What lies beneath: An inside look at corporate CLOs collateral

Transcription

1 What lies beneath: An inside look at corporate CLOs collateral Efraim Benmelech a Harvard University and NBER Jennifer Dlugosz b Harvard University Victoria Ivashina c Harvard Business School This draft: November, 2008 Abstract Since 2000 collateralized loan obligations (CLOs) have been the dominant source of capital for the high-yield corporate loan market. Despite the widespread belief that the rise of CLOs has led to riskier lending, there is little evidence to support it. In this paper, we address this issue by examining the quality of individual loans held by CLOs. We find that, overall, securitized loans perform no worse than unsecuritized loans in terms of accounting returns, credit rating changes, and market-assessed probability of default. However, within a CLO portfolio, loans originated by the bank that acts as the CLO underwriter significantly underperform the rest of the loan portfolio. Keywords: Structured finance, collateralized loan obligations, syndicated loans We thank Harvard Ph.D. Seminar participants for helpful comments, and Division of Research at Harvard Business School for research support. a Harvard University, Littauer Center, Cambridge, MA effi_benmelech@harvard.edu. b Harvard University, Littauer Center, Cambridge, MA jdlugosz@hbs.edu. c Harvard Business School, Baker Library 233, Boston, MA vivashina@hbs.edu.

2 The 2007 financial crisis brought to a close an extended period of growth in structured finance markets. Collateralized debt obligations (CDOs), special-purpose vehicles created to invest in pools of non-investment grade securities, suffered a major blow to their reputation after a record-breaking wave of downgrades and bank losses tied to CDOs. Global CDO issuance in the first half of 2008 fell to 10 percent of the amount issued during the same period in While most of the downgrades affected CDOs that were collateralized by sub-prime mortgages, there is lingering concern that strong demand for securitizable assets may have led to risky lending in the corporate sector as well, and that deterioration in credit quality might force a re-evaluation of some of the corporate loan backed CDOs (CLOs). Indeed, CLOs played a key role in financing billions of dollars of private equity firms leveraged buyouts around the world. Forty percent of all buyout deals done between 1997 and 2007 took place after 2004 (Stromberg, 2008; Kaplan and Stromberg, 2008), a period that coincided with heavy CLO issuance. A Wall Street Journal article from spring 2007 warned that CLOs could be next to default: 5 We are witnessing a loan market rife with liquidity and disproportionate power in the hands of borrowers, arrangers, and financial sponsors Investors searching for higher yields have put so much money into CLOs that even weak companies can get loans at relatively low interest rates These days, banks that arrange large buyout financings hold on to very little of the loans themselves. 4 Securities Industry and Financial Markets Association (SIFMA), Global CDO Market Issuance Data. 5 In what follows we use indistinctively CLO, CDO and structured investment portfolio to refer to collateralized loan obligations. 1

3 Bank underwriting standards have slipped as banks have become mere intermediaries. 6 In this paper, we investigate if securitization led to risky lending in the corporate loan market by examining the performance of loans held in CLOs portfolios. Focusing on the sample of loans that are likely to be held by institutional investors, we find that borrowers whose loans are securitized are more leveraged than unsecuritized borrowers at origination. However, controlling for firm characteristics, securitization does not predict poor future performance in terms of accounting measures, credit ratings, or market-assessed probability of default. This result might not be surprising because a CLO has a portfolio underwriter in addition to portfolio manager, both responsible for screening the quality of the collateral. However, some lenders not only originate loans in the primary market, but also underwrite, or even manage, CLO portfolios. Underwriting a CLO potentially provides a lending institution with the opportunity to pass bad loans to CLO investors, as there is one less screener of loan quality at the CLO level. Loans that appear in a CLO underwritten by the same lead arranger have lower credit ratings and ROA than other securitized loans before they are purchased by the CLO. Yet even after controlling for observables at the time of purchase, same bank loans underperform other securitized loans. They have significantly lower industry-adjusted ROA in the year of securitization, and they have significantly larger increases in CDS spreads in the longer-term, around two years post-securitization. They are no more likely to be downgraded postsecuritization than other securitized loans but they are less likely to be upgraded. 6 Easy Money: Behind the Buyout Surge, a Debt Market Booms -- CLOs Spark Worries of Volatility and Risk; Loan Standards Loosen, The Wall Street Journal, 26 June

4 Our findings cast doubt on the commonly held belief that corporate loans sold to CLOs are worse quality than unsecuritized loans. However, we find that agency problems exist for a subset of securitized loans, when a loan is purchased by a CLO underwritten by the loan arranger. This implies that underwriters potentially play an important role in certifying the quality of assets in securitized pools. Our paper is related to a growing literature that argues that securitization has some negative consequences. 7 Keys, et al (2008) show that subprime mortgages that are securitized are more likely to default than a set of mortgages with similar characteristics that have a lower probability of securitization. Our paper also relates to the extensive literature on financial intermediation and how loan sales and syndications affect lenders incentives to screen and monitor borrowers (Diamond and Rajan, 2001; Drucker and Puri, 2007; Ivashina, 2007; Sufi, 2007). Finally, it relates to Drucker and Mayer (2008) which shows that underwriters of mortgage-backed securities have valuable private information about their quality. The rest of the paper is organized as follows. First section provides institutional details on CLOs and the process of securitizing corporate loans. Sections two lays out the hypotheses and section three describes the data. Sections five presents the empirical results, and section six concludes. I. Institutional Details The first CLO completed by a U.S. bank was structured in late Since then, the CLO market has experienced explosive growth, reaching $180 billion in Table I provides statistics on the size of the CLO market relative to the CDO market as a whole. From Several papers have documented the benefits of securitization. Loutskina (2006) and Loutskina and Strahan (2007) show that securitization decreases the sensitivity of lending to banks financial conditions. 3

5 through the second quarter of 2008, 31 percent of CDOs issued globally were CLOs backed by high yield corporate loans. 50 percent of CDO global issuance was structured finance CDOs. 8 In this paper, we focus on CDOs backed by high-yield loans. 9 Most of these loans are syndicated; that is, they are originated by a lead bank which retains only a fraction of the loan, and sells the rest to the other banks and institutional investors. It is estimated that structured investment vehicles represent approximately sixty percent of institutional participation in the primary loan market. 10 In addition to acquiring parts of loans at the syndication, CLOs can also purchase loans in the secondary market. Figure I illustrates the securitization process. A CLO manager puts together a portfolio of corporate loans and issues securities to investors backed by the principal and interest payments from the loans. The CLO itself is a bankruptcy-remote special-purpose vehicle (SPV) set up by the issuer or collateral manager, usually an investment management company, to acquire the loans and issue the securities. As in other types of CDOs, the distinctive feature of a CLO is the tranching of its liability structure. A CLO has several classes of investors whose claims to the underlying assets are prioritized. As principal and interest payments are made on the underlying loans, proceeds are distributed to CLO investors in order of seniority. Investors are impacted by defaults in the underlying collateral pool only after subordinate classes have been exhausted. [FIGURE 1] 8 Structured finance CDOs are those backed by collateral that is itself structured (e.g. asset-backed securities, credit default swaps, or other CDOs). Most mortgage-backed CDOs would fall into this category because mortgages are usually pooled into pass-through securities, which are then purchased by CDOs. 9 Benmelech and Dlugosz (2008) collect data on all CDOs backed by loan collateral in the S&P RatingsDirect database and find that the weighted average rating of collateral pools ranges from BB to CCC+. Investment-grade collateral is more commonly observed in synthetic CDOs, which use credit default swaps as collateral. 10 Loan Syndications and Trading Association, The Handbook of Loan Syndications and Trading. The McGraw-Hill Compaies, Inc., New York, NY. 4

6 A bank can structure a CLO backed by loans it originated to reduce its risk exposure. More often, however, CLOs are structured by non-lending institutions that acquire loans on the open market. The motivation for these arbitrage CLOs is that the issuer earns fees from the spread between the relatively high-yielding assets and the low-paying liabilities. For example, Benmelech and Dlugosz (2008) show that the median CLO invests in a portfolio of B+ rated loans on average and funds itself with 73% AAA-rated liabilities, 8% unrated equity, and the rest in notes rated AA-BBB. Spreads on B-rated loans have ranged from 250 to 350 basis points over Libor in the recent past while the average AAA-rated CLO tranche pays 32 basis points over Libor. In arbitrage CLOs the most junior tranche (equity tranche) is typically sold to investors rather than being retained by the CLO manager. Thus, the CLO manager rarely has a stake in the SPV, and instead receives a fixed fee for collateral selection, and management. In addition to the collateral manager, a CLO has an underwriter that is responsible for assessing the risk of the collateral, structuring the deal and working with the rating agencies. While the collateral manager has formal authority over asset selection, the underwriter likely exerts influence over collateral choice. CLO underwriters are primarily banks. For 10% of the securitized loans in our sample, the loan lead arranger is the same as the CLO underwriter. CLOs differ from other institutions that participate in the high yield loan market because their demand for assets is driven almost solely by spread and rating. The CLO manager s goal is to structure a transaction that achieves the minimum cost of funding for the highest-yielding collateral. A CLO s cost of funding is largely determined by the ratings given to the notes that it issues. Rating agency models use only basic indicators of credit quality to assess the default risk of CLOs underlying collateral. At least one rating agency uses only rating, maturity, seniority, 5

7 jurisdiction, and industry to compute an expected loss distribution for the underlying collateral. 11 In other words, loans within a given rating class are treated as equally risky. Selecting loans that are better quality along dimensions are unmeasured by the model (e.g., covenants) provide the CLO with no cost of funding advantage vis-a-vis the rated noteholders. Meanwhile, selecting better quality loans would likely reduce the yield on the assets, leaving less to pay equity investors their required return (approximately 15%). Depending on how sophisticated equity investors are whether they analyze the assets individually or not - CLOs might skimp on unmeasured dimensions of loan quality in order to increase the yield on their collateral pools. II. Testable Hypotheses To receive a loan companies undergo due diligence and are regularly monitored in the process of lending relationship. Thus, the constant flow of private information about the firm gives banks an information advantage over the public capital markets participants. A large empirical literature, dating back to Fama (1985) and James (1987), support this claim. Among institutional investors who buy loans, CLOs probably have the weakest incentives to screen and monitor borrowers. The rating model uses only the most basic indicators of borrower quality to measure collateral quality and rate the deal, primarily the rating of the underlying assets, so CLOs incentive to perform extensive credit analysis is muted. Unlike other investment managers, whose returns may be carefully examined and benchmarked to other managers, CLO managers earn yearly management fees on their deals no matter how they perform. Once a deal has been rated and issued, the fee-maximizing manager faces two constraints. Deterioration in collateral quality (i.e., downgrades in the collateral) can force the manager to buy new collateral or else pay down notes. Hence a CLO manager cares about the 11 Benmelech and Dlugosz (2008) and Coval, Jurek, and Stafford (2008) provide extensive detail on rating models 6

8 ratings of his collateral assets because too many downgrades can terminate a deal. The manager also faces a reputational constraint. When assets in the collateral pool miss payments or default, the deal s equityholders bear the loss. If equityholders do not earn an adequate return, the manager may have difficulty selling the equity tranche in future deals. Despite reputational constraints, the rating model and the compensation structure for CLO managers indicate that CLOs probably put less effort into borrower analysis than other institutional investors. Hypothesis 1: If banks have inside information about the loans they originate, and they expect CLOs to be the least discriminating of investors, they will sell worse quality loans to CLOs. We test whether, controlling for observables at the time of the sale, loans sold to CLOs perform worse than institutional loans without CLO investors. We use three measures of performance to test this prediction: borrower accounting returns, credit rating changes, and changes in market-assessed probability of default (measured with CDS spreads). If the presence of CLO investors in a loan predicts poor performance, one interpretation is the information story described above. However, we cannot determine whether CLOs are being deceived about loan quality without knowing the price they paid for the loans. If CLO collateral managers are aware of the agency problem, they will adjust the price they are willing to pay for any of their underwriter s loans to account for the lemons problem. Kroszner and Rajan (1994), studying securities underwriting prior to the passage of Glass-Steagall in 1933, found that investors were aware of the conflicts of interest involved when a bank lends to a company and underwrites its securities. They found that securities underwritten by universal banks were discounted relative to securities issued by pure investment banks. 7

9 Some banks participate on both sides of the loan market arranging loans and underwriting (or even managing) CLOs that buy them. When a bank underwrites a CLO, this presents the greatest opportunity to unload bad loans, because there is one fewer monitor of loan quality at the CLO level. In other words, even if CLOs do not end up with worse quality loans than other institutions on average, they may end up with worse quality loans when they buy them from the underwriter of their deal. Although the collateral manager has formal authority over what collateral to buy, the CLO underwriter has the opportunity to exert influence. 75% of the securitized loans in our sample are purchased at syndication by CLOs. When a bank is arranging a loan and looking for investors to fill the syndicate, it might contact some of the CLOs it has underwritten. 12 Hypothesis 2: Loans that appear in a CLO underwritten by their lead arranger ( same bank loans) are worse quality than other securitized loans. We test whether same bank loans perform worse than other securitized loans, controlling for observables at the time of purchase. Again we use three measures of performance - accounting returns, credit rating changes, and changes in CDS spreads. We expect to find greater agency problems when the collateral manager is relatively inexperienced or has a weaker relationship with the underwriting bank, and we test for these interactions as well. It is worth noting that most loans sold to CLOs are syndicated loans. The syndicated loan market has established mechanisms for dealing with the agency problems created by risksharing, namely the requirement that the lead bank (the primary information gatherer) retain a 12 Most CLO transactions make continue to make some sales and purchases of collateral after they are issued. This is described in more detail in the next section. 8

10 share of the loan (Ivashina, 2007; Sufi, 2007). Despite the lead bank share mechanism, banks might decide to underwrite marginal loans if they bring in lucrative fees (e.g., by generating advisory revenue on M&A or LBO deals) or help to maintain relationships with an important client, like a private equity shop. III. Data A. Sample construction We assemble a dataset that identifies loans contained in specific CLOs. To the best of our knowledge, this is the first study that looks inside a CDO s portfolio and identifies the individual assets in its collateral pool. Our starting sample is all institutional loans to U.S. companies in the Reuters DealScan database identified as Term loan B or C. 13 We also include all loans that have hedge funds, mutual funds, pension funds, distressed funds or structured financial vehicles in the lending syndicate. We use two methods to identify CLOs loan holdings. CLOs often invest in a loan at origination. The list of the original lenders and other information available at the loan origination is collected by DealScan. Thus, we check the names of the syndicate investors to determine if a piece of a given loan was acquired by a CLO. The identity of the investors is cross-checked with the list of outstanding CDOs. Looking at the primary market data allows us to capture only a fraction of the loans held by structured financial vehicles, because loans can also be acquired on the secondary loan market. We detect secondary market purchases using SEC filings of loan amendments. A material loan amendment, such as change in spread, pricing grid, repayment schedule, maturity or loan amount requires unanimous approval of all lenders. The amendments are typically disclosed as part of the SEC filings, with signatures and identities of the lenders appearing at the 13 Term loans are installment loans (like mortgages or student loans) typically issued for specific corporate purposes. 9

11 bottom of the document. 14 We collect the first amendment for each loan in our starting sample and search the signers for entities that appear to be CLOs. If we do not find any CLO investors in DealScan or on the amendment, a loan is labeled as unsecuritized in the period between the loan origination and loan amendment. Overall we find 398 securitized loans: for 284 (71 percent) we detect CLO investors only through DealScan; for 113 (28 percent) we find CLO investors through amendments in addition to the CLO investors picked up by DealScan; and for 1 loan we identify CLO investors only through amendments. There are multiple CLO investors per loan. Specifically, we identify 5,988 loan-clo pairs covering 1,586 unique investors and 398 unique loans. 231 institutional loans that were amended did not have CLO investors and were defined as unsecuritized. For our key analysis, we need CLO characteristics. Thus, we collect the CLO s issue date, underwriter, and collateral manager from one of three sources: (1) Reuters CDO pipeline, (2) S&P s Quarterly CDO Deal List, and (3) S&P RatingsDirect. The deal list and RatingsDirect have substantial overlap but there are some transactions that appear exclusively in one or the other. The deal list summarizes all global CDOs rated by S&P from September 1994 to March RatingsDirect is a real-time database of the agency s ratings which allows us to identify more current deals but it drops information on securities when they mature or have their ratings withdrawn. Reuters tracks CLOs that invest in loans more generally, regardless of what agency rated them. Most of the CLOs in our sample were issued between 2000 and On average, we identify five loans per CLO. The median size of a CLO issued during that period was $460 million (Benmelech and Dlugosz, 2008) and the average minimum assignment size in the 14 A discussion on the requirements of the syndicate voting and public disclosure of the amendments can be found in Ivashina and Sun (2007). 10

12 institutional loan market is $5 million, so five loans represent roughly 5 percent of the collateral pool. Prior research shows that the typical CLO invests 90 percent of its capital in senior secured loans and the remainder in bonds, structured finance securities, and other loans. Additionally, most loans purchased by CLOs are leveraged loans with ratings in the BB or B range. 15 The securitized loan sample we have collected here conforms to that description. 99 percent of loans in our sample are senior secured and the average rating is BB-. Cash-flow CLOs purchase loans for collateral at various points in their lifecycle. The initial collateral pool is typically percent ramped up at the time the transaction closes and 100 percent ramped six months thereafter. Most CLOs are structured as revolving pools that allow the manager to turnover some fraction of the collateral for the first five to seven years of the twelve year life of the transaction (the reinvestment period ). A standard provision in the deals is that managers may sell securities that meet certain conditions (defaulted, credit improvement/credit reduction, etc) and make a limited amount of discretionary sales each year (usually percent). Slightly more than half of the loan-clo observations in our dataset represent supplemental collateral purchases. 54 percent have the loan origination date more than 180 days (six months) after the CLO issue date, indicating that the loan was probably not a part of the original collateral pool. 16 B. Potential selection biases The first part of our analysis is based on comparison of securitized (treatment group) and unsecuritized (control group) loans; therefore, it is important to consider whether our data 15 Benmelech and Dlugosz (2008) find that CLOs are typically backed by collateral pools with a weighted average rating of BB-/B+/B. Many restrict the amount of securities rated below CCC+ to 5-7 percent of the pool, suggesting that the average loan put in a CLO has a BB or B rating percent of loan-clo observations have the loan origination date after the CLO issue date. 11

13 collection method has introduced a selection bias. Our control group is constrained to the existence of a loan amendment. However, our treatment group includes amended loans and unamended loans that had CLO investors at the origination. To the degree that loans with and without loan amendments could be fundamentally different this would introduce a bias in the results. However, it is unclear if presence on an amendment reflects positive or negative news. 17 If observable amendments are a reflection of successful renegotiations (as opposed to failed renegotiations that are not observable) then our control group is on average of better quality. Alternatively, if most of the firms soliciting amendments and receiving amendments are troubled firms then our treatment group is on average of better quality. We address this issue by reexamining the results in the subsample where treatment and control group were constrained to the sample with loan amendments. Although we collect all first loan amendments, there is potential concern about misclassifying loans as unsecuritized (type I error) in that we only detect CLO ownership when an amendment requires unanimous agreement. Only material loan amendments require the approval of all lenders. A covenant waiver can be approved by a majority vote. Given that existence of a material loan amendment is likely to be correlated with loan quality, we could experience further selection problems. Specifically, we are concerned that all of the loans in our analysis are eventually securitized, but because some of them do not face a material amendment they are classified as unsecuritized. However, this is unlikely to be the case, given that the majority of our loan-clo pairs are identified via DealScan rather than loan amendments, which should mitigate selection concerns. In the full sample, 84 percent of loan-clo observations 17 Ivashina and Sun (2007) find that on average abnormal return on the stock or secondary loan market around loan amendments is zero as a result of offsetting reactions within the sample. 12

14 come from DealScan data and 16 percent come from loan amendments. At the loan-level, there is only one securitized loan whose identification as such relies purely on amendment data. In the second part of our analysis, we compare the quality of securitized loans where the loan arranger is the same as the CLO underwriter to securitized loans in general. Since this test is done within the subset of securitized loans, it is insulated from the selection concerns above. C. Summary statistics Table II, Panel A presents summary statistics on the loans in our sample. Institutional loans are large loans made to large borrowers, with a median loan size of $350 million and the median borrower having assets of $875 million. Securitized loans are not drastically different from unsecuritized loans in terms of loan and borrower characteristics. The median loan pays an all-in-drawn spread of 275 basis points over Libor and 40 percent of the loans in the sample are used to finance LBOs. The median borrower had sales of $638 million at the time the loan facility closed and implied leverage (deal size/sales at loan closing) of Table II, Panel B presents a more extensive description of the borrowers, using Compustat data for the fiscal year ending immediately prior to loan origination. Borrowers who get institutional loans have high industry-adjusted leverage and profitability. Panel B also compares loan and borrower characteristics across the securitized and unsecuritized subsets. Borrowers whose loans were securitized have significantly higher leverage, higher z-scores, and are larger than unsecuritized borrowers. Table III examines the same data in a multivariate setting. We run a probit where the dependent variable is a dummy that indicates whether a loan was securitized and the independent variables are loan and borrower characteristics at origination. Larger borrowers with higher 13

15 industry-adjusted leverage are more likely to have their loans sold to a CLO. A one standard deviation increase in the log of borrower assets (1.3) is associated with a 6.5 percentage point increase in the probability of being securitized. A one standard deviation increase in industry adjusted leverage (0.33) is associated with a 5.2 percentage point increase in the probability of securitization. The probability of securitization in our sample is 63%. It has been suggested in the media that CLOs are a source of easy money funding corporate activities indiscriminately. We test this hypothesis by including dummies for loan purpose in the regression. LBO loans and M&A loans are no more likely to be securitized. 18 Debt repayment loans are less likely to be securitized. Loans sold to CLOs appear ex-ante riskier than unsecuritized loans in some respects (higher leverage) and less risky in others (larger companies). When we constrain the sample to amended loans (specification 5), leverage at origination is no longer a significant determinant of securitization. [TABLES II & III] IV. Results A. Hypothesis I: Does Securitization Predict Performance? Collateral managers can observe ex-ante loan and borrower characteristics, so the results of the previous section do not indicate information asymmetry problems. In this section we test whether loans sold to CLOs are unobservably worse quality than loans sold to other institutions. To do this, we need to examine whether securitization predicts future performance, controlling for observables at the time the CLO buys the loan. We use three different measures of performance: (1) borrower ROA (2) changes in borrower credit rating and (3) changes in the 18 However, the number of CLO investors in a given loan is positively correlated with the LBO dummy. 14

16 borrower s CDS spreads as a proxy for the company s probability of default. We match each loan-clo pair in our data (treatment loans) to comparable unsecuritized loans (control loans) and compare performance around the date the treatment loan was purchased by the CLO. The idea is to compare each loan purchased by a CLO to other loans it might have purchased instead. Our matching process requires the following: Matched loan is arranged by the same lead arranger as the securitized loan in question Matched loan must be outstanding at the time the treatment loan was securitized. Matched loan must have similar time remaining (+/- 1.5 yrs) Matched loan must have been originated around the same time (+/- 2.5yrs) Matched loan must have similar rating (+/- 2 rating levels) We do not observe the actual date a CLO purchases a loan, but we proxy for it using the loan origination date and the date the CLO was issued. We set the securitization date for a loan- CLO pair equal to the later of the loan origination date and the CLO issue date. For 70% of loan-clo observations, the loan began after the CLO was issued, so the securitization date is the same as the loan origination date. We are able to find at least one matched loan for 2,538 loan- CLO observations. 19 The median loan-clo observation matched to three unsecuritized loans. We call the date the treatment loan was securitized the event date and we measure the performance of the treatment loan and the control loans in a window around this date. Table IV compares borrower ROA around the event date for loans chosen by CLOs and matched loans that could have been chosen instead. Borrowers whose loans are purchased by 19 Requiring an issue date for the CLO reduces our loan-clo observations down from 5,988 to 3,

17 CLOs outperform matched loan borrowers unconditionally, as well as when controlling for observable characteristics around the event date. Table IV, Panel A contains the univariate results. Borrowers whose loans are sold to CLOs have significantly smaller assets, higher leverage, and lower Q than the matched-loan borrowers (all industry-adjusted). However, they have higher industry-adjusted ROA than the matched loan borrowers in the year before the event and up to two years afterwards. Table IV, Panel B explores this comparison in a multivariate setting. Controlling for loan and borrower characteristics at the time a loan is purchased by a CLO, loans sold to CLOs outperform control loans in the event year and for two years afterwards. A 1 standard deviation increase (0.09) in the borrower s industry-adjusted ROA in the event year is associated with a 4.0 percentage point increase (0.09*0.44) in the probability that the loan was securitized. A similar relationship holds between each of the next two years ROA and the probability of securitization. Borrowers whose loans are sold to CLOs perform better than expected in each of the three years after securitization, controlling for observables at the time they were chosen. The performance differential at t+1 and t+2 should be interpreted carefully because survivor bias is large. The number of observations drops from 8,424 at time t to 7,169 at t+1, to 5,703 at t+2. As a robustness check, we run the regressions again, conditioning on survival to t+2. Among borrowers that survive until t+2, securitized loans outperform unsecuritized control loans significantly at t+1 and t We need to be cautious though because of the possible selection bias in our sample. Our securitized loans are a mix of amended and unamended loans while our unsecuritized loans are all amended. If amended loans are worse quality than unamended loans, this would bias us towards finding that loans sold to CLOs are better. As a robustness check, we re-run the 20 The coefficients and z-stats on ROA at t, t+1, and t+2 in this robustness check are (-0.69), 0.54 (5.42), and 0.51 (6.68), respectively. 16

18 regressions in Panel B, limiting the sample to amended loans. The coefficient on ROA in the year of securitization remains positive and significant; the coefficients on ROA in the following two years are not significant. 21 Together these results suggest that securitized borrowers outperform unsecuritized borrowers in terms of ROA. The window of outperformance depends on the sample used but, certainly, the performance of securitized loans is no worse than that of the matched loans. [TABLE IV] Next, we examine whether securitization predicts borrower downgrades or upgrades in credit ratings. Credit ratings are salient measures of loan performance for CLOs, because the model used by rating agencies to rate CLO securities uses the ratings of the underlying assets as inputs. A CLO that experiences too much deterioration in the ratings of its underlying assets will be forced to shore up credit support (by buying additional collateral or paying down liabilities), or else suffer downgrades in its issued notes. We compare the upgrade and downgrade frequency of loans sold to CLOs with unsecuritized, matched loans in Table V. Rating changes are measured from a scale that combines Moody s and S&P senior secured ratings for the borrowers. We convert each respective scale into a numerical one and set the borrower rating equal to the lower of the two ratings available on a given date. Credit watch negative or positive is valued at 0.5 of a rating class. Table V, Panel A presents the univariate results. We find that securitized loans are more likely to be downgraded and less likely to be upgraded than unsecuritized, matched loans. Once we control for loan and borrower characteristics at the event date (Table V, Panel B), loans sold to CLOs are no more likely to be downgraded than control loans but significantly less likely to be upgraded. Being purchased by a CLO decreases the 21 The coefficients and z-stats on ROA in time t, t+1, and t+2 respectively are: 0.58 (2.82), 0.09 (0.67), and (- 1.25). 17

19 probability that a loan is upgraded within the next two years, by one to nine percentage points depending on the specification. The results are qualitatively the same when we constrain the sample to amended loans. [TABLE V] The third way we measure performance is with credit default swap (CDS) spreads. CDS prices measure the cost an investor would have to pay to insure against a company s default. As a company s default risk rises, its CDS spread increases. The advantage of CDS data over accounting data is that CDS contracts will often continue to trade if a company is taken private, say, in an LBO. A large fraction of the loans in our sample are LBO loans, so measuring borrower performance with accounting returns in a long-term window encounters survivor bias issues. Using CDS prices as a measure of performance reduces survivor bias; however, it limits analysis to the largest companies in the sample because only large companies have liquid CDS contracts. The analysis for this section is ongoing, as we are in the process of getting improved CDS data from Markit. In summary, loans financed by CLOs are observably different from other institutional loans at origination. Loans with CLO investors finance larger borrowers that are relatively highly leveraged compared to their industry. However, controlling for observables at the time they are purchased, securitized loans outperform similar loans from the same lead arranger that are unsecuritized. One interpretation of this finding is that CLOs have skill in picking loans. Although borrowers they choose are more highly leveraged, they have above average profitability for their type. A caveat is that our matching process assumes that if a CLO did not buy the loan we observe it buying, its alternatives would have been other loans with similar characteristics from the same arranging bank. 18

20 B. Hypothesis II: When the CLO Underwriter is a Loan Arranger The findings of the previous section indicate that agency problems in securitization may be less widespread than commonly believed. Loans sold to CLOs do not, on average, perform worse than loans without CLO investors. Instead, by some measures, securitized loans outperform a matched sample of unsecuritized loans ex-post. In this section, we examine the performance of a subset of securitized loans for which we expect agency problems to be particularly pronounced. Some CLO underwriters participate on both sides of the loan market arranging loans and underwriting (and sometimes managing) CLOs that purchase loans for collateral. This presents the greatest opportunity for the bank to sell poor quality loans to a CLO, because there is one fewer monitor of loan quality at the CLO level. A bank that underwrites CLOs and arranges loans faces a conflict of interest between its duty to the collateral manager and its desire to find investors for its syndicates. For example, when a bank is arranging a loan, it might contact CLOs to see if they want to invest. It would be easiest for the bank to deceive its own CLOs (the CLOs it has underwritten) about loan quality, rather than other CLOs whose independent underwriters will be assessing the loan. We separate loan-clo observations into two groups: cases where the loan arranger is the CLO underwriter ( same bank ) and cases where the loan arranger is different from the CLO underwriter ( different bank ). Our hypothesis is that loans arranged by the same bank that underwrote the CLO will perform worse post-securitization than other securitized loans. We compare post-securitization performance along three dimensions: borrower ROA, credit rating changes, and CDS spreads. Table VI compares ROA for same bank loans and different bank loans in a three-year window after securitization. The univariate results in Table VI, Panel A indicate that same bank 19

21 loans perform unconditionally better than other securitized loans two years after securitization. However, there is substantial survivor bias two years after securitization. The sample size drops from 2,771 to only 1,931. Later, we will examine long-term performance more carefully using a measure that should not have survivor bias CDS spreads. Table VI, Panel B tests whether same bank loans perform worse ex-post, controlling for observable characteristics at the time of securitization. We find that borrowers whose loans show up in a CLO underwritten by their lead arranger perform significantly worse than other loans in the year of securitization. A one standard deviation increase in ex-post industry-adjusted ROA is associated with a 5 percentage point decrease in the likelihood that a loan shows up in a CLO underwritten by its lead arranger. This is economically large, since the underlying probability of securitization by lead arranger in our sample, conditional on securitization, is 26% (104/398 loans). The coefficient on ROA two years post-securitization goes in the opposite direction, suggesting that same bank borrowers outperform other borrowers in the long-term. However there is significant survivor bias in the t+2 sample; we lose 32% of our original observations. For this reason, we do not place much faith in this estimate. We will examine CDS prices as another measure of ex-post performance that is less subject to survivor bias and may be a better measure of long term performance. [TABLE VI] Table VII examines whether securitization by lead arranger predicts future rating changes. Table VII, Panel A shows that same bank borrowers are no more likely to be downgraded post-securitization than other securitized borrowers, but they are significantly less likely to be upgraded. Table VII, Panel B presents the multivariate results. After controlling for loan and borrower characteristics at the time of securitization, same bank borrowers are no more likely to be upgraded or downgraded one to two years after securitization. 20

22 [TABLE VII] Finally, we compare whether same bank borrowers experience larger CDS price increases than do other securitized borrowers. We use one short-term window (3-9 months postsecuritization) and one long-term window (9-21 months post-securitization). 22 Table VIII, Panel A presents the univariate results. CDS spreads increase significantly more for same bank borrowers 9-21 months after securitization. In a regression setting (Table VIII, Panel B), same bank borrowers experience an increase in CDS spreads 11 percentage points higher than other securitized borrowers on average. CDS spreads are a good measure of long-term ex-post performance because they do not suffer from the same survival bias as accounting measures. Even if a company undergoes an LBO, its CDS contracts will continue trading. Forty percent of the loans in our sample are LBO loans, so measuring long-term performance with accounting returns is difficult. On the other hand, CDS prices are only available for the largest companies in our sample. [TABLE VIII] Results suggest that conflicts of interest exist when banks that arrange loans also underwrite CLOs. Loans that appear in CLOs underwritten by their lead arranger perform worse than other securitized loans, controlling for observables at the time of purchase. If the explanation for this finding is that lending banks use their position as CLO underwriter to offload loans for which they have a private negative signal of quality, the agency problem should be more pronounced when the CLO manager is inexperienced and, perhaps, easier to fool. We test 22 We use 3-9 months instead of 0-6 months because it preserves data (loan-clo observations increase by 43% from 162 to 231). Our CDS data, from Datastream, begins in January 2003 for most borrowers. 21

23 for an interaction effect between the same bank dummy and CLO manager experience but it is insignificant. 23 The remaining question is one of interpretation. If loans purchased from the CLO underwriter are worse quality, do CLO managers overpay for these loans or are they aware of the conflict of interest? If CLO collateral managers are aware of the agency problem, they would rationally demand a lemons discount on loans purchased from their underwriter. Borrowing terms from Kroszner and Rajan, who studied conflicts of interest in security issuance by universal banks in the 1920s, are CLO managers naïve investors or are they rational discounters? Our finding that manager experience does not seem to matter suggests that the rational discounting interpretation may be the correct one. V. Conclusion Following the subprime mortgage crisis in 2007, the market for CDOs collapsed. Although most of the downgrades and defaults of highly-rated CDO securities were concentrated among CDOs with mortgage collateral, investors emerged wary of CDOs in general. Several recent papers examine what went wrong with the rating process. In this paper we examine a different concern - whether securitization led to risky lending in the corporate loan market. This is the first paper to test for agency problems in securitization using data that identifies the individual assets used as collateral in a CDO. Contrary to expectations, we find no evidence that loans sold to CLO investors are worse quality than loans sold to other institutional investors. However, we do find evidence of agency problems in a particular set of transactions. When a CLO purchases loans arranged by its underwriter, those loans underperform relative to other 23 Results available upon request from authors. 22

24 securitized loans, controlling for observables at the time of purchase. Our results provide broader insights about structured finance products. Securitization by itself does not necessarily lead to adverse selection in collateral quality. However, in the presence of agency problems and a lack of independent monitoring, collateral pools are likely to contain worse quality assets. 23

25 References: Benmelech, Effi and Jennifer Dlugosz, 2008, The Alchemy of CDO Credit Ratings, Working paper. Coval, Josh, Jakub Jurek, and Erik Stafford, 2008, The Economics of Structured Finance, Journal of Economic Perspectives, forthcoming. Diamond, Douglas, and Raghuram Rajan, 2001, Liquidity Risk, Liquidity Creation, and Financial Fragility: A Theory of Banking, Journal of Political Economy 109, Drucker, Steven, and Manju Puri, 2007, On Loan Sales, Loan Contracting and Lending Relationships, Working paper. Drucker, Steven, and Christopher Mayer, 2008, Inside Information and Market Making in Secondary Mortgage Markets, Working paper. Fama, Eugene, 1985, What s Different About Banks?, Journal of Monetary Economics 15, Ivashina, Victoria, 2007, Asymmetric Information Effects on Loan Spreads, Journal of Financial Economics, forthcoming. Ivashina, Victoria, and Zheng Sun, 2007, Institutional Stock Trading on Loan Market Information, Working paper, Harvard Business School. James, Christopher, 1987, Some Evidence on the Uniqueness of Bank Loans, Journal of Financial Economics 19, Kaplan, Steven, and Per Stromberg, 2008, Leveraged Buyouts and Private Equity, Journal of Economic Perspectives, forthcoming. Keys, Benjamin, Tanmoy Mukherjee, Amit Seru, and Vikrant Vig, 2008, Did Securitization Lead to Lax Screening? Evidence From Subprime Loans, Working paper. 24

26 Kroszner, Randall, and Raghuram Rajan, 1994, Is the Glass-Steagall Act Justified? A Study of the U.S. Experience with Universal Banking Before 1933, American Economic Review 84, Loutskina, Elena, 2006, Does Securitization Affect Bank Lending: Evidence from Bank Responses to Funding Shocks, Working paper. Loutskina, Elena, and Philip Strahan, 2007, Securitization and the Declining Impact of Bank Finance on Loan Supply: Evidence from Mortgage Acceptance Rates, Working paper. Standard & Poor s, 2002, Global Cash Flow and Synthetic CDO Criteria. Stromberg, Per, 2008, The New Demography of Private Equity, Working paper, Stockholm School of Economics. Sufi, Amir, 2007, Information Asymmetry and Financing Arrangements: Evidence from Syndicated Loans, Journal of Finance 62,

27 FIGURE I HYPOTHETICAL CLO SECURITIZATION Loan A (Lead arranger: Bank1) Loan B (Lead arranger: Bank2) CLO SPV Portfolio of Loans Cash flows from assets Proceeds from sale of securities to investors AAA tranche AA tranche BBB tranche Equity tranche Investors Agents: CLO manager CLO underwriter 26

28 TABLE I GLOBAL CDO MARKET ISSUANCE BY UNDERLYING COLLATERAL This table provides statistics on CDO issuance by type of underlying collateral. High yield loans are borrowers with senior unsecured ratings below Baa3 from Moody s or BBB- from S&P. Structured Finance CDOs are backed by structured collateral, i.e. these are portfolios of portfolios of securities. Most mortgage-backed CDOs (or collateralized mortgage obligations) fall into this category because mortgages are typically pooled into pass-through securities, which are then purchased as collateral by CDOs. The data comes from Securities Industry and Financial Markets Association (SIFMA). Total Issuance By Underlying Collateral: High Yield Loans Structured Finance Year $MM $MM % $MM % ,803 71, , , , , , , , H 36,807 17, , Total 1,363, , ,

29 TABLE II SUMMARY STATISTICS: UNSECURITIZED VS. SECURITIZED LOANS Loan variables are constructed using DealScan. Most loans are structured in several facilities. The assignment minimum is a minimum loan commitment for the syndicate participants. Borrower sales at close is taken at the time the loan closed. Implied leverage is the size of the loan divided by the borrower s sales at close. All-in-drawn spread is defined as total (fees and interest) annual spread paid over Libor for each dollar drawn down from the loan net of upfront fees. Senior debt rating and loan rating are S&P ratings; a larger number represents a lower rating. The performance pricing dummy indicates that the spread on the loan is tied to the firm's financial indicators like leverage and/or interest coverage. The LBO, debt repayment, and M&A dummies indicate the purpose of the loan. In Panel B, financial data corresponds to the fiscal year that ended just prior to loan origination. Industries are measured at the 2-digit SIC code level and industry adjusted figures have the industry median subtracted for the corresponding fiscal year. Compustat variables: Debt-to-assets is Total Long-Term Debt plus Debt in Current Liabilities divided by Total Assets. Loan size-to-assets is the facility size from DealScan divided by Compustat s Total Assets. Interest coverage is EBITDA divided by Interest and Related Expense. Return on assets is EBITDA/Assets. Q is calculated as [Assets + Market Value of Equity Book Value of Equity Deferred Taxes]/Assets. Z-score is 1.2*[Working Capital/Assets] + 1.4*[Retained Earnings/Assets] + 3.3*[EBIT/Assets] + 0.6*[Market Value of Equity/Book Value of Liabilities] *[Sales/Assets]. Return on equity is Income Before Extraordinary Items divided by Common Equity. Equity volatility is the volatility assumption used for calculating option values for financial reporting in the given year (optvol). Same bank indicates that a loan was sold to a CLO underwritten by its lead arranger. Significance at the 1, 5, and 10 percent levels is indicated by ***, **, and *, respectively. Panel A: Loan terms Full sample Unsecuritized loans Securitized loans Total Different bank Same bank Number of loans Deal size ($mil) median Facility size ($mil) median Number of facilities median Assignment min. ($mil) median Borrower sales at close ($mil) - median Implied leverage median All-in-drawn spread (bps) 25 th percentile Median th percentile Senior debt rating of borrower 25 th percentile Median th percentile Loan rating 25 th percentile Median th percentile Performance pricing (dummy) mean LBO loan (dummy) mean Debt repayment loan (dummy) mean M&A loan (dummy) mean