Risk in Capital Structure Arbitrage 2006

Download as pdf or txt
Download as pdf or txt
You are on page 1of 46

Risk in Capital Structure Arbitrage∗

Stephen M. Schaefer Ilya A. Strebulaev


London Business School Graduate School of Business
Stanford University

First version: May 2006


We would like to thank Inquire Europe for their generous financial support and European Credit Management
for help with the data. Svetlana Kolvalskaia provided valuable research assistance. We are responsible for all
remaining errors. Address for correspondence: Graduate School of Business, Stanford University, 518 Memorial
Way, Stanford, CA, 94305. E-mail: [email protected].
Risk in Capital Structure Arbitrage

Abstract
Capital structure arbitrage attempts to profit from inconsistencies in the
relative pricing of a firm’s liabilities and typically combines the firm’s
straight or convertible with its equity. Using a large database on US
corporate debt, we examine the risks of portfolios of capital structure
arbitrage positions under a variety of hedging strategies but focussing
particularly on hedges that include the issuing firm’s equity. We also ex-
amine strategies that include variables, such as the Fama-French factors,
momentum and the S&P, that previous studies have found to be signifi-
cantly related to corporate debt returns. Second, we use index data to ask
whether the relation between corporate debt and these “non-structural”
variables is a feature of only recent data or is also evident in long-run
data. Finally, we examine the surprisingly low sensitivity of corporate
debt and, particularly, high yield debt, to interest rate movements.

Keywords: Capital structure arbitrage, hedging, risk management, structural models,


credit risk.
I Introduction

Capital structure arbitrage is a trading strategy that attempts to exploit mispricing between

a company’s liabilities, most commonly between equity and straight or convertible debt. In

recent years such strategies have become increasingly popular, particularly among hedge funds,

possibly as a result of the development of the credit default swap market that has allowed market

participants to take short positions in a credit risk more easily (Currie and Morris (2002)).

While the mechanics of capital structure arbitrage are now widely understood, there is little

formal evidence on its risk. Structural models provide one framework for analyzing the risk but,

in the classical models of Merton (1974), Leland (1994), Leland and Toft (1996) and others, a

correctly constructed triangular hedge between a company’s debt, its equity, and riskless debt

should be risk free. On the other hand, recent events in the credit markets suggest that there are

times when capital structure arbitrage may be very risky. For example, when S&P downgraded

Ford and GM in early 2005, Kirk Kerkorian chose that moment to announce a tender offer

for around 5% of GM’s stock, resulting in the prices of GM’s debt and equity going opposite

directions and creating substantial losses for capital structure arbitrageurs.1

The academic literature on capital structure arbitrage is limited. Yu (2005) analyzes conver-

gence trades involving credit default swaps (CDS) and equity. He finds these “quite risky” and,

in particular, prone to large losses when CDS spreads rise quickly. In an analysis of a number of

fixed-income arbitrage strategies, Duarte, Longstaff, and Yu (2005) analyze CDS-based capital

structure arbitrage and find that these produce promising Sharpe ratios of approximately 0.8.

Agarwal and Naik (2000) analyze the performance of hedge funds following a variety of strategies

including capital structure arbitrage. Schaefer and Strebulaev (2005) investigate hedge ratios of
1
A report by Reuters on 19 May 2005 said: “Market gyrations after a cut in Ford’s and General Motors’ credit
ratings turned predictive models on their heads and, according to market talk, produced huge losses at hedge funds
that had effectively sold the companies’ equity and bought their debt”.

2
corporate debt against equity and find that, at the level of individual bonds, the quality of the

hedge is quite poor, particularly for bonds with higher levels of credit risk.

Unlike Yu (2005) and Duarte et. al. (2005), the focus of this paper is not the profitability of

capital structure arbitrage but its risk and, specifically, its risk at a portfolio level. While Schaefer

and Strebulaev (2005) and others have found that the returns on an individual corporate bond

are not well explained by the issuing firm’s equity and riskless debt, it remains an open question

as to whether a significant fraction of the remaining risk is diversifiable. Encouragingly, in his

analysis of CDS, Yu (2005) finds that “. . . when the individual trades are aggregated into monthly

capital structure arbitrage portfolio returns, the strategy appears to offer attractive Sharpe ratios

. . . ”. (Yu (2005), p. 32).

The focus of this study is, then, the risk of capital structure arbitrage, and the effectiveness of

alternative strategies for managing this risk.2 In this context, it is important to point out that our

calculations are carried on portfolios that are chosen in a mechanical, i.e., passive manner while,

in practice, capital structure arbitrageurs choose positions because they judge the debt and equity

to be relatively mispriced. It is possible that these latter positions may be systematically more

or less risky than the passive positions we study precisely because of the presence of mispricing

but we lack the data on actual capital structure arbitrage positions that would be required to

decide this question.

The paper addresses three main questions. First, how much of the risk of hedged positions

is diversifiable? This question is important because the earlier work of Schaefer and Strebu-

laev (2005) shows that, while the statistical relation between corporate debt and equity can be

identified quite well in large samples, the degree of sensitivity is typically quite small and the
2
We – along with capital structure arbitrageurs – would also like to be able to say something about the average
profitability of this activity, i.e., the αj ’s, however, the relatively short period covered by our data makes this
impractical.

3
fraction of the variability in return accounted for by equity is also small. For example, Table

IV of Schaefer and Strebulaev reports that the average “beta” of BBB debt against the issuing

firm’s equity is 0.04. However, even though the t-statistic on this average is large (approximately

13.0) this arises because the size of the cross-section is large rather than because, at the level of

an individual bond, equity accounts for a large fraction of the return variability. Indeed, in these

same regressions Schaefer and Strebulaev report that, compared with a regression that includes

only a Treasury bond, including equity increases the R2 by only 5% (from 43% to 48%).

While this small increase in the R2 shows that equity is not a very effective hedge at the

level of the individual corporate bond, the high t-statistic for the average beta suggests that at

least some significant part of the remaining variability is diversifiable. This is the issue that we

examine in this paper by computing the risk of portfolios of capital structure arbitrage positions,

i.e., where many bonds are hedged simultaneously against riskless debt and the many underlying

equities.

In discussing these results we also examine the relation between the risk of hedged positions

and the credit quality of corporate bonds. Previous results on credit spreads on individual bonds

suggests that, for bonds with low credit ratings, equity and riskless debt account for less than

50% of the return variability. For example, Collin-Dufresne et al. (2001) find that, for the BBB

and BB categories, less than 20% of the variability of credit spreads is accounted for by the

return on the firm’s equity and changes in the riskless yield curve. We therefore investigate how

the diversifiability of the remaining risk varies with credit risk.

Of course, as many authors have pointed out (e.g., Collin-Dufresne et al. (2001)), finding that

equity and riskless debt provide a poor hedge for corporate debt is inconsistent with the standard

structural approach to credit risk. Accordingly, the second issue we address is the contribution of

other instruments to the management of the risk in capital structure arbitrage positions. Collin-

4
Dufresne et al. (2001) and Elton et al. (2001) find that corporate bond returns are related to

the Fama-French HML and SMB factors. Again, since these results have been obtained at the

level of individual bonds, we assess their importance in managing risk at the portfolio level.

A natural way in which to estimate the composition of the hedge portfolio is to estimate a

regression of the excess return on the corporate bond on a vector of excess returns on hedge

instruments. The betas obtained in this way represent the positions in the hedging instruments

that provide the minimum variance in-sample hedge. However, to be implementable, the hedge

ratios must be estimated using only information available up to the time of portfolio formation.

Therefore, as well as an in-sample analysis we also calculate the corresponding results when the

betas used to construct a hedged position are estimated using only prior data (out-of-sample

analysis).

Much of the empirical work on credit risk has focussed on changes in credit spreads (e.g.,

Collin-Dufresne et. al. (2001)) rather than on rates of return. Although this difference may

appear minor, the sensitivity of credit spreads to the rates of return on hedging instruments has

no clear portfolio construction interpretation. In contrast, the corresponding sensitivities of the

rate of return on corporate debt give the composition of the minimum variance hedging portfolio.

Perhaps because of the difficulty of linking an analysis of credit spreads to the normative problem

of hedge construction, none of these studies investigate the out-of-sample properties that we give

in Section III.5 3 .

The analysis in Section III provides what is perhaps the main result of the paper which is
3 ∗
The excess return on a position in a corporate bond hedged with a Treasury bond of equal duration, rj,t , is
(approximately) equal to:

rj,t = M Dj ∆y − M Dj ∆yT = M Dj ∆Sj
where M Dj is the modified duration of both the corporate and Treasury bonds, ∆y the change in the yield-to-
maturity on the corporate, ∆yT the change in the yield on the Treasury and ∆Sj the change in the corporate minus

Treasury spread. Thus rj,t , the excess return on a position in a corporate bond hedged against the corresponding
Treasury bond is – approximately – proportional to the change in the credit spread.

5
that much of the risk in hedging individual corporate bonds against the equity of the issuing

firm and the riskless term structure turns out to be diversifiable. What this means is that the

return on a portfolio of corporate bonds and is explained by the riskless term structure and a

portfolio containing the equity of the issuing firms to a much greater extent than has previously

been observed. Although our analysis is in no sense a test of the structural approach to credit

risk, this finding is clearly supportive of the structural approach.

In Section IV we turn our attention to the variables such as VIX and the Fama-French SM B

and HM L factors that previous studies have found to be related to credit spreads. In almost all

cases these studies have been based on individual bond data and, as a result of data availability,

have usually spanned a period of less than ten years. In the second part of our study we use

long-run data on bond indices to establish whether these variables have a long-run relation to

corporate bond returns or whether the results that others have obtained are particular to the

relatively short and recent periods that have been studied. Our results show that the sensitivity of

corporate debt to the Fama-French factors is indeed a persistent feature of the data, particularly

for bonds with lower credit ratings.

Finally, in Section V we turn our attention to the interest rate sensitivity of corporate debt.

It has been reported elsewhere in the literature (for example, see Fons (1990), Gautier and

Goodman(2003)) that the empirical interest rate sensitivity, or duration of corporate debt is

significantly lower than its McCaulay duration. Schaefer and Strebulaev (2005) show that the

lower empirical duration is not explained by the Merton model even with the inclusion of interest

rate uncertainty. Section V therefore provides some further evidence on this puzzle. Our results

suggest, first, that the relation between the empirical interest rate sensitivity of corporate debt

and conventional (McCaulay) duration varies significantly across bonds and, second, that at least

some of this variation is explained by a bond’s maturity, its credit exposure and whether it has

6
an investment grade credit rating. Finally, Section VI concludes.

II Sample Selection and Data Description

In this section we describe the individual bond data used in the first part of the paper and the

bond index data used in the second part.

We use monthly prices on individual corporate bonds that are included either in the Merrill

Lynch Corporate Master index or the Merrill Lynch Corporate High Yield index. These indexes

include most rated U.S. publicly issued corporate bonds. The data cover the period from Decem-

ber 1996 to December 2003 and contain more than 380,000 bond-month observations on about

2140 issuers and 10,400 issues. Detailed information on each bond is obtained from the Fixed

Income Securities Database (FISD) as provided by LJS Global Services; equity and treasury

bond returns are from CRSP. For riskless rates of return we use constant-maturity US Trea-

sury monthly returns as reported by CRSP. Further details are given in Schaefer and Strebulaev

(20045).

Our analysis is performed on a subsample that includes bonds satisfying the following criteria:

(1) we can match the bond return data with CRSP and COMPUSTAT (this allows us to use

about 62% of the total number of observations); (2) the bond is issued by a U.S. company and

denominated in $U.S.;4 (3) it is possible to match unambiguously the bond issuer with a company

in CRSP using the CUSIP; (4) the bond is issued by a non-financial corporation; (5) the bond

is straight (i.e. does not have any option-like embedded features); (6) the bond has at least 25

consecutive monthly price observations; (7) the bond has an initial maturity of at least four years.

Table I gives summary statistics for the subsample of 1370 bonds that meet these criteria.
4
More specifically, the company is of the U.S. origin according to the FISD definition. In particular, its head-
quarters should be located in the U.S. and it is subject to U.S. legal practice.

7
All rating categories (from AAA to CCC) are represented in the sample. The categories with the

largest numbers of bonds are A and BBB (603 and 539 respectively) but there are also quite large

samples of AA (127) and BB (50) bonds. The mean time-to-maturity of decreases – as might be

expected – for lower credit ratings but it is interesting to note that the median maturity actually

increases. The median nominal amount in issue declines from $300 million for AAA to $175

million for B.

The index data used in Section IV are the Merrill Lynch (ML) investment grade and high yield

indices and their associated sub-indices. For the Corporate Master Index, qualifying bonds must

have at least one year remaining term to maturity, a fixed coupon schedule, at least $150 million

outstanding and be rated investment grade based on a composite of Moody’s and S&P ratings.

The index is re-balanced on the last calendar day of the month. Issues that meet the qualifying

criteria are included in the index for the following month. Issues that no longer meet the criteria

during the course of the month remain in the index until the next month-end re-balancing at

which point they are dropped from the index. For the US High Yield Master Index the minimum

nominal amount outstanding is $100 million and bonds must be rated below investment grade,

again based on a composite of Moody’s and S&P. Our data on the Corporate Master Index start

in December 1975 and on the High Yield Index in September 1986. Both indices are rebalanced

on the last calendar day of the month.

We also report results on investment grade sub-indices for (i) individual credit ratings (AAA,

AA, A and BBB) and (ii) maturity intervals (1-5 years, 5-10 years, 10-15 years and over 15 years).

For the sub-investment grade we use sub-indices for the BB, B and CCC rating categories.

8
III Hedging Portfolios of Corporate Debt

III.1 Estimating Hedge Portfolio Composition

Consider a capital structure arbitrage position at time t in which a corporate bond issued by

firm j is hedged with number of instruments. The latter may include firm j’s equity, one or more

riskless (i.e., Treasury) bonds, the S&P index and so forth. The excess return on this position

may be written as:

rj,t − β j rth (1)

where rj,t is the excess return on the corporate bond from time t − 1 to t, rth is the vector of

excess returns on the hedging instruments and β j is a vector of amounts (in dollars) of the hedging

instruments that are sold short per dollar of firm j’s debt that is held. We refer to the elements

of β j as “hedge ratios”; accordingly, the second term in this expression, β j rth , represents the

excess return on the hedging portfolio and the entire expression the excess return on the hedged

portfolio.

The excess return on this hedged position may also be regarded as the hedging error which,

without loss of generality, may be written as the sum of its mean value αj and a zero mean

variable εj,t :

αj + εj,t = rj,t − β j rth , (2)

or

rj,t = αj + β j rth + εj,t . (3)

9
A natural way to estimate the hedge ratios is via regression. This has the disadvantage that it

treats the sensitivities, β j , as constants while, in general, they are not. Nonetheless the approach

has several advantages compared with the alternative of using a model. The first, clearly, is

simplicity. Second, it allows us to include in a straightforward way hedging instruments such as

the S&P and the Fama-French factors that find no place in standard models.

Our approach, therefore, has two steps. First, at the level of the individual bond, we estimate

the hedging portfolio composition and the hedging errors. Second, we aggregate these errors to

calculate the portfolio hedging error and its risk.

III.2 Implementation

In this Section we describe the calculation of the risk of portfolios of capital structure arbitrage

positions, i.e., positions in corporate debt hedged with equity.

The first step is to identify those bonds that are eligible for inclusion in month t. We define

these as bonds that (1) were in the ML database in month t; (2) had at least four years to

maturity in month t; (3) had at least 20 (monthly) return observations in the data set; (4) had

no call or other option-related features, i.e., were “straight”.

Our portfolio formation rules must accommodate the fact that a bond rated AA, say, in

January 1997 may, at some later date, have a different rating. We must also deal with entry to

and exit from the database. Accordingly, we present results based on two alternative eligibility

rules. Under the first (“5A”) a bond is eligible for inclusion in the portfolio for credit rating

category K in month t if it belongs to K at the end of the first month in our analysis (January

1997, month τ ) and no account is taken of (i) subsequent changes in rating or (ii) subsequent

entry into the database of bonds belonging to K. Under the second, simpler rule, (“5B”), a bond

is eligible for inclusion in the portfolio for credit rating category K in month t if it belongs to

10
credit rating category K in month t.

Applying one of these two conditions completes the set of five eligibility conditions. The

results for these two cases turn out to be substantially similar.

III.3 Hedged Portfolio Returns: In-Sample

The in-sample results are calculated as follows. For each bond in the database meeting criteria

(1) to (4) above, (i.e., “eligible”) we first estimate the hedge ratios from equation (3) by regressing

the excess return on each eligible bond j in month t, rj,t , on the excess returns on a set of hedging

instruments. These regressions use all the available data from January 1997 to December 2003.

We run the in-sample procedure in the following way. For each portfolio formed in month t, we

estimate hedge regressions (as specified in the next subsections) for each bond in this portfolio

between t and either December 2003 or the last date the bond is recorded in the data set. We

next calculate the portfolio hedging errors by averaging the hedging errors for individual bonds

within each month and finally calculate the variance of the portfolio hedging errors over time .

We follow this procedure for each portfolio formed between January, 1997 and December, 1999

and report the average results over all portfolios.

We present results for four alternative sets of hedging instruments: (i) a 10-year and a one-

year Treasury bond; (ii) set (i) together with the firm’s equity; (iii) set (ii) together with the

S&P, the Fama-French factors (SMB & HML) and the momentum mimicking portfolio; (iv) set

(i) together with the S&P estimate.

Let KA B
t [Kt ] denote the set of bonds that is eligible for inclusion in the hedged portfolio

for credit category K in month t under rule 5A [5B]. The corresponding number of bonds is

denoted NtA [NtB ]. Under rule 5A, the hedged portfolio at date t invests a fraction 1/NtA in

each of the eligible bonds and takes a short position in the hedging instruments of 1/NtA times

11
 (equation (3)). The (demeaned) portfolio hedge
the corresponding vector of estimated betas, β j

error for credit category K in month t under rule 5A, εK,P,A


t , is therefore given by:

   h
 
εK,P,A = 1/NtA 
rj,t − β j rt − 1/NtA αj (4)
t
j∈KA
t j∈KA
t

= 1/NtA εj,t (5)
j∈KA
t

The expressions under rule 5B are exactly equivalent.

We now measure the time series variance of the portfolio hedging error, εK,P,A
t , and express

this as a fraction of the variance of the unhedged portfolio excess returns. This fraction, which

we denote H K,A measures the (in)effectiveness of the hedge:

V ar(εK,P,A )
H K,A = t
, (6)
V ar(rtK,P,A )

where:


rtK,P,A = 1/NtA rj,t . (7)
j∈KA
t

The smaller value of H K,A the better the hedge.

III.4 Results: In-sample

Table II shows the results on in-sample hedging effectiveness using the eligibility rule 5A. The first

column, “All”, gives the results for a portfolio containing all the eligible bonds; the remaining

columns gives the results for particular credit ratings.

The first four rows show H K,A , defined above, for the four alternative sets of hedging instru-

12
ments. The final two rows show the maximum and minimum number of bonds in the portfolio

in any one month.

For a portfolio composed of all the eligible bonds hedged against one- and ten-year Treasury

bonds only, the value of H K,A , the variance of the portfolio hedge errors as a fraction of the

variance of portfolio returns is 49.6%. Including individual firm equity reduces this fraction by

almost a half to 27.5%. This is equivalent to R2 of over 70%. Including the S&P, the Fama-French

HML and SMB factors and the momentum mimicking factor reduces the factional error variance

still further to 18.2%.

In the fourth row the hedging instruments are the one- and ten-year Treasury bonds and the

S&P index. Comparing this result to the second row we can see that replacing firm equity as

a hedging instrument with the S&P produces a much worse hedge. When the S&P is used, the

reduction in variance from the “Treasury Curve only” case for the “All” bond portfolio is only

around 10% (49.6% to 39.6%) as compared with over 20% when firm equity is used.

It is no surprise that the importance of firm equity in hedging is much greater for lower

credit rating categories and, in particular, non-investment grade bonds than for bonds with little

credit and, therefore, equity exposure. For AAA and AA the inclusion of firm equity reduces

the percentage error variance from 18.7% to 17.9% and 23.3% to 19.4% respectively. Hedging

with the S&P rather than firm equity gives 15.2% for AAA and 19.6% for AA. (The hedging

error variance for AAA is actually lower with the S&P than with firm equity but the sample

size is very small). However, for BBB and even more for BB, the inclusion of firm equity makes

a very substantial difference. For BBB the error variance percentage is 47.2% when hedging

against the Treasury curve alone and this falls to 27.5% when firm equity is included. For BB

the corresponding figures are 92.9% and 26.9%, in other words including firm level equity reduces

the error variance percentage by over 60%. The inclusion of the “other” instruments (the S&P,

13
Fama-French factors and momentum) reduces the error variance further (by around 5% to 10%

of the total variance) but the improvement is small compared to the inclusion of firm equity.

The results for the alternative eligibility rule (“5B”), given in Table III, are not substantially

different and, although not quite as large as in Table II, the reduction in the risk when firm equity

is included rather than the S&P is still substantial. For the BBB category, for example, hedging

against the Treasury curve alone produces a hedging error variance of 64.4% of the unhedged

returns. Including firm equity reduces this by over 30%, i.e., by almost half, to 33.1% while using

the S&P gives a reduction of only around 8% to 56.2%. For BB the results are similar: including

firm equity reduces the error variance from 86.0% (using only Treasury bonds) to 47.2%, while

with Treasury bonds and the S&P the percentage error variance falls by less than 10% to 77.1%.

In summary, therefore, Tables II and III show that there is a substantial benefit to hedging

bonds in the lower credit rating categories with the equity of the issuing firm rather than simply

with a broad equity index. Since firm equity provides a relatively poor hedge at the level of the

individual firm, the results imply that much of the risk of capital structure arbitrage positions is

diversifiable.

The results also provide some modest support for the structural view of credit risk under

which, of course, changes in the value of firm equity and in the riskless yield curve should explain

100% of the return variability on corporate debt. While certainly not upheld precisely in the

data, this implication holds much better at the portfolio level than for individual bonds. What

accounts for the noise in the returns at the individual bond level remains unclear: it could be a

“missing factor”, “noisy” mispricing or simply noise in the price data.

14
III.5 Out-of-sample hedging error estimation

For the out-of-sample estimates we start by forming the same portfolio as for the in-sample

estimation (we report here only portfolios constructed under rule 5A) at date τ and choose only

the bonds that are in the database for at least three years. For each bond we use observations

up to three years from month τ and run the following individual regressions:

rj,t = αj,τ ,τ +36 + β j,τ ,τ +36 rth + εj,t , t = τ , ..., τ + 36. (8)

In other words, time-series regressions are run for 36 months, and the notation β j,τ ,τ +36 underlines

the fact that we estimate the hedging regression over the interval from τ to τ + 36.

Then, for each bond that has more than 36 observations we estimate the out-of-sample hedging

error for month τ + 37 in the following way:

 
εj,τ +37 = rj,t − α 
 j,τ ,τ +36 + β h
j,τ ,τ +36 τ +37 .
r (9)

We then re-estimate the hedging regression but including the 37th month and then calcu-

late the out-of-sample hedging error for the 38th month. We continue this procedure until the

last month the bond is recorded in the data set. We then compute the out-of-sample portfolio

hedging errors and their time series variance. As for the in-sample estimation, we form overlap-

ping portfolios between January, 1997 and December, 1999, and report averages over all these

portfolios.

Table IV contains our out-of-sample results. The risk of these positions stems from two

sources: first, as before, fluctuations in the bond price that are not accounted for by the hedge

instruments and second, errors in predicting the hedge ratios. These results simply use the raw

15
hedge ratios from the regressions on prior data; these are inevitably noisy.5

The important result that emerges from this analysis is that the broad pattern of results is

quite similar to that in Tables II and III. The error variances are, inevitably, larger in Table

IV, but, again particularly for bonds with higher credit exposure, hedging with firm equity

substantially dominates hedging with the S&P. For BBB, for example, hedging with firm equity

reduces the error variance from 79.7% (yield curve only) to 53.5% while, with the S&P the

corresponding reduction is only just over 1% to 78.3%. For BB, the difference is, once again,

even more dramatic. Using firm equity the reduction from the “yield curve only” case is from

106% to 47.6% while with the S&P it is reduced only to 92.2%. In all these case, just as in the

earlier results, including the Fama-French factors etc. reduces the risk further but the impact is

small compared with the effect of including firm equity.

IV Analysis of Bond Indices

The results of the previous section demonstrate that a firm’s equity provides a better hedge for

its debt than has previously been supposed in the literature. While this is, as we have pointed

out, supportive of the structural model of credit risky debt, the sensitivity of corporate debt to

instruments such as the Fama-French factors is not predicted in the standard structural “story”.6

Both Collin-Dufresne et al. (2001) and Elton et al. (2001) – the two studies that have so

far examined this question – use the Warga (1998) database and analyze very similar periods.

This leaves open the question of whether the sensitivity of corporate debt to the Fama-French
5
It should be possible to improve the precision of the predicted hedge ratios by making use of cross-sectional
information and later versions of this paper will attempt to do this. Thus the results in Table IV should be viewed
as representing a lower bound on the effectiveness of the hedging strategies.
6
While the “other” variables such as the Fama-French factors do not appear in standard structural models,
it is not difficult to envisage ways in which other factors could be important. In the standard structural model
uncertainty derives only from the value of the firm and the riskless yield curve. However, it is possible that, for
example, one or both of the recovery rate and the level of the default boundary exhibit variability that is at least
to some degree dependent on the Fama-French factors.

16
and other factors is persistent phenomenon or is limited to a particular period. This is the issue

that we now address and, because reliable long-run price data on individual corporate bonds are

not available, we use indices. For investment grade bonds, indices have been available since the

mid-1970s and, for high yield debt, since the mid-1980s.

Tables V and VI give summary statistics on the equity related data and the Treasury data

respectively and Tables VII and VIII on the Merrill Lynch corporate bond indices. The difference

between these last two tables is the period covered by the data. In Table VII the statistics are

computed using all the available data for each series. However, although all the series have the

same end date, they have different start dates; for example, the broad investment grade index

(column one) is available from December 1975 while the single credit category non-investment

grade indices (columns 11 - 13) are available only from January 1997.7 Accordingly, Table

VIII presents summary statistics for the maximum common period of the data that is available

(January 1997 - December 2004).

Section V discusses the interest rate sensitivity of corporate bonds in some detail. Here

the issue is their sensitivity to the remaining three factors and Table IX shows the results of

regressing the one-month excess return on each of the indices on the return on the 10-year

benchmark Treasury (RF 10Y ), the return on the S&P and the two Fama-French factors, HM L

and SM B. These sensitivities are significant in almost all cases. For the broad investment grade

(IG All ) index, for example, the t-statistic on the S&P is 5.53 and on HML and SMB it is 3.18

and 3.68 respectively. Both the sensitivities and their significance are higher for the broad junk

bond index (HY all ) and this pattern is mirrored for the sub-indices.8
7
The start date for each index is given in row 7.
8
One reason that corporate debt may be sensitive to the S&P, HML and SMB is because the firm’s underlying
assets (and therefore its equity) are sensitive to these factors. In regressions for individual bonds, including the
firm’s equity controls for this channel of sensitivity and it is possible to distinguish between sensitivity to SMB,
say, that derives from the firm’s assets and sensitivity that is specific to the bond. However, no such control is
possible in an analysis of a bond index and so it is not possible to say for sure whether the sensitivity to SMB etc.
derives from firm’s underlying assets or is specific to the corporate bond market.

17
Table X shows the corresponding results for the period from January 1997 to December 2004;

this is he longest period for which data are available for each of the 13 indices. Overall, the

sensitivities are similar to those in Table IX and the significance levels are only slightly lower

despite the much shorter time period for investment grade indices (and the correspondingly

smaller number of observations). In Table XI we re-run the regression on all available monthly

data including changes in VIX, the inflation rate (CPI ) and the momentum mimicking factor

(UMD). The sensitivities to the S&P, HML and SMB – particularly for junk bonds – are not

significantly changed.

A concern consistently expressed about data on corporate bonds is their low level of liquid-

ity and the consequent potential for observed prices to be unreliable and, in particular, non-

synchronous (Fisher (1959), Chacko et. al. (2005)).

To study the robustness of our results with respect to the illiquidity of corporate bonds, we

perform the following experiment. We assume that we observe each of the independent regressors

every month, but observe the true value of corporate bonds less frequently. In particular, at date

0 (the starting date of the index or of the regression subperiod) we observe the true value of the

index. In month t the observed value of the index is It . Then, in month t, with probability p,

∗ , is observed, and, with probability 1 − p the observed value is equal


the true value, It+1 = It+1

to the last observed value, It+1 = It . If p = 1, corporate bonds are as liquid as regressors. In the

benchmark robustness case, we take p = 4/12, i.e. on average the true value is observed every

three months. We then run regressions many times (simulating the p process anew with each

regressions) and report average results in Table XII.

For the junk bond index, for example, the sensitivity to SMB is 0.24 with a t-value of 4.33

compare to 0.23 with a t-value of 6.77 with monthly data. For the S&P and HML the level

of the sensitivity is lower and HML in particular becomes insignificant for all investment grade

18
indices. Interestingly, however, SMB remains highly significant, indeed for the investment grade

indices the t-statistics are higher on SMB than on the S&P. Overall, therefore, and despite the

undoubted illiquidity of the market, there is no evidence that the results are strongly sensitive

to the illiquidity of corporate bonds.9

Finally, in Table XIII we present the results of regressions similar to those in Tables IX and X

but for the sub-periods 1975-85, 1985-90 and 1990-95. The question we wish to address is whether

the sensitivities to SMB and HML that are significant in regressions for the entire sample period

and for the period since the mid-1990’s are also significant for earlier years.

For junk bonds the only data available for these earlier years is the broad index HY All

and this is available only from 1990. Nonetheless, we see that the sensitivity to SMB is highly

significant in both the 1985-90 and 1990-95 sub-periods. HML is significant in the earlier period

and just insignificant for the later period. For investment grade bonds, however, neither SMB

nor HML is significant in the 1985-90 subperiod and, over 1975-85 HML is significant but SMB

is significant for only one of the four maturity buckets.

In summary, and consistent with earlier results, high yield bonds have highly significant

sensitivities to SMB and HML in all periods. For investment grade bonds the sensitivities are

lower and not always significant. On some occasions SMB appears more significant than HML

and on others the reverse is true. In general the sensitivity to both factors increases with declining

credit quality.
9
Although corporate bonds are infrequently traded (Chacko et. al. (2005)), the prices used are quotations
rather than transactions and, for this reason, it is moot as to whether these are good or poor proxies for the prices
at which transactions would take place.

19
V The Duration Puzzle

V.1 The Literature

A striking feature of the price of corporate debt is that its sensitivity to the riskless yield curve is

significantly smaller than would be predicted by its conventional (“McCaulay”) duration (Fons

(1990), Gauthier and Goodman (2003), Schaefer and Strebulaev (2005)). This same phenomenon

can also be understood in terms of the negative sensitivity of credit spreads to changes in the

riskless rate (Longstaff and Schwartz (1995), Collin-Dufresne et. al. (2001)).

A negative relation between the credit spread and the riskless rate is, as Longstaff and

Schwartz (1995) point out, a natural consequence of the structural model of credit risk since

an increase in the riskless rate decreases the (riskless) present value of the firm’s debt and so

increases the distance-to-default. So far, however, there is no evidence that structural models

are able to account for the size – as distinct from the direction – of the interest rate sensitivity

of the spread (or, equivalently, the difference between the empirical and conventional duration of

corporate debt). For example, using a the Merton model with stochastic interest rates, Schaefer

and Strebulaev (2005) find that the durations predicted by the model are much higher than those

observed empirically.

V.2 The Empirical Duration of Corporate Debt

In this section we do not attempt – still less claim – to solve this puzzle. What we try to do,

however, is document the effect in some detail and outline some issues that future research on

this topic might address.

Consider a regression of the type described by equation (3) in which the hedging instruments

are the firm’s equity and a Treasury bond with T years-to-maturity:

20
rf rf,T
rj,t = αj + β E E
j rt + β j rt + εj,t . (10)

rf,T
where rtE is the excess return on the firm’s equity, β E
j is the sensitivity to equity, rt is the

excess return on a T -year Treasury and β rf


j is the sensitivity to the return on the Treasury .

The results of this regression, with T = 5 years and for the 1370 individual bonds in our full

sample (see Section II), are reported in Table XIV. If the spread were constant and the yield

curve were to experience parallel shifts, the instantaneous value of β rf


j would be the ratio of the

duration of the corporate bond to the duration of the Treasury.

For all 1370 bonds, the average value of the coefficient on the Treasury return is 0.77,10 in

other words, the empirical duration of a corporate bond is, on average, about 77% of the duration

of a 5-year Treasury bond. From table I, the average McCaulay duration of the bonds in the

sample is 5.63 years and, since this is necessarily larger than the duration of a five-year Treasury,

the empirical duration of corporate bonds must indeed be lower than its McCaulay duration.

A problem with the specification of the regression in Equation 10 is that the duration of both

the corporate and Treasury bond attenuate with time and, almost certainly, at different rates.

Consequently the theoretical value of β rf


j in the regression is also not constant over time. A

better (though still imperfect) specification that accommodates changes in durations over time

is:

  
Dj,t
rj,t = αj + β E E ∗
j rt + β j rtrf,T + εj,t . (11)
Dtrf,T

where Dj,t is the McCaulay duration of corporate bond j at time t and Dtrf,T the McCaulay

duration of Treasury bond time t. In this case the coefficient β ∗j measures the empirical duration
10
In the table the coefficients are reported in basis points (i.e., multiplied by 100)

21
of bond j as a percentage of its McCaulay duration.

The results of this regression are given in Table XV. For the entire sample (“All”), we

find that, on average, a corporate bond has a duration that is around two-thirds (0.67) of its

McCaulay duration. However, as we see from the results for individual credit ratings, the results

for investment grade and sub-investment grade bonds are quite different.

Investment grade bonds have durations that are in the region of 70% to 75% of their McCaulay

durations. For BB bonds, however, this figure is much lower (0.43) and for B grade bonds the point

estimate is actually negative (though insignificantly different from zero). By way of comparison,

Fons (1990), using data from 1979 - 1988, obtains estimates of 74% for AAA, 62% for A, 48% for

BBB, 55% for BB and 39% for B.

If a bond has a duration that is a fraction β ∗ of its McCaulay duration, then the sensitivity

of its credit spread to changes in the Treasury rate is (β ∗ − 1). Using index data, Longstaff and

Schwartz (1995) estimate the sensitivity of spreads by sector. For utility bonds they estimate

the sensitivity of spreads to the 30-year Treasury yield as -0.18, implying a value of β ∗ of 0.82;

their corresponding implied estimate of β ∗ for industrial bonds is 0.37 and for railroad bonds,

0.18.11 Collin-Dufresne et. al (2001) obtain estimates that imply values of β ∗ of around 0.85 for

investment grade, 0.7 for BB and 0.14 for B. Thus, both our results and those found previously

suggest that the empirical duration of corporate debt is lower than its McCaulay duration.12

Table XVI repeats the analysis in Table XV but for maturity sub-samples. For the three

maturity bands that are less than 12 years (1-5 years, 5-8 years and 8-12 years), the values of β ∗

are higher than for the entire sample. For example, for A rated bonds, for which the estimate
11
The data used by Longstaff and Schwartz are Moody’s indices.
12
The estimates of β ∗ obtained by these earlier studies vary widely. There are many potential reasons for this
but two possibilities are: (i) the use of quite different data sets and sample periods, and (ii) the use of different
proxies for the Treasury rate (the 5-year rate in Table XV, the 10-year rate in Collin-Dufresne et. al. (2001) and
the 30-year rate in Longstaff and Schwartz (1995))

22
of β ∗ for the entire sample is 0.71, we obtain estimates of 0.76 (0-5 years), 0.85 (5-8 years) and

0.79 (8-12 years) but only 0.55 for the 12-15 year bucket. The results are much the same for the

other investment grade credit ratings: bonds up to 12 years have durations that are in the region

of 80%-85% of their McCaulay durations while the results for longer maturities are around 55%

to 65%.13

For non-investment grade bonds the sensitivities are uniformly lower (although the sample

sizes for B-grade bonds are small). For BB we obtain 0.57 for 0-5 years and 5-8 years, 0.34 for

8-12 years and 0.23 for 12-25 years.

V.3 The Cross-Section of Empirical Durations

For the entire sample of 1370 bonds,14 Figure 1 shows the values of the interest rate sensitivity,

β rf
j , estimated in equation (10) against McCaulay duration (“Average duration”) .

The figure is quite informative. A large number of those bonds with McCaulay durations of

around eight years or less lie on a “boundary” that is approximately the 45 degree line. In other

words, for these bonds, empirical and McCaulay duration are approximately equal. Next, bonds

that have a McCaulay duration above around eight years – long term corporate bonds – have

interest rate sensitivities that are substantially below their McCaulay duration. The behavior

of these bonds appears markedly different from the shorter bonds. Finally, for the shorter –

less than eight year duration – bonds, there is a certain amount of scatter about the 45 degree

boundary with a few bonds having an anomalously high interest rate sensitivity while, at the

same time, a much larger number have lower interest rate sensitivity.

Figure 2 shows a subset of the same data: those bonds with ratings of AAA, AA and A.
13
The corresponding result for the AAA category is 0.10 but this outlier should be treated with caution since
the sample size i this case is just one.
14
less 11 outliers

23
The figure is similar to Figure 1 but there are some clear differences. First, bonds with Mc-

Caulay durations lower than about eight years, a greater proportion of the empirical durations

lie roughly on the 45 degree boundary; in other words there are fewer outliers. Similarly, there

are proportionally fewer bonds with long McCaulay durations and very low sensitivities.

Figure 3 shows bonds with a rating of BBB and below, i.e., those that appear in Figure 1

but not in Figure 2. Here, a much smaller proportion lie on the boundary. This sub-sample also

includes the small number of bonds, visible in Figure 1, that have short “McCaulay” durations

and anomalously high interest rate sensitivity. Altogether, a large proportion of these bonds have

lower empirical durations than McCaulay durations.

In summary, these figures show that the low relative duration that bonds have on average is

not a feature that is uniform across the market. Many bonds, particularly higher quality bonds

with durations of under around eight years McCaulay duration, have empirical durations that

are quite close to their McCaulay durations. Others, particularly longer term and lower quality

bonds have empirical durations that are much lower than McCaulay duration.

Finally Table XVII shows the results of a cross-sectional regression of the determinants of

 ∗ (equation XV). The table reports the results of three regressions. In all three there is
β j

a strongly negative relation between relative empirical duration and duration itself. In other

words, as we have seen, empirical duration is, on average, a smaller fraction of McCaulay the

longer the McCaulay duration. Second, there is a strong negative relation between relative

empirical duration and quasi-market leverage, firm volatility and the coupon rate. All three

variables are proxies for credit exposure and the negative sign on each of them implies that

increasing credit risk lowers relative empirical duration. Finally, regressions (2) and (3), that

include an investment grade dummy, show that, all else equal and investment grade bond has a

relative empirical duration that is higher by about 15% of its McCaulay duration.

24
VI Conclusions

The main result in this paper is encouraging for those charged with managing the risk of capital

structure arbitrage positions. Previous work on individual bonds, in particular, Collin-Dufresne

et. al. (2001) has suggested that using the issuing firm’s equity to hedge corporate debt was

unlikely to be effective. At the portfolio level, however, the picture is different since much of the

residual risk at the individual bond level turns out to be diversifiable. The differences between

the portfolio results described here and those for individual bonds are particularly striking for

lower credit ratings. For the BB bonds, for example, the average R2 at the individual bond level

in a regression of returns on Treasury returns and firm equity, is 33%. At the portfolio level the

corresponding figure is 73%. (Both figures are for an in-sample analysis).

Second, we confirm the findings of Elton et al. (2001) and Collin-Dufresne et. al. (2001)

that corporate debt returns are significantly related to the Fama-French factors, SMB and HML.

In particular we find that these results hold over long periods and over sub-periods. High yield

bonds are highly sensitive to the Fama-French factors in all periods; investment grade bonds

are less sensitive and, in some periods, their sensitivities are statistically insignificant. From a

hedging perspective, however, the importance of the Fama-French factors is less clear. Within

sample, hedging against the Fama-French factors naturally reduces portfolio risk but the same is

not true out-of-sample and it appears that the hedge ratios of corporate bonds against the Fama-

French factors may not be sufficiently predictable for hedging against these factors to make a

useful impact on portfolio risk.

Third, and finally, we find that the empirical duration of corporate debt is on average markedly

lower than its McCaulay duration but that the degree of difference varies significantly by credit

rating and maturity. For low grade debt the difference is substantial. The precise reason for this

25
difference is a topic for further research.

26
References
Agarwal and Naik, 2000., “Performance Evaluation of Hedge Funds with Option-Based and
Buy-and-Hold Strategies”, Working Paper, London Business School - Institute of Finance and
Accounting
Chacko, George, Peter Hecht and Marti Subrahmanyam, 2005, “The Determinants of
Liquidity in the Corporate Bond Market: An Application of Latent Liquidity”, Working Paper,
Harvard University.
Collin-Dufresne, Pierre, Robert S. Goldstein, and J. Spencer Martin, 2001, “The
Determinants of Credit Spread Changes”, Journal of Finance, 56, 2177–2207.
Currie and Morris, 2002, “And now for Capital Structure Arbitrage”, Euromoney, December,
38-43.
Duarte, Jefferson, Francis Longstaff and Fan Yu, 2004, “Risk and Return in Fixed
Income Arbitrage: Nickels in Front of a Steamroller?”, Review of Financial Studies, forthcoming.
Elton, Edwin J., Martin J. Gruber, Deepak Agrawal, and Christopher Mann, 2001,
“Explaining the Rate Spread on Corporate Bonds”, Journal of Finance, 56, 247–277.
Fisher, Lawrence, 1959, “Determinants of Risk Premiums on Corporate Bonds”, Journal of
Political Economy, 67: 217-237.
Fons, Jerome S., 1990, “Default Risks and Duration Analysis”, in The High Yield Debt Market,
ed.: E.I.Altman, Irwin Professional Publications.
Gauthier, Laurent and Laurie Goodman, 2003,“Risk-Return Trade-Offs on Fixed Income
Asset Classes”, in Frank J. Fabozzi, Professional Perspectives on Fixed Income Portfolio Man-
agement, Volume 4, John Wiley & Sons, 1st edition, .
Leland, Hayne, 1994, ”Corporate Debt Value, Bond Covenants, and Optimal Capital Struc-
ture,” Journal of Finance 49, 1213-1252.
Leland, Hayne and Klaus Toft, 1996, ”Optimal Capital Structure, Endogenous Bankruptcy,
and the Term Structure of Credit Spreads”, Journal of Finance, 51, 987-1019.
Longstaff, Francis A., and Eduardo S. Schwartz, 1995, “A Simple Approach to Valuing
Risky Fixed and Floating Rate Debt”, Journal of Finance, 50, 789–819.
Merton, Robert C., 1974, “On the Pricing of Corporate Debt: The Risk Structure of Interest
Rates”, Journal of Finance, 29, 449–470.
Schaefer, Stephen M. and Ilya A. Strebulaev, 2005, “Structural Models of Credit Risk
are Useful: Evidence from Hedge Ratios on Corporate Bonds”, working paper.
WARGA, A, 1998, “Fixed Income Database”, University of Houston, Houston, Texas.
Yu, Fan, 2005, “How Profitable Is Capital Structure Arbitrage?” Financial Analysts Journal,
forthcoming.

27
Table I
Summary statistics for Sample of Bonds used in Analysis

This table reports summary statistics for the bonds used in the analysis. The period covered is
12.1996–12.2003. The sample selection criteria are: (1) we can match the bond return data with
CRSP and COMPUSTAT (this allows us to use about 62% of the total number of observations);
(2) the bond is issued by a U.S. company and denominated in $U.S.;15 (3) it is possible to match
unambiguously the bond issuer with a company in CRSP using the CUSIP; (4) the bond is issued
by a non-financial corporation; (5) the bond is straight (i.e. does not have any option-like embedded
features); (6) the bond has at least 25 consecutive monthly price observations; (7) the bond has
an initial maturity of at least four years. Where relevant, statistics are first calculated for each
bond and then for averaged across bonds. Each bond is classified by rating on the date of its first
occurrence in the data set. T − t is the time to maturity remaining on the date of each observation
and is given in calendar years. The annual coupon rate is in percent. The nominal value (of the
amount outstanding) is in million $US dollars.

All AAA AA A BBB BB B CCC

No. Bonds 1370 23 127 603 539 50 26 2


No. Issuers 396 7 34 158 185 24 10 2
Mean T − t 9.50 10.16 9.45 10.13 9.14 7.11 7.39 2.48
Median T − t 5.71 4.58 5.25 5.71 5.81 5.61 6.90 2.48
Mean Duration 5.63 6.11 5.55 5.81 5.55 4.87 4.95 2.38
Median Duration 4.73 3.93 4.52 4.78 4.76 4.61 4.76 2.38
Mean Coupon 7.44 6.47 6.83 7.21 7.72 8.25 9.51 4.81
Median Coupon 7.20 6.63 6.75 7.00 7.45 8.00 9.50 4.81
Mean Nominal Value 260.26 324.90 315.54 271.46 254.22 252.19 219.74 229.14
Median Nominal Value 200 300 250 200 200 200 175 200

28
Table II
Hedging Effectiveness: In-Sample Estimates (Eligibility rule 5A)

Each of the first four rows of the table shows H K,A , the variance of the hedged portfolio as a
percentage of the variance of the unhedged portfolio for a particular set of hedging instruments
(shown in the first column) when the eligibility rule 5A is used. Here “Yield Curve (YC)” means the
one- and ten-year Treasury bonds; “Yield Curve + firm equity” means the yield curve instruments
and, for each bond, the equity of the issuing firm; “YC + equity + other” means the yield curve
instruments, the firm’s equity and, in addition, the S&P, the Fama-French factors (SMB & HML)
and the momentum mimicking portfolio; “Yield Curve + S&P” means the yield curve instruments
and the S&P. T indicates the number of months of data used in the calculations and min (max) the
minimum (maximum) number of bonds in the portfolio in any one month.

All AAA AA A BBB BB B CCC


Yield Curve only 49.6 18.7 23.3 33.4 47.2 92.9 95.0 95.1
Yield Curve + firm equity 27.5 17.9 19.4 26.0 27.5 26.9 48.1 94.1
YC + Equity + other 18.2 12.9 16.0 17.8 19.7 20.5 42.1 83.3
Yield Curve + S&P 39.6 15.2 19.6 27.6 43.8 74.9 88.6 94.3
T 84 84 84 84 84 84 84 84
N min 200 4 30 77 70 14 0 0
N max 726 14 67 289 278 67 10 1

Table III
Hedging Effectiveness: In-Sample Estimates (Eligibility rule 5B)

Each of the first four rows of the table shows H K,B , the variance of the hedged portfolio as a
percentage of the variance of the unhedged portfolio for a particular set of hedging instruments
(shown in the first column) when the eligibility rule 5B is used. Here “Yield Curve (YC)” means the
one- and ten-year Treasury bonds; “Yield Curve + firm equity” means the yield curve instruments
and, for each bond, the equity of the issuing firm; “YC + equity + other” means the yield curve
instruments, the firm’s equity and, in addition, the S&P, the Fama-French factors (SMB & HML)
and the momentum mimicking portfolio; “Yield Curve + S&P” means the yield curve instruments
and the S&P. T indicates the number of months of data used in the calculations and min (max) the
minimum (maximum) number of bonds in the portfolio in any one month.

All AAA AA A BBB BB B CCC


Yield Curve only 51.1 19.6 20.4 39.4 64.4 86.0 96.4
Yield Curve + firm equity 31.3 18.9 17.5 31.4 33.1 47.2 50.6
YC + Equity + other 19.9 13.7 14.4 21.0 20.6 39.0 44.1
Yield Curve + S&P 43.6 17.1 18.5 35.4 56.2 77.1 87.1
T 66.5 66.5 66.5 66.5 66.5 66.5 66.5 66.5
N min 267.7 4.1 32.6 111.1 102.8 12.3 0.5 0.0
N max 574.4 11.9 53.1 226.6 233.7 41.5 7.6 0.0

29
Table IV
Hedging Effectiveness: Out-of-sample Estimates

Each of the first four rows of the table shows the variance of the hedged portfolio as a percentage
of the variance of the unhedged portfolio. The first column shows the hedging instruments that are
used. Here “Yield Curve (YC)” means the one and ten-year Treasury bonds are used as hedging
instruments; “Yield Curve + firm equity” means the yield curve instruments and, for each bond,
the equity of the issuing firm; “YC + equity + “other”” means the instruments in the previous case
and, in addition, the S&P, the Fama-French factors (SMB & HML) and the momentum mimicking
portfolio. T indicates the number of months of data used in the calculations and min (max) the
minimum (maximum) number of bonds in the portfolio in any one month. [Add - out of sample]

All AAA AA A BBB BB B CCC


Yield Curve only 90.0 33.8 41.4 73.2 79.7 106.0 104.3 92.9
Yield Curve + firm equity 60.4 34.3 38.8 62.6 53.5 47.6 69.8 3.7
YC + Equity + other 58.9 37.3 45.2 72.8 55.9 41.8 73.2 41.9
Yield Curve + S&P 76.6 30.0 37.7 64.3 78.3 92.2 104.0 72.4
T 48 48 48 48 48 48 48 48
N min 200 4 30 77 70 14 0 0
N max 424 14 53 159 154 41 7 1

Table V
Descriptive Statistics for Equity Related Indices: Jan 1976 - December 2003

The table gives summary statistics for the equity related indices used in the analysis. The
variables included are the S&P 500 (SP500), the consumer price index (CPI), the Fama-French
SMB, HML and momentum (UMD) factors, the CRSP one-month Treasury bill rate (RF) and
the Fama-French one-month Treasury bill rate (RF/FF). The statistics on rates of return are
monthly.

SP500 CPI SMB HML UMD RF RF/FF


Mean monthly return 0.85 0.36 0.28 0.44 0.88 0.50 0.50
Total return 13.61 3.48 2.19 3.95 15.58 5.70 5.72
St. dev. 4.36 0.33 3.25 3.14 4.26 0.26 0.25
Min -21.76 -0.46 -16.69 -12.03 -25 0.06 0.06
Max 13.18 1.52 21.49 13.75 18.38 1.52 1.35
Index start 7512 7512 7512 7512 7512 7512 7512
N. Obs. 348 348 349 348 348 349 349

30
Table VI
Descriptive Statistics for CRSP Treasury Indices: Jan 1976 - December 2003

The table gives summary statistics for CRSP constant maturity (“Fixed Term”) Treasury
indices. The statistics on rates of return are monthly.

30Y 20Y 10Y 7Y 5Y 2Y 1Y 3M


Mean monthly return 0.77 0.82 0.75 0.75 0.71 0.64 0.61 0.55
Total return 12.17 14.66 12.45 12.56 11.18 9.16 8.39 6.75
St. dev. 3.32 3.11 2.43 2.09 1.73 0.99 0.61 0.30
Min -10.49 -9.36 -6.68 -7.04 -5.80 -3.69 -1.72 0.07
Max 13.31 15.23 10 10.75 10.61 8.42 5.61 2.13
Index start 7512 7512 7512 7512 7512 7512 7512 7512
N. Obs. 349 349 349 349 349 349 349 349

31
Table VII
Descriptive Statistics on Merrill Lynch Corporate Bond Indices: All Years

The table gives summary statistics on the Merrill Lynch corporate bond indices described in
Section II. The first column refers to the broad IG index that includes all maturities and the
next four columns to sub-indices chosen on the basis of maturity. The next four columns (AAA
to BBB) are credit rating sub-indices of the broad IG index. The remaining four columns give
statistics on high yield debt with the first column a broad index and the remaining three chosen
on the basis of credit rating. Note that for some indices data on maturity, duration and yield
are available only for recent years and that the indices do not all share a common start date
(see row 7).

IG all IG 1-5Y IG 5-10Y IG 10-15Y IG 15+Y AAA AA A BBB HY all BB B CCC


Mean monthly return 0.79 0.72 0.86 0.82 0.86 0.69 0.70 0.71 0.72 0.76 0.60 0.49 0.54
Total return 14.57 9.80 16.34 14.97 17.32 3.70 3.78 3.82 3.90 5.03 1.76 1.55 1.57
St. dev. 2.07 1.07 2.53 2.38 2.66 1.36 1.35 1.35 1.40 1.96 1.76 2.51 3.85
Min -7.36 -3.17 -6.52 -7.39 -8.52 -4.38 -3.65 -4.39 -4.17 -7.74 -8.15 -7.15 -11.73
Max 11.97 6.60 33.49 13.19 13.94 4.51 4.91 4.62 4.54 8.68 3.95 7.85 11.54

32
N. Bonds 3981 1085 1159 404 1117 175 564 1695 1302 1028 496 639 18
Index start 7512 7803 7701 7604 7512 8901 8901 8901 8901 8609 9701 9701 9701
Maturity 12.46 3.06 7.52 12.28 25.81 12.50 11.87 11.24 12.40 9.17 8.94 7.37 8.12
Duration 6.45 2.70 5.62 7.82 10.24 6.61 6.36 6.21 6.57 5.51 5.56 5.18 6.91
Yield 8.00 7.29 8.04 8.22 8.73 6.80 6.84 7.09 7.76 11.65 8.42 10.75 4.62
N. Obs. 349 322 336 345 349 192 192 192 192 220 96 96 96
Table VIII
Descriptive Statistics on Merrill Lynch Corporate Bond Indices: Jan 1997 - Dec 2004

The table reports statistics for the same Merrill Lynch indices as in Table VII and for the
maximum common period of availability across all the indices (January 1997 - December 2004).

IG all IG 1-5Y IG 5-10Y IG 10-15Y IG 15+Y AAA AA A BBB HY all BB B CCC


Mean monthly return 0.62 0.54 0.64 0.70 0.71 0.61 0.61 0.62 0.61 0.55 0.60 0.49 0.54
Total return 1.79 1.67 1.83 1.92 1.93 1.78 1.78 1.79 1.78 1.65 1.76 1.55 1.57
St. dev. 1.37 0.71 1.49 1.97 2.26 1.36 1.26 1.37 1.48 2.26 1.76 2.51 3.85
Min -4.22 -1.61 -4.73 -6.74 -8.16 -4.38 -3.65 -4.39 -4.17 -7.74 -8.15 -7.15 -11.73
Max 3.70 2.07 4.13 5.40 5.70 3.68 3.60 3.64 3.86 6.33 3.95 7.85 11.54
N. Bonds 3774 1339 1194 200 1042 118 431 1705 1521 1370 496 639 18
Index start 9701 9701 9701 9701 9701 9701 9701 9701 9701 9701 9701 9701 9701
Maturity 11.12 3.06 7.53 12.09 27.29 11.42 9.49 10.76 12.05 7.95 8.94 7.37 -0.10
Duration 6.29 2.77 5.92 8.18 11.55 6.44 5.71 6.18 6.61 5.27 5.56 5.18 6.91
Yield 6.28 5.47 6.42 6.78 7.30 5.68 5.64 6.05 6.80 10.68 8.42 10.75 4.62
N. Obs. 96 96 96 96 96 96 96 96 96 96 96 96 96

33
Table IX
Corporate Bond Index Hedging Regressions: All Available data

The table shows the results of regressing the excess return on corporate bond indices on the
excess return on the 10-year Treasury bond and the S&P index and on the Fama-French SMB
and HML factors. The regressions are calculated using all the available data for each series
and, as a result, are based on different numbers of observations and periods.

IG all HY all IG 1-5Y IG 5-10Y IG 10-15Y IG 15+Y AAA AA A BBB BB B CCC


Constant 0.14 0.38 0.41 0.31 0.06 0.01 0.20 0.22 0.22 0.20 0.19 0.12 0.06
( 2.62) ( 3.31) (11.62) ( 2.58) ( 0.95) ( 0.19) ( 7.43) ( 6.85) ( 6.05) ( 3.92) ( 1.18) ( 0.53) ( 0.16)
RF10Y 0.74 0.16 0.36 0.60 0.86 0.93 0.65 0.64 0.63 0.59 0.22 0.07 -0.12
(35.88) ( 2.96) (27.07) (12.83) (37.54) (33.57) (51.65) (43.22) (37.89) (24.62) ( 3.04) ( 0.65) (-0.74)
SP 0.07 0.25 0.03 0.05 0.09 0.11 0.04 0.05 0.05 0.10 0.24 0.31 0.43
( 5.53) ( 9.46) ( 3.01) ( 1.68) ( 6.17) ( 6.22) ( 5.86) ( 5.69) ( 6.06) ( 7.81) ( 6.57) ( 5.77) ( 5.17)
HML 0.06 0.16 0.04 0.13 0.09 0.07 0.02 0.02 0.03 0.06 0.16 0.12 0.29
( 3.18) ( 4.02) ( 3.31) ( 2.93) ( 4.38) ( 3.00) ( 2.52) ( 1.80) ( 2.33) ( 3.43) ( 3.49) ( 1.72) ( 2.74)
SMB 0.06 0.23 0.04 0.12 0.08 0.09 0.02 0.02 0.04 0.08 0.17 0.20 0.37

34
( 3.68) ( 6.77) ( 3.28) ( 3.18) ( 4.63) ( 4.18) ( 3.13) ( 2.54) ( 3.95) ( 5.24) ( 4.33) ( 3.57) ( 4.18)
R2 0.81 0.37 0.72 0.36 0.83 0.79 0.94 0.91 0.89 0.79 0.34 0.31 0.28
N 347 219 320 334 343 347 191 191 191 191 96 96 96
Table X
Corporate Bond Index Hedging Regressions: January 1997 - December 2004

The table shows the results of regressing the excess return on corporate bond indices on the
excess return on the 10-year Treasury bond and the S&P index and on the Fama-French SMB
and HML factors. The regressions are calculated using the data from January 1997 to December
2004, the maximum period for which data is available for all the series.

IG all HY all IG 1-5Y IG 5-10Y IG 10-15Y IG 15+Y AAA AA A BBB BB B CCC


Constant 0.18 0.15 0.36 0.18 0.06 -0.01 0.20 0.24 0.20 0.14 0.19 0.12 0.06
( 2.74) ( 0.71) ( 8.93) ( 2.60) ( 0.71) (-0.10) ( 4.68) ( 5.03) ( 3.35) ( 1.52) ( 1.18) ( 0.53) ( 0.16)
RF10Y 0.60 0.11 0.30 0.66 0.88 0.90 0.63 0.57 0.61 0.59 0.22 0.07 -0.12
(20.00) ( 1.17) (16.92) (21.98) (23.19) (14.93) (33.56) (26.91) (22.55) (14.58) ( 3.04) ( 0.65) (-0.74)
SP 0.08 0.30 0.01 0.08 0.11 0.19 0.05 0.04 0.06 0.12 0.24 0.31 0.43
( 5.53) ( 6.35) ( 1.13) ( 5.14) ( 5.49) ( 6.06) ( 4.79) ( 3.57) ( 4.47) ( 6.02) ( 6.57) ( 5.77) ( 5.17)
HML 0.05 0.15 0 0.05 0.09 0.11 0.03 0.02 0.03 0.07 0.16 0.12 0.29
( 2.32) ( 2.58) ( 0.22) ( 2.51) ( 3.45) ( 2.67) ( 2.59) ( 1.36) ( 1.72) ( 2.58) ( 3.49) ( 1.72) ( 2.74)
SMB 0.06 0.21 0.01 0.06 0.10 0.14 0.03 0.03 0.05 0.09 0.17 0.20 0.37

35
( 3.74) ( 4.18) ( 1.41) ( 3.44) ( 4.76) ( 4.18) ( 3.37) ( 2.50) ( 3.22) ( 4.09) ( 4.33) ( 3.57) ( 4.18)
R2 0.81 0.33 0.75 0.84 0.85 0.71 0.92 0.88 0.84 0.70 0.34 0.31 0.28
N 96 96 96 96 96 96 96 96 96 96 96 96 96
Table XI
Corporate Bond Index Hedging Regressions: All Available data

The table shows the results of regressing the excess return on corporate bond indices on the
excess return on the 10-year Treasury bond and the S&P index, VIX, the Fama-French SMB
and HML factors, the CPI and the momentum factor (UMD) . The regressions are calculated
using all the available data for each series and, as a result, are based on different numbers of
observations and periods.

IG all HY all IG 1-5Y IG 5-10Y IG 10-15Y IG 15+Y AAA AA A BBB BB B CCC


Constant 0.23 0.51 0.37 0.22 0.04 0.16 0.20 0.22 0.23 0.23 0.25 0.28 0.10
( 3.89) ( 2.66) ( 9.13) ( 3.79) ( 0.57) ( 1.45) ( 5.03) ( 4.67) ( 4.33) ( 3.04) ( 1.23) ( 0.95) ( 0.23)
RF10Y 0.64 0.19 0.32 0.69 0.88 0.91 0.66 0.65 0.65 0.61 0.25 0.13 -0.06
(31.90) ( 2.89) (23.26) (34.07) (34.45) (23.43) (47.47) (39.30) (35.37) (22.91) ( 3.28) ( 1.21) (-0.35)
SP 0.05 0.27 0 0.04 0.06 0.10 0.02 0.03 0.03 0.08 0.21 0.25 0.40
( 3.81) ( 6.02) ( 0.19) ( 3.10) ( 3.59) ( 3.65) ( 2.49) ( 2.52) ( 2.77) ( 4.62) ( 4.11) ( 3.51) ( 3.48)
VIX -0.02 0.03 -0.01 -0.02 -0.02 -0.04 -0.02 -0.03 -0.02 -0.01 -0.02 -0.01 0.04
(-1.70) ( 0.61) (-0.83) (-1.71) (-1.18) (-1.37) (-2.38) (-2.50) (-1.93) (-0.76) (-0.35) (-0.14) ( 0.39)

36
HML 0.03 0.14 0 0.04 0.06 0.06 0.02 0.02 0.02 0.05 0.15 0.09 0.26
( 2.28) ( 3.11) ( 0.23) ( 2.59) ( 3.27) ( 2.10) ( 1.93) ( 1.39) ( 1.73) ( 2.85) ( 3.12) ( 1.28) ( 2.40)
SMB 0.05 0.24 0.01 0.04 0.07 0.10 0.02 0.02 0.04 0.08 0.17 0.21 0.39
( 4.00) ( 6.32) ( 1.11) ( 3.86) ( 4.97) ( 4.67) ( 2.55) ( 1.89) ( 3.58) ( 5.03) ( 4.31) ( 3.78) ( 4.41)
CPI 0.01 -0.28 0.13 -0 0.24 -0.25 0.07 0.04 0.03 -0.06 -0.09 -0.28 0.26
( 0.05) (-0.51) ( 1.15) (-0.00) ( 1.13) (-0.78) ( 0.62) ( 0.26) ( 0.22) (-0.27) (-0.15) (-0.34) ( 0.20)
UMD -0.02 -0.09 -0.01 -0.02 -0.01 -0.05 -0.01 -0 -0.02 -0.03 -0.04 -0.10 -0.13
(-2.30) (-3.19) (-1.45) (-2.51) (-1.35) (-3.09) (-1.26) (-0.43) (-2.22) (-2.58) (-1.62) (-2.58) (-2.19)
R2 0.87 0.39 0.77 0.88 0.89 0.79 0.94 0.91 0.89 0.78 0.34 0.33 0.29
N 177 177 177 177 177 177 177 177 177 177 95 95 95
Table XII
Corporate Bond Index Hedging Regressions: All Available data, Three-monthly observations

The table shows the results of regressing the excess return on corporate bond indices on the
excess return on the 10-year Treasury bond and the S&P index and on the Fama-French SMB
and HML factors. A three-month differencing interval is used for calculating returns. The
regressions are calculated using all the available data for each series and, as a result, are based
on different numbers of observations and periods.

IG all HY all IG 1-5Y IG 5-10Y IG 10-15Y IG 15+Y AAA AA A BBB BB B CCC


Constant 0.34 0.86 1.19 0.64 0.16 -0.04 0.55 0.57 0.57 0.50 0.38 -0.30 -0.70
( 2.09) ( 2.08) (10.16) ( 4.77) ( 0.89) (-0.18) ( 6.04) ( 4.57) ( 4.36) ( 2.65) ( 0.60) (-0.38) (-0.49)
RF10Y 0.84 0.31 0.43 0.72 0.95 1.05 0.72 0.70 0.69 0.66 0.25 0.41 0.24
(27.77) ( 3.06) (20.54) (29.22) (28.05) (27.49) (31.94) (22.59) (21.33) (14.19) ( 1.54) ( 2.04) ( 0.65)
SP 0.04 0.26 -0.01 0.02 0.05 0.07 0.02 0.03 0.04 0.09 0.27 0.33 0.55
( 1.93) ( 5.48) (-0.85) ( 1.17) ( 2.07) ( 2.72) ( 1.55) ( 2.03) ( 2.62) ( 3.98) ( 3.81) ( 3.79) ( 3.44)
HML 0.02 0.12 0.01 0.02 0.03 0.02 0.01 0.02 0.02 0.03 0.17 0.09 0.25
( 0.98) ( 1.99) ( 0.30) ( 1.09) ( 0.90) ( 0.78) ( 0.44) ( 1.13) ( 1.29) ( 1.35) ( 2.41) ( 1.08) ( 1.53)

37
SMB 0.08 0.24 0.03 0.06 0.09 0.11 0.03 0.05 0.06 0.10 0.21 0.23 0.40
( 3.46) ( 4.33) ( 1.95) ( 3.18) ( 3.58) ( 3.98) ( 2.97) ( 2.94) ( 3.72) ( 4.27) ( 3.17) ( 2.85) ( 2.70)
R2 0.88 0.44 0.81 0.90 0.88 0.88 0.95 0.90 0.89 0.80 0.36 0.38 0.31
N 114.00 72.00 105.00 109.00 112.00 114.00 62.00 62.00 62.00 62.00 31.00 31.00 31.00
Table XIII
Corporate Bond Index Hedging Regressions: All Available data

The Table shows sub-period regressions for the regression described in Table ??.

IG all HY all IG 1-5Y IG 5-10Y IG 10-15Y IG 15+Y AAA AA A BBB

Panel (a) 1975-85


RF10Y 0.87 0.41 0.60 0.88 1.05
(18.62) (13.05) ( 4.55) (16.98) (19.05)
SP 0.08 0.08 -0.01 0.15 0.09
( 2.01) ( 2.77) (-0.05) ( 3.34) ( 2.05)
HML 0.12 0.14 0.30 0.19 0.13
( 2.42) ( 3.73) ( 1.91) ( 3.04) ( 2.21)
SMB 0.03 0.05 0.39 0.07 0.03
( 0.64) ( 1.39) ( 2.47) ( 1.23) ( 0.46)
R2 0.81 0.75 0.22 0.80 0.82
N 108.00 81.00 95.00 104.00 108.00

38
Panel (b) 1985-90
RF10Y 0.66 0.25 0.28 0.56 0.74 0.82 0.68 0.70 0.73 0.69
(17.54) ( 3.31) (13.53) (15.42) (15.27) (14.57) (11.30) (12.28) ( 9.10) (14.26)
SP 0.04 0.17 0.00 0.02 0.06 0.06 0.02 0.04 0.04 0.03
( 1.78) ( 5.19) ( 0.27) ( 0.76) ( 2.09) ( 1.75) ( 0.65) ( 1.12) ( 0.82) ( 1.25)
HML 0.05 0.27 0.02 0.03 0.09 0.06 0.08 0.04 0.06 0.04
( 0.93) ( 3.18) ( 0.56) ( 0.62) ( 1.44) ( 0.87) ( 0.92) ( 0.51) ( 0.52) ( 0.60)
SMB 0.05 0.28 0.01 0.02 0.07 0.08 0.03 0.09 0.11 0.12
( 1.11) ( 4.05) ( 0.54) ( 0.44) ( 1.25) ( 1.24) ( 0.47) ( 1.25) ( 1.11) ( 2.05)
R2 0.86 0.53 0.78 0.83 0.83 0.81 0.94 0.95 0.90 0.96
N 60.00 40.00 60.00 60.00 60.00 60.00 12.00 12.00 12.00 12.00

Panel (c) 1990-95


RF10Y 0.69 0.28 0.35 0.72 0.88 0.83 0.69 0.75 0.68 0.62
(24.34) ( 1.96) (10.70) (20.95) (24.94) (18.96) (26.13) (27.50) (22.56) (15.89)
SP 0.02 0.22 -0.02 0.01 0.01 0.07 0.01 0.01 0.02 0.04
( 1.28) ( 3.01) (-1.21) ( 0.63) ( 0.47) ( 2.83) ( 0.59) ( 0.62) ( 1.16) ( 2.08)
HML 0.02 0.16 0.02 0.04 0.01 0.03 -0.01 -0.00 0.02 0.06
( 1.43) ( 1.84) ( 0.83) ( 1.72) ( 0.27) ( 1.01) (-0.59) (-0.01) ( 1.32) ( 2.47)
SMB 0.04 0.44 0.00 0.05 0.05 0.07 0.02 0.02 0.04 0.08
( 2.36) ( 5.28) ( 0.08) ( 2.35) ( 2.20) ( 2.60) ( 0.98) ( 0.98) ( 2.24) ( 3.49)
R2 0.95 0.49 0.75 0.93 0.95 0.93 0.95 0.96 0.94 0.90
N 59.00 59.00 59.00 59.00 59.00 59.00 59.00 59.00 59.00 59.00
Table XIV
Empirical Duration of Corporate Debt

The table reports the results of the following regression


rf rf,T
rj,t = αj + β E E
j rt + β j rt + εj,t .
The coefficients are average values across the bonds in the sample (and multiplied by 100). The
data used are the 1370 bonds of our full sample for the entire sample period from December 1996
to December 2003. N gives the average number of time-series observations in each regression
and the number in parentheses is the number of bonds in the sample for each credit category.

All AAA AA A BBB BB B


Intercept -0.15 -0.19 -0.19 -0.18 -0.18 0.02 0.62
(-4.42) (-4.11) (-6.36) (-6.17) (-4.10) ( 0.24) ( 2.41)
rtrf,5 77.71 88.84 85.63 84.13 80.32 40.87 -11.83
(30.93) (28.35) (40.72) (38.39) (24.58) ( 7.40) (-0.65)
rtE 4.26 0.75 1.60 3.48 4.31 9.72 15.85
(15.75) ( 1.30) ( 5.02) (12.46) (13.54) (19.99) (15.95)
R̄2 0.49 0.64 0.61 0.52 0.45 0.32 0.35
N 47.02 57.30 53.28 45.23 47.60 46.96 41.27
(1370) (23) (126) (620) (466) (107) (26)

39
Table XV
Empirical Duration of Corporate Debt as Fraction of McCaulay Duration

The table reports the results of the following regression


  
E E ∗ Dj,t rf,T
rj,t = αj + β j rt + β j rt + εj,t .
Dtrf,T
The coefficients are average values across the bonds in the sample (and multiplied by 100). The
data used are the 1370 bonds of our full sample for the entire sample period from December 1996
to December 2003. N gives the average number of time-series observations in each regression
and the number in parentheses is the number of bonds in the sample for each credit category.

All AAA AA A BBB BB B


Intercept -0.16 -0.20 -0.21 -0.20 -0.18 -0.01 0.61
(-4.78) (-4.60) (-7.19) (-6.77) (-4.12) (-0.09) ( 2.36)
β ∗j 66.91 74.41 76.01 70.71 69.51 43.27 -17.38
(31.70) (34.03) (48.32) (41.14) (24.30) ( 9.13) (-0.98)
βE
j 4.14 0.46 1.46 3.40 4.16 9.62 15.80
(15.63) ( 0.86) ( 4.79) (12.43) (13.27) (20.02) (15.76)
R̄2 0.52 0.70 0.66 0.55 0.48 0.34 0.35
N 47.02 57.30 53.28 45.23 47.60 46.96 41.27
(1370.00) (23.00) (126.00) (620.00) (466.00) (107.00) (26.00)

40
Table XVI
Empirical Duration of Corporate Debt as Fraction of McCaulay Duration: by
Maturity. Maturities are on the first date of bond entering the data set. For
details see Table XV and text

.
PANEL A: 0–5 YEARS
All AAA AA A BBB BB B CCC
Intercept 0.02 -0.01 -0.02 0.01 0.00 0.06 0.87 0.44
( 1.11) (-0.21) (-1.65) ( 0.80) ( 0.03) ( 1.05) ( 1.48) ( 1.26)
β ∗j 73.86 82.54 80.04 75.89 75.90 56.97 -12.59 -15.12
(29.34) (26.29) (42.10) (35.76) (17.44) ( 7.00) (-0.22) (-0.32)
βE
j 1.20 0.29 -0.08 0.87 1.84 0.44 19.12 -0.83
( 8.65) ( 0.75) (-0.43) ( 6.38) (10.00) ( 1.06) ( 6.78) (-0.34)
R̄2 0.61 0.84 0.78 0.63 0.58 0.38 0.37 -0.03
N 31.47 29.43 31.44 31.26 32.06 30.73 34.25 32.00
(372.00) ( 7.00) (36.00) (197.00) (104.00) (22.00) ( 4.00) ( 2.00)
PANEL B: 5–8 YEARS
All AAA AA A BBB BB B CCC
Intercept -0.12 -0.15 -0.18 -0.18 -0.19 0.03 1.42
(-3.98) (-4.62) (-8.74) (-7.36) (-4.97) ( 0.32) ( 3.22)
β ∗j 77.36 92.18 89.07 85.18 83.60 57.38 -105.15
(28.93) (26.00) (47.87) (38.25) (24.31) ( 8.01) (-3.23)
βE
j 3.39 -0.06 0.58 2.21 3.25 7.51 15.78
(15.93) (-0.14) ( 2.48) (10.06) (12.60) (14.17) (10.85)
R̄2 0.59 0.77 0.75 0.64 0.57 0.38 0.39
N 47.58 62.75 55.24 48.84 45.64 45.65 35.86
(300.00) ( 4.00) (25.00) (107.00) (123.00) (34.00) ( 7.00) ( 0.00)
PANEL C: 8–12 YEARS
All AAA AA A BBB BB B CCC
Intercept -0.21 -0.25 -0.21 -0.28 -0.24 0.05 0.31
(-5.54) (-4.22) (-4.50) (-8.40) (-4.62) ( 0.39) ( 1.06)
β ∗j 72.40 83.73 80.86 79.33 73.48 34.00 23.85
(32.13) (23.43) (32.80) (43.25) (22.63) ( 4.87) ( 1.40)
E ret 4.23 0.70 1.94 3.54 4.46 6.27 14.67
(14.34) ( 0.99) ( 4.69) (11.20) (11.65) ( 8.15) (14.38)
R̄2 0.52 0.69 0.61 0.58 0.47 0.29 0.35
N 52.51 68.80 62.61 50.25 53.36 48.63 42.08
(343.00) ( 5.00) (36.00) (146.00) (120.00) (24.00) (12.00) ( 0.00)
PANEL D: 12–25 YEARS
All AAA AA A BBB BB B CCC
Intercept -0.33 10.43 -0.43 -0.39 -0.26 -0.24 -0.41
(-5.36) ( 3.33) (-5.83) (-6.38) (-3.00) (-1.16) (-0.73)
β ∗j 52.15 10.43 66.30 55.38 56.03 22.87 17.27
(24.72) ( 3.33) (26.23) (28.04) (17.90) ( 3.24) ( 0.85)
E ret 7.79 10.43 2.67 5.08 6.20 26.70 14.95
(15.40) ( 3.33) ( 3.24) ( 8.63) ( 9.29) (22.22) ( 4.92)
R̄2 0.40 10.43 0.54 0.42 0.35 0.36 0.16
N 58.06 49.00 71.00 52.81 61.28 63.26 60.00
(167.00) ( 1.00) (12.00) (80.00) (53.00) (19.00) ( 2.00) ( 0.00)

41
Table XVII
Cross-Sectional Analysis of Estimated Ratio of Empirical Duration as Fraction of
McCaulay Duration. The Table shows the results of a cross sectional regression
with β ∗j , estimated in equation XV as dependent variable.

(1) (2) (3)


Constant 166.00 114.30 114.99
(23.31) (22.92) (23.13)
Duration -2.47 -2.77 -3.23
(-3.89) (-8.96) (-26.59)
Time to maturity -0.19 -0.19
(-0.77) (-1.61)
Quasi-Market Leverage -70.11 -39.75 -39.74
(-12.41) (-13.15) (-13.14)
Firm Volatility -142.39 -75.00 -75.01
(-8.92) (-8.88) (-8.88)
Coupon rate -3.67 -1.61 -1.62
(-4.70) (-3.93) (-3.94)
Inv.Grade Dummy 15.36 15.29
( 8.20) ( 8.16)
R2 0.21 0.48 0.48
N 1359.00 1331.00 1331.00

42
Average Duration vs BetaR for Straight bonds

200

150

100

BetaR
50

−50

0 5 10 15 20
Average Duration

Figure 1. Relationship between empirical duration and McCaulay duration, esti-


mated in equation 10, for straight bonds. Sample size: 1359. McCaulay duration is
modified duration averaged over the bond observations. Outliers are eliminated (maturities near
100 years and extreme β r observations.

43
Average duration vs BetaR for Straight IG bonds

200

150

100

BetaR
50

−50

0 5 10 15 20
Average Duration

Figure 2. Relationship between empirical duration and McCaulay duration, esti-


mated in equation 10, for straight bonds with rating: AAA,AA and A. Sample size:
760 observations. Bond duration is modified duration. Outliers are eliminated (maturities near
100 years and extreme β r observations.

44
Average bond duration vs BetaR forStraight BBB and below

200

150

100
BetaR

50

−50

0 5 10 15 20
Average duration

Figure 3. Relationship between empirical duration and McCaulay duration, esti-


mated in equation 10, for straight bonds with rating: BBB and below. Sample size:
569 observations. Bond duration is modified duration. Outliers are eliminated (maturities near
100 years and extreme β r observations.

45

You might also like