maturity and corporate loan pricinghomepages.rpi.edu/home/17/wuq2/yesterday/public... · maturity...

24
The Financial Review 39 (2004) 55--77 Maturity and Corporate Loan Pricing Aron A. Gottesman Lubin School of Business, Pace University Gordon S. Roberts Schulich School of Business, York University Abstract We investigate the relation between corporate loan spreads and maturity to test whether lenders are compensated for longer maturity loans (tradeoff hypothesis) or limit their exposure by forcing riskier borrowers to take short-term loans (credit-quality hypothesis). Earlier studies reject the tradeoff hypothesis. We use the LPC DealScan database to create a matched sample of pairs of loans to the same borrower on the same day holding credit quality constant. We perform mean of difference tests and cross-sectional and regression analyses, and find evidence supporting both the tradeoff and credit quality hypotheses. Keywords: bank, borrower, loan, contract terms JEL Classification: G21 Corresponding author: CIBC Professor of Financial Services, Schulich School of Business, York Univer- sity, 4700 Keele Street, Toronto, Ontario, Canada M3J 1P3; Phone: (416) 736-2100, X77953; Fax: (416) 736-5687; E-mail: [email protected]. Financial support for this project came from the Social Sciences and Humanities Research Council of Canada and the CIBC Professorship in Financial Services at the Schulich School of Business at York University. The authors benefited from comments from Larry Wall, the editor, an anonymous referee, seminar participants at the Symposium on Bank Lending at the Eastern Finance Association Meeting, the Midwest Finance Association Meeting, the Multinational Finance Society Conference, Pace University, and York University. The authors thank Kamphol Panyagometh for excellent research assistance. 55

Upload: voliem

Post on 11-May-2018

216 views

Category:

Documents


2 download

TRANSCRIPT

The Financial Review 39 (2004) 55--77

Maturity and Corporate Loan PricingAron A. Gottesman

Lubin School of Business, Pace University

Gordon S. Roberts∗Schulich School of Business, York University

Abstract

We investigate the relation between corporate loan spreads and maturity to test whetherlenders are compensated for longer maturity loans (tradeoff hypothesis) or limit their exposureby forcing riskier borrowers to take short-term loans (credit-quality hypothesis). Earlier studiesreject the tradeoff hypothesis. We use the LPC DealScan database to create a matched sampleof pairs of loans to the same borrower on the same day holding credit quality constant. Weperform mean of difference tests and cross-sectional and regression analyses, and find evidencesupporting both the tradeoff and credit quality hypotheses.

Keywords: bank, borrower, loan, contract terms

JEL Classification: G21

∗Corresponding author: CIBC Professor of Financial Services, Schulich School of Business, York Univer-sity, 4700 Keele Street, Toronto, Ontario, Canada M3J 1P3; Phone: (416) 736-2100, X77953; Fax: (416)736-5687; E-mail: [email protected].

Financial support for this project came from the Social Sciences and Humanities Research Council ofCanada and the CIBC Professorship in Financial Services at the Schulich School of Business at YorkUniversity. The authors benefited from comments from Larry Wall, the editor, an anonymous referee,seminar participants at the Symposium on Bank Lending at the Eastern Finance Association Meeting, theMidwest Finance Association Meeting, the Multinational Finance Society Conference, Pace University,and York University. The authors thank Kamphol Panyagometh for excellent research assistance.

55

56 A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77

1. Introduction

The design and pricing of loan contracts has recently attracted considerable re-search interest. This line of research seeks to understand the process through whichcorporate borrowers choose public, private nonbank, and/or private bank debt. Inaddition, it endeavors to explain how borrowers and lenders settle on key loan con-tract features, such as maturity and collateral, and how these features are priced.More broadly, work in this area is important in operationalizing theories of finan-cial intermediation based on the premise that loan contract design decisions aim toovercome asymmetries between borrower private beliefs and market assessments ofcredit quality, and are influenced by reputation and relationship effects (Diamond,1993; Houston and James, 2001; among others). Prior empirical work, as reviewed inDennis, Nandy, and Sharpe (2000), has enhanced understanding of why certain loanfeatures, such as maturity and collateral, appeal to firms with particular characteris-tics, such as high growth and uncertain risk profiles.

Prior research can be grouped into two principal types of studies. First, in thecorporate finance literature, a number of studies explore the process through whichborrowers choose different proportions of short-term, long-term, collateralized, andnoncollateralized debt. Employing balance sheet data, the studies successfully predictthese choices at the firm level based on the borrower’s risk profile, growth potential,and other characteristics. Examples include Barclay and Smith (1995a, 1995b) andStohs and Mauer (1996). Second, in the financial institutions literature, researchersfocus on contract feature choices made by borrowers when negotiating new loansfrom banks and other private lenders. Examples include Strahan (1999) and Dennis,Nandy, and Sharpe (2000). Prior research recognizes that three decisions are madesimultaneously: the choice of loan spread over LIBOR, time to maturity, and thedecision whether to secure the loan using collateral. For example, Dennis, Nandy, andSharpe (2000) introduce a simultaneous equation model of determinants of contractterms and spreads in revolving credit arrangements between borrowers and banks.

Despite the advances in these prior studies, the link between loan spread andmaturity remains open to interpretation because this link reflects two opposing effects.A borrower issuing short-term debt can face costly liquidation at maturity, motivatingthe borrower to choose longer-term debt. At the same time, lenders prefer short-termdebt to control agency problems, such as asset substitution and underinvestment.1 Asa result, borrowers are willing to incur, and lenders demand, higher spreads for loanswith longer maturity. Underlying this positive relation between spreads and maturityis the hypothesis that lenders are willing to offer long-term loans to risky borrowersat higher spreads (tradeoff hypothesis). More formally, the tradeoff hypothesis statesthat the term structure for bank loans is upward sloping, consistent with empirical

1 See Myers (1977) and Barnea, Haugen, and Senbet (1980) for discussions of asset substitution and un-derinvestment. Similar arguments pertain to collateralized debt as a means of controlling agency problems.A contrary view under which collateral heightens agency costs is in John, Lynch, and Puri (2002).

A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77 57

results for corporate bonds (Helwege and Turner, 1999) and for bank loans (Coleman,Esho, and Sharpe, 2002).

An alternative to the tradeoff hypothesis holds that, instead of offering longer-term loans at higher rates, lenders limit their exposure through forcing riskier bor-rowers to take short-term loans (credit quality hypothesis). At the same time, lessrisky borrowers signal their credit quality by taking short-term loans (Flannery, 1986;Kale and Noe, 1990), suggesting a negative relation between spreads and maturity.Empirical work by Strahan (1999) finds such a negative relation for his sample of re-volving and term loans. Dennis, Nandy, and Sharpe (2000, p. 102) also find evidenceto support a negative relation for revolving loans, and interpret it as “inconsistent withthe tradeoff hypothesis.”2

These empirical findings are puzzling in light of the evidence of a positivelysloped term structure for corporate bonds. Further, their interpretation overlooks thepossibility that both the tradeoff and credit quality hypotheses reflect reality. Whennegotiating with individual firms, a bank is willing to accept the spread-maturitytradeoff. However, when structuring its loan portfolio, the bank will limit exposurethrough truncating the tradeoff by forcing riskier borrowers to take short-term loans.As a result, both the tradeoff and credit-quality hypotheses can hold at the sametime; the former at the firm level in comparing alternative packages of contract terms,and the latter between firms, as Helwege and Turner (1999) note for corporate bondspreads. When investigating the relations between corporate loan spreads and matu-rity, a number of important factors should be controlled, such as loan type and lendercharacteristics. For example, Coleman, Esho, and Sharpe (2002) note that firms bor-rowing with revolving loans are larger and less levered than firms borrowing withterm loans. An important implication is that it is crucial to conduct tests separatelyon revolving and term loans to isolate the borrower effects.

This study exploits the existence of a large sample of borrowers engaging inmultiple loan contracts with the same lead and participant banks on the same day inthe Loan Pricing Corporation DealScan database. Using this sample, the paper teststhe tradeoff and credit quality hypotheses using matched pairs of loans, where bothloans in each matched pair are made to the same borrower, on the same date, by anidentical syndicate of lenders. This approach allows the paper to hold constant theborrower and lender characteristics. To test the tradeoff and credit quality hypotheses,we focus on fee and interest rate spread variables. Following earlier work, the keymeasure of cost for a loan facility is “all-in-spread-drawn,” defined by Loan PricingCorporation (1999) as “the basis point coupon spread over LIBOR plus the annualfee and upfront fee, spread over the life of the loan.” Under the tradeoff hypothesis,we expect this measure to increase as time to maturity increases, while the credit-quality hypothesis predicts it will decrease with longer maturity. Mean of differencetests identify significantly larger spreads associated with longer-maturity, term loans,

2 Berger and Udell (1990) also find such a negative relation.

58 A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77

suggestive of a positively sloped term structure. Cross-sectional analyses conductedon paired samples demonstrate that this finding is robust.

To demonstrate that the credit-quality effect works jointly with the tradeoffeffect, we next remove the control for the former by dropping our matching method-ology and reexamine our sample of loans employing regression analysis. We show thatpooled regressions similar to those discussed above lack robustness due to inadequatecontrols for lender and credit-quality effects. Selected regressions display negativecoefficients for maturity, consistent with the credit-quality hypothesis. These find-ings strongly support our argument that individual borrowers face a positively slopedterm structure of lending rates. The negative slope found in prior research arises fromthe credit-quality effect: lenders force riskier borrowers to take short-term loans.

The rest of this study is organized as follows. Section 1 describes the data andsampling method. Section 2 presents univariate tests of the statistical relation betweenprice and maturity. Section 3 describes multivariate regression analysis and presentsthe associated results. In Section 4, we offer conclusions.

2. Data and sampling method

To test the credit quality and tradeoff hypotheses, two data samples are created:a large sample of loans, and a subset of matched pairs. The matched pair subset isformed to isolate the tradeoff effect from the impact of loan quality. Matching entailscreating a set of paired observations of two loan deals, differing in maturities, extendedto the same company on the same day. We control observable features including thepresence or absence of collateral, loan size, and the credit quality of the company.Unobservable features are controlled through matching the two loans in the same dayfor the same company.

We use the Loan Pricing Corporation DealScan database to extract loan dealsinitiated between 1988 and 1999, between U.S. banks and nonfinancial U.S. firms.Typically, a loan deal consists of a number of dissimilarly designed loans with commonlead and participant banks, designated “facilities,” made to the same borrower on agiven date. We eliminate any facility for which the “term facility maturity” or “all-in-spread-drawn” variables are missing.

We perform our tests separately for term loans and revolving loans. Revolvingcredit arrangements specify a maximum amount that can be borrowed for the du-ration of the arrangement. The actual amount borrowed varies from this maximumamount, and there is a commitment fee associated with the unused portion of thecredit arrangement. The DealScan database specifies loans that are revolving loans,and we treat other loans as term loans. As noted in Coleman, Esho, and Sharpe (2002),differences in borrower and lender characteristics between term and revolving loanssuggest that empirical tests should be performed separately. Of the facilities that re-main following the previous filtering, 11,817 are term loans and 18,358 are revolvingloans. We designate this sample of term and revolving loan observations as “full.” Toprepare for the matched pairs test, we eliminate any loan deal for which the times to

A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77 59

Table 1

Breakdown of facilities in the LPC DealScan database and the sample selection process

Facilities Pairs

(1) All facilities over the period 1970–1999 66,491(2) All facilities in (1), excluding deals associated with finance companies,

non-U.S. companies, or non-U.S. banks47,342

(3) All term facilities in (2), excluding facilities for which the“all-in-spread-drawn” or “term facility maturity” variable is missing

11,817

(4) All revolving facilities in (2), excluding facilities for which the“all-in-spread-drawn” or “term facility maturity” variable is missing

18,358

(5) All facilities in (3), excluding deals where the times to maturity of all facilitiesare identical, or only a single facility exists

3,112

(6) Pairs formulated based on facilities in (5) 1,765(7) All facilities in (4), excluding deals where the times to maturity of all facilities

are identical, or only a single facility exists832

(8) Pairs formulated based on facilities in (7) 433

maturity of all facilities associated with the loan deal are identical, or only a singlefacility exists. Of the remaining facilities, 3,112 are term loans and 832 are revolvingloans. We designate this sample as “paired.”

We next form pairs of facilities where each pair contains a shorter and longermaturity loan from the same deal. If only two facilities are associated with a loandeal, we simply identify one facility as shorter time to maturity and the other aslonger. If more than two facilities are associated with a loan deal, we treat everytwo facilities with different times to maturity as a pair. For example, consider adeal with three facilities, designated A, B, and C. The times to maturity of the threefacilities (in months) are 60, 72, and 84, respectively. For the matched pair sample,we create three unique pairs: A with B, B with C, and A with C. Since pairs areformed separately for the revolving and term loan samples, both elements of eachpair are either revolving or term.3 This methodology produces 1,765 term and 433revolving loan pairs, representing approximately 10% of the total number of U.S.loans available in the database.4 Table 1 details the sample selection process.

Table 2 specifies the variables extracted and the respective units of measure. Aswell, the table reports the number of facilities for which each variable is nonmissing

3 While we do not include a restriction that both elements of the pairs share the same collateralization status,our tests demonstrate that there are no significant statistical or economic differences in collateralizationstatus between the elements of the pairs.

4 As a test of robustness, we also formed a sample of matched pairs using an alternative methodology. Thealternative sample was created through matching the longest maturity facility for each loan deal againstthe shortest maturity facility. This methodology resulted in only a single matched pair for each loan dealwith more than a single facility. The empirical results associated with this alternative sample, which wedo not report, are generally consistent with the results for the sample of matched pairs created through themethodology we describe in this paper.

60 A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77Ta

ble

2

The

vari

able

s,th

eva

riab

lety

pe,a

ndth

enu

mbe

rof

faci

litie

sfo

rw

hich

the

vari

able

isno

nmis

sing

The

vari

able

s,th

eva

riab

lety

pe,a

ndth

enu

mbe

rof

faci

litie

sfo

rw

hich

the

vari

able

isno

nmis

sing

are

repo

rted

for

both

the

full

and

pair

edsa

mpl

es,f

orbo

thth

ete

rman

dre

volv

ing

loan

sam

ples

.As

wel

l,th

eta

ble

repo

rts

the

num

ber

ofpa

irs

form

edfr

omth

epa

ired

sam

ple

for

whi

chth

eva

riab

leis

nonm

issi

ngfo

rbo

thel

emen

ts,f

orth

ete

rmsa

mpl

e,an

dfo

rth

esu

bset

ofob

serv

atio

nsco

nsis

ting

ofre

volv

ing

loan

s.V

aria

ble

defi

nitio

nsar

eas

follo

ws:

TFC

MA

Tis

the

term

faci

lity

mat

urity

.RA

TE

AIS

Dis

the

rate

sal

l-in

-spr

ead-

draw

n.R

AT

EA

ISU

isth

era

tes

all-

in-s

prea

d-un

draw

n.R

AT

EPR

IMis

the

rate

spr

ime

spre

ad.R

AT

EL

IBO

isth

era

tes

LIB

OR

spre

ad.F

EE

UPF

RO

isth

efe

esup

fron

t.FE

EC

OM

isth

efe

esco

mm

itmen

t.FE

EL

Cis

the

fees

LC

.FE

EC

AN

isth

efe

esca

ncel

latio

n.ln

(AM

TFC

SIZ

)is

the

ln(A

mou

ntfa

cilit

ysi

ze).

CO

LL

AT

isth

eco

llate

raliz

ed/n

onco

llate

raliz

eddu

mm

yva

riab

le.B

WM

Dis

the

Bor

row

erM

oody

Seni

orD

ebtR

atin

g.

Full

sam

ple

Pair

edsa

mpl

e

Num

ber

offa

cilit

ies

Num

ber

offa

cilit

ies

Num

ber

ofpa

irs

for

whi

chth

eva

riab

lefo

rw

hich

the

vari

able

for

whi

chth

eva

riab

leis

nonm

issi

ngis

nonm

issi

ngis

nonm

issi

ngfo

rbo

thel

emen

ts

Non

revo

lvin

gR

evol

ving

Non

revo

lvin

gR

evol

ving

Non

revo

lvin

gR

evol

ving

Var

iabl

eV

aria

ble

type

sam

ple

sam

ple

sam

ple

sam

ple

sam

ple

sam

ple

TFC

MA

TM

onth

s11

,817

18,3

583,

112

832

1,76

543

3R

AT

EA

ISD

Bas

ispo

int

11,8

1718

,358

3,11

283

21,

765

433

RA

TE

AIS

UB

asis

poin

t1,

809

14,1

5811

362

017

281

RA

TE

PRIM

Bas

ispo

int

7,87

914

,321

1,92

468

91,

069

353

RA

TE

LIB

OB

asis

poin

t8,

712

14,2

162,

680

624

1,51

430

9FE

EU

PFR

OB

asis

poin

t4,

133

6,24

51,

237

320

621

148

FEE

CO

MB

asis

poin

t1,

154

11,5

6317

852

948

233

FEE

LC

Bas

ispo

int

207

5,87

727

188

537

FEE

CA

NB

asis

poin

t59

699

813

333

5512

ln(A

MT

FCSI

Z)

Dol

lar

11,8

1718

,358

3,11

283

21,

765

433

CO

LL

AT

Dum

my

7,27

111

,833

2,00

353

21,

128

268

BW

MD

Ass

igne

dva

lue

2,76

94,

165

834

146

490

77

A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77 61

for both the full and paired samples, and the number of pairs for which the variableis nonmissing for both elements.

Variables that measure rates include all-in-spread-drawn (RATEAISD), all-in-spread-undrawn (RATEAISU), the prime spread (RATEPRIM), and the LIBORspread (RATELIBO). The all-in-spread-drawn is the sum of the coupon spread andannual fee. The all-in-spread-undrawn is the sum of the commitment fee and annualfee. The prime spread is the spread over the prime rate, while the LIBOR spread isthe spread over the LIBOR rate.

Variables that measure fees include upfront fees (FEEUPFRO), commitmentfees (FEECOM), letter of credit fees (FEELC), and cancellation fees (FEECAN).Upfront fees are one-time fees that are typically collected at the close of the deal.Commitment fees are charged on the unused portion of the credit. Letter of credit feesare the annual fees charged related to the issuance of letters of credit. Cancellationfees are charged against termination or reduction of the commitment.

A number of variables measure other loan characteristics. TFCMAT is the termto maturity of the loan facility, measured in months. ln(AMTFCSIZ) is the naturallogarithm of the loan facility size. COLLAT is a dummy variable that controls forcollateralization, and is equal to one if the loan is secured with collateral and zerootherwise. BWMD is the Moody’s senior debt rating for borrowers, and is our measureof credit quality. A cardinal ranking that decreases in risk and ranges from 8 (riskiest)to 28 (least risky) was established, corresponding to the Moody’s letter values.

Table 3 reports the mean value and standard deviation for each variable, for theterm and revolving subsets of both the full and paired samples. For each variable, weexamine whether means differ significantly between term and revolving loans withinthe full sample and again within the paired sample. All differences between term andrevolving loans are statistically significant for both the full and paired samples, withthe exception of FEECAN for the paired sample. Within the full sample, economicdifferences are small between means of term and revolving loans, with the exceptionof RATEISD. In the paired sample, revolving loans are shorter, exhibit lower spreads,are less likely to be collateralized, and are higher quality. These differences aresignificant within the paired sample. They are generally consistent with Coleman,Esho, and Sharpe (2002). Further, we examine how sample characteristics shift whenwe introduce pairing. For term loans, comparing the features of the full sample withthe paired sample reveals that the loans in the paired sample have longer maturitiesand higher spreads. They are larger, more likely to be collateralized, and of lowercredit quality. Turning to revolving loans reveals that compared to the full sample,paired loans are similar in maturity, have somewhat lower spreads, and are less likelyto be collateralized. Borrowers in the paired sample are slightly smaller.

3. Univariate analysis

We begin our analysis by counting the number of times that the longer maturityfacility of the matched pairs is associated with a higher, identical, or lower value for

62 A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77Ta

ble

3

The

mea

nva

lue

and

stan

dard

devi

atio

nfo

rea

chva

riab

le

The

mea

nva

lue

and

stan

dard

devi

atio

nfo

rea

chva

riab

lear

ere

port

edfo

rbo

thth

efu

llan

dpa

ired

sam

ples

,for

both

the

term

and

revo

lvin

glo

ansa

mpl

es.V

aria

ble

defi

nitio

nsar

eas

follo

ws:

TFC

MA

Tis

the

term

faci

lity

mat

urity

.RA

TE

AIS

Dis

the

rate

sal

l-in

-spr

ead-

draw

n.R

AT

EA

ISU

isth

era

tes

all-

in-s

prea

d-un

draw

n.R

AT

EPR

IMis

the

rate

spr

ime

spre

ad.

RA

TE

LIB

Ois

the

rate

sL

IBO

Rsp

read

.FE

EU

PFR

Ois

the

fees

upfr

ont.

FEE

CO

Mis

the

fees

com

mitm

ent.

FEE

LC

isth

efe

esL

C.F

EE

CA

Nis

the

fees

canc

ella

tion.

ln(A

MT

FCSI

Z)

isth

eln

(Am

ount

faci

lity

size

).C

OL

LA

Tis

the

colla

tera

lized

/non

colla

tera

lized

dum

my

vari

able

.B

WM

Dis

the

Bor

row

erM

oody

Seni

orD

ebtR

atin

g. Full

sam

ple

Pair

edSa

mpl

e

Term

sam

ple

Rev

olvi

ngsa

mpl

eTe

rmsa

mpl

eR

evol

ving

sam

ple

Stan

dard

Stan

dard

Mea

ndi

f.te

stSt

anda

rdSt

anda

rdM

ean

dif.

test

Mea

nde

viat

ion

Mea

nde

viat

ion

T-St

atM

ean

devi

atio

nM

ean

devi

atio

nT-

Stat

TFC

MA

T52

.08

115.

2851

.34

24.8

923

.28∗

∗∗69

.18

28.1

643

.98

27.8

74.

85∗∗

∗R

AT

EA

ISD

243.

6914

0.26

223.

7511

7.58

15.3

8∗∗∗

280.

6888

.67

213.

8412

2.52

25.5

0∗∗∗

RA

TE

AIS

U20

.01

21.8

939

.43

20.1

32.

63∗∗

∗46

.29

35.1

736

.60

19.7

4−2

9.13

∗∗∗

RA

TE

PRIM

103.

6395

.50

77.3

580

.55

20.6

7∗∗∗

136.

2577

.41

66.6

883

.07

35.4

2∗∗∗

RA

TE

LIB

O19

6.58

113.

3818

6.45

93.7

324

.81∗

∗∗26

6.21

75.6

716

8.97

94.6

227

.66∗

∗∗FE

EU

PFR

O77

.38

87.4

569

.50

67.5

03.

46∗∗

∗75

.05

97.7

959

.77

59.8

010

.60∗

∗∗FE

EC

OM

34.2

822

.03

41.0

116

.08

5.25

∗∗∗

49.4

427

.81

38.1

116

.25

−3.7

4∗∗∗

FEE

LC

135.

6180

.15

185.

5981

.49

−1.6

9∗15

0.83

71.4

717

2.13

81.8

1−4

.67∗

∗∗FE

EC

AN

196.

8312

8.15

189.

7212

9.74

2.12

∗∗18

1.77

118.

5914

6.32

103.

690.

94L

n(A

MT

FCSI

Z)

17.3

31.

9417

.36

1.67

7.59

∗∗∗

17.4

41.

6517

.00

1.71

−10.

43∗∗

∗C

OL

LA

T0.

870.

330.

870.

347.

72∗∗

∗0.

960.

200.

840.

3711

.74∗

∗∗B

WM

D17

.69

3.72

17.4

93.

64−9

.73∗

∗∗15

.53

2.52

18.6

43.

69−3

.56∗

∗∗

∗∗∗ I

ndic

ates

stat

istic

alsi

gnif

ican

ceat

the

0.01

leve

l.∗∗

Indi

cate

sst

atis

tical

sign

ific

ance

atth

e0.

05le

vel.

∗ Ind

icat

esst

atis

tical

sign

ific

ance

atth

e0.

10le

vel.

A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77 63

the rates-all-in-spread-drawn (RATEAISD) variable, relative to the shorter maturityfacility (Table 4). These values are reported for both the term and revolving samples,for all matched pairs, and are reported separately for matched pairs associated withborrowers that have Moody’s senior debt ratings of A, B, C, and unrated. We alsoreport the percentage higher, identical, and lower, and the number of pairs for eachsample.

For the term sample, we find strong evidence that the longer maturity facilitiesare associated with higher spreads than the shorter maturity facilities. Overall, for 71%of the matched pairs the longer maturity facility is associated with higher spreads thanthe shorter maturity facility, while the reverse is true for only 6% of the matched pairs.For the A, B, C, and unrated borrowers, we find that for 45.5%, 77.1%, 91.7%, and68.6% of the matched pairs the longer maturity facility is associated with higherspreads than the shorter maturity facility, respectively. The reverse is true for only18%, 2%, 4%, and 8% of the matched pairs, respectively.

The results are more ambiguous for the revolving sample. Overall, 26% ofmatched pairs exhibit higher spreads for the longer maturity facility over the shorterfacility, while the reverse is true in 17% of the matched pairs. More interestingly,spreads are identical in 57% of the matched pairs. This result appears consistent forborrowers rated A, B, C, and unrated, with identical spreads in 65%, 61%, 67%, and56% of matched pairs, respectively.

We next perform mean of difference tests for the pairs formed from the termand revolving loan paired samples. For each variable, we calculate the mean of thedifference for all observations, for all pairs for which the variable is available. Wecalculate the Student’s T-statistic and Wilcoxon Signed Rank Test to test whether thepositive association of maturity with loan spreads posited by the tradeoff hypothesisis supported. The results of these tests are reported in Table 5.

By construction, there is a large difference for the “term facility maturity” vari-able, TFCMAT, for both samples. For the term sample, longer maturity loan facilitiesare, on average, approximately 22 months longer in maturity than shorter maturityfacilities. For the revolving sample, the difference is approximately 30 months. Thesedifferences are statistically significant at the 1% level.

There is strong evidence that the longer maturity facilities are associated withhigher rates for the term sample (Panel A). All-in-spread-drawn, prime spread, andLIBOR spreads are approximately 28.3, 22.9, and 36.3 basis points larger than shortermaturity facilities, respectively, at the 1% level. As well, the longer maturity facilitiesare associated with all-in-spread-undrawn spreads that are approximately 4.6 basispoints larger than shorter facility maturity facilities at the 5% level. The difference inrates associated with the revolving sample (Panel B) is weaker, with the all-in-spread-undrawn rate approximately 1.8 basis points larger for the longer maturity facilities,statistically significant at the 1% level. There are no other statistically significantdifferences in rates for the revolving sample.

There is no statistically significant difference in loan commitment or cancellationfees for either sample. All-in-spread-drawn includes both spread and fees, and the

64 A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77

Tabl

e4

Cou

ntin

gte

sts

The

num

ber

oftim

esth

atth

elo

nger

mat

urity

faci

lity

ofa

mat

ched

pair

has

ahi

gher

,ide

ntic

al,o

rlo

wer

valu

efo

rth

era

tes

alli

nsp

read

undr

awn

(RA

TE

AIS

D)

vari

able

rela

tive

toth

esh

orte

rm

atur

ityfa

cilit

yis

repo

rted

.The

seva

lues

are

repo

rted

for

both

the

term

and

revo

lvin

gsa

mpl

es,f

oral

lm

atch

edpa

irs,

and

are

repo

rted

sepa

rate

lyfo

rmat

ched

pair

sas

soci

ated

with

borr

ower

sw

ithM

oody

’sse

nior

debt

ratin

gsof

A,B

,C,a

ndun

rate

d.W

eal

sore

port

the

perc

enta

gehi

gher

,id

entic

al,l

ower

,and

the

num

ber

ofpa

ired

obse

rvat

ions

for

each

sam

ple.

All

A-r

ated

B-r

ated

C-r

ated

Unr

ated

Term

Rev

olvi

ngTe

rmR

evol

ving

Term

Rev

olvi

ngTe

rmR

evol

ving

Term

Rev

olvi

ng

Num

ber

ofpa

irs

1,76

543

311

1745

557

243

1,27

535

6H

ighe

r1,

253

113

55

351

1022

187

597

Iden

tical

398

247

411

9335

12

300

199

Low

er11

473

21

1112

10

100

60

Perc

enta

gehi

gher

0.71

0.26

45.5

029

.40

77.1

017

.50

91.7

033

.30

68.6

027

.20

Perc

enta

geid

entic

al0.

230.

570.

360.

650.

200.

610.

040.

670.

240.

56Pe

rcen

tage

low

er0.

060.

170.

180.

060.

020.

210.

040.

000.

080.

17

A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77 65

Table 5

Mean of difference tests

Panels A and B report the results for the term and revolving loan samples, respectively. Variable defini-tions are as follows: TFCMAT is the term facility maturity. RATEAISD is the rates all-in-spread-drawn.RATEAISU is the rates all-in-spread-undrawn. RATEPRIM is the rates prime spread. RATELIBO is therates LIBOR spread. FEEUPFRO is the fees upfront. FEECOM is the fees commitment. FEELC is thefees LC. FEECAN is the fees cancellation. ln(AMTFCSIZ) is the ln(Amount facility size). COLLAT isthe collateralized/noncollateralized dummy variable.

Number of The mean Wilcoxon Mean, Mean,nonmissing of the Student’s signed rank shorter longer

pairs difference T statistic statistic maturity maturity

Panel A: Mean of difference tests, term loans

TFCMAT 1,765 21.66 49.50∗∗∗ 77,9247.50∗∗∗ 58.99 80.65RATEAISD 1,765 28.32 20.89∗∗∗ 36,7126.50∗∗∗ 266.40 294.72RATEAISU 17 4.56 2.52∗∗ 28.00∗∗ 23.88 28.44RATEPRIM 1,069 22.85 16.77∗∗∗ 97,775.50∗∗∗ 126.72 149.57RATELIBO 1,514 36.32 37.08∗∗∗ 350,749.00∗∗∗ 249.15 285.47FEEUPFRO 621 −14.45 −3.47∗∗∗ −11,027.50∗∗∗ 81.40 66.95FEECOM 48 −0.29 −0.29 −0.50 44.89 44.60FEELC 5 0.00 N/A N/A 140.00 140.00FEECAN 55 0.00 N/A N/A 187.76 187.76ln(AMTFCSIZ) 1,765 0.05 2.49∗∗ 33,527.50∗ 17.45 17.50COLLAT 1,128 0.00 0.58 1.00 0.96 0.96

Panel B: Mean of difference tests, revolving loans

TFCMAT 433 29.81 28.36∗∗∗ 46,980.50∗∗∗ 30.15 59.96RATEAISD 433 −0.49 −0.04 1,063.50 213.37 212.88RATEAISU 281 1.80 2.87∗∗∗ 840.00∗∗∗ 35.14 36.94RATEPRIM 353 2.20 0.83 407.00 65.09 67.29RATELIBO 309 2.16 1.10 899.50∗∗∗ 164.48 166.64FEEUPFRO 148 −2.06 −0.86 13.50 61.01 58.95FEECOM 233 1.23 1.92∗ 306.50∗∗∗ 37.69 38.92FEELC 37 0.34 0.03 13.00 193.75 194.09FEECAN 12 1.04 1.00 0.50 141.47 142.51ln(AMTFCSIZ) 433 0.41 7.19∗∗∗ 16,273.50∗∗∗ 16.79 17.20COLLAT 268 0.01 0.38 2.00 0.84 0.85

∗∗∗ Indicates statistical significance at the 0.01 level.∗∗ Indicates statistical significance at the 0.05 level.∗ Indicates statistical significance at the 0.10 level.

nonsignificance of fees suggests that differences are due to spreads. The commitmentfee is not statistically different for the term sample. For the revolving sample, thecommitment fee is approximately 1.2 basis points higher, significant at the 10%level. For the term sample, the upfront fee is approximately 14.5 basis points smallerfor the longer maturity facilities, significant at the 1% level. No significant differencein upfront fee was found for the revolving sample.

66 A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77

The difference of means test for both the term and revolving samples suggests thatlonger maturity facilities are larger than shorter maturity facilities. The average naturallogarithm of the longer maturity facilities is approximately 0.05 and 0.41 larger thantheir shorter maturity counterparts for the term and revolving samples, respectively.These differences are significant at the 5% level for the term sample and at the 1%level for the revolving sample. To facilitate economic interpretation, we translatethe differences into dollars. For the term sample (Panel A), the mean difference inloan size is approximately $1.05 million, relatively minor compared to an averagesize for the shorter maturity loan of approximately $37.88 million. For the revolvingsample (Panel B), the mean difference in loan size is approximately $1.51 million.This amount is also relatively minor compared to an average size for the shortermaturity loan of approximately $19.63 million. These differences in size betweenthe longer and shorter maturity facilities do not pose a threat to our methodologybecause any bias introduced is in the direction of rejecting our hypothesized tradeoffeffect.5

Finally, there is no statistically significant difference between the percentagesof facilities collateralized for the longer and shorter maturity components of both theterm and revolving samples. The lack of significance is not surprising, as 96% ofterm loans and 84% of term loans are collateralized.6

Highlighting our most important results, our counting (Table 4) and paired tests(Table 5) contradict finding in earlier studies of a negative relation between spreadsand maturity for bank loans, and reinforces Helwege and Turner’s (1999) findingsfor corporate bonds. For the term loans we find a significant, positive relation andfor the revolving loans there is no significant relation. This suggests that the termstructure of credit spreads is flat initially (for short-maturity revolving loans) andthen becomes upward sloping (for long-maturity term loans). These paired tests areconvincing as facility size, collateral, and bond rating as well as lead and participantbanks are controlled perfectly through the matching technique we employ.

4. Multivariate analysis

In Section 3, we demonstrate that longer maturity facilities are associated withhigher spreads. The evidence of significant association between spread and maturitywithin pairs supports the tradeoff hypothesis. However, several outstanding issuesremain. First, the mean of difference tests identified significant differences in fa-cility size between the longer and shorter components. While we argued that these

5 Strahan (1999) reports a weak negative relation between loan size and all-in-spread-drawn. As a result,the greater size of the longer loan in our pairs should reduce spread.

6 The evidence relating collateralization and spreads is mixed. Bond rating agencies assign collateralizeddebt a higher rating than noncollateralized debt of the same issuer suggesting that collateral reduces yield.In contrast, John, Lynch, and Puri (2002) find that collateralized debt has higher yields after controllingfor bond ratings.

A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77 67

significant differences do not weaken the evidence that higher spreads associatedwith longer maturity facilities are due to the longer maturity characteristic, a formaltest is required to confirm that other characteristics associated with the longer ma-turity facilities explain the higher spreads. Second, if the spread difference betweenthe longer and shorter facilities is attributable to the difference in maturity, is thisdifference in spread invariant to the riskiness of the firm? The tradeoff hypothesissuggests that the spread difference should widen as the borrower becomes riskier.We address these issues through performing cross-sectional and regression analyses.Further, an important theme of this paper is that the positive relation between spreadsand loan maturity at the borrower level can be obscured in pooled regressions. Asdiscussed earlier, this occurs because pooling introduces credit quality effects acrossfirms. We test for the impact of pooling through conducting regression analysis.

4.1. Cross-sectional analysis

In Table 6, we report cross-sectional mean of difference tests conditioned ontwo variables: loan size and bond rating. For both the term and revolving loan pairedsubsets, we calculate the difference of the loan size variable between the longermaturity and shorter maturity facilities for each pair. We then place pairs into oneof ten groups of equal size, where the first group represents the pairs with greatestdifferences, and each subsequent grouping contains pairs with smaller differences.We then perform mean of difference tests for the RATEAISD and loan size variablesfor all pairs within each group. We also report the average bond rating values for thepairs within the group.

In the lower panel of Table 6, we repeat this approach through placing pairsinto one of ten groups on the basis of the bond rating value for the pair, and againperform mean of difference tests for the RATEAISD and loan size variables for allpairs within each group. We cannot perform a similar analysis based on groupings onthe basis of the COLLAT variable. This is because the vast majority of pairs displayno difference for the COLLAT variable, as the vast majority of loan facilities arecollateralized, effectively resulting in only two groups.

For the term subset, the grouping on the basis of the difference in the loansize variable demonstrates that the longer maturity facility is associated with higherspreads than the shorter maturity facility regardless of the difference in the loansize variable, with the exception of one grouping. Further, the difference in ratespread increases across groups (moving down the upper panel of Table 6) as thelonger maturity facilities become increasingly smaller than the shorter maturity fa-cilities. For group 10, the rate spread associated with the longer maturity facility is,on average, approximately 34.19 basis points larger than the shorter maturity facility.Conversely, for group 1, the rate spread associated with the longer maturity facil-ity is, on average, approximately 0.13 basis points larger than the shorter maturityfacility.

For the revolving loan subset, the grouping on the basis of the difference in

68 A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77

Table 6

Cross-sectional mean of difference tests

This table shows mean of difference tests for the RATEAISD and ln(AMTFCSIZ) variables, and themean values of the BWMD variable, for ten groups, for the revolving and term samples. Findings areorganized into two groupings: (1) based on the difference of means value of the ln(AMTFCSIZ) variable,and (2) based on the Mean BWMD variable. Variable definitions are as follows: RATEAISD is the rates all-in-spread-drawn. ln(AMTFCSIZ) is the ln(Amount facility size). BWMD is the Borrower Moody SeniorDebt Rating. Diff RATEAISD and Diff ln(AMTFCSIZ) are the differences in the values of the RATEAISDand ln(AMTFCSIZ) variables, respectively, between the longer maturity and shorter maturity facilities foreach pair.

Revolving Term

Diff Diff Mean Diff Diff MeanGroups RATEAISD ln(AMTFCSIZ) BWMD RATEAISD ln(AMTFCSIZ) BWMD

Diff ln(AMTFCSIZ)1 −29.60∗∗∗ 2.43∗∗∗ 18.80 0.13 1.71∗∗∗ 15.442 8.59 1.61∗∗∗ 18.86 29.21∗∗∗ 0.83∗∗∗ 15.443 −5.81 1.23∗∗∗ 16.25 27.77∗∗∗ 0.50∗∗∗ 15.394 −15.23 0.82∗∗∗ 19.50 36.29∗∗∗ 0.25∗∗∗ 15.265 −8.15 0.57∗∗∗ 17.63 24.97∗∗∗ 0.05∗∗∗ 16.076 −7.80 0.29∗∗∗ 17.90 32.70∗∗∗ −0.02∗∗∗ 15.677 5.60 0.01∗∗ 21.29 31.34∗∗∗ −0.20∗∗∗ 15.438 14.29∗∗ −0.29∗∗∗ 18.63 31.08∗∗∗ −0.42∗∗∗ 15.379 15.56∗∗∗ −0.76∗∗∗ 18.50 34.81∗∗∗ −0.73∗∗∗ 15.0510 22.05∗∗ −1.69∗∗∗ 15.78 34.19∗∗∗ −1.46∗∗∗ 16.50

Mean BWMD1 1.09 0.20 24.00 10.61 −0.01 20.142 7.03 0.05 22.88 31.73∗∗∗ −0.15 17.273 −6.88 0.60∗ 21.13 31.48∗∗∗ 0.04 16.244 −7.63 0.16 20.00 32.91∗∗∗ −0.13 16.005 2.81 1.02∗ 19.38 33.78∗∗∗ 0.11 15.806 −3.13 0.22 18.75 42.09∗∗∗ 0.19∗∗ 15.007 −10.00 0.42 16.75 32.40∗∗∗ 0.12 15.008 −10.94 0.02 15.00 38.77∗∗∗ −0.03 14.189 −6.25 0.48 13.75 28.57∗∗∗ 0.01 13.45

10 31.00 −0.71 11.00 37.24∗∗∗ 0.24∗ 12.00

∗∗∗ Indicates statistical significance at the 0.01 level.∗∗ Indicates statistical significance at the 0.05 level.∗ Indicates statistical significance at the 0.10 level.

the loan size variable only finds differences in spreads when the differences in theloan size variable are large. For group 10, the rate spread associated with the longermaturity facility is, on average, approximately 22.05 basis points larger than theshorter maturity facility. Conversely, for group 1, the rate spread associated with thelonger maturity facility is, on average, approximately 29.60 basis points smaller thanthe shorter maturity facility.

For the term subset, the grouping on the basis of the bond rating demonstrates

A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77 69

that, as predicted under the tradeoff hypothesis, riskier firms are associated withlarger differences in spreads between longer maturity and shorter maturity facilities.The five riskiest term groups (6 through 10) have an average RATEAISD differenceof means value of approximately 35.81 basis points, while the five least risky termgroups (groups 1 through 5) have an average RATEAISD difference of means valueof approximately 28.10 basis points. Further, the least risky term group, group 1, doesnot have a significant difference of means associated with the RATEAISD variable.For the revolving subset, the grouping on the basis of bond rating variable does notfind statistically significant differences in spreads.

4.2. Regression analysis

Regression tests do not incorporate the pairing technique employed in Section 3to isolate the impact of the tradeoff effect. There we found strong support for thetradeoff hypothesis contrary to the results of prior studies using pooled regressions.Because the regression technique pools observations across companies, it allows forthe operation of the credit-quality effect. Thus, in conducting regression tests, weare testing for the net impact of the two effects. If we find a negative coefficient formaturity, this will be evidence supporting the presence of both effects, correcting theconclusions of prior papers that only the credit-quality effect holds.

We investigate these issues through performing regressions relating loan spreadsand maturity, while controlling for other loan contract characteristics. Unlike Dennis,Nandy, and Sharpe (2000) and Coleman, Esho, and Sharpe (2002), our regressionspecifications do not account for simultaneity in the choice of contract terms suchas collateral and maturity and the determination of the spread. Hence, a negativecoefficient on maturity would support the presence of both effects. Panels A and Bof Table 7 present the correlations between the variables used in the regression testsfor the full and pooled paired samples, respectively, and reflect the collinearity notedin earlier research.

To test the core spread-maturity relation, all of the regression tests use “all-in-spread-drawn” as the dependent variable and “term facility maturity” as an indepen-dent variable. Other independent variables are included to control for facility size,the presence of a collateral provision, and risk. The regression model tested is asfollows:

RATEAISD = a0 + a1 × TFCMAT + a2 × ln(AMTFCSIZ)

+ a3 × COLLAT + a4 × BWMD (1)

This model is estimated using four methods, with each method imposing differ-ent restrictions. The first method tests the relation between spreads and the time tomaturity variable exclusively. The second method controls for facility size as well.The third method also controls for the presence of a collateral provision. The fourthmethod controls for riskiness as well.

Three types of regressions are performed: First, we run the regression tests for

70 A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77

Table 7

Correlations

This table shows correlations between variables for the full sample (Panel A), pooled observations associ-ated with the paired sample (Panel B), and pair data associated with the paired sample (Panel C). Variabledefinitions are as follows: RATEAISD is the rates all-in-spread-drawn. TFCMAT is the term facility ma-turity. ln(AMTFCSIZ) is the ln(Amount facility size). COLLAT is the collateralized/noncollateralizeddummy variable. BWMD is the Borrower Moody Senior Debt Rating. Diff RATEAISD is the differencein the value of RATEAISD between the longer maturity and shorter maturity facilities for each pair. DiffRATEAISD, Diff TFCMAT, and Diff ln(AMTFCSIZ) are the differences in the values of the RATEAISD,TFCMAT, and ln(AMTFCSIZ) variables, respectively, between the longer maturity and shorter maturityfacilities for each pair.

Panel A: Full sample

RATEAISD TFCMAT ln(AMTFCSIZ) COLLAT BWMD

RATEAISD 1.0000 0.0015 0.2297 −0.3344 0.3489TFCMAT 0.0015 1.0000 −0.2001 −0.2081 0.1333ln(AMTFCSIZ) 0.2297 −0.2001 1.0000 0.1768 −0.1604COLLAT −0.3344 −0.2081 0.1768 1.0000 0.2618BWMD 0.3489 0.1333 −0.1604 0.2618 1.0000

Panel B: Paired sample, pooled observations

RATEAISD TFCMAT ln(AMTFCSIZ) COLLAT BWMD

RATEAISD 1.0000 −0.0665 0.0024 −0.2001 0.2828TFCMAT −0.0665 1.0000 −0.1585 −0.2360 0.1349ln(AMTFCSIZ) 0.0024 −0.1585 1.0000 0.1545 −0.1868COLLAT −0.2001 −0.2360 0.1545 1.0000 0.1308BWMD 0.2828 0.1349 −0.1868 0.1308 1.0000

Panel C: Paired sample, pair data

Diff Diff DiffRATEAISD TFCMAT ln(AMTFCSIZ) BWMD

Diff RATEAISD 1.0000 0.3058 0.0879 0.2535Diff TFCMAT 0.3058 1.0000 −0.1964 −0.0035Diff ln(AMTFCSIZ) 0.0879 −0.1964 1.0000 0.0093BWMD 0.2535 −0.0035 0.0093 1.0000

all data, using the debt-rating variable, BWMD, to control for differences in riskiness.Second, we rerun the regressions for the paired data using differences within pairs.Third, as a robustness test, we form two subsamples based on whether the debt ratingis reported by the DealScan database, and repeat the regressions for each subsample.

4.2.1. Regression analysis, entire sample

We perform regression methods one through four for both the term and therevolving sample. For each of the samples, we run the regressions for all nonmissingfacilities. We estimate the regressions for the full and paired samples.

The results of the regression tests are in Table 8. Overall, consistent with our pre-

A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77 71

diction that pooling would confound the results, the regression tests do not revalidatethe positive relation between rate spread and maturity identified in our paired tests.Starting with the term paired sample in the upper half of Table 8, the coefficientsassociated with the TFCMAT variable are significant and positive for regressions 2and 3, at the 1% and 5% levels, respectively, but lack significance in regressions 1 and4. The coefficients associated with ln(AMTFCSIZ) are negative in regressions 2 and3, significant at the 1% level, and insignificant in regression 4. The ln(AMTFCSIZ)coefficient value is −15.56 for regression 2 and −13.97 for regression 3. The co-efficients associated with COLLAT are positive for regressions 3 and 4, significantat the 1% level. The COLLAT coefficient value is 52.71 for regression 3 and 65.68for regression 4. The coefficient associated with BWMD, in regression 4, is 10.17.The intercept ranges from 281.75 (regression 1) through 530.74 (regression 2). Forthe full sample of term loans (lower half of Table 8), two regressions have a pos-itive coefficient for TFCMAT, one displays a negative coefficient and one showsno significance for maturity. Once again, pooling undermines the robustness of ourresults.

Next, we turn to the revolving loans regression tests. Again, our benchmark forcomparison remains the paired tests that show no significant difference in spreadsand suggest that the term structure is flat. The paired sample regressions are alsoinsignificant, with one exception. The comparisons in Table 3 show that movingfrom the paired revolving loan sample to the full sample does not increase matu-rity significantly but increases size and reduces the percentage collateralized. Theseconfounding effects are associated with a switch in sign for the TFCMAT coef-ficient, which turns significantly negative in three of four full sample, revolvingloan regressions. Put another way, because they treat jointly determined variablesas exogenous, regressions can only imperfectly control for other variables impactingspreads while collinearity among these variables causes regression coefficients to lackrobustness.

The significant coefficients associated with the TFCMAT variable are generallypositive for the paired sample and negative for the full sample. In relation to thiseffect, note that the paired sample only includes firms that have loan deals withmultiple facilities, while the full sample permits single facility loan deals. Hence, analternative explanation for the differences in the signs of the coefficients is that firmsthat receive multiple facilities differ fundamentally from firms that receive singlefacilities.

4.2.2. Regression analysis, paired differences

In this section we report results obtained through replicating our regressionsfor the paired sample using differences in the variables rather than levels for spread,maturity, and size, while retaining the level as a measure of bond rating. We seekto develop a third comparison to the cross-sectional analysis and regressions pre-sented earlier. We further investigate these relations through performing regressions

72 A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77

Tabl

e8

Reg

ress

ion

anal

ysis

,ent

ire

sam

ple

Thi

sta

ble

show

sre

gres

sion

ofR

AT

EA

ISD

agai

nst

TFC

MA

T,ln

(AM

TFC

SIZ

),C

OL

LA

T,an

dB

WM

Dfo

rth

ete

rmsa

mpl

ean

dre

volv

ing

sam

ple,

for

both

the

pool

edpa

ired

and

full

sam

ples

.V

aria

ble

defi

nitio

nsar

eas

follo

ws:

RA

TE

AIS

Dis

the

rate

sal

l-in

-spr

ead-

draw

n.T

FCM

AT

isth

ete

rmfa

cilit

ym

atur

ity.

ln(A

MT

FCSI

Z)

isth

eln

(Am

ount

faci

lity

size

).C

OL

LA

Tis

the

colla

tera

lized

/non

colla

tera

lized

dum

my

vari

able

.BW

MD

isth

eB

orro

wer

Moo

dySe

nior

Deb

tR

atin

g.

Reg

ress

ion

Inte

rcep

tT

FCM

AT

ln(A

MT

FCSI

Z)

CO

LL

AT

BW

MD

Num

ber

ofob

serv

atio

nsA

dj-R

2Pr

(F)

Pair

edsa

mpl

eA

.Ter

m1

281.

75∗∗

∗−0

.02

3,11

2−0

.000

30.

784

253

0.74

∗∗∗

0.31

∗∗∗

−15.

56∗∗

∗3,

112

0.07

29<

0.00

013

461.

55∗∗

∗0.

20∗∗

−13.

97∗∗

∗52

.71∗

∗∗1,

708

0.08

94<

0.00

014

351.

78∗∗

∗0.

20−0

.13

65.6

8∗∗∗

−10.

17∗∗

∗52

80.

1683

<0.

0001

B.R

evol

ving

122

1.73

∗∗∗

−0.2

083

20.

0010

0.17

152

865.

50∗∗

∗0.

36∗∗

∗−3

9.38

∗∗∗

832

0.27

94<

0.00

013

609.

09∗∗

∗−0

.00

−28.

71∗∗

∗12

1.65

∗∗∗

532

0.36

68<

0.00

014

536.

94∗∗

∗0.

14−1

0.08

71.8

5∗∗∗

−13.

72∗∗

∗80

0.57

23<

0.00

01

Full

sam

ple

C.T

erm

124

2.21

∗∗∗

0.03

∗∗11

,817

0.00

050.

0107

284

4.34

∗∗∗

0.04

∗∗∗

−34.

78∗∗

∗11

,817

0.23

08<

0.00

013

568.

69∗∗

∗−0

.10∗

∗−2

3.48

∗∗∗

119.

55∗∗

∗7,

271

0.27

09<

0.00

014

547.

02∗∗

∗0.

03−8

.61∗

∗∗90

.91∗

∗∗−1

4.40

∗∗∗

1,68

80.

4468

<0.

0002

D.R

evol

ving

120

6.61

∗∗∗

−0.0

8∗∗∗

18,3

580.

0031

<0.

0001

293

8.43

∗∗∗

0.00

−41.

90∗∗

∗18

,358

0.31

41<

0.00

013

681.

62∗∗

∗−0

.29∗

∗∗−3

0.56

∗∗∗

108.

07∗∗

∗11

,833

0.39

82<

0.00

014

692.

52∗∗

∗−0

.21∗

∗∗−2

0.69

∗∗∗

77.6

1∗∗∗

−10.

28∗∗

∗2,

503

0.53

54<

0.00

01∗∗

∗ Ind

icat

esst

atis

tical

sign

ific

ance

atth

e0.

01le

vel.

∗∗In

dica

tes

stat

istic

alsi

gnif

ican

ceat

the

0.05

leve

l.∗ I

ndic

ates

stat

istic

alsi

gnif

ican

ceat

the

0.10

leve

l.

A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77 73

Table 9

Regression analysis, paired differences

This table shows regressions of Diff RATEAISD against Diff TFCMAT, Diff ln(AMTFCSIZ), and BWMDfor the term sample and revolving sample. Variable definitions are as follows: RATEAISD is the rates all-in-spread-drawn. Diff RATEAISD, Diff TFCMAT, and Diff ln(AMTFCSIZ) are the differences in the valuesof the RATEAISD, TFCMAT, and ln(AMTFCSIZ) variables, respectively, between the longer maturityand shorter maturity facilities for each pair. BWMD is the Borrower Moody Senior Debt Rating.

Diff Diff Number ofRegression Intercept TFCMAT ln(AMTFCSIZ) BWMD observations Adj-R2 Pr(F)

A. Term1 41.26∗∗∗ −0.60∗∗∗ 1765 0.0366 <0.00012 40.46∗∗∗ −0.54∗∗∗ −6.70∗∗∗ 1765 0.0461 <0.00013 86.51∗∗∗ −0.68∗∗∗ −2.07 −2.69∗∗∗ 490 0.1227 <0.00014 28.77∗∗∗ −8.64∗∗∗ 1765 0.0166 <0.00015 80.51∗∗∗ −5.63∗∗∗ −3.12∗∗∗ 490 0.0505 <0.0001

B. Revolving1 −1.98 0.07 433 −0.0017 0.59312 −0.14 0.16 −11.20∗∗∗ 433 0.0444 <0.00013 26.31 −0.39∗∗ −7.43∗∗ −0.78 77 0.0826 0.02574 4.48 −10.75∗∗∗ 433 0.0432 <0.00015 7.29 −8.05∗∗ −0.35 77 0.0427 0.0740

∗∗∗ Indicates statistical significance at the 0.01 level.∗∗ Indicates statistical significance at the 0.05 level.∗ Indicates statistical significance at the 0.10 level.

relating loan spreads to differences in maturity, differences in size, and riskiness. Theregression model tested is as follows:

Diff RATEAISD = a0 + a1 × Diff TFCMAT

+ a2 × Diff ln(AMTFCSIZ) + a3 × BWMD, (2)

where Diff RATEAISD, Diff TFCMAT and Diff ln(AMTFCSIZ) are the differencesin the RATEAISD, TFCMAT, and ln(AMTFCSIZ) variables between the longer andshorter maturity elements of each matched pair. As before we test the model es-timated using five methods, with each method adding variables in turn. Panel Cof Table 7 presents the correlations between the variables used in the regressiontests. The regressions are performed separately for the term and revolving loansamples.

The results of the regression tests are reported in Table 9 and demonstrate that forboth the term and revolving loan samples, differences in size (Diff ln(AMTFCSIZ))and bond rating of the pair (BWMD) are generally associated with significant neg-ative coefficients. For these two variables, the regressions of differences in Table 9validate the cross-sectional analysis. Turning to maturity, the significant coefficientsassociated with the Diff TFCMAT variable are uniformly negative (with only a single

74 A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77

coefficient significant for the revolving loan sample). As in our earlier regressionresults, inadequate controls for risk allow the impact of credit quality to dominate.

The intercept for these regression equations can be interpreted as the averagedifference in RATEAISD, given the various control variables. Thus, regression 5provides the average difference after controlling for facility size and riskiness. Theintercept is largest when both size and rating are controlled. This validates our keyresult that lenders are compensated for longer maturity loans.7

4.2.3. Regression analysis, rated versus unrated subsamples

This section reports results of tests conducted for the two subsamples that wereformed based on whether the DealScan database reported the debt rating. Droppingthe bond rating variable, BWMD, we rerun regressions 1 through 3. In particular,we seek to ascertain whether the coefficients obtained for some of the regressions inTable 8 are robust to the reporting of bond ratings in the DealScan database.

Beginning with the paired sample in Panel A of Table 10, comparison of theresults against those in the upper half of Table 8 demonstrates that the coefficientsfor size, ln(AMTFCSIZ), uniformly carry negative signs, while the coefficients forthe collateral dummy, COLLAT, uniformly carry positive signs, as in Table 8. Weconclude that the impacts of size and collateral on spreads are robust to the alter-native specification for risk controls introduced here. Focusing on the coefficientfor maturity (TFCMAT), a similar comparison between tables reveals that signifi-cant coefficients (in Panel A of Table 10) are generally positive, with the excep-tion of a negative coefficient associated with regression 1 for unrated revolvingloans.

Panel B of Table 10 carries on these comparisons for the full sample. We findthat the signs of the size variable and the collateral dummy are exactly as in Table 8.Interestingly, the significant TFCMAT coefficients associated with unrated facilitiesare negative for both the term and revolving loan samples, and are negative for the ratedrevolving sample as well. The finding of significant positive coefficients associatedwith the TFCMAT variable for the paired sample and negative coefficients for the fullsample is similar to our results for the regression tests of the entire sample. As notedearlier, an alternative explanation for the differences in the signs of the coefficientscould be that firms that receive multiple facilities differ fundamentally from firmsthat receive single facilities.

In summary, our regressions suggest two conclusions. First, credit spreads de-crease with borrower size, and are higher for collateralized borrowers. These relationsare robust to whether the debt rating is reported by the DealScan database. Second,similarly to pooling across maturities, pooling across credit ratings masks the impactof the credit-quality effect.

7 We thank Larry Wall, the editor, for noting this interpretation.

A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77 75

Table 10

Regression analysis, rated versus unrated subsamples

This table shows regressions of RATEAISD against TFCMAT, ln(AMTFCSIZ), COLLAT, and BWMDfor the term sample and revolving sample, based on the pooled paired sample (Panel A) and the full sample(Panel B), for rated and unrated loans. Variable definitions are as follows: RATEAISD is the rates all-in-spread-drawn. TFCMAT is the term facility maturity. ln(AMTFCSIZ) is the ln(Amount facility size).COLLAT is the collateralized/noncollateralized dummy variable. BWMD is the Borrower Moody SeniorDebt Rating.

Panel A: Paired sample

Number ofRegression Intercept TFCMAT ln(AMTFCSIZ) COLLAT observations Adj-R2 Pr(F)

Term: Rated1 227.47∗∗∗ 0.53∗∗∗ 834 0.0282 <0.00012 339.36∗∗∗ 0.56∗∗∗ −6.21∗∗∗ 834 0.0374 <0.00013 223.39∗∗∗ 0.34∗∗ −3.34 84.59∗∗∗ 528 0.0977 <0.0001

Term: Unrated1 292.54∗∗∗ −0.11 2263 0.0007 0.11252 559.47∗∗∗ 0.27∗∗∗ −17.08∗∗∗ 2263 0.0802 <0.00013 508.21∗∗∗ 0.18∗ −15.59∗∗∗ 34.98∗∗ 1176 0.0850 <0.0001

Revolver: Rated1 120.37∗∗∗ 0.58∗ 146 0.0126 0.09352 807.12∗∗∗ 0.82∗∗∗ −37.87∗∗∗ 146 0.3186 <0.00013 444.99∗∗∗ 0.41 −20.33∗∗∗ 114.83∗∗∗ 80 0.4135 <0.0001

Revolver: Unrated1 239.77∗∗∗ −0.30∗ 686 0.0037 0.05932 848.39∗∗∗ 0.26∗ −37.98∗∗∗ 686 0.2194 <0.00013 632.77∗∗∗ −0.07 −30.18∗∗∗ 125.22∗∗∗ 428 0.3474 <0.0001

Panel B: Full sample

Number ofRegression Intercept TFCMAT ln(AMTFCSIZ) SECURE observations Adj-R2 Pr(F)

Term: Rated1 108.69∗∗∗ 1.59∗∗∗ 2789 0.1714 <0.00012 681.99∗∗∗ 1.33∗∗∗ −30.05∗∗∗ 2789 0.2670 <0.00013 387.11∗∗∗ 0.29∗∗∗ −15.75∗∗∗ 139.12∗∗∗ 1688 0.3668 <0.0001

Term: Unrated1 261.24∗∗∗ −0.01 9020 0.0000 0.28552 815.81∗∗∗ 0.01 −32.83∗∗∗ 9020 0.2062 <0.00013 595.63∗∗∗ −0.23∗∗∗ −23.83∗∗∗ 103.86∗∗∗ 5574 0.2113 <0.0001

Revolver: Rated1 157.76∗∗∗ −0.25∗∗∗ 4165 0.0023 0.00112 922.83∗∗∗ 0.31∗∗∗ −42.26∗∗∗ 4165 0.2894 <0.00013 604.26∗∗∗ −0.12 −26.84∗∗∗ 109.42∗∗∗ 2503 0.4698 <0.0001

Revolver: Unrated1 222.89∗∗∗ −0.06∗∗∗ 14169 0.0025 <0.00012 922.36∗∗∗ 0.00 −40.84∗∗∗ 14169 0.2662 <0.00013 697.68∗∗∗ −0.33∗∗∗ −31.52∗∗∗ 109.21∗∗∗ 9324 0.3478 <0.0001

∗∗∗ Indicates statistical significance at the 0.01 level.∗∗ Indicates statistical significance at the 0.05 level.∗ Indicates statistical significance at the 0.10 level.

76 A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77

5. Conclusions

This paper examines the relations between corporate loan spreads and maturity.Using the Loan Pricing Corporation DealScan database, we empirically test whetherlenders are compensated for engaging in longer maturity loans (tradeoff hypothesis)or, alternatively, limit exposure through forcing riskier borrowers to take short-termloans (credit-quality hypothesis).

We create a large sample of matched pairs of loans of different maturities madeby the same lead and participant banks, on the same day to the same company. Sincethe risk of the borrower is identical by construction within each pair, we are able tocontrol borrower and lender characteristics. Our matched pairs analysis demonstratesthat the longer loans have higher spreads than shorter loans, supporting the tradeoffhypothesis. For term loans we find a significant, positive relation, but no significantrelation for revolving loans. This suggests that the term structure of credit spreads isflat for short-maturity revolving loans and then becomes upward sloping for long-maturity term loans. These paired tests are convincing as facility size, collateral, andbond rating as well as lead and participant banks are controlled perfectly through thematching technique we employ.

Further tests suggest that the credit-quality hypothesis is also supported. Settingaside the matching technique, and with it the control on credit quality, we conduct re-gressions on a larger sample. The significant coefficients associated with the maturityvariable are generally positive for the paired sample and negative for the full sam-ple. This suggests that both tradeoff and credit-quality effects are present. However,there is a possibility that endogeneity and collinearity issues disrupt these inferences.An alternative explanation for the differences in the signs of the coefficients is thatthe firms in the paired sample that receive multiple facilities are somehow differentfrom firms in the full sample receiving single facilities. Exploration of this alternativeexplanation is left for future research.

When negotiating with an individual firm, a bank is willing to accept a spread-maturity tradeoff, causing the borrower to face an ascending term structure of loanspreads. However, when structuring its loan portfolio, the bank can limit exposurethrough truncating the tradeoff by forcing riskier borrowers to take short-term loans.

References

Barclay, M.J. and C.W. Smith, Jr., 1995a. The maturity structure of corporate debt, Journal of Finance 50,609–631.

Barclay, M.J. and C.W. Smith, Jr., 1995b. The priority structure of corporate liabilities, Journal of Finance50, 899–917.

Barnea, A., R. Haugen, and L. Senbet, 1980. A rationale for debt maturity structure and call provisions inan agency theoretic framework, Journal of Finance 35, 1223–1234.

Berger, A. and G. Udell, 1990. Collateral, loan quality, and bank risk, Journal of Monetary Economics 25,21–42.

Coleman, A.D.F., N. Esho, and I.G. Sharpe, 2002. Do bank characteristics influence loan contract terms?Working paper, University of New South Wales.

A. A. Gottesman and G. S. Roberts/The Financial Review 39 (2004) 55–77 77

Diamond, D., 1993. Seniority and maturity of debt contracts, Journal of Financial Economics 53, 341–368.Dennis, S., D. Nandy, and I.G. Sharpe, 2000. The determinants of contract terms in bank revolving credit

agreements, Journal of Financial and Quantitative Analysis 35, 87–110.Flannery, M., 1986. Asymmetric information and risky debt maturity choice, Journal of Finance 41, 18–38.Helwege, J. and C.M. Turner, 1999. The slope of the credit yield curve for speculative-grade issuers,

Journal of Finance 54, 1869–1884.Houston, J.F. and C.M. James, 2001. Do relationships have limits? Banking relationships, financial con-

straints, and investment, Journal of Business 74, 347–374.John, K., A.W. Lynch, and M. Puri, 2002. Credit ratings, collateral, and loan characteristics: implications

for yield, Journal of Business.Kale J. and T.H. Noe, 1990. Risky debt maturity choice in a sequential game equilibrium, Journal of

Financial Research 13, 155–166.Loan Pricing Corporation, 1999. DealScan Users’ Manual (Loan Pricing Corporation, New York).Myers, S., 1977. The determinants of corporate borrowing, Journal of Financial Economics 5, 147–176.Stohs, M.H. and D.C. Mauer, 1996. The determinants of corporate debt maturity structure, Journal of

Business 69, 279–312.Strahan, P.E., 1999. Borrower risk and the price and nonprice terms of bank loans. Working paper, Federal

Reserve Bank of New York.