On the Securitization of Student Loans and the Financial Crisis of 2007–2009 by Maxime Roy Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy at Tepper School of Business Carnegie Mellon University May 16, 2017 Committee: Burton Hollifield (chair), Adam Ashcraft, Laurence Ales and Brent Glover External Reader: Pierre Liang À Françoise et Yvette, mes deux anges gardiens. abstract This dissertation contains three chapters, and each examines the securitization of student loans. The first two chapters focus on the underpricing of Asset-Backed Securities (ABS) collateralized by government guaranteed student loans during the financial crisis of 2007–2009. The findings add to the literature that documents persistent arbitrages during the crisis and doing so in the ABS market is a novelty. The last chapter focuses on the securitization of private student loans, which do not benefit from government guarantees. This chapter concentrates on whether the disclosure to investors is sufficient to prevent the selection of underperforming pools of loans. My findings have normative implications for topics ranging from the regulation of securitization to central banks’ exceptional provision of liquidity during crises. Specifically, in the first chapter, “Near-Arbitrage among Securities Backed by Government Guaranteed Student Loans,” I document the presence of near-arbitrage opportunities in the student loan ABS (SLABS) market during the financial crisis of 2007–2009. I construct near-arbitrage lower bounds on the price of SLABS collateralized by government guaranteed loans. When the price of a SLABS is below its near-arbitrage lower bound, an arbitrageur that buys the SLABS, holds it to maturity and finances the purchase by frictionlessly shorting short-term Treasuries is nearly certain to make a profit. The underpricing on some SLABS relative to Treasuries exceeded 22% during the crisis. In the second chapter, “SLABS Near-Arbitrage: Accounting for Historically Unprecedented Macroeconomic Events,” I analyze whether the risks associated with unprecedented macroe- conomic events, such as exceptionally high inflation or default by the government on its loan guarantee, could explain the large underpricing of SLABS relative to Treasuries observed during the financial crisis of 2007–2009. Using data on inflation caps, interest rate swaps and interest rate basis caps, and comparing the price dynamics of SLABS to other securities benefiting from a similar government guarantee, I find that for 90% of SLABS, the aforementioned risks explain at most 25% of the near-arbitrage gaps. In the third chapter, “Securitization with Asymmetric Information: The Case of PSL-ABS” (joint with Adam Ashcraft), we empirically analyze the adverse selection of loans in the private student loan (PLS) ABS market. Using loan-level data, we demonstrate the potential for an issuer of PSL-ABS to select loans in such a way that could result in materially adverse outcomes for investors (credit rating downgrades or market value losses). We find that an issuer could increase pool losses on the non-cosigned portion of securitized pools by 6%–20% among pre-crisis deals and by 16%–36% among post-crisis deals while still matching the pool characteristics disclosed to investors. The shifts in pool losses are achieved by exploiting the coarseness of the disclosure and by jointly overrepresenting unseasoned loans in the low credit score region and overrepresenting seasoned loans in the high credit score region. We present multiple additional channels for adverse selection of private student loans that could substantially increases losses without altering the i disclosed characteristics of PSL-ABS deals (e.g. overrepresenting college drop-outs, the share of which is known to the securitizer but not disclosed). The existence of such channels indicates that our estimates of ABS issuers’ ability to affect pool performance via loan selection at the time of securitization should be interpreted as lower bounds. ii Acknowledgements I am grateful for the guidance and advice provided by Burton Hollifield throughout my efforts to produce this dissertation. I am indebted to Adam Ashcraft for giving me the opportunity to work at the Federal Reserve Bank of New York as a PhD intern and later as a contractor. Adam went far beyond his duties and the completion of the third Chapter would not have been possible without his generosity. I am thankful to Laurence Ales who did a great job helping me prepare for the job market and helped improve this dissertation. I would also like to thank Brent Glover and Pierre Liang for raising questions and providing valuable advice that helped improve this dissertation. Other faculties that stood out because of their advice or suggestions on parts of this dissertation include: Antje Berndt, Stephen Karolyi and Jack Stecher. I would also like to thank Chris Sleet, Yaroslav Kryukov, Rick Greene, Duane Seppi and Fallaw Sowell. Thank you for helping me progress through the PhD program or for going out of your way to pose a supportive action. Finally, a big thank you to my friends and classmates who helped me improve this dissertation or helped me progress through the PhD program. From proof-reading and editing to modeling advice, I greatly appreciate the help provided by Artem Neklyu- dov, Batchimeg Sambalaibat, David Schreindorfer, Andrés Bellofatto, Carlos Ramírez, Alexander Schiller, Emilio Bisetti, Ben Tengelsen and Christopher Reynolds. I received a Doctoral Fellowship from the Social Sciences and Humanities Research Council of Canada (SSHRC). I thank the SSHRC for its support. I also acknowledge financial support from the William Larimer Mellon Fellowship at Carnegie Mellon University. iii Contents 1 Near-Arbitrage among Securities Backed by Government Guaranteed Student Loans 1 1.1 Introduction.....................................1 1.2 Sources of cash flow on SLABS..........................4 1.3 Benchmark no-arbitrage lower bounds on the price of simplified SLABS.. 11 1.4 Near-arbitrage lower bound on the price of SLABS.............. 14 1.4.1 Simulations, overcollateralization and relation with analytical lower bounds.................................... 14 1.4.2 Abandoning the simplifying assumptions............... 19 1.4.3 Upper bound on servicing fees...................... 21 1.4.4 Examples of SLABS-Treasury near-arbitrage.............. 28 1.5 Normative implications.............................. 35 1.5.1 Central banks’ exceptional measures of liquidity provision..... 35 1.5.2 Asset purchase program.......................... 36 1.5.3 Fire-sale insurance............................. 38 1.6 Conclusion...................................... 39 1.7 Appendix...................................... 41 1.7.1 SLABS that satisfy all selection criteria................. 41 1.7.2 Proof of Proposition 1........................... 43 1.7.3 Estimation of parameters of the interest rate model.......... 49 1.7.4 Data and computation of the servicing cost difference between delinquent and current borrower..................... 51 1.7.5 Near-arbitrage lower bounds on SLABS deals with a stepdown date 52 2 SLABS Near-Arbitrage: Accounting for Historically Unprecedented Macroeco- nomic Events 55 2.1 Introduction..................................... 55 2.2 Insuring against inflation risk........................... 56 v 2.3 Basis risk....................................... 60 2.4 Default on government guarantees........................ 69 2.5 Conclusion...................................... 71 2.6 Appendix...................................... 74 2.6.1 SBA PC price indexes........................... 74 3 Securitization with Asymmetric Information: The Case of PSL-ABS 76 3.1 Introduction..................................... 76 3.2 Related Literature and Lower Bound Interpretation.............. 79 3.3 Data Description.................................. 84 3.3.1 Pool characteristics data.......................... 84 3.3.2 Loan-level performance data....................... 86 3.4 Shifts in Pool Losses via Selection........................ 89 3.4.1 Matched pool characteristics and empirical constraints........ 90 3.4.2 Estimation.................................. 94 3.4.3 Forming loss-maximizing and random pools.............. 98 3.4.4 Shifts in pool losses............................ 104 3.4.5 Matched credit score, performance and interest rate......... 105 3.5 Concluding Remarks................................ 106 3.6 Appendix...................................... 108 3.6.1 Disclosure on Credit Score at Origination................ 108 3.6.2 Type of Data Used to Construct Pool Level Parameters........ 108 3.6.3 Loan Level Variables from CCP: Raw and Derived.......... 109 3.6.4 Distribution across Seasoning Groups.................. 111 3.6.5 Matched Pool Characteristics....................... 112 3.6.6 Geometric Solution to a Simplified Issuer’s Problem......... 114 3.6.7 Forming Loss-Maximizing Pool (Post-Crisis Deals).......... 127 Bibliography 128 vi Chapter 1 Near-Arbitrage among Securities Backed by Government Guaranteed Student Loans 1.1 introduction The financial crisis of 2007-2009 presented several challenges for central banks in perform- ing their role of liquidity provider of last resort. In the preceding decade, the origination of consumer loans became increasingly reliant on their indirect sale to investors purchas- ing asset-backed securities (ABS). Most ABS markets experienced sharp declines in prices during the crisis. Simultaneously, the cost of raising funds to originate