Citation
Essays in Debt Structure and Maturity Structure

Material Information

Title:
Essays in Debt Structure and Maturity Structure
Creator:
Badoer, Dominique C
Place of Publication:
[Gainesville, Fla.]
Florida
Publisher:
University of Florida
Publication Date:
Language:
english
Physical Description:
1 online resource (134 p.)

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Business Administration
Finance, Insurance and Real Estate
Committee Chair:
JAMES,CHRISTOPHER
Committee Co-Chair:
HOUSTON,JOEL F
Committee Members:
NARANJO,ANDY
HAMILTON,JONATHAN H
Graduation Date:
8/9/2014

Subjects

Subjects / Keywords:
Assets ( jstor )
Business structures ( jstor )
Capital structure ( jstor )
Credit ratings ( jstor )
Creditors ( jstor )
Debt ( jstor )
Financial investments ( jstor )
Investment credit ( jstor )
Loans ( jstor )
Term loans ( jstor )
Finance, Insurance and Real Estate -- Dissertations, Academic -- UF
creditor-concentration -- debt-maturity -- debt-structure -- distress-costs -- gap-filling
Genre:
bibliography ( marcgt )
theses ( marcgt )
government publication (state, provincial, terriorial, dependent) ( marcgt )
born-digital ( sobekcm )
Electronic Thesis or Dissertation
Business Administration thesis, Ph.D.

Notes

Abstract:
This dissertation examines the determinants of corporate debt maturity choice as well as the drivers of corporate debt specialization. The first part of this dissertation examines the importance of gap filling behavior, in which highly rated issuers fill gaps in the supply of long-term government bonds, as a determinant of very long-term corporate debt issues, using a large sample of individual corporate bonds and term loans issued by public U.S. companies between 1987 and 2009. It documents that gap filling behavior is more prominent in the very long end of the maturity spectrum where the required risk capital makes it difficult for arbitrageurs to smooth out supply shocks. Moreover, it provides evidence that changes in the supply of long-term government bonds not only affect the choice of maturity but also the level of corporate borrowing. The second part of this dissertation analyzes why some firms borrow primarily using one type of debt, and thereby specialize their debt structure, while other firms diversify across different debt types, and how this tendency for specialization is related to financial distress costs. It documents that debt specialization is only loosely correlated with creditor concentration (i.e. the number of creditors holding the debt). Additionally, it provides evidence that firms shift their debt structure away from dispersedly held debt towards more concentrated debt in response to declining operating performance. Furthermore, it documents that companies that primarily rely on concentrated intermediated bank debt recover faster from industry wide downturns than firms with more diverse debt structures, but that their reliance on intermediated bank debt leaves them more vulnerable to solvency and liquidity shocks to the banking sector. ( en )
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis:
Thesis (Ph.D.)--University of Florida, 2014.
Local:
Adviser: JAMES,CHRISTOPHER.
Local:
Co-adviser: HOUSTON,JOEL F.
Electronic Access:
RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2016-08-31
Statement of Responsibility:
by Dominique C Badoer.

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Embargo Date:
8/31/2016
Resource Identifier:
968131481 ( OCLC )
Classification:
LD1780 2014 ( lcc )

Downloads

This item has the following downloads:


Full Text

PAGE 1

ESSAYS IN DEBT STRUCTURE AND MATURITY STRUCTURE By DOMINIQUE C. BADOER A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2014

PAGE 2

© 2014 Dominique C. Badoer

PAGE 3

To my family Your love and support have not gone unnoticed

PAGE 4

4 ACKNOWLEDGMENTS I want to sincerely thank Christopher James, my dissertation committee chair, for his advice, guidance , and support throughout the doctoral program. His contribution to my research and my growth as an aca demic has been invaluable. I also want to thank my dissertation committee, Joel Houston, Andy Naranjo, and Jonathan Hamilton for guiding this dissertation and providing valuable feedback. Finally, I am especially grateful to all my friends and colleagues f or taking the time to read and comment on the individual chapters of my dissertation and for their encouragement and support throughout the doctoral program.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURE S ................................ ................................ ................................ .......... 9 ABSTRACT ................................ ................................ ................................ ................... 10 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 12 Overview of Chapter 2 ................................ ................................ ............................ 13 Overview of Chapter 3 ................................ ................................ ............................ 15 2 THE DETERMINANTS OF LONG TERM CO RPORATE DEBT ISSUANCES ....... 18 Motivation ................................ ................................ ................................ ............... 18 Background: Corporate Maturity Choice and the Limits of Arbitrage ...................... 25 Data ................................ ................................ ................................ ........................ 30 Empirical Evidence ................................ ................................ ................................ . 33 Descriptive Statistics ................................ ................................ ........................ 33 Empirical Models of Debt Maturity Choice ................................ ........................ 36 Linear Models of Debt Maturity Choice ................................ ............................ 37 Multinomial Logit Models of Debt Maturity Choice ................................ ............ 40 Gap Filling and the Propensity to borrow long term ................................ .......... 43 Endogeneity Concerns ................................ ................................ ............................ 46 Event Study ................................ ................................ ................................ ...... 46 Corporate Supply Response ................................ ................................ ............ 48 Chapter 2 Concluding Remarks ................................ ................................ .............. 49 3 DEBT SPECIALIZATION AND CREDITOR CONCENTRATION ............................ 72 Motivation ................................ ................................ ................................ ............... 72 Data ................................ ................................ ................................ ........................ 80 Debt Structure and Measures of Debt Specialization ................................ ....... 81 Loan level Measures of Creditor Concentration ................................ ............... 82 Major Shifts in Debt Structure and Creditor Structure ................................ ...... 83 Distressed Industries and firm level Measures of Creditor Concentration ........ 85 Empirical Analysis ................................ ................................ ................................ ... 86 Descriptive St atistics ................................ ................................ ........................ 86 The Relationship between Debt Specialization and Creditor Concentration .... 90 Substantial Shifts in Debt Structure and Creditor Structure .............................. 95 Debt Specialization, Creditor Concentration, and Distress Costs ..................... 99

PAGE 6

6 Chapter 3 Concluding Remarks ................................ ................................ ............ 103 4 CONCLUSION ................................ ................................ ................................ ...... 119 A PPENDIX : DATA AND VARIABLE DESCRIPTIONS ................................ ................ 121 LIST OF REFERENCES ................................ ................................ ............................. 131 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 134

PAGE 7

7 LIST OF TABLES Table page 2 1 Annual implied price volatilities. ................................ ................................ .......... 51 2 2 Summary statistics by final maturity of the issue. ................................ ............... 52 2 3 Maturities by S&P issuer rating. ................................ ................................ .......... 54 2 4 Maturities by S&P issuer rating (excluding credit line s). ................................ ..... 55 2 5 Linear models of debt maturity choice. ................................ ............................... 56 2 6 Linear models of debt maturity choice (different subsamples). ........................... 57 2 7 Multinomial logit model of debt maturity choice. ................................ ................. 58 2 8 Marginal effects of multinomial logit model. ................................ ........................ 59 2 9 Logit models of issuance. ................................ ................................ ................... 60 2 10 Marginal effects of logit model. ................................ ................................ ........... 61 2 11 Annual change in total debt. ................................ ................................ ............... 62 2 12 Event study relative changes in yields around Treasury announcement. ........ 63 2 13 Natural experiment all issues. ................................ ................................ .......... 64 2 14 Natural experiment issues rated A AAA. ................................ ......................... 65 2 15 Natural experiment issues rated BBB. ................................ ............................. 66 3 1 Summary statistics by rating category. ................................ ............................. 105 3 2 Summary statistics Dealscan sample (investment grade rated fir ms). ............. 107 3 3 Summary statistics Dealscan sample (non investment grade rated firms). ...... 108 3 4 Linear models of syndicate size (investment grade rated firms). ...................... 110 3 5 Linear models of syndicate size (non investment grade rated firms). ............... 111 3 6 Logit mode ls of institutional loan tranche. ................................ ......................... 112 3 7 Summary statistics for firms with major shifts between public and bank debt. . 113 3 8 Logit models of major shifts in debt structure. ................................ .................. 114

PAGE 8

8 3 9 Instrumental variable probit model of major shifts in debt structure. ................. 115 3 10 Fraction of floating rate debt as a proxy for fraction of bank debt. .................... 116 3 11 Summary statistics for firms in distressed industries. ................................ ....... 117 3 12 Logit models of operating performance for firms in distressed industries. ........ 118 A 1 Variable definitions for Chapter 2. ................................ ................................ .... 125 A 2 Variable definitions for Chapter 3. ................................ ................................ .... 128

PAGE 9

9 LIST OF FIGURES Figure page 2 1 Monthly time series of TSY1 and TSYMAT. ................................ ....................... 67 2 2 Monthly time series of TSY5 and TSYMAT. ................................ ....................... 68 2 3 Monthly time series of TSY20 and TSYMAT. ................................ ..................... 69 2 4 Monthly time series of TSY20 and TSYGDP. ................................ ..................... 70 2 5 Average composition of long term debt issues around Treasury e vent. ............. 71

PAGE 10

10 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy ESSAYS IN DEBT STRUCTURE AND MATURITY STRUCTURE By Dominique C. Badoer August 2014 Chair: Christopher James Major: Business Administration This dissertation examines the determinants of corporate debt maturity choice as well as the drivers of corporate debt specialization. The first part of this dissertation examines the importance of gap filling behavior, in which highly rated issuers fill gaps in the supply of long term government bonds, as a determinant of very long term corporate debt issues, using a large sample of individual corporate bonds and term loa ns issued by public U.S. companies between 1987 and 2009. It documents that gap filling behavior is more prominent in the very long end of the maturity spectrum where the required risk capital makes it difficult for arbitrageurs to smooth out supply shocks . Moreover, it provides evidence that changes in the supply of long term government bonds not only affect the choice of maturity but also the level of corporate borrowing . The second part of this dissertation analyzes why some firms borrow primarily using one type of debt, and thereby specialize their debt structure, while other firms divers ify across different debt types, and how this tendency for specialization is related to financial distress costs. It documents that debt specialization is only loosely c orrelated with creditor concentration (i.e. the number of creditors holding the debt). Additionally, it provides evidence that firms shift their debt structure away from dispersedly held debt

PAGE 11

11 towards more concentrated debt in response to declining operatin g performance. Furthermore, it documents that companies that primarily rely on concentrated intermediated bank debt recover faster from industry wide downturns than firms with more diverse debt structures, but that their reliance on intermediated bank debt leaves them more vulnerable to solvency and liquidity shocks to the banking sector.

PAGE 12

12 CHAPTER 1 INTRODUCTION This dissertation examine s what determines corporate debt maturity choice as well as what drives the tendenc y of many firms to rely exclusively on one type of debt for their financing. In the first part of this dissertation, Chapter 2, we analyze corporate debt maturity choice . In particular, we examine why a substantial portion of debt issued by U.S. public fir ms has maturities of 20 years or more. W e provide evidence that gap filling behavior, in which high grade issuers fill gaps in the supply of long term government bonds, is an important determinant of these very long term issues. Specifically, we find that the likelihood if issuing debt with a maturity of 20 years or more is negatively and significantly related to the supply of long term Treasury bonds. Moreover, we provide evidence that changes in the supply of long term government bonds not only affect the maturity choice of firms, but also the overall level of corporate borrowing. In the second part of this dissertation, Chapter 3, we analyze why some firms specialize their debt structure by borrow ing primarily through one type of debt, such as senior bond s or term loans, while other firms diversify across different debt types , and more specifically, how this tendency for debt specialization is related to financial distress costs. We argue that debt specialization alone is unlikely to mitigate financial distress costs, unless it is accompanied by greater creditor concentration (i.e. a reduction in the number of lenders holding the debt). Consistent with this notion, we find that firms shift their debt structure away from dispersedly held public debt towar ds more tightly held intermediated bank debt in response to a decline in operating performance. Furthermore, we document that companies that rely primarily on intermediated bank

PAGE 13

13 debt recover faster from industry wide distress than firms with more diverse d ebt structures, but that their reliance on banks leaves them more vulnerable to solvency and liquidity shocks to the banking sector. Overview of Chapter 2 Between 1987 and 2009 a substantial number of public U.S. companies issued very long term debt (with maturities greater than 20 years) that was not tied to a particular tangible asset. Given that we document a mismatch of over 20 years between the average maturity of long term debt issues and the average maturity of long term issuer s hese unsecu red long term issues are difficult to explain in the context of agency cost theories of maturity choice, such as Myers (1997), which predict that issuers will match the maturity of their liabilities to the maturity of their assets. In this chapter we addre ss to what extent these long term issues can be explained by variation in credit market conditions, such as the supply of long term government debt or the term structure of interest rates, and to what extent changing credit market conditions affect the pro pensity of firms to borrow. We are particularly interested in the relationship between the supply of long term government debt and corporate debt maturity choice given recent work by Greenwood, Hanson, and Stein (2010). In their theoretical model importan t classes of investors, such as pension funds and insurance companies, have a preference for relatively safe long term debt, and given a shock to the supply of long term government bonds arbitrageurs have limited ability to fill the gap because of the high costs of capital associated with trades at the long end of the term structure. As a result , highly rated corporate issuers have a comparative cost advantage over arbitrageurs and attempt to exploit the differences in the expected return on short versus lo ng term bonds, created

PAGE 14

14 by the shock to the supply in long term government bonds, by issuing more long term debt. We empirically examine the determinants of corporate debt maturity choice at the firm level by using a large dataset of corporate loan and secu rity issues from Dealscan and SDC Platinum for the 1987 to 2009 period . We estimate both linear as well as multinomial logit models relating debt maturity choice to credit market conditions. Overall , our results suggest that market conditions, and particul arly the supply of long term debt issues. Specifically, we find a negative and significant relationship between the supply of long term government bonds and corporate debt maturity choice in the 20 year plus segment, but we find no evidence of gap filling in the very short end of the market. Moreover, b ecause the model by Greenwood, Hansen, and Stein (2010) implies that gap filling should be most prominent among close substi tutes to Treasury bonds, we also explore which types of securities have the highest supply elasticity with respect to changes in the supply of long term government bonds. Consistent with expectations, we document that supply elasticities are greatest for r ated fixed rate non callable bonds and bonds by issuers with a credit rating between A and AAA. In order to address endogeneity concerns we use the fraction of government debt to GDP as an instrument for the supply of long term Treasures in our linear mod els of debt maturity choice, and in addition , we exploit the suspension of 30 year Treasury bonds in 2001 as a natural experiment. Overall, the results from these additional tests confirm that gap filling is an important and statistically significant deter minant of corporate debt maturity choice for highly rated issuers.

PAGE 15

15 We further provide evidence that while the supply of long term government debt predicts long term corporate debt issuances for highly rated issuers, this corporate supply response does not come at the expense of a substitution between long term and short term debt issues. Specifically, for firms with a credit rating between A and AAA we find a significant and negative relationship between changes in firm borrowing and the supply of long term Treasuries. Overview of Chapter 3 Between 2002 and 201 1 almost half of U.S. public firms obtained at least 90 % or more of their debt financing through one form of debt alone. For instance, firms with an investment grade credit rating rely primarily on senior unsecured bonds for their debt financing, whereas a significant number of lower rated and unrated firms also limit their borrowing to one source, such as term loans, credit lines, senio r or subordinated bonds . While the current empirical literature has documented this tendency for debt specialization, it does not provide clear evidence as to what drives firms to rely on one source of debt alone. In this chapter we take a closer look at why some firms borrow primarily through one type of debt and why others prefer to diversify across different debt types. We are particularly interested in the relationship between debt specialization and creditor concentration (i.e. the number of creditors holding the debt) and how these factors are related to financial distress costs. We argue that both debt specialization and creditor concentration are important drivers in mitigating financial distress costs, and that efforts by firms to mitigate distress should be most effective if a reduction in the number of different claim types is accompanied by a reduction in the number of lenders.

PAGE 16

16 We empirically examine the relationship between financial distress costs and debt structure using detailed debt structur e information from Capital IQ for U.S. public firms from 2002 through 2011. We proxy for creditor concentration by using information on lenders in syndicated loans from Dealscan as well as distinguishing between dispersedly held public bonds and private pl acements in corporate debt structures through the presence of a firm level credit rating. Consistent with the notion that efforts to mitigate financial distress costs involve a reduction in the number of debt claims as well as a reduction in the number of lenders, we find that a significant number of firms substantially alter their debt structure in response to declining operating performance. Specifically, we use a series of logit models to estimate the likelihood of firms shifting away from dispersedly he ld public debt towards more bank debt for a sample of rated firms. W e document that the likelihood of companies shifting away from public debt towards more bank debt is EB ITDA to sales growth. Moreover, we find that this increasing reliance on bank debt is mostly driven by increasing tightly held intermediated bank debt and not dispersedly held institutional bank debt. Furthermore, we mitigate endogeneity concerns by instru menting our measure of corporate operating performance with a trade weighted U.S. Dollar index and an economic policy index in an instrumental variables probit model . Our results are robust to instrumenting operating performance with these two indices. We next analyze whether debt specialization is associated with lower costs of distress and whether this relationship is stronger for companies with more concentrated

PAGE 17

17 creditor holdings. To measure costs of distress we identify firms in industries experiencing economic distress, as measured by steep declines in stock prices and negative sales growth, and examine whether differences in debt structure and creditor concentration prior to the distress period are related to operating performance during the distress p eriod. Our primary measures of operating performance are industry adjusted growth in EBITDA to sales ratio and CAPX to sales ratios. In logit models , in which we estimate the likelihood of above industry median operating performance during the distress pe riod, we find that companies that rely exclusively on intermediated bank debt recover faster from industry wide downturns, as measured by a statistically significant increase in the likelihood of above industry median operating performance, than firms that have more diverse debt structures or that rely heavily on institutional bank debt. However, our results also suggest that firms relying primarily on intermediated bank debt do not benefit from this specialization during solvency and liquidity shocks to th e banking sector.

PAGE 18

18 CHAPTER 2 THE DETERMINANTS OF LONG TERM CORPORATE DEBT ISSUANCES Motivation While the average maturity of corporate borrowing is about 5 years, companies sometimes issue very long term bonds not tied to a particular tangible asset . Perhaps bonds. 1 More recently , Apple issued $3 billion in 30 year fixed rate bonds in 2013 . 2 While 100 year bonds are rare, unsecured bond offerings with maturities of more than 20 years are common. For example, between 1987 and 2009, 472 firms covered by Compustat made 1620 unsecured bond offerings with maturities of more than 20 years and a total face value (in 2009 dollars) of $590 billion (the median maturity of these long term offerings was just over 30 years). Virtually all of these issuers were also active in the short end of the maturity spectrum and regularly issued debt with ma turities of less than five years. As we outline below, very long term unsecured debt issues and significant year to year variation in the maturity of issues are difficult to explain in the context of contracting cost theories of maturity choice in which th e average maturity of assets and the importance of growth options drive debt maturity choice. 3 In this paper we examine the determinants of long term debt issues using a dataset of bond issues and bank borrowing by public firms during the 1987 through 1 The original offering was for $150 million but was increased to $300 million due to unexpectedly strong investor demand. The bonds were priced at 80 basis points over 30 year Treasury bonds. 2 In May 2013 Apple issued $3 billion of 30 year bonds with a fixed rate of 3.85%. These bonds were priced at 100 basis points over the 30 year Treasury rate. 3 See for example Myers (1977), Barclay and Smith (1995 ) , and Guedes and Opler (1996) for discussi ons of agency cost explanations of maturity choice.

PAGE 19

19 2009 time period. We address three related questions. First, to what extent are very long term corporate debt issues a response to changes in credit market conditions, such as changes in the maturity composition of U.S. Treasury debt or changes in the term str ucture of interest rates? Second, what types of borrowers have the greatest supply elasticity with respect to changes in credit market conditions? Third, to what extent do changes in the maturity composition of Treasury bonds and the term structure of inte rest rates affect the propensity of firms to borrow? Our analysis is motivated by recent work by Greenwoo d, Hanson , and Stein , and Stein (2010) develop a theory in which corporat e issuers respond to shocks in the supply of short and long term Treasury bonds. The basic idea is that important classes of investors, such as pension funds and insurance companies, have a preference for relatively safe long term assets. Given a shock to the supply of long term government bonds, the cost and availability of risk capital limits the ability of arbitrageurs to fill the gap. As a result, bond yields can stray from the yields implied by the expectations hypothesis. Corporate issuers, particular ly those with investment grade credit ratings, attempt to exploit differences in the expected return on short versus long term bonds leading to supply shifts in the long term corporate market offsetting changes in the supply of long term government bonds. We hypothesize that gap filling is likely to be a more important determinant of very long term corporate borrowing (20 years or more) than for shorter term borrowing. There are a couple of reasons we expect this to be the case. First, the gap filling hypot hesis is based on highly rated corporate issuers having a comparative cost

PAGE 20

20 advantage in arbitraging rate differences arising from shifts in the maturity composition of government debt. An important factor affecting the cost of arbitrage activity is the amo unt of capital arbitrageurs must commit to a position and the amount of capital available to them. Capital constraints include limits on leverage and costs of raising external equity. 4 Given capital constraints, arbitrage costs are likely to vary directly with the amount of risk capital needed. The amount of risk capital needed is in turn likely to vary with the price volatility associated with a position. As shown in Table 2 1 , the average daily implied price volatility of Treasury bonds increases with the ir maturity, with the implied price volatility associated with 20 year Treasury bonds averaging over five times the daily price volatility of two year bonds. Thus, for any value at risk (VAR) level the amount of risk capital required for trades in very lon g term Treasuries is orders of magnitude larger than in the short end of the market. The second reason we expect gap filling to be an important determinant of long term issues is that very long term issues are difficult to explain in the context of agency cost theories of maturity choice, which predict issuers will match the maturity of their liabilities to the maturity of their assets (see Myers (1977)). In our sample, long term issues (20 plus years) average 30 years while the average maturity of long ter assets is about 8 years a mismatch of over 20 years. Moreover, for long term issuers we find little evidence that maturity of debt issues is related to within firm variation in asset maturity. In contrast, maturity matching appears more preval ent for short term issuers. For example, among firms that never issue very long term debt, the median 4 See Gromb and Vayanos (2010) for a survey of the theoretical literature on financing constraints and the limits of arbitrage.

PAGE 21

21 maturity of assets matches quite closely the median maturity of debt issued (3 years versus 4.19 years). Our analysis is based on a detailed sample of ind ividual corporate loans and debt security issues collected from the SDC and Dealscan databases. These data allow us to focus on the determinants of debt maturity choice at the individual firm level rather than on time series variation in aggregate debt iss uance. This distinction is important, since information and agency costs potentially limit the ability of many firms to exploit perceived gap filling opportunities. Moreover, gap filling predicts that corporations will issue forms of debt that are close su bstitutes to Treasury securities. By using firm level panel data on issuances rather than aggregated data we are better able to identify the effects of state variables that vary over time. Moreover, using firm level data we can examine how changes in marke t conditions affect not only the choice of debt maturity but also the propensity to borrow long term. Issuance data have the further advantage of enabling us to measure the maturity of debt offerings more precisely. In particular, previous studies of corpo rate maturity choice generally classify debt as either short term (under one or five years in maturity) or long term (over one or five years). 5 This choice is motivated in part by data limitations since Compustat classifies debt according to maturity but o nly up to five years and the year. Using more granular data on maturity choice allows us to examine whether gap filling occurs within specific segments of the term structu re. 5 For example, see Greenwood, Hanson, and Stein (2010) and Barclay and Smith (1995).

PAGE 22

22 We begin by discussing why the cost of arbitrage is likely to be greater in the very long term segment of the market. We show that price volatility and margin requirements suggest that the amount of risk capital needed for arbitrage trades in the long end of the market is significantly greater than in the shorter end of the term structure. We next examine the characteristics of long term debt issuers. Given the potential agency costs associated with issuing long term debt, it is perhaps not surprising that we find that all but a few long term (20 years or more) bond issues are by firms with investment grade ratings. Offerings by these issuers are distributed fairly uniformly over the maturity spectrum. For example, in our sample firms rated AA or higher issue about as many bonds with maturities under five years as they do bonds with a maturity of 20 years or more. The fact that investment grade issues are not clustered in a particular segment of the maturity spectrum is consistent with the hypothesis tha t the maturity choice of these firms is more sensitive to market conditions than for lower rated firms. We examine the determinants of maturity choice by estimating models relating maturity choice to market conditions, including the maturity composition of Treasury debt. Controlling for firm characteristics we find a negative and significant relationship between the maturity composition of corporate debt issues and the maturity composition of outstanding Treasury debt. More importantly, consistent with the idea that arbitrage costs are higher in the long term segment of the market, we find that gap filling is a significant determinant of maturity choice only in the 20 year plus segment of the corporate market. Indeed, we find no evidence of gap filling in th e short end of the

PAGE 23

23 market (one year or less). In addition, we find that the supply elasticity is greatest for securities that are closer substitutes for long term Treasury bonds. For instance, we find that the supply elasticity is greatest for investment g rade fixed rate non callable bonds . For highly rated firms we find that their long term bond issues are quite sensitive to changes in the supply of long term Treasuries. For example, we find that a 1% decrease in the proportion of outstanding Treasury issu es with maturities of 20 years o r more is associated with a 2. 6 % increase in the likelihood of a highly rated corporate issue of 20 years or more. Our results are consistent with those of Krishnamurthy and Vissing Jorgensen (201 1 , 201 2 ) who find little effect of changes in supply of Treasuries on corporate yields except for bonds rated A or better. We conduct an analysis of covariance (ANCOVA) to decompose variation in maturity choice attributable to different factors. We find that the majority of the va riance in maturity choice over time for highly rated firms that is not attributable to firm fixed effects is attributable to gap filling measures, which explain about 6.9 % of the variation, and not to changes in credit risk premia or the term structure of interest rates, which only explain about 3.2 % and 0.5 % of the total variation respectively. For firms rated BBB changes in credit risk premia explain the majority of variance in maturity choice, after firm fixed effects, and gap filling measures only accou nt for a relatively small portion of the overall variance (about 2.3 % ). We also find evidence that the supply of long term Treasury debt predicts corporate long term debt issuances and does not just lead to a substitution between long term debt and short t erm debt issues. Consistent with the gap filling argument we

PAGE 24

24 find a relationship between changes in firm borrowing and the supply of long term Treasury debt only for investment grade issuers. One concern with using shifts in the composition of Treasury deb t to analyze gap filling is that changes in the supply of Treasury debt may not be exogenous in the context of corporate debt maturity choice. Specifically, it might be the case that changes in the supply of Treasury debt respond endogenously to the same f actors that influence corporate maturity choice (albeit in the opposite direction). Our ability to focus on narrow segments of the term structure and on issuers with different credit ratings mitigates concerns with endogeneity. Nevertheless, we address end ogeneity concerns by year bonds in 2001 on 30 year corporate bond issues. As discussed later, this decision was not expected by investors and thus serves as a natural expe riment to examine the effect of a supply shock on the issuance of long term corporate bonds. Consistent with the announcement being unexpected we find a positive and significant decrease in yields of long term Treasury bonds when the suspension is announce d. Moreover, consistent with gap filling, we find a significant increase in the issuance of highly rated 30 year corporate bonds, relative to 20 year bond issues, subsequent to the suspension of 30 year Treasury bond issues. Our analysis contributes to the literature in several ways. First, using issuance data we provide evidence that corporate gap filling is limited to the very long end of the term structure. Second, using individual firm issuance data we examine in a rigorou s debt issued. Third, our analysis provides insights into the effects of policy initiatives

PAGE 25

25 such as the recent Federal Reserve Board large scale asset purchases (LSAP s) also referred to as quantitative easing. The various stages of recent LSAPs have been term Treasury Bonds and mortgage backed securities (MBS). 6 These policies were designed to spur long term borrowing by increasing reserves and changing the relative supply of long term government bonds available to private investors. 7 Several recent papers (See Swanson (2011) as well as Krishnamurthy and Vissing Jorgensen (2011)) provide evidence tha t the spillover effect on the prices of other assets has been limited. For example, Krishnamurthy and Vissing Jorgensen (2011) find that the impact of the term Treasury bonds on corporate yields has been limited to the yi eld on very high grade corporate debt. We provide firm level evidence that suggests that the spillover effects of these programs affect both the maturity choice as well the propensity of highly rated firms to borrow. We find no evidence that changes in the supply of long term Treasury bonds are related to the likelihood of issuing long term debt or the borrowing propensity of firms with below A credit ratings. Background: Corporate Maturity Choice and the Limits of Arbitrage Our analysis is motivated by two strands of the literature on corporate maturity choice. The first is based on agency and information problems. As discussed in Guedes and Opler (1996), the risk of being unable to roll over short term debt because of 6 See Krishnamurthy and Vissing Jorgensen (201 2 ) for a discussion of these initiatives. These initiatives 7 Fed Funds and short term Treasury rates were near z ero between December 2008 and the end of 2013. As a result, further monetary accommodation through lowering short term target interest rates was not feasible. Consequently, the Fed attempted to provide further stimulus through the purchase of long term Tre asury Bonds and debt as well private sector MBSs. For example, in a press release by the Board of Governors of the Federal Reserve System (dated March 18 2009) conditions in private credit markets, the Committee decided to pu rchase up to $300 billion of longer term

PAGE 26

26 adverse economic shocks or a deteriora tion of firm quality provides an incentive for firms to issue long term debt (see also Diamond (1991)). While rollover risk creates a preference for long term borrowing, adverse selection and agency costs serve to screen low quality firms out of the long e nd of the market. 8 Agency costs, in particular debt overhang and asset substitution problems, create incentives for lower quality firms to match the maturity of their borrowing to the maturity of their assets (see Myers (1977)). The second strand of the li terature comprises recent work on market segmentation and the effects of changes in the supply of Treasury securities on corporate borrowing costs (see for example Vayanos and Vila (2009), Greenwood and Vayanos (2010), G reenwood, H anson, and S tein (2010) , Krishnamurthy and Vissing Jorgensen (201 1 ), and Krishnamurthy and Vissing Jorgensen ( 201 2 ) ). 9 This literature is based on the idea that bond markets are segmented and that this segmentation is a ies with specific attributes. For example, Vayanos and Vila (2009) develop a model in which investor preferences lead to risk premia that vary with bond maturity. G reenwood, H anson, and S tein (2010) extend the model by Vayanos and Vila (2009) by introducin g firms who respond to excess demand in segments of the term structure by issuing bonds with durations that offset changes in supply of Treasury securities with the same duration. Krishnamurthy and Vissing Jorgensen (201 2 ) argue that bond markets are more narrowly segmented 8 In a recent paper, He and Xiong (2012) present a model in which debt maturity plays an important role over risk. They show that while sho rter term individual bonds are less risky, a shorter maturity of all bonds issued by a firm can exacerbate roll over risk and thus overall credit risk. The basic idea is that shorter overall maturity of debt forces equity holders to absorb losses more quic kly thus reducing the default threshold. 9 Our analysis is also motivated the market timing literature . S ee for instance Baker and Wurgler (2000) and Baker, Greenwood, and Wurgler (2003) that argue limits to arbitrage create opportunities for firms to profit by timing their debt and equity issues.

PAGE 27

27 and that only the highest rated corporate bonds provide viable substitutes for Treasury securities in meeting the demand for safe, long term assets. These models predict that corporate gap filling will be positively related to the cost o f arbitrage and limited to highly term assets (as in Krishnamurthy and Vissing Jorgensen (201 2 ) ), or because high agency costs limit the supply response of less creditworthy borro wers (as in G reenwood, H anson, and S tein (2010) ). Krishnamurthy (2010) identifies three considerations for every debt market purchase: risk capital, the haircuts in the repo market, and counterparty risk. The amount of risk capital and repo haircuts are li kely to be higher for trades involving long term Treasury securities for at least two reasons. First, unless changes in short term spot rates are negatively correlated with expected future spot rates, price volatility will increase with the duration of the bond. Consistent with this expectation, we find that the daily implied price volatility of Treasury bonds increases with the maturity of the bonds. Specifically, we computed the annual averages of implied daily price volatilities from call options on futu res contracts associated with Treasury bonds with maturities of two, five, ten and 20 years over the period from 1994 to 2009. 10 As shown in Table 2 1, the average implied daily price volatility of 20 year Treasury bonds is nearly 10 % during the 1994 through 2009 time period while the average daily implied volatility of two year Treasury Bonds is less than 2% . Moreover, the range of annual average volatilities is much greater in the long end than in the short end of the market. The hig her price 10 We obtain similar results if we compute the average of annual realized 90 day volatilities of the corresponding Treasury bond futures. Information on Treasury Bond futures and call options on Treasury Bond futures is available from Bloomberg beginning in 1994.

PAGE 28

28 volatility of long term Treasuries implies that more risk capital must be devoted to trades involving long term Treasuries. Second, and consistent with the greater price volatility of long term Treasuries, margin requirements (i.e. liquid capital requirements) are higher for positions in long dated Treasuries. For example, as outlined in Title 17 , Section 402.2 of the Code of Federal Regulations , the required margin, or market risk haircut, for broker dealers in Treasury Securities is increasing i n the duration and term to maturity of their bond holdings. For example, the net position haircut factor for coupon bonds with a ma turity of over 15 years is 4.5 % versus 0.12 % for three month Treasuries. 11 A recent example of the limits of arbitrage in the very long end of the term structure is the negative 30 year swap spread that developed in November 2008. The 30 year swap spread measures the difference between the rate on 30 year Treasury bonds and the fixed rate associated with a 30 year fixed for float ing (LIBOR) swap. Negative spreads are anomalous since arbitrageurs can essentially enter into zero interest rate risk transactions by purchasing Treasuries financed by repos and then entering into a swap to pay fixed and receive floating. Krishnamurthy (2 010) attributes the negative spread to the amount of risk capital needed to support such a trade until convergence. What is interesting in the context of our analysis is that during the period from 2006 to 2009, negative swap spreads only appeared for 30 y ear Treasuries (and 11 See 17 CFR 402.2 Capital requirements for registered government securities brokers and dealers and specifically 17 CFR § 402. 2(f)(1) for the haircut schedule. Note that repo haircuts also tend to be higher when collateral consists of long term Treasury bonds. As discussed in Copeland, Martin, and Walker (2010) haircuts in the repo market and in particular in the tri party repo m arket reflect primarily the strength of the counterparty and secondarily the price risk of the underlying collateral. Repo haircuts set by clearing houses reflect the market risk of the underlying collateral. Clearing houses sit in the middle of a trade, a ssuming the counterparty risk involved when two parties (or members) trade. When the trade is registered with a clearing house it becomes the legal counterparty to the trade, ensuring the financial performance.

PAGE 29

29 not for example for ten or five year Treasuries). This finding is consistent with the hypothesis that arbitrage costs are higher in the very long end of the market. 12 Previous empirical studies on the relationship between corporate debt maturity choice and government debt maturity measure government debt maturity as share of government debt outstanding with a maturity of one year or more or, to address endogeneity concerns, the ratio of total government debt to GDP (see G reenwood, H anson , and S tein (2010) and Krishnamurthy and Vissing Jogensen (201 2 )). A potential concern with using these broad measures when analyzing corporate maturity is that they may not capture changes in the supply of Treasury debt in the very long end of the term st ructure (where we argue corporate issuers are likely to have the greatest comparative advantage in exploiting arbitrage opportunities created by changes in the supply of Treasury securities). More important, as discussed later, using more granular data to measure the maturity composition allows us to address concerns that broad measures may be correlated with other factors , such as credit spreads , that may affect corporate maturity choice. To illustrate the concern with using the proportion of Treasury debt maturing in over one year as a proxy for changes in the supply of long term Treasury debt we calculated monthly series of the proportion of total outstanding Treasury debt in various 13 As shown in Figure 2 1 through 12 In untabulated results we show that while s wap spreads contracted for 5 and 10 year Treasuries between 2006 and 2009, they were consistently positive except in the very long end of the market. 13 CRSP reports monthly pricing information, the remaining outstanding principal amounts, and the remaining time to maturity for most outstanding marketable treasury securities over our sample period. Since CRSP provides outstanding principal amounts that have been adjusted for repurchases and follow on offerings of existing securities by the Treasury, we are a ble to create monthly series of the fraction of outstanding principal in various maturity buckets. For a small number of observations during our sample period, the outstanding principal amounts are missing in the CRSP database. Therefore, where

PAGE 30

30 Figure 2 3 , the fraction of Treasury debt over one year (TSY1) as well as fraction of Treasury debt over five years (TSY5) exhibit a very different pattern over time than both the weighted average maturity of Treasury debt or the fraction of Treasury debt over 20 years (the segment of the term structure frequently associated with institutional demand by insuranc e companies and pension plans) . 14 Indeed, over the 2001 to 2006 time period (which coincides with the moratorium of Treasury bond issue s over 20 years), the fraction of total Treasury debt over one year increased while the average maturity of Treasury debt (as well as the fraction of Treasury debt over 20 years) declined substantially. In contrast, as shown Figure 2 3 and Figure 2 4 , the share of Treasury debt over 20 years (TSY20) tracks both the weighted average of maturity of Treasury debt (TSYMAT) as well as the total Treasury debt to GDP (TSYGDP) very closely. To isolate changes in the supply of Treasury debt in the very long term seg ment of the term structure, we focus our analysis on changes in the supply of Treasury debt greater than 20 years. However given concerns with endogeneity we also instrument the fraction of long term Treasury debt with the total Treasury debt to GDP and ex ploit a natural experiment involving the suspension of 30 year Treasury bond issues. Data We examine these issues using a database of corporate borrowing by public U.S. companies over the period from 1987 through 2009, which we describe in detail in necessary w e follow Greenwood and Vayanos (2008) and replace missing outstanding principal amounts with the amounts observed in the previous month. 14 Pension fund demand for long term assets has been cited as one factor contributing to relatively low long term intere st rates in the U.S. and Europe since 2001. See Ahrend, Catte and Price (2006). See also Greenwood and Vanyanos (2010) for evidence concerning UK pension fund demand for long term assets.

PAGE 31

31 the at tached Appendix . 15 We collect information on individual corporate debt issues from Thompson Reuters LPC Dealscan and Thomson Reuters SDC Platinum databases. From Dealscan we obtain information on corporate loans originated by bank and certain non bank lende rs, whereas from SDC we obtain information on new debt issues of non convertible debt securities, debt shelf registrations, U.S. Rule 144A non convertible debt, and medium term note programs. Using a combination of data fields available from SDC as well as text searches on the issue description, we exclude from our analysis asset or mortgage backed debt, secured debt, pass through securities, equipment trust certificates, lease obligations, convertible debt, preferred stock that has been misclassified as d ebt, equity linked certificates, and perpetual debt. We excluded these issues because the maturity of issue is likely to match the economic life of the underlying collateral (for example the cash flows of the underlying assets determine the life of the pas s through securities). We limit our sample to U.S. dollar denominated debt issues and bank borrowing. loans is limited prior to 1987 (see Chava and Roberts (2008) for a d iscussion). We focus on U.S. companies because of homogeneity with respect to tax laws, and because we would expect U.S. companies to be more susceptible to changes in the structure than foreign issuers. 16 Furthermore, follow ing prior literature on capital structure and maturity choice we exclude financial firms (SIC codes 6000 6999). We exclude financial firms both to align our analysis with previous 15 Table A 1 in the appendix contains the full list and description of all variables used in the empirical section. 16 For example, US tax law limits the tax deductibility of interest on perpetual bonds and for bonds in which the maturity date exceeds forty years if payments are contingent on earnings.

PAGE 32

32 studies and because regulatory restrictions and capital requirements create incentives for asset and liability matching among financial firms. 17 However, our results are robust to including financial firms in the sample. We account for inflation by adjusting all dollar onsumer price index (all urban consumers), unless specifically mentioned otherwise. We supplement our issuance database with firm level financial data from Compustat for the fiscal year end immediately prior to the date of debt issuance, which we describe in detail in the A ppendix , as well as with monthly data on the supply of long term Treasury securities from CRSP and monthly data on credit market conditions from the F 18 Our primary measure for the supply of long term Treasu ry securities is the fraction of outstanding Treasury debt maturing in over 20 years (TSY20), whereas our primary measures of credit market conditions are year BBB and AAA rated corporate bond indi ces and the term structure (or term premium) measured by the spread between the percentage yields of 10 year and 6 month Treasury securities. The advantage of using the BBB a measure of the credit risk premiu m in the long end of the term structure and is available on a monthly basis for our entire sample period. The disadvantage of this measure is that AAA bond yields and thus the BBB AAA yield spread may reflect changes in the supply of long term treasuries, or more troubling, affect the 17 Unlike credit risk, under current bank capital regulation there is no formulaic relationship between interest rate risk and minimum capital requirements. However, interest rate risk is one factor used to ss the adequacy of bank capital. 18 We merged Dealscan with Compustat using a link file program used by Chava and Roberts (2008). We updated their program for our sample period.

PAGE 33

33 Jorgensen (201 2 ) find that the BBB AAA spread is negatively related to the ratio of Treasury debt to GDP (their proxy for the supply of safe long term assets). Co nsistent with Krishnamurthy and Vissing Jorgensen (201 2 ), we find a negative and statistically significant relationship between the BBB AAA spread and the fraction of Treasury debt maturing in 20 years or more (TSY20) (the correlation is .422). However, w e also find a negative correlation between credit spreads and the fraction of short term Treasury debt ( .589). Nevertheless, to address this concern, we also measure credit risk premiums by the Corporate Industrial Bond Indices for BB and BBB rated bonds with a maturity of 10 years. We obtain this measure on a monthly basis through Bloomberg from February 1996 to April 2007. In contrast to the BBB AAA spread results we find a positive relationship between the BB BBB spread and the fraction of Treasury debt maturing in 20 years or more. Additionally, when instrumenting TSY20 with the ratio of total Treasury debt to GDP there appears to be no statistically significant relationship between TSY20 and the BB BBB spread, thus mitigating concerns that the BB BBB spread captures changes in the supply of long term Treasuries. Empirical Evidence Descriptive Statistics Our sample consists of 29,110 corporate loans from Dealscan and 10,040 debt issu es from SDC by 5,041 individual companies from 1987 to 2009. Table 2 2 provides descriptive statistics concerning the deal characteristics and financial characteristics of the issuing firms in our sample for issues in various maturity buckets. The maturity buckets are similar to those used by Guedes and Opler (1996). Notice that that the

PAGE 34

34 proportion of offerings that consist of bank loans declines with maturity. For example, virtually all borrowing under one year is from financial institutions, with the prop ortion declining to just over 3% for issues with maturities of 20 years or more. One reason why the proportion of intermediated debt is so large for offerings under one year is that we do not have information on commercial paper issued by firms in our samp le. However, even in the one to five year bucket the percentage of intermediated debt is greater than 90 % . Finally, notice that just over half the issues with maturities greater than five years are callable. Turning to issuer characteristics, as shown in T able 2 2 , with the exception of issues that are under a year, issuer size (measured by assets or market value of equity), age and the weighted average maturity of issuer assets all increase with the term to maturity of the issue. 19 For firms with issues of between one and ten years issuer leverage is generally higher, whereas the frequency of dividend payments as well as the fractions of issuers with an investment grade rating are lower than for issues with under one year or over 20 years in maturity. Consis tent with agency and information problems limiting very long term issues, 92 % of 20+ year issues were made by firms with investment grade ratings and 89 % of issues in this maturity bucket were made by dividend paying firms. Overall these results suggest th at better quality issuers inhabit the very short and very long term portion of the maturity spectrum. 19 We measure asset maturity using the same methodology as Stohs and Mauer (1996 ). Specifically, we define the value weighted average asset maturity as the book value weighted average maturity of current assets and long term assets. The maturity of current assets is measured as current assets divided by costs of goods sold and the mat urity of long term assets is measured as net property, plant and equipment divided by depreciation expense. Alternatively, we have also excluded cash from current assets to measure the maturity of current assets. Our main results remain robust to this alte rnative specification.

PAGE 35

35 If agency and information costs limit maturity choice, we would expect higher grade issuers to distribute their offerings more widely across the maturity spectrum than unrated firms or firms with more speculative grade ratings. To further examine this issue we grouped debt issues results are reported in Table 2 3 and Table 2 4 . Table 2 3 displa ys the results for the full sample of issues while Table 2 4 excludes revolving credit facilities from the sample . As shown, issuers with credit ratings lower than investment grade (BB or below) and nonrated issuers tend to cluster their issues in the midd le part of the term structure. Notice that the very long term issues (20 years +) are almost the exclusive providence of investment grade issuers. Only 49 out of a total of 1620 long term issues were made by firms with credit ratings below BBB. Moreover, a s shown in Table 2 4 , 20+ bond issues make up about 18 % of the non revolving debt of these issuers. As discussed in the next section, we find evidence of a corporate supply response with respect to 20+ year bond issues. Based on these findings we divided f irms in our sample into two groups : short term issuers, i.e. firms that only issue debt with maturity at issue of less than 20 years, and long term issuers, i.e. firms that have at least one issue with a maturity of more than 20 years. 20 We find that only a bout 12% of the debt issued by these firms is long term. Indeed, long term issuers spread their issuance across the maturity spectrum, issuing more frequently in the under one year maturity bucket than short term issuers. Long term issuers are larger, less levered, have higher ROA (EBIT/Assets) and are more likely to have an investment grade rating 20 Our findings concerning long term issuers are not sensitive to whether we define long term issuers based on issuance activity during the entire sample period or whether we limit the sample of firms to those that issue within prior five years.

PAGE 36

36 than short term issuers. Overall the evidence suggests long term issuers have greater flexibility in their choice of maturity than short term issuers. Empirical Models of Debt Maturity Choice As discussed earlier, we examine three questions related to gap filling. The first is whether long term debt issues are more sensitive to market conditions , and in particular the supply of long term Treasury securities , than short and intermediate term debt issues. The second is whether the maturity choice for high quality issuers is more sensitive to changes in the supply of Treasury securities than for lower rated issuers. The third is whether changes in the supply of Treasu ries affects the propensity of firms to borrow. We begin the analysis by estimating a linear model of maturity choice similar to the ones in Barclay and Smith (1995) and Guedes and Opler (1996) but include controls for credit risk spreads, term structure s preads (or term premia) and the maturity composition of Treasury debt. Unlike these previous studies we use panel data on monthly debt issues and bank borrowing to better identify the impact of credit market conditions on the maturity choice. While estimat ing a linear model of maturity choice facilitates a comparison of our results to earlier studies, a concern with a linear model is that, given higher costs of arbitrage in the long end of the market, the relationship between long term borrowing and the mat urity composition of Tr easury debt is likely to be non linear. To address this concern, we estimate a multinomial logit model of maturity choice in various maturity segments. In addition to examining whether shifts in the supply of Treasuries are related to the propensity of issuing long and short term debt, we estimate unconditional models of debt issuance.

PAGE 37

37 Linear Models of Debt Maturity Choice In the first set of regressions we examine the relationship between term to maturity and credit market conditions controlling for firm and industry characteristics as we ll as firm fixed effects. Table 2 5 present estimates of the linear model using four different measures of the maturity composition of Treasury debt. To control for potential shifts in the demand for l ong term safe assets and other macro factors we include indicator variables corresponding to five year intervals in our sample. 21 As shown, we find no significant relationship between the maturity of corporate debt issues and TSY1. However, consistent with gap filling we find a negative and significant relationship between the maturity of corporate issues and both the average maturity of Treasury debt (TSYMAT) and the fraction of Treasury debt with maturities of 20+ years (TSY20). Changes in the supply of lo ng term Treasuries have an economically significant effect on the average maturity of corporate issues. The coefficient estimate in Column (3) implies that a 1% decrease in TSY20 is associated with a 0.45 year increase in the average maturity of corporate debt issues. 22 Before discussing alternative regression specifications, it is interesting to note that the sign and significance on controls for issuer characteristics are generally consistent with contracting cost explanations of maturity choice (see Guede s and Opler (1996)). In particular, we find that the coefficient estimates for firm size (Log(MV of Equity)), deal size (Log(Deal Amount)) profitability (EBIT to Total Assets) and credit 21 The time dummies correspond to 1991 1995, 1996 2000, 2001 2005, 2006 2009 with the omitted time period being 1987 1990. 22 Because we estimate the model with 5 year fixed effects, the coefficient estimates should be interpreted in terms of var iation in TSY20 within 5 year intervals .The standard deviations of the monthly TSY20 within 5 year intervals ranged from 0.003 to 0.021 over our sample period.

PAGE 38

38 quality (IG Rating Dummy) are all positive and statistically signific ant. Because we estimate the regression including firm fixed effects, this implies that firms increase the maturity of their debt issues in periods when they are larger, raise more debt, are more profitable, and when they have an investment grade credit ra ting. Notice however, that we find no significant relationship between the maturities of debt issued and the estimating the regression with accounting for firm fixed ef fects. If we exclude the firm fixed effects we find a positive and significant relationship between debt maturity and As shown in Table 2 5 debt maturities are negatively related to credit risk spreads. On e concern with using the BBB AAA spread as a control variable in the maturity choice regression is that the composition of Treasury debt may affect AAA yields and thus the BBB AAA spread. As a result, the BBB AAA spread may also reflect shifts in the suppl y of long term Treasury bonds. To address this concern , in untabulated results we estimate the maturity choice regression using the percentage spread between BB and BBB bond yields obtained from Bloomberg. We assume that lower rated bonds are poorer substi tutes for Treasury bonds and thus a better control to identify the effect of changes in the supply of Treasuries on corporate maturity choice. However, as discussed above a drawback of the BB BBB spread measure is that it is only available for a portion of our sample period (1996 2007). As in Table 2 5 , we find a negative and significant relationship between the maturity of debt issues and TSYMAT and TSY20. Furthermore, the coefficient estimates on TSY20 remain negative and significant even when instrumenti ng it with the ratio of Treasury debt to GDP.

PAGE 39

39 If the negative relationship between the maturity of corporate debt issues and the maturity of Treasury debt is the result of gap filling we would expect the supply response to be greater for issues that are cl oser substitutes for Treasury debt. We therefore expect that the supply elasticity to be greater for firms rated A or better than for firms with BBB (recall virtually all 20+ debt is issued by investment rated firms). Also, since bank and institutional ter m loans are typically floating rate, they are not likely to be not close substitutes for fixed rate Treasury debt. Thus, we would expect a greater supply response if we exclude term loans from the sample. We also expect non callable bonds to be closer subs titutes for non callable Treasury bonds than callable corporate debt. Table 2 6 provides estimates of the linear model of maturity choice estimated over different subsamples of our data. For purposes of comparison we focus on subsamples of firms with rated debt. For brevity we only report the coefficient estimates for the credit market variables and only using TSY20 as a measure of shifts in Treasury supply (we obtain similar results using TSYMAT). Column (1) presents estimates of the linear model estimated for all firms with rated debt. In Column (2) we control for whether an issue is a bond and find a significantly greater supply response for bonds over bank debt (the difference in the coefficient estimates on TSY20 are significant at the .06 level). Addit ionally, we find a significantly greater supply response, in terms of magnitude, when we estimate the regression using samples that exclude term loans (no bank) and control for callable debt as shown in Column (4) (the difference in the estimates is signif icant at the .01 level). In Columns (6) and (7) we divide the sample further into firms with debt rated A AAA and firms with BBB rated debt (recall virtually all

PAGE 40

40 long term debt issues are done by investment grade firms). As shown, we find the supply respon se for A AAA rated firms is over twice as large as the response of BBB firms (the difference in coefficient estimates is significant at the .02 level). Overall, consistent with the gap filling hypothesis we find the coefficient estimates on TSY20 are more negative when estimated over samples that contain debt issues that are closer substitutes for Treasury bonds. How important are gap filling and credit market conditions generally in explaining maturity choice? We address this question using an analysis of covariance (ANCOVA) to decompose the variation in maturity choice over time attributable to various factors. For this analysis we use same methodology as Lem m on, Roberts, and Zender (2008) to compute the fraction of the total Type III partial sum of square s associated with each factor in the linear maturi ty choice models shown in Table 2 5. In other words, we (firm, year, TSY20 etc.) by dividing the partial sum of squar es for each effect by the aggregated partial sum of squares across all effects in the model. Our findings can be summarized as follows: For all of the models firm fixed effects account for most of the explained variation in maturity choice. In terms of cre dit market factors, TSY20 explains the greatest amount of within firm variability in maturity (about 7 % ) for highly rated issuers. When we estimate model without firm fixed effects, TSY20 still explains most of the variation for highly rated issuers (about 13 % ) after industry fixed effects. Multinomial Logit Models of Debt Maturity Choice What segment of the term structure is gap filling most prevalent in? While the above linear models are helpful in establishing the mean relationship between corporate and government debt maturity there is no reason to assume that this relationship is

PAGE 41

41 linear, or that gap filling occurs across the entire maturity spectrum. Moreover, as discussed earlier, given market segmentation we would expect corporate gap filling to be mo st prevalent in the very long end of the term structure where the demand for safe long term assets is relatively constant and where arbitrage costs are high (see Krishnamurthy and Vissing Jorgensen (201 2 ) and Greenwood and Vayanos (2010)). To investigate t his issue we estimate a multinomial logit model, relating the likelihood of debt issuances in various segments of the term structure to market conditions and the supply of Treasury bonds. The advantage of the multinomial model is that it allows us to accou nt for non linear effects of both firm characteristics and gap filling on maturity choice. The disadvantage is that it provides an estimate of the supply response conditional on the issuance of debt. However, as we discuss a shift in Treasury supply may not be limited to substituting away from short term debt to long term debt. In estimating the multinomial model, we model the decision to issue short term debt with maturities in (0,1) years, short term to medium term debt with maturities in [1,5) years, medi um term debt with maturities in [5,10) years, medium to long term debt with maturities in [10,20) years, and very long term debt with maturities in [20,...) years. 23 To estimate a multinomial logit models one needs to select a base or reference category to which the coefficient estimates relate. While theoretically it does not matter which of the alternative maturity ch oices we use, we selected the 5 to 10 year segment since this is the segment in which there is the most issuance activity and the segment 23 We also estimate a set of logit models. For the logit models we divide the term structure int o intervals reflecting short term maturities in (0,1] and long term maturities in [20,...) years. The results of the logit estimates are similar to the multinomial logit regressions.

PAGE 42

42 tha t most closely matches the average asset maturity for the firms in our sample. Interpretation of coefficient estimates is in relation to the chosen base category. 24 Table 2 7 contains the estimates of the multinomial logit model. The specification is simila r to the one used by Guedes and Opler (1996). Again, for brevity we report estimates using TSY20 to measure the supply of long term Treasury bonds, but our results are similar if we use TSYMAT to measure changes in the supply of long term Treasuries. As sh own, we find that changes in the supply of long term Treasuries affect maturity choice in a nonlinear fashion. Note that the coefficient estimate for TSY20 is negative and statistically significant only for maturities in [20,...) years relative to the base category. This indicates that the likelihood of firms issuing long term deb t relative to medium term debt decreases as the fraction of long term Treasury debt increases. However, as the estimated coefficients only allow us to make statements with relation to the assumed base case we also estimate a series of marginal effects in Table 2 8 , estimated at the mean firm characteristics for A AAA rated firms. The marginal effects estimated with respect to the fraction of long term government debt are negative an d highly statistically significant for the likelihood of issuing debt with a maturity greater than 20 years. The marginal effect estimates indicate the economic magnitude of gap filling is significant. For a 1% decrease in the fraction of 20 year treasurie s we find a 2.6 % increase in the likelihood of 20 year bon d issues. 25 The estimate d marginal effects term 24 See for instance Cameron and Trivedi (2005) for more on multinomial mod els. 25 Because we estimate the model with 5 year fixed effects, the variation in TSY20 should be interpreted as variation within the corresponding 5 year periods. The standard deviations of the monthly TSY20 within 5 year intervals ranged from 0.003 to 0.0 21 over our sample period.

PAGE 43

43 borrowing leads to highly rated firms substituting away from long te rm bonds in to short and medium term borrowing. In untabulated results w e also estimate the multinomial logit model separately for A AAA rated firms and BBB rated firms, as highly rated firm are expected to be more actively engaged in gap filling . Similar t o the results reported in Table 2 7 and Table 2 8 , the estimated marginal effect for the fraction of long term government debt is negative and statistically significant for the likelihood of issuing 20+ years corporate debt for A AAA rated issuers. Moreover, consistent with our prior results we find that the economic magnitude of the effect of TSY20 on the likelihood if issuing 20+ years corporate debt is over two times greater for A AAA rated issuers than for BBB rated issuers. Gap Filling and the Propensity to b orrow long t erm Our analysis thus far has focused on the relationship between maturity choice and changes in the relative supply of long term Treasury debt. Specifically, we have focused on maturity choice conditional on issuing debt. However, gap fil ling is not necessarily limited to substitution effects between corporate short term and long term debt with respect to the supply of long term Treasuries. Gap filling incentives may also r one of primary term corporate borrowing through purchases of long term Treasury bonds. To address this issue, we obtain the full time series of annual Compustat data from 1 986 to 2009 for all rated firms in our original sample of debt issuances. We then estimate a series of logit models where the dependent variable takes on a value of one if the fiscal year in Compustat was one in which long term (20+ years) debt was issued by the corresponding firm (as identified in our original dataset) and zero otherwise, as

PAGE 44

44 well as logit models where the dependent variable takes on a value of one if short term (one year or less) debt was issued and zero otherwise. The independent variable s for our logit models are the same as for our linear models in Table 2 5 but are lagged by one period. We also include 12 month averages of the credit market variables and, depending on the model, either TSY1 or TSY20. Because the Compustat data are annua l we do not include year fixed effects, but use indicator variables for 5 year time periods in order to control for changes in other unobserved macro variables. This specification allows us examine whether changes in Treasury supply predicts corporate debt issues and if so in what segment of the term structure. Table 2 9 provide s the logit regression estimates of the relationship between corporate debt issues and changes in the relative supply of short and long term Treasury debt and Table 2 10 displays the corresponding marginal effects . As shown in Column (1) of Table 2 9 we find no significant relationship between short term debt issues and changes in the supply of short term Treasury debt. In contrast, and consistent with our findings concerning maturity choice, we find a negative and significant relationship between long term debt issues and the fraction of long term Treasury debt outstanding in Column (2) of Table 2 9 . Specifically, as shown in Column (2) of Table 2 10 the marginal effect estimated for changes in TSY20 implies that a 1% decrease in TSY20 increases the likelihood of issuing long term debt by about 0.6 % for highly rated firms. Moreover, as can be seen in Columns (3) and (4) of Table 2 9 and Table 2 10 , when we include interaction terms for highly rated issuers or restrict our sample to rated firms that issued long term debt at any point in time from our original issuance sample, the marginal effects of a change in long term Treasuries outstanding

PAGE 45

45 on the likelihood of issuing long term debt increase both in significance and magnitude. For instance, a 1% decrease in TSY20 (within a five year interval) increases the likelihood of issuing long term debt by about 1.4 % for highly rated long term issuers. The macro control variables are included in the logit model to mitigate concerns that TSY20 may pick up the effects of shifts in the terms structure , credit risk spreads or changes in overall economic activity on the decision to issue long term debt. The coefficient estimates on the firm specific c ontrol variables generally have the expected sign. For example, larger firms (as measured by the market value of equity) more profitable firms and firms with longer term assets are more likely to issue long term debt. Firms with stronger growth options, as measure by the market to book ratio are less likely to issue long argument that long term debt exacerbates under investment problems for firms with significant growth options. Overall these results are consistent with agency cost theories of corporate debt maturity. To further examine the impact of shifts in Treasury supply on borrowing, we estimate a linear model with firm fixed effects of the annual changes in total debt outstanding for the rated f irms in our sample. The dependent variable in these models is the annual change in total debt outstanding scaled by t he lagged book value of assets. 26 We include the same macro control variables as in the previous regressions to mitigate concern that TSY20 is picking up the effects of changes in term structure or credit risk premiums. In addition, since we are examining the change in total debt outstand ing we include controls for firm profitability, leverage, investment spending, firm size, and asset 26 Specifically we define total debt outstanding as the sum of the Compustat variables DLTT (Total Long Term Debt) and DLC (Debt in Current Liabilities)

PAGE 46

46 tangibility in the regression. 27 Estimates of these models are reported in Table 2 11 . As shown in Columns (2) and (3), consistent with the results from the logit models we find a significant relationship between changes in corporate borrowing and the relative supply of long term Treasury bonds when firms are highly rated and when they have issued long term debt at any point in our sample. Indeed, as shown we find no significant relationship between corporate borrowing and the supply of long term Treasury bonds when firms are rated BBB or less. Finally, the estimated sign on the firm specific control variables are generally as expected. In particular, we find a positive relationship between changes in total debt and both investment spending (measured by net capital expenditures relative to total assets) and firm profitability. In addition, we find a negative relationship between changes in total debt and firm overall leverage. Endogeneity Concerns Event Study Although our data allow us to focus our tests with respect to gap filling on narrow segments of the term structure and on issuers with different credit ratings, concerns regarding endogeneity might remain. In this section we address these concerns by year bonds in 2001. Based on news articles and press statements by the U.S. Treasury around the announcement of the suspensio n, it does not appear that the suspension term corporate and 27 These are controls that are typically included in empirical studies of debt issuance. See for example Leary and Roberts (2005). Since we examine changes in total debt rather than maturity choice, we do not include average asset maturity as a control variab le. Including average asset maturity as a control variable yields results similar to those reported in Table 2 11 .

PAGE 47

47 government debt. Rather, it appears that the foremost reason for the suspension was to reduce long term interest rates. 28 To examine the impact of the suspension on Treasury yields we conducted an event study on daily relative changes in yields of Treasury securities using a constant mean return model (see Chapter 4 of Campbell, Lo, and MacK inlay (1997)). If the suspension of 30 year treasury bonds constituted an unexpected supply shock then we would expect to find a statistically significant decrease in yield relatives. For our event study we obtain data on daily yields for constant maturity Treasury period from 10/29/2001 to 11/02/2001, the week in which the Treasury announced the suspension, and the estimation window as the one year period prior to the event fr om 10/31/2000 to 10/26/2001. We denote the daily relative change in yield to maturity for a treasury security: (2 1) We then estimate the following constant mean return model over the estimation window using OLS (2 2) And proceed by calculating the daily abnormal relative changes in yie lds as the sample residuals over the event window. 28 In the U.S. Treasury press release from 10/31/2001 on the November Quarterly refunding, Peter Fisher, the undersecretary of the Treasury, a rgued that maintaining the issuance levels of 30 year bonds would be unnecessary and expensive to taxpayers. See also coverage by CNN Money on 10/31/2001 and by the November 2001 issue of The Economist. As discussed in the Economist article, critics of th e of an economic downturn, it was the wrong time to suspend long term Treasury issues.

PAGE 48

48 Table 2 12 summarizes the results of the event study. The cumulative abnormal returns (CAR) for 10, 20 and 30 year Treasury bonds are reported in C olumns ( 1 ) , ( 2 ) and ( 3 ) respectively. Consistent with a supply shock we find a positive and statistically significant a bnormal return on 30 year Treasuries. Consistent with market segmentation the decline in yi elds is significant only for 30 year Treasury bonds. Specifically, as shown the weekly retu rn for the 30 year bond is 5.61% which is significantly different from zero at the 1% level. In contrast the return on 20 year Treasury bonds is 2.17 % which is not significant at the 10 % level. Corporate Supply Response We next examine the corporate supply response to the suspension of 30 year Treasuries by examining changes in the fraction of corporate long term issuers with original maturity of 30 years or more. We calculate the annual fraction of total debt issued with maturities in [30,...) years to total debt issued with maturities in [20,...) years for each year in our sample for debt issues rated between A and AAA a nd for issues rated BBB. Figure 2 5 illustrates the mean fractions for the four years prior to the year bonds as well as the mean fractions for the four years following it. As expected , the increase in the proportion of issues exceeding 30 years relative to issues exceeding 20 years is far greater for A AAA rated bonds than for BBB rated bonds. Specifically, the increase for issues rated A AAA wa s about 20 % in absolute terms while for issues rated BBB it was approximately 5 % in absolute terms. To examine the corporate supply response to the suspension of 30 year Treasury bond issues we conduct a simple regression discontinuity analysis on the quar terly fraction of total debt issued with maturities in [30,...) years to total debt issued with maturities in [20,...). The results of this analysis are reported in Table 2 13 through

PAGE 49

49 Table 2 15 . 29 Consistent with gap filling we find positive and statistica lly signi ficant coefficients on the post event dummy variable for the sample consisting of highly rated debt issues (Table 2 14) , but not on the sample consisting of BBB rated issues (Table 2 15) . Additionally, the coe fficient estimates for the post event du mmy correspond to an increase of between 25 % and 29 % (in absolute terms) in the pr oportion of issues exceeding 30 years relative to issues exceeding 20 years, which is in line with the estimates from Figure 2 5 . Overall , consistent with gap filling we find that the elimination of 30 year Treasury bonds is associated with an increase in the proportion of long term bonds with maturities of 30 year or more. Chapter 2 Concluding Remarks In this paper we examine the determinants of long term debt issues and spec ifically how variation in the supply of long term Treasury bonds affect s the issuance of long term corporate bonds. Overall we find that debt issues of 20 years or more are common for highly rated firms. Consistent with gap filling, we find that highly rat issuance of long term bonds is inversely related to the proportion of outstanding Treasury bonds with maturities of 20 years or more. Moreover, we find very little evidence that issuances of shorter maturity bonds are related to changes in the su pply of short or long term Treasury bonds. Our evidence is consistent with recent work by Krishnamurthy and Vissing Jorgensen (201 2 ) that suggests a premium associated with relatively safe long term bonds that varies with the supply of highly rated long te rm 29 Recall that the BBB AAA spread is calculated for 30 year bonds, and the term premium as the yield difference between 10 year and 6 month Treasuries. In simple univariate regressions where the dependent variable is the quarterly fraction of +30 to +20 year debt we find no statistically significant coefficients on the B BB AAA spread and term premium variables.

PAGE 50

50 bonds. Our results suggest that shifts in the supply of long term Treasuries impact corporate bond issues in a relatively narrow segment of the term structure. Overall our results suggest that market conditions and particularly the supply of long term T reasuries are important determinants of high grade issuers timing of long term debt issues. Moreover, consistent with agency theories of maturity choice we find that very long term unsecured debt issues are a choice available principally to A rated or bett er firms. Finally, we find that changes in the supply of long term Treasuries are negatively related to the propensity of high grade firms to borrow.

PAGE 51

51 Table 2 1. Annual implied price volatilities . 20 Year 10 Year 5 Year 2 Year 1994 10.42 7.45 4.95 1.94 1995 9.41 6.65 4.59 2.17 1996 9.98 6.75 4.72 2.03 1997 8.83 5.48 3.73 1.68 1998 8.66 5.71 4.09 2.06 1999 8.82 6.05 4.18 3.05 2000 9.32 6.52 4.13 1.60 2001 9.46 7.10 4.76 2.19 2002 10.98 7.93 5.47 2.15 2003 12.04 8.04 5.38 2.02 2004 10.01 6.92 4.70 1.87 2005 8.08 5.10 3.41 1.42 2006 6.79 4.25 2.86 1.26 2007 7.32 5.10 3.74 1.78 2008 13.50 9.22 6.68 3.00 2009 15.60 8.95 5.46 1.76 Overall Average 9.97 6.71 4.56 1.97 Number of Observations 3969 This table presents the annual means of daily implied volatilities associated with the corresponding futures on U.S. Treasury securities. Daily implied volatilities for the underlying securities are calculated from a weighted average of the volatilities of the two closest call options. For all securities, the contract used is the closest pricing contract month that is expiring at least 20 business days from the corresponding observation date. 20 year Treasury bond futures refer to U.S. Treasury bonds that, if callable, are not callable for at least 15 years from the first day of the delivery month or, if not callable, have a maturity of at least 15 years from the first day of the delivery month. 10 year Treasury notes futures refer to futures on notes maturi ng at least 6 1/2 years, but not more than 10 years, from the first day of the delivery month. 5 year Treasury notes futures refer to U.S. T notes that have an original maturity of not more than 5 years and 3 months . and a remaining maturity of not less t han 4 y ea rs and 2 mo nths . as of the 1st day of the delivery month. 2 year Treasury Notes futures refer to futures on U.S. Treasury notes that have an original maturity of not more than 5 years and 3 months and a remaining maturity of not less than 1 year a nd 9 months from the first day of the delivery month but not more than 2 years from the last day of the delivery month.

PAGE 52

52 Table 2 2. Summary statistics by final maturity of the issue . (0,1) Years [1,5) Years [5,10) Years [10,20) Years [20,...) Yea rs Mean Median N Mean Median N Mean Median N Mean Median N Mean Median N Years to Final Maturity 0.87 1.00 5453 3.37 3.01 18051 6.62 6.35 10158 11.06 10.03 3868 31.23 30.01 1620 0.55 0.16 5453 0.27 0.09 18051 0.35 0.18 10158 0.32 0.23 3868 0.36 0.27 1620 Bank Debt Dummy 1.00 1.00 5453 0.91 1.00 18051 0.66 1.00 10158 0.14 0.00 3868 0.03 0.00 1620 SDC Call Flag 0.00 0.00 18 0.20 0.00 1692 0.55 1.00 3431 0.62 1.00 3334 0.56 1.00 1565 9.94 2.10 5448 6.08 0.67 18030 7.55 1.42 10151 13.49 4.06 3865 26.22 10.42 1620 MV Equity ($ 10.41 1.55 5323 5.30 0.56 17455 6.99 1.06 9773 13.39 3.29 3772 21.19 8.37 1601 Market 0.30 0.26 5323 0.32 0.27 17455 0.34 0.32 9773 0.32 0.30 3772 0.30 0.27 1601 0.28 0.18 5438 0.23 0.12 17952 0.17 0.08 10133 0.17 0.10 3855 0.18 0.13 1619 0.37 0.28 5442 0.34 0.24 17963 0.25 0.16 10135 0.23 0.17 3856 0.24 0.20 1619 0.44 0.37 5442 0.44 0.37 17963 0.34 0.26 10135 0.31 0.26 3856 0.30 0.28 1619 0.50 0.45 5442 0.51 0.48 17963 0.42 0.38 10135 0.38 0.35 3856 0.36 0.35 1619 0.56 0.55 5442 0.58 0.61 17963 0.51 0.50 10135 0.46 0.46 3856 0.43 0.44 1619 1.77 1.40 5323 1.64 1.33 17455 1.63 1.36 9773 1.66 1.40 3772 1.63 1.35 1601 Dividend Dummy 0.56 1.00 5448 0.41 0.00 18030 0.47 0.00 10151 0.68 1.00 3865 0.89 1.00 1620 IG Rating Dummy 0.47 0.00 5453 0.25 0.00 18051 0.30 0.00 10158 0.58 1.00 3868 0.92 1.00 1620 CP Rating Dummy 0.35 0.00 5453 0.18 0.00 18051 0.22 0.00 10158 0.44 0.00 3868 0.69 1.00 1620 5.71 3.41 5161 5.04 3.02 16951 5.89 4.00 9541 7.33 5.05 3588 8.41 6.28 1528 Firm Age 25.30 16.78 5365 19.48 12.05 17879 20.11 12.23 10058 27.06 21.62 3837 38.61 34.94 1614 0.06 0.08 5443 0.07 0.08 18005 0.09 0.09 10121 0.10 0.10 3859 0.11 0.10 1620 This table presents issue level summary statistics for the full sample split up by different maturity buckets. The sample consists of all U.S. Dollar denominated debt issues by public companies bet ween 1987 and 2009 and is obtained from SDC and Dealscan . In order to mitigate the influence of extreme outliers continuous variables have been winsorized at the 1st and 99th percentile (fractions between 0 and 1 ) , symbol. For illu strative purposes dollar amounts that are normally measured in millions throughout the paper have been displayed in billions where indicated by .

PAGE 53

53 Table 2 2. Continued . (0,1) Years [1,5) Years [5,10) Years [10,20) Years [20,...) Years Mean Median N Mean Median N Mean Median N Mean Median N Mean Median N Moodys BBB AAA 30Y 0.93 0.84 5453 0.88 0.82 18051 0.88 0.83 10158 0.90 0.83 3868 0.86 0.78 1620 Termstructure 10y 6mth 1.50 1.41 5453 1.44 1.27 18051 1.30 1.05 10158 1.41 1.16 3868 1.37 1.15 1620 TSY20 0.09 0.10 5453 0.09 0.10 18051 0.09 0.10 10158 0.09 0.10 3868 0.09 0.10 1620 Number of Firms 2067 4156 2861 1408 472 Number of LT Firms 341 412 427 404 472 Number of AA AAA Firms 68 98 93 81 68 Number of BBB A Firms 514 706 650 544 372 This table presents issue level summary statistics for the full sample split up by different maturity buckets. The sample consists of all U.S. Dollar denominated debt issues by public companies between 1987 and 2009 and is obtained from SDC and Dealscan . I n order to mitigate the influence of extreme outliers continuous variables have been winsorized at the 1st and 99th percentile (fractions between 0 and 1 ) symbol. For illustrative purposes dollar amounts that are normally measured in millions throughout the paper have been displayed in billions where indicated by .

PAGE 54

54 Table 2 3. M aturities by S&P issuer rating . (0,1) Years [1,5) Years [5,10) Years [10,20) Years [20,...) Years Total AAA Rated 75 69 96 78 48 366 AA Rated 252 397 298 274 210 1431 A Rated 1071 1749 1192 958 690 5660 BBB Rated 1088 2258 1405 909 522 6182 BB Rated 270 1942 1728 541 45 4526 B Rated 201 1168 1193 260 4 2826 CCC Rated 32 97 70 10 0 209 CC Rated 3 15 14 2 0 34 Default 15 39 23 0 0 77 Unrated 2446 10317 4139 836 101 17839 Total 5453 18051 10158 3868 1620 39150 This table presents the frequency distribution of individual debt issues by S& P issuer rating and issue maturity for the full sample of all U.S. Dollar denominated debt issues by public U.S. companies between 1987 and 2009 and is obtained from SDC and Dealscan .

PAGE 55

55 Table 2 4. Maturities by S&P issuer rating (excluding credit lines) . (0,1) Years [1,5) Years [5,10) Years [10,20) Years [20,...) Years Total AAA Rated 4 45 84 78 48 259 AA Rated 7 233 247 273 210 970 A Rated 84 856 933 957 688 3518 BBB Rated 123 715 1025 899 521 3283 BB Rated 65 506 1179 537 45 2332 B Rated 54 300 824 255 4 1437 CCC Rated 10 35 50 10 0 105 CC Rated 1 8 9 2 0 20 Default 4 18 16 0 0 38 Unrated 486 2433 2419 806 98 6242 Total 838 5149 6786 3817 1614 18204 This table presents the frequency distribution of individual debt issues by S& P issuer rating and issue maturity for the full sample of all U.S. Dollar denominated debt issues by public U.S. companies between 1987 and 2009 and is obtained from SDC and Dealscan . The sample is further restricted to exclude credit lines.

PAGE 56

56 Table 2 5. Linear models of debt maturity choice . (1) (2) (3) (4) FE FE FE IV TSY1 13.318 (1.54) TSYMAT 1.941 *** (6.10) TSY20 45.008 *** 154.793 *** (5.40) (10.25) Moodys BBB AAA 30Y 2.407 *** 3.264 *** 3.174 *** 4.070 *** (5.74) (8.48) (8.29) (10.70) Termstructure 10y 6mth 0.444 *** 0.402 *** 0.420 *** 0.425 *** (4.53) (4.27) (4.47) (4.55) Total GDP 4Q Growth 18.027 ** 32.329 *** 32.235 *** 65.794 *** (2.38) (3.82) (3.74) (7.64) Log(Deal Amount) 0.233 * 0.240 * 0.225 * 0.208 *** (1.71) (1.76) (1.66) (2.77) 0.311 ** 0.323 ** 0.298 * 0.276 * (2.02) (2.09) (1.92) (1.78) Market 1.751 *** 1.218 * 1.445 ** 0.606 (2.61) (1.82) (2.15) (0.80) 1.632 1.712 * 1.737 * 1.957 * (1.57) (1.67) (1.69) (1.69) 0.031 0.027 0.028 0.019 (0.73) (0.65) (0.67) (0.56) 0.184 0.010 0.043 0.283 * (1.05) (0.06) (0.25) (1.68) Dividend Dummy 0.317 0.319 0.319 0.349 (0.95) (0.97) (0.97) (0.96) IG Rating Dummy 1.792 *** 2.005 *** 1.969 *** 2.387 *** (4.24) (4.78) (4.67) (6.41) 2.813 *** 2.945 *** 3.168 *** 3.687 *** (2.64) (2.79) (2.97) (3.45) Constant 0.031 20.214 *** 14.542 *** 27.987 *** (0.01) (7.66) (7.50) (12.37) 5 Year FE Yes Yes Yes Yes Adj. R Square 0.123 0.126 0.125 0.095 Observations 16440 16440 16440 16440 This table presents linear models of debt maturity choice that account for firm fixed effects. The dependent variable is Years to Maturity. The sample consists of all U.S. Dollar denominated debt issues (excluding credit lines) by public companies between 1987 and 2009 and is obtained from SDC and Dealscan . All models are linear models and Model (4) is an instrumental variables model where TSY20 has been instrumented with total Treasury Debt to GDP. In order to mitigate the influence of extreme outliers con tinuous variables have been winsorized at the 1st and 99th percentile (fractions between 0 and 1 ) , symbol. Absolute values of t statistics are in parentheses below the corresponding coefficient estimates and standard errors have be en clustered by firm. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively .

PAGE 57

57 Table 2 6. Linear models of debt maturity choice (different subsample s) . (1) (2) (3) (4) (5) (6) (7) Rated Rated No Bank No Bank No Bank & No Call A AAA BBB TSY20 50.438 *** 46.871 *** 79.868 *** 46.360 *** 131.987 *** 122.491 *** 54.511 ** (4.65) (4.87) (5.41) (3.18) (5.03) (4.88) (2.43) TSY20 x Bonds 15.624 * (1.94) TSY20 x No Call 86.679 *** (6.96) Termstructure 10y 6mth 0.482 *** 0.584 *** 0.216 0.433 ** 0.550 * 0.477 ** 0.449 (3.98) (4.12) (1.40) (2.57) (1.72) (2.03) (1.51) Termstructure 10y 6mth x Bonds 0.406 *** (2.80) Termstructure 10y 6mth x No Call 0.222 (0.84) Moodys BBB AAA 30Y 3.595 *** 2.281 *** 3.761 *** 4.152 *** 7.513 *** 3.754 *** 5.684 *** (7.62) (4.63) (6.47) (7.54) (4.90) (3.84) (5.92) Moodys BBB AAA 30Y x Bonds 1.319 ** (2.18) Moodys BBB AAA 30Y x No Call 2.758 ** (2.51) 5 Year FE Yes Yes Yes Yes Yes Yes Yes Adj. R Square 0.088 0.111 0.030 0.067 0.056 0.011 0.035 Observations 10782 10782 7989 7989 3885 4287 2915 This table presents linear models of debt maturity choice that account for firm fixed effects. For brevity only the independent variables of interest are displayed. The dependent variable is Years to Maturity. The sample consists of all U.S. Dollar denomin ated debt issues (excluding credit lines) by public companies with a long term S&P credit rating, between 1987 and 2009 and is obtained from SDC and Dealscan . Columns (1) and (2) show the results for the rated sample. Columns (3) and (4) exclude bank loans from the sample, and Column (5) displays results for rated firms but excludes bank loans and callable debt. Column (6) shows results for a subsample of firms with a credit rating of at least 'A ', and Column (7) for firms with a credit rating of BBB. In o rder to mitigate the influence of extreme outliers continuous variables have been winsorized at the 1st and 99th percentile (fractions between 0 and 1 ) , symbol. Absolute values of t statistics are in parentheses below the correspon ding coefficient estimates and standard errors have been clustered by firm. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively .

PAGE 58

58 Table 2 7. Multinomial logit model of debt maturity choice . (1) (2) (3) (4) (5) (0,1) Years [1,5) Years [5,10) Years [10,20) Years TSY20 13.984 *** 0.523 0.000 0.646 16.611 *** (3.90) (0.29) (.) (0.32) (5.07) Moodys BBB AAA 30Y 0.774 *** 0.296 *** 0.000 0.094 1.102 *** (4.11) (2.93) (.) (0.90) (6.72) Termstructure 10y 6mth 0.099 ** 0.049 * 0.000 0.037 0.133 *** (1.97) (1.93) (.) (1.35) (3.46) Total GDP 4Q Growth 9.056 ** 2.489 0.000 3.829 * 12.428 *** (2.24) (1.20) (.) (1.76) (4.11) Log(Deal Amount) 0.112 *** 0.338 *** 0.000 0.007 0.066 ** (3.29) (21.28) (.) (0.40) (2.50) 0.341 *** 0.039 ** 0.000 0.105 *** 0.200 *** (9.34) (2.24) (.) (5.50) (7.07) Market 1.124 *** 0.038 0.000 0.573 *** 0.766 *** (5.43) (0.33) (.) (4.03) (2.84) 3.049 *** 1.667 *** 0.000 0.056 0.674 (9.92) (7.96) (.) (0.19) (1.14) 0.009 0.018 *** 0.000 0.025 *** 0.048 *** (1.22) (4.45) (.) (6.60) (8.71) 0.137 *** 0.008 0.000 0.050 0.179 *** (2.74) (0.27) (.) (1.51) (3.20) Dividend Dummy 0.081 0.190 *** 0.000 0.203 *** 0.392 *** (0.78) (3.52) (.) (3.45) (3.57) IG Rating Dummy 0.696 *** 0.329 *** 0.000 0.532 *** 2.288 *** (5.32) (4.94) (.) (7.86) (17.70) 0.545 * 0.339 ** 0.000 0.434 *** 0.132 (1.84) (2.19) (.) (2.60) (0.54) Constant 1.849 *** 0.704 ** 0.000 1.330 *** 1.652 *** (3.13) (2.36) (.) (4.05) (3.24) 5 Year FE Yes Yes Yes Yes Yes Pseudo R Square 0.104 Observations 16440 This table presents coefficient estimates of a multinomial logit model of debt maturity choice. The sample consists of all U.S. Dollar denominated debt issues (excluding credit lines) by public companies between 1987 and 2009 and is obtained from SDC and Deals can . The dependent variable takes on a value of 0 if the maturity of a debt issue is in (0,1) years, 1 if maturity in [1,5) years, 2 if maturity in [5,10) years, 3 if maturity in [10,20) years, and 4 if maturity in [20,...) years. The base category for the multinomial model is chosen to be maturities in [5,10) years . In order to mitigate the influence of extreme outliers continuous variables have been winsorized at the 1st and 99th percentile (fractions between 0 and 1 ) , symbol. Absolute values of t statistics are in parentheses below the corresponding coefficient estimates . ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively .

PAGE 59

59 Table 2 8. Marginal effects of multinomial logit model . (1) (2) (3) (4) (5) (0,1) Years [1,5) Years [5,10) Years [10,20) Years Moodys BBB AAA 30Y 0.019 *** 0.089 *** 0.047 *** 0.020 0.176 *** (4.98) (5.92) (2.62) (1.01) (7.58) [0.00] [0.02] [0.02] [0.02] [0.02] Termstructure 10y 6mth 0.003 ** 0.014 *** 0.007 0.004 0.020 *** (2.53) (3.77) (1.59) (0.70) (3.72) [0.00] [0.00] [0.00] [0.01] [0.01] TSY20 0.344 *** 0.674 ** 0.859 ** 0.692 * 2.569 *** (4.70) (2.42) (2.48) (1.74) (5.46) [0.07] [0.28] [0.35] [0.40] [0.47] Observations 4287 4287 4287 4287 4287 This table presents marginal effects estimates for the multinomial logit model in Table 2 7. The marginal effects are evaluated at the mean values for firms rates A AAA. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively .

PAGE 60

60 Table 2 9. Logit models of issuance . (1) (2) (3) (4) ST Rated LT Rated LT Rated LT Issuer L1 TSY1 Avg. 12m 11.306 (1.55) L1 TSY20 Avg. 12m 6.973 * 1.179 2.144 (1.80) (0.29) (0.51) L1 A Rating x TSY20 Avg. 12m 7.731 *** 8.023 *** (2.84) (2.87) L1 Termstructure 10y 6mth Avg. 12m 0.074 0.067 0.109 * 0.093 (0.66) (1.28) (1.66) (1.37) L1 A Rating x Termstructure 10y 6mth Avg. 12m 0.099 0.087 (1.42) (1.21) L1 Moodys BBB AAA30Y Avg. 12m 0.123 0.937 *** 0.938 *** 0.902 *** (0.23) (3.57) (3.03) (2.83) L1 A Rating x Moodys BBB AAA30Y Avg. 12m 0.221 0.267 (0.74) (0.87) L1 Total GDP 4Q Growth 3.053 11.187 *** 10.603 *** 9.643 *** (0.47) (3.34) (3.16) (2.82) L1 A Rating or Higher 0.946 ** 0.740 * (2.28) (1.74) L1 IG Rating Dummy 0.131 1.472 *** (0.69) (9.50) L1 Log(MV of 0.056 0.599 *** 0.673 *** 0.365 *** (0.95) (17.66) (20.26) (9.93) L1 Market 0.126 0.570 * 0.014 0.610 * (0.29) (1.79) (0.05) (1.84) 0.510 2.186 *** 2.195 *** 2.108 *** (0.69) (3.20) (3.30) (3.00) 0.010 0.032 *** 0.033 *** 0.019 ** (0.63) (4.00) (4.07) (2.29) 0.022 0.344 *** 0.446 *** 0.351 *** (0.24) (5.37) (6.87) (5.23) L1 Dividend Dummy 0.080 0.261 ** 0.632 *** 0.106 (0.45) (2.15) (5.35) (0.88) 0.683 0.595 0.549 0.355 (0.95) (1.48) (1.35) (0.85) Constant 11.269 ** 7.781 *** 8.123 *** 2.705 ** (2.28) (7.87) (8.14) (2.31) 5 Year FE & Industry FE Yes Yes Yes Yes Pseudo R Square 0.027 0.198 0.188 0.065 Observations 17765 18170 18170 7134 This table presents logit models for short term and long term debt issuance and the corresponding marginal effects. The sample is taken from Compustat for all rated US firms that issued non revolving debt as identified in our original sample. Capital structure variables and credit market variables are lagged by one year. Column (1) presents results where the dependent variable takes on a v alue of 1 if the year was one in which short term debt was issued , (0,1]Y , but not long term debt , [20,...)Y. For Columns (2) (4) the dependent variable takes on a value of 1 if the year was one in which long term debt was issued. For Column (4) the sample has been further reduced to firms that issued long term debt at any point in our sample . C ontinuous variables have been winsorized at the 1st and 99th percentile (fractions between 0 and 1 ) , where symbol. Absolute values of t statistics are in parentheses below the corresponding coefficient estimates . ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively .

PAGE 61

61 Table 2 10. Marginal effects of logit model . (1) (2) (3) (4) ST Rated LT Rated LT Rated LT Issuer L1 TSY1 Avg. 12m 0.139 (1.54) L1 TSY20 Avg. 12m 0.611 * 0.793 ** 1.368 ** (1.80) (2.15) (2.40) L1 Termstructure 10y 6mth Avg. 12m 0.001 0.006 0.001 0.001 (0.66) (1.28) (0.17) (0.10) L1 Moodys BBB AAA30Y Avg. 12m 0.002 0.082 *** 0.064 ** 0.086 ** (0.23) (3.59) (2.44) (2.12) Observations 4761 4879 4879 3548 This table presents marginal effects estimates for the logit model s estimated in Table 2 9 . The marginal effects are evaluated at the mean values for firms rates A AAA. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 62

62 Table 2 11. Annual change in total debt . (1) (2) (3) Rated Rated LT Issuer L1 TSY20 Avg. 12m 0.184 ** 0.106 0.123 (2.30) (1.16) (1.12) L1 A Rating x TSY20 Avg. 12m 1.069 *** 0.708 *** (7.75) (5.06) L1 Termstructure 10y 6mth Avg. 12m 0.011 *** 0.011 *** 0.011 *** (8.17) (6.50) (4.66) L1 A Rating x Termstructure 10y 6mth Avg. 12m 0.001 0.002 (0.66) (0.80) L1 Moodys BBB AAA30Y Avg. 12m 0.014 * 0.018 ** 0.019 (1.93) (2.01) (1.45) L1 A Rating x Moodys BBB AAA30Y Avg. 12m 0.015 0.024 * (1.33) (1.71) L1 Total GDP 4Q Growth 0.724 *** 0.699 *** 0.571 *** (8.09) (7.65) (5.34) L1 A Rating or Higher 0.109 *** 0.094 *** (5.96) (4.54) L1 IG Rating Dummy 0.003 (0.53) 0.041 *** 0.041 *** 0.019 *** (9.00) (9.06) (3.67) L1 0.109 *** 0.107 *** 0.134 *** (4.73) (4.62) (3.82) 0.002 0.011 0.045 (0.07) (0.42) (1.44) L1 Book 0.582 *** 0.590 *** 0.373 *** (24.15) (24.47) (11.28) L1 Dividend Dummy 0.031 *** 0.030 *** 0.008 (4.02) (3.95) (0.59) 0.094 *** 0.093 *** 0.052 *** (16.84) (17.06) (7.46) 0.445 *** 0.448 *** 0.464 *** (8.87) (8.93) (6.15) Constant 0.869 *** 0.838 *** 0.512 *** (16.99) (16.54) (6.54) Adj. R Square 0.241 0.244 0.138 Observations 19182 19182 7397 This table presents linear models, that account for firm fixed effects, where the dependent variable is the change in total debt over lagged total assets , (winsorized at the 1% and 99% level). The sample is annual data taken from Compustat for all rated US firms that iss ued non revolving debt as id entified in our original sample . Capital structure variables and credit market variables are lagg ed by one year. Columns (1) (3) present results for all firms with an S&P credit rating. Column (3) presents results for firms with an S&P credit rating that issued long term debt at any point in our sample period. A Rating is an indicator variable that ta kes on a value of 1 if the firm has an S&P credit rating of 'A ' or higher. Continuous variables have been winsorized at the 1st and symbol. Absolute values of t statistics are in parent heses below the corresponding coefficient estimates and standard errors have been clustered by firm . ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 63

63 Table 2 12. Event study relative changes in yields around Treasury annou ncement . (1) (2) (3) 10 Year 20 Year 30 Year Cummulative Abnormal Returns 0.030 0.022 0.056 T Statistic 1.103 1.067 2.997 *** P Value 0.271 0.287 0.003 This table presents the results of an event study on the relative changes in daily yields of various U.S. Treasury securities. The event study has been conducted following Chapter 4 of Campbell, Lo, and MacKinlay (1997) and estimates cum ulative abnormal returns for constant return models. The estimation window constitutes 247 business days and ranges from 10/31/2000 to10/26/2001. The event window constitutes 5 business days and ranges from 10/29/2001 to 11/02/2001. The constant return mod els are estimated for 10, 20, and 30 year constant maturity Treasury rates as indicated in the corresponding columns. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively .

PAGE 64

64 Table 2 13. Natural experiment all issues . (1) (2) (3) (4) Qtr. Mean TSY20 0.039 0.965 0.654 0.471 (0.02) (0.44) (0.28) (0.18) Qtr. Mean Spread Moodys BBB AAA 30Y 0.022 0.033 (0.38) (0.58) Qtr. Mean Termstructure 10y 6mth 0.017 0.018 (0.64) (0.67) Year [2002,...) Dummy 0.245 ** 0.235 ** 0.188 0.171 (2.22) (2.11) (1.32) (1.17) Time Trend 0.001 0.002 (0.69) (0.85) Constant 0.569 *** 0.427 0.603 ** 0.420 (3.14) (1.51) (2.40) (1.30) R Square 0.230 0.237 0.239 0.247 Observations 92 92 92 92 This table presents results of ordinary least squares regressions on the long term debt composition of new debt issues to estimate the effect of the U.S. Treasury's decision to suspend the issuance of long term bonds. The dependent variable is the quarterly to total as (issue quarter 187). The sample consists of all U.S. Dollar denominated debt issues (excluding credit lines) by public U.S. companies and is obtained from SDC and Dealscan . All models calculate heterosk edasticity robust standard errors. Absolute values of t statistics are presented in parentheses below the corresponding coefficient estimates. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively .

PAGE 65

65 Table 2 14. Natural experiment issues rated A AAA . (1) (2) (3) (4) Qtr. Mean TSY20 1.202 0.488 1.023 0.546 (0.73) (0.25) (0.41) (0.21) Qtr. Mean Spread Moodys BBB AAA 30Y 0.060 0.056 (0.98) (0.92) Qtr. Mean Termstructure 10y 6mth 0.009 0.008 (0.27) (0.25) Year [2002,...) Dummy 0.292 *** 0.300 *** 0.256 * 0.263 * (2.69) (2.71) (1.78) (1.77) Time Trend 0.001 0.001 (0.52) (0.35) Constant 0.490 *** 0.600 ** 0.447 * 0.525 * (2.75) (2.32) (1.70) (1.73) R Square 0.173 0.176 0.184 0.186 Observations 91 91 91 91 This table presents results of ordinary least squares regressions on the long term debt composition of new debt issues to estimate the effect of the U.S. Treasury's decision to suspend the issuance of long term bonds. The dependent variable is the quarterly as (issue quarter 187). The sample consists of all U.S. Dol lar denominated debt issues (excluding credit lines) with a rating between A and AAA by public U.S. companies and is obtained from SDC and Dealscan . All models calculate heterosk edasticity robust standard errors. Absolute values of t statistics are present ed in parentheses below the corresponding coefficient estimates. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively .

PAGE 66

66 Table 2 15. Natural experiment issues rated BBB . (1) (2) (3) (4) Qtr. Mean TSY20 5.516 3.911 6.274 4.714 (1.62) (1.01) (1.47) (1.02) Qtr. Mean Spread Moodys BBB AAA 30Y 0.041 0.030 (0.48) (0.35) Qtr. Mean Termstructure 10y 6mth 0.013 0.014 (0.36) (0.41) Year [2002,...) Dummy 0.126 0.143 0.160 0.184 (0.56) (0.65) (0.6)0 (0.70) Time Trend 0.003 0.003 (0.89) (0.87) Constant 1.191 *** 0.941 ** 1.289 *** 1.031 * (3.30) (2.00) (2.83) (1.90) R Square 0.135 0.145 0.138 0.147 Observations 84 84 84 84 This table presents results of ordinary least squares regressions on the long term debt composition of new debt issues to estimate the effect of the U.S. Treasury's decision to suspend the issuance of long term bonds. The dependent variable is the quarterly proceeds raised from new issu as (issue quarter 187). The sample consists of all U.S. Dollar denominated debt issues (excluding credit lines) with a rating of BBB by public U.S. companies and is obtained from SDC and Dealscan . All models calculate heteroskedasticity robust standard errors. Absolute values of t statistics are presented in parentheses below the corresponding coefficient estimates. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively .

PAGE 67

67 Figure 2 1. Monthly time series of TSY1 and TSYMAT. This figure presents the monthly time series of the fraction Treasury securities maturing in over one year (TSY1) and compares it to the value weighted average maturity of outstanding Tre asury debt (TSYMAT).

PAGE 68

68 Figure 2 2 . Monthly time series of TSY 5 and TSYMAT. This figure presents the monthly time series of the fraction Treasury securities maturing in over five year s (TSY 5 ) and compares it to the value weighted average maturity of outstanding Treasury debt (TSYMAT).

PAGE 69

69 Figure 2 3 . Monthly time series of TSY 20 and TSYMAT. This figure presents the monthly time series of the fraction Treasury securities maturing in over twenty year s (TSY 20 ) and compares it to the value weighted average maturity of outstanding Treasury debt (TSYMAT).

PAGE 70

70 Figure 2 4 . Monthly time series of TSY20 and TSYGDP . This figure presents the monthly time series of the fraction Treasury securities maturing in over twenty years (TSY20) and compares it to the fraction of total Treasury debt, as measured by outstanding principal, divided by GDP (TSY GDP ).

PAGE 71

71 Figure 2 5. Average composition of long term debt issues around Treasury even t . This figure represents the 4 year means for the annual fraction of total debt issued years. The sample is broken down into debt issues with a rating between A and AAA and issues with a rating of BBB. Means are calculated over the period of 1998 2001 and over the period of 2002 2005 respectively .

PAGE 72

72 CHAPTER 3 DEBT SPECIALIZATION AND CREDITOR CONCENTRATION Motivation Why do some firms rely almost exclusively on one type of debt (for example unsecured public debt) while others diversify across as number of different types of debt? Rauh and Sufi (2010) use a sample of publicly traded firms with rated debt outstanding and find that corporate debt structure varies across credit quality, with investment grade rated firms borrowing mostly senior unsecu red debt from arms length sources, whereas lower rated firms on average tend to have multi tiered debt structures, both in terms of the number of different classes as we ll as their priority structure . 1 Moreover, they find that in response to rating downgra des, firms tend to diversify their funding sources. In contrast, a recent study by Colla, Ippolito, and Li (2013) finds that the propensity to borrow primarily through one class of debt is not just limited to highly rated firms, and that a substantial numb er of unrated and lower rated companies exhibit Colla, Ippolito, and Li (2013) use debt specialization as a proxy for creditor concentration and argue that debt specialization is associated with lower costs of financial distress. However, if specialization is associated with lower distress costs, as Colla, Ippolito, and Li (2013) argue, why do companies appear to migrate towards more complex debt structures following a ratings downgrade? In this paper we attempt to explain these seemingly conflicting patterns in debt specialization. We begin by distinguishing debt specialization, as measured by the 1 Rauh and Sufi (2010) argue that their empirical results are in line with theoretical models of credit quality and debt structure such as Park (2000). In these models lower rated firms can only obtain external financing if a senior claimant monitors the borrowe r, but at the same time monitoring incentives can be preserved by only holding a limited amount of the overall capital structure.

PAGE 73

73 number of different classes of debt a firm uses, from creditor concentration, as measured by the number of lenders within a class of debt. We show that empirically specialization and concentration are only loosely correlated. For example, as firms shift from public debt to intermediated bank debt, their degree of specialization may decline while their creditor We next examine the relationship between financial distress costs and both debt specialization and creditor concentration. The hypothesis that companies specialize in order to mitigate the costs associated with distres s by reducing the amount of different debt classes in their capital structure is motivated by the literature on troubled debt restructurings. For instance, Asquith , Gertner, and Scharfstein . (1994) hypothesize that firms with more complex debt structures a re more difficult to restructure outside of bankruptcy because of hold out problems and incentive conflicts between different creditor classes. Such conflicts across different debt classes can arise because unimpaired senior or secured lenders often have l ittle incentive to renegotiate outside of Chapter 11, as any concessions they make effectively constitute a wealth transfer to more junior claim holders. Creditor concentration is also likely to be an important driver of financial distress costs as free rider and hold out problems make dispersedly held claims difficult to restructure outside of bankruptcy. Consistent with t his argument, studies by Gilson, John, and Lang (1990) as well as by Gertner and Scharfstein (1991) provide evidence that, conditional on financial distress, firms with dispersedly held public bonds are less likely to restructure their claims outside of bankruptcy than firms that rely primarily on bank debt. However, Demiroglu and James (2013) provide evidence that hold out

PAGE 74

74 problems make dispersedly held senior bank claims even more difficult to restructure outside of bankruptcy than public debt. To empirically examine the relationship between financial distress costs and debt structure we rely on two datasets which have both been augment ed with detailed debt structure and priority information for non financial U.S. public firms obtained from S&P Capital IQ for the period of 2002 through 2011. The first dataset consists of detailed lender information for a large sample of syndicated and si ngle lender loans from Thomson Reuters LPC Dealscan at their origination, while the second dataset is panel data containing detailed annual financial information from Compustat. Using these two datasets we are able to distinguish between debt specializatio n and creditor concentration. We begin by analyzing the relationship between debt specialization and creditor concentration using the dataset on loan origination from Dealscan. Specifically, we examine this relationship by estimating both linear and non lin ear models of bank loan creditor concentration at the origination of the loan. As outlined in the following section , we follow Colla, Ippolito, and Li (2013) and use as our primary measure of debt specialization an indicator variable (Excl90) that takes on a value of one if at least 90 % comprised of one class of debt. 2 In order to measure creditor concentration we construct two different measures using the Dealscan database: the 2 Specifically, as outlined in the Data section , the classes of debt that Capital IQ distinguishes are commercial paper, drawn credit lines, term loans, senior bonds, subordinated bonds, capital leases, and other debt.

PAGE 75

75 number of lenders in a loan syndicate and whether or not a loan package contains an institutional loan tranche (identified as Term B Loans). 3 4 Overall we find that debt specialization is not highly correlated with lender concentration. Indeed, we find no statistically significant relationship between sp ecialization and creditor concentration for investment grade rated firms nor for unrated or below investment grade rated firms once other firm characteristics and loan characteristics are controlled for. Specifically, we find that neither the syndicate siz e nor the likelihood of issuing an institutional loan tranche are significantly impacted by a we discuss in our empirical analyis , this finding is robust to alternative measures o f debt specialization, such as the normalized Herfindahl Hirschman Index (HHI) (also used by Colla, Ippolito, and Li (2013) ) as well as the fraction of bank and other debt claims. The lack of a positive correlation between debt specialization and creditor concentration suggests that any test of whether reliance on one debt class is driven by lowering distress costs needs to include measures of creditor concentration in order distinguish between the effects that result from a reduction in the number of outst anding debt classes and the effects that are driven by greater creditor concentration. If distress costs are a primary driver of corporate debt structure choices, we would expect companies to shift their debt structures away from dispersedly held public de bt towards more concentrated intermediated bank debt in response to declining 3 This c lassification is motivated by Standard & P oor e to the U.S. Loan Market (2012 ) which argues that Term B loans are typically held by institutional inv estors. Moreover, Nadauld and Weisbach (2012) document that Term Loan B facilities are significantly more likely to be securitized through CLOs than their pro rata counterparts. 4 An additional measure would be a Herfindahl index of ownership using the fra ctions of the loans held by different lenders. Unfortunately, Dealscan coverage of lender shares is not very extensive.

PAGE 76

76 operating performance well before the onset of financial distress. We examine this question by following an approach similar to Rauh and Sufi (2010) and focusing on firms whose d ebt structure changes substantially but whose leverage ratios remain relatively constant. Specifically, using data from Compustat and Capital IQ we identify firms that have a long term credit rating by S&P and that shift from relying primarily on dispersed ly held public debt to bank debt. 5 A rated company is defined as making a large major change if its annual change in the fraction of public debt is in the lowest decile and was compensated by at least 90 % through a corresponding increase in the fraction of bank debt. Consistent with the idea that firms shift away from dispersedly held debt towards more concentrated intermediated bank debt in response to deteriorating operating performance well outside of states of severe financial distress, we find that EBI TDA to sales growth is negatively and significantly related to the likelihood of a firm shifting away from public debt towards more bank debt. We further document that this increase in bank debt mostly happens through intermediated bank debt (non instituti onal bank debt) as over 80 % of the firms that switch to more bank debt do not issue institutional Term B loans in the year of the shift. However, one concern with using EBITDA to sales growth to analyze changes in corporate debt structures is that changes in operating performance are unlikely to be exogenous in the context of corporate debt composition choice. We address this concern by estimating an instrumental variable probit model in which we use the annual average of monthly returns of a trade weighted U.S. Dollar 5 We use the presence of a credit rating as a proxy for whether a firm has public debt outstanding. For rated firms we define disperse dly held public debt to be the sum of senior bonds, subordinated bonds and commercial paper. We further define bank debt as the sum of drawn credit lines and term loans.

PAGE 77

77 index (TXWEB) as well as the annual average of monthly returns of an economic policy uncertainty index (EPU) as instruments for EBITDA to sales growth. In line with the above results, we find a significant and negative relationship between the likelihood of shifting from public debt towards more bank debt and the instrument when instrumenting EBITDA to sales growth with TXWEB and EPU. We next examine whether debt specialization is associated with lower costs of distress and whether this relatio nship is stronger for companies with more concentrated creditor holdings. To measure costs of distress we follow an approach similar to Opler and Titman (1994) and identify industries that have experienced economic distress, as measured by steep annual dec lines in the median industry stock price and negative median annual sales growth. We then investigate whether differences in debt structure composition and creditor concentration prior to the distress period are related to operating performance during the distress period. We measure operating performance similar to Andrade and Kaplan (199 8) through industry adjusted growth in EBITDA to sales, capital expenditures to sales, as well as asset growth. Because coverage in Capital IQ only begins in 2002, the majority of firm years based on this definition of distressed industries are clustered w ithin the 2008 to 2009 period when solely relying on Capital IQ data. This is concerning, since relying crises in the banking system, as lenders might be unable or unwilling to accommodate borrowers facing economic distress by extending new debt financing to them or by renegotiating the terms of existing debt contracts. For instance, Cornett et al. (2011) document that as liquidity dried up during the recent financial crisis, banks that were

PAGE 78

78 more exposed to liquidity risk significantly reduced their lending activities. Moreover, a study by Santos (201 1 ) provides evidence that the large losses experienced by banks during the financial crisis led them to be more restrictive in t heir corporate lending. Therefore, in order to extend the sample we use the fraction of floating rate debt, measured as the fraction of long term debt that is tied to the prime rate from Compustat, as a proxy variable for the fraction of bank debt within a discussed below, this approach is motivated by the fact that about 97 % of loans in Dealscan over our sample period were floating rate, compared to 90 % of public bond issues being fixed rate. 6 Additionally, when estimating a simpl e model for the fraction of period, the fraction of floating rate debt alone explains about 90 % variation as measured by its normalized partial . Further more, using the fraction of floating rate debt as a proxy for bank debt allows us to extend the sample period from 1999 to 2011. In order to measure bank creditor concentration prior to the distress period we control for whether a firm issued an institutio nal bank loan in any of the three years prior to the industry wide downturn. Our results are consistent with the hypothesis that debt specialization accompanied by greater creditor concentration is associated with lower distress costs. Using the fraction o f floating rate debt as a proxy for the fraction of bank debt, we find that among non investment grade rated firms (i.e. unrated and below investment grade rated) prior to the financial crisis of 2008, those which rely exclusively on intermediated bank deb t prior to the distress period recover significantly faster from industry wide 6 The figure for public bonds is obtained from the SDC Platinum Database used in Chapt er 2 of this dissertation .

PAGE 79

79 downturns than those with more diversified debt structures or firms that rely more heavily on institutional bank debt. 7 8 For instance, prior to the financial crisis firms relyi ng solely on floating rate debt are about 16 % more likely to experience above industry median EBITDA to sales growth following an industry wide downturn than firms which rely exclusively on fixed rate debt. Furthermore, we document that the likelihood of a bove industry median CAPX to sales growth is about 11 % greater for firms which rely exclusively on floating rate debt than for those that rely on fixed rate debt. Moreover, consistent with the idea that specialization in dispersedly held debt by itself doe s not mitigate hold out problems and incentive conflicts between lenders, we find no statistically significant difference in the likelihood of above industry median EBITDA to sales growth or CAPX to sales growth, between firms that issue institutional term loans and firms that do not rely on floating rate debt at all. Consistent with the notion that reliance on bank debt might expose borrowers to bank solvency or liquidity shocks, our results further indicate that while firms that specialize in intermediate d bank debt, as measured by the fraction of floating rate debt for firms that do n o t issue institutional term loans, tend to recover faster from industry wide downturns prior to the financial crisis, this is no longer the case during the financial crisis o f 2008 and 2009. Specifically, for firms facing industry wide distress between 2008 and 2009 we find no statistically significant difference in the likelihood of above industry median EBITDA to Sales growth, above industry median CAPX to sales 7 We use the year 2008 as the cutoff for the financial crisis based on the findings by Ivashina and Scharfstein (2010) who document that new loan issuance contracted dramatically beginning with the first quarter of 2008. 8 Throug hout this paper we will refer to unrated firms and firms rated below investment grade collectively as non investment grade rated firms or lower rated firms.

PAGE 80

80 growth, or a bove industry median asset growth, between firms that rely exclusively on floating rate debt and those that do not. Overall this study contributes to the literature on debt structure in several ways. First, our results indicate that there is little empirical support that debt specialization is a proxy for measures of creditor concentration and thus, specialization alone is unlikely to mitigate conflicts between creditors of the same debt class. Second, we docume nt that firms tend to shift from dispersedly held debt to more concentrated debt as a result of declining operating performance well before suffering from severe financial distress. Third, our findings indicate that debt specialization in and of itself doe s not appear to mitigate distress costs, as measured by industry adjusted operating performance during industry wide downturns, unless specialization is accompanied by greater creditor concentration. Lastly, we provide evidence consistent with the idea tha t firms specializing in bank debt were exposed to the recent banking crisis and recovered slower from industry wide downturns than prior to the financial crisis. Data The basic data for this study are panel data obtained from the Compustat Fundamental Annu al file for a large sample of U.S. public firms over the period from 2002 to 2011 and, as outlined below, they are subsequently augmented with annual information on debt structure and debt priority from S&P Capital IQ as well as with measures of debt conce ntration from Thomson Reuters LPC Dealscan. 9 Our choice of sample period is driven by the fact that coverage of debt structure information by Capital IQ is limited before 2002. To ascertain that our results are not driven by firms with 9 Table A 2 in the appendix contains a full description of all the variables used in the empirical section.

PAGE 81

81 limited access to capital markets, we restrict this sample to companies with positive total debt and at least $100 million in total assets and $10 million in sales (measured in 2011 dollars). While about 55 % of firms with assets of less than $100mn special ize, most of them specialize in bank debt or capital leases, almost half of them have interest coverage ratios below one, and given their size they are unlikely to have access to public debt markets or syndicated loans. 10 Furthermore, following prior litera ture on capital structure we exclude financial firms (SIC codes 6000 6999) and utilities (SIC codes 4900 4999) due to the regulatory restrictions that may influence the debt structure choices of firms in these industries. We account for inflation by adjust ing all index (all urban consumers), unless specifically mentioned otherwise. Debt Structure and Measures of Debt Specialization We augment the data from Compustat with fir m level data on debt structure obtained from Capital IQ. As Colla, Ippolito, and Li (2013) discuss, and following their notation, Capital IQ distinguishes between seven mutually exclusive debt types: commercial paper (CP), drawn credit lines (DC), term loa ns (TL), senior bonds and notes (SBN), subordinated bonds and notes (SUB), capital leases (CL), and other debt (Other). Following Colla, Ippolito, and Li (2013) , we discard observations for which the difference in total debt between Compustat and Capital I Q exceeds 10 % of total debt 10 Between 2002 and 2009 the median size for syndicated loans to non investment grade rated U.S. i ndustrial firms covered by Deals can was about USD 100mn with only 25% of loans being below USD 30mn. For institutional term loans the median size was about USD 200mn with about 13% being below USD 30mn. For public issues covered by SDC the median issue size was USD 270mn.

PAGE 82

82 (about 15 % of observations). 11 Additionally, we discard observations for which the sum of the seven debt fractions is not between 0.98 and 1.02 (about 10 % of observations). Similar to Colla, Ippolito, and Li (2013) our main measu re of debt specialization is an indicator variable of specialization Excl90 that takes on a value of one if a firm obtains more than 90 % of its debt financing from one debt type alone. We also use their second measure of debt specialization which is the no rmalized Herfindahl Hirschman Index (HHI) defined as: (3 1) Where ( 3 2 ) The HHI is normalized between zero and one and higher values of the HHI indicate Loan l evel Measures of Creditor Concentration In order to examine the relationship between debt specialization and measures of de bt concentration we merge the annual financial and debt structure data from Compustat and Capital IQ with detailed loan level information for a large sample of single lender and syndicated loans originated between 2002 and 2011 from Thomson Reuters LPC Dea lscan. This merge is conducted using a link file based on Chava and 11 Specifically, we discard observations for which

PAGE 83

83 Roberts (2008). 12 Dealscan also provides deal level information on loans that are negotiated and grouped together as part of one loan package. For the purpose of this study, we focus on ter m loans and revolvers from completed and closed deals and discard observations with missing information on syndicate composition, loan amount, maturity, and whether the loan is secured (about 25 % of observations). Using the information on syndicate members at loan origination provided by Dealscan we create two different measures of debt concentration: the number of syndicate members at origination of the loan and whether a deal contains at least one institutional loan tranche identified as Term Loan B facil ities (including term loan facilities modified with a letter different than A). While the first measure is a straightforward measure of creditor dispersion the second measure is motivated by the fact that Term B loans are typically marketed to non bank ins titutional investors whereas revolving lines of credit and Term A loans are typically held by banks and finance companies. 13 Furthermore, as research by Nadauld and Weisbach (2012) suggests, Term Loan B facilities are significantly more likely to be securit ized through CLOs than credit lines or non institutional term loans. 14 Major Shifts in Debt Structure and Creditor Structure To examine whether firms substantially shift their debt structures away from dispersedly held debt towards more concentrated credito r structures as a result of 12 We are grateful to Michael Roberts for providing the University of Florida with an updated version of this link file. 13 See for instance Standard & P oor e to the U.S. Loan Market (2012 ) for a discussion of different loan ty pes and the types of investors to which they are marketed. 14 An additional measure would be a Herfindahl index of ownership using the fractions of the loans held by different lenders. Unfortunately, Dealscan coverage of lender shares is not very extensive and would reduce the sample size by another 66%.

PAGE 84

84 deteriorating operating performance we follow an approach similar to Rauh and Sufi (2010). Specifically, we use the merged panel dataset of annual Compustat and Capital IQ data to analyze major shifts between public debt and ban k debt for firms with relatively stable leverage ratios between 2003 and 2011. We restrict the sample to firms that have a long term credit rating by S& P, since these are the most likely candidates to issue public debt or to have public debt outstanding, and for these rated firms we define public debt to be the sum of senior bonds, subordinated bonds and commercial paper. We similarly define bank debt as the sum of drawn credit lines and term loans. We define companies as having stable leverage ratios if the absolute value of their annual change in total debt over lagged total assets does not exceed 2.5 % . We further identify firms with substantial shifts f rom dispersedly held debt to more concentrated debt if their annual change in the fraction of public debt is in the lowest decile and if at least 90 % of that change is compensated by a corre sponding increase in bank debt. 15 For the companies that shift from public debt to bank debt we obtain information on whether this shift came as a result of early repurchases, scheduled principal payments, or firms exercising their right to call their outstanding bonds prior to maturity from their 10 K filings and reclass ify observations that are a result of data misclassification s in Capital IQ as well as observations that are not a substitution between public and bank debt. Our primary measure of operating performance is the annual change (gr owth) in EBITDA to sales rati o. 16 In order to ascertain that our results are not driven by large outliers in the EBITDA to sales growth variable we also measure operating performance 15 For robustness tests we define major shifts from bank debt to public debt in a similar manner. 16 Specifically:

PAGE 85

85 through an indicator variable for whether firms experienced negative annual EBITDA to sales growth. How ever, as mentioned above, a further concern with the change in annual EBITDA to sales ratio is that it is unlikely to be exogenous in the context of debt composition choice. Therefore we use two macro level variables as instruments for the growth in EBITDA to sales ratio: the annual average monthly returns of the Economic Policy Uncertainty Index (EPU) developed by Baker , Bloom, and Davis (2013), as well as the annual average monthly returns of the Broad Trade Weighted U.S. Dollar Index (TXWEB) from the Fed 17 Distressed Industries and firm level Measures of Creditor Concentration To analyze the relationship between debt specialization and creditor concentration with distress costs we use the panel dataset of Compustat d ata to identify firms in economically distressed industries. We follow a methodology similar to Opler and Titman (1994) by defining distressed industries for the years from 1999 through 2009. We define industries by three digit SIC codes and consider an in dustry to be distressed if the median annual stock return in CRSP was less than 20 % and the median annual sales growth was negative. Furthermore, we restrict the sample to industries that have not experienced distress in the previous two years and for whi ch there are at least five firms in Compustat for the corresponding year. We then further restrict the sample to unrated and below investment grade rated firms, and to firms that are not in Chapter 11 during the year of industry wide distress as identified by the UCLA LoPucki Bankruptcy Research Database. Simila r to Andrade and Kaplan (1998) we measure distress costs through changes in operating performance using industry 17 The data for the EPU Index wer e obtained from www.policyuncertainty.com and the data for the TXWEB Index were obtained from http://research.stlouisfed.org/fred2

PAGE 86

86 adjusted changes in operating margins (EBITDA/Sales), capital expenditure margins (CAP X/Sales) as well as asset growth. Specifically, we analyze changes in operating performance from the year of industry distress to two years following distress and adjust these values by subtracting the corres ponding industry median values. 18 While these are indirect measures of distress costs they should capture whether companies are downsizing by selling assets or whether they are curtailing investment. However, using the above definition of industry distress and solely relying on measures of debt structure from Capital IQ results in a small sample of firms for which the years of industry wide distress are clustered within the 2008 to 2009 period. Therefore, instead of using debt structure data from Capital IQ we proxy for the fraction of bank debt in a firm measured as the fraction of long term debt that is tied to the prime rate from Compustat. This approach allows us to extend the sample period to 1999 to 2009 and results in a larger sample of f irm years, 18 % of which are in the 200 8 to 2009 period. Furthermore, we proxy for dispersedly held institutional bank debt by identifying firms that issued Term B loans in the current fiscal year or in the previous two fiscal years by using the Dealscan database. 19 Empirical Analysis Descriptive Statistics Our basic sample consists of over 14,000 firm year observations by about 2900 individual companies for which we have detailed financial information from Compustat 18 Specifically: , , adjusted by median industry values. 19 Throughout this study we define Term B loans as all term loan facilities modified with a letter different than A.

PAGE 87

87 and debt structure information from Capital IQ. Despite the longer time period this study covers , the sample is slightly smaller than the one by Colla, Ippolito, and Li (2013) because we restrict our sample to firms with assets of $100 million or more and sales of at least $10 million . As discussed in the previous section , we impose this size constra int in order to mitigate concerns that our results might be driven by firms with limited access to capital markets. Without the size constraint, and over the same time period, the number of observations would be comparable to theirs. Table 3 1 provides des criptive statistics concerning the financial characteristics as well as the debt structure and priority structure for the firms in our sample. In order to highlight the differences in debt structure between firm s of different rating classes, we split the s ample into firms with an investment grade credit rating (BBB or above), firms rated below investment grade (BB or below), and firms without a credit rating. Despite the difference in firm size our sample is comparable to the one Colla, Ippolito, and Li (20 13) both in terms of the financial characteristics of sample firms as well as in terms of their debt structure characteristics. For instance, the financial cha racteristics displayed in Table 3 1, such as the average leverage ratio, profitability (measured by EBITDA to total assets), tangibility (measured by PPE to total assets) as well as the fraction of firms that pay dividends are similar to those reported by Colla, Ippolito, and Li (2013). Both the average and the median firm in our sample are far from b eing financially distressed as the median interest coverage ratios, as measured by EBITDA to interest expense, are over eleven for investment grade rated firms, almost four for below investment grade rated firms, and about nine for unrated firms.

PAGE 88

88 Turning t o debt structure and priority structure, Table 3 1 also display s descriptive statistics for the debt structure and priority structure of the different debt and priority classes identified by Capital IQ. Consistent with the findings of Rauh and Sufi (2010) investment grade rated firms rely heavily on arms length sources of financing. For instance, the average investment grade rated company obtains almost 75 % of its debt through senior bonds and almost all of its debt is equal in priority with an average of 9 2 % being senior and unsecured. Moreover, the average non investment grade rated firm seems to have a multi tiered debt structure, both in terms of the classes of debt it employs as well as the priority structure of its different debt claims. 20 Specifically, below investment grade rated firms, on average, obtain about 30 % of debt their financing from bank debt , 40 % from senior bonds and 20 % from subordinated bonds. Unrated firms, on average, rely slightly more on bank financing and obtain about 50 % of their d ebt from banks and about 26 % and 7 % from senior and subordinated bonds respectively. Furthermore, the average priority structure becomes more complicated as credit ratings decline, with almost half of the debt of lower rated firm being senior and secured, about 40 % senior unsecured, and the remainder coming from various kinds of subordinated debt classes. While the average and priority debt structure of firms in our sample is in line with the findings of Rauh and Sufi (2010), a substantial number of firms a cross the credit quality spectrum obtain their financing primarily through one form of debt. As shown in Table 3 1 , over 42 % of investment grade rated firms, about 35 % of below investment grade rated firms, and about 55 % of unrated firms obtain 90 % or more of their debt from 20 We will refer to unrated firms and firms rated below BBB collectively as non investment grade, or alternatively as firms of lower credit rating throughout this paper.

PAGE 89

89 one debt class alone. Table 3 1 further breaks this tendency for specialization down by the individual debt classes as reported by Capital IQ. As can be seen, it is quite rare for highly rated firms to specialize in anything oth er than senior bonds. Indeed, almost 39 % of investment grade rated firms obtain virtually all their debt financing from senior bonds alone. On the other hand, while a greater percentage of lower rated firms tend to specialize they also do so across a varie ty of different debt classes. While about 9 % of all below investment grade rated firms and about 26 % of all unrated firms do obtain 90 % or more of their debt financing in one form of bank debt, over 19 % of below investment grade companies specialize in sen ior bonds and about 6 % in subordinated bonds, whereas the corresponding percentages for unrated firms are 14 % and 4 % respectively. While it is the case that smaller firms and firms with less debt tend to be more likely to specialize, the different types of debt firms specialize in are not solely determined by firm size and leverage. When splitting the combined sample of below investment grade rated firms and unrated firms into size quartiles (based on total assets) and leverage quartiles (based on book leve rage) there is still considerable within quartile variation in the types of debt firms specialize in. For instance, for the smallest size quartile about 30 % of firms specialize in one form of bank debt while about 15 % specialize in senior bonds or subordin ated bonds. For the largest size quartile the percentage of firms specializing in bank debt is about 9 % while about 21 % specialize in senior or subordinated bonds. The corresponding numbers for the first and fourth leverage quartile are of comparable magni tude.

PAGE 90

90 The Relationship between Debt Specialization and Creditor Concentration As discussed earlier, we argue that unless debt specialization is positively correlated with creditor concentration, exclusively relying on one form of debt will not be sufficien t to mitigate distress costs because conflicts within debt classes might still persist. However, there is no reason to assume that such a positive correlation actually exists given that companies specialize in a variety of different debt classes, some of w hich can have a highly dispersed creditor structure. In this subsection we analyze whether measures of debt specialization are indeed correlated with measures of creditor concentration. Our identification strategy relies on a detailed dataset of lender inf ormation from single lender and syndicated loans at the time of their origi nation with which we are able to measure the creditor con centration of new loan issues. We then pro ceed to estimate linear and non linear models relating our measures of debt special ization to the two different measures of creditor concentration discussed in the prior section: the number of different lenders and whether a loan package includes an institutional loan tranche. Admittedly, one concern with identifying the relationship bet ween debt specialization and creditor concentration using lender information for syndicated loans alone, instead of including creditor information on all types of debt, is that it might increase the probability of our tests falsely accepting the null hypot hesis o f no relationship. However, as we outline below, our results are robust to controlling for the level of bank debt instead of using the more general measures of specialization, such as Excl90 and HHI, which should help to at least partially mitigate this concern. Our sample consists of approximately 5300 loans from Dealscan. Table 3 2 and Table 3 3 display the firm characteristics and debt structure information for firm years associated with a loan observation in Dealscan, and compare them to firm yea rs without

PAGE 91

91 an associated observation in Dealscan. Motivated by the differences in debt specialization for firms of different rating classes, the sample is further split up into rating classes with Table 3 2 displaying the results of investment grade rated firms, and Table 3 3 for fir ms rated below investment grade and firms without a credit ra ting. As can be seen from Table 3 2 the two samples seem to be fairly similar for investment grade rated firms. Indeed, there do not appear to be any statistically or economically significant differences between the two samples. On the other hand, given the fact that Dealscan primarily covers large syndicated loans, it is perhaps not surprising to find that among unrated and below investment grade rated firms there are several differences between the t wo samples. For instance, Table 3 3 shows that among unrated and below investment grade rated firms those covered by Dealscan are slightly larger, tend to be more profitable, and tend to have more tangible assets. Moreover, non investment grade rated firms covered by Dealscan tend to have slightly more diverse debt structures, as measured by both the HHI and Excl90, and tend to rely more on bank debt financing. Even though the fraction of unrated and below investment grade r ated firms that rely on one form of debt for their financing is about 7 % lower when compared to the Compustat sample, over 17 % of below investment grade rated firms and over 12 % of unrated firms rely exclusively on senior bonds. Moreover, their reliance on subordinated bonds is only slightly lower than for the Compustat sample with over 5 % and 2.5 % respectively. In the first set of regressions we estimate the relationship between debt specialization and creditor concentration, measured as the logarithm of one plus the number of lenders in a loan syndicate, controlling for loan and firm characteristics as

PAGE 92

92 well as Fama French industry fixed effects and year fixed effects. 21 22 Since t he covariates might affect creditor concentration differently for companies of different credit quality, and because firms of lower credit quality tend to be more heterogeneous with respect to the class es of debt they specialize in, we estimate the models separately for investment grade rated borrowers and non investment grade rated borrowers. However, our results are robust to further distinguishing between unrated firms and firms rate d below investment grade. Table 3 4 presents the estimates of these line ar models for firms with an investment grade credit rating . As can be seen in Columns (1) through (5), debt specialization is not associated with a significant effect on the syndicate size of loans obtained by companies with investment grade credit ratings . Indeed, even in the univariate regression in Column (1) there is no significant correlation between the two. Rather, syndicate size appears to be most strongly related to loan amount and loan maturity as well as firm size. Similarly, as shown in Table 3 5 , there is little evidence of a significant relationship between specialization and creditor concentration for lower rated firms. While the univariate regression in Column ( 1 ) displays a negative and significant coefficient estimate for Excl90, the multiv ariate specifications in Columns ( 2 ) through ( 5 ) indicate that Excl90 does not have any explanatory power, both in terms of 21 We include year fixed effects in order to capture unobserved macro economic factors that might affect syndicate size and composition. However, the results are qualitatively similar when excluding the year fixed effects. 22 We use the logarithm of one plus the number of syndicate members in order to limit the influence of large outliers in syndicate size. However , the results are robust to estimating the linear models with the number of syndicate members as the dependent variable.

PAGE 93

93 statistical signific ance and in terms of partial , once loan and firm characteristics are taken into consideration. 23 While the re sults in Table 3 4 and Table 3 5 suggest that debt specialization is at best loosely correlated with the number of lenders at origination, a potential concern with this measure is that it not accurately reflects the actual concentration of creditors. Notic e for instance, that the coefficient estimate on the Term Loan B indicator variable in Columns ( 4 ) and ( 5 ) of Table 3 5 is negative and significant even though Term B loans tend to be more dispersedly held, because they can be traded and sold on the second ary market or they can be securitized through CLOs after origination. Moreover, the results discussed so far have primarily focused on the creditor composition of individual loans. However, often loans are negotiated together as part of a larger deal, or l oan package, which involves several different loan tranches. While these tranches might be held by different lenders, might have different maturities, interest rates or principal payments, they are typically all governed by the same credit agreement and su bject to the same set of contrac tual covenants. 24 To address this issue we use an additional measure of creditor concentration which relates more to the types of lenders participating in a syndicate than their absolute number and which is associated with th e likelihood of being more dispersedly held: whether a loan package includes an institutional tranche. 23 A potential concern with the se regressions is that measures of debt specialization could be strongly correlated with the other covariates. In untabulated results we show that this does not appear to be the case. Specifically, regressions with measures of debt specialization as the de pendent variable indicate that the other covariates only explain about 13% of the total variation in debt specialization for non investment grade rated and less than 1% for investment grade rated firms. 24 See for instance Sufi (2007)

PAGE 94

94 Table 3 6 analyzes the relationship between debt specialization and creditor concentration at the loan package level by estimating a series of logit mode ls where the dependent variable takes on a value of one if the loan package includes an institutional loan tranche. As institutional investors participate primarily in the lower rated and high yield segmen t of the corporate loan market we only estimate the models for non investment grade rated firms. Consistent with the prior findings, Columns (1) through (5) indicate that the coefficient estimate for Excl90 has no explanatory power for the likelihood that firms obtain an institutional bank loan. While the binary nature of Excl90 makes the results easy to interpret, the cutoff defining specialization at the 90 % level of total debt is admittedly arbitrary. Therefore, in untabulated results we also estimate the models in Table 3 4 through Table 3 6 using an alternative measure of debt specialization the normalized HHI measure defined by Colla, Ippolito, and Li (2013). We find qualitatively similar results when using the HHI instead of Excl90. Specifically, even when using the HHI measure of specia lization we find no statistically significant relationship between HHI and the number of syndicate members. There is however, a weakly significant relationship between HHI and the likelihood of issuing an institutional loan tranche ( at the 5 % and 8 % signif icance level). However, the economic magnitude of an increase in specialization, as measured by HHI is rather small. For instance, a one standard deviation increase in HHI is only associated with a 1% decrease in the likelihood of obtaining an institutiona l loan tranche. As outlined above, one potential concern with the above tests for estimating the relationship between debt specialization and creditor concentration is that our

PAGE 95

95 identification strategy revolves solely around lender concentration in bank loa ns and does not identify the creditor structure in the other classes of debt firms employ. Specifically, it could be the case that the creditor concentration in the other non bank debt classes that firms specialize in is not correlated with the concentrati on of lenders within their bank debt claims. While the quality of the data prohibit s us from compl etely addressing this concern, we partially address it by distinguishing between specialization in different classes of debt and making sure that the results are not driven by firms that specialize in forms of non bank debt. To that end, in untabulted results, we estimate a series of mode ls similar to the ones in Table 3 4 and Table 3 5 and replace the Excl90 measure with a series of indicator variables taking on a value of one if a firm obtains more than 90 % of its debt financing through the correspo nding debt type. Specifically, we include an indicator variable if the main source of financing is bank debt, as measured by the sum of drawn revolvers and term loa ns, and a second indicator variable if the main source is any other non bank class of debt. We find no statistically significant relationship between the se measures of specialization and the number of syndicate members for highly rated firms. For firms tha t are unrated o r rated below investment grade we find that firms that specialize in debt sources other than bank debt have significantly smaller lending syndicates (at the 5 % level). However, the economic magnitude on syndicate size is rather insignificant , given that the average syndicate consists of about seven lenders and that firms specializing in non bank debt, on average, have about 5 % smaller syndicates. Substantial Shifts in Debt Structure and Creditor Structure As outlined in the introduction, if d istress costs are a primary driver of debt structure choice, then we would expect companies to shift from public debt to more

PAGE 96

96 intermediated bank debt in response to deteriorating operating p erformance. In this subsection we test this hypothesis by analyzin g major annual shifts between public debt and bank debt for a set of companies whose leverage stays relatively constant (absolute change in total debt over lagged ass ets is not greater than 2.5 % ). We restrict the sample to firms that have a long term credi t rating by S&P, since these are the most likely candidates to issue public debt or to have public debt outstanding, and define public debt to be the sum of senior bonds, subordinat ed bonds and commercial paper. We similarly define bank debt as the sum of draw n credit lines and term loans. We define a shift from public debt towards more bank debt as substantial if the annual change in the fraction of public debt is in the lowest decile and was compensated at least to 90 % by a corresponding increase in bank debt (and vice versa for major shifts from bank debt to public debt). 25 Table 3 7 presents firm level summary statistics for the year prior to a substantial shift from public debt to bank debt and compares them to the year end of the year in which the shift took place. As can be seen, a major shift in debt structure towards more bank debt does not appear to be associated with firms moving from the investment grade category to the below investment grade or unrated categories. Moreover, neither the average nor the median firm appears to be in financial distress as median interest coverage ratios are well above four in both cases. However, operating margins as measured by EBITDA to s ales appear to be about 2.5 % lower for firms that moved from public debt to bank debt. Table 3 7 further shows that these major shifts between different debt classes are indeed quite substantial. For instance, the median firm that 25 The lowest decile in the annual change in the fraction of public debt for all rated firms with stable leverage ranges from 1 to 0.08 with a mean change of 0.33 and a median change of 0.17.

PAGE 97

97 increased its bank debt did so by reducing the fraction of public debt by about 23 % , which corresponds to a decrease in public debt of about $210 million . Moreover, on average about 55 % of the increase in bank debt was financed through the increasing use of revolving lines of credit whereas only about 18 % of all firms increased their bank debt through the use of institutional term loans (Term B loans). In order to analyze whether declining operating performance is related to the likelihood of shifting between different debt classes, we estimate a series of logit models in Table 3 8 relating the likelihood of a firm shifting from public debt towards more bank debt to different measures of operating performance. We control for firm characteristics such as size, leverage, whether the firm has an investment gra de credit rating, and the tangibility of assets. Because firms with outstanding callable debt might call their bonds in order to refinance at more favorable terms, and not in response to de clining operating performance, we further control for the term stru cture of interest rates and include year fixed effects. 26 Consistent with the idea that firms shift to more concentrated forms of debt when their operating performance declines, the coefficient estimates on EBITDA to sales growth in Columns (1) to (3) are n egative and highly statistically significant, even when controlling for the term structure of interests rates and year fixed effects in Column (3). Moreover, this effect is not driven by outliers in the EBITDA to sales growth variable as the coefficient es timates on Neg. EBITDA/Sales Gr. 1Y in Columns (4) to (6), which is an indicator variable for whether the experienced negative EBITDA to sales growth over the fiscal year, is positive and statistically significant at the 1% level as well. 26 We k filings. We document that in 41% of cases bonds were called, in 29% of cases they were repurchased and in 41% of cases the shift was due to scheduled principal payments. Note that these are not mutually exclusive reasons.

PAGE 98

98 The effect of a d ecline in EBITDA to sales growth is economically significant as well. For instance, for the model in Column (5) experiencing a year with negative EBITDA to sales growth is associated with significantly increasing the estimated likelihood of a firm shifting away from public debt towards more bank debt from 2.9 % to about 5.7 % . 27 We further document that the decline in operating performance that is associated with the likelihood that companies substantially shift their debt structures away from public towards more bank debt is not merely a one time shock, but rather seems to be a persistent decline. For instance, when estimating the models in Columns (1) to (3) of Table 3 8 with the forward looking EBITDA to sales growth over the next two years as the dependent variable of interest, instead of the curren t year EBITDA to sales growth, we find negative coefficient estimates that are significant at the 2.1 % through 6.1 % level. A potential concern with the model specifications in Table 3 8 is that changes in operati ng performance are not exogenous with respect to debt structure cho ices. Therefore, we estimate an instrumental variables probit model for the likelihood to shift away from public debt towards more bank debt, where we instrument EBITDA to sales growth with two macro level variables: the annual average monthly returns of the Economic Policy Uncertainty Index (EPU) developed by Baker, Bloom, and Davis (2013), as well as the annual average monthly returns of the Broad Trade Weighted U.S. Dollar Inde x (TXWEB). Column (1) of Table 3 9 displays the results of a first stage linear model for the firms in our sample where the dependent variable is the annual change in the EBITDA to sales ratio. The coefficient estimate on the annual average 27 In order to estimate the marg inal effects the likelihood of shifting from public debt to bank debt is estimated at the mean values of the corresponding covariates.

PAGE 99

99 monthly return on TXWEB is negative and significant at the 3 % level, which indicates that for the firms in our sample profit margins decline when the value of the U.S. Dollar appreciates. On the other hand, the coefficient estimate on the annual average monthly return on EPU is ins ignificant, but the two coefficients are jointly significant at the 4 % level. Moreover, the partial of TXWEB and EPU jointly expl ains almost all of the overall . Column (2) of Table 3 9 shows the results of the instrumental variable probit model. Co nsiste nt with the findings from Table 3 8 , the coefficient estimate on the instrumented EBITDA to sales growth variable is negative and statistically significant. As an additional robustness test we also verify that major shifts from bank debt towards more dispersedly held public debt are not associated with deteriorating operating performance. In untabulated results we estimate a series of logit mode ls similar to the ones in Table 3 8 , but where the dependent variable takes on a value of one if the company significantly increased its public debt and simultaneously reduced its bank debt. Consistent with expectations, we do not find any significant relationship between EBITDA to sales growth and the li kelihood of shifting towards more public debt. Debt Specialization, Creditor Concentration, and Distress Costs The findings in the prior subsections illustrate that there is little empirical evidence in favor of a positive correlation between specializatio n and creditor concentration, and that major shifts from more dispersedly held debt toward more tightly held intermediated bank debt are associated with companies responding to unanticipated declines in operating performance (well outside of financial dist ress). Motivated by these findings, we analyze whether debt specialization and lender

PAGE 100

100 concentration are related to measures of distress costs. Specifically, as outlined in the motivation , we argue that debt specialization by lower rated firms should only b y reflective of their efforts to reduce the costs associated with distress to the extent that this specialization simultaneously occurs with greater creditor concentration. In order to test this hypothesis we identify unrated and below investment grade rat ed firms in industries facing economic distress between 1999 and 2009, following an approach similar to Opler and Titman (1994). Specifically, as outlined in the data section , we define industries by 3 digit SIC codes and consider an industry distressed if its median annual stock return was less than 20 % and if its median annual sales growth was negative. We then use within industry variation in creditor concentration and debt structure for the year prior to the industry wide economic downturn to estimate logit models, where we measure distress costs through changes in operating performance during the distress period. Specifically, our dependent variables of interest are indicator variables for whether a firm experienced above industry median growth in EBIT DA to sales, capital expenditures to sales and total assets over the two year period following the year of an industry wide downturn. 28 Our main measure of debt structure is the fraction of floating rate debt to total debt obtained from Compustat. As discus sed above, we use this measure as a proxy the sample size and to avoid clustering of distressed firms within the financial crisis time period. Our choice of a proxy variab le for bank debt is further motivated by the fact that over our sample period about 97 % of loans in Dealscan have floating rate interest rates, 28 Specifically: , , adjusted by median industry values.

PAGE 101

101 whereas about 90 % of public bond issues in SDC Platinum have fixed rate coupons. For our main measure of credito r concentration we further proxy between dispersedly held institutional bank debt and more tightly held intermediated bank debt by including an indicator variable for whether the firm issued Term B loans in the three fiscal years prior to industry wide dis tress. To illustrate that the fraction of floating rate debt is indeed highly correlated with the fraction of bank debt, we estimate a simple model using the Capital IQ data on debt structure in Table 3 10 . As shown in Columns (1) and (2) of Table 3 10 , the fraction of floating rate debt is highly statistically significant and positively related to the fraction of bank debt. Including the fraction of floating rate debt into the regression also dramatically increases the overall from 0.10 to over 0.38. Moreover, when conducting an analysis of covariance following the methodology of Lemmon, Roberts , and Zender (2008) we find that the fraction of floating rate debt explains about 90 % of the variation in bank debt, as m easured by its normalized partia l . The industry distress sample consists of 629 companies without an investment grade credit rating. 29 Table 3 11 presents summary statistics for the firms in distressed industries for the year prior to distress and compares them to firms without an inv estment grade credit rating in industries that never experience a downturn. As can be seen, the two samples are fairly similar with respect to size, market to book ratio, interest coverage ratio, as well as ratings distribution. However, firms in distresse d industries tend to have lower profitabi lity, as measured by EBITDA to s ales and EBITDA to total assets, tend to have lower capital expenditures, and tend to rely less on bank 29 About 14% of firms face industry wide distress between 1999 and 2001, 58% in 2002, 10% between 2003 and 2007, and 18% in 2008.

PAGE 102

102 debt for their debt financing, as measured by the fraction of floating rate deb t. On the other hand, even though firms in distressed industries tend to rely less on bank debt, the fraction of firms that issue institutional bank debt in the previous three years is similar between the two groups. Table 3 12 presents the results of the logit models analyzing the relationship between operating performance following and industry wide downturn and debt structure and creditor concentration prior to industry distress. For two of our three measures we document that, prior to the financial cris is of 2008, the fraction of floating rate debt is positively and significantly related to the likelihood of above median industry performance. Specifically, Columns (1) and (3) indicate that companies that rely exclusively on floating rate debt are about 1 6 % more likely to experience above industry median EBITDA to sales growth and about 11 % more likely to experience above industry median CAPX to sales growth than companies that rely exclusively on fixed rate debt. Moreover, when the fraction of floating ra te debt is interacted with an indicator variable for whether the firm issued institutional bank loans in Columns (2) and (4), the joint significance of the two coefficient estimates is not statistically different from zero at conventional significance leve ls, which is consistent with the idea that dispersedly held institutional bank debt is harder to renegotiate and restructure than tightly held intermediated bank debt. Taken together, these results indicate that companies that relied more heavily on interm ediated bank debt prior to the financial crisis recovered significantly faster from industry wide distress than firms that relied on more diversified debt structures or those that relied more heavily on institutional bank debt. To the extent that our measu res of operating performance proxy for distress costs this would be

PAGE 103

103 consistent with the hypothesis that debt specialization in and of itself does not mitigate distress costs that arise from within creditor class conflicts. As outlined in the motivation of this paper there is evidence suggesting that banks curtailed their lending during the recent financial crisis, for instance because they were experiencing large losses or due to efforts to increase their cash holdings in order to mitigate their exposure to liquidity risk. It is therefore likely that borrowers that relied exclusively on bank debt for their financing might have been exposed to this banking crisis, in the sense that bank lenders might not have been willing to accommodate them by renegotiating the terms of their loans or by imposing more restrictive amendments when they violate covenants. In all our regressions pertaining to firms in distressed industr ies, we have accounted for this possibility by including an indicator variable (Fin. Crisis), w hich takes on a value of one if the year of industry distress was in 2008 or 2009, and interactions thereof with our measures of debt structure and concentration. Notice that the coefficient estimates for the interaction term between the fraction of floati ng rate debt and Fin. Crisis are negativ e in all of the models in Table 3 12 (with the exception of Column (4) where it is positive but statistically insignificant). Additionally, the joint significance of the fraction of floating rate debt with the intera ction term is not statistically significant at conventional levels. This indicates that firms whose industries faced economic distress during the financial crisis did not benefit from relying on intermediated bank debt at all, compared to firms with more d iverse debt structure s . Chapter 3 Concluding Remarks In this paper we examine the relationship between debt specialization and creditor concentration and the extent to which they are related to financial distress

PAGE 104

104 costs. We argue that distress costs are mos t likely mitigated when debt specialization is combined with a greater concentration of lenders. Overall, our findings are consistent with the notion that only when it is accompanied by greater creditor concentration does the exclusive reliance on one debt type help to mitigate distress costs. Specifically, we find no significant correlation between measures of debt specialization and measures of creditor concentration, but we document that companies with large declines in operating performance are signific antly more likely to substantially shift their debt structure away from dispersedly held public debt towards more tightly held intermediated bank debt. Moreover, these major shifts towards more concentrated forms of debt seem to occur well before states of financial distress. Furthermore, our results suggest that firms specializing in concentrated intermediated bank debt recover faster from an industry wide downturn than firms that specialize in other forms of debt or those that do not specialize at all. Ho wever, our results also indicate that the benefits from relying exclusively on concentrated intermediated bank debt might come at the cost of leaving these firms more vulnerable to solvency and liquidity shocks in the banking sector.

PAGE 105

105 Table 3 1. Summary statistics by rating category . Investment Grade Below Investment Grade Unrated Mean Median N Mean Median N Mean Median N 13665.130 6656.359 2355 3198.796 1621.484 3664 810.541 397.577 8278 Book 0.252 0.238 2355 0.430 0.395 3664 0.223 0.177 8278 Market 0.206 0.178 2355 0.429 0.392 3664 0.216 0.144 8278 Market 1.814 1.568 2355 1.433 1.263 3664 1.805 1.454 8278 0.199 0.171 2355 0.158 0.141 3664 0.020 0.109 8278 EBITDA 0.157 0.149 2355 0.119 0.116 3664 0.100 0.113 8278 20.443 11.160 2338 6.674 3.658 3642 33.613 9.167 7757 Dividend Payer 0.816 1.000 2355 0.285 0.000 3664 0.278 0.000 8278 0.311 0.246 2355 0.353 0.296 3664 0.273 0.190 8278 Below Investment Grade 0.000 0.000 2355 1.000 1.000 3664 0.000 0.000 8278 Unrated Firm 0.000 0.000 2355 0.000 0.000 3664 1.000 1.000 8278 RD Dummy 0.545 1.000 2355 0.323 0.000 3664 0.465 0.000 8278 HHI 0.696 0.719 2355 0.633 0.577 3664 0.759 0.868 8278 Excl90 0.427 0.000 2355 0.355 0.000 3664 0.554 1.000 8278 Bank Debt 0.129 0.024 2355 0.328 0.239 3664 0.502 0.523 8278 Drawn Revolvers 0.080 0.000 2355 0.099 0.000 3664 0.284 0.007 8278 Term Loans 0.049 0.000 2355 0.228 0.023 3664 0.218 0.000 8278 Senior Bonds 0.741 0.837 2355 0.428 0.392 3664 0.260 0.000 8278 Subord. Bonds 0.024 0.000 2355 0.202 0.000 3664 0.074 0.000 8278 Capital Leases 0.011 0.000 2355 0.018 0.000 3664 0.107 0.000 8278 Comm. Paper 0.049 0.000 2355 0.000 0.000 3664 0.002 0.000 8278 Other Debt 0.046 0.002 2355 0.024 0.000 3664 0.054 0.000 8278 This table presents summary statistics for the full sample of firms merged between Compustat and Capital IQ. The full sample consists of all firms with at least $100mn in total assets and $10mn in sales (2011 dollars). It is further split into firms with a n investment grade credit rating by S&P (BBB and above), those rated below investment grade (BB and lower), and those without a credit rating by S&P. Variables have been winsorized at the 1% and 99% level where

PAGE 106

106 Table 3 1. Cont inued . Investment Grade Below Investment Grade Unrated Mean Median N Mean Median N Mean Median N Total Senior Debt 0.974 1.000 2266 0.788 1.000 3428 0.920 1.000 8089 Total Senior Secured Debt 0.056 0.000 2266 0.385 0.315 3424 0.523 0.590 8064 Total Senior Unsecured Debt 0.918 0.998 2266 0.404 0.324 3424 0.400 0.093 8064 Total Subordinated Debt 0.025 0.000 2266 0.212 0.000 3428 0.079 0.000 8089 Sen. Sub. Debt 0.012 0.000 2266 0.154 0.000 3428 0.024 0.000 8089 Subordinated Debt 0.011 0.000 2266 0.050 0.000 3428 0.050 0.000 8089 Junior Debt 0.002 0.000 2266 0.008 0.000 3428 0.005 0.000 8089 Total Secured Debt 0.056 0.000 2266 0.399 0.330 3428 0.533 0.651 8089 Total Unsecured Debt 0.943 1.000 2266 0.601 0.671 3428 0.467 0.349 8089 Priority HHI 0.902 0.997 2266 0.670 0.629 3424 0.865 1.000 8064 Drawn Revolvers Excl90 0.014 0.000 2355 0.010 0.000 3664 0.150 0.000 8278 Term Loans Excl90 0.005 0.000 2355 0.074 0.000 3664 0.112 0.000 8278 Senior Bonds Excl90 0.386 0.000 2355 0.198 0.000 3664 0.140 0.000 8278 Subord. Bonds Excl90 0.006 0.000 2355 0.060 0.000 3664 0.042 0.000 8278 Capital Leases Excl90 0.000 0.000 2355 0.004 0.000 3664 0.076 0.000 8278 Comm. Paper Excl90 0.004 0.000 2355 0.000 0.000 3664 0.001 0.000 8278 Other Debt Excl90 0.012 0.000 2355 0.008 0.000 3664 0.034 0.000 8278 This table presents summary statistics for the full sample of firms merged between Compustat and Capital IQ. The full sample consists of all firms with at least $100mn in total assets and $10mn in sales (2011 dollars). It is further split into firms with a n investment grade credit rating by S&P (BBB and above), those rated below investment grade (BB and lower), and those without a credit rating by S&P. Variables have been winsorized at the 1% and 99% level where .

PAGE 107

107 Table 3 2. Summ ary statistics Dealscan sample (investment grade rated firms) . Investment Grade Rated Dealscan Compustat Mean Median Mean Median Mean Diff. 12.542 5.842 14.137 7.183 1.596 ** Book 0.258 0.246 0.249 0.236 0.009 Market 0.214 0.186 0.203 0.174 0.010 * Market 1.759 1.529 1.837 1.589 0.078 ** 0.196 0.165 0.200 0.175 0.004 0.153 0.148 0.159 0.149 0.006 ** 17.776 10.566 21.562 11.440 3.786 ** Dividend Payer 0.808 1.000 0.819 1.000 0.011 0.322 0.251 0.307 0.244 0.015 Below Investment Grade 0.000 0.000 0.000 0.000 0.000 Unrated Firm 0.000 0.000 0.000 0.000 0.000 RD Dummy 0.516 1.000 0.557 1.000 0.040 * HHI 0.701 0.730 0.694 0.715 0.007 Excl90 0.435 0.000 0.423 0.000 0.011 Bank Debt 0.131 0.025 0.127 0.023 0.004 Drawn Revolvers 0.085 0.000 0.078 0.000 0.007 Term Loans 0.046 0.000 0.049 0.000 0.003 Senior Bonds 0.740 0.844 0.742 0.834 0.002 Subord. Bonds 0.026 0.000 0.023 0.000 0.003 Capital Leases 0.011 0.000 0.011 0.000 0.000 Comm. Paper 0.052 0.000 0.048 0.000 0.005 Other Debt 0.039 0.002 0.049 0.003 0.010 * Total Senior Debt 0.971 1.000 0.975 1.000 0.004 Total Senior Secured Debt 0.064 0.000 0.053 0.000 0.012 Total Senior Unsecured Debt 0.907 0.996 0.923 0.999 0.015 * Total Subordinated Debt 0.028 0.000 0.024 0.000 0.004 Sen. Sub. Debt 0.011 0.000 0.012 0.000 0.001 Subordinated Debt 0.015 0.000 0.010 0.000 0.005 Junior Debt 0.002 0.000 0.002 0.000 0.000 Total Secured Debt 0.064 0.000 0.053 0.000 0.012 Total Unsecured Debt 0.936 0.999 0.947 1.000 0.011 Priority HHI 0.884 0.991 0.910 0.998 0.025 *** Drawn Revolvers Excl90 0.013 0.000 0.014 0.000 0.002 Term Loans Excl90 0.004 0.000 0.005 0.000 0.001 Senior Bonds Excl90 0.387 0.000 0.386 0.000 0.001 Subord. Bonds Excl90 0.010 0.000 0.004 0.000 0.006 Capital Leases Excl90 0.000 0.000 0.000 0.000 0.000 Comm. Paper Excl90 0.010 0.000 0.001 0.000 0.009 ** Other Debt Excl90 0.010 0.000 0.013 0.000 0.003 This table presents summary statistics for the Compustat sample and for firms that issued loans between 2002 and 2011 from Dealscan. The full sample consists of all firms with at least $100mn in total assets and $10mn in sales (2011 dollars). It is further restricted to firms with an investment grade credit rating by S&P (BBB and above ) . For illustrative purposes dollar amounts that are normally measured in millions throughout the . ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively .

PAGE 108

108 Table 3 3. Summary statistics Dealscan sample (non investment grade rated firms) . Below Investment Grade Unrated Dealscan Compustat Dealscan Compustat Mean Median Mean Median Mean Diff. Mean Median Mean Median Mean Diff. 3.303 1.624 3.152 1.619 0.151 0.898 0.508 0.790 0.379 0.108 ** Book 0.427 0.386 0.431 0.399 0.003 0.242 0.205 0.218 0.169 0.024 *** Market 0.418 0.381 0.433 0.396 0.015 * 0.230 0.164 0.212 0.138 0.018 *** Market 1.463 1.296 1.420 1.240 0.044 * 1.737 1.461 1.821 1.452 0.084 *** 0.182 0.141 0.147 0.141 0.035 ** 0.127 0.120 0.005 0.106 0.132 *** 0.126 0.118 0.116 0.115 0.009 *** 0.126 0.124 0.094 0.110 0.032 *** 6.682 3.775 6.670 3.613 0.012 29.858 9.337 34.530 9.147 4.672 * Dividend Payer 0.303 0.000 0.278 0.000 0.026 0.291 0.000 0.275 0.000 0.017 0.373 0.318 0.345 0.285 0.028 *** 0.312 0.225 0.264 0.185 0.048 *** Below Investment Grade 1.000 1.000 1.000 1.000 0.000 0.000 0.000 0.000 0.000 0.000 Unrated Firm 0.000 0.000 0.000 0.000 0.000 1.000 1.000 1.000 1.000 0.000 RD Dummy 0.306 0.000 0.330 0.000 0.024 0.342 0.000 0.494 0.000 0.152 *** HHI 0.603 0.530 0.647 0.600 0.044 *** 0.725 0.778 0.767 0.892 0.042 *** Excl90 0.302 0.000 0.378 0.000 0.076 *** 0.492 0.000 0.568 1.000 0.076 *** Bank Debt 0.326 0.253 0.328 0.229 0.002 0.543 0.626 0.493 0.498 0.050 *** Drawn Revolvers 0.114 0.008 0.093 0.000 0.021 *** 0.345 0.155 0.270 0.000 0.075 *** Term Loans 0.212 0.019 0.235 0.024 0.024 ** 0.198 0.000 0.223 0.000 0.025 *** Senior Bonds 0.416 0.381 0.434 0.402 0.017 0.260 0.014 0.260 0.000 0.000 Subord. Bonds 0.215 0.000 0.196 0.000 0.019 * 0.065 0.000 0.076 0.000 0.011 * Capital Leases 0.017 0.000 0.019 0.000 0.002 0.075 0.000 0.115 0.000 0.040 *** Comm. Paper 0.000 0.000 0.000 0.000 0.000 0.001 0.000 0.002 0.000 0.002 *** Other Debt 0.025 0.000 0.024 0.000 0.001 0.056 0.000 0.053 0.000 0.003 This table presents summary statistics for the Compustat sample and for firms that issued loans between 2002 and 2011 from Dealscan. The full sample consists of all firms with at least $100mn in total assets and $10mn in sales (2011 dollars). It is further restricted to firms rated below investment grade ( BB and below ) , and to firms without a credit rating . Variables normally measured in millions throughout the paper have been displayed in billions where **, and * denote significance at the 1%, 5%, and 10% levels, respectively .

PAGE 109

109 Table 3 3. Continued. Below Investment Grade Unrated Dealscan Compustat Dealscan Compustat Mean Median Mean Median Mean Diff. Mean Median Mean Median Mean Diff. Total Senior Debt 0.776 0.998 0.793 1.000 0.017 0.929 1.000 0.918 1.000 0.010 * Total Senior Secured Debt 0.354 0.278 0.399 0.326 0.045 *** 0.498 0.477 0.529 0.622 0.030 ** Total Senior Unsecured Debt 0.422 0.371 0.396 0.288 0.027 * 0.431 0.207 0.392 0.070 0.039 *** Total Subordinated Debt 0.224 0.000 0.206 0.000 0.018 0.071 0.000 0.081 0.000 0.010 * Sen. Sub. Debt 0.182 0.000 0.141 0.000 0.041 *** 0.027 0.000 0.024 0.000 0.004 Subordinated Debt 0.034 0.000 0.057 0.000 0.023 *** 0.038 0.000 0.053 0.000 0.015 *** Junior Debt 0.008 0.000 0.008 0.000 0.001 0.005 0.000 0.005 0.000 0.000 Total Secured Debt 0.366 0.293 0.413 0.343 0.047 *** 0.510 0.506 0.538 0.685 0.028 ** Total Unsecured Debt 0.634 0.708 0.587 0.656 0.047 *** 0.489 0.494 0.461 0.315 0.028 ** Priority HHI 0.634 0.552 0.686 0.674 0.052 *** 0.841 0.988 0.870 1.000 0.029 *** Drawn Revolvers Excl90 0.008 0.000 0.011 0.000 0.003 0.182 0.000 0.142 0.000 0.040 *** Term Loans Excl90 0.060 0.000 0.080 0.000 0.020 ** 0.088 0.000 0.118 0.000 0.030 *** Senior Bonds Excl90 0.171 0.000 0.211 0.000 0.039 *** 0.120 0.000 0.145 0.000 0.025 *** Subord. Bonds Excl90 0.053 0.000 0.063 0.000 0.010 0.025 0.000 0.046 0.000 0.021 *** Capital Leases Excl90 0.003 0.000 0.005 0.000 0.002 0.044 0.000 0.083 0.000 0.040 *** Comm. Paper Excl90 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.000 0.001 ** Other Debt Excl90 0.007 0.000 0.009 0.000 0.002 0.034 0.000 0.034 0.000 0.000 This table presents summary statistics for the Compustat sample and for firms that issued loans between 2002 and 2011 from Dealscan. The full sample consists of all firms with at least $100mn in total assets and $10mn in sales (2011 dollars). It is further restricted to firms rated below investment grade (BB and below), and to firms without a credit rating. Variables normally measured in mill denote significance at the 1%, 5%, and 10% levels, respectively .

PAGE 110

110 Table 3 4. Linear models of syndicate size (investment grade rated firms) . (1) (2) (3) (4) (5) Excl90 0.050 0.005 0.041 (0.89) (0.12) (1.03) Revolver 0.167 *** 0.168 *** (2.82) (2.84) Log(Loan Amt) 0.262 *** 0.263 *** (7.26) (7.38) Log(1+Yrs Maturity) 0.301 *** 0.304 *** (3.09) (3.11) Secured Loan 0.349 *** 0.349 *** (4.26) (4.33) 0.212 *** 0.212 *** 0.079 *** 0.077 *** (8.27) (8.03) (4.60) (4.48) Book 0.223 0.220 0.238 0.206 (0.99) (1.01) (1.14) (0.98) Market 0.118 *** 0.117 *** 0.106 *** 0.103 *** (3.17) (3.20) (4.01) (3.76) 1.184 ** 1.183 ** 0.397 0.386 (2.36) (2.34) (0.84) (0.82) Dividend Payer 0.108 * 0.107 * 0.123 ** 0.116 ** (1.77) (1.80) (2.26) (2.13) 0.239 0.239 0.098 0.097 (1.47) (1.47) (0.76) (0.76) RD Dummy 0.121 0.121 0.082 0.085 (0.98) (0.98) (0.90) (0.95) Constant 2.581 *** 1.137 *** 1.138 *** 0.021 0.027 (59.62) (3.10) (3.09) (0.05) (0.06) Industry FE No Yes Yes Yes Yes Year FE No Yes Yes Yes Yes Adj. R Square 0.000 0.148 0.147 0.394 0.394 Observations 917 917 917 917 917 This table presents linear models where the dependent variable is Log(1+Syndicate Size). The sample consists of all term loans and revolvers by firms with total assets of at least $100mn and $10mn in sales (2011 dollars) and is further restricted to firms with an investment grade credit rating (rated BBB or above). Variables have been winsorized at statistics are presented in parentheses below the corresponding coefficient estimates and standar d errors have been clustered by firm and year. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively .

PAGE 111

111 Table 3 5. Linear models of syndicate size ( non investment grade rated firms) . (1) (2) (3) (4) (5) Excl90 0.124 *** 0.012 0.019 (4.71) (0.43) (0.88) Revolver 0.039 0.038 (0.91) (0.88) Term Loan B 0.319 *** 0.318 *** (5.13) (5.15) Log(Loan Amt) 0.255 *** 0.255 *** (21.94) (21.86) Log(1+Yrs Maturity) 0.245 *** 0.244 *** (6.46) (6.47) Secured Loan 0.018 0.018 (0.57) (0.56) 0.289 *** 0.289 *** 0.117 *** 0.115 *** (13.63) (13.55) (6.59) (6.44) Book 0.238 ** 0.245 ** 0.152 ** 0.162 ** (2.54) (2.51) (2.29) (2.43) Market 0.026 * 0.026 * 0.010 0.011 (1.75) (1.80) (0.80) (0.88) 0.952 *** 0.952 *** 0.485 *** 0.485 *** (4.63) (4.61) (3.03) (3.01) Dividend Payer 0.110 *** 0.110 *** 0.071 ** 0.071 ** (2.92) (2.92) (2.02) (2.02) 0.111 0.111 0.080 0.081 (1.40) (1.40) (1.21) (1.22) Unrated Firm 0.087 * 0.087 * 0.049 0.049 (1.77) (1.78) (1.23) (1.24) RD Dummy 0.081 *** 0.079 *** 0.056 ** 0.054 ** (2.65) (2.73) (2.12) (2.11) Constant 1.914 *** 0.485 ** 0.475 ** 0.711 *** 0.695 *** (52.16) (2.18) (2.14) (3.32) (3.28) Industry FE No Yes Yes Yes Yes Year FE No Yes Yes Yes Yes Adj. R Square 0.007 0.330 0.329 0.448 0.448 Observations 4383 4383 4383 4383 4383 This table presents linear models where the dependent variable is Log( 1+Syndicate Size). The sample consists of all term loans and revolvers by firms with total assets of at least $100mn and $10mn in sales (2011 dollars) and is further restricted to firms rated below investment grade (BB or lower) and firms without a credit rating. Variables have statistics are presented in parentheses below the corresponding coefficient estimates and standard errors have been clustered by firm and year. ***, * *, and * denote significance at the 1%, 5%, and 10% levels, respectively .

PAGE 112

112 Table 3 6. Logit models of institutional loan tranche . (1) (2) (3) (4) (5) Excl90 0.709 *** 0.191 0.165 (5.90) (1.56) (1.30) Log(Deal Amt) 1.301 *** 1.300 *** (12.03) (12.00) 0.116 * 0.103 * 0.701 *** 0.711 *** (1.83) (1.66) (7.37) (7.57) Book 2.393 *** 2.292 *** 2.204 *** 2.118 *** (7.57) (7.52) (7.61) (7.41) Market Book 0.216 *** 0.205 *** 0.298 *** 0.288 *** (3.68) (3.71) (4.01) (4.04) 2.901 *** 2.907 *** 0.626 0.628 (2.74) (2.74) (0.58) (0.57) Dividend Payer 0.497 *** 0.495 *** 0.596 *** 0.595 *** (3.76) (3.80) (3.71) (3.71) 1.651 *** 1.666 *** 1.560 *** 1.570 *** (5.91) (6.09) (5.52) (5.60) Unrated Firm 0.866 *** 0.869 *** 0.886 *** 0.882 *** (6.17) (6.15) (5.50) (5.54) RD Dummy 0.042 0.022 0.083 0.100 (0.28) (0.14) (0.46) (0.56) Constant 1.105 *** 1.554 ** 1.379 ** 3.041 *** 2.894 *** (8.63) (2.50) (2.26) (3.80) (3.74) Industry FE No Yes Yes Yes Yes Year FE No Yes Yes Yes Yes Pseudo R Square 0.018 0.202 0.203 0.305 0.305 Observations 3830 3809 3809 3809 3809 This table presents logit models where the dependent variable takes on a value of 1 if the loan package contains an institutional loan tranche (Term B Loan) and 0 otherwise. The sample consists of all term loans and revolvers by firms with total assets of at least $100mn and $10mn in sales (2011 dollars) and is further restricted to firms without an investment grade credit rating (rated BB or below or unrated). Variables have been statistics are presented in parentheses below the corresponding coefficient estimates and standard errors have been clustered by firm and year. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively .

PAGE 113

113 Table 3 7. Summary statistics for firms with major shifts between public and bank debt . Year Prior to Shift Year of Shift Mean Median N Mean Median N Mean Diff. 5395.194 2870.797 86 5190.790 2732.367 86 204.404 0.069 0.032 86 0.067 0.030 86 0.003 Book 0.342 0.316 86 0.345 0.330 86 0.003 Market 0.328 0.295 86 0.352 0.317 86 0.024 Market 1.436 1.195 86 1.427 1.197 86 0.008 EBITDA 0.207 0.170 86 0.182 0.143 86 0.025 EBITDA 0.134 0.121 86 0.126 0.117 86 0.007 9.174 4.525 86 10.736 4.910 85 1.562 Dividend Payer 0.523 1.000 86 0.512 1.000 86 0.012 0.355 0.267 86 0.349 0.287 86 0.006 Below Investment Grade 0.628 1.000 86 0.640 1.000 86 0.012 Unrated Firm 0.000 0.000 86 0.012 0.000 86 0.012 Downgrade 0.103 0.000 68 0.118 0.000 85 0.015 RD Dummy 0.291 0.000 86 0.279 0.000 86 0.012 Issued Term B in FY 0.105 0.000 86 0.186 0.000 86 0.081 HHI 0.587 0.528 86 0.541 0.479 86 0.045 Excl90 0.221 0.000 86 0.151 0.000 86 0.070 Bank Debt 0.212 0.131 86 0.460 0.381 86 0.249 *** Public Debt (Rated) 0.755 0.828 86 0.520 0.596 86 0.235 *** Drawn Revolvers 0.117 0.000 86 0.255 0.158 86 0.138 *** Term Loans 0.095 0.000 86 0.205 0.045 86 0.110 *** Senior Bonds 0.544 0.678 86 0.421 0.521 86 0.123 ** Subord. Bonds 0.185 0.000 86 0.091 0.000 86 0.094 ** Capital Leases 0.009 0.000 86 0.007 0.000 86 0.002 Comm. Paper 0.026 0.000 86 0.008 0.000 86 0.018 * Other Debt 0.024 0.000 86 0.013 0.000 86 0.011 Total Senior Debt 0.817 1.000 83 0.915 1.000 84 0.097 *** Total Senior Secured Debt 0.211 0.084 83 0.326 0.171 84 0.114 ** Total Senior Unsecured Debt 0.606 0.810 83 0.589 0.702 84 0.017 Total Subordinated Debt 0.183 0.000 83 0.085 0.000 84 0.098 *** Sen. Sub. Debt 0.139 0.000 83 0.063 0.000 84 0.077 ** Subordinated Debt 0.043 0.000 83 0.020 0.000 84 0.023 Junior Debt 0.000 0.000 83 0.002 0.000 84 0.002 Total Secured Debt 0.211 0.084 83 0.328 0.178 84 0.117 ** Total Unsecured Debt 0.789 0.912 83 0.672 0.817 84 0.117 ** Priority HHI 0.697 0.733 83 0.726 0.823 84 0.030 Drawn Revolvers Excl90 0.000 0.000 86 0.070 0.000 86 0.070 ** Term Loans Excl90 0.000 0.000 86 0.058 0.000 86 0.058 ** Senior Bonds Excl90 0.186 0.000 86 0.023 0.000 86 0.163 *** Subord. Bonds Excl90 0.035 0.000 86 0.000 0.000 86 0.035 * Capital Leases Excl90 0.000 0.000 86 0.000 0.000 86 0.000 Comm. Paper Excl90 0.000 0.000 86 0.000 0.000 86 0.000 Other Debt Excl90 0.000 0.000 86 0.000 0.000 86 0.000 This table presents summary statistics for firms that experience major shifts between public and private debt while holding leverage stable (2.5% band). Firms are further restricted to have at least $100mn in total assets and $10mn in sales (2011 dollars). Variables have been winsorized at the 1% and . ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively .

PAGE 114

114 Table 3 8. Logit models of major shifts in debt structure . (1) (2) (3) (4) (5) (6) 3.734 *** 3.897 *** 4.031 *** (2.65) (2.70) (2.64) Neg EBITDA/Sales Gr 1Y 0.725 *** 0.714 *** 0.701 *** (3.12) (3.07) (2.97) 0.057 0.045 0.038 0.034 0.025 0.017 (0.54) (0.43) (0.36) (0.32) (0.24) (0.16) L1 Book 0.505 0.624 0.676 0.390 0.501 0.532 (0.77) (0.94) (1.01) (0.59) (0.76) (0.79) L1 Market 0.196 0.250 0.301 0.167 0.220 0.259 (1.03) (1.25) (1.46) (0.88) (1.11) (1.26) L1 Dividend Payer 0.091 0.054 0.086 0.072 0.038 0.068 (0.35) (0.21) (0.33) (0.28) (0.15) (0.26) 0.053 0.097 0.124 0.158 0.204 0.234 (0.11) (0.20) (0.26) (0.33) (0.42) (0.48) L1 IG Rating Dummy 0.190 0.150 0.123 0.261 0.219 0.198 (0.61) (0.48) (0.39) (0.84) (0.70) (0.62) L1 RD Dummy 0.534 ** 0.516 ** 0.531 ** 0.527 ** 0.510 ** 0.524 ** (2.05) (1.98) (2.03) (2.03) (1.97) (2.01) Term Structure (Avg12m) 0.198 ** 0.135 0.189 ** 0.008 (2.29) (0.20) (2.18) (0.01) Constant 2.232 ** 1.984 ** 2.403 2.808 *** 2.539 *** 3.319 (2.44) (2.15) (1.11) (3.02) (2.70) (1.54) Year FE No No Yes No No Yes Pseudo R Square 0.029 0.036 0.056 0.035 0.041 0.060 Observations 1853 1853 1853 1853 1853 1853 This table presents logit models for the likelihood that firms that experience major shifts between public and private debt while holding leverage stable (2.5% band). The sample consists of rated firms with stable leverage that have at least $100mn in tota l assets and $10mn in sales (2011 dollars). The dependent variable takes on a value of one if the firm shifted from public debt to bank debt, and zero otherwise. EBITDA to Sales Growth 1Y is defined as . Neg EBITDA/Sales Gr 1Y is an indicator variable that takes on a value of one if the firm's EBITDA to Sales Growth 1Y is negative. The remaining variables are lagged by one year. Variables have been winsorized at the 1% t statistics are presented in parentheses below the corre sponding coefficient estimates. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 115

115 Table 3 9. Instrumental variable probit model of major shifts in debt structure. (1) (2) OLS IV Probit Avg. TXWEB Return 12mth 1.120 ** (2.22) Avg. EP U Return 12mth 0.077 (0.89) 10.384 *** (3.34) 0.002 0.036 (1.00) (1.04) L1 Book Debt 0.002 0.148 (0.15) (0.59) L1 Market 0.005 * 0.004 (1.68) (0.06) L1 Dividend Payer 0.005 0.033 (1.01) (0.37) 0.002 0.048 (0.26) (0.30) L1 IG Rating Dummy 0.000 0.044 (0.06) (0.41) L1 RD Dummy 0.003 0.122 (0.65) (0.91) Constant 0.007 0.722 (0.43) (1.12) Adj. R Square 0.002 Observations 1853 1853 This table presents an instrumental variable probit model for the likelihood that firms experience a major shift away from public debt towards more bank debt while holding leverage constant (2.5% band). Column (1) presents an OLS model where the dependent variable is EBITDA to Sales Growth 1Y, and Column ( 2) presents the results of the instrumental variable probit model where EBITDA to Sales Growth 1Y has been instrumented with Avg. TXWEB Return 12mth and Avg. EPU Return 12mth. The remaining variables are lagged by one year. Variables have been winsorized a t the 1% statistics are presented in parentheses below the corre sponding coefficient estimates . ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 116

116 Table 3 10. Fraction o f floating rate debt as a proxy for fraction of bank debt. (1) (2) Base Model Float Model 0.604 *** (67.16) 0.051 *** 0.043 *** (12.35) (12.69) Market 0.038 *** 0.013 *** (9.48) (3.91) Book 0.097 *** 0.149 *** (4.93) (9.20) 0.024 0.052 *** (1.39) (3.63) RD Dummy 0.095 *** 0.046 *** (10.41) (6.03) Dividend Payer 0.010 0.004 (1.08) (0.46) EBITDA to T otal A ssets 0.453 *** 0.158 *** (12.03) (5.05) Unrated 0.151 *** 0.085 *** (14.08) (9.56) Constant 0.795 *** 0.564 *** (23.99) (20.41) Adj R Square 0.103 0.387 Observations 9724 9724 This table presents linear models where the dependent variable is the fraction of bank debt to total debt obtained from Capital IQ. The sample consists of all firms that have total assets of at least $100mn and $10mn in sales (in 2011 dollars). The sample is further restricted to firms that do not have an investment grade credit rating (unrated or rated BB and below). Frac. Float Debt is the fraction of long term debt tied to the prime rate and is obtained from Compustat. Variables have been winsorized at t he 1% and statistics are presented in parentheses below the corresponding coefficient estimates. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 117

117 Table 3 11. Summary stat istics for firms in distressed industries. Distress Sample Non Distress Sample Mean Median N Mean Median N Mean Diff. 1668.187 649.849 629 1508.549 581.161 7442 159.638 Book 0.309 0.265 629 0.352 0.310 7442 0.043 *** Market 1.762 1.331 526 1.795 1.381 6031 0.033 EBITDA 0.092 0.107 629 0.111 0.125 7442 0.019 ** 0.102 0.109 629 0.120 0.124 7442 0.018 *** 25.532 4.318 592 28.796 4.792 7177 3.263 0.091 0.042 629 0.152 0.044 7442 0.061 *** Dividend Payer 0.254 0.000 626 0.274 0.000 7425 0.020 0.260 0.204 629 0.339 0.267 7442 0.079 *** Below Investment Grade 0.345 0.000 629 0.315 0.000 7442 0.030 Unrated 0.655 1.000 629 0.685 1.000 7442 0.030 RD Dummy 0.507 1.000 629 0.343 0.000 7442 0.164 *** 0.312 0.128 629 0.362 0.260 7442 0.049 *** TermB Issued in Prior 3Y 0.149 0.000 629 0.165 0.000 7442 0.015 This table presents summary statistics for firms in distressed industries for the year prior to distress, and compares them to firm years by firms that have never experienced industry wide distress. The sample is further restricted to firms which have at l east $100mn in total assets and $10mn in sales (2011 dollars), do not have an investment grade credit rating (unrated or rated BB and below), and that were not in Chapter 11 during the year of industry wide distress. An industry is defined as distressed if the median annual stock return was less than 20%, the median annual sales growth was negative, and if the industry consisted of at least 5 firms in the corresponding year in Compustat. Variables have been winsorized at the 1% and 99% level where indicate d ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

PAGE 118

118 Table 3 12. Logit models of operating performance for firms in distressed industries. (1) (2) (3) (4) (5) (6) EBITDA EBITDA CAPX CAPX Asset Asset 0.653 ** 0.574 ** 0.463 * 0.580 ** 0.274 0.349 (2.48) (2.07) (1.80) (2.12) (1.08) (1.30) 0.026 0.859 0.143 (0.03) (1.00) (0.17) Fract Float x 0.987 1.370 ** 0.324 0.131 1.060 * 1.340 * (1.63) (1.97) (0.54) (0.19) (1.70) (1.89) 0.987 1.275 0.962 (1.02) (1.23) (0.97) Fin. Crisis 0.370 0.342 0.085 0.095 0.093 0.105 (1.22) (1.11) (0.28) (0.31) (0.31) (0.35) TermB Issued in Prior 3Y 0.428 0.133 0.270 (0.88) (0.27) (0.56) Unrated 0.049 0.124 0.070 0.008 0.257 0.228 (0.22) (0.54) (0.32) (0.04) (1.17) (1.02) Book 0.303 0.538 0.309 0.502 0.039 0.037 (0.77) (1.31) (0.80) (1.26) (0.10) (0.09) 0.153 * 0.119 0.136 0.119 0.025 0.014 (1.78) (1.35) (1.60) (1.36) (0.30) (0.17) 0.464 0.464 0.785 0.725 0.786 * 0.760 * (1.03) (1.01) (1.52) (1.39) (1.80) (1.71) RD Dummy 0.257 0.234 0.085 0.091 0.216 0.222 (1.46) (1.33) (0.49) (0.52) (1.25) (1.28) EBITDA 2.899 *** 2.913 *** (4.76) (4.78) 2.507 *** 2.495 *** (3.05) (3.05) Constant 1.264 * 1.040 0.479 0.369 0.595 0.644 (1.82) (1.47) (0.70) (0.52) (0.87) (0.93) Pseudo R Square 0.047 0.053 0.029 0.036 0.013 0.015 Observations 629 629 629 629 629 629 This table presents logit models of operating performance for firms in distressed industries. In Columns (1) and (2) the dependent variable takes on a value of one if the firm experienced above industry median EBITDA to Sales Growth over the two years foll owing industry wide distress, measured as adjusted by the corresponding industry median. In Columns (3) and (4) the dependent variable takes on a value of one if the firm experienced above indus try median CAPX to Sales Growth over the two years following the industry downturn, measured as adjusted by the corresponding industry median. In Columns (5) and (6) the dependent variable takes on a va lue of one if the firm experienced above industry median asset growth over the two years following industry wide distress, measured as adjusted by the corresponding industry median. All independent variable s with the exception of Fin. Crisis are measured in the year prior to industry distress. Fin. Crisis is an indicator variable that takes on a value of one if the year of industry distress was 2008 or 2009. Variables have been winsorized at the 1% and 99% l statistics are presented in parentheses below the corresponding coefficient estimates. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively .

PAGE 119

119 CHAPTER 4 CONCLUSION In this dissertation we analyze the determinants of long term corporate debt issues as well as the relationship between debt specialization, creditor concentration, and financial distress costs. In the first part of this dissertation we analyze how variati on in credit market conditions, and in particular variation in the supply of long term government bonds, explain long term corporate issues with maturities of 20 years or more. Overall, the results in Chapter 2 suggest that market conditions, and particul arly the supply of long term Treasuries, are important determinants of long term corporate debt issues for highly rated issuers. Moreover, we document that gap filling is restricted to a relatively narrow segment of the term structure. Specifically, we fin d a negative and significant relationship between debt maturity choice and the supply of long term government bonds, a result which is robust to our instrumental variables estimation as well as the natural experiment of the suspension of 30 year Treasuries in 2001, but no significant relationship between debt maturity choice and the supply of short term government bonds. Additionally, we provide evidence that for highly rated issuers corporate gap filling behavior does not result in a substitution between s hort term and long term debt but affects their overall propensity to borrow. In Chapter 3 we examine the relationship between debt specialization and creditor concentration and the extent to which they are related to financial distress costs. We argue that distress costs are most likely mitigated when debt specialization is combined with a greater concentration of lenders. Overall, our findings are consistent with the notion that only when it is accompanied by greater creditor concentration does

PAGE 120

120 the exclusi ve reliance on one debt type help to mitigate distress costs. Specifically, we find no significant correlation between measures of debt specialization and measures of creditor concentration, but we document that companies with large declines in operating p erformance are significantly more likely to substantially shift their debt structure away from dispersedly held public debt towards more tightly held intermediated bank debt. This finding is further confirmed in an instrumental variables probit model, ther eby mitigating concerns about the endogeneity of our results. Moreover, these major shifts towards more concentrated forms of debt seem to occur well before states of financial distress. Furthermore, our results suggest that firms specializing in concentra ted intermediated bank debt recover faster from an industry wide downturn than firms that specialize in other forms of debt or those that do not specialize at all. However, our results also indicate that the benefits from relying exclusively on concentrate d intermediated bank debt might come at the cost of leaving these firms more vulnerable to solvency and liquidity shocks in the banking sector.

PAGE 121

121 APPENDIX DATA AND VARIABLE DESCRIPTIONS Data and Variable Descriptions for Chapter 2 The basic dataset of corporate debt issues used in Chapter 2 is constructed from the Thompson Reuters LPC Dealscan and Thomson Reuters SDC Platinum databases. It contains U.S. dollar denominated debt issues and bank borrowings by public U.S. firms between 1987 and 2009. Our choice of this time period is driven by the fact that (2008) for a discussion). We focus on U.S. companies because of homogeneity with respect to tax laws, and because we would expect U.S. companies to be more issuers. Furthermore, we exclude financial firms from our sample which we identify by the SIC codes 6000 through 699 9. All dollar amounts in this study have been adjusted index (all urban consumers), unless specifically mentioned otherwise. A detailed description of the variables used in Chapter 2 as well as their short hand notation in the tables can be found in Table A 1. Data on Individual Debt Issues We collected information on individual corporate debt issues from Thompson Reuters LPC Dealscan and Thomson Reuters SDC Platinum datab ases. Dealscan provides information on corporate loans, also referred to as facilities, originated by bank and certain non bank lenders (such as life insurance companies, hedge funds and other institutional investors). Since these loans are often negotiate d and grouped together as part of an entire loan package, we only focus on individual facilities from completed or

PAGE 122

122 closed deals and further restrict our sample to loans with non missing (positive) loan amounts. From SDC we obtained information on new debt issues of non convertible debt securities, debt shelf registrations, U.S. Rule 144A non convertible debt, and medium term note programs. Using a combination of data fields available from SDC as well as text searches on the issue description, we exclude fr om our analysis asset or mortgage backed debt, secured debt, pass through securities, equipment trust certificates, lease obligations, convertible debt, preferred stock that has been misclassified as debt, equity linked certificates, and perpetual debt. W e exclude these issues because the maturity of issue is likely to match the economic life of the underlying collateral (for example the cash flows of the underlying assets determine the life of the pass through securities). We restrict our sample to debt i ssues where the issue amount is positive and not missing. 1 While Dealscan primarily covers corporate loans, a small portion of the deals from Dealscan consists of private placements and publicly underwritten bonds. To avoid duplication, we manually compare SDC with all private placements and publicly underwritten bonds from Dealscan and discard duplicate entries from Dealscan. We consider an entry in Dealscan to be a duplicate if there is an observation in SDC with issue and maturity dates within a month of the corresponding dates in Dealscan and if the issue amount in SDC is of similar magnitude to the one in Dealscan. 2 1 We define issue amount or deal amount as the loan amount, if the observation is from Dealscan, and as the total proceeds raised (including over allotment options) if the observation is from SDC. 2 It should be noted however, that the overlap of the two databases is relatively small . Out of over 32,000 deals from Dealscan and SDC only about 560 were classified by us as being duplicates.

PAGE 123

123 Firm Level Data Annual File and merge them with our dataset on debt issuances. We merged Dealscan with Compustat using a link file program used by Chava and Roberts (2008), which we updated to cover our longer sample period. For each debt issue in our combined Dealscan and SDC sample we obtain financial information concer ning the issuer from Compustat for the fiscal year end immediately prior to the date of debt issuance and exclude observations for which the most recent Compustat data are more than 367 days from the issuance date. A detailed description of the firm level variables used and their construction can be found in Table A 1. Data on U.S. Government Debt Maturity Structure We collect information on outstanding marketable U.S. government debt from the CRSP monthly treasury database. CRSP reports monthly pricing inf ormation, the remaining outstanding principal amounts, and the remaining time to maturity for most outstanding marketable treasury securities over our sample period. As CRSP provides outstanding principal amounts that have been adjusted for repurchases and follow on offerings of existing securities by the Treasury, we are able to construct the primary variables of interest for this study on a monthly basis over our sample period. Specifically, we construct the fraction of outstanding Treasury debt maturing in over 20 years (TSY20), the fraction of Treasury debt maturing in over one year (TSY1), and the value weighted average maturity of outstanding principal payments (TSYMAT). Unfortunately, for a small number of observations during our sample period the out standing principal amounts are missing in the CRSP monthly treasury database. Therefore, where appropriate, we follow Greenwood and Vayanos (2008) and replace

PAGE 124

124 missing principal amounts with the outstanding amounts observed in the previous month. Credit Mar ket Data We use several measures of credit market conditions. The first is the spread (or year BBB and AAA rated corporate bond indices, which we obtain on a monthly basis from the Federal Reserve with a maturity of 10 years, which we obtain on a monthly basis from Bloomberg from 1996 t hrough 2007. We also control for the term structure of interest rates (or term premium) by including a variable representing the percentage difference between the yields of 10 year and 6 month Treasuries. We obtain these data on a monthly basis fro m the year note is the longest maturity that was regularly issued by the Treasury throughout our sample period. For example, regular 20 year offerings were eliminated in 1986 and issues of 30 year bonds completely suspended from 2002 to 2006.

PAGE 125

125 Table A 1. Variable definitions for Chapter 2. Variable Label Description Bank Debt Dummy Indicator variable; Takes on a value of one if a debt issue is bank debt. Deal Amount Loan amount if the issue is in Dealscan, and total proceeds raised (incl. over allotment option) if the observation is in SDC. Measured in $mn. Log(Deal Amount) Natural logarithm of "Deal Amount". SDC Call Flag Indicator variable; Takes on a value of one if a debt issue is callable according to SDC. Only pertains to observations from SDC. Years to Final Maturity Years until final maturity of a debt issue. A Rating or Higher Indicator variable; Takes on a value of one if the firm has a long term credit rating by S& P of "A " or higher. Measured by the Compustat variable "splticrm". Book Debt Ratio Book debt ratio. Measured by the Compustat variables ((dltt+dlc)/at). CAPX to Total Assets Ratio of capital expenditures to total assets. Measured by the Compustat variab les (capx/at). Com. Paper Rating Dummy Indicator variable for the presence of a commercial paper rating. Takes on a value of one if the firm has a short term credit rating by S&P. Measured by the Compustat variable "spsticrm". Dividend Dummy Indicator variable; Takes on a value of one if the firm declared dividends on common stock. Measured by the Compustat variable "dvc". EBIT to Total Assets Earnings before interest and taxes, scaled by total assets. Measured by Compustat variables ((ib+xint+txt)/at) . Firm Age Years from the ipo date (Compustat variable "ipodate") or years from the first date in CRSP if "ipodate" is missing. Fraction of Debt due 1Y Fraction of debt due within 1 year to total debt. Measured by Compustat variables ((dlc)/(dlc+dltt)). Fraction of Debt due 2Y Fraction of debt due within2 years to total debt. Measured by Compustat variables ((dlc+dd2)/(dlc+dltt)). Fraction of Debt due 3Y Fraction of debt due within 3 years to total debt. Measured by Compustat variables ((dlc+dd2+dd3)/(d lc+dltt)). Fraction of Debt due 4Y Fraction of debt due within 4 years to total debt. Measured by Compustat variables ((dlc+dd2+dd3+dd4)/(dlc+dltt)). Fraction of Debt due 5Y Fraction of debt due within 5 years to total debt. Measured by Compustat variables ((dlc+dd2+dd3+dd4+dd5)/(dlc+dltt)). Industry FE Industry fixed effects according to Fama French industry classification. IG Rating Dummy Indicator variable; Takes on a value of one if the firm has a long term credit rating by S&P of "BBB " or h igher or if it has a short term credit rating by S&P of "A 3" or higher. Measured by the Compustat variables "splticrm" and "spsticrm". Log(MV of Equity) Natural logarithm of "Market Value of Equity". Log(Sale) Natural logarithm of "Sales". Market to Book Ratio Market to book ratio. Measured by Compustat variables ((lt txditc+prcc_f*csho+preferred)/at). Where "preferred" is measured by "pstkl" or "pstkrv" or "pstk". Market to Book Same as "Market to Book Ratio". Simply abbreviation. Market Value of Equity Market value of equity. Measured by Compustat variables (prcc_f*csho). MV Equity Same as "Market Value of Equity". Simply an abbreviation. Market Debt Ratio Market debt ratio. Measured by the Compustat variables ((dltt+dlc)/(dltt+dlc+prcc_f*csho )).

PAGE 126

126 Table A 1. Continued. Variable Label Description Past LT Issuer Flag Indicator variable; Takes on a value of one if the firm issued debt with at least 20 years to maturity in prior years. Calculated from our sample of debt issuances. PPE to TA Property, plant and equipment, scaled by total assets. Measured by Compustat variables (ppent/at). R&D Missing Indicator variable; Takes on a value one if research and development expense in Compustat ("xrd") is missing. R&D to TA Research and develop ment expense, scaled by total assets. Measured by Compustat variables (xrd/at) and set to zero if (xrd) missing. STD EBIT Growth 2DSIC Industy earnings volatility measure; Measured as the annual standard deviation of growth in "EBIT to Total Assets" by 2 digit SIC codes. Value Weighted Asset Maturity Asset maturity measured as Stohs and Mauer (1996); Book value weighted average maturity of curr ent assets and long term assets. Measured by Compustat variables (act/(act+ppent))*(act/cogs))+(ppent/(act+ppent))*(ppent/dp) VW Asset Maturity Same as "Value Weighted Asset Maturity". Simply abbreviation. Value Weighted Asset Maturity (No Cash) Same definition as "Value Weighted Asset Maturity" but with cash and short term investments ("che") subtracted from current assets ("act"). Year [2002,...) Dummy Indicator variable. Takes on a value of one if the year of the observation is between 2002 and 200 9. 5 Year FE Time period fixed effects. Indicator variables for the following time periods: 1991 1995, 1996 2000, 2001 2005, 2006 2009 with 1987 1990 being the omitted category. A Rating x Moodys BBB AAA30Y Avg. 12m Interaction term between "A Rating or Higher" and "Moodys BBB AAA30Y Avg. 12m" A Rating x Termstructure 10y 6mth Avg. 12m Interaction term between "A Rating or Higher" and "Termstructure 10y 6mth Avg. 12m" A Rating x TSY20 Avg. 12m Interaction term between "A Rating or Higher" and "TSY20 Avg . 12m" Moodys BBB AAA 30Y Difference between the percentage yields of Moody's 30 year BBB and AAA rated corporate bond indices measured monthly. Moodys BBB AAA30Y Avg. 12m 12 month average of the difference between the percentage yields of Moody's 30 year BBB and AAA rated corporate bond indices measured monthly. Moodys BBB AAA 30Y x Bonds "Moodys BBB AAA 30Y" interacted with an indicator variable for whether a debt issue is a bond. Moodys BBB AAA 30Y x No Call "Moodys BBB AAA 30Y" interacted wi th an indicator variable for whether a debt issue non callable (as measured by SDC). SP 10Y Spread BB BBB Difference between the percentage yields of S&P Creditweek Corporate Industrial Bond Indices for BB and BBB rated bonds. Termstructure 10y 6mth Difference between the percentage yields of 10 year and 6 month treasury securities measured monthly. Termstructure 10y 6mth Avg. 12m 12 month average of the difference between the percentage yields of 10 year and 6 month treasury securities measured monthly. Termstructure 10y 6mth x Bonds "Termstructure 10y 6mth" interacted with an indicator variable for whether a debt issue is a bond. Termstructure 10y 6mth x No Call "Termstructure 10y 6mth" interacted with an indicator variable for whether a debt issue non callable (as measured by SDC). Total Assets Total assets. Measured by Compustat variable "at". Total GDP 4Q Growth Growth in real GDP over the past 4 quarters (fraction) meas ured quarterly.

PAGE 127

127 Table A 1. Continued. Variable Label Description TSY1 Fraction of Treasury debt maturing in over a year measured monthly. TSY1 Avg. 12m 12 month average of the fraction of Treasury debt maturing in over a year measured monthly. TSY20 Fraction of Treasury debt maturing in over 20 years measured monthly. TSY20 Avg. 12m 12 month average of the fraction of Treasury debt maturing in over 20 years measured monthly. TSY20 x Bonds "TSY20" interacted with an indicator variable for whether a debt issue is a bond. TSY20 x No Call "TSY20" interacted with an indicator variable for whether a debt issue non callable (as measured by SDC). TSYMAT Weighted average maturity of Treasury debt. Qtr. Mean Moodys BBB AAA 30Y Quarterly average of "Moodys BBB AAA 30Y". Qtr. Mean Termstructure 10y 6mth Quarterly average of "Termstructure 10y 6mth". Qtr. Mean TSY20 Quarterly average of "TSY20".

PAGE 128

128 Variable Descriptions for Chapter 3 Table A 2. Variable definitions for Chapter 3. Variable Label Description Below Investment Grade Indicator variable that takes on a value of one if a firm has a S&P long term credit rating of BB or below and 0 if unrated or rated BBB and above. Measured by Compustat variable "splticrm". Book Debt Ratio Book debt ratio. Measured by the Compustat variables ((dltt+dlc)/at). CAPX to Sales Ratio of capital expenditures to sales. Measured by Compustat variables (capx/sale). Cash to Total Assets Ratio of cash and short term investments to total assets. Measured by Compustat variables (che/at). Dividend Payer Indicator variable; Takes on a value of one if the firm declared dividends on common stock. Measured by the Compustat variable "dvc". Downgrade Indicator variable that takes on a valu e of one if a firm was downgraded compared to the prior fiscal year. Measured by Compustat variable "splticrm". EBITDA to Interest Expense Interest coverage ratio. Measured by Compustat variables (ebitda/xint). EBITDA to Sales Ratio of EBITDA to sales. M easured by Compustat variables (ebitda/sale) EBITDA to Sales Growth 1Y One year growth in EBITDA to sales ratio. Measured as EBITDA to Total Assets Ratio of EBITDA to total assets. Measured by Compustat variables (ebitda/at). F1 EBITDA to Sales Growth 2Y Two year growth in EBITDA to sales ratio. Measured as Float x TermB 3Y x FinCrisis Tripple interaction term between "Fract Float Rate Debt", "TermB Issued in Prior 3Y" and "Fin. Crisis". Fract Float Rate Debt Fraction of floating rate debt (tied to prime rate) to total debt. Measured by Compustat variables (dltp/(dlc+dltt)). Fract Float x FinCrisis Interaction term between "Fract Float Rate Debt" and "Fin. Crisis". Fract Float x TermB 3Y Interaction term between "Fract Float Rate Debt" and "TermB Issued in Prior 3Y". IG Rating Dummy Indicator variable that takes on a value of one if a firm has a S&P long term credit rating of BBB or above and 0 i f unrated or rated BB and below. Measured by Compustat variable "splticrm". Industry FE Industry fixed effects according to Fama French industry classification. Issued Term B in FY Indicator variable that takes on a value of one if the firm issued a term loan B in the corresponding fiscal year. Loan issuance is obtained from Dealscan and I consider term loans to be term B loans if the are modified with any letter other than 'A'. Log (Total Assets) Natural logarithm of 'Total Assets'. Market Book Ratio Market to book ratio. Measured by Compustat variables ((lt txditc+prcc_f*csho)/at). Market Debt Ratio Market debt ratio. Measured by the Compustat variables ((dltt+dlc)/(dltt+dlc+prcc_ f*csho)). Neg EBITDA/Sales Gr 1Y Indicator variable that takes on a value of one if "EBITDA to Sales Growth 1Y" is less than 0. PPE to Total Assets Ratio of net property plant and equipment to total assets. Measured by Compustat variables (ppent/at). RD Dummy Indicator variable that takes on a value of one if the firm has positive reseach and development expenses, as measured by "xrd", and zero if "xrd" is zero or missing.

PAGE 129

129 Table A 2. Continued. Variable Label Description T ermB Issued in Prior 3Y Indicator variable that takes on a value of one if the firm issued a term loan B in the current fiscal year or the previous two fiscal years. Loan issuance is obtained from Dealscan and I consider term loans to be term B loans if the a re modified with any letter other than 'A'. Total Assets Total assets. Measured by Compustat variable "at". Unrated Indicator variable that takes on a value of one if the firm did not have a long term credit rating by S&P as measured by the Compustat variable "splticrm". Bank Debt Fraction of bank debt to total debt. Sum of "Term Loans" and "Drawn Revolvers". Capital Leases Fraction of capital leases to total debt. Capital Leases Excl90 Indicator variable that takes on a value of one if the firm obt ained at least 90% of its total debt from capital leases. Comm. Paper Fraction of commercial paper to total debt. Comm. Paper Excl90 Indicator variable that takes on a value of one if the firm obtained at least 90% of its total debt from commercial paper . Drawn Revolvers Fraction of drawn credit lines to total debt. Drawn Revolvers Excl90 Indicator variable that takes on a value of one if the firm obtained at least 90% of its total debt from drawn revolvers. Excl90 Indicator variable if a firm receives 90% of its total debt from one of the following seven categories: term loans, drawn credit lines, senior bonds, subordinated bonds, commercial paper, capital leases, other debt. HHI Normalized Herfindahl Hirschman Index defined as in Colla et al. (2013) and in Section 2 of the paper. Scaled between 0 and 1. Other Debt Fraction of other debt to total debt. Other Debt Excl90 Indicator variable that takes on a value of one if the firm obtained at least 90% of its total debt from other debt. Public Debt (R ated) For rated firms only. Fraction of public debt to total debt, constructed as the sum of commercial paper, senior bonds and subordinated bonds. Senior Bonds Fraction of senior bonds to total debt. Senior Bonds Excl90 Indicator variable that takes on a value of one if the firm obtained at least 90% of its total debt from senior bonds. Subord. Bonds Fraction of subordinated bonds to total debt. Subord. Bonds Excl90 Indicator variable that takes on a value of one if the firm obtained at least 90% of it s total debt from subordinated bonds. Term Loans Fraction of term loans to total debt Term Loans Excl90 Indicator variable that takes on a value of one if the firm obtained at least 90% of its total debt from term loans. Junior Debt Fraction of junior d ebt to total debt. Priority HHI Normalized Herfindahl Hirschman Index, scaled between 0 and 1. Constructed from the following categories: total senior secured debt, total senior unsecured debt, senior subordinated debt, subordinated debt, junior debt. Sen. Sub. Debt Fraction of senior subordinated debt to total debt. Subordinated Debt Fraction of subordinated debt to total debt. Total Secured Debt Fraction of secured debt to total debt. Total Senior Debt Fraction of total senior debt to total debt. S um of "Total Senior Secured Debt" and "Total Senior Unsecured Debt". Total Senior Secured Debt Fraction of senior secured debt to total debt. Total Senior Unsecured Debt Fraction of senior unsecured debt to total debt. Total Subordinated Debt Fraction of total subordinated debt. Sum of "Subordinated Debt" and "Sen. Sub. Debt". Total Unsecured Debt Fraction of total unsecured debt to total debt.

PAGE 130

130 Table A 2. Continued. Variable Label Description Log(1+Yrs Maturity) Natural logarithm of the one plus the years to maturity of the loan tranche. Log(Deal Amt) Natural logarithm of the total deal amount of the entire loan package. Log(Loan Amt) Natural logartihm of the amount of the loan tranche. Revolver Indicator variable that takes on a value of one if the loan tranche is a credit line. Secured Loan Indicator variable that takes on a value of one if the loan is secured. Term Loan B Indicator variable that takes on a value of one if the loan is a term B loan. I consider term loans to be term B loans if the are modified with any letter other than 'A'. Fin. Crisis Indicator variable that takes on a value of one if the year of distress in 2008 or 2009. Term Structure (Avg12m) Annual average of the monthly difference between the percentage yields on 10 year and 6 month Treasuries. Year FE Year fixed effects. Avg. EPU Return 12mth Annual average of monthly returns on the economic policy incertainty Index (measured in decimals). Avg. TXWEB Return 12mth Annual average of monthly returns on the trade weighted U.S. Dollar Index (measured in decimals).

PAGE 131

131 LIST OF REFERENCES Ahrend, Rudiger, Pietro A. Catte , and Robert Price, 2006, Factors behind low long term interest rates, OECD Working Paper No. 2006/18 . Andrade, Gregor, and Steven N. Kaplan, 1998, How costly is financial (not economic) distress? Evidence from highly leveraged transactions that became di stressed, Journal of Finance 53, 1443 1493. Asquith, Paul, Robert Gertner, and David Scharfstein, 1994, Anatomy of financial distress: an examination of junk bond issuers, Quarterly Journal of Economics 109, 625 658. Baker, Malcolm, Robin Greenwood, and Jeffrey Wurgler, 2003, The maturity of debt issues and predictable variation in bond returns, Journal of Financial Economics 70, 261 291. Baker, Malcolm, and Jeffrey Wurgler, 2002, Market timing and capital structure, Journal of Finance 57, 1 32. Baker, Scott R., Nicholas Bloom, and Steven J. Davis, 2013, Measuring economic policy uncertainty, Working Paper . Barclay, Michael J., and Clifford W. J. Smith, 1995, The maturity structure of corporate debt, Journal of Finan ce 50, 609 631. Board of Governors of the Federal Reserve System, 2009, Press Release (March 18) . Cameron, Collin A., and Pravin K. Trivedi, 2005, Microeconomi c s methods and applications (Cambridge University Press). Campbell, John Y., Andrew W. Lo, a nd A. C. MacKinlay, 1997, The econometrics of financial markets (Princeton University Press). Chava, Sudheer, and Michael R. Roberts, 2008, How does financing impact investment? The role of debt covenants, Journal of Finance 63, 2085 2121. CNN Money, 200 1, U.S. kills 30 ye ar bond, CNN Money (October 31) . Code of Federal Regulations, Title 17: Section 402.2 capital requirements for registered government securities brokers and dealers . Colla, Paolo, Filippo Ippolito, and Kai Li, 2013, Debt specializatio n, Journal of Finance 68, 2117 2141. Copeland, Adam, Antoine Martin, and Michael Walker, 2010, The tri party repo market before the 2010 reforms, Staff Report, F ederal Reserve Bank of New York .

PAGE 132

132 Cornett, Marcia M., Jamie J. McNutt, Philip E. Strahan, and Hassan Tehranian, 2011, Liquidity risk managment and credit supply in the financial crisis, Journal of Financial Economics 101, 297 312. Demiroglu, Cem, and Christopher M. James, 2013, "Bank" loan owner sh ip and troubled deb t restructurings, Working Paper . Diamond, Douglas W., 1991, Debt maturity structure and liquidity risk, Quarterly Journal of Economics 106, 709 737. The Economist, 2001, Cut sh ort, The Economist (November 1) . Gertner, Robert, and David Scharfstein, 1991, A theory of workouts and the e ffects of reorganization law, Journal of Finance 46, 1189 1222. Gilson, Stuart C., Kose John, and Larry H. P. Lang, 1990, Troubled debt restructurings: an empirical study of private reorganization of firms in default, Journal of Financial Economics 27, 31 5 353. Greenwood, Robin, Samuel Hanson, and Jeremy C. Stein, 2010, A gap filling theory of corporate debt maturity choice, Journal of Finance 65, 993 1028. Greenwood, Robin, and Dimitry Vayanos, 2008, Bond supply and excess b ond returns, NBER Working Pap er . Greenwood, Robin, and Dimitry Vayanos , 2010, Price pressure in the government bond market, American Economic Review 100, 585 590. Gromb, Denis, and Dimitry Vayanos, 2010, Limits of arbitrage, Annual Review of Financial Economics 2, 251 275. Guedes , Jose, and Tim C. Opler, 1996, The determinants of the maturity of corporate debt issues, Journal of Finance 51, 1809 1833. He, Zhiguo, and Wei Xiong, 2012, Rollover risk and credit risk, Journal of Finance 67, 391 430. Ivashina, Victoria, and David Sch arfstein, 2010, Bank lending during the financial crisis of 2008, Journal of Financial Economics 97, 319 338. Krishnamurthy, Arvind, and Annette Vissing Jorgensen, 2012, The aggregate demand for Treasury debt, Journal of Political Economy 120, 233 267. K rishnamurthy, Arvind, and Annette Vissing Jorgensen, 2011, The effects of Quantitative Easing on interest rates: channels and implications for policy, Brookings Papers on Economic Activity , 215 287.

PAGE 133

133 Krishnamurthy, Arvind, 2010, How debt markets have malfu nctioned in the crisis, Journal of Economic Perspectives 24, 3 28. Leary, Mark T., and Michael R. Roberts, 2005, Do firms rebalance their capital structure? Journal of Finance 60, 2575 2619. Lemmon, Michael L., Michael R. Roberts, and Jaime F. Zender , 2008, Back to the beginning: persistence and the cross section of corporate capital structure, Journal of Finance 63, 1575 1608. Myers, Stewart C., 1977, Determinants of corporate borrowing, Journal of Financial Economics 5, 147 175. Nadauld, Taylor D. , and Michael S. Weisbach, 2012, Did securitization affect the cost of corporate debt? Journal of Financial Economics 105, 332 352. Opler, Tim C., and Sheridan Titman, 1994, Financial distress and corporate performance, Journal of Finance 49, 1015 1040. Park, Cheol, 2000, Monitoring and structure of debt contracts, Journal of Finance 55, 2157 2195. Rauh, Joshua D., and Amir Sufi, 2010, Capital structure and debt structure, Review of Financial Studies 23, 4242 4280. Santos, João A. C., 2011, Bank corpora te loan pricing following the subprime crisis, Review of Financial Studies 24, 1916 1943. Standard & Poor's Rating Services, 2012, Leverage commentary & data: a guide to the U.S. loan market . Stohs, Mark H., and David C. Mauer, 1996, The determinants of corporate debt maturity structure, Journal of Business 69, 279 312. Sufi, Amir, 2007, Information asymmetry and financing arrangements: evidence from syndicated loans, Journal of Finance 62, 629 668. Swanson, Eric T., 2011, Let's twist again: a high freq uency event study analysis of Operation Twist and its implications for QE2, Brooki ngs Papers on Economic Activity , 151 207. U.S. Department of the Treasury, 2 001, Press Release (October 31) . Vayanos, Dimitr y, and Jean Luc Vila, 2009, A preferred habitat model of the term structure of int erest rates, NBER Working Paper .

PAGE 134

134 BIOGRAPHICAL SKETCH Dominique is a CFA cha r terholder and earned his MSc from the Swiss Federal Institute of Technology, majoring in mathem atics. Dominique joined the PhD program in finance at the University of Florida in the fall of 2009. After completing his PhD in the summer of 2014, he will join the finance department at the University of Missouri as an assistant professor of finance.