Complexity and Loan Performance: Evidence from the Securitization of Commercial Mortgages

Transcription

1 Complexity and Loan Performance: Evidence from the Securitization of Commercial Mortgages Craig H. Furfine Kellogg School of Management, Northwestern University Between 2001 and 2007, the complexity of commercial mortgage-backed securities (CMBS) increased substantially. The median size of commercial mortgage loan pools tripled and the median number of AAA-rated tranches doubled. I examine whether deal complexity is related to loan performance by analyzing a sample of approximately 40,000 commercial mortgage loans from 334 CMBS deals. I find that loan performance is worse for loans in more complex securitizations. However, neither the price of a deal s securities nor a deal s risk retention reflected that complexity correlates with lower loan quality. These findings present a challenge for theories of optimal security design. (JEL G14, G21, G23) One of the many facets of the recent financial crisis was the breakdown of securitization markets. This paper focuses on one such market, the market for commercial mortgage-backed securities (CMBS). At its peak in 2007, new issues of CMBS reached $191.7 billion before collapsing to less than $11 billion in 2008 and literally $0 in Not only did the primary market for CMBS literally disappear, losses on existing commercial mortgages grew dramatically, with delinquencies on loans securitized between 2005 and 2007 being roughly twice that found in similar loans sold between 2001 and There have been a number of reasons proposed as to why securitization markets fared so poorly during the financial crisis. Among the most common explanations are those related to incentive problems among the parties to the securitization process the originators, the underwriters, The author would like to thank Sumit Agarwal, Mike Fishman, David Matsa, Jonathan Parker, Mitchell Petersen, Amit Seru, Joshua Rauh (editor), and two anonymous referees for their helpful comments. Send correspondence to Craig H. Furfine, Department of Finance, Kellogg School of Management, 2001 Sheridan Road, Evanston, IL 60208; telephone: (847) c-furfine@kellogg.northwestern. edu. 1 Author s calculations derived from the Commercial Mortgage Alert CMBS database. 2 See Realpoint (2010). ß The Author Published by Oxford University Press on behalf of The Society for Financial Studies. All rights reserved. For Permissions, please journals.permissions@oup.com. doi: /rcfs/cft008 Advance Access publication January 17, 2014

2 Complexity and Loan Performance and the investors. According to common perception, firms originating mortgages quickly sold them, relieving them of any downside risk if a mortgage borrower ultimately defaulted. Similarly, underwriters pooling mortgage loans quickly passed along the risk of default to the investors of mortgage-backed securities (MBS). This originate to distribute model is believed to have led to originators becoming lax in their screening of risks, thereby reducing the quality of assets being securitized. As expressed by the Financial Crisis Inquiry Commission (2011), Collapsing mortgagelending standards and the mortgage securitization pipeline lit and spread the flame of contagion and crisis. Thus, it is no surprise that after the fact, financial regulators and policymakers have focused on risk-retention or skin in the game requirements as part of the reform of financial markets specified by the 2010 Dodd-Frank Wall Street Reform and Consumer Protection Act. 3 Although misaligned incentives and the resulting lax lending standards were undoubtedly an important determinant of the recent crisis in mortgage finance and beyond, this paper identifies another channel by which the securitization process might have influenced the likelihood that securitized loans default. The analysis focuses on complexity. In the years immediately preceding the financial crisis, the complexity of the typical commercial mortgage securitization increased substantially. For example, the median size of the pool of loans serving as collateral increased from just over $1 billion in 2001 to over $3 billion in The tranching of CMBS deals increased significantly too. I document a secular increase in the median number of tranches in a CMBS deal from 19 in 2001 to 26 in Much of this increased tranching occurred within securities initially receiving AAA ratings, with the median number of AAA-rated tranches increasing from five to ten over the same period. Increasing complexity has been argued to have played an important role in the duration and severity of the financial crisis (Caballero and Krishnamurthy 2008; Arora et al. 2009), in part by influencing how financial instruments were assigned credit ratings (Coval, Jurek, and Stafford 2009) and by affecting the efficiency of financial market trading (Carlin, Kogan, and Lowery 2013). As a complement to this previous work, this paper emphasizes that the complexity of a securitization reflects choices made by the underwriter, who decides not only which risky assets to pool, but also how to structure the securities being issued. It is important to emphasize that the complexity chosen by a securitization deal s underwriter is observable to investors. Thus, much of the existing theoretical literature might predict that deal complexity, like 3 This component of the law attempts to align the incentives of the various parties involved in securitizations by requiring securitizing institutions to retain no less than 5% of the underlying credit risk in the pool of risky assets being securitized. 155

3 Review of Corporate Finance Studies / v 2 n other observables, would become incorporated into an optimal security design problem whose equilibrium prices and quantities of various securities would reflect not only complexity, but all other characteristics of the loan pool and the deal. The key contribution of this paper is to demonstrate that within the CMBS market, there are four empirical findings that when considered together, are hard to reconcile within the context of such an optimal security design problem. These findings are: 1. Loans packaged in more complex securitizations defaulted more than would be expected given their observable characteristics. 2. Neither the price of a deal s securities nor a deal s risk retention (i.e., skin in the game) reflected that greater complexity correlates with lower loan quality. 3. The complexity loan quality relationship appears unrelated to ex ante underwriter reputation. 4. Underwriters originated and securitized high quality loans that they placed into deals demonstrating a complexity loan quality relationship. Complexity does not correlate with loan quality in deals without underwriter-originated loans. The first finding states the key result of the paper. Controlling for observable loan characteristics, subsequent movements in commercial property prices, and the identities of key institutions involved in the deal, I find that loan performance is worse for loans packaged in more complex securitizations. By itself, this does not imply that underwriters tricked their investors. That is, since complexity is observable, investors could incorporate complexity into the price that they are willing to pay for a deal s securities or the amount of risk they insist is being retained by the issuer. However, the second finding suggests that neither of these adjustments appear to have been made. Thus, the first two findings suggest that the CMBS market was in an equilibrium in which investors were unaware that complex deals contained loans of lower quality. One might have thought that such equilibrium would be difficult to maintain given that underwriters securitize many risky assets over time. Thus, underwriters would have an incentive to not mislead investors because of reputational concerns. The third finding, however, suggests that the complexity loan quality relationship is not being driven by deals underwritten by underwriters with low ex ante reputation. However, it could still be the case that underwriters placing lower quality loans into more complex deals may have had concerns regarding reputational losses should this relationship be discovered. If so, they could take additional steps in putting together loan pools to enhance their reputation. The fourth finding provides evidence suggesting that this was indeed done. Underwriters originated high-quality loans and placed them into complex deals containing other 156

4 Complexity and Loan Performance low-quality loans. There is no complexity loan quality relationship in securitization deals without underwriter-originated loans. The remainder of the paper is organized as follows. In Section 1, I describe the parties to a commercial mortgage securitization and emphasize how the related literature on security design is somewhat at odds with my findings. In Section 2, I introduce the data and explore how various aspects of commercial mortgage securitization evolved in the years preceding the recent financial crisis. In Section 3, I present the key empirical finding of a complexity loan quality relationship. In Section 4, I present additional empirical evidence suggesting that investors were unaware of the complexity loan quality relationship. I conclude in Section Parties to the Securitization of Commercial Mortgages and Theoretical Predictions of Their Interactions The parties involved in the securitization of commercial mortgages are similar to those involved in the securitization of any asset. First, there are the originators of loans, which in this paper, will be mortgages secured by commercial property. Such property typically consists of office buildings, retail establishments, industrial properties, apartment buildings, and other specialized real estate like hotels, medical facilities, or self-storage facilities. As part of the loan review process, these originators accumulate both hard and soft information. For example, various hard-information metrics, such as a loan s loan-to-value (LTV) ratio and debt service coverage ratio (DSCR) 4 are collected. At the same time, originators also collect soft, qualitative information regarding the borrower s likelihood of repayment. In a commercial mortgage context, this soft information might relate to information the originator collects regarding the probability that the tenants in the underlying collateral property will renew their leases. Because soft information is difficult to pass along to others in the securitization chain, originators will have information regarding a loan s quality that is superior to that of underwriters and investors. Having originated a commercial mortgage loan, the originator decides whether to keep the loan or sell it to another party. A commercial mortgage loan that is sold is typically securitized. The decision to keep or sell a loan involves evaluating the interaction of buyers and sellers of a commercial mortgage loan in an environment where the seller has superior information regarding loan quality. The theoretical literature discussed below describes a tradeoff faced by originators in this environment. On the one hand, originators wish to sell loans to free up resource constraints 4 The debt-service coverage ratio measures the ratio of the income generated by the property (through rents collected, etc.) to the debt service required by the loan. Thus, higher values of DSCR, all else equal, imply a safer loan. 157

5 Review of Corporate Finance Studies / v 2 n so that they may make additional profitable loans. On the other hand, originators may refrain from selling loans because they face a lemon s discount that arises from their informational advantage. The second party to the securitization process is the underwriter. The underwriter purchases commercial mortgages from one or more originators and at the same time, might originate loans too. The underwriter then determines which loans to place in a pool and how to structure the securities to sell to investors. Commercial mortgage loans placed into CMBS pools typically forbid loan prepayments, either through outright contractual bans, high prepayment penalties, or yield maintenance or defeasance requirements. 5 As a result, the underwriter s structuring of a CMBS deal focuses on default risk, which was addressed by a traditional senior-subordinate tranche structure of the securities. The underwriter s objective is straightforward. As long as the underwriter can sell securities for an amount greater than the cost to acquire the loans (net of transactions cost), the underwriter will continue to acquire loans and securitize them. Whether due to underwriter experience in buying and pooling mortgages or because of their own origination business, CMBS underwriters maintain an informational advantage over the investors of the securities. As will be described below, investors can be encouraged to purchase securities in which they are at an informational disadvantage if the underwriter retains some of the risk of the underlying pool. The intuition is that an underwriter signals a favorable opinion regarding loan quality by retaining some of the underlying risk. In equilibrium, it is the loan pools about which investors are most uncertain that require the highest degree of risk retention. The third party to the securitization process is a community of investors. These investors purchase the securities sold by the underwriter and receive cash flows backed by the payments received on the loans in the pool. Investors in CMBS are heterogeneous. Many investors of CMBS buy only the most senior, AAA-rated bonds, and thus can be viewed as being simply demanders of safe and liquid securities. In practice, these are the largest investors of CMBS and are typically financial institutions and money managers looking for a fixed income alternative to government bonds. At the other end of the credit spectrum would be investors that buy the most risky securities in the offering. These investors, known as B-piece investors, are high-yield investors with the commercial real estate expertise necessary to understand the risks inherent in the pool of underlying loans. As will be emphasized in the description of the empirical analysis, since the B-piece investors are purchasing the securities most exposed to default risk, they essentially replace the underwriter as the party with skin in the game. Between the 5 Defeasance requires a borrower seeking to prepay a securitized loan to place Treasury securities into the pool in an amount that would generate the originally promised principal, along with interest payments. 158

6 Complexity and Loan Performance institutional investors looking for fixed income securities and the commercial real estate experts who seek high yields in exchange for careful underwriting and analysis would typically be a set of mezzanine loan investors, who are a cross between the investors at either end of the capital structure. Although the risk-return tradeoff faced by each type of investor is rather different, all investors share the objective to acquire securities at no more than a fair risk-adjusted price. Interactions between these three participants in the securitization process have received substantial theoretical attention. 6 In a full-information model, where individual loan quality is observable by all, the originator of the loan, the underwriter of the deal, and the investors would all agree on the value of each loan, and thus, on the value of the deal s securities. Thus, the starting point of theoretical work examining optimal security design (i.e., the best way to sell claims on risky assets) necessarily involves consideration of information asymmetries among these three parties. For example, there are models of asymmetric information in an environment where buyers of risky assets are made up of differentially informed types of investors. For example, in Gorton and Pennachi (1990), uninformed investors would optimally form an intermediary that splits cash flows into riskless debt and risky equity. This allows the uninformed to trade riskfree securities to satisfy their liquidity needs without fearing losses that would be incurred by trading with better-informed investors. Gorton and Pennachi (1993) extend these predictions by modeling why the pooling of risky assets may be optimal. In their model, asset pooling helps to eliminate the problems associated with the information asymmetry between differently informed investors since it reduces the advantage of investors who are informed regarding an asset s risks. Thus, trading claims on a pool of assets minimizes the losses that liquidity-driven investors would incur when trading with others with superior information. Boot and Thakor (1993) propose a similar rationale for the tranching of securities. In their model, the issuance of securities with different seniorities backed by the same pool of risky assets is optimal because it makes informed trading of the informationally-sensitive security more profitable. Informed investors can better understand the risks of the information-sensitive security and are thus likely to value the security more highly than the uninformed. Uninformed investors prefer an informationallyinsensitive security, which they can acquire at its fundamental value. Researchers have also developed asymmetric information models where the asymmetry is between the owner/originators of a pool of 6 For empirical studies testing the importance of information asymmetries in securitization markets, see Ambrose, Yavas, and Sanders (2008), Downing et al. (2008), Keys et al. (2010), Purnanandam (2010), Titman and Tsyplakov (2010), Agarwal et al. (2011), An, Deng, and Gabriel (2011), Agarwal, Chang, and Yavas (2012), Benmelech, Dlugosz, and Ivashina (2012), Demiroglu and James (2012), and Keys, Seru, and Vig (2012). 159

7 Review of Corporate Finance Studies / v 2 n risky assets and an imperfectly informed investor. Leland and Pyle (1977) demonstrate that an entrepreneur can signal the quality of a project by agreeing to retain some of the underlying risk in that project. Since risk retention is costly, the signaling mechanism is credible. Riddiough (1997) applies this idea in the context of asset-backed security (ABS) design. In his model, the issuer can increase the proceeds from securitization by creating multiple securities, or tranches, with differing levels of exposure to the issuer s private information. The issuer then sells the least informationally-sensitive securities to avoid an adverse selection discount. Moreover, pooling assets that are not perfectly correlated can provide some diversification benefits and thus reduce the lemons discount. In a similar context, DeMarzo and Duffie (1999) present a model where an issuer signals his private information by retaining a portion of the security offered to investors. The design of offered securities reflects a tradeoff between the cost of risk retention and cost of illiquidity arising from informational sensitivity. DeMarzo (2005) applies this framework to the ABS market. His model explains the tranching of ABS as the result of the issuer optimally retaining the most informationally-sensitive portion of the security. Pooling arises from the consideration of two opposing forces. On the one hand, pooling risky assets is undesirable to the issuer due to an information effect, since it eliminates the issuer s asset-specific informational advantage. On the other hand, pooling is beneficial due to a diversification effect, since it allows issuance of securities that are less sensitive to an issuer s private information, thereby enhancing liquidity by alleviating the adverse selection problem. Finally, there are information asymmetries between the creator of a risky asset (e.g., mortgage originator) and secondary market investors in those loans. In Parlour and Plantin s (2008) model, banks have multiple reasons for wishing to sell their loans to outside investors. First, it could be a means by which the bank raises liquidity so that it can take advantage of investment opportunities. Second, the selling bank may be trying to take advantage of its private information. In this setup, the liquidity of the loan market is endogenously related to these two motivations for originators to sell. Hartman-Glaser et al. (2012) propose one way to address this information asymmetry between originators and investors. In their model, security design and sale take place together, yet the revelation of information regarding asset quality is observable over time. This allows a financial contract that incorporates the timing of payments as a critical design mechanism. In this dynamic setting, originators desire optimal effort, otherwise investors could effectively punish them over time and thus affect their future business opportunities. Malamud et al. (2013) extend this framework to consider a model where the optimal securitization structure involves investors paying originators over time but ceasing 160

8 Complexity and Loan Performance to do so after the first default. This mechanism improves originators incentive to screen. So what does the equilibrium would look like in the CMBS market? One would expect that information asymmetries among originators, underwriters, and investors would lead to pooling of loans, the tranching of securities for different investors, and the retention of risk (i.e., skin in the game) by the underwriter. In particular, the informational advantage of underwriters is understood by the collective set of investors and should therefore be reflected in the deal s risk retention, the design of securities sold, and the prices at which the securities will sell. The empirical findings of this paper, however, are not easily reconciled with this view of equilibrium in the market for securitized commercial mortgages. In particular, loans serving as collateral in more complex securitizations are more likely to become nonperforming. There is currently no theory to suggest why the complexity of a deal should be related to loan quality. Further, since complexity is observable, it should be reflected in a deal s security design. However, my evidence on risk retention and security pricing suggests that investors were unaware that more complex deals contained lower quality loans. This suggests an equilibrium in which investors are not able to decipher the true quality of the commercial mortgage loans being securitized. What sustains this equilibrium? One possibility may be the fact that investors simply did not appreciate that complexity was important. When commercial mortgage securitization began, deal structures were plain vanilla, which allowed investors to devote nearly all of their attention to analyzing the details of the underlying loans. In the years immediately preceding the financial crisis, however, deal complexity increased dramatically and investors may simply have continued to analyze loan pools as they had been doing in the past. A second possibility is that investors understood that both loan characteristics and deal complexity were relevant to learn about loan quality, but they had limited time. This notion is consistent with Hirshleifer et al. s (2009) model of distraction. In their model, firms optimally set their disclosure policies in an environment where investors have limited attention. With only so much time to analyze a securitization deal, investors optimally allocate time between details of the individual loans and the details related to the securities structure. In such a model, rational investors would equate the marginal benefits of an additional minute studying the loan pool with the marginal benefit of another minute studying the deal structure. Because greater complexity increases the marginal benefit of studying deal structure, a rise in complexity would cause investors to shift their limited attention away from an analysis of individual loans. Importantly, however, equilibrium in such a limited attention model 161

9 Review of Corporate Finance Studies / v 2 n might not lead to complete information revelation regarding the quality of the loan pool. Although this might explain why investors may not have learned that complexity provides information regarding loan quality, it cannot explain why ex post, underwriters would not suffer a loss of valuable reputation once the complexity loan quality relationship was revealed. That is, in a model of repeated interactions between underwriters and investors, one might expect an equilibrium in which underwriters are discouraged from exploiting their information advantage because of reputational concerns. That is, reputational concerns should limit an underwriter s desire to use complexity as a way to sell loans of low quality. The empirical evidence in this paper, however, suggests that reputational concerns were insufficient to prevent underwriters from introducing lower quality loans into more complex securitizations. In particular, measuring the reputation of the underwriter as the ex ante significance of that underwriter in the market suggests no connection between reputation and the complexity loan quality relationship. This suggests that underwriters may not have believed this activity raised sufficiently large reputational concerns. The evidence that the creators of complex financial products acted in ways consistent with a limited concern for reputation has strong support anecdotally (Lewis 2010). Thus, it may be reasonable to assume that underwriters believed that investors would be slow to discover or might never discover that loans in more complex deals were of lower quality. After all, it was only because there was a systematic decline in property prices that widespread commercial mortgage default and losses on CMBS became an issue. Thus, from an ex ante perspective that widespread loan defaults are unlikely, underwriters may simply have believed the marginal benefit of selling low quality loans exceeded the marginal cost of diminished reputation. Another reason why ex ante reputation may be unrelated to the complexity loan quality relationship is that underwriters can affect their reputation in ways other than placing low-quality loans in more complex deals. In particular, underwriters can affect their reputation by strategically mixing loans of different quality from different originators. Underwriters may have believed that reputational damage from a complexity loan quality relationship would be limited if the unexpected losses accruing to investors could be traced to loans originated by others. Consistent with this idea, I find that loans originated by the underwriter perform better than expected, whereas loans originated by someone other than the underwriter perform worse than would have been predicted by the readily observable characteristics of the loans. Further, the complexity loan quality relationship exists only among loans in securitizations where the underwriter has placed his own loans. 162

10 Complexity and Loan Performance 2. Deal Complexity and the Evolution of the CMBS Market In this study, I examine a sample of deals from the CMBS database maintained by Commercial Mortgage Alert, a commercial real estate finance trade publication. From that database, 357 deals were identified that were issued between 2001 and Each of these 357 deals was then searched on Bloomberg, and for 334 of these deals, Bloomberg contained the deal s Prospectus Supplement and the most recent monthly servicer report. The Prospectus Supplements contain two types of information: (1) information on the securities issued and (2) information on the underlying commercial mortgages that are in the collateral pool at the time the deal was issued. With respect to the information on the securities, the data lists, among other facts, each security s face value and the credit rating as assigned at issue by the leading rating agencies and the identity of the lead underwriter. To provide an example of the security structure information available, Table 1 reports the relevant information contained in the Prospectus Supplements for a particular deal underwritten by Lehman Brothers and UBS denoted LBUBS 2006 C1. For this particular deal, there were 34 securities issued that were backed by the cash flows arising from a common set of commercial mortgage loans. Of these, seven principal-receiving classes of securities initially received a AAA-rating from Standard and Poor s. The largest security issued in this deal, Class A-4, consisted of $1.1 billion of securities. The Prospectus Supplements also contain standardized information on the at-issue characteristics of the loans, including each loan s original balance, amortization period (if any), interest rate, loan-to-value (LTV) ratio, and debt service coverage (DSCR) ratio, as well as the location and type of the underlying collateral property. Panels A and B in Figure 1 demonstrate that in the years leading up to the financial crisis, the observable quality of securitized commercial mortgages declined. That is, CMBS pools securitized between 2005 and 2007 had higher median LTVs and lower median DSCRs than pools securitized in 2003 and Declines in observable loan quality reflect, in part, loosening of lending standards at commercial mortgage originators. A decline in lending standards alone, however, would not necessarily lead to problems in the securitization process. That is because originators, underwriters, and investors were able to observe this decline and according to theory, should have adjusted the details of the securitizations accordingly. 7 It is important to emphasize that the empirical analysis is not an examination of the determinants of loan performance per se. Rather, the purpose of the study is to determine whether loans placed in more complex deals were more likely to default when commercial real estate markets faltered during the recent financial crisis. Thus, loans that had matured or had been fully defeased in the years before the crisis are not central to the analysis, and deals issued before 2001 would have been made up almost entirely of such loans. 163

11 Review of Corporate Finance Studies / v 2 n Table 1 Sample securitization Class Amount (Millions of $) % of Deal Coupon rate Initial rating A % AAA A % AAA A % AAA A-AB % AAA A-4 1, % AAA A-M % AAA A-J % AAA B % AAþ C % AA D % AA- E % Aþ F % A G % 5.5 A- H % BBBþ J % BBB K % BBB- L % BBþ M % BB N % BB- P % Bþ Q % B S % B- T % NR IUU % NR IUU % NR IUU % NR IUU % NR IUU % NR IUU % NR IUU % NR IUU % NR IUU % NR X-CP(IO) 2, AAA X-CL(IO) 2, AAA This table lists the set of securities and their characteristics that were collateralized by the loans pooled for the LBUBS 2006 C-1 securitization, one of the 334 deals being examined. Class distinguishes different securities. Amount is the original face value issued of the particular class. % of Deal expresses the original face value as a fraction of the total deal size. Coupon rate is the promised interest rate on the security. Initial rating refers to the rating given to the security at issue by Standard and Poor s. All data come from the deal s Prospectus Supplement. However, at the same time that lending standards were being loosened, originators reliance on hard information was rising. Following Rajan, Seru, and Vig (Forthcoming), I report in Table 2 the R-squared from a series of interest rate models. In the baseline model, loan interest rates are regressed on a loan s LTV and DSCR. In the extended model, I include controls for the size of the loan, the type of collateral property, and the state in which the property securing the loan is located. The final column reflects the extended model estimated only for loans originated by the 17 leading originators in the sample, each of which originated at least 1,

12 Complexity and Loan Performance Figure 1 Developments in commercial mortgage securitization. Panels A D depict the evolution over time of LTV, DSCR, risk retention, and size, respectively. Panels E H depict the evolution over time of pool tranching, AAA tranches, pool loan concentration, and an indicator for using only two rating agencies, respectively. Pool loan-to-value is defined as the weighted (by original loan balance) average LTV (loan balance divided by property value) of all the underlying commercial mortgage loans, calculated at the time the pool was created. Pool DSCR is the weighted (by original loan balance) average DSCR (property net income to required debt service ratio) of all the underlying commercial mortgage loans in the pool, calculated at the time the pool was created. Pool risk retention is the fraction of the face value of securities sold that originally received a below investment grade rating. Pool size is the total mortgage balance of the (continued) 165

13 Review of Corporate Finance Studies / v 2 n Figure 1 Continued loans in the pool, calculated at the time the pool was created. Pool tranching counts the number of unique securities whose cash flows were backed by a common pool of loans. AAA tranches counts the number of unique securities within a deal to receive a AAA rating at issuance. Pool loan concentration is the sum of the squared loan shares where each loan share is defined as a loan s initial balance divided by initial pool size. Only two agencies are equal to one if the securities issued as part of a CMBS issue were rated by only two out of the three major rating agencies Moody s, Standard and Poor s, and Fitch, and is equal to 0, otherwise. In Panels A-G, the box graphed represents the inter-quartile range, with the line in the middle of the box depicting the median value. The dashes outside of the boxes measure the region of continuity of the distribution, calculated as 150% of the distance above (below) the 75 th (25 th ) percentile unless that value exceeds the maximum (minimum) value observed in the data. Dots outside of the dashes represent outlier observations in the data. Panel H plots only the fraction of deals within a year that are rated by only two agencies. 166

14 Complexity and Loan Performance Table 2 Reliance of interest rates on LTV and DSCR Origination Year Observations Base Model Model with additional controls Observations Controls and limited to leading originators % 14.63% % % 13.71% % % 15.35% % % 21.54% % The table reports the adjusted R-squared from regressions of the commercial mortgage loan interest rate on the loan s initial loan to value (LTV) ratio and its debt service coverage ratio (DSCR). In the base model, only these two variables are included. The model with additional controls includes the log of the size of the loan, indicator variables for the type of collateral property, and indicators for the state in which the collateral property is located. The final column reports the findings from the extended regression, but for a limited sample of loans originated by the 17 banks in the sample that originated at least 1,000 loans. loans in the sample. Table 2 demonstrates that observable variables explain more of the variation in loan interest rates in 2006 and 2007 relative to earlier years. Rajan, Seru, and Vig (Forthcoming) interpret this finding as evidence that in the years before the financial crisis, originators increasingly relied upon hard information. These findings suggest that while banks were loosening standards, they were conceivably exerting less effort to collect soft information that would be predictive of future default. According to the security design literature, if less soft information was being collected, then the underwriter would seemingly have less of an information advantage over investors. This may explain why at the same time that the use of hard information was rising, risk retention was falling. Panel C in Figure 1 describes the evolution of B-piece size, which as will be argued below, is a measure of risk retention in the context of commercial mortgage securitization. In 2001, nearly 8% of the typical CMBS deal was below investment grade and purchased by specialized investors. By 2004, this fraction had fallen to approximately 3%. Put another way, in the years immediately preceding the financial crisis, approximately 97% of the face value of CMBS received an investment grade rating. 8 Another reason why risk retention may have fallen in the years before the crisis is that while the risk of any individual loan may be increasing, the risk of the overall pool may be declining due to increases in pool diversification. Panel D in Figure 1 is suggestive of this possibility by graphing the evolution of pool size. Relative to the size of CMBS loan pools from early in the decade, pools formed in 2007 were typically more 8 The fraction of the face value of securities sold to B-piece investors is, by construction, equal to the subordination level of the lowest investment-grade security. 167

15 Review of Corporate Finance Studies / v 2 n than three times as large, with the median pool size approaching $3 billion, nearly twice the size of deals just a few years earlier. The data also demonstrates that in the years immediately preceding the financial crisis, the complexity of the typical CMBS deal rose as well. First of all, as shown in Panel D in Figure 1, deals became larger, which may reasonably proxy for complexity. Larger deals incorporate more loans, more types of underlying collateral (e.g., retail, office, apartments), and more geographic variation. Analysis of such deals therefore requires a wider range of understanding of commercial real estate fundamentals. Another measure of deal complexity is the number of tranches in the deal. Greater tranching complicates the deal structure since it makes the nonlinearity of payoffs more extreme, and therefore more sensitive to one s assumptions regarding underlying mortgage performance. Panel E in Figure 1 demonstrates that in the years before the financial crisis, tranching complexity increased, too. Overall, the median CMBS pool had 19 tranches in 2001, but by 2007, this number had grown to 26. Panel F in Figure 1 demonstrates that most of this increase is attributable to a rise in the number of AAA-rated tranches. Over the sample period, the median number of AAA-rated tranches doubled from 5 to 10. Existing theories of security design would have difficulty explaining the large number of identically rated securities being backed by the same loan pool. Typically, tranching is viewed as a response to investor clienteles seeking different levels of risk. In the data, however, tranching is largely being done within securities of identically rated risk. This suggests that future theoretical work might wish to contemplate different dimensions in which investor clienteles might vary. For example, purchasers of AAA-rated CMBS differ in ways reflecting design features other than credit risk. The ability to design a security to meet the demand of one or a small number of investors is evidenced by the fact that many AAA-rated CMBS tranches were $10 $20 million, whereas the largest AAA-rated tranches often exceeded $1 billion in face value. An example of a design feature that varies across multiple AAA-rated securities within a given deal is a difference in payment priority, which impacts a tranche s weighted average expected life. Within a given deal, the shortest lived AAA-rated CMBS might be expected to be repaid in as little as three years, while others would likely last ten years or longer. Another security design feature that might proxy for complexity is the concentration of the underlying loan pool. The more concentrated the loan pool, the more that investors with limited attention might focus on the ten largest loans for which they are given the most information. Typically, the largest ten loans comprise over 50% of the entire mortgage pool, despite the fact that the typical pool would contain over 100 individual loans. Panel G in Figure 1 graphs the evolution of pool loan concentration, defined as the sum of the squared loan shares where 168

16 Complexity and Loan Performance each loan share is defined as a loan s initial balance divided by the pool s total initial size. This measure of complexity has no obvious trend in the years leading up to the financial crisis. A final proxy for deal complexity can be inferred from the underwriter s choice of rating agencies. 9 At the time the CMBS offering is being marketed to investors, deal structures are presented to multiple rating agencies simultaneously. Underwriters prefer agencies that are more generous with their highest ratings. Skreta and Veldkamp (2009) formally model the underwriter rating agency interaction and predict that as the complexity of a deal increases, rating agency opinion regarding the creditworthiness of any given security issued as part of the deal is likely to become more varied. This disagreement could incentivize security issuers to choose only agencies willing to give the highest rating. Therefore, the number of agencies selected to rate the deal is going to be positively correlated with the extent of agreement across agencies, and agreement is less likely when deals are complex. Note that this result is independent of the quality of the underlying loan pool. If all rating agencies understand that a pool contains low-quality loans, securities backed by the pool would still receive identical ratings from each agency. The ratings would be lower on average (or subordination levels would be higher for any given rating), but nonetheless would be identical across agencies. Panel H in Figure 1 depicts how the frequency of having only two agencies rate a CMBS deal changed over time. Receiving ratings from two agencies was the norm in the CMBS market throughout much of the previous decade. In the years immediately preceding the financial crisis, however, it became more common to have three agencies rate a deal. In part, this change may reflect the increasing relevance of a third agency, Fitch Ratings, rather than a market-wide decline of complexity. Thus, the data indicate that two of the four proxies for deal complexity increased in the years preceding the financial crisis at the same time that observable loan quality declined. As my main empirical finding is that complexity is positively correlated with ex post loan performance, the time series correlation of measures of complexity and observable measures of loan quality raises the concern that the empirical finding is spurious. This issue will be addressed in the empirical analysis, in particular by including dummy variables for each quarter of the sample period. That is, the empirical result regarding a complexity loan quality relationship is found in a cross-section of deals that came to the market during the same calendar quarter. Despite the fact that time series movements in complexity will be incorporated into the empirical models, it remains useful to reflect 9 See also Stanton and Wallace (2010) and Ashcraft et al. (2011) for analysis of the role played by rating agencies in the securitization process. 169

17 Review of Corporate Finance Studies / v 2 n upon why the features of commercial mortgage securitization changed in the way that they did. Theory suggests two possibilities. First, it could be that the incentives of the various participants were changing. Recall the evidence on risk retention, which indicated that the role of the B-piece investor was falling at the same time that complexity was rising. As mentioned, this may result from the fact that the use of soft information was falling. Nonetheless, this finding is somewhat counterintuitive if one believes that complexity provides a means by which underwriters have an information advantage over investors with limited attention. In a model that incorporates a link between complexity and information asymmetry, one would predict that the role of the B-piece investor would have risen too. Another possible explanation of the decline in risk-retention is that B-piece investors saw a substantial increase in the liquidity of their investments at the same time that complexity was increasing. This was driven by the proliferation of commercial real estate collateralized debt obligations (CRE CDOs). These new financial instruments were a form of second-round securitizations, where B-pieces from multiple CMBS deals served as the assets that backed the issuance of CDO securities. 10 Because B-piece investors saw a way to exit their investments that had not previously existed, it is possible that they had less incentive to monitor the risks in the original CMBS loan pool. In the extreme case, when CRE CDOs were established at the same time as the CMBS, B-piece buyers would literally flip ownership of the B-piece into the CDO. Thus, in practice, CMBS with B-pieces in CDOs had no investor with skin in the game. Theory also suggests a second possibility as to why the complexity of CMBS increased in the years preceding the financial crisis. It could have been driven by increased demand by a wider array of investor clienteles, who may have demanded additional complexity in the form of specialized tranches with particular cash flow characteristics. As described, the very small AAA-rated tranches are suggestive of investor clienteles, a suggestion supported by the fact that such small tranches were often privately placed. An increase in clienteles for AAA-rated CMBS may also have arisen due to market conditions. Credit spreads in the years prior to the financial crisis were small, and fixed income investors may have convinced themselves that specially catered AAA-rated CMBS were a way to pick up yield without sacrificing the safety as measured by its credit rating. 3. Deal Complexity and Loan Performance This section formally documents the correlation between deal complexity and loan performance that exists after controlling for variation across 10 The typical CRE CDO combined CMBS B-pieces with other real estate debt, such as mezzanine loans, to form a pool of risky assets that were then securitized. 170

18 Complexity and Loan Performance time, as well as for the variation in observable loan characteristics. Loan performance data comes from each CMBS deal s most recent monthly servicer report, which describes each loan s current performance status. Note that this data reflects each loan s status as of the end of the month summarized by the report. As the data were collected manually, the process extended over a period of weeks between March and April, I refer to a loan s status as of the first quarter of More precisely, it measures how each loan was described in the most recently available servicer report, which reflects either February or March, 2010 remittances. Note that due to the need to collect this data manually, loan performance is only observed at this single point in time. 11 Table 3 exemplifies the loan-level data collected from the Prospectus Supplements and from the most recently available servicer report for five of the 120 loans contained in the securitization described in Table 1. Recall, the underwriting characteristics are from when the collateral pool was finalized, but the status of the loan is as of the first quarter of The sample of loans was further trimmed by dropping loans secured by multiple properties or loans that were part of a pari passu structure. 12 The final sample contains 40,172 loans from 334 different deals. The dependent variable in the initial analysis is an indicator variable that equals 1 if as of the first quarter of 2010, a given loan is nonperforming, and is equal to 0, otherwise. Nonperforming is defined as a loan that is late, delinquent, currently in foreclosure, or has been through foreclosure as of the first quarter of My focus is on whether loan performance is related to the complexity of CMBS deals. The previous discussion motivates my empirical proxies measuring such complexity. I measure pool size as the log of the total dollar value of the underlying loan pool. I measure the number of tranches in the deal, but also consider additional specifications where I distinguish tranches according to whether they were initially AAA-rated, below-investment-grade rated, or rated somewhere in between. I measure pool concentration as the sum of the squares of each loan s share of the overall pool. Finally, I construct a measure of rating agency inferred complexity that takes into account that in my sample, all deals are rated by at least two agencies out of the leading three rating agencies Moody s, Standard and Poor s, and Fitch. Thus, I measure rating agency 11 Because the data measures loan performance at a single point in time, it is impossible to identify loans that had experienced previous late payments, distress, or any other measures of nonperformance. 12 Pari passu loans, where a single large loan was split into different deals, would mechanically introduce correlation across the performance of individual loan observations, since multiple loans would link to the performance of the same underlying property. Robustness checks indicate that the empirical results are not sensitive to the inclusion or omission of such loans. 13 Extensive robustness checking confirms that alternative multinomial and ordered analyses with various definitions of loan nonperformance lead to the same qualitative conclusions. 171