal face value, initial overcollateralization level, delinquency test threshold, issue date, servicer, trustee, issuer of the MBS, and the originator of the mortgage, if available. Tranche level information includes issue spread over 1-month LIBOR, tranche face value, weighted average life, credit rating, subordination level, whether the tranche is wrapped by insurance, the insurance provider, whether the tranche is privately placed, and the issue and transaction prices, if available. Third, my dataset also includes the performance of the deal after issuance: the 30,

11

60, and 90-day delinquency rates; the percentage of bankruptcies, real estate owned, and foreclosed properties; the cumulative loss rate; the principal and interest payments to each tranche; and the factor of the deal.8 This information is updated monthly. Table 2 ABOUT HERE Table 2 presents an example of the deal structure and collateral characteristics of Merrill Lynch Investors Trust 2006-HE5 from my data sample.9 Of its 14 tranches in this subordinated structure, one tranche, MLMI 2006-HE5 A2D, is a component of ABX.HE.AAA 2007-1.10 The subordination level measures the percentage of mezzanine and junior tranches, and the overcollateralization level of the deal, which serves to absorb losses before the AAA tranche is a¤ected. This deal totaled $1.324 billion and the amount of subordinated tranches was $246 million, or 17.9% of the deal. Combined with a 4.2% initial overcollateralization level, the subordination level for the AAA tranche was 22.1%. Tranches are usually classi…ed into three groups: the senior class including A1 to A2D, the mezzanine class, from M1 to M6, and the junior class, from B1 to B3. Note that A1 and A2 are backed by di¤erent collateral pools; this structural technique is called cross-collateralization. A1 is backed by conforming loans that conform to the GSEs’ loan limits, while A2 is backed by nonconforming loans. This technique is adapted to achieve an e¢ cient way of implementing credit enhancements. As can be seen, the most junior tranche, B3, has a 4.2% of subordination level, which is the initial overcollateralization level provided by Merrill Lynch Investors Trust. The weighted average life di¤ers among AAA tranches, making it possible to satisfy investors’heterogeneous demand for deal maturity. The issuer can produce securities with di¤erent risk pro…les, risk premiums, and maturities out of an illiquid asset pool –this is one of the economic bene…ts of securitization. 8

I will explore this data for my paper, “Did the Originate-to-Distribute Model contribute to the Financial Crisis? : Evidence from the Performance of Subprime MBSs”. 9 The issuer has to submit the prospectus to SEC when it issues MBS. The following link provides the prospectus of Merrill Lynch Investors Trust 2006-HE6: http://www.secinfo.com/dsvr4.vbPq.htm#1stPage. 10 There are three vintages of ABX.HE; 2006-1, 2006-2 and 2007-1. Due to the market disruption, no vintages were issued afterwards.

12

B.

Deal, Tranche and Collateral Characteristics

Evolution of Qualities of underlying Collateral Characteristics.— Panel A of Table 3 provides summary statistics of average collateral characteristics for the 1,391 deals in the sample. Note that I calculate summary statistics by excluding any missing data. Average credit scores fell by 11 points, from 633 in 2004 to 622 in 2007. The average LTV ratio increased slightly from 76% to 79.5% over the same sample period. Note that the changes to average credit score and LTV ratio are lower in Demyanyk and Van Hemert (2010). For example, in their sample using First American Corelogic’s LoanPerformance dataset, the average credit score fell from 618.3 to 613.2, and the average LTV rose from 82.8% to 83.6% during the same period. Separately, the initial weighted average coupon in my sample increased from 7.3% to 8.6%, and the average loan size rose from $136.3 thousand to $178.1 thousand. The average loan age is fairly constant over time and it shows that it takes six months from loan origination to MBS issuance. See Figures 7-15 for scatter plots and median splines of these variables for an overview of the evolution of the subprime MBS market. The percentage of mortgages with limited documentation in the pool rose from 29.9% to 35.8%. In contrast, the proportion of loans originated in California decreased from 31.5% to 20.1%. Note that the proportion of FRM contracts decreased from 41.8% to 34.1%, which makes it particularly important to evaluate the credit risk of subprime MBSs given interest rate change shocks to mortgagors on the interest rate reset date. In contrast, when we split the mortgage contract by occupancy type or family type, the time variations are relatively small. The percentage of owner occupancy and single-family type mortgages are dominant and the averages are nearly constant during the sample period. We can split the purpose of a mortgage loan into two types: purchase and re…nance. And there are two reasons for re…nancing: equity take-out to capitalize on house price appreciation, or pure re…nance for reducing mortgage interest costs. Note that while mortgages for new home purchases fell from 40.3% to 27.1%, mortgages for equity take-out by re…nancing 13

increased from 48.6% to 62.1%. When evaluating the risk measures (aside from the proportion of mortgages originated in California), subprime MBSs clearly became riskier between 2004 and 2007. However, the deterioration of average collateral characteristics is not dramatic. As can be seen in Figures 7 and 8, to some extent all risk measures were standardized at the issuance stage, which suggests that issuers in the subprime MBS market have target pool characteristics when they construct the mortgage pool for each deal. Standard & Poors (2009) de…nes characteristics of archetypical subprime mortgage pools, and it adjusts its risk assessment for each loan as the loan or borrower characteristics deviate from the archetype. Table 3 ABOUT HERE Change of Deal Structure.— Panel B of Table 3 presents the average structural characteristics by focusing on credit enhancement mechanisms. First, the average subordination level for AAA-rated tranches increased throughout the period, from 16.9% to 23.9%, re‡ecting that the underlying collateral characteristics became riskier. This has implications on the issue amount of AAA-rated subprime tranches. Figure 5 illustrates the average subordination level and proportion of AAA-rated tranches by year. It is evident that the issue amount of AAA-rated tranches in the subprime MBS market monotonically decreased from 83.09% to 76.14% during the sample period. Meanwhile, the initial overcollateralization level increased from 1.7% to 4.6%. Overcollateralization is a type of an issuer-provided credit enhancement, because the issuer transfers an asset pool that has a market value that exceeds the amount paid by the SPV. The amount of the overcollateralization is a form of equity and is equal to the di¤erence between the par value of the assets transferred and the price paid. Also, overcollateralization is the subordination level of the most junior tranche in the deal. Before the junior-most tranche absorbs a loss from any defaults of collateral, losses are taken by the overcollateralization piece. The Dodd-Frank Wall Street Reform and Consumer Protection Act of 2010, which was signed into law on July 21, 2010, includes securitization reform. Federal banking agencies, 14

including the O¢ ce of the Comptroller of the Currency, the Federal Reserve, and the Federal Deposit Insurance Corporation, together with the Securities and Exchange Commission (SEC), are required to jointly promulgate regulations requiring securitizers (issuers) of assetbacked securities and mortgage-backed securities to retain a portion of the credit risk (at least 5%) in securitized assets sold to investors. In other words, the federal banking agencies and SEC require the issuer provide at least 5% of the initial overcollateralization level. After observing that the average initial overcollateralization level was below 5% in the subprime MBS market, securitization reform of risk retention was made binding and consequently has economic e¤ects on the behavior of issuers. This development is among one of the key reasons to investigate the evolution of the subprime MBS market itself. Lastly, the delinquency test threshold decreased from 43.2% to 33.4% over the period. The delinquency test functions as protection and in‡uences payment priority. If the percentage of delinquent loans in a mortgage pool is above the test threshold level, then the mezzanine and junior tranche investors cannot receive principal and interest payments even if it is after the lock-out period. If the deal fails to pass the delinquency test, then all payments go to the senior tranche investor. A lower delinquency test level provides more protection for the senior tranche. By all measures, on average, there is more credit protection inserted into deal structures. The average deal size is calculated among deals that contain a senior AAA tranche. As seen in the Merrill Lynch Investors Trust 2006-HE5 example, it is a common feature to have two collateral groups in one deal; they each support di¤erent tranches. Since Bloomberg reports the deal size from the size of each collateral group, the size of actual deals that contain two collateral groups are larger than the average deal size reported in Table 3. On average, there are 13 tranches in subprime MBSs during the sample period. Change of Credit Spread among Di¤ erent Rating Groups.— Panel C of Table 3 reports the average tranche characteristics by rating groups. I classify the tranche by using S&P credit rating and split tranches into four categories: AAA, AA, A, and BBB.

15

For example, I combine AA+, AA, and AA- rated tranches into AA group for the analysis purpose. Figures 12-15 present issue spread (basis points) over 1 month LIBOR by rating groups over time. As can be seen, the issue spreads had become tighter in all credit rating groups between 2004 and 2006. For example, the average issue spread of AAA rated subprime tranche fell from 31.10 basis points in 2004 to 16.05 basis points in 2006. And the average issue spread of BBB rated subprime tranche decreased from 251.51 basis points to 159.98 basis points over the same period. Together with Figure 6 showing the dramatic market expansion between 2004 and 2006, this implies that the subprime MBS market expansion is driven by demand factor. While the subprime MBS issuance increased, and the price of subprime MBS became expensive since the spreads fell. However, it is not clear what factors lead the expansion of subprime MBS demand and would be a future research question in this literature. Next section of Panel C in Table 3 presents the average subordination level by rating groups. As can be seen, the average subordination level increased in all credit rating groups. Using the average subordination level, it is possible to calculate the proportion of rating groups. For example, there exists 20.83% of subordination level for AAA rated tranche on average in the last column. It means that the proportion of AAA rated tranche is 79.17% (=100%-20.83%). Also the di¤erence between subordination level for AAA rated tranche and AA rated tranche is 7.29% (=20.83%-13.4%) and thus the proportion of AA rated tranche is 7.29%. Note that the …rst two sections of Panel C illustrate that the high risk is compensated with high risk premium.

IV. A.

Empirical Results

Designing Information Insensitive Security

In this section, I investigate the design of an information-insensitive security by focusing on three credit enhancement mechanisms in subprime MBS deals: the subordination level for AAA tranches, the initial overcollateralization level, and the delinquency test threshold. 16

First, I begin by estimating the following baseline speci…cations.

(1)

Subordinationi =

0

+ (Collateral Characteristics)i + ci + dy + "i

where Subordinationi is the % of subordinated amount to the AAA tranche in the ith deal, collateral characteristics include deal average credit score, LTV ratio, weighted average coupon rate, % of ARM, % of mortgages originated in California, whether the tranche is wrapped by insurance, ci is a vector of issuer …xed e¤ects, dy is a vector of year …xed e¤ects, and "i is the regression residual. I report the results from estimating di¤erent speci…cation of regression equation (1). Standard errors reported in parentheses are heteroskedasticity robust and clustered at the issuer level throughout. Table 4 ABOUT HERE Note that for the analysis, I use deals issued before August 2007, resulting in excluding 33 deals. While the …nancial and capital markets had detected problems and deterioration in the subprime mortgage sector from the end of 2006, the real e¤ect on the subprime MBS market including other securitization markets emerged in August 2007. Including MBSs issued after August 2007 hinders precise analysis of the determinants of issue spread since market participants’behavior changed drastically after that point. The …rst two columns in Table 4 report the coe¢ cients from estimating a simple version of regression (1) without any year …xed e¤ects or issuer …xed e¤ects. The reason for estimating a regression only with credit scores and LTV ratios is that those two variables are the most important to predict the performance of mortgages and thus are critical measures when the loan is originated. As would be expected, collateral characteristics a¤ect the subordination level for AAA tranches. The coe¢ cients on credit score and LTV are always statistically signi…cant at the one percent level among di¤erent speci…cations. For example, a deal backed by collateral with lower average credit scores and higher LTV ratios has a higher subordination level. 17

In the rest of the speci…cations reported in Table 4, I add other collateral characteristics, an insurance dummy, and the issuer and year …xed e¤ects. As can be seen, I …nd that riskier collateral characteristics are associated with higher subordination levels. Note that if a AAA tranche is wrapped with insurance and a monoline insurance company guarantees the principal and interest payment to AAA tranche investors unless the insurance company becomes insolvent itself, then this could reduce subordination level by 8-9%. With the insurance dummy, the collateral characteristics can explain the variation of subordination levels by 50% as shown in the third column. This implies that the credit risk of collateral is re‡ected in the size of the bu¤er of a AAA tranche. Since the bu¤er absorbs a possible loss, a AAA tranche can be independent of credit risk of collateral to some extent. After controlling the collateral characteristics, the issuer …xed e¤ects and year …xed e¤ects have relatively less explanatory power on subordination level. It can be interpreted that since subordination level is one of the key structural elements of securitization, the asset pool backing the security is the most important determinant. Table 5 repeats the analysis in Table 4 using the initial overcollateralization level as a dependent variable with the same set of independent variables used in the equation (1).

(2)

Overcollaterlizationi =

0

+ (Collateral Characteristics)i + ci + dy + "i

Table 5 ABOUT HERE Among the collateral characteristics, average credit score and LTV ratio are economically and statistically signi…cant in di¤erent speci…cations of the regression equation (2). I also con…rm that credit score and LTV are considered the most important factors when market participants evaluate the credit risk of a deal. The last two columns in Table 5 report the results with issuer …xed e¤ects and year …xed e¤ects. As mentioned above, overcollateralization is an issuer provided credit enhancement. Predictably, the issuer …xed e¤ects are the most important factor to determine the overcollateralization level and increase adjusted 18

R-squared by 17%. This suggests the existence of reputation e¤ect among subprime MBS issuers. If issuers have varying reputations in the MBS market, then an issuer with a bad reputation has to provide more overcollateralization than an issuer with a good one in order to prevent an adverse selection problem between the issuer and the investor to some extent. Reputation concern is also visible from the fact that market participants monitor the performance of deals under issuer shelves. I continue the analysis in Table 6 by examining the e¤ect of collateral characteristics on delinquency test thresholds.

(3) Delinquency test thresholdi =

0

+ (Collateral Characteristics)i + ci + dy + "i

Table 6 ABOUT HERE As can be seen in Table 6, collateral characteristics a¤ect delinquency test thresholds. However, these results are distinct from the subordination and overcollateralization regressions in several aspects. First, credit score is not economically and statistically signi…cant. The expected sign of the coe¢ cient of credit score is positive, but the estimates show a negative sign in all speci…cations except the …rst one. Second, the year …xed e¤ects have no marginal explanatory power. Third, with the independent variables at hand, I can explain the variation of delinquency test thresholds at about 30% at most, which is less than the previous cases. One possible explanation about the distinct results is because the delinquency test threshold is a credit enhancement mechanism that carries an indirect cost for the issuer. The subordination level and overcollateralization are direct costs. A large proportion of subordinated tranches means that the issuer has to pay out a high interest rate to investors or issue a smaller proportion of AAA tranches. Also, since the initial overcollateralization is a form of equity, higher overcollateralization means more commitment to the deal by the issuer. Even though the test thresholds do not cost directly, the issuer cannot set the threshold extremely low because it means the issuer has to pay more compensation to 19

the investors in junior tranches. Furthermore, as seen in the Figure 11, the delinquency test thresholds do not show large variations during the sample period even though the average trend was downward. It suggests the threshold level is standardized to some extent, and thus is di¢ cult to explain with collateral characteristics. The e¤ ect of Originator-Issuer a¢ liation on the design.— Based on the relationship between the mortgage originator and the MBS issuer there are two types of deals. The …rst type is the deal issued by the mortgage originator such as Ameriquest Bank, New Century Financial, and Countrywide Financial. The second type is the case when the issuer of the deal is di¤erent from the mortgage originator. The issuer, for example Goldman Sachs or Carrington Capital, purchases mortgage pools from the wholesale market and packages the mortgages into MBSs. The motivation for each type of deal is di¤erent. In the case of the former, it is to reduce funding costs or to satisfy risk-based capital requirements. In contrast, the latter case is to generate arbitrage opportunities or fee income. For example, according to New Century Financial’s 2005 10-K …ling, it reports that it had three types of secondary market transactions in 2005: (i) securitizations structured as …nancings total 20.8%, or $10.961 billion of such transactions, (ii) securitization structured as sales total 12.2%, or $6.442 billion, and (iii) subprime whole loan sales total 62.2%, or $32.816 billion. This means that New Century Financial kept $10.961 million of the MBSs it issued on its balance sheet and sold $6.442 million of MBSs to investors. This is the …rst type of the deal described above. In addition, the 10-K noted that New Century Financial sold $6.3 billion of loans to Carrington Capital Management, LLC, and $5.8 billion of loans to Morgan Stanley Mortgage Capital Inc., during the same year. This means that Carrington Capital and Morgan Stanley Mortgage Capital produced subprime MBSs out of New Century’s subprime loans. This is the second type of deal mentioned above. Thus it would be interesting to know whether the originator-issuer relationship a¤ects the design of MBSs, especially credit enhancements. This question concerns whether market participants expect there is an adverse selection problem between the originator and the

20

issuer. Obviously the subprime originator has superior information about its mortgages than a third-party because the originator evaluates the credit risk of mortgages in the …rst place. Therefore, it is possible that the originator sells bad loans to a third-party in the wholesale market and packages good loans into MBSs under its name and then sells them to investors or keeps them on its balance sheets. In other words, it is possible that the deal packaged by a third-party is exposed to adverse selection. For testing this question, I add the variable xi indicating whether the mortgage originator is the same as the issuer of the MBS in equation (1) and (2). If the market participants re‡ect the relationship between the originator and the issuer in the credit enhancement, then the estimate of

1

in the equation

(4) has to be negative and signi…cant.

(4)

yi =

0

+

1 xi

+

0 2 (Collateral

Characteristics)i + ci + dy + "i

where yi is either subordination level or the initial overcollateralizatoin level, xi is an indicator whether the mortgage originator is the issuer of MBS. In the case of subordination, the estimates of

1

are economically signi…cant but sta-

tistically insigni…cant. And xi has no marginal explanatory power. In the case of overcollateralization, the estimates of

1

are economically signi…cant, but statistically are not, at

the 5% level. This suggests that MBS design does not re‡ect the relationship between the originator and the issuer. B.

Pricing Information Insensitive Security

Next, I test pricing in my information-insensitivity hypothesis by regressing the issue spread of AAA-rated tranches on the collateral characteristics with tranche and deal characteristics. This analysis focuses on what factors investors take into account when they price an information-insensitive security at the issuance stage. I include the same set of collateral characteristics used in the credit enhancements regression except the insurance dummy. Ad-

21

ditionally, I use the subordination level of each tranche, and deal and tranche characteristics. I replace the year …xed e¤ects with quarter …xed e¤ects for capturing bond market conditions precisely on the issue date. And I use tranches issued before August 2007, similar to my previous analysis. (5) SpreadAAA = i

0 0 + 1 (Subordination)i + 2 (Collateral

Characteristics)i + Xi ++ci +dq +"i

is the tranche spread (basis points) which is the ‡oater over 1 month where SpreadAAA i LIBOR, collateral characteristics include deal average credit score, LTV ratio, weighted average coupon rate, % of ARM in the collateral, % of mortgages originated in California, Xi denotes the tranche and deal characteristics including the weighted average life of each tranche, the deal face amount, and the indicator whether the deal is privately placed or not, ci is a vector of issuer …xed e¤ects, dq is a vector of quarter …xed e¤ects, and "i is the regression residual. I report the results from estimating di¤erent speci…cations of the regression equation (5). Standard errors reported in parentheses are heteroskedasticity robust and clustered at the issuer level throughout. Table 7 ABOUT HERE The …rst column in Panel A of Table 7 reports the simplest version of regression (5) with subordination level and quarter …xed e¤ects. Since subordination is direct protection for AAA-rated tranches from the loss of underlying collateral, it could be expected that a higher subordination level leads to a lower risk premium. The coe¢ cient of subordination level is economically signi…cant, but not statistically signi…cant. This implies that if a tranche is classi…ed as AAA-rated (which requires adequate loss-protection), then investors do not take into account subordination level. In other words, in the case of AAA-rated tranches, credit rating is a su¢ cient condition for subordination level. Hence, in the …rst column, most of the explanatory power comes from quarter …xed e¤ects, which re‡ect subprime MBS market

22

conditions, especially the risk appetite of investors, and the supply and demand of subprime MBSs during the sample period. The second column represents the regression results after adding the underlying collateral characteristics. Except for the average LTV ratio, other collateral characteristics are not statistically signi…cant. While the LTV ratio is statistically signi…cant at the 5% level, the coe¢ cient of LTV ratio is not economically signi…cant. For example, if the deal is backed by collateral with a high LTV ratio and thus has a higher credit risk, then the tranche has to pay a higher spread, all other things being equal. However, the sign of the coe¢ cient of LTV ratio is positive. Also, note that comparing the …rst and second columns, the increment of adjusted R-squared is only 0.4%. In sum, the collateral characteristics are not economically and statistically signi…cant and also have no marginal explanatory power on the issue spread of AAA-rated subprime tranches. Column 3 in Panel A of Table 7 repeats the analysis in Column 1 with the addition of tranche and deal characteristics. The results demonstrate that the weighted average life is the most crucial determinant for the issue spread of AAA-rated tranches. For example, a tranche with a month longer weighted average life, pays and additional two basis points. Moreover, the coe¢ cient of the private placement dummy is economically and statistically signi…cant at the 10% level. A private placement deal is designed for a speci…c investor group and is not available to the public. Thus, the requests of a speci…c investor group are re‡ected in the deal and tranche structure. And the results imply that if a deal is privately placed, then the tranche pays seven basis points more than a publicly available deal. Note that the increment of adjusted R-squared is 36% from Column 1 to Column 3 and I can explain the issue spread by 60%. As can be seen, the results in Column 4 and 5 remain qualitatively and quantitatively the same. The issuer …xed e¤ects do not play an important role in AAA spread determination. I conduct a joint hypothesis test that all coe¢ cients of collateral characteristics are zeros with the speci…cation of Column 4. If investors do not consider collateral information at all,

23

the null hypothesis has to be accepted or statistically insigni…cant.

H0 :

21

=

22

=

23

=

24

=0

This hypothesis is accepted at the 1% level but rejected at 5%, since the p-value is 0.0397. In other words, all collateral characteristics are jointly insigni…cant at 1%, but signi…cant 5%. It is common that a joint hypothesis test is di¢ cult to accept. Taking account of the fact that the test includes four independent variables, the results are supportive of my information-insensitivity hypothesis. After combining the individual coe¢ cient test results and analyzing the increment of R-squared, the collateral characteristics are not important factors to explain the spread of AAA-rated tranches. In summary, the issue spread of AAArated tranches is insensitive to the collateral characteristics. In Panel B of Table VI, I estimate the following speci…cation in spirit with Kwan (1996) and Collin-Dufresne, Goldstein, and Martin (2001).

(6)SpreadAAA = i

0

+

1 (Subordination)i

+

0 2 (Collateral

+ 1 (10-year T reasury Y ield) +

1 (Y

Characteristics)i + Xi +

ield Curve Slope)

+ 2 (M onthly S&P return) + ci + "i

where 10-year T reasury Y ield is the monthly average of the 10-year Treasury yield, Y ield Curve Slope is the monthly average of yield di¤erence between the 10-year Treasury and 2-year Treasury, and M onthly S&P return is the average return of S&P index. I replace the quarter …xed e¤ects from the regression equation (5) with the …nancial variables that can re‡ect bond market conditions. I match the monthly average of the …nancial variables to the deal by the issue month. For example, if the deal is issued in March 2005, then I assign the corresponding averages of the 10-year Treasury yield, the yield curve slope, and the monthly S&P return.

24

The results of Panel B demonstrate similar results with Panel A. The coe¢ cients of the 10-year Treasury yield and the yield curve slope are economically and statistically signi…cant at the 1% level. For example, a 1% increase in the 10-year Treasury yield is associated with a nine basis point increase of the issue spread. And if the average yield di¤erence between the 10-year Treasury and 2-year Treasury increases 1%, then the spread increases four basis points. In contrast, the S&P return is not signi…cant in every speci…cation. Note that the dependent variable is the ‡oater over the 1-month LIBOR. Total interest payments depend on the market interest rate movement re‡ected in the LIBOR and the ‡oater. The ‡oater re‡ects the risk appetite of investors on the issue date. Thus, the signi…cance of the 10-year Treasury yield has to be interpreted as co-movement rather than causality. During the period when the spreads of AAA-rated tranches decreased, 10-year Treasury yields increased. At the same time, due to the larger increments of the 2-year Treasury yield, the yield curve slope experienced a downward movement. Even though the relationships between …nancial variables (which represent bond market conditions) and issue spreads are correlations, the …nancial variables are surprisingly good substitutes for the time …xed e¤ects. And the results are qualitatively and quantitatively the same as the results of Panel A, and hence are robust. The pricing after August of 2007.— For studying what happened after the market disruption in August 2007, I repeat the same analysis with the tranches issued after August 1, 2007 by using the regression equation (5). There were 31 deals; subsequently the subprime MBS market completely disappeared. There are 110 AAA-rated tranches that can be used for the analysis. As can be seen in Panel C of Table 7, the regression results are not robust among di¤erent speci…cations, and therefore are di¢ cult explaining the determinants of the spread. Even though the estimates of weighted average life and percentage of WAC are statistically and economically signi…cant, the estimates of other independent variables are either statistically or economically insigni…cant. Note that including issuer …xed e¤ects increases adjusted R-squared by 19%. In contrast to the analysis with the sample issued before August 2007, the issuer became an important consideration when investors priced

25

deals. These contrasting results reveal the profound change of market sentiment. To summarize the results in Table 7, investors did not take into account collateral characteristics when they priced AAA tranches before August 2007. It is no di¤erent from investors’ over-reliance on credit ratings when they decide to invest. This result can be interpreted as evidence that the subprime MBS market had functioned in the way it was supposed to. It becomes clearer with the following analysis about BBB spreads. C.

Di¢ culties for pricing Information Sensitive Security

In this subsection, I focus on the pricing of an information-sensitive security by analyzing the spreads of BBB-rated tranches. The control variables are the same as in the regression equation (5). (7) SpreadBBB = i

0 0 + 1 (Subordination)i + 2 (Collateral

Characteristics)i + Xi ++ci +dq +"i

One empirical challenge for analyzing the spread of BBB tranches is that said tranches could be sold at discounted prices and the prices at issuance were rarely recorded. However, when I analyze the BBB spreads assuming that issue price is equal to par, the results must be less sensitive to the case when BBB tranches are sold at a discount. In other words, if my results are sensitive to collateral information, then the real case must be even more sensitive to the credit risk of collateral. Using this dominance condition, it is worthwhile to study BBB spreads with the same speci…cations in my AAA spread analysis. Table 8 ABOUT HERE Table 8 represents the results estimating the regression equation (7). The results are exactly opposite to the results of AAA-rated tranche spreads. First, the subordination level is always economically and statistically signi…cant at the 1% level. For example, a 1% increase of subordination level or the initial overcollateralization level reduces the issue spread of BBB-rated tranches by 16 basis points. By providing 26

protection for BBB-rated tranches, the issuer can pay less risk premium to the investors in such tranches. Second, collateral characteristics have more explanatory power than the issue spread of AAA-rated tranches. Collateral characteristics except for the percentage of ARMs are economically signi…cant in all speci…cations. Statistical signi…cance of collateral characteristics varies among di¤erent speci…cations, but the individual coe¢ cients are more likely to be signi…cant than the issue spreads of AAA-rated tranches. Third, deal characteristics are not important in the determination of the issue spreads of BBB-rated tranches. In particular, the weighted average life is neither economically nor statistically signi…cant. This suggests that there is much more uncertainty about principal payment to BBB-rated tranches, thus it is more important whether the collateral can generate enough cash payment to the SPV. If there is no uncertainty about the principal payment, then the weighted average life a¤ects the issue spread in the case of AAA-rated tranches. Fourth, issuer …xed e¤ects are larger than the issue spread of AAA-rated tranches. Including issuer …xed e¤ects increase by 6.7%. These results imply that investors treat AAA-rated tranches nearly homogeneously. However, investors acknowledge the riskiness of BBB-rated tranches and take into account available information about the credit risk of a deal. In this sense, the issue spread of BBB-rated tranches is sensitive to information about the collateral. Fifth, adjusted R-squared is smaller than the issue spread of AAA-rated tranches. Despite the collateral and issuer information having more explanatory power than in the case of the issue spread of AAA-rated tranches, the adjusted R-squared is at most 42%. This implies that there is more uncertainty in the performance of BBB-rated tranches, with a higher probability of write-o¤s during the life of the deal, and thus there exists more unknowns in the determination of issue spreads of BBB-rated tranches. This could be as a result of the nature of information-sensitive securities.

27

D.

Pricing AA and A rated security

In this …nal section, I provide an analysis of the determinants of issue spread from the perspective of a comparison among credit rating class. Table 9 shows the results regressing the issue spread on subordination level and time …xed e¤ects. Note that except for AAArated tranches, subordination level is signi…cant at the 1% level. And as the tranche is more junior, the impact of subordination level on the issue spread monotonically increases. For example, a 1% increase of subordination level reduces the issue spread by 1.6 basis points for AA-rated tranches, but 4.6 basis points for A-rated tranches. This means that investors consider protection for the tranche increasingly as they invest in more junior tranches. Overall, time …xed e¤ects are one of the most important factors in the determinants of issue spread. The downward trend of issue spreads between 2005 and 2006 re‡ects that risk appetite increased over the period. In other words, investors perceived subprime MBSs less risky over that time. Tables 10-13 present the regression results after adding the underlying collateral characteristics. Two points are worth noting. First, the individual collateral characteristics have more impact on the issue spread as the tranche is more junior. Second, when the collateral characteristics are added, increments of adjusted R-squared are 0.3% in the AAA-rated tranche’s case, but 5% in the AA-rated tranche, 4% in the A-rated tranche and 6% in the BBB-rated tranche.

V.

Conclusion

What is the purpose of the securitization process? That is a central question of this paper. Using detailed information on subprime MBSs between 2004 and 2007, …rst I document what happened to the subprime securitization market before the global …nancial crisis. I found out the overall quality of collateral deteriorated, but the deterioration was not dramatic. I show that more credit enhancement mechanisms were provided for AAArated tranches on average. Also, I …nd that credit enhancements are associated with the 28

credit risk of underlying mortgages. Thus, it is di¢ cult to conclude that ex-ante qualities of AAA-rated subprime tranches had deteriorated. Next, I show that the issue spread of AAA-rated tranches is insensitive to information about the credit risk of the deal. In contrast, I …nd that the issue spread of BBB-rated tranches re‡ects the credit risk of the deal and thus is sensitive to information about the deal. This is the main reason that most of the BBB subprime MBS tranches went into a re-securitization process in order to recreate information-insensitive securities. My results are consistent with the hypothesis that market participants designed and priced AAA-rated tranches as information-insensitive before the …nancial crisis. Therefore, in conclusion, this paper empirically shows that securitization is the creation of information-insensitive securities.

29

REFERENCES Adrian, Tobias, and Hyun Song Shin. 2010. “The Changing Nature of Financial Intermediation and the Financial Crisis of 2007-2009.” Federal Reserve Bank of New York, Sta¤ Report No.439. Ashcraft, Adam B., and Til Schuermann. 2008. Understanding the Securitization of Subprime Mortgage Credit. New York: Now Publishers Inc. Bank for International Settlement. 2005. “The Role of Ratings in the Structured Finance: Issues and Implications.”Global Financial System Working Group Report. Benmelech, E¢ , and Jennifer Dlugosz. 2009. “The Credit Rating Crisis.”NBER Working Paper No.w15045. Bernanke, Ben S. 2000. Essays on the Great Depression. NJ: Princeton University Press. Bernanke, Ben S. 2007. Housing, Housing Finance, and Monetary Policy. Speech at the Federal Reserve Bank of Kansas City’s Economic Symposium. Bernanke, Ben S. 2010. Causes of the Recent Financial and Economic Crisis. Testimony in front of the Financial Crisis Inquiry Commission. Bhardwaj, Geetesh, and Rajdeep Sengupta. 2008. “Where’s the Smoking Gun? A Study of Underwriting Standards for US Subprime Mortgages.”Federal Reserve Bank of St. Louis Working Paper, No.2008-036D. Brunnermeier, Markus K. 2009. “Deciphering the Liquidity and Credit Crunch 20072008.”Journal of Economic Perspective, 23(1): 77-100. Calomiris, Charles W. 2009. “The Subprime Turmoil: What’s Old, What’s New, and What’s Next.”The Journal of Structured Finance, 15(1): 6-52. Collin-Dufresne, Pierre, Robert S. Goldstein, and J. Spencer Martin. 2001. “The Determinants of Credit Spread Changes.”Journal of Finance, 56(6): 2177-2207. Dang, Tri Vi, Gary B. Gorton, and Bengt Holmstrom. 2010. “Opacity and the Optimality of Debt for Liquidity.”Working Paper. DBRS. 2009. Methodology: Residential Mortgage-Backed Securities Transactions. Demyanyk, Yuliya, and Otto Van Hermert. 2009. “Understanding the Subprime Mortgage Crisis.”Review of Financail Studies, doi: 10.1093/rfs/hhp033. 30

Dell’Ariccia, Giovanni, Deniz Igan, and Luc Laeven. 2008. “Credit Booms and Lending Standards: Evidence from the Subprime Mortgage Market.” International Monetary Fund Working Paper, No. 08-106. Fabozzi, Frank J., Anand K. Bhattacjarya, and William S.Berliner. 2009. Mortgage-Backed Securities: Products, Structuring and Analytical Techniques. NJ: Wiley. Fabozzi, Frank J., and Vinod Kothari. 2008. Introduction to Securitization. NJ: Wiley. Fabozzi, Frank J., and Dennis Vink. 2010. “Looking beyond Credit Rating: Factors Investors Consider in Pricing European Asset-Backed Securities.”Working Paper. Financial Crisis Inquiry Commission. 2010a. “Securitization and the Mortgage Crisis.” Preliminary Sta¤ Report. Financial Crisis Inquiry Commission. 2010b. “Shadow Banking and the Financial Crisis.” Preliminary Sta¤ Report. Financial Crisis Inquiry Commission. 2010c. “Credit Derivarives and Mortgage-Related Derivatives.” Preliminary Sta¤ Report. Goodman, Laurie S., Shumin Li, Dougals J. Lucas, Thomas A. Zimmerman, and Frank J. Fabozzi. 2008. Subprime Mortgage Credit Derivatives. NJ: Wiley. Gorton, Gary B. 2008a. “Information, Liquidity, and the (Ongoing) Panic of 2007.”American Economic Review, Papers and Proceedings, Volume 9: 567-572. Gorton, Gary B. 2008b. “The Panic of 2007.” Proceedings of the 2008 Jackson Hole Conference, Federal Reseve Bank of Kansas City Symposium. Gorton, Gary B. 2010. Slapped by the Invisible Hand, the Panic of 2007. New York: Oxford University Press. Gorton, Gary B., and Andrew Metrick. 2009a. “Securitized Banking and the Run on the Repo.”Yale University School of Management Working Paper. Gorton, Gary B., and Andrew Metrick. 2009b. “Haircuts.” Yale University School of Management Working Paper. Gorton, Gary B., and Nicholas S. Souleles. 2005. “Special Purpose Veichles and Securitization.”In The Risks of Financial Institutions, ed. Rene Stulz and Mark Carey, Chicago: University of Chicago Press. 31

Gorton, Gary B., and George Pennacchi. 1990. “Financial Intermediaries and Liquidity Creation.”Journal of Finance, 45(1): 49-72. Keys, Benjamin J., Tanmoy Mukherjee, Amit Seru, and Vikrant Vig. 2010. “Did Securitization Lead to Lax Screening? Evidence from Subprime Loans.” The Quarterly Journal of Economics, 125: 307-362. Kothari, Vinod. 2006. Securitization. NJ: Wiley. Kwan, Simon H. 1996. “Firm-speci…c Information and the Correlation between Individual Stocks and Bonds.”Journal of Financial Economics, 40: 63-80. Mortgage Banker’s Association. 2009. The 2009 Mortgage Market Statistical Annual: Volune I and II. Bethesda: Inside Mortgage Finance Publication. Mayer, Christopher, Karen Pence, and Shane M. Sherlund. 2009. “The Rise in Mortgage Defaults.”Journal of Economic Perspectives, 23(1): 27-50. Markit. 2008. Index Methodology for the ABX.HE Index for the Sub-Prime Home Equity Sector. Nadauld, Taylor B., and Shane M. Sherlund. 2009. “The Role of the Securitization Process in the Expansion of Subprime Credit.” Federal Reserve Board, Washington, D.C., Finance and Economics Discussion Series, Working Paper 2009-28. New Century Financial Corportaion. 10-K for 12/31/2005. Securities and Exchange Commission(SEC). http://www.secinfo.com/dR7Km.v8d.htm. Nomura Fixed Income Research. 2006. “MBS Market: Concepts and Topics.”American Securitization Forum. Shin, Hyun Song. 2010. Risk and Liquidity. New York: Oxford University Press. Standard & Poor’s. 2007. “Current State of the Securitization Market.”ASF/ESF Global Summit on the State of the Securitization Industry. Standard & Poor’s. 2009. “Request For Comment: U.S. RMBS Rating Methodology And Assumptions For Prime Jumbo, Alternative-A, And Subprime Loans.”Rating Directs. United States Senate Committee on Banking, Housing, and Urban A¤airs. 2010. Brief Summary of The Dodd-Frank Wall Street Reform and Consumer Protection Act.

32

A

Appendix: Description of Variables

For reference, the following is a list of variables used in the paper and a brief description of each variable. A. Collateral Characteristics: The collateral characteristics are the weighted average risk measures of a mortgage pool backing a security. 1. Credit Score is a numerical grade of the credit history of a borrower that makes use of information about payment history, amounts owed, the length of time a borrower has received credit, new credit, and types of credit used. Credit bureau scores are often called FICO scores because most credit bureau scores used in the U.S. are produced from software developed by Fair Isaac and Company. A FICO score is between 300 and 850 and exhibits a left-skewed distribution with 720 as a median. 2. LTV ratio(Loan-to-Value, %) is calculated by dividing a mortgage loan by its house value. The ratio represents the relative size of a down payment in relation to the loan, suggesting the extent of the borrower’s leverage and equity level. 3. WAC(Weighted Average Coupon, %) represents the weighted average gross interest rate of the mortgage pool at its issue date. 4. Loan Size($1000) represents the average loan balance. It is one of the criteria for GSE’s purchasing guideline. In 2010, the single-family mortgage loan limit of Fannie Mae is $417,000 for a general area and $729,750 for high-cost areas. If the loan is larger than the limit, it is called a nonconforming loan. If the loan meets the limit, it is called a conforming loan. 5. Loan Age(month): The age of a mortgage is calculated by subtracting the MBS issue date from the date of origin. 6. % of Limited Documentation: According to Fabozzi and Kothari(2008), “Loan o¢ cers typically required applicants to report and document income, employment status, and …nancial resources. Part of the application process routinely involved compiling documents such as tax returns and bank statements for use in the underwriting process.” Limited documentation means that the applicant provides insu¢ cient documentation compared with the standard. 7. % of California represents the percentage of mortgage loans which were secured by mortgaged properties located in California. Geographical concentration is one of the important considerations in measuring the credit risk of MBS. 8. % of FRM(Fixed Rate Mortgage)/ARM(Adjustable Rate Mortgage) denotes the interest rate type of a mortgage loan. A …xed rate mortgage has a constant interest rate over the life of the loan. In contrast, the interest payment of an adjustable rate mort-

33

gage depends on the index and the margin. Generally, after an initial …xed-rate period, the interest rate on an adjustable-rate mortgage loan will adjust semi-annually on each adjustment date to equal the sum of six-month LIBOR and the gross margin for that mortgage loan, subject to periodic and lifetime limitations. ARM type includes hybrid ARM and negative amortization ARM. The interest rate reset date is an important consideration due to payment shock. 9. % of Owner Occupancy/Invest describes the occupancy type. It represents whether a borrower buys a house for occupancy purposes or investment purposes. In the latter case, a borrwer does not intend to reside in the house. 10. % of Single Family Home/Multi-Family Home describes the type of home. If the home is structured to allow more than one family to live, such as a duplex or condominium, then it is called a multi-family home. 11. % of Purchase/Equity take out/Re…nance describes the purpose of a mortgage loan. If a borrower applies for a loan to purchase a new house, then it is counted as a purchase type. If a borrower reapplies the mortgage loan which is larger than the original loan, then it is classi…ed as an equity take-out type. If a borrower reapplies the mortgage loan which is same balance as the original loan, then it is classi…ed as a re…nance type. According to Gorton(2008b), a majority of subprime mortgage borrowers depend on the re…nancing option and the eradication of the re…nancing option to subprime mortgage borrowers contributed to the rise of default rates.

34

Table 1 Bear Stearns Asset Backed Securities Trust 2007 Issuing Behavior Series

Issue Date

OC

Subordination

Spread of AAA

FICO

LTV

% of ARM

HE1

1/29/2007

2.25

21.80

0.12

611

81.44

79.12

HE2

2/27/2007

3.00

21.40

0.1

612

84.12

78.28

HE3

3/28/2007

5.55

26.25

0.12

624

83.21

70.46

HE4

4/13/2007

5.45

26.15

0.12

612

80.99

75.2

HE5

5/29/2007

5.05

24.90

0.09

608

80.41

71.04

HE6

8/29/2007

8.49

39.26

1.25

593

78.57

69.66

HE7

9/17/2007

5.73

34.07

1

603

75.64

46.16

Note: Table 1 represents the issuing behavior of Bear Stearns Asset Backed Securities Trust during 2007. According to the 2009 Mortgage Market Statistical Annual, Bear Stearns placed 6th among the top subprime MBS issuers with 4.3% market share in 2007. There were …ve deals before the market disruption and two afterwards. As can be seen, the spread of AAA-rated tranches rose from 12 basis points to above 100 basis points. Note that before the market interruption, the issue spread of A-rated tranches was 100 basis points. And the subordination level of AAA tranches rose from an average of 25%, to above 35%.

35

Table 2 An Example of a Subprime MBS Deal: Merrill Lynch Investors Trust 2006-HE5 Deal and Tranche Structure (Issue Date 2006/09/28) This table represents an example of a subprime mortgage backed securities in the sample. Merrill Lynch Investors Trust notes the issuer name. 2006-HE5 notes the series number of the MBS and is short for the …fth Home Equity deal issued in 2006. There are 14 tranche in the deal. Tranche

Original Amount

Rating

Collateral Group

A1

170,000,000

22.1

2.31

15

AAA

CONF/G1

A2A

480,000,000

22.1

1

6

AAA

NCON/G2

A2B

150,000,000

22.1

2

11

AAA

NCON/G2

A2C

200,000,000

22.1

3.47

15

AAA

NCON/G2

A2D

78,000,000

22.1

8.96

24

AAA

NCON/G2

M1

50,000,000

18.4

3.74

26

AA+

ALL

M2

42,000,000

15.4

4.67

30

AA

ALL

M3

29,000,000

13.3

8.72

33

AA

ALL

M4

25,000,000

11.5

5.06

38

AA-

ALL

M5

23,000,000

9.8

5.01

41

A+

ALL

M6

22,000,000

8.2

4.96

47

A

ALL

B1

21,000,000

6.7

4.9

80

A-

ALL

B2

19,000,000

5.3

4.83

100

BBB+

ALL

B3

15,000,000

4.2

4.75

200

BBB

ALL

All

Subordination

WAL

Spread

1,324,000,000

This table below represents the collateral information backing Merrill Lynch Investors Trust 2006-HE5 Deal.

Average Collateral Characteristics FICO

LTV

ARM(%)

Invest(%)

Lim(%)

Multi(%)

WAC

Cal(%)

Loan Size

CONF/G1

612

78.38

78.59

10.94

4.3

9.46

8.35

20.94

177.42

NCON/G2

634

78.35

78.86

3.89

6.2

8.67

8.10

32.36

210.87

ALL

631

78.35

78.82

5.00

5.9

8.79

8.14

30.56

205.34

36

Table 3 Summary Statistics Panel A. Mean Collateral Characteristics by Year

Interest rate type Occupancy type House type Purpose of loan

2004

2005

2006

2007

All

Credit Score

633.0

628.4

622.2

622.3

625.4

LTV(%)

76.0

79.2

79.3

79.5

78.6

WAC(%)

7.3

7.4

8.3

8.6

7.9

Loan size($1000)

136.3

148.4

164.1

178.1

156.1

Loan age(month)

5.7

5.5

4.6

6.0

5.3

% of Limited documentation

29.9

31.6

33.3

35.8

32.7

% of California

31.5

25.3

22.0

20.1

24.0

% of FRM

41.8

27.3

27.1

34.1

31.2

% of ARM

58.2

72.7

72.8

65.9

68.8

% of Owner occupancy

92.0

92.1

91.8

91.3

91.8

% of Invest

6.2

6.6

5.8

6.7

6.3

% of single family

74.6

74.7

72.7

74.3

73.8

% of multi-family

7.4

6.9

7.2

7.3

7.2

% of Purchase

40.3

41.0

37.7

27.1

36.6

% of Equity take-out

48.6

50.7

56.1

62.1

55.0

% of Re…nance

9.5

8.8

7.3

11.1

8.9

2004

2005

2006

2007

All

Subordination

16.9

20.1

21.9

23.9

20.7

Overcollateralization

1.7

2.0

2.6

4.6

2.7

Delinquency test threshold

43.2

38.4

35.9

33.4

37.6

% of Insurance

12.54

4.43

2.38

9.24

6.33

% of private placement

4.74

5.20

10.58

24.10

10.28

Deal Size(million)

579

574

526

544

538

Number of Tranches

11.1

13.6

14.4

13.3

13.3

Panel B. Deal Characteristics by Year

37

Table 3-continued

Panel C. Tranche Characteristics by Rating Group 2004

2005

2006

2007

All

Issue Spread over 3 Month LIBOR (bps) by year AAA

31.10

24.86

16.05

27.16

22.89

AA

87.80

51.10

36.78

60.78

51.55

A

157.94

90.71

64.72

113.29

93.77

BBB

251.51

192.86

159.98

206.64

189.94

Subordination Level (%) AAA

16.61

19.50

21.52

22.84

20.83

AA

9.81

12.68

13.75

15.26

13.54

A

5.60

6.91

8.08

9.42

7.91

BBB

3.17

4.18

5.04

5.53

4.75

Tranche Size ($ millions) AAA

183

164

157

152

161

AA

33.6

29.5

26.4

28.1

28.2

A

19.1

15.6

14.4

14.9

15.3

BBB

13.3

13.8

10.6

12.0

11.8

Weighted Average Life (year) AAA

3.29

3.45

3.48

3.45

3.44

AA

5.31

5.13

5.35

5.50

5.35

A

5.08

4.88

4.92

5.08

4.98

BBB

4.86

4.59

4.66

4.77

4.70

38

Table 4 Subordination and Credit Risk of Collateral Dependent Variable: Subordination level of AAA tranche Credit Score LTV

(1)

(2)

(3)

(4)

(5)

-0.099**

-0.054**

-0.060**

-0.057**

-0.061**

(0.020)

(0.013)

(0.012)

(0.012)

(0.009)

0.474**

0.321**

0.349**

0.377**

0.368**

(0.055)

(0.049)

(0.046)

(0.054)

(0.056)

2.084**

1.806**

1.516**

1.171+

(0.509)

(0.440)

(0.428)

(0.610)

0.051**

0.033**

0.034*

0.027*

(0.015)

(0.011)

(0.016)

(0.012)

0.068**

0.063*

0.048*

0.061*

(0.024)

(0.024)

(0.020)

(0.023)

-9.196**

-8.474**

-8.539**

(1.685)

(1.818)

(1.858)

WAC % of ARM % of California Insurance Issuer Fixed E¤ects

No

No

No

Yes

Yes

Year Fixed E¤ects

No

No

No

No

Yes

N

738

738

738

738

738

0.300

0.405

0.513

0.578

0.591

Adjusted R

2

Note: This table presents the coe¢ cient estimates and standard errors in parenthese for subordination regressions with robust standard errors, clustered within issuer shelves. **, *, and + denote statistical signi…cance at the 1%, 5%, and 10% levels, respectively.

39

Table 5 Overcollateralization and Credit Risk of Collateral Dependent Variable: Overcollateralization Credit Score LTV

(1)

(2)

(3)

-0.017**

-0.008+

-0.009+

-0.014**

(0.005)

(0.005)

(0.005)

(0.004)

0.094**

0.049*

0.069**

0.079**

(0.019)

(0.022)

(0.021)

(0.016)

0.535**

0.429*

-0.130

(0.162)

(0.164)

(0.136)

0.005

0.005

(0.005)

(0.006)

(0.004)

0.002

-0.001

0.012*

(0.008)

(0.007)

(0.005)

WAC % of ARM % of California Year 2005

(4)

-0.003

0.833** (0.223)

Year 2006

1.492** (0.285)

Year 2007

2.856** (0.406)

Issuer Fixed E¤ects

No

No

Yes

Yes

N

715

715

715

715

Adjusted R2

0.095

0.131

0.302

0.438

Note: This table presents the coe¢ cient estimates and standard errors in parenthese for overcollateralization regressions with robust standard errors, clusterd within issuer shelves. **, *, and + denote statistical signi…cance at the 1%, 5%, and 10% levels, respectavely.

40

Table 6 Delinquency Test Thresholds and Credit Risk of Collateral Dependent Variable: Delinquency Test Thresholds Credit Score LTV

(1)

(2)

(3)

(4)

0.031+

-0.045+

-0.028

-0.023

(0.018)

(0.025)

(0.019)

(0.020)

-0.899**

-0.626**

-0.679**

-0.678**

(0.135)

(0.121)

(0.154)

(0.170)

-3.648**

-2.880**

-2.431*

(0.608)

(0.746)

(1.001)

-0.077

-0.076

-0.066

(0.049)

(0.051)

(0.049)

-0.092+

-0.113**

-0.124**

(0.047)

(0.039)

(0.040)

WAC % of ARM % of California Issuer Fixed E¤ects

No

No

Yes

Yes

Year Fixed E¤ects

No

No

No

Yes

N

661

661

661

661

0.192

0.259

0.323

Adjusted R

2

0.323

Note: This table presents the coe¢ cient estimates and standard errors in parenthese for delinquency test thresholds regressions with robust standard errors, clusterd within issuer shelves. **, *, and + denote statistical signi…cance at the 1%, 5%, and 10% levels, respectavely.

41

Table 7 Determinants of Issue Spread of AAA-rated Subprime MBS with Quarter Fixed E¤ects

Panel A. Dependent Variable: Issue Spread of AAA-rated Subprime MBS (bps) (1) Subordination

(3)

(4)

(5)

-0.021

-0.032

-0.046

-0.054

-0.157

(0.065)

(0.070)

(0.071)

(0.069)

(0.097)

0.002

0.002

0.010

(0.016)

(0.015)

(0.020)

-0.156*

-0.116*

-0.105+

(0.075)

(0.054)

(0.062)

1.120

1.264

1.529+

(1.204)

(0.829)

(0.825)

0.025

-0.0004

0.018

(0.025)

(0.021)

(0.021)

0.006

0.000

-0.047+

(0.024)

(0.022)

(0.024)

Credit Score LTV WAC % of ARM .

(2)

% of California Weighted

2.615**

2.629**

2.648**

Average Life

(0.084)

(0.088)

(0.088)

Deal Face

-0.000*

-0.000*

-0.000

(0.000)

(0.000)

(0.000)

7.836*

6.922*

7.873*

(3.271)

(2.655)

(3.218)

Private Placement Constant

27.450**

28.073*

21.037**

19.995+

13.111

(2.512)

(11.372)

(2.308)

(10.844)

(11.558)

Issuer Fixed E¤ects

No

No

No

No

Yes

N

2985

2985

2985

2985

2985

0.236

0.240

0.604

0.607

0.630

Adjusted R

2

Note: This table presents the coe¢ cient estimates and standard errors in parentheses for issue spread of AAA-rated subprime MBS regressions with robust standard errors. All speci…cations include quarter …xed e¤ects. The sample used in this analysis includes tranches issued before August of 2007.

42

Table 7-continued Determinants of Issue Spread of AAA-rated Subprime MBS with Financial Variables Panel B.

Dependent Variable: Issue Spread of AAA-rated Subprime MBS (bps) (1)

(2)

(3)

(4)

(5)

-0.067

-0.061

-0.093

-0.087

-0.165

(0.059)

(0.077)

(0.066)

(0.073)

(0.107)

-9.072**

-8.390**

-8.993**

-8.625**

-8.256**

(0.734)

(1.317)

(0.788)

(1.202)

(1.329)

4.224**

4.013**

4.336**

4.146**

4.306**

(0.729)

(0.646)

(0.579)

(0.534)

(0.570)

Lagged Monthly

-12.569+

-8.606

-10.594

-9.456

-6.259

S&P index return

(6.424)

(7.778)

(6.953)

(8.404)

(7.746)

-0.020

-0.020

-0.012

(0.018)

(0.016)

(0.021)

-0.069

-0.026

0.016

(0.080)

(0.064)

(0.072)

-0.529

-0.399

-0.406

(0.918)

(0.641)

(0.693)

0.003

-0.022

-0.006

(0.029)

(0.026)

(0.026)

0.017

0.011

-0.036

(0.026)

(0.023)

(0.032)

Subordination 10-year Treasury yield Yield Curve Slop

Credit Score LTV WAC % of ARM % of California Weighted

2.614**

2.624**

2.644**

Average Life

(0.081)

(0.083)

(0.084)

Deal Face

-0.000*

-0.000*

-0.000+

(0.000)

(0.000)

(0.000)

8.200**

7.564**

8.523**

(2.995)

(2.591)

(3.142)

Private Placement Constant N Adjusted R

2

62.412**

80.701**

54.590**

72.020**

68.921**

(4.204)

(11.943)

(5.401)

(9.548)

(10.449)

2960

2960

2960

2960

2960

0.184

0.187

0.556

0.558

0.582

43

Table 7-continued Determinants of Issue Spread of AAA-rated Subprime MBS after August of 2007

Panel C. Dependent Variable: Issue Spread of AAA-rated Subprime MBS (bps) Subordination

(1)

(2)

(3)

(4)

(5)

1.322

0.652

2.338

1.666

0.363

(1.711)

(1.817)

(1.614)

(1.775)

(3.019)

0.904*

0.895*

0.486

(0.348)

(0.351)

(0.458)

-3.759+

-4.505+

-3.170

(2.068)

(2.288)

(2.492)

40.832*

37.981*

63.188**

(16.398)

(16.813)

(15.389)

0.077

0.086

0.427

(0.207)

(0.185)

(0.314)

0.800+

0.648

0.969*

(0.443)

(0.468)

(0.436)

Credit Score LTV WAC % of ARM % of California Weighted

5.318**

5.020**

4.562**

Average Life

(0.598)

(0.902)

(0.986)

Deal Face

-0.000

-0.000

0.000

(0.000)

(0.000)

(0.000)

-2.248

-10.073

-22.725

(17.067)

(12.817)

(27.323)

Private Placement Constant

45.096

-581.021+

12.064

-529.296+

-733.641*

(51.637)

(288.687)

(50.252)

(256.470)

(301.802)

Issuer Fixed E¤ects

No

No

No

No

Yes

N

110

110

110

110

110

Adjusted R2

0.011

0.189

0.149

0.282

0.577

Note: This table presents the coe¢ cient estimates and standard errors in parentheses for issue spread of AAA-rated tranche regressions with robust standard errors. All speci…cations include quarter …xed e¤ects. The sample used in this analysis includes tranches issued after August of 2007.

44

Table 8 Determinants of Issue Spread of BBB-rated Subprime MBS Dependent Variable: Issue Spread of BBB-rated Subprime MBS (bps) Subordination

(1)

(2)

(3)

(4)

(5)

-8.176**

-13.039**

-8.621**

-13.438**

-16.248**

(1.615)

(2.226)

(1.367)

(2.084)

(1.990)

-0.287

-0.272

-0.398*

(0.229)

(0.219)

(0.180)

1.402*

1.415*

2.167**

(0.589)

(0.598)

(0.603)

16.500**

15.782**

10.421

(5.007)

(4.714)

(6.550)

-0.478*

-0.393*

-0.595*

(0.203)

(0.193)

(0.236)

0.527*

0.628**

0.440*

(0.222)

(0.208)

(0.204)

Credit Score LTV WAC % of ARM % of California Weighted

-7.643+

-1.609

-1.670

Average Life

(3.992)

(3.984)

(3.648)

Deal Face

-0.000**

-0.000**

-0.000

(0.000)

(0.000)

(0.000)

14.081

8.400

19.960

(9.8989)

(10.589)

(12.552)

Private Placement Constant

254.394**

233.143

306.828**

241.429+

348.208**

(7.933)

(147.635)

(23.828)

(127.443)

(117.691)

Issuer Fixed E¤ects

No

No

No

No

Yes

N

1704

1704

1704

1704

1704

Adjusted R2

0.278

0.345

0.301

0.357

0.423

Note: This table presents the coe¢ cient estimates and standard errors in parenthese for issue spread of BBB-rated subprime MBS regressions with robust standard errors, clusterd within issuer shelves. **, *, and + denote statistical signi…cance at the 1%, 5%, and 10% levels, respectavely. All speci…cations include quarter …xed e¤ects. The sample used in this analysis includes tranches issued before August of 2007.

45

Table 9 The E¤ects of Subordination and Quarter Fixed E¤ects on Issue Spread by Rating Group Dependent Variable: Issue Spread (bps) S&P Credit Rating AAA

AA

A

-0.021

-1.587**

-4.647**

-8.176**

(0.065)

(0.291)

(1.158)

(1.615)

27.450**

83.260**

163.228**

254.394**

(2.512)

(6.540)

(8.931)

(7.933)

N

2985

2117

1797

Adjusted R2

0.236

0.422

0.372

Subordination Constant

BBB

1704 0.278

Table 10 The E¤ects of Collateral Characteristics on Issue Spread by Rating Group Dependent Variable: Issue Spread (bps) S&P Credit Rating AAA

AA

A

BBB

-0.032

-2.181**

-6.975**

-13.039**

(0.0704)

(0.268)

(1.413)

(2.226)

0.002

-0.105

-0.211

-0.287

(0.016)

(0.063)

(0.199)

(0.229)

-0.156*

0.448+

0.858

1.402*

(0.075)

(0.243)

(0.546)

(0.589)

1.120

5.373

8.745

16.500**

(1.204)

(3.971)

(6.476)

(5.007)

0.025

-0.195+

-0.403

-0.478*

(0.025)

(0.115)

(0.242)

(0.203)

0.006

0.241**

0.507*

0.527*

(0.024)

(0.069)

(0.195)

(0.222)

28.073*

87.019

189.945

233.143

(11.372)

(56.072)

(146.884)

(147.635)

N

2985

2117

1797

1704

Adjusted R2

0.240

0.474

0.412

0.345

Subordination Credit Score LTV WAC % of ARM % of California Constant

46

Table 11 The E¤ects of Deal Characteristics on Issue Spread by Rating Group Dependent Variable: Issue Spread (bps) S&P Credit Rating AAA

AA

A

BBB

-0.046

-1.622**

-4.931**

-8.621**

(0.071)

(0.266)

(1.143)

(1.367)

Weighted

2.615**

0.063

-7.477

-7.643+

Average Life

(0.084)

(0.792)

(7.932)

(3.992)

Deal Face

-0.000*

-0.000**

-0.000**

-0.000**

(0.000)

(0.000)

(0.000)

(0.000)

7.836*

16.038**

38.763**

14.081

(3.271)

(5.208)

(12.124)

(9.899)

21.037**

86.961**

217.305**

306.828**

(2.308)

(10.079)

(37.666)

(23.828)

N

2985

2117

1797

1704

Adjusted R2

0.604

0.444

0.416

0.301

Subordination

Private Placement Constant

Note: This table presents the coe¢ cient estimates and standard errors in parenthese for issue spread regressions by rating groups with robust standard errors, clusterd within issuer shelves. **, *, and + denote statistical signi…cance at the 1%, 5%, and 10% levels, respectavely. All speci…cations include quarter …xed e¤ects. The sample used in this analysis included tranches issues before August of 2007.

47

Table 12 The E¤ects of Collateral and Deal Characteristics on Issue Spread Dependent Variable: Issue Spread (bps) S&P Credit Rating AAA

AA

A

BBB

-0.054

-2.176**

-6.957**

-13.438**

(0.069)

(0.270)

(1.661)

(2.084)

0.001

-0.093

-0.199

-0.272

(0.015)

(0.066)

(0.179)

(0.219)

-0.116*

0.534**

1.013*

1.415*

(0.054)

(0.193)

(0.494)

(0.598)

1.264

4.334

4.973

15.782**

(0.829)

(3.346)

(4.671)

(4.714)

-0.0004

-0.170

-0.355+

-0.393*

(0.021)

(0.118)

(0.197)

(0.193)

0.000

0.257**

0.575**

0.628**

(0.022)

(0.067)

(0.201)

(0.208)

Weighted

2.629**

-0.361

-4.581

-1.609

Average Life

(0.088)

(0.592)

(9.707)

(3.984)

Deal Face

-0.000*

-0.000*

-0.000**

-0.000**

(0.000)

(0.000)

(0.000)

(0.000)

6.922*

11.503**

33.556**

8.400

(2.655)

(4.256)

(10.514)

(10.589)

19.995+

83.869

230.371*

241.429+

(10.844)

(57.293)

(88.594)

(127.443)

N

2985

2117

1797

1704

Adjusted R2

0.607

0.486

0.446

0.357

Subordination Credit Score LTV WAC % of ARM % of California

Private Placement Constant

Note: This table presents the coe¢ cient estimates and standard errors in parenthese for issue spread regressions by rating groups with robust standard errors, clusterd within issuer shelves. **, *, and + denote statistical signi…cance at the 1%, 5%, and 10% levels, respectavely. All speci…cations include quarter …xed e¤ects. The sample used in this analysis includes tranches issued before August of 2007.

48

Table 13 The E¤ects of Issuer and Collateral and Deal Characteristics on Issue Spread Dependent Variable: Issue Spread (bps) S&P Credit Rating AAA

AA

A

BBB

-0.157

-2.5195**

-8.231**

-16.248**

(0.097)

(0.2977)

(1.763)

(1.990)

0.010

-0.1022

-0.375+

-0.398*

(0.020)

(0.0683)

(0.216)

(0.180)

-0.105+

0.5322*

1.385*

2.167**

(0.062)

(0.2015)

(0.635)

(0.603)

1.529+

5.2613

-2.558

10.421

(0.825)

(3.7904)

(5.323)

(6.551)

0.018

-0.1827

-0.529*

-0.595*

(0.021)

(0.1129)

(0.215)

(0.236)

-0.047+

0.1776*

0.380*

0.440*

(0.024)

(0.0715)

(0.182)

(0.204)

Weighted

2.648**

-0.3488

-8.429

-1.670

Average Life

(0.088)

(0.5868)

(8.390)

(3.647)

Deal Face

-0.000

-0.0000

-0.000+

-0.000

(0.000)

(0.0000)

(0.000)

(0.000)

7.873*

18.0879**

40.641**

19.960

(3.218)

(3.1334)

(10.188)

(12.552)

13.111

67.9281

494.972**

348.208**

(11.558)

(68.6490)

(130.585)

(117.691)

N

2985

2117

1797

1704

Adjusted R2

0.630

0.523

0.519

0.423

Subordination Credit Score LTV WAC % of ARM % of California

Private Placement Constant

Note: This table presents the coe¢ cient estimates and standard errors in parenthese for issue spread regressions by rating groups with robust standard errors, clusterd within issuer shelves. **, *, and + denote statistical signi…cance at the 1%, 5%, and 10% levels, respectavely. All speci…cations include issuer …xed e¤ects and quarter …xed e¤ects. The sample used in this analysis included tranches issued before August of 2007.

49

3,000 0

$ in Billions 1,000 2,000

Figure 1 Mortgage-Related Securities Issuance

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Agency

Non-Agency

Source: Securities Industry and Financial Markets Association(2010)

0

100

$ in Billions 200 300

400

500

Figure 2 Non-Agency MBS Issuance by Mortgage Type

2000

2001

2002

2003

2004

Prime Subprime Source: Inside Mortgage Finance(2009)

50

2005

2006 Alt-A

2007

2008

0

50

(bps)

100

150

Figure 4 1st AAA-rated Subprime MBS Issue Spread over Time

01jan2004

01jan2005

01jan2006 Issue date

1st AAA tranche issue spread

51

01jan2007 Median spline

01jan2008

100

Figure 5 Percentage of Subordination and AAA-rated Tranche 16.91

21.90

23.86

79.88

78.10

76.14

2005

2006

2007

0

20

40

(%)

60

80

83.09

20.12

2004

AAA-rated Tranche

Subordination

0

200

$ in Billions 400

600

Figure 6 Subprime Mortgage Origination and MBS Issuance

2000

2001

2002

2003

2004

Subprime Mortgage Origination Source: Inside Mortgage Finance(2009)

52

2005

2006

2007

2008

Subprime MBS Issuance

750 500

550

Credit Score 600 650

700

Figure 7 Average Credit Score over Time

01jan2004

01jan2005

01jan2006 Issue date

01jan2008

Median spline

Figure 8 Average LTV over Time

60

70

LTV (%) 80

90

100

Credit Score

01jan2007

01jan2004

01jan2005

01jan2006 Issue date

LTV

01jan2007

Median spline

53

01jan2008

0

Subordination level (%) 20 40

60

Figure 9 Subordination Level over Time

01jan2004

01jan2005

01jan2006 Issue Date

Subordination level

01jan2007

01jan2008

Median spline

0

Initial Overcollateralization (%) 5 10 15

20

Figure 10 Initial Overcollateralization over Time

01jan2004

01jan2005

01jan2006 Issue date

Initial Overcollateralization

54

01jan2007 Median spline

01jan2008

100 0

Delinquency Test Threshold (%) 20 40 60 80

Figure 11 Delinquency Test Threshold over Time

01jan2004

01jan2005

01jan2006 Issue date

Delinquency Test Threshold

01jan2007

01jan2008

Median spline

0

50

(bps) 100

150

200

Figure 12 AAA-rated Subprime MBS Issue Spread over Time

01jan2004

01jan2005

01jan2006 Issue date

AAA-rated tranche issue spread

55

01jan2007 Median spline

01jan2008

0

100

(bps) 200

300

400

Figure 13 AA-rated Subprime MBS Issue Spread over Time

01jan2004

01jan2005

01jan2006 Issue date

AA-rated tranche issue spread

01jan2007

01jan2008

Median spline

300 200 100 0

(bps)

400

500

Figure 14 A-rated Subprime MBS Issue Spread over Time

01jan2004

01jan2005

01jan2006 Issue date

A-rated tranche issue spread

56

01jan2007 Median spline

01jan2008

0

100

(bps) 200 300

400

500

Figure 15 BBB-rated Subprime MBS Issue Spread over Time

01jan2004

01jan2005

01jan2006 Issue date

BBB-rated tranche issue spread

57

01jan2007 Median spline

01jan2008