Preliminary and Incomplete

The Seasons of Money: ABS/MBS Issuance and the Convenience Yield

Lei Xie*† Yale School of Management First Draft: 2012/09/30

Abstract Theoretical research suggests that earning the convenience yield carried by safe assets could be one of the most important driving forces for the boom in the securitization market that preceded the financial crisis of 2007-2009. This paper empirically tests this hypothesis by examining the relationship between a high frequency ABS/MBS issuance series and a new proxy for the convenience yield. I use two shocks as instruments: the seasonal fluctuation of the convenience yield and the variation in Treasury issuance. Both are independent of the securitization market, but they are correlated with the convenience yield. I find that ABS/MBS issuers react to the change in the convenience yield, i.e., they issue more ABS/MBS when the expected convenience yield is high, and vice versa. A similar phenomenon can be found in another market for private safe assets, the ABCP market. This phenomenon does not exist, however, in markets for risky debt, such as the corporate bond market.

*

Tel: 203-737-0714. Email: [email protected]. I am indebted to my dissertation committee, Nicholas Barberis, Gary Gorton, Andrew Metrick and Justin Murfin for many helpful discussions and guidance. I would also like to thank Wenxi Jiang and Hongjun Yan for helpful comments. All remaining errors are my own. †

1

I.

Introduction

The recent boom and bust of the securitization market was one of the most dramatic events in the post-war history of U.S. financial markets. One explanation for the unprecedented boom was the huge demand for safe, money-like assets from 2002 to 2006 (Caballero and Krishnamurthy (2009), Bernanke, Bertaut, DeMarco and Kamin (2011)). Using pooling and tranching technology, financial institutions assembled traditionally illiquid and risky assets together and financed them by issuing securities. The vast majority of these securities were sold as safe assets and earned a premium, namely, the convenience yield (Gorton and Metrick (2012b), Gennaioli, Shleifer, and Vishny (2011), Stein (2012)). Due to problems with data availability, scant empirical work has tested the explanation that the emergence and boom of the securitization market was an endogenous private-sector response to the increasing demand for money-like assets. Using a novel dataset, this paper empirically examines the connection between the securitization market and the convenience yield. Recent research points out that safe, liquid, and money-like assets carry a convenience yield; in other words, yields on these assets will be lower than a theoretical risk-free rate. For example, Krishnamurthy and Vissing-Jorgensen (2010) and Greenwood, Hanson, and Stein (2012) document a large convenience yield carried by Treasuries. Gorton and Pennacchi (1990), and Dang, Gorton, and Holmström (2009) argue that, in theory, these assets are information-insensitive, in the sense that the potential profit for informed transacting parties is close to zero. Therefore they are ideal media of exchange and can be used as collateral in financial transactions. Investors are willing to accept a lower yield for the convenience services provided by these assets. One of the main purposes of securitization is to produce these kinds of safe assets (Gorton and Metrick (2012b)). By pooling assets, issuers can minimize the idiosyncratic risks of underlying assets. At the same time, tranching technology makes it possible for an issuer to strip off the information-sensitive part from an asset pool and sell only the information-insensitive part to investors. Typically, about 85% of a private-label assetbacked securities (ABS) or mortgage-backed securities (MBS) deal will be rated as

2

AAA,1 the highest rating that private debt can have. In the past twenty years, about 8 trillion private-label AAA-rated ABS/MBS has been issued.2 AAA-rated ABS/MBS, to some extent, can replace Treasuries and provide money-like services. Banks and other institutional investors acquire a large amount of AAA-rated ABS/MBS as liquidity reserves (e.g., Erel, Nadauld and Stulz (2011), Manconi, Massa and Yasuda (2010)). Moreover, highly rated ABS and MBS are widely accepted as collateral for repo transactions and asset-backed commercial paper (ABCP). (Gorton and Metrick (2012a), Acharya, Schnabl, and Suarez (2012) and Covitz, Liang, and Suarez (2012)). The ABS/MBS issuers could therefore earn a convenience premium. This paper tests the hypothesis that earning the convenience yield carried by safe assets could be one of the most important driving forces for the boom in the securitization market. I do so by examining the relationship between ABS/MBS issuance and the timevarying convenience yield. If the hypothesis is correct, ABS/MBS issuers may react to the change in the convenience yield. Specifically, they may issue more ABS/MBS when the convenience yield is high, and vice versa. However, the long-term trend in ABS/MBS issuance could be affected by many economic factors, so it is hard to identify clearly the influence of the convenience yield. Therefore, this paper focuses instead on the short-term deviation of the issuance series from its long-term trend. In order to get high-frequency ABS/MBS issuance data, I hand collected a large dataset from Bloomberg, which includes more than 20,000 ABS/MBS deals and 300,000 tranches. The total issuance in this dataset is more than 11 trillion, and the sample period ranges from 1978 to 2011. This dataset effectively covers most privatelabel ABS/MBS issued in U.S. market.3 With this dataset, I am able to construct a daily series for the total ABS/MBS issuance for more than twenty years. Next, I use an H-P filter and piecewise regressions to model the long-term trend for this daily series. And then I calculate the short-term deviation by subtracting the long-term trend.

1

Detailed discussion can be found in Section III.A. See discussion in Section III.A. 3 The data coverage issue will be discussed in section III.A. 2

3

Research to date has often used the spreads between the yield on Treasuries and the yields on private safe assets as measures for the convenience yield. However, the prices and issuance of ABS/MBS are jointly determined. If I find that ABS/MBS issuance the ABS/MBS-Treasury spreads move in the same direction, then, for example, it could be caused by the shift of the supply curve for ABS/MBS rather than by the change in the convenience yield. A positive supply shock may result in increase in both ABS/MBS issuance and the ABS/MBS-Treasury spreads. Therefore, I construct a measure for the convenience yield that is independent of the ABS/MBS market. Specifically, this measure is the rate spread between general collateral repurchase agreement (GC repo hereafter) and Treasuries. The GC repo is the repo transaction that uses Treasuries as collateral. A GC repo is as safe as Treasuries, but it cannot provide as many the money-like functions as offered by Treasuries. Historically, the GC repo rate is 36 bps higher than the Treasury yield. My interpretation for the difference is the compensation for the convenience services. GC repo is a money market instrument and its price should be unrelated to the supply for ABS/MBS. Hence GCTreasury spread is an ideal proxy for the convenience yield. Next, I use the seasonal variation in the convenience yield as an instrument variable. The seasonal variation in the convenience yield is predictable for ABS/MBS issuers and is exogenous to the ABS/MBS market. I confirm the existence of the seasonality of the convenience yield by showing that the seasonal pattern repeatedly appears in independent subsamples. Then I regress the detrended ABS issuance series on the expected convenience yield obtained from a first-stage seasonality regression. The result shows that ABS/MBS issuance is positively correlated with the expected convenience yield. This suggests that ABS/MBS issuers attempt to “time” the market, i.e., they issue more ABS/MBS when the expected convenience yield is high, and versa vice. To cross-check that the results are not driven by the common seasonal fluctuation that may exist in both debt and money markets, I repeat the test but replace ABS/MBS issuance by corporate bond issuance and net ABCP issuance. Corporate bonds are usually considered to be risky debts, and ABCP are usually considered to be safe debts. I find that corporate bond issuance is negatively correlated with the expected convenience yield, 4

while net ABCP issuance is positively correlated with it. This result indicates that only safe asset issuers, rather than all debt issuers, respond to the variation in the convenience yield. The second shock I employ is the variation in Treasury issuance. Treasury issuance is a supply shock to the safe asset market. Krishnamurthy and Vissing-Jorgensen (2010) finds that the convenience yield is negatively correlated with the supply of Treasuries in the long-term. I confirm their results using short-term data. The previous three days’ Treasury issuance has a significant negative effect on today’s convenience yield. ABS issuers react to this supply shock, too: They issue fewer ABS/MBS when the fitted convenience yield, which is obtained from the first-stage Treasury issuance regression, is low; in other words, ABS/MBS issuers attempt to “fill the gap”, i.e., they issue more ABS/MBS when Treasury issuance is low. This is consistent with the findings of Gorton, Lewellen, and Metrick (2010), which show that, in the long-term, the share of private safe assets in total financial assets is negatively correlated with the share of Treasuries. The sum of the shares of private and public safe assets is almost constant over 60 years. My paper shows a high-frequency and dynamic version of this “constant safe asset share.” This paper is related to several strands of the literature. First, as Gorton and Metrick (2012b), Gennaioli, Shleifer, and Vishny (2011), Stein (2010), and other papers argue, one of the main purposes of securitization is to produce safe assets. This paper supports this hypothesis by showing that ABS issuers react to the change in the convenience yield, the premium paid to safe assets. Second, this paper is related to the research on the interactions between public and private debt. Greenwood, Hanson, and Stein (2010) documents that corporate bond issuers actively adjust their debt maturity to fill gaps created by government debt. Gorton, Lewellen, and Metrick (2010) shows that, in the long-term, public and private safe assets are substitutes. My results suggest that ABS/MBS issuers also try to time the short-term price fluctuation caused by the change in government bond issuance. Third, an emerging literature focuses on the convenience yield. Krishnamurthy and Vissing-Jorgensen (2010) demonstrates a negative relationship between Treasuries outstanding and the AAA-Treasury spread. Greenwood, Hanson, and Stein (2012) shows Treasury Bills earn a higher convenience yield than do Treasury 5

Notes and Bonds. They also document the seasonality in the convenience yield and Treasury issuance. Sunderam (2012), which is most closely related to this paper, shows that ABCP outstanding is negatively correlated with the ABCP-Treasury spread at the weekly level. His results are consistent with the findings of my paper. The rest of this paper is organized as follows. Section II discusses the measure for the convenience yield. Section III presents data sources and summary statistics. Section IV shows the seasonality of the convenience yield and its stability. Section V demonstrates the relationship between ABS issuance and the convenience yield. Section VI explores the effect of Treasury issuance on the convenience yield and ABS issuance. Section VII shows some robustness checks, and Section VII concludes. II.

Measure for convenience yield

The convenience yield or convenience premium is the difference between a theoretical risk-free rate and yields on Treasury securities or other similar safe assets. This premium is the compensation to money-like convenience services provided by these safe assets. (Krishnamurthy and Vissing-Jorgensen (2010), Stein(2012), Gorton and Metrick (2011), Gorton (2010)).

Dang, Gorton and Holmström (2009) argue these safe assets are

information-insensitive, i.e., no one can benefit from acquiring private information when she trades these assets. As a result, no one has the incentive to produce private information and the adverse selection problem will be minimized. Investors are willing to accept these assets without doing costly due-diligence. Therefore, information-insensitive assets are ideal media of exchange and are widely accepted as collateral in financial transactions. The previous research suggests several proxies for the convenience yield, such as the yield spread between AAA-rated corporate bonds and Treasury bonds (Krishnamurthy and Vissing-Jorgensen (2010)), the difference between n-week Treasury bills yield and a theoretical yield based on an extrapolation of the rest of the yield curve (Greenwood, Hanson and Stein(2012), and the yield difference between private short-term debt (ABCP) and Treasury bills (Sunderam (2012)).

6

However, these proxies may not serve the purpose of this paper. First, I intend to test how ABS/MBS issuers react to the change in the convenience yield. If the yield on ABS/MBS is included in the estimation of the convenience yield, endogeneity will be a serious problem. Suppose I use the yield spread between ABS/MBS and Treasuries as the proxy for the convenience yield. Even if the Treasury yield were a constant, we may still observe a positive relationship between the ABS/MBS-Treasury spread and ABS/MBS issuance, exactly because ABS/MBS prices react to the change in the supply of ABS/MBS, rather than that issuers attempt to time the convenience yield. Hence I need a proxy which is outside of the ABS/MBS market. Moreover, using the yield spreads between other private safe assets, e.g. AAA corporate bonds or ABCP, and Treasury securities to measure the convenience yield could be problematic too. Because these spreads actually capture the yield difference between Treasuries and its private substitution, the relationship between these spreads and the convenience yield is unclear. For example, if there is a positive shock to the demand of safe assets, the convenience yield carried by Treasuries and private safe assets should both increase. But without more specific model, it is hard to tell whether the yield difference between Treasuries and private safe assets should increases or decreases. Therefore, I need a measure that is independent of all private safe assets. This paper proposes a new measure for the convenience yield. The idea is to find a riskfree instrument that does not carry money-like services. GC repo is very close to this definition. In a GC repo transaction, lenders are willing to accept any of a variety of Treasury securities as collateral. The collateral is general rather specific. Because GC repo is secured by Treasuries, the lending is free of any counterparty or credit risk. And since the collateral is not limited to any specific security type, the repo rate will not be affected by the supply or demand shock to certain Treasury category. On the other hand, comparing to a holder of Treasuries, a GC repo lender loses some flexibility. For example, the lender cannot terminate the repo transaction at any time, but a holder can sell Treasuries at any time he wants. Although in some cases, the collateral can be rehypothecated to the third party, but it is not as easy as using Treasuries held by himself, especially if the lender is a small institutional investor. Unlike Treasuries, it is hard to use 7

GC repo as a medium of exchange. GC repo is unable to perform the functions of money. Historically, the GC repo rate is 36 basis points lower than the Treasury yield, on average. This difference does not contain any counterpart or credit risk premium and it is more like to be the compensation for the money-like flexibility of Treasuries. Furthermore, the GC-Treasury spread is independent of the ABS/MBS yield. So we have an unambiguous prediction for the relationship between the GC-Treasury spread and the convenience yield carried by private safe assets: When the GC-Treasury spread increases, which means the convenience yield for Treasuries increases, at the same time, as the best private substitution for Treasuries, the convenience yield carried by ABS/MBS should also increases. Last, from the data part, both Treasury and GC repo are highly active market and there is no stale price problem. Moreover, GC rate is available from 1991. Its sample period is much longer than that of another widely used risk-free rate, OIS, which starts from 2001. Therefore, for the purpose of this paper, the GC-Treasury spread is an ideal proxy for the convenience yield. Besides the GC-Treasury spread, I also look into the widely used TED spread, i.e. the spread between the LIBOR and the Treasury yield. But the LIBOR is not entirely riskfree: the inter-bank lending is not secured. So this spread includes the compensation for counterparty risk too. I decompose TED spread to two parts, the GC-Treasury spread is for convenience yield and the LIBOR-GC spread is for counterparty risk premium. All the three spreads will be examined in the later tests. III.

Data

A. ABS and MBS data This paper focuses on private-label ABS/MBS, i.e. ABS/MBS issued by banks and other institutions rather than government sponsored agencies. The ABS/MBS data is downloaded from Bloomberg. This dataset includes both alive and matured deals. The earliest deal dates back to 1978. The total issuance amount of this dataset is about 11 trillion. Both tranche-level and deal-level information are available. For each tranche, this dataset provides issue date, original balance, ratings, securities type and other 8

characteristics. In the deal level, it gives summary statistics for collateral quality, e.g. average credit score, LTV etc, and also the performance of collateral after issuance. The richness of this dataset offers a great chance to explore the private-label securitization market. About the detail of this dataset and corresponding data clean process, please refer to the data appendix of this paper. Panel A of Table I presents the total issuance amount of all private-label ABS/MBS for each year from 1978 to 2010. The first column shows the issuance of AAA-rated tranches. The second and third columns show the issuance of other non-AAA investment grade tranches and non-investment grade tranches. The total issuance of all tranches and the percentage of AAA-rated tranches are presented in the last two columns. To check the coverage of this dataset, I compare my annual issuance series with that provided by Securities Industry and Financial Markets Association (SIFMA) 4. The two series are plotted in Figure I. The correlation between them is 99% and the difference between them is less than 10% in most years. The Bloomberg dataset covers most private-label ABS/MBS issued in U.S. market. Private-label ABS/MBS market starts from late 1970s as a tiny part of fixed-income market. It starts to take off after 1983 stimulated by the Federal Reserve’s high interest policy in the middle 1980s. From 1983 to 1988, the market size increases from 2.3 billion to 66.4 billion. Although having been through two short recessions, private-label ABS/MBS market keeps growing in the whole 1990s. At the end of 2000, it is already a significant part of the debt market. After 2000, this market starts a round of unprecedented growth. From 2000-2006, in just six years, the market size quintuples from 386 billion to 1,654 billion. Exceeding Treasury and corporate bonds, it becomes the largest debt category. After the crisis of 2007-2009, this market suddenly collapses. In 2010, the total issuance drops to 225 billion, which means the industry shrinks by 87% in 4 years and goes back to its level in 1996. Unlike corporate bonds, one ABS/MBS deal backed by the same collateral will be divided to different tranches and be given different ratings. The ratings range from AAA 4

http://www.sifma.org/research/statistics.aspx, SFIMA provides aggregated statistics for different sectors of U.S. fixed-income market.

9

to no-rating. Before the crisis of 2007-2009, on average 86.3% ABS/MBS has been rated as AAA, 10.6% has been rated as other non-AAA investment grade and 3.1% has been rated as non-investment grade. This is a market dominated by AAA-rated securities. In contrast, corporate bonds is a very different market, in which only 9.2% bonds are rated as AAA and 28.9% bonds are below investment grade.5 The private-label ABS/MBS market can be divided to several subsectors depending on their underlying collateral. For ABS, the largest collateral categories are credit cards, home equities and auto loans. For MBS, the largest two categories are residential mortgage securities (RMBS) and commercial mortgage securities (CMBS). Subprime mortgage loans, through in essence are more like RMBS, are often labeled as “Home Equity”. Considering the substantial role played by subprime loans in the recent crisis, I separate them from normal home equity ABS and put them into a new category “Subprime”. Panel B of Table II presents issuance statistics for the seven categories. For ABS, home equity is the largest category. The total issuance is about 1.5 trillion. It is followed by auto loans and credit cards, which have 1.1 trillion and 0.92 trillion issuance respectively. RMBS, which are mainly backed by alt-a and jumbo etc. prime residential mortgages, is the largest part of the private-label securitization market. The total issuance is 4.3 trillion. The total issuance of subprime MBS is about 1.1 trillion. One interesting fact is that although from category to category the underlying assets are very different, the average percentage of AAA-rated securities (hereafter AAA ratio) for all categories is consistently high. Even for the category with lowest AAA ratio, CMBS, 77.3% tranches are rated as AAA. B. Spreads data The interest rate data comes from Bloomberg too. Table A.1 presents description and Bloomberg tickers for all interest rates. One thing worth mentioning is that for GC repo, Bloomberg provides both repo rate and reverse repo rate. They are similar to ask and bid price of stocks. Over this paper’s sample period, the average spreads between repo and

5

The average is calculated from 1990-2007.

10

reverse repo is 7 basis points. In the rest of the paper, I use the average of repo and reverse repo rate as the GC repo rate. Table II presents summary statistics for three spreads used in this paper: the LIBORTreasury, LIBOR-GC and GC-Treasury spread. Panel A shows the summary statistics for one-month spreads. Mean, standard deviation, 25%, 50% and 75% percentiles are reported. For the purpose of this paper, I mainly focus pre-crisis period. The sample covers from June 1991 to June 2007. The crisis periods will be discussed in Section VII. The average LIBOR-Treasury spread across whole sample is 52.1 basis points (bps). It can be decomposed to two parts: the average LIBOR-GC spread is 16.3 bps and the average GC-Treasury spread is 35.7 bps. The convenience premium accounts for about two thirds of the TED spread and the risk premium accounts for about one third of it. The volatility of the Ted spread also mainly comes from the convenience premium part. The standard deviation of the GC-Treasury spread is 28.4 basis points and the standard deviation of the LIBOR-GC spread is only 10.2 basis points. I divide the whole to two parts, 1991-1999 and 2000-2007 to check the time-series variation of spreads. From the first to the second period, the LIBOR-GC spread does not change too much (15.3 bps to 17.3 bps), but GC-Treasury spreads decreases by about 33%, from 42.7 bps to 28.8 bps. It may reflect the effect of financial innovations, like rehypothecation, as reactions to the lack of collateral. When repo collateral can be rehypothecated, the extra flexibility provided by holding Treasuries is undermined. Panel B shows statistics for three-month spreads. The three-month LIBOR-GC spread is higher than the one-month spread, which reflects the increasing counterparty risk with a longer maturity. In contrast, the three-month GC-Treasury spread is smaller than the onemonth one. It is consistent with the finding of Greenwood, Hanson and Stein(2012), which shows that convenience yield is larger for Treasuries with shorter maturities. Since the convenience yield carried by one-month Treasuries is higher, in the main tests I will focus on one-month spread and also check three-month spread as a robustness check. C. Corp bond/ABCP

11

Corporate bond issuance data is obtained from Mergent FISD database 6 . The total issuance by year is summarized in Panel A of Table III. AAA-rated, non-AAA investment grade and non-investment grade corporate bond issuance are listed separately. Unlike ABS/MBS, very few corporate bonds are rated as AAA. Before 2000, the percentage of AAA-rated corporate bonds is never higher than 5%. During the credit boom of 2002-2006, the AAA ratio for corporate bonds is 14.1%, which is higher than historical average but is still much lower than the AAA ratio for ABS/MBS. Corporate bond is hardly a market for safe asset investors. Asset-backed commercial paper (ABCP) conduits are a special type of CP issuers. Such a conduit is a special purpose vehicle (a legal entity) that buys asset-backed securities, financing this by issuing commercial paper. Most ABCP are required to obtain the necessary ratings and have back-up liquidity facilities in case they have problem to renew their commercial paper. (Acharya, Schnabl and Suarez (2011), Covitz, Liang, and Suarez (2012)). Hence they are considered to be very close to risk-free. The ABCP outstanding data is from Federal Reserve. The total outstanding is presented in Panel B of Table III. Its market size doubles in the credit boom of 2002-2006 and collapses during the crisis. In 2010, the total outstanding of ABCP is only half of the level in 2001. D. Treasury issuance CRSP provides daily outstanding for each Treasury bill/bond. But if one Treasury security matures in this month, the change in outstanding of this security will not be updated until the end of the month. Hence I download the maturity dates for all Treasuries from website of the Treasury department and adjust the maturity dates in CRSP. Then I aggregate the outstanding of all Treasury securities together to a single total outstanding series. To check the accuracy of the data, I compare the quarterly issuance series I got and the series from the flow of funds tables. The two series are plotted in Figure II. Clearly they are highly correlated and the difference is small. IV.

Seasonal Variation of the Convenience Yield

6

I only include securities with bond_type in CMTN, NT, CS, CPIK, CZ, USID, CCPI, CCOV, CMTZ, CDEB.

12

To test the relationship between ABS/MBS issuance and the convenience yield, we need a shock that is outside of the securitization market. It should not be affected by the supply or demand on ABS/MBS. On the other hand, this shock needs to be predictable. Issuing an ABS/MBS deal requires a long time to prepare. The whole process may cost several months. Hence it will be hard for ABS/MBS issuers to react to short-term unpredictable variations in the convenience yield. We need something predictable and can be planned in advance by issuers. In this paper, I employ the seasonal variation in the convenience yield as the shock. The seasonal variation of interest rates is hardly a new thing. In the period before the Federal Reserve System there were seasonal spikes in interest rates when cash had to move from cities to rural areas for planting season and then later for harvesting season. After the establishment of the Federal Reserve System, variation caused by harvest season becomes less important, but other periodic fluctuations emerge. For example, the cashflow demand induced by regulatory and operationally reasons (Griffiths and Winters (2005)), and the payment of estimated taxes by corporations. Another possibility is “window dressing”. Allen and Saunders (1992) find that banks’ total assets significantly increase at the quarter-end. Banks use money market instruments to finance these new assets, which drives up the federal fund rates. Recently Gorton, Metrick and Xie (2012) documents the seasonality for a variety of money market instruments. Greenwood, Hanson and Stein(2012) shows that the Treasury yield and the supply of Treasuries also have seasonality. To examine the seasonality for the convenience yield, I regress the GC-Treasury spreads on 365 date dummies and plot the fitted values in Figure III. There is considerable seasonal variation in the GC-Treasury spread. For example, at the end of March, the spread is as low as 15 bps. But on December, the spread could be more than 60 bps. So far I still not sure whether the seasonal variation presented here is a stable relationship. To test whether this pattern is repetitive, I divide the sample to two parts: 1991-1999 and 2000-2007. For each subsample, I separately run a regression on date dummies and plot

13

the fitted values in Figure IV7. If the spread has no seasonality, the two series of fitted values should be independent. However, from the picture it is clear that two series are positively correlated. The correlation coefficient is 0.4 and highly significant. It indicates that the seasonal pattern for the GC-Treasury spread is stable over time. V.

ABS/MBS Issuance and the Convenience Yield Seasonality

As discussed before, a major part of private-label ABS/MBS tranches are rated as AAA. They are assumed risk-free and information-insensitive. In some circumstances, they can replace Treasuries and provide some money-like functions. For example, there is an active repo markets which accepts AAA-rated ABS/MBS as collateral (Gorton and Metrick (2010)). Many commercial banks hold AAA-rated ABS/MBS as liquidity reserve (Erel, Nadauld and Stulz (2011)). It is one of the best private substitutions for Treasuries. However, to directly test the long-term relationship between ABS/MBS issuance and the convenience yield will be hard. First, private-label ABS/MBS is relative new market. We only have about 20 year’s reliable data. Second, before the crisis of 2007-2009, the longterm trend for the securitization market is almost monotonically increasing. Finally, other economic factors, like the improvement in the credit evaluation system, may play a large role in the determination of the market’s long-term growth. However, it is hard to control for these factors and to separate the influence of the convenience yield, especially given the limited observations we have. Therefore, this paper is intended to examine the short-term deviation from its long-term trend. I focus on high-frequency daily rather than annual data, which gives me much more observations. On the other hand, I do not need to control for the change in economic environment if I only examine the short-term deviation, because the change in other fundamental factors will be captured by the long-term trend. A. Benchmark Test

7

As discussed in the summary statistics part, the level of GC-Treasury spread changes over time. So I first demean the level of the spread by regress it on a bunch of year dummies.

14

First, I take log of all issuance series. Then I use Hodrick–Prescott filter to separate the short-term deviation from the long-term trend8. Alternatively, I also use piecewise linear regressions to capture long-term trend, that is, I use a linear function to fit ABS/MBS issuance in each year and then connect these lines together. Figure V plots the long-term trend series generated by both methods. The two lines are very close. Next, the long-term trend part of the series will be subtracted from the original series. Only residuals will be kept and used in the later tests.

The summary statistics for the residual series are

presented in Table X. Then I run following regressions: ( where

(

trend.

̂

)

̂

)

is the residual of log ABS issuance separated from the long-term is the fitted value from the first stage seasonality regression. It is the

expected spread on certain date t inferred from the spread’s seasonal fluctuation. Panel A of Table V presents the regression results for H-P filter detrended series. In the first three columns, three fitted spreads are used as independent variables: the LIBORTreasury spread (TED), the LIBOR-GC spread and the GC-Treasury spread. The LIBOR-Treasury spread is the sum of the latter two spreads. The coefficient on the LIBOR-Treasury spread is positive and significant (t=3.91). When we further look at the two components, we will find the results are mostly driven by the GC-Treasury spread (t=3.8), which is the measure for the convenience yield. The coefficient on the LIBORGC spread, the proxy for counterparty risk premium, is only marginally significant (t=1.83). In column (5), I put both spreads into the regression. The coefficient on the GCTreasury spread is almost unchanged (t=3.36) and the coefficient on the LIBOR-GC spread becomes insignificant (t=0.53). The results imply that ABS/MBS issuers will react to the expected change in the convenience yield. They tend to issue more when the premium attached to safe assets is higher.

8

λ is set as 12960000 to match daily issuance data.

15

Column 4 shows the results using t-1 gross GC-Treasury spread rather than fitted value. The coefficient is positive but not significant (t=0.54). As discussed before, issuing an ABS/MBS deal requires a long time to prepare. Therefore, issuers are only able to react to the predicable part of the variation in the convenience yield, but fail to react to the unpredictable part. From column (5) to (7), I add several control variables into regressions. First, I add lagged ABS issuance to control for autocorrelation. Second, Gorton, Metrick and Xie (2012) finds that spreads for repo and other money market instruments are abnormally high on the beginning, middle and end of each month. So I add three day dummies into regression. It shows that ABS issuances in these three days are significantly higher than normal days, but it cannot explain the effect of the GC-Treasury spread. Similarly, I add a quarter-end dummy into regression. The dummy is equal to 1 if the date is in the last week of each quarter. The coefficient on the GC-Treasury spread is still significant after controlling for quarter-end effect. In Panel B of Table 5, I use piecewise regressions to model the long-term trend for ABS/MBS issuance and repeat all regressions. The results are very close to Panel A. Hence in the following tests, I only present the results based on H-P filter. B. Out-of-sample Test It is possible that both the GC-Treasury spread and all debt issuance have seasonality and they coincide with each other. To check whether the results are driven by the common seasonality appeared in all debt markets, I do two out-of-sample tests. I run the same regressions for corporate bond issuance and net ABCP issuance. As shown before, corporate bond market is very different from ABS/MBS market. Most corporate bonds are below AAA rating and not considered as safe assets. Therefore, corporate bond issuers have little incentive to time the convenience yield. On the other hand, ABCP is another kind of private safe assets. ABCP issuers may adjust their issuance according to the variation in the convenience yield. Hence if the hypothesis of this paper is true, we should observe corporate bond issuance is not positively correlated with the GC-Treasury spread, but ABCP issuance is. However, if the results are driven by a seasonal pattern 16

that is common for all debts, the three series should all be positively correlated with the GC-Treasury spread. The results for corporate bonds are presented in Panel A of Table VI. The corporate bond issuance series has been taken log too and only the deviation from the long-term trend has been kept. Then I run the same regression as column (1)-(4) in Table V. For all three spreads, the coefficients are negative and significant. The results for ABCP are showed in Panel B of Table VI. Because most ABCP are very short-term and issuers usually continuously issue new commercial papers to finance mature ones, the issuance series may not be able to reflect the supply change in the market. Therefore I use the change in weekly ABCP outstanding rather than issuance as the dependent variable. I do the same logging and detrending process on ABCP series. The result shows that the GC-Treasury spread is positively correlated with ABCP outstanding change and the coefficient is highly significant (t=3.7). The two out-of-sample tests show that only the issuers of private safe assets will react to the change in the convenience yield. Interestingly, the supply of risky assets actually is negatively correlated with the convenience yield. The common seasonal fluctuation in debt markets cannot explain these results. VI.

Treasury Issuance and ABS Issuance

In this section, I explore another shock to convenience yield: the variation in the Treasury issuance. If the GC-Treasury spread represents the market price for convenience services provided by Treasuries, the change in the supply of Treasuries should affect the spread. Although Treasury market has been considered to be highly liquid and deep, the Treasury issuance or redemption may still has a short-term price effect (Lou, Yan and Zhang (2010)). Therefore ABS issuers may react to the short-term price shock. Private safe assets may be temporarily crowded out by public safe assets. I will check this hypothesis in a two-step test: first I examine whether the change in Treasury outstanding will affect the GC-Treasury spread. Second, I further whether Treasury issuance instrumented GCTreasury spread will influence the ABS issuance. A. Treasury Issuance and the GC-Treasury Spread 17

As discussed in Section III.D., I use the change in aggregated Treasury outstanding to measure net Treasury issuance. Similarly to other series, the Treasury issuance series has been detrended using H-P filter and only the deviation from the long-term trend will be examined. I first regress the GC-Treasury spread on lagged detrended Treasury issuance. The results are reported in the first three columns in Panel A of Table VII. Up to three days, the lagged Treasury issuance has a significant negative effect on the GC-Treasury spread. The results confirm that the short-term price effect of Treasury issuance and redemption. The fitted values of this regression will be used in the next stage test. B. Treasury Issuance and ABS Issuance, Second stage Next, I regress the detrended ABS issuance on the fitted GC-Treasury spread obtained from the first stage regression. Panel B of Table VII shows the regression results. First column is the benchmark regression. The coefficient on the spread is significantly negative (t=3.91). From the second to fourth column, I add lagged ABS/MBS issuance and date dummies into regressions. The coefficient on the spread is hardly changed. All of them are statistically significant. In column (5)-(7) of Panel B of Table VII, I directly test the relationship between ABS issuance and Treasury issuance. The result shows that lagged one and two day Treasury issuance are negatively correlated with ABS issuance. The results suggest that private issuers will react to government’s action. Treasury issuance (public safe assets) will drive down the prices for safe assets and crowd out private safe assets. Gorton, Lewellen and Metrick (2011) shows that in long-term, the sum of private and public safe assets is a fixed percentage of total financial assets. In other words, the shares of public and private safe assets are negatively related. The results shown here suggest that this relationship may also exist in short-term. As a substitution to Treasuries, ABS/MBS is more likely to be issued when there are less Treasuries available. ABS/MBS issuers try to fill the “gap” of Treasuries. It is similar to the results found by (Greenwood, Hanson and Stein (2010)). C. Seasonal Effect Revisit: Treasury Issuance and the Seasonality of Convenience Yield

18

For the purpose of this paper, I treat the seasonal variation in the GC-Treasury spread as exogenous. But it will be interesting to explore the possible explanation. The previous research (e.g. Greenwood, Hansen and Stein (2012)) points out that Treasury issuance has seasonality too. Tax flows are not evenly distributed among different dates and government’s expenses are not steady either. The Treasury Department needs to issue Treasury bills to smooth government’s cash flow. Treasury issuance will be higher in certain dates and be lower in others. For example, after Apr 15th, when government gets a large amount of tax inflows, the Treasury Department will redeem Treasuries or cut down the issuance. The seasonal fluctuation of Treasury issuance could be one important factor causing the seasonal variation of the convenience yield. To explore the seasonality of Treasury issuance, I regress detrended Treasury issuance series on 365 date dummies. In Figure VI, I plot the date dummies fitted GC-Treasury spread and Treasury issuance together. Noticeably, the two series are negatively correlated. When Treasury issuance is higher, the convenience yield is low. D. Seasonal and non-seasonal effect of Treasury Issuance The correlation between the seasonality of Treasury issuance and the convenience yield raises several questions: first, will the seasonal variation of Treasury issuance affect ABS/MBS issuance? Second, can the results in Section V be fully explained by the seasonality of Treasury issuance? In other words, is the seasonality of the convenience yield just a reflection of Treasury issuance, or the variation of the convenience yield provides incremental information? Third, after controlling for seasonal variation, will the non-seasonal part of Treasury issuance affect ABS issuance? To answer above questions, I run a series of regressions in Panel C of Table 7. First, I regress ABS issuance on the date dummies fitted Treasury issuance series. The first three columns report the results for one-, three-, and five-day lagged date-fitted Treasury issuance. The coefficients on one to three days lagged date-fitted Treasury issuance are all negative and significant. It suggests that ABS issuers will react to the seasonal variation of Treasury issuance: they issue more ABS when they expect Treasury issuance is lower. 19

Second, I put both fitted Treasury issuance and fitted GC-Treasury spread into one regression. It shows that both variables are in their original direction and significant. Although Treasury issuance is one important factor causing the seasonality of the GCTreasury spread, it cannot explain all effect of the GC-Treasury spread on ABS issuance. There may be other periodic factors that also affect the convenience yield, like high demand for safe assets at the end of month-end/quarter-end, etc. ABS/MBS issuers may attempt to take advantage of these opportunities too. Next, I will check whether non-seasonal variation in Treasury issuance will affect ABS issuance. I look into the residuals of Treasury issuance series after excluding seasonal variation. It is the part that cannot be inferred from historical data, i.e., the unexpected shocks. I regress the GC-Treasury spread on the lagged residual of Treasury issuance and report the result in column (5)-(6) in Panel C of Table VII. Surprisingly, ABS issuance is correlated with the non-seasonal spread change too. ABS issuers are able to know future Treasury issuance in about next two weeks9. Hence they may be able to “nudge” the issuance date a little bit. To see this point more clearly, we can compare column (2) and column (6) in Panel C of Table VII. The first one is how ABS issuers react to the seasonal variation, or the expected change in Treasury issuance. The second one is the reaction to non-seasonal variation, or the unexpected change. In the first regression, the coefficients on all three lags are negative, which suggest that the effect is lasting and it does not reversed. In the second regression, the coefficient on three-day lagged Treasury issuance reverts to positive and the magnitude is close to the sum of the coefficients on one- and two-day lags. It suggests that the effect of unexpected Treasury issuance shock will reverse in short-time. A ABS/MBS issuer is only able to move the issuance date to a couple of days later, but cannot cancel the deal. VII.

Robustness Checks

In this section, I run a bunch of robustness checks for the previous tests. First, I look at alternative convenience yield measures. Second, I examine the seasonal effect in both

9

For example, the announcement date of new treasury auction is about one week before the auction. And issue day is about one week after the auction. So market participators are able to know the treasury issuance in the future two weeks.

20

monthly level and intra-month level. Third, I include crisis period into the analysis. Fourth, I use rolling regression to infer seasonal effect. A. Alternative convenience yield measure In this section I examine two alternative convenience yield measures. First, I replace the GC repo rate by another risk-free rate, Overnight Index Swap (OIS) rate. Second, instead of one-month GC-Treasury spread, I look at three-month spread. OIS is the fixed interest paid in an interest swap contract to exchange floating overnight federal fund rate. As a derivative contract, counterparties in this transaction only exchange interest payment rather than principals. Hence OIS rate carries no principal risk and it should equal to the geometric average of future overnight federal fund rates. Federal fund market is unsecured. The counterparty risk in this market is small but not zero. Historically, OIS rate is several bps higher than GC rate. I treat it as the second best proxy for risk-free rate. However, since OIS rate is only available after 2001, its sample period is much shorter than the GC repo rate. Correspondingly, the noise in the estimation of its seasonality is expected to be larger. I repeat the regressions in column (3) of Panel A, Table V and column (1) of Panel B, Table VII. The dependent variables for both regressions are detrended ABS issuance. The independent variable for the first regression is the seasonal variation of OIS-Treasury spread and the second one is Treasury issuance instrumented OIS-Treasury spread. The results are presented in Panel A of Table VIII. In both regressions, the OIS-Treasury spread is positively correlated with ABS issuance. The coefficient on Treasury issuance instrumented spread is strongly significant (t=3.52) and the coefficient on date dummies instrumented spread is marginal insignificant (t=1.63). The insignificant result for the seasonal effect could be caused by the noise in the estimation from a much smaller sample. Section VII.C will show that the results become significant when we include more data into analysis. Panel B of Table VIII repeats main regressions using three-month spread instead of onemonth spreads. All tests are robust to three-month spreads. Both date dummies and

21

Treasury issuance instrumented GC-Treasury spreads are significantly correlated with ABS/MBS issuance. B. Inter-month and intra-effect In the previous tests, I use 365 date dummies to estimate seasonal effect. In this section, I decompose the seasonality into two parts. One is monthly seasonal effect. For example, convenience yield is lower in March and higher in June. The other is intra-month seasonal effect, or calendar-effect. For example, the convenience yield may be higher on certain dates of each month. Here I use two groups of date dummies as instruments. The first is a set of 12 monthly dummies and the second is a set of 30 daily dummies. I regress spreads on these two sets of date dummies and then put fitted values into the second stage regression. The results are shown in Panel C&D of Table VIII. For monthly effect, the coefficients on all three spreads, the LIBOR-Treasury, LIBOR-GC and GC-Treasury spreads are significantly positive. For the intra-month effect, the coefficients on the LIBOR-Treasury and GC-Treasury spreads are significantly positive, but the coefficient on the LIBOR-GC spread is significant negative. In both monthly and intra-month level, the measure for the convenience yield, the GC-Treasury spread, has significant effect on ABS issuance. C. Crisis period So far all analysis of this paper are restricted to the pre-crisis period because I want to examine the effect in the normal period. However, it would be interesting to see whether the results will change if we include crisis period into analysis. And it also gives us extra three year data. The results including crisis period are presented in Panel E of Table VIII. First three columns repeat the tests in Panel A of Table V. After including crisis period, the GC-Treasury spread is still significantly positive (t=4.24) and the coefficient is almost unchanged. But the LIBOR-GC spread is not significant anymore (t=1.03). As argued before, the LIBOR-GC spread may capture counterparty risk premium rather than the convenience yield. Again, we see how the effect of counterparty risk premium is different from the effect of the convenience yield. In the following columns, the results for 22

Treasury issuance instrumented GC-Treasury spread, date dummies instrumented and Treasury issuance instrumented OIS-Treasury spread are presented. The result for the GC-Treasury spread is still robust. After including more observations, the results for the OIS-Treasury become much stronger and significant (t=2.09 and 6.02 respectively). D. Rolling regression In Section V, I use the whole sample to estimate seasonality. But ABS issuers do not know future GC-Treasury spreads, which may cause a forward-looking problem. To make sure the data is available for issuers before estimation, I employ a rolling regression. The data of year t-1 to t-8 is used to estimate coefficients on date dummies and the fitted value will be applied to year t. The results are showed in Panel F of Table VIII. Although I lose 7 years observations in the rolling regression estimation, fitted GC-Treasury spread is still positively correlated with ABS issuance (t=1.9). VIII. Conclusion The boom of the securitization market is one of most interesting puzzles in recent financial research. Theoretical papers suggest that to earn the convenience yield carried by safe assets could be one important reason. This paper examines the relationship between high frequency ABS/MBS issuance series and a convenience yield proxy inferred from money market interest rates. Two shocks that are outside of the securitization market are used as instruments: the seasonal fluctuation of the convenience yield and the variation in Treasury issuance. The results show that ABS/MBS issuers attempt to time the market: they issue more ABS/MBS when the convenience yield is high, and vice versa. Particularly, the ABS/MBS issuers will react to governments’ actions. More ABS/MBS is issued when there is less Treasuries available.

23

References Acharya, Viral V., Philipp Schnabl, and Gustavo Suarez, 2012, Securitization without risk transfer, forthcoming, Journal of Financial Economics.

Allen, Linda, and Anthony Saunders, 1992, Bank window dressing: Theory and evidence, Journal of Banking & Finance 16, 585-623.

Bernanke, Ben S., Carol C. Bertaut, Laurie DeMarco, and Steven B. Kamin , 2011, International Capital Flows and the Return to Safe Assets in the United States, 20032007, FRB International Finance Discussion Paper No. 1014. .

Caballero, Ricardo J., and Arvind Krishnamurthy, 2009, Global Imbalances and Financial Fragility, American Economic Review 99, 584-588.

Covitz, Daniel M., Nellie Liang, and Gustavo A. Suarez, 2012, The evolution of a financial crisis: panic in the asset-backed commercial paper market, Journal of Finance, forthcoming.

Dang, Tri V., Gary Gorton, and Bengt Holmström, 2010, Opacity and the Optimality of Debt for Liquidity Provision, Working paper.

Erel, Isil, Taylor D. Nadauld, and RenéM. Stulz, 2011, Why Did U.S. Banks Invest in Highly-Rated Securitization Tranches?

Gennaioli, Nicola, Andrei Shleifer, and Robert W. Vishny, 2011, A Model of Shadow Banking, Working paper.

24

Gennaioli, Nicola, Andrei Shleifer, and Robert Vishny, 2012, Neglected risks, financial innovation, and financial fragility, Journal of Financial Economics 104, 452-468.

Gorton, Gary B., Stefan Lewellen, and Andrew Metrick, 2012, American Economic Review: Papers & Proceedings, forthcoming.

Gorton, Gary, Andrew Metrick, and Lei Xie, 2012, The flight from maturity, Working paper.

Gorton, Gary, and Andrew Metrick, 2010a, Haircuts, Federal Reserve Bank of St. Louis Review, 507-520.

---. 2010b, Regulating the Shadow Banking System, Brookings Papers on Economic Activity 41, 261-312.

---. 2012a, Securitized banking and the run on repo, Journal of Financial Economics 104, 425-451.

---. 2012b, Securitization, The Handbook of the Economics of Finance, forthcoming.

Gorton, Gary, and George Pennacchi, 1990, Financial Intermediaries and Liquidity Creation, The Journal of Finance 45, 49-71.

Greenwood, Robin M., Samuel G. Hanson, and Jeremy C. Stein, 2012, A ComparativeAdvantage Approach to Government Debt Maturity, Working Paper.

Greenwood, Robin M., Samuel G. Hanson, and Jeremy C. Stein, 2010, A Gap-Filling Theory of Corporate Debt Maturity Choice, The Journal of Finance 65, 993-1028. 25

Griffiths, Mark D., and Drew B. Winters, 2005, The Turn of the Year in Money Markets: Tests of the Risk‐Shifting Window Dressing and Preferred Habitat Hypotheses, The Journal of Business 78, 1337-1364.

Krishnamurthy, Arvind, and Annette Vissing-Jorgensen, 2007, The Demand for Treasury Debt, Working paper.

---. 2012, Short-term Debt and Financial Crises: What we can learn from U.S. Treasury Supply, Working paper.

Manconi, Alberto, Massimo Massa, and Ayako Yasuda, 2010, The Behavior of Intoxicated Investors: The role of institutional investors in propagating the crisis of 2007-2008, Working paper.

Stein, Jeremy C., 2012, Monetary Policy as Financial Stability Regulation, The Quarterly Journal of Economics 127, 57-95.

Sunderam, Adi, 2012, Money Creation and the Shadow Banking System, Working paper.

26

Figure I

1800

Total Issuance of ABS/MBS, cross-check

1600 1400 1200

1000 800 600 400 200 0 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 This paper's sample

SIFMA data

Figure II

Treasury Outstanding, cross-check 10000.00 9000.00 8000.00 7000.00

6000.00 5000.00 4000.00 3000.00 2000.00 1000.00 0.00

CRSP Data

Flow of Funds Data

27

Figure III

Fitted LIBOR_GC Spread 50 45 40 35 30 25 20 15 10 5 0

1/2

2/2

3/2

4/2

5/2

6/2

7/2

8/2

9/2

10/2

11/2

12/2

Figure IV

Seasonal Pattern Comparison, GC-Treasury 120 100 80 60 40 20

0 1/2

2/2

3/2

4/2

5/2

6/2

7/2

8/2

9/2

10/2

11/2

-20 -40 GC_Treasury (1991-1999)

GC_Treasury (1999-2007)

28

12/2

Figure V

Long-term Trend in ABS/MBS Issuance 25

20

15

10

5

0 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 Piecewise Regression

H-P Filter

Figure VI

Seasonality: Treasury Issuance and Convenience Yield (30-Days Moving Average) 60

0.0015

50

0.001

40

0.0005

30

0

20

-0.0005

10

-0.001

0 1/31

2/28

3/31

4/30

5/31

6/30

7/31

GC-Treasury Spread

29

8/31

9/30

-0.0015 10/31 11/30 12/31

Treasury Issuance

Table I ABS/MBS Issuance

1978 1979 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

AAA-rated

Non-AAA Investment Grade

Non-investment

Total

AAA as a % of Total

0.2 0.1 2.1 4.9 9.8 27.6 59.5 56.4 35.9 61.1 84.9 130.0 138.8 129.2 129.9 176.3 254.6 351.3 325.1 295.8 454.8 552.6 730.1 928.3 1339.3 1407.3 1253.9 159.9 192.9 90.5

0.3 0.0 0.0 0.1 0.3 1.7 4.6 8.2 17.1 15.7 18.3 19.5 18.6 17.7 15.2 23.3 31.2 53.4 35.9 36.3 47.3 48.8 73.0 111.8 172.6 197.2 128.1 14.4 12.5 65.1

0.0 0.0 0.2 0.4 0.1 0.3 0.5 1.8 0.9 2.0 3.2 4.4 6.0 5.2 4.8 8.8 11.7 23.4 15.1 16.6 17.6 21.6 23.7 24.3 45.6 50.1 35.2 15.5 54.5 69.5

0.5 0.1 2.3 5.4 10.2 29.6 64.7 66.4 53.9 78.8 106.5 153.8 163.4 152.1 149.8 208.4 297.5 428.1 376.1 348.7 519.8 622.9 826.8 1064.3 1557.5 1654.7 1417.2 189.8 259.9 225.1

39% 100% 92% 91% 97% 93% 92% 85% 67% 78% 80% 84% 85% 85% 87% 85% 86% 82% 86% 85% 88% 89% 88% 87% 86% 85% 88% 84% 74% 40%

30

Table II Panel A: Summary Statistics for Spreads (One-Month)

Whole sample

1991-1999

2000-2007

Mean

Std. Dev.

P25

P50

P75

LIBOR-Treasury

52.1

32.4

27.4

44.5

67.7

LIBOR-GC

16.3

10.2

11.5

15.0

19.5

GC-Treasury

35.7

28.4

14.2

29.0

50.0

LIBOR-Treasury

58.1

31.0

36.5

51.2

71.6

LIBOR-GC

15.3

10.4

9.8

14.5

19.2

GC-Treasury

42.7

26.6

24.5

36.0

56.0

LIBOR-Treasury

46.1

32.8

22.1

34.4

63.5

LIBOR-GC

17.3

9.8

12.5

15.5

19.8

GC-Treasury

28.8

28.5

8.4

17.1

44.5

Panel B: Summary Statistics for Spreads (Three-Month)

Whole sample

1991-1999

2000-2007

Mean

Std. Dev.

P25

P50

P75

LIBOR-Treasury

47.4

22.8

28.8

45.6

60.8

LIBOR-GC

21.1

10.1

15.5

19.5

24.5

GC-Treasury

26.3

19.1

10.5

24.5

38.0

LIBOR-Treasury

52.9

19.2

39.3

51.0

64.8

LIBOR-GC

21.4

9.3

16.3

20.3

25.0

GC-Treasury

31.5

17.9

17.5

29.5

42.0

LIBOR-Treasury

41.8

24.7

22.0

36.0

55.0

LIBOR-GC

20.7

10.9

15.0

18.5

23.5

GC-Treasury

21.1

18.9

5.5

15.5

33.0

31

Table III Panel A: Corporate Bond Issuance Year

AAA-rated

Non-AAA Investment Grade

Non-investment

Total

AAA as a % of Total

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

1.2 1.3 2.4 8.2 7.5 14.1 7.7 9.0 26.1 24.3 34.7 65.4 117.3 95.6 116.6 160.4 313.3 173.6 140.4 226.6 27.7 2.8

21.0 49.2 97.9 194.6 106.3 204.9 227.8 294.9 495.0 507.4 675.5 856.8 622.9 655.7 580.8 587.4 840.4 864.0 628.0 966.9 713.4 105.4

64.8 107.9 127.1 118.0 64.4 63.7 120.2 232.1 300.8 234.7 173.8 223.6 162.1 296.6 338.3 260.5 328.5 471.7 297.1 435.8 493.8 42.4

87.0 158.4 227.4 320.7 178.3 282.6 355.7 536.1 822.0 766.4 884.0 1145.8 902.4 1047.9 1035.7 1008.4 1482.2 1509.3 1065.6 1629.3 1234.8 150.6

1.3% 0.8% 1.1% 2.6% 4.2% 5.0% 2.2% 1.7% 3.2% 3.2% 3.9% 5.7% 13.0% 9.1% 11.3% 15.9% 21.1% 11.5% 13.2% 13.9% 2.2% 1.8%

32

Panel B: Outstanding of Treasury and ABCP Year

Total Treasury Outstanding

ABCP Outstanding

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

2180.6 2434.9 2739.0 2974.4 3111.0 3292.1 3444.7 3408.8 3272.8 3165.2 2830.7 2827.8 3058.3 3398.9 3699.7 3807.2 3884.7 4050.8 5253.4 6692.4 8236.8 9128.2

697.4 693.1 678.5 714.1 886.9 1125.3 831.8 704.5 451.6 377.8 350.6

33

Table IV Table 4: Correlation Between Predicted Spreads of Two Subsamples Spreads

Correlation

P-value

LIB_TR GC_TR TR_GC

0.47 0.4 0.52

<0.01 <0.01 <0.01

34

Table V Panel A: ABS/MBS Issuance on the Fitted Convenience Yield Measure (Seasonal Effect), Detrended Using H-P Filter (1) Fitted LIBOR-Treasury

(2)

(3)

(4)

(5)

(6)

(7)

(8)

0.057*** (3.80)

0.019 (0.53) 0.054*** (3.36)

0.003 (0.54)

0.050*** (3.59) 0.114*** (6.62)

0.046*** (3.29) 0.094*** (5.56) 10.552*** (27.44) 6.074*** (8.43) 5.398*** (9.89)

0.035*** (2.63) 0.102*** (6.06)

0.047*** (3.91)

Fitted LIBOR-GC

0.063* (1.83)

Fitted GC-Treasury ABS Issuance lag_1 day_1 day_15 day_30 Quarter end Constant Observations

-2.424*** (-3.66)

-1.001* (-1.73)

-2.025*** (-3.49)

-2.234*** (-3.23)

0.023 (0.09)

-1.758*** (-3.33)

-2.321*** (-4.33)

3.630*** (8.17) -1.531*** (-3.17)

4,044

4,044

4,044

4,044

3897

4,043

4,043

4,043

35

Panel B: ABS/MBS Issuance on the Fitted Convenience Yield Measure (Seasonal Effect), Detrended Using Piecewise Regressions (1) Fitted LIBOR-Treasury

(2)

(3)

(4)

(5)

(6)

(7)

(8)

0.057*** (3.79)

0.019 (0.53) 0.054*** (3.36)

0.003 (0.45)

0.050*** (3.58) 0.116*** (6.70)

0.046*** (3.28) 0.096*** (5.65) 10.552*** (27.39) 6.079*** (8.45) 5.390*** (9.85)

0.035*** (2.61) 0.104*** (6.16)

0.047*** (3.86)

Fitted LIBOR-GC

0.060* (1.74)

Fitted GC-Treasury ABS Issuance lag_1 day_1 day_15 day_30 Quarter end Constant Observations

-2.432*** (-3.65)

-0.986* (-1.68)

-2.046*** (-3.52)

-2.234*** (-3.23)

0.024 (0.09)

-1.773*** (-3.35)

-2.336*** (-4.36)

3.612*** (8.05) -1.546*** (-3.19)

4,044

4,044

4,044

4,044

3897

4,043

4,043

4,043

36

Table VI

Panel A: Corporate Bond Issuance on the Fitted Convenience Yield Measure (Seasonal Effect) Fitted LIBOR-Treasury

(1) -0.030*** (-4.38)

Fitted LIBOR-GC

(2)

Observations

(4)

-0.036* (-1.81) -0.028*** (-3.09) 1.583*** (4.17) 4,044

-0.058*** (-3.00)

Fitted GC-Treasury Constant

(3)

1.595*** (4.33)

0.954*** (2.99)

-0.033*** (-3.89) 1.200*** (3.83)

4,044

4,044

4,044

Panel B: ABCP Issuance on the Fitted Convenience Yield Measure (Seasonal Effect) Fitted LIBOR-Treasury

(1) 0.000*** (3.67)

Fitted LIBOR-GC

(2)

Observations

(4)

0.0002* (1.95) 0.0001*** (2.91) -0.0077*** (-4.19) 335

0.000*** (3.44)

Fitted GC-Treasury Constant

(3)

-0.007*** (-3.73)

-0.005*** (-3.23)

0.000*** (3.44) -0.006*** (-3.48)

335

335

335

37

Table VII Panel A: Treasury Issuance as Instrument, First Stage Treasury Issuance Lag_1

(1)

(2)

(3)

-181.732*** (-3.18)

-254.081*** (-3.02) -179.198** (-2.20)

35.712*** (25.24)

35.694*** (25.24)

-292.470*** (-2.99) -270.061** (-2.33) -191.611** (-2.15) 35.692*** (25.25)

3,897

3,896

3,895

Treasury Issuance Lag_2 Treasury Issuance Lag_3 Constant Observations

38

Panel B: Treasury Issuance as Instrument, Second Stage Fitted GC-Treasury Spread (Issuance)

(1)

(2)

(3)

(4)

0.422*** (3.91)

0.375*** (3.56) 0.114*** (6.52)

0.291*** (2.88) 0.095*** (5.50) 10.539*** (26.91) 5.882*** (8.16) 5.385*** (9.89)

0.318*** (3.00) 0.101*** (5.94)

ABS Issuance lag_1 Day_1 Day_15 Day_30 Quarter end

(5)

(6)

(7)

-97.318*** (-3.33)

-171.260*** (-5.23) -186.667*** (-5.85)

3.688*** (8.75)

Treasury Issuance Lag_1

-15.044*** (-3.92)

-13.374*** (-3.57)

-11.061*** (-3.08)

-11.663*** (-3.10)

0.016 (0.09)

0.017 (0.10)

-161.594*** (-4.73) -162.894*** (-4.30) 49.419 (1.48) 0.018 (0.10)

4,041

4,041

4,041

4,041

4,043

4,042

4,041

Treasury Issuance Lag_2 Treasury Issuance Lag_3 Constant Observations

39

Panel C: Seasonality of Treasury Issuance and ABS Issuance Treasury Issuance Lag_1 (Seasonal Part)

(1)

(2)

(3)

(4)

-107.463

-131.947*

-128.630*

-126.144*

(-1.57)

(-1.89)

(-1.84)

(-1.81)

-469.861***

-472.062***

-457.946***

(-6.48)

(-6.44)

(-6.32)

-95.472

-98.640

-73.421

(-1.30)

(-1.32)

(-0.99)

Treasury Issuance Lag_2 (Seasonal Part) Treasury Issuance Lag_3 (Seasonal Part) Treasury Issuance Lag_4 (Seasonal Part)

(5)

(6)

-95.358***

-112.451***

(-2.92)

(-3.00)

33.217 (0.46)

Treasury Issuance Lag_5 (Seasonal Part)

95.781 (1.43)

Fitted GC-Treasury

0.052*** (3.47)

Treasury Issuance Lag_1 (Non-Seasonal Part) Treasury Issuance Lag_2 (Non-Seasonal Part)

-46.535 (-1.18)

Treasury Issuance Lag_3 (Non-Seasonal Part)

131.418*** (3.66)

Constant

Observations

0.016

0.019

0.020

-1.845***

0.016

0.018

(0.09)

(0.11)

(0.12)

(-3.16)

(0.09)

(0.10)

4,043

4,041

4,039

4,041

4,043

4,041

40

Table VIII Panel A: OIS-Treasury Spreads (1) Fitted OIS-Treasury (Seasonal)

(2)

0.022 (1.63)

Fitted OIS-Treasury (Issuance) Constant

-0.506 (-1.46)

0.209*** (3.52) -5.023*** (-3.53)

Observations

4,044

4,041

Panel B: Three-month Spreads (1) Fitted LIBOR-Treasury (Seasonal)

(2)

(3)

(4)

(5)

0.063*** (2.76)

Fitted LIBOR-GC (Seasonal)

0.065 (1.49)

Fitted GC-Treasury (Seasonal)

0.073** (2.49)

Fitted GC-Treasury (Issuance)

0.455*** (3.40)

Fitted OIS-Treasury (Seasonal)

0.069*** (3.78)

Fitted OIS-Treasury (Issuance) Constant Observations

(6)

-2.958*** (-2.72)

-1.346 (-1.42)

-1.908** (-2.46)

-11.943*** (-3.41)

-1.426*** (-3.37)

0.506*** (3.57) -10.681*** (-3.58)

4,044

4,044

4,044

4,041

4,044

4,041

41

Panel C: Monthly (1) Fitted LIBOR-Treasury (Seasonal)

(2)

(3)

0.057*** (4.07)

Fitted LIBOR-GC (Seasonal)

0.192*** (4.57)

Fitted GC-Treasury (Seasonal) Constant Observations

-2.933*** (-3.83)

-3.117*** (-4.28)

0.063*** (3.39) -2.242*** (-3.18)

4,044

4,044

4,044

(2)

(3)

Panel D: Daily (1) Fitted LIBOR-Treasury (Seasonal)

0.208*** (3.19)

Fitted LIBOR-GC (Seasonal)

-0.406*** (-2.87)

Fitted GC-Treasury (Seasonal) Constant Observations

42

-10.797*** (-3.16)

6.660*** (2.87)

0.446*** (5.48) -15.899*** (-5.43)

4,044

4,044

4,044

Panel E: Including Crisis (1) Fitted LIBOR-Treasury (Seasonal)

(2)

(3)

(4)

(5)

0.031*** (3.06)

Fitted LIBOR-GC (Seasonal)

0.022 (1.03)

Fitted GC-Treasury (Seasonal)

0.061*** (4.24)

Fitted GC-Treasury (Issuance)

0.596*** (5.29)

Fitted OIS-Treasury (Seasonal)

0.024** (2.09)

Fitted OIS-Treasury (Issuance) Constant Observations

(6)

-1.754*** (-2.96)

-0.466 (-0.99)

-2.134*** (-4.02)

-20.884*** (-5.31)

-0.708** (-1.99)

0.727*** (6.02) -21.080*** (-6.05)

4,932

4,932

4,932

4,929

4,932

4,929

Panel F: Rolling Regressions (1) Fitted LIBOR-Treasury (Seasonal)

(2)

(3)

0.023** (2.06)

Fitted LIBOR-GC (Seasonal)

0.053* (1.88)

Fitted GC-Treasury (Seasonal) Constant Observations

43

-1.379* (-1.90)

-0.936* (-1.70)

0.025* (1.90) -1.086* (-1.74)

2,398

2,398

2,398