How Do Corporate Bond Holders Interpret Idiosyncratic Risk?

How Do Corporate Bond Holders Interpret Idiosyncratic Risk?

Victor Fanga Yee Choon Limb a,b Department of Accounting and Finance Monash University, Australia

Chien-Ting Lin International Graduate School of Management Division of Business and Enterprise University of South Australia, Australia

Chia-Cheng Ho Department of Finance College of Business Administration National Chung Cheng University Taiwan

Abstract

The objective of this paper is to address issues in the area of default pricing. Past research document inadequacies in estimating the premium received by debt holders. One of the foremost reasons is that the idiosyncratic risks borne by debt holders are not well approximated, and this stems from a lack of understanding of default probabilities. Acknowledging the problem, this study adopts an exploratory approach in developing an alternative approach that seeks to explain the price behavior of corporate bonds. This approach simply improvises two areas of finance: event studies and options pricing. An amalgamation of both areas introduces a stochastic element in the prediction of corporate bond prices; and by means of empirical tests, the resulting estimates of this approach appear to be suitable for default pricing. Another contribution resulting from the exposits of our approach is to deduce that debt holders do not respond homogenously to default risk. Instead, they respond in two manners that are bounded by the undertakings of the firm. In the first manner, debt holders exhibit a symmetric response to information about the undertakings of the firm, such that they resemble equity holders. The second manner portrays asymmetric responses. Evidence to these contrasting observations is independently documented in past literature, but somehow it has never been synthesized. Thus, this study attempts to point out the differences and demonstrate how it can impinge default pricing.

i

1.

Introduction

Past studies demonstrate that yield spreads are peculiarly difficult to predict. Eom et al. (2004) provides a comprehensive comparison of structural and reduced-form models. Under specified conditions, each proposed reduced-form model provides more accurate estimates than Merton (1974). More importantly, it is documented by most studies that firm volatility has little impact on credit spreads. Thus far, Campbell and Taksler (2003) is the latest and most significant study1 to find evidence in support of Merton (1974). It is interesting to note that firm volatility is frequently estimated by simply taking the standard deviation of realized or expected returns. This does not allow the firm value process to be independent of systematic factors, such that the idiosyncrasies of a firm’s assets cannot be adequately captured. Because of this, we suspect that bond yield spread movements resulting from a firm’s undertakings cannot be properly estimated. While standard deviation or absolute values are proxies for expected risk, it is however, not able to take into account the information content of a firm’s undertakings. As such, it is highly possible that a vectorized variable is more likely to provide better explanatory powers. The empirical findings of event studies in bond prices support our argument2. Therefore, we propose an alternative approach for the firm value process. This approach combines the market model and the contingent claims analysis3. As the market model possesses Markov properties, thus, it is highly consistent with the firm value process specified in the contingent claims analysis. Most studies which are variations of Merton (1974) such as Campbell and Taksler (2003), and Jones et al. (1984) disregard information contents of assets. Our objective in this case, is to demonstrate that the information content of a firm’s undertakings is crucial in determining responses of bond holders. To verify the responses of bond holders to information content, the volatility of assets is also investigated. Furthermore, several measures of robustness are instituted along with our

1

Refer to Eom et al. (2004) for an extensive review of the empirical findings of recently proposed models. Some prominent studies include Katz (1974), Grier and Katz (1976), Hettenhouse and Sartoris (1976), Weinstein (1977) Handjinicolaou and Kalay (1984), Hand et al. (1992) and Dhillon and Johnson (1994). 3 A commonly tested option pricing model initiated by Merton (1974). 2

2

approach: (i) firms are analyzed according to several financial characteristics and bond endowments, (ii) our approach is compared with the mean adjusted return model employed by Campbell and Taksler (2003) and (iii) different estimation parameters are employed to minimize inferential biasness. The findings are as follows: (i) Bond holders of financial institutions respond to information content while bond holders of non-financial institutions respond to asset volatility. (ii) Higher rated (A category) financial institutions tend to signal more information content to bond holders as compared to lower rated financial institutions (B category). However, the caveat here is that higher rated firms in the sample are financial institutions. Further analysis suggests that higher rated non-financial institution bond prices respond only to asset volatility, while lower rated non-financial institution bond prices respond to both the information content and asset volatility, albeit in varying degrees. (iii) Bond prices with low coupons are more responsive to information content than those with high coupons. (iv) Returns of bond with longer maturity are more sensitive to information content (v) Information content has limited impact on firms with consistently decreasing leverage ratios, while there are no signs of asset volatility. While bond prices of firms that have increasing leverage ratios are responsive to asset volatility. (vi) Decreasing market-tobook ratios induces bond price responses to asset volatility, while bond prices with increasing ratios are associated with information content. (vii) Lagged responses are frequently recorded for information content, while asset volatility tends to have a more contemporaneous impact on bond returns. Several benefits can be derived from research of this nature. First, a better estimation of price risk is possible through the understanding of idiosyncratic risk. Although bonds are known to be more sensitive to systematic risk, there is only limited knowledge about the relation between firm specific information and bond prices. Even then, evidence about their relation is incoherent. Secondly, understanding risks that are unsystematic to our portfolios is ipso facto a route to improved diversification. As a matter of fact, default risk and price risk are frequently neglected in bond portfolios. This is because of the fixed returns of bonds are assumed to be relatively less risky than other investments. Unfortunately, this is only true under different circumstances. For example, to circumscribe from price risk, an 3

investor will have to hold the bond till maturity and receive principal payment in full. In this case, the investor will not be able to rebalance his portfolio throughout the life of the bond. However, by holding a bond for longer periods, chances of default of the firm issuing this bond tend to be higher. This is commonly taken for granted as bond holders rely on credit ratings as a proxy for default. Such negligence has occasionally resulted in unprecedented losses; and evidence can be drawn from the defaults of investment grade bonds issued by Enron and Worldcom. Therefore, it will be paradoxical to assume that both risks are immaterial since they coexist and are in fact a manifestation of each other. The proceeding discussion comprises of 4 sections, starting with a thorough and meticulous definition of the underlying theory in section 2. Section 3 details the research design and data. Sections 4 and 5 present the empirical findings and conclusion respectively.

2.

The Model

This study draws on the basic idea that event studies provide a unique perspective to default pricing. According to Merton (1974) debt holders should be sensitive to the undertakings of a firm. For this matter, prices and yields of bonds issued by the firm is expected to vary according to any idiosyncratic information. Since these variations are also commonly measured in event studies, it is argued that these measurements can be extrapolated to Merton (1974)’s model.

First, let us consider a simple expectations model

R j ,t = Et ( Rt Φ t −1 ) + ε j ,t

(1)

where the return R on a security j at time t is linearly related to the expected returns at time t, which is conditional on information set Φ available to investors at t-1. Thus the residuals of the model ε j,t can be referred as the unexpected returns of security j. According to the efficient market hypothesis, this term E (ε j ,t Φt ) is zero. If this is not true, then returns on the security must also be conditional upon information Φt, which is essentially, information 4

about the firm that is only available to investors at time t. Thus, for the model to be operational, it is rewritten as R j , t = α j + β j R m ,t + ε j , t

(2)

where Rm ,t is the return on the market, α j is the mean of historical returns and β j is the sensitivity of the firm to the market index. This regression model is also widely known as the market model. Its applications are mainly based on seminal work by Ball and Brown (1968) and Fama, Fisher, Jansen and Roll (1969). For the market model to be applicable, it must follow statistical assumptions of a normal regression model, where R j ,t and R m ,t are normally distributed and ε j ,t is also independently and normally distributed with mean of zero and variance σ2. Finally, the estimates of ε j ,t are obtained by rewriting the market model as

εˆ j ,t = R j ,t − αˆ j − βˆ j Rm,t

(3)

where R j ,t and R m ,t are observed returns on security j and the market m at time t respectively, and αˆ j and βˆ j are estimates based on past information about the security j. With these estimates, the equation posits that if E (e j ,t Φ t ) ≠ 0 , then unanticipated information must be received and acted upon by investors at time t to make an excess return ( εˆ j ,t ).

2.1

Extrapolating the market model to Merton model

One of the assumptions in contingent claim analysis is that the value of the firm, V, satisfied the stochastic differential equation4: dV = (αV − C )dt + σVdz

(4)

Basing on the above and other assumptions, Merton (1974) derives a yield spreads model:

[ (

)]

[ (

)]

1  1  R ( τ ) − r = − ln Φ h2 d , σ V2 τ + Φ h1 d , σ V2 τ  5 τ  d  4

(5)

Refer to Merton (1974) for other assumptions.

5

where R (τ ) is the yield to maturity on the bond and r is the return on a risk free bond of similar endowments, τ is the time to maturity of the bond R, Ф(x) is the standard normal cumulative distribution function of h1 (.) and h2 (.) . According to Merton (1974), the variable that remains undefined in the model is σ V2 . It represents the volatility of the firm’s assets and it follows an Ito process. The extrapolation of the market model to Merton (1974) involves two important characteristics of Ito’s process. First, the drift term in equation (4) changes over time according to investor’s future expectations. These expectations are based only on past information. Second, variance of the value of the firm possesses Markov properties, such that it is stochastic, and independent of past information. These two characteristics are identical to the properties of the basic expectations model of equation (1), such that both equations have highly similar implications. Therefore, if parameters αˆ j and βˆ j are unbiased estimates of α j and β j in equation (1), then, the expected returns of time t can be estimated by the information set Φ T 6. If this holds, then the excess return εˆ j,t must be an unbiased estimator of the variance of the firm. It follows that the distributional assumptions of the associated variables are identical and thus, parallel inferences can be drawn within7. Therefore, the above a priori allows the stochastic variability of the value of the firm to be approximated by the parameters of the market model of equation (3).

2.2

Obtaining approximations for the alternative model

To operationalize the alternative approach, first it involves deciding the estimation period for which αˆ j and βˆ j to be approximated by ordinary least squares (OLS). In order to

capture past information Φ T about the firm in the past 1 year, the length of the estimation

5 6 7

Refer to Merton (1974) for a detailed derivation of the equation. Refers to the total set of past information Φ collected over days T which is required to estimate α j and

βj

.

See Appendix C for details.

6

window should consist of 250 days8. The pre-announcement window is set at 20 days between the estimation period and the event day. This is to ensure that prior expectations are controlled for. Thus the estimation window stretches from days -270 to -21 and the preannouncement window from days -21 to -1. In our approach, we assume that every day is an event day. Thus, on event day t 0,1 , ε j,t will be obtained from the market model of equation (3). The same procedure is applied to subsequent days t 0, 2 ......t 0,n following day t 0,1 , which is again treated as an event day. Here, n denotes the nth day in the stock data series and 0 represents an event day, so, if the series is 5 years, n will be 1250 where t o ,1250 . In other words, for a stock series from 1998 to 2003, 1250 αˆ j and βˆ j will be estimated, and 1250 of εˆ j,t will be obtained to form an excess return series of 1250 observations.

This procedure resembles a typical moving average model except that an independent OLS is being executed on a daily (revolving) basis for the entire data series. The resulting excess return series consisting of εˆ j,t from T0,1 to T1,1250 is assumed to be an unbiased proxy for the variability of the firm, and hence can be extrapolated to Merton’s model as shown in equation (5).

2.3

Implementation of the alternate approach

2.3.1 The regression model

As demonstrated, obtaining the excess return series εˆ j ,t allows the daily impact of firm specific information on bond prices to be measured via Merton’s model of equation (5). Unfortunately, as experienced by previous studies, a direct implementation of Merton’s model has proven to be problematic. For instance, Campbell and Taksler (2003) have made a series of unrealistic assumptions about the leverage ratios of the firm in order to operate Merton’s model. To avoid similar problems, a simple regression will be employed with ε j ,t as an explanatory variable (see Collin-Dufresne et al, 2002). 8

It is common for estimation periods in event studies to span a year although reasons have rarely been provided.

7

Since the regression model seeks explanatory powers as Merton (1974), the choice of independent variables must possess similar characteristics to those of equation (5). Otherwise, adequate control measures must be instituted in order not to deviate away from theory. Moreover, since this study is exploratory in nature, it will be most appropriate to maintain parsimonious approach.

The following regression model is specified according to the conditions set by Merton (1974).

RBPj ,t = β1 + β 2 Bindret t + β3 IRt + β 4 S i ,t + β5 S i ,t −1 + β 6 S i ,t − 2 + ηb ,t

(7)

where,

RBPj ,t = logarithmic returns on the clean price of a bond issued by firm j at time t. Our study is based only on investment grade bonds that may tend to display far less variations as compared to those of non-investment grades. Thus, it is of utmost importance to take these issues into account when interpreting and analyzing the regression output.

Bindrett = the logarithmic returns of the bond index at time t While the Merton model subtracts r, a risk free bond with similar endowments from the corporate bond (maturity and coupon matching) to control for systematic influences, the same technique cannot be applied to bond prices. Thus, a control variable is included in the regression. In this case, the Australian bond index (ABI) compiled by Datastream is used as a proxy for systematic fluctuations as it aggregates all the bonds traded in the domestic market. The ABI disregards the special features of each bond while taking into account capital appreciation and coupon payments.

IRt = the prevailing market interest rate at time t Bond yields are found to be related to future expectations of interest rates. Therefore, an interest rate proxy must be included to control for short-term interest rate movements. Variable r in Merton’s model provides such an approximation. Here, the Australian 38

month Bank Bill Swap Rate (BBSW) controls for interest rate variations. In addition, a combination of Bindret t and IRt captures the effects of level and slope of term structure, which is found by Longstaff and Schwartz (1995) to be inversely related to credit spreads (or default risk). S j ,t = excess return as a proxy for asset volatility ε Vj ,t , at time t as discussed in previous

section. To understand and measure the effect of asset volatility, several transpositions of ε Vj ,t will be tested. This will be discussed in a later section. S i ,t − n = the lagged excess return as a proxy for the lagged effects of asset volatility ε Vj ,t at

time t-1, and t-2. Substantial evidence about stock returns leading bond returns by Kwan (1996) suggest that stocks tend to take a less time in impounding information. A similar effect may be observed in asset volatility and bond prices. Thus, the lagged effects for two days are also included9.

2.3.2 Empirical tests of asset volatility under different sub-sample groups To increase the robustness of the test, the sample is grouped according to several characteristics of the bond or firm that may influence by the asset volatility. (i)

Financial and non-financial institutions Financial institutions have a far more complex capital structure as compared to nonfinancial institutions and bond holders may react differently to the volatility of their

9

Before the regression model is executed, a Durbin-Watson test is conducted. Unsurprisingly, autocorrelation is present in majority of the bonds. Thus, to minimize the effect, autoregressive terms up to 3 days are included in the model. Very rarely will a bond price series require adjustments up to 3 days, while most will require a day or two. For convenience, these terms will not be presented in the regression output as they do not offer value to the analysis

9

asset returns. Therefore, it is necessary to differentiate these two groups particularly in Australian markets where financial institutions are the dominant issuers. To further the analysis, financial institutions are segregated into banks and non-banks10 where necessary. (ii)

Standard and Poor’s credit ratings of individual firms It is widely documented in both event studies and default price modeling that bond holders respond accordingly to credit rating changes. To account for such effects firms will be grouped accordingly.

(iii)

Maturity Corporate bonds of various maturities are found to react differently to asset volatility (Collin-Dufresne et al.(2001)). Hence, the sample will be grouped according to various maturity groups.

(iv)

Coupon rate The coupon effects are not directly observed in bond prices. Thus, this needs to be addressed by grouping bonds into their respective coupon rates. Besides, bond holders of different coupons rates may not react in the same manner.

(v)

Leverage According to Merton (1974), a firm defaults when leverage ratio approaches unity. In this respect, bond holders should be concerned with increased leverage ratios. Hence we expect bond prices be responsive to information content and asset volatility.

(vi)

Market-to-book value

Collin-Dufresne et al. (2001) finds that a jump in firm value increases credit spreads. This is consistent with Merton’s view of bond holders as sellers of put options. Unfortunately, jumps in value of firms in the sample are infrequent. Alternatively, the changes in

10

Financial institutions that do not fall in the category of banks in GICS will be known as non-banks.

10

investor’s perceived value of the firm can be sampled by grouping according to various market-to-book ratios.

2.3.3 Transpositions of asset volatility The estimated term ε Vj ,t may be interpreted in several ways in order to investigate bond price behavior.

Transposition 1 First, according to the information content hypothesis, a symmetric response can be expected from both equity and bond holders. That is, favorable (unfavorable) news about the firm that induces a positive (negative) abnormal return is expected to cause a corresponding positive (negative) response. Thus, ε Vj ,t and bond price changes are expected to have a positive relation on the event day (transposition 1 denoted by [PE] in the regression output).

Transposition 2 Second, the constant term αˆ j , that is estimated based on realized returns of 270 days prior to the event day captures investors’ future expectations about the firm’s undertakings is assumed to be normally distributed with a mean of zero. According to Elton (1999), the normality assumption is deficient, in a sense that αˆ j may be subjected to a permanent (or long-term) and significant change when the impact of information shocks is overwhelming. For instance, a firm announces a potentially successful joint venture that induces positive sentiments may subsequently lead to a permanent or long-term increase (positive) in αˆ j . Elton (1999) points out that if independent observations are made in periods when true αˆ j is significantly different, measurements of abnormal returns may be severely flawed. That is, when a firm receives a substantial number of permanent shocks such that αˆ j is changed permanently, the market model tends to be falsely rejected. Indeed, a visual inspection of a random sample11 of daily αˆ j for 5 firms in 5 years suggests that fluctuations in αˆ j can be extreme in independent intervals. This suggests that its impact cannot be ignored while 11

This random sample consists of 2 financial institutions and 3 non-financial institutions.

11

estimating abnormal returns. However, since our approach mimic a moving average, it is likely to circumscribe the problem noted by Elton (1999) and hence, this provides motivation to investigate the impact of αˆ j (transposition 2 denoted by [PE+A] in the regression output) on bond prices. This can be achieved by comparing the coefficients of asset volatility ε Vj ,t without αˆ j , and ( ε Vj ,t + αˆ j ). If the inclusion of αˆ j provides noticeably different results and inferences, it thus indicates that αˆ j generated by the alternative method does not have a mean of zero. More importantly, it will suggest that historical mean rather than variance, as exposited by Kwan (1996), is indeed more important in determining bond price and yields.

Transpositions 3 and 4 This transposition involves treating ε Vj ,t as pure signals of negative information about the firm’s assets. As bond holders only receive fixed payments, they are hypothesized to be averse to all forms of variability in a firm’s undertakings such that it is immaterial to them whether expected returns of the firm are positive or negative12. In other words, as long as firm returns are not stable, the prospects of firms meeting their fixed obligations to debt holders will be undermined. The concepts of implied volatility and historical standard deviations used in default pricing models13 provide similar intuition in this regard. To avoid further complicating the estimation procedures in this study, ε Vj ,t can be used to test this hypothesis by simply taking its absolute value ε Vj ,t , and hereafter known as asset deviation/volatility (transposition 3, denoted by [ABS(PE)] in the regression output). To take into account the impact of αˆ j , ε Vj,0 + αˆ j will also be tested (transposition 4, denoted

by [ABS(PE+A)] in the regression output). The actual effect of these two transpositions can be observed in Appendix B.

The values of [ABS(PE)] and

[ABS(PE+A)] are

abnormally large and positive. Skewness and kurtosis are also high. However, it is important to note that in the actual time series, the magnitude of each observation remains the same. The only difference is that these values do not possess a vector (or direction). 12

Another way of distinguishing between information content and asset deviation is that the former has a vector (direction) and the latter does not. This means that information content predicts bond and equity holders to react symmetrically and asset deviation expects an asymmetric response. 13 The most common type of proxy is implied volatility, which is in fact calculated based on historical standard deviations

12

2.3.4 Comparing our approach to Campbell and Taksler (2003) model Transpositions 5 and 6 One of the main objectives of this study is to propose a new approach in estimating firm

volatility. Thus, assessing the performance of ε Vj ,t relative to a benchmark is imperative. To facilitate a fair comparison, while staying within scope and brevity, the highly effective method employed by Campbell and Taksler (2003) will be replicated in this study. They use the mean-adjusted model14, which essentially restricts αˆ j to zero and βˆ j to one when estimating abnormal returns. By restricting αˆ j to zero, the authors assume that bond holders are sensitive to all past information captured within the estimation window. Furthermore, restricting βˆ j to 1 is highly unrealistic as the sensitivity of firms is known to vary over time15. Hence, making a comparison is crucial in determining the robustness of the model in this study. The exact procedure of Campbell and Taksler (2003) is being implemented to the sample firms in this study, and the resulting abnormal returns π Vj, 0 (transposition 5, denoted by [MAR(180)] in the regression output) will be used as a proxy for ε Vj ,t . In addition, estimation periods of 270 days will be used in this study (transposition 6, denoted by [MAR(270)] in the regression output). The purpose is to prevent any biasness that may arise from differing estimation parameters of both models.

2.3.5 Hypothesis testing and expected coefficient signs Bindret

The bond index return is expected to be positively related to bond price. Thus, the expected coefficient sign is positive (+).

14

Brown and Warner (1985) defines the Mean Adjusted Returns Model as the difference between actual return and the average of security return for an estimation period. 15 This phenomenon is being widely documented in asset pricing literature, which is out of the scope of this study.

13

BBSW3 The bank-bill swap rate is expected to have an inverse relationship with bond price. The expected coefficient sign is negative (-). Transpositions of ε Vj ,t

[PE], [PE+A], [MAR(180)] and [MAR(270)] are tests for the information content hypothesis such that positive information about the firm is expected to induce a positive (symmetrical) response from bond holders. Hence the expected sign is positive (+). While [ABS(PE)] and [ABS(PE+A)] are tests for asset volatility such that bond holders are expected to react negatively (asymmetrical) to all information. Therefore, the expected sign is negative (-). 2.3.6 Statistical Testing

The above methodology involves running numerous regressions. For each transposition of volatility estimate, a regression model is required for each bond. Thus, for 88 bonds, 704 regression output needs to be analyzed. This impedes meaningful statistical inference, and thereby, requires an aggregate statistic. Hence, throughout the next section, the reported coefficient values and their associated t-statistics will be calculated in accordance with the method used in Collin-Dufresne et al. (2001):   1 t − statistics =  Σ nj β j n  

   1  nj Σ nj (β j − β j )   n −1 

(14)

where, n = the number of bonds in the analysis j = jth bond

β j = coefficient value of bond j

14

3

Data

Daily corporate bond prices are obtained from Datastream for a period from 1 July 1998 to 31 March 2004. Most bonds do not start and expire within this period although either situation can occur. Corporate bond issues are much thinner as compared to other developed markets. As such, similar sampling periods that are used in other studies may not be replicated. Nonetheless, this period provides an interesting basis for the study as Australia records strong corporate bond growth (refer to Figure 1). Figure 1 Total face value of long-term issues by type of borrowers from 1998 to 2003. Historical data are obtained from the Reserve Bank of Australia. Offshore issues have been substantially higher than domestic issues and have exceeded government issues since June 2000.

250

200

150

$b

Government issues

100 Domestic private issues

50

Private offshore Issues

0 Ju

3.1

n-9

8

De

c -9

8 Ju

n-9

9

De

c -9

9 Ju

0 n-0

De

c -0

0 Ju

1 n-0

De

c -0

1 Ju

2 n-0

De

c -0

2 Ju

n-0

3

De

c -0

3

Sampling criteria

The following selection criteria are being applied to all the corporate bonds available in Australia i)

Issuers must be locally incorporated and listed on the Australian Stock Exchange.

ii)

Issuers must issue straight debt bond that pay a fixed coupon. 15

iii)

Issuers must be assigned an investment grade by Standard and Poor’s and firm ratings are used instead of bond ratings in order to reflect credit worthiness of the firm rather than external characteristics.

iv)

Bonds must not be convertible, not callable and not puttable16. These features tend to confound the relation between asset information and prices of the bonds.

v)

Bonds must not be backed by assets and thus truly reflect the creditability or asset volatility of the issuer.

vi)

Bonds must be denominated in Australian dollars.

vii)

No defaults or rating revisions of more than a grade have occurred prior to the sample period. For example, from AA to B or BBB to C

viii)

Only bonds with more than 100 transactions within the sample period are being included in the analysis.

ix)

Thinly traded bonds are being excluded to avoid non-synchronous and estimation problems.

x)

Transactions that are missing and inconsistent with issue and expiry dates are also being excluded.

3.2

Final bond sample description and bond characteristics

Following the strict selection criteria, only 22 firms (see Table 1) are available for the study (15 firms that issued bonds were excluded in the filtering process). These 22 firms have a total of 88 bonds suitable for the study, of which, each bond has an average of 650 daily observations (see Table 1). These observations are recorded in Datastream as market price of the bond17. The price is calculated based on the average daily prices transacted in the market, less accrued interest18. The coupon rates of these bonds are fairly well distributed, with 23 bonds paying rates between 5% and 6%, 30 bonds paying between 6% and 7%, and the last 35 bonds pay 16

Jones et al. (1984) and Duffee (1998) argues that callable bonds and straight bonds react differently to the same term structure of interest rates. 17 The market price of a bond is synonymous to its clean or gross price. 18 Definitions are obtained from Datastream-extranet.

16

coupons more than 7%. Maturities of these bonds mainly fall within 4 to 6 years (52 bonds). Only 24 bonds have short maturities of 1 to 3 years, while 12 bonds have longer maturities of 7 years and above (se Table 1). Lastly, all firms within the sample are given long-term investment grade ratings, between AA and BB+ (see Table 2). Transactions Firm Australia Gas Light AMP Australia New Zealand Bank BHP Billiton Boral Caltex Commonwealth Bank of Australia Coca-Cola Amatil Coles Myer John Fairfax MacQuarie Bank National Australia Bank Origin Energy Publishing and Broadcasting Qantas SouthCorp Stockland Trust Suncorp-Metway Telstra WestField Trust WestPac Banking Corp Woolworths 22 Total

No. of Bonds 2 6 5 3 1 1 15 3 2 1 1 17 1 1 1 2 5 5 6 3 5 2 88

Maturity

Average No. of 1 to 3 No. of transaction per bond for years Transactions each firm 1653 3456 3368 1754 637 339 10601 2282 2181 246 617 12880 505 748 617 1355 1812 2508 3896 1911 2685 1181 57232

827 576 674 585 637 339 707 761 1091 246 617 758 505 748 617 678 362 502 649 637 537 591 650

2 1

1 4

1 10

2 2 1

24

4 to 6 years 2 4 3 2 1 7 3 2 1 7 1 1 1 1 2 3 2 3 5 1 52

7 years and above

1 1

4

5% to 6%

6% to 7%

2 2

5 1

1 3 2 1 1 1 5

7% and above 1 1 1 2

5

2 1 5

1 1 3

3 2 1 2

1 12

23

1 3

9 1 1 1 2

2 2

1 5 1 2 1 35

2 1 1 30

Table 1This table presents the distribution of bonds according to the number of firms, maturity and coupon size.

17

Firm

GICS Sector

GICS Industry Group (sub-industry)

Standard and Poor's Longterm Credit Rating

Australia Gas Light

Utilities

Utilities

A

AMP

Financials

Insurance

BBB+

Australia New Zealand Bank

Financials

Banks

AA-

BHP Billiton

Mateirals

Mateirals

A+

Boral

Materials

Materials (construction)

BBB+

Caltex

Energy

Energy

BBB+

Commonwealth Bank of Australia

Financials

Banks

AA-

Coca-Cola Amatil

Consumer staples

Food, beverage and tobacco

A-

Coles Myer

Consumer staples

Food and staples retailing

BBB

John Fairfax

Consumer discretionary

Media

BBB

MacQuarie Bank

Financials

Diversified financials

A

National Australia Bank

Financials

Banks

AA-

Origin Energy

Energy

Energy

A-

Publishing and Broadcasting

Consumer discretionary

Media

A-

Qantas

Industrials

Transportation

BBB+

SouthCorp

Consumer staples

Food, beverage and tobacco

BB+

Stockland Trust

Financials

Real estate (Investment trusts)

A-

Suncorp-Metway

Financials

Diversified financials

A

Telstra

Telecommunications Services

Telecommunications Services

A+

WestField Trust

Financials

Real estate (Investment trusts)

A

WestPac Banking Corp

Financials

Banks

AA-

Woolworths

Consumer staples

Food and staples retailing

A-

Table 2 The table presents the name, the sector in which it is classified under GICS sector and industry, and credit ratings by Standard and Poor’s. Throughout the analysis, these firms are further grouped according to “banks”, “financial services” and “non-financials”. “Banks” refer to authorized deposit taking institutions, “financial services” refer to financials that are not classified as “banks”, and all other firms fall under “nonfinancials”.

3.3 Stock prices and Industry classifications

Daily closing stock prices of the 22 firms are also retrieved from Datastream. The sampling period for each firm is from January 1997 to March 2004. This includes the estimation period required for the model to obtain αˆ j and βˆ j . The thin trading problem also occurs in stock transactions, but fortunately, they tend to occur on early dates (near the start of 1997). As such, corresponding bond transactions will also have to be excluded under the sampling criteria to avoid non-synchronous problem.

18

The sample firms are being classified according to sectors (GICS), industry groups and sub-industries. Out of 22 firms, 9 are belong to the financials sector; of which, 4 are being categorized as banks and 4 others are non-banks. The remaining 13 firms are from various industries such as utilities, energy and general service providers (see Table 2).

4. Empirical results 4.1 Issuer type 4.1.1 All firms

In Table 3, the coefficient values of S for [PE] and [PE+A] are highly statistically significant, suggesting a strong support for information content hypothesis. The coefficient values are 0.0034 and 0.0034 respectively indicating that in general, bond holders respond contemporaneously to firm specific information. Nonetheless, statistical significance of coefficients of S(-1) for [PE] and [PE+A], each with values of 0.0017 further indicating that bond holders also tend to respond to past information, albeit not beyond the second day of information arrival. Since the coefficient values of [PE+A] are statistically similar to [PE], it can be assumed that the influences of past returns αˆ j has been minimized. However, the coefficients of S for [MAR(180)], [MAR(270)] of mean-adjusted model are not statistically significant, but they are statistically significant for S(-1). This implies that bond holders react to information much slower under the mean-adjusted model. On the whole, the price behavior of corporate bond holders in Australia appears to be well explained by the information content hypothesis while the S for [ABS(PE)] and [ABS(PE+A)], tests for asset volatility, does not seem to be of any influence. By comparison, [PE] appears to be a better estimate than [MAR(180)] and [ MAR(270)] given that the [PE] coefficient values are larger and statistically significant.

19

Transposition All Firms

1

2

3

PE

PE+A

-0.0001 -5.8085

-0.0001 -5.9865

-0.0001 -3.3930

0.3019 8.1602

0.3035 8.2141

-0.0305 -12.3833

S

4

6

MAR(180)

MAR(270)

-0.0001 -3.5061

-0.0001 -5.9219

-0.0001 -5.9996

0.3047 8.2258

0.3047 8.2252

0.3042 8.2308

0.3003 8.0993

-0.0312 -12.3823

-0.0308 -12.5551

-0.0308 -12.5272

-0.0307 -12.5220

-0.0300 -12.2099

0.0034 4.7292

0.0034 4.7171

-0.0017 -1.6719

-0.0015 -1.5047

0.0002 0.2581

0.0001 0.0975

S(-1)

0.0017 2.5756

0.0017 2.5334

0.0001 0.0949

0.0003 0.3743

0.0012 2.4861

0.0012 2.3704

S(-2)

-0.0002 -0.2398

-0.0002 -0.2458

-0.0001 -0.1248

-0.0004 -0.3177

-0.0003 -0.5168

-0.0003 -0.5698

0.4307 0.4236 88 57232 22

0.4378 0.4308

0.4291 0.4222

0.4308 0.4238

0.4312 0.4242

0.4276 0.4195

C BINDRET BBSW3Y

R-squared Adjusted R-squared Number of bonds Number of Observations Number of firms

ABS(PE) ABS(PE+A)

5

Table 3 Issuer type: all firms. This table presents the coefficient estimates and the associated t-statistics (immediately beneath) of all firms in the sample. Transpositions 1 to 2 test for the information content, 3 and 4 test for asset deviation; 5 and 6 refer to the estimates of the mean-adjusted model. The leftmost column displays the regression equation (7) RBPj,t = β1 + β2 Bindret + β3 BBSW3 + β4 Sj,t + β5 Sj,t-1 + β6 Sj,t-2 + εt , where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate

4.1.2 Non-financial institutions

Table 4 shows the regression results that exclude financial institutions. Only coefficient values for [ABS(PE)] and [ABS(PE+A)] on the event day are statistically significant at 5% level (-0.0075 and -0.0072 respectively). The negative coefficient, as hypothesized, indicates that bond holders tend to go short in response to asset volatility. This is highly consistent with the propositions of Merton (1974). However, it can be inferred that bond that are issued by non-financial institutions are not influenced by specific information content since coefficients of S for [PE] and [PE+A] are not statistically significant.

20

1

2

3

Non-Financial Institutions

PE

PE+A

ABS(PE)

C

-0.0001 -2.0169

-0.0001 -2.1085

0.0001 0.7722

0.0001 0.6627

-0.0001 -2.0732

-0.0001 -2.0711

BINDRET

0.3610 5.5605

0.3663 5.6818

0.3671 5.6907

0.3670 5.6919

0.3653 5.6999

0.3654 5.7005

BBSW3Y

-0.0345 -6.6032

-0.0355 -6.8438

-0.0355 -6.9201

-0.0356 -6.9232

-0.0354 -6.8296

-0.0354 -6.8314

S

0.0011 0.7722

0.0011 0.7546

-0.0075 -3.0449

-0.0072 -3.0530

-0.0028 -1.7488

-0.0028 -1.7778

S(-1)

0.0013 0.6394

0.0012 0.5763

-0.0022 -0.7700

-0.0019 -0.6489

0.0008 0.5771

0.0009 0.6032

S(-2)

-0.0008 -0.3498

-0.0009 -0.4014

-0.0012 -0.4092

-0.0014 -0.4567

-0.0001 -0.0139

0.0001 0.0121

0.4258 0.4186 26 39838 13

0.4270 0.4202

0.4254 0.4189

0.4268 0.4200

0.4274 0.4207

0.4276 0.4208

Transposition


4

5

ABS(PE+A) MAR(180)

6 MAR(270)

Table 4. Issuer type: non-financial institutions. This table presents the coefficient estimates and the associated t-statistics (immediately beneath) of the sample excludes financial institutions. Transposition 1 to 2 test for the information content, 3 and 4 test for asset deviation; 5 and 6 refer to the estimates of the meanadjusted model. The leftmost column displays the regression equation (7) RBPj,t = β1 + β2 Bindret + β3 BBSW3 + β4 Sj,t + β5 Sj,t-1 + β6 Sj,t-2 + εt , where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate.

4.1.3 Financial institutions

Given that financial institutions have highly leveraged balance sheets, bond holders are more readily respond to information about their performance. However, a comparison between highly rated financial institutions and low rated non-financial institutions has been ignored by past literature. Frequently, financial institutions are either excluded from samples, or little control is implemented. Thus, such a comparison will be feasible with a relatively more manageable sample size. In Table 5, the price behavior exhibited by bond holders of financial institution contrasts, in an interesting manner, to that of non-financial institutions. The coefficient values of S and S(-1) are statistically significant for all transpositions of [PE] and [MAR] regressions, with the exception of [ABS(PE)] and [ABS(PE+A)]. The coefficient values are also higher than 21

those of non-financial institutions; although, R-squared is marginally lower. With regards to information content and asset deviation, the statistical results imply that bond holders react in a mutually exclusive manner. That is, bond holders of financial institutions respond exclusively to information content, while bond holders of non-financial institutions respond exclusively to asset volatility. This is a strong indication of how bond holders can treat information differently such that it warrants a distinction.

1

2

3

Financial Institutions

PE

PE+A

ABS(PE)

C

-0.0001 -6.9965

-0.0001 -7.0540

-0.0001 -7.5237

-0.0001 -7.3085

-0.0001 -6.8651

-0.0001 -7.1752

0.2727 6.2424

0.2728 6.2443

0.2740 6.2506

0.2741 6.2500

0.2741 6.2569

0.2686 6.1148

-0.0288 -10.8546

-0.0292 -10.6329

-0.0287 -10.8362

-0.0287 -10.8028

-0.0286 -10.8766

-0.0277 -10.5139

S

0.0041 5.2293

0.0042 5.2551

0.0007 0.8294

0.0009 1.0317

0.0013 2.2771

0.0011 2.0698

S(-1)

0.0020 4.6114

0.0021 4.5541

0.0003 0.4403

0.0005 0.7657

0.0015 4.2153

0.0014 3.6937

S(-2)

-0.0003 -0.6184

-0.0003 -0.5304

0.0005 0.5717

0.0003 0.3066

-0.0008 -1.7654

-0.0008 -2.0594

0.3655 0.3057 62 17394 9

0.3734 0.3266

0.3722 0.3249

0.3724 0.3253

0.3731 0.3263

0.3699 0.3222

Transposition

BINDRET BBSW3Y


4

5

ABS(PE+A) MAR(180)

6 MAR(270)

Table 5. Issuer type: financial institutions. This table presents the coefficient estimates and the associated tstatistics (immediately beneath) of all firms after non-financial institutions are removed. Transposition 1 to 4 tests for the information content, 5 and 6 tests for asset deviation; 7 and 8 refers to the estimates of the meanadjusted model. The leftmost column displays the regression equation (7) RBPj,t = β1 + β2 Bindret + β3 BBSW3 + β4 Sj,t + β5 Sj,t-1 + β6 Sj,t-2 + εt , where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate.

4.2

Credit ratings

4.2.1 AA and AA-

This is the highest rating class in the Australian markets consisting of 4 largest commercial banks (see Table 2). The statistical results are similar to the test for financial-institutions 22

presented in the previous section. From Table 6, coefficient values of S for [PE] and [PE+A] are statistically significant at 0.0047 each, while coefficient values of S(-1) are statistically significant at 0.0019 each, thus indicating that the information content can explain the bond price behaviors. The regression results of [MAR(180)] and [MAR(270)] appear less reliable and less consistent with the information content hypothesis as only the coefficients of lagged S(-1) display evidence. Lastly, the test for asset volatility [ABS(PE)] and [ABS(PE+A)] is only significant at S(-2) with coefficient values of 0.0024 and 0.0023 respectively. This suggests that bond holders delay their responses with respect to asset volatility.

1

2

3

AA/AA- rated firms

PE

PE+A

ABS(PE)

C

-0.0001 -5.7457

-0.0001 -5.8261

-0.0001 -6.8128

-0.0001 -6.7997

-0.0001 -5.5813

-0.0001 -5.7812

BINDRET

0.2236 5.2691

0.2236 5.2717

0.2248 5.2685

0.2248 5.2695

0.2255 5.2794

0.2180 5.0871

BBSW3Y

-0.0306 -9.5619

-0.0305 -9.5153

-0.0306 -9.5599

-0.0305 -9.5185

-0.0303 -9.5419

-0.0294 -9.2739

S

0.0047 4.9323

0.0047 4.9402

0.0015 1.5135

0.0018 1.7432

0.0011 1.8069

0.0010 1.6785

S(-1)

0.0019 3.9841

0.0019 3.8749

0.0005 0.5625

0.0006 0.7380

0.0015 3.8631

0.0015 3.8670

S(-2)

0.0005 1.1230

0.0005 1.1270

0.0024 2.0417

0.0023 2.0704

-0.0001 -0.4069

-0.0003 -0.7588


0.3808 0.3736 42 29534 4

0.3818 0.3745

0.3797 0.3724

0.3807 0.3733

0.3810 0.3736

0.3777 0.3680

Transposition

4

5

ABS(PE+A) MAR(180)

6 MAR(270)

Table 6. Credit rating: AA and AA-. This table presents the coefficient estimates and the associated tstatistics of all firms in the sample with AA and AA- credit ratings which happen to only consist of banks. Transposition 1 to 2 test for the information content, 4 and 5 test for asset deviation; 5 and 6 refer to the estimates of the mean-adjusted model. The leftmost column displays the regression equation (7) RBPj,t = β1 + β2 Bindret + β3 BBSW3 + β4 Sj,t + β5 Sj,t-1 + β6 Sj,t-2 + εt , where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate.

23

4.2.2 A+, A and A-

This category is made up of 7 non-financial firms and 4 financial service firms. Owing to the mixture of firms, the statistical results are also confounded as shown in panel 1 of Table 7. Both hypotheses are being supported by statistically significant coefficients of S for [PE], [PE+A], [ABS(PE)] and [ABS(PE+A)] with values of 0.0036, 0.0036, -0.007 and 0.0068 respectively. Due to the mixed results, it will not be possible to draw any reliable inferences. The composition effect is the most likely reason for this inconsistency. As such the 4 firms providing financial services is excluded from the sample and the test is rerun. The results presented in panel 2 of Table 7 justify the composition effect suggested above. Only coefficients of S for [ABS(PE)] and [ABS(PE+A)] are statistically significant, with values of -0.011 and -0.0106 respectively indicating that price changes of the bonds of these firms are highly associated with asset volatility while information content play a smaller role. These results are highly consistent with those reported in Table 4. Above that, the most intriguing aspect is that the values of the coefficient of firms with ratings of A+, A and A- excluding financials (panel 2 of Table 7) are larger (negatively) than those reported in Table 4. This suggests that higher graded non-financial firm bonds are more sensitive to firm volatility as compared to the average graded non-financial firm bonds. This is incongruent with the notion that bonds that issued by lower rated firms tend to be more volatile since their underlying assets are less stable. To confirm these findings, the same testing procedure will have to be applied to lower rated firms.

24

Panel 1 T ransposition A +/A /A - rated firm s

1

2

3

4

5

6

PE

PE +A

A B S(PE )

A B S(PE +A )

M A R (180)

M A R (270)

-0.0001 -2.9644

-0.0001 -3.1217

-0.0001 -0.7016

-0.0001 -0.8200

-0.0001 -3.0612

-0.0001 -3.1047

B IN D R E T

0.4293 5.7313

0.4333 5.8162

0.4350 5.8346

0.4349 5.8309

0.4328 5.8201

0.4328 5.8257

B B SW 3Y

-0.0292 -6.2716

-0.0310 -6.3896

-0.0299 -6.4693

-0.0301 -6.4779

-0.0301 -6.4768

-0.0300 -6.4711

S

0.0036 2.6351

0.0036 2.6101

-0.0070 -3.0189

-0.0068 -3.0834

-0.0004 -0.2945

-0.0006 -0.4152

S(-1)

0.0016 0.9197

0.0015 0.8694

0.0018 0.7733

0.0025 1.0414

0.0002 0.1936

0.0002 0.1499

S(-2)

-0.0021 -1.1913

-0.0021 -1.2201

-0.0032 -1.3465

-0.0037 -1.4931

-0.0014 -0.9361

-0.0012 -0.8077

0.4958 0.4890 30 17727 11

0.5131 0.5066

0.4937 0.4873

0.4965 0.4901

0.4961 0.4897

0.4941 0.4875

C

R -squared A djusted R -squared N um ber of bonds N um ber of O bservations N um ber of firm s

Panel 2 A +/A /A - rated firm s excluding financial institutions

1

2

3

4

5

6

PE

PE +A

A B S(PE )

A B S(PE +A )

M A R (180)

M A R (270)

-0.0001 -1.6701

-0.0001 -1.8208

-0.0001 -0.0401

-0.0001 -0.2048

-0.0001 -1.7723

-0.0001 -1.7777

B IN D R E T

0.4093 4.8364

0.4167 4.9832

0.4180 5.0057

0.4175 5.0029

0.4154 4.9992

0.4157 5.0007

B B SW 3Y

-0.0345 -5.0852

-0.0360 -5.3482

-0.0360 -5.4133

-0.0362 -5.4266

-0.0360 -5.3584

-0.0360 -5.3583

S

0.0029 1.4436

0.0029 1.4168

-0.011 -3.1658

-0.0106 -3.1689

-0.0027 -1.2055

-0.0027 -1.2417

S(-1)

0.0009 0.3231

0.0008 0.2650

0.0032 0.8254

0.0038 0.9653

-0.0007 -0.3322

-0.0006 -0.2947

S(-2)

-0.0009 -0.3090

-0.0010 -0.3323

-0.0017 -0.4186

-0.0021 -0.4877

0.0003 0.1075

0.0004 0.1727

0.4420 0.4352 16 10879 7

0.4432 0.4368

0.4415 0.4354

0.4432 0.4367

0.4434 0.4369

0.4434 0.4370

C

R -squared A djusted R -squared N um ber of bonds N um ber of O bservations N um ber of firm s

Table 7. Credit rating: A+/A/A-. This table presents the coefficient estimates and the associated t-statistics of all firms with a credit rating of either A+/A/A. Panel 1 presents all firms with this rating and panel 2 presents the sample when banks and financial services are removed. Transposition 1 to 2 tests for the information content, 3 and 4 tests for asset deviation; 5 and 6 refers to the estimates of the mean-adjusted model. The leftmost column displays the regression equation (7) RBPj,t = β1 + β2 Bindret + β3 BBSW3 + β4 Sj,t + β5 Sj,t-1 + β6 Sj,t-2 + εt, where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate.

25

4.2.3

BBB and BB+

Before excluding financial firms from this group19, verification is conducted with all firms with BBB and BB+ ratings. The results are reported in panel 1 of Table 8. Only coefficients depicting the information hypothesis is statistically significant, albeit having a minor effect. The coefficient values of S(-2) for [PE] and [PE+A] are 0.0028 each, and S(2) for [MAR(180)] and [MAR(270)] are 0.0023 and 0.0021 respectively. Thus, in Australia, B-rated bonds are not subjected to asset volatility unlike A-rated bonds. Also, on the average, a two day lag can be expected, that is B-rated bond returns are rather slow in response to information content. . Panel 2 of Table 8 reports the results when firms in the financial sector are excluded. As anticipated, more evidence of information content is present as the coefficient values of S(2) for [PE], [PE+A], [MAR(180)] and [MAR(270)] are higher and statistically significant than those of A+, A and A- rated non-financials: 0.0033, 0.0032, 0.0027 and 0.0027 respectively. On the other hand, the coefficients of S for [ABS(PE)] and [ABS(PE+A)] are statistically significant suggesting that changes in bond prices are also influenced by asset volatility. These coefficients appear to have decreased (in absolute term) in comparison to those of non-financials with A+, A and A- ratings (in Table 7 Panel 2). This is anomalous to previous studies20 (for examples, Collin-Dufresne et al., 2001 and Campbell and Taksler, 2003) that frequently record the association between lower credit ratings and firm volatility. Our results may suggest that bond investors will only pay attention to asset volatility of the non-financial firms when their credit ratings are high, but for lower credit rating non-financial firms, they rely on both forms of information. This is highly consistent with the fact that speculation is more active in lower rated bonds. ..

19 20

None of the banks in the full sample falls in this category. For brevity of discussion, only recent studies will be mentioned.

26

Panel 1 Transposition BBB/BB+ rated firms C

1

2

3

4

5

6

PE

PE+A

ABS(PE)

ABS(PE+A)

MAR(180)

MAR(270)

-0.0001 -3.2196

-0.0001 -3.2547

0.0001 0.1832

0.0001 0.2720

-0.0001 -3.2311

-0.0001 -3.3940

BINDRET

0.2749 4.4110

0.2753 4.4127

0.2748 4.4259

0.2751 4.4280

0.2747 4.4118

0.2737 4.3803

BBSW 3Y

-0.0332 -6.2505

-0.0332 -6.2468

-0.0332 -6.2858

-0.0333 -6.2780

-0.0330 -6.2890

-0.0320 -5.8384

S

-0.0007 -0.7876

-0.0006 -0.7395

-0.0009 -0.6972

-0.0011 -0.8499

-0.0016 -1.6331

-0.0017 -1.7309

S(-1)

0.0011 1.4154

0.0012 1.3936

-0.0015 -1.1094

-0.0017 -1.2969

0.0013 1.3631

0.0011 1.2289

S(-2)

0.0028 3.9024

0.0028 4.0650

-0.0034 -1.1743

-0.0034 -1.1889

0.0023 3.6164

0.0021 3.4967


0.4329 0.4259 15 9468 7

0.4338 0.4272

0.4317 0.4248

0.4324 0.4258

0.4348 0.4282

0.4281 0.4215

1

2

3

4

5

6

PE

PE+A

ABS(PE)

ABS(PE+A)

MAR(180)

MAR(270)

Panel 2 BBB/BB+ rated firms excluding financial institutions C

-0.0001 -3.1120

-0.0001 -3.1365

0.0000 0.6389

0.0000 0.7071

-0.0001 -3.0909

-0.0001 -3.0575

BINDRET

0.3302 3.6515

0.3307 3.6538

0.3301 3.6549

0.3305 3.6581

0.3300 3.6515

0.3301 3.6518

BBSW 3Y

-0.0346 -4.7061

-0.0347 -4.7024

-0.0346 -4.7579

-0.0346 -4.7445

-0.0344 -4.7358

-0.0344 -4.7322

S

-0.0009 -0.8076

-0.0008 -0.7931

-0.0025 -2.6499

-0.0028 -2.6742

-0.0024 -1.7916

-0.0024 -1.8089

S(-1)

0.0004 0.3596

0.0004 0.3637

-0.0025 -1.1308

-0.0025 -1.2450

0.0009 0.6108

0.0009 0.5796

S(-2)

0.0033 2.9565

0.0032 3.0146

-0.0061 -1.3115

-0.0059 -1.3163

0.0027 2.8342

0.0027 2.9235


0.4046 0.3980 9 6012 5

0.4057 0.3995

0.4044 0.3980

0.4052 0.3989

0.4066 0.4005

0.4071 0.4009

Table 8. Credit rating: BBB and BB+. This table presents the coefficient estimates and the associated tstatistics of all firms with BBB and BB+ credit ratings. Panel 1 presents all firms and panel 2 displays the sample when banks and financial services are removed. Transposition 1 to 2 tests for the information content, 3 and 4 tests for asset deviation; 5 and 6 refers to the estimates of the mean-adjusted model. The leftmost column displays the regression equation (7) RBPj,t = β1 + β2Bindret + β3BBSW3 + β4Sj,t + β5Sj,t-1 + β6Sj,t-2 + εt , where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate.

27

4.3 Coupons 4.3.1 Bin 1: bonds with coupons between 4% and 6%

From Table 9, only S for [PE] and [PE+A] are statistically significant with coefficient values of 0.0022 each. S(-1) for [MAR(180)] is also statistically significant at 5%. Therefore, information content hypothesis provides adequate explanation for smaller coupons. Transposition

1

2

3

All firms 4% < C < 6%

PE

PE+A

ABS(PE)

C

4

5

ABS(PE+A) MAR(180)

6 MAR(270)

-0.0001 -3.5546

-0.0001 -3.7041

-0.0001 -2.5971

-0.0001 -2.7680

-0.0001 -3.7426

-0.0001 -3.5980

BINDRET

0.2739 4.6590

0.2739 4.6605

0.2753 4.6554

0.2754 4.6546

0.2756 4.6929

0.2620 4.3986

BBSW3Y

-0.0265 -6.1763

-0.0264 -6.1524

-0.0263 -6.2619

-0.0263 -6.2081

-0.0261 -6.1495

-0.0244 -5.8889

S

0.0022 2.9270

0.0022 2.9581

-0.0011 -0.5861

-0.0009 -0.5186

-0.0001 -0.0863

-0.0002 -0.2087

S-1

0.0016 1.8515

0.0016 1.8551

0.0001 0.0532

0.0003 0.1724

0.0014 1.9366

0.0015 2.0237

S-2

-0.0012 -0.7415

-0.0012 -0.7305

0.0002 0.0819

-0.0001 -0.0020

-0.0009 -0.7298

-0.0010 -0.8697

0.4727 0.4671 23 15330 9

0.4727 0.4671

0.4718 0.4662

0.4719 0.4663

0.4735 0.4678

0.4660 0.4564


Table 9. Coupon size bin 1: Bonds with coupons between 4% and 6%. This table presents the coefficient estimates and the associated t-statistics (immediately beneath) of all bonds in the sample. Transposition 1 to 2 test for the information content, 3 and 4 test for asset deviation; 5 and 6 refer to the estimates of the meanadjusted model. The leftmost column displays the regression equation (7) RBPj,t = β1 + β2 Bindret + β3 BBSW3 + β4 Sj,t + β5 Sj,t-1 + β6 Sj,t-2 + εt , where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate.

28

4.3.2 Bin 2: bonds with coupons between 6% and 8%

Information content explains the response of bond holders on event day. Coefficients of S for [PE] and

[PE+A] displayed in Table 10 are statistically significant at 1%, with

coefficient values of 0.0055 each. These values are significantly larger than those of bin 1, suggesting that although information content hypothesis explains bond prices well; and larger coupons may not necessarily be less sensitive to firm specific information. Transposition All firms 6% < C < 8%

1

2

3

PE

PE+A

C

-0.0001 -5.3008

-0.0001 -5.3730

-0.0001 -4.8555

BINDRET

0.3788 -5.3008

0.3834 -5.3730

BBSW3Y

-0.0345 -6.8704

S

4

ABS(PE) ABS(PE+A)

5

6

MAR(180)

MAR(270)

-0.0001 -4.6330

-0.0001 -5.3470

-0.0001 -5.5633

0.3846 -4.8555

0.3849 -4.6330

0.3845 -5.3470

0.3838 -5.5633

-0.0365 -7.0257

-0.0353 -7.0509

-0.0354 -7.0326

-0.0354 -7.1420

-0.0347 -6.9128

0.0055 3.6717

0.0055 3.6749

0.0002 0.0883

0.0003 0.1924

0.0010 1.0604

0.0008 0.8785

S-1

0.0012 1.6007

0.0013 1.5155

0.0003 0.1548

0.0005 0.3255

0.0004 0.5756

0.0003 0.3199

S-2

0.0002 0.2022

0.0002 0.2621

0.0007 0.4253

0.0003 0.1930

-0.0003 -0.3761

-0.0003 -0.3401


0.4722 0.4652 29 19545 15

0.4932 0.4861

0.4682 0.4614

0.4725 0.4655

0.4719 0.4653

0.4672 0.4604

Table 10. Coupon size bin 2: Bonds with coupons between 6% and 8%. This Table presents the coefficient estimates and the associated t-statistics (immediately beneath) of all bonds with coupon sizes between 6% and 8%. Transposition 1 to 2 test for the information content, 3 and 4 test for asset deviation; 5 and 6 refer to the estimates of the mean-adjusted model. The leftmost column displays the regression equation (7) RBPj,t = β1 + β2 Bindret + β3 BBSW3 + β4 Sj,t + β5 Sj,t-1 + β6 Sj,t-2 + εt , where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate.

29

4.3.3 Bin 3: bonds with coupons above 8%

The results in Table 11 show that both information content and asset volatility hypotheses are rejected for bonds with coupons larger than 8%. This suggests that returns on bond that have higher coupon rates are not influenced by information content and asset volatility.

Transposition All firms above 8%

1

2

3

PE

PE+A

-0.0001 -1.1322

-0.0001 -1.1233

0.0001 0.1409

BINDRET

0.1443 1.8196

0.1446 1.8188

BBSW3Y

-0.0214 -2.9555

S

4

6

MAR(180)

MAR(270)

0.0001 0.0607

-0.0001 -1.1791

-0.0001 -1.1254

0.1444 1.8292

0.1443 1.8205

0.1433 1.8096

0.1431 1.8046

-0.0214 -2.9543

-0.0212 -2.9066

-0.0214 -2.9422

-0.0211 -2.9454

-0.0212 -2.9699

-0.0001 -0.1404

-0.0002 -0.2803

-0.0035 -1.2229

-0.0036 -1.2771

-0.0016 -1.0765

-0.0016 -1.1029

S-1

0.0013 1.2841

0.0012 1.1749

-0.0018 -0.6611

-0.0013 -0.4977

0.0019 1.7949

0.0019 1.8089

S-2

0.0011 0.7573

0.0010 0.6919

-0.0044 -1.0645

-0.0045 -1.1236

-0.0002 -0.1510

-0.0003 -0.2060


0.2244 0.2147 10 6355 4

0.2237 0.2142

0.2246 0.2152

0.2260 0.2165

0.2288 0.2185

0.2273 0.2175

C

ABS(PE) ABS(PE+A)

5

Table 11. Coupon size bin 4: Bonds with coupons larger than 8%. This Table presents the coefficient estimates and the associated t-statistics (immediately beneath) of all bonds with coupon sizes larger than 8%. Transposition 1 to 2 test for the information content, 3 and 4 test for asset deviation; 5 and 6 refer to the estimates of the mean-adjusted model. The leftmost column displays the regression equation (7) RBPj,t = β1 + β2 Bindret + β3 BBSW3 + β4 Sj,t + β5 Sj,t-1 + β6 Sj,t-2 + εt , where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate.

30

In summary, the above results (Tables 10 and 11) show that bondholders respond to information content and asset volatility only when they hold bonds that have lower coupon rates. 4.4 Maturity 4.4.1 Bin 1: 1 to 3 years

Table 12 reports the effects of firm specific information on returns of bond with shorter maturities. The coefficient values of S(-1) for PE, [PE+A] are 0.0016 each and are statistically significant at 1%. This suggests that bondholders respond to information content is not contemporaneous. The coefficients of S(-1) for [ABS(PE)] and [ABS(PE+A)] are statistically significant at 5% level, with values of 0.0017 each, suggesting that the response to asset volatility is lagged. While S(-1) for [MAR(180)] and [MAR(270)] are also statistically significant at 5% level (coefficient values are 0.0009 each). Thus, the results suggest that bond returns in this category appear to be influenced by information content of a firm and asset volatility with lagged effect.

31

Transposition All firms 1 to 3 years

1

2

3

PE

PE+A

-0.0001 -3.6914

-0.0001 -3.8278

-0.0001 -4.4346

0.1043 4.7886

0.1045 4.8043

-0.0177 -13.4290

S

4

6

MAR(180)

MAR(270)

-0.0001 -4.3221

-0.0001 -3.7648

-0.0001 -3.8335

0.1046 4.7684

0.1047 4.7768

0.1052 4.7596

0.1043 4.8067

-0.0177 -13.1599

-0.0179 -13.3245

-0.0178 -13.1905

-0.0177 -13.9222

-0.0178 -13.6480

0.0009 1.2288

0.0009 1.2227

0.0008 0.7891

0.0008 0.7169

0.0005 0.6082

0.0005 0.6390

S-1

0.0016 3.8755

0.0016 3.8819

0.0017 2.3162

0.0017 2.2303

0.0009 2.3732

0.0009 2.3314

S-2

0.0001 0.1228

0.0001 0.2416

-0.0008 -0.6681

-0.0009 -0.7453

-0.0005 -1.2413

-0.0005 -1.2950

R-squared Adjusted R-squared S.E. of regression Number of bonds Number of Observations Number of firms

0.3062 0.2947 0.0007 24 11239 9

0.3327 0.3211 0.0007

0.3039 0.2923 0.0007

0.3059 0.2944 0.0007

0.3066 0.2953 0.0007

0.3082 0.2963 0.0007

C BINDRET BBSW3Y

ABS(PE) ABS(PE+A)

5

Table 12. Maturities of 1 to 3 years: bin 1. This table presents the coefficient estimates and the associated tstatistics (immediately beneath) of all bonds with maturities between 1 to 3 years. Transposition 1 to 4 tests for the information content, 5 and 6 tests for asset deviation; 7 and 8 refers to the estimates of the meanadjusted model. The leftmost column displays the regression equation (7) RBPj,t = β1 + β2 Bindret + β3 BBSW3 + β4 Sj,t + β5 Sj,t-1 + β6 Sj,t-2 + εt , where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate.

4.4.2 Bin 2: 4 to 6 years

Most corporate bond issues in Australia fall within this maturity range. From Table 13, it is clear that the information content hypothesis explains the behavior of returns of bond in this maturity bin. The coefficients of S for [PE] and [PE+A] are statistically significant at 1%, with coefficient values of 0.0038 each. It is apparent that the bonds in this bin react to information content instantaneously, which is more rapid as compared to those in bin 1. Another noteworthy point, is that the coefficients of S or S(-1) for [MAR(180)] and [MAR(270)] are not statistically significant.

32

Transposition 4 to 6 years

1

2

3

PE

PE+A

-0.0001 -7.5763

-0.0001 -7.9628

-0.0001 -5.4364

0.3097 9.0426

0.3123 9.1501

-0.0330 -13.2502

S

4

6

MAR(180)

MAR(270)

-0.0001 -5.6359

-0.0001 -7.8979

-0.0001 -7.9755

0.3135 9.1772

0.3137 9.1763

0.3128 9.1958

0.3070 8.9147

-0.0341 -13.0981

-0.0335 -13.5925

-0.0335 -13.5341

-0.0333 -13.4589

-0.0322 -12.7518

0.0038 4.2343

0.0038 4.2265

-0.0016 -1.1998

-0.0015 -1.1347

0.0004 0.4662

0.0002 0.2504

S-1

0.0019 1.9232

0.0019 1.8596

0.0008 0.8213

0.0011 1.0780

0.0010 1.4540

0.0009 1.2646

S-2

0.0006 0.8574

0.0006 0.8446

-0.0002 -0.1694

-0.0003 -0.2381

0.0006 0.8234

0.0005 0.6714


0.4844 0.4793 52 36927 20

0.4850 0.4801

0.4830 0.4780

0.4845 0.4795

0.4846 0.4797

0.4784 0.4718

C BINDRET BBSW3Y

ABS(PE) ABS(PE+A)

5

Table13. Maturities of 4 to 6 years: bin 2. This table presents the coefficient estimates and the associated tstatistics (immediately beneath) of all bonds in the sample with maturities between 4 to 6 years. Transposition 1 to 4 tests for the information content, 5 and 6 tests for asset deviation; 7 and 8 refers to the estimates of the mean-adjusted model. The leftmost column displays the regression equation (7) RBPj,t = β1 + β2 Bindret + β3 BBSW3 + β4 Sj,t + β5 Sj,t-1 + β6 Sj,t-2 + εt , where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate

4.4.3 Bin 3: 7 years or more

Similar to bins 1 and 2, this bin rejects the asset deviation hypothesis and provides strong evidence to fully support information content. From Table 14, the coefficient estimates of S for [PE] and [PE+A] are statistically significant at 5% level with values of 0.0066 and 0.0066 respectively. More importantly, these coefficients are larger than those in bin 2; suggesting that corporate bonds with longer maturities are much more sensitive to firm specific information.

33

Transposition 7 years and above

1

2

3

PE

PE+A

-0.0001 -1.9758

-0.0001 -2.0077

-0.0001 -0.2290

BINDRET

0.6488 4.0067

0.6488 4.0078

BBSW3Y

-0.0440 -4.1318

S

4

6

MAR(180)

MAR(270)

-0.0001 -0.3211

-0.0001 -2.0198

-0.0001 -2.0067

0.6498 4.0055

0.6493 4.0001

0.6497 4.0083

0.6493 4.0080

-0.0440 -4.1310

-0.0440 -4.0940

-0.0442 -4.1219

-0.0438 -4.1371

-0.0439 -4.1504

0.0066 2.4883

0.0066 2.5088

-0.0062 -1.8221

-0.0054 -1.5209

-0.0006 -0.3456

-0.0006 -0.3519

S-1

0.0015 0.9628

0.0017 1.0530

0.0006 0.136

0.0014 0.3035

0.0021 1.2979

0.0024 1.4369

S-2

-0.0003 -0.1330

-0.0003 -0.1132

-0.0052 -1.0563

-0.0062 -1.2987

-0.0009 -0.4150

-0.0005 -0.2625

0.4797 0.4734 12 6193 7

0.4796 0.4735

0.4793 0.4733

0.4809 0.4749

0.4808 0.4743

0.4807 0.4744

C


ABS(PE) ABS(PE+A)

5

Table 14 Maturities of 7 years or more: bin 3. This table presents the coefficient estimates and the associated t-statistics (immediately beneath) of all bonds in the sample with maturities more than 7 years. Transposition 1 to 2 test for the information content, 3 and 4 test for asset deviation; 5 and 6 refer to the estimates of the mean-adjusted model. The leftmost column displays the regression equation (12) RBPj,t = β1 + β2 Bindret + β3 BBSW3 + β4 Sj,t + β5 Sj,t-1 + β6 Sj,t-2 + εt , where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate.

In short, the results suggest that asset deviation hypothesis is being rejected according to differing maturity structures while information content hypothesis is able to consistently explain the behavior of bond investors. That is, the longer the maturity of a corporate bond, the more sensitive it is to information content. While under the means-adjusted returns models ([MAR(180)] and [MAR(270)] ), the firm specific information does not explain the bond price movements throughout the analysis. 4.5 Leverage 4.5.1 Group 1: decreasing leverage ratio

The statistical output for all firms (excluding banks) with decreasing leverage ratios is reported in Table 15. The only coefficients that are statistically significant at 5% level are 34

S(-2) for [PE+A], MAR(180) and MAR(270). The results suggest that returns of bonds in this category do not respond to information contemporaneously.

Transposition

1

2

4

5

6

PE

PE+A

ABS(PE) ABS(PE+A)

MAR(180)

MAR(270)

-0.0001 -4.2423

-0.0001 -4.2988

-0.0001 -3.0054

-0.0001 -3.0616

-0.0001 -4.2676

-0.0001 -4.4791

BINDRET

0.3069 4.9636

0.3070 4.9671

0.3074 4.9788

0.3073 4.9841

0.3059 4.9874

0.3053 4.9594

BBSW3Y

-0.0310 -6.3113

-0.0310 -6.2952

-0.0311 -6.3230

-0.0311 -6.3025

-0.0310 -6.3160

-0.0301 -5.9578

S

-0.0001 -0.0294

0.0001 0.0099

-0.0022 -1.2196

-0.0023 -1.3040

-0.0020 -1.7995

-0.0021 -1.9026

S-1

0.0001 0.1547

0.0001 0.1552

0.0031 1.4992

0.0032 1.4534

-0.0004 -0.4672

-0.0005 -0.5948

S-2

0.0016 1.9541

0.0016 1.9699

-0.0020 -1.0567

-0.0022 -1.1039

0.0017 2.3620

0.0016 2.2104


0.4633 0.4567 21 12828 9

0.4639 0.4574

0.4616 0.4551

0.4621 0.4555

0.4648 0.4583

0.4591 0.4527

Decreasing leverage ratio C

3

Table 15. Decreasing leverage ratio: group 1. This table presents the coefficient estimates and the associated t-statistics (immediately beneath) of all firms with decreasing leverage ratios from 1997 to 2003. Transposition 1 to 2 test for the information content, 3 and 4 test for asset deviation; 5 and 6 refer to the estimates of the mean-adjusted model. The leftmost column displays the regression equation (7) RBPj,t = β1 + β2 Bindret + β3 BBSW3 + β4 Sj,t + β5 Sj,t-1 + β6 Sj,t-2 + εt , where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate.

4.5.2 Group 2: increasing leverage ratio

This category is presented in Table 16. Both hypotheses cannot be rejected as coefficients of S for [PE], [PE+A], [ABS(PE)], [ABS(PE+A)] are statistically significant at 5% level. Their corresponding values are 0.0051, 0.0051, -0.0096 and -0.0092 respectively. The findings are not surprising as the sample in this category includes financial firms whose bond holders are very responsive to information content as reported in the earlier section. 35

Transposition Increasing leverage ratio

1

2

3

PE

PE+A

-0.0001 -2.4023

-0.0001 -2.5250

0.0001 0.6628

BINDRET

0.5080 5.1205

0.5141 5.2283

BBSW3Y

-0.0327 -5.3555

S

4

6

MAR(180)

MAR(270)

0.0001 0.5527

-0.0001 -2.4953

-0.0001 -2.5328

0.5157 5.2426

0.5159 5.2420

0.5140 5.2338

0.5138 5.2376

-0.0354 -5.5642

-0.0337 -5.6121

-0.0340 -5.6261

-0.0339 -5.6087

-0.0338 -5.6103

0.0051 2.9744

0.0051 2.9465

-0.0096 -3.2670

-0.0092 -3.3003

0.0006 0.3173

0.0004 0.2225

S-1

0.0030 1.2642

0.0030 1.2195

-0.0009 -0.2851

-0.0001 -0.0289

0.0018 1.0525

0.0017 0.9764

S-2

-0.0025 -0.9898

-0.0026 -1.0205

-0.0045 -1.2627

-0.0051 -1.4000

-0.0018 -0.8144

-0.0015 -0.7103

0.5366 0.5306 21 12279 7

0.5379 0.5325

0.5335 0.5282

0.5383 0.5329

0.5374 0.5319

0.5341 0.5285

C


ABS(PE) ABS(PE+A)

5

Table 16 Increasing leverage ratio: group 2. This table presents the coefficient estimates and the associated t-statistics (immediately beneath) of all firms with increasing leverage ratios from 1997 to 2003. Transposition 1 to 2 tests for the information content, 3 and 4 tests for asset deviation; 5 and 6 refers to the estimates of the mean-adjusted model. The leftmost column displays the regression equation (7) RBPj,t = β1 + β2 Bindret + β3 BBSW3 + β4 Sj,t + β5 Sj,t-1 + β6 Sj,t-2 + εt , where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate.

4.5.3 Group 3: increasing leverage ratio excluding all financial firms

There are a total of 5 firms and 14 bonds in this category. Table 17 presents the results that do not include the financial firms. The coefficients of S for [ABS(PE)] and [ABS(PE+A)] are statistically significant at 5% level suggesting that bondholders of non-financial firms are only sensitive to asset volatility.

36

Transposition Excluding all financial firms Increasing leverage ratio C

1

2

3

PE

PE+A

ABS(PE)

4

5

ABS(PE+A) MAR(180)

6 MAR(270)

-0.0001 -1.3133

-0.0001 -1.4110

0.0001 1.1393

0.0001 0.9856

-0.0001 -1.3958

-0.0001 -1.4003

BINDRET

0.4224 4.2571

0.4320 4.4232

0.4327 4.4378

0.4324 4.4350

0.4312 4.4335

0.4315 4.4346

BBSW3Y

-0.0394 -4.9650

-0.0413 -5.3345

-0.0413 -5.4209

-0.0416 -5.4557

-0.0411 -5.3364

-0.0411 -5.3399

S

0.0033 1.5132

0.0033 1.4745

-0.0122 -2.9602

-0.0117 -2.9651

-0.0023 -0.8564

-0.0023 -0.8745

S-1

0.0026 0.7000

0.0024 0.6426

-0.0014 -0.3121

-0.0008 -0.1721

0.0014 0.5548

0.0015 0.5953

S-2

-0.0004 -0.0936

-0.0005 -0.1275

-0.0034 -0.6024

-0.0036 -0.6264

0.0004 0.1378

0.0006 0.1823

0.4287 0.4214 14 9172 5

0.4305 0.4241

0.4294 0.4232

0.4317 0.4253

0.4304 0.4238

0.4304 0.4239


Table17. Increasing leverage ratio excluding banks: group 3. This table presents the coefficient estimates and the associated t-statistics (immediately beneath) of all firms excluding banks and financial services with increasing leverage ratios from 1997 to 2003. Transposition 1 to 2 test for the information content, 3 and 4 test for asset deviation; 5 and 6 refer to the estimates of the mean-adjusted model. The leftmost column displays the regression equation (7) RBPj,t = β1 + β2 Bindret + β3 BBSW3 + β4 Sj,t + β5 Sj,t-1 + β6 Sj,t-2 + εt , where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate.

4.6 Market-to-book value 4.6.1 Bin 1: decreasing market-to-book ratio for full sample

In Table 18, only the coefficients of S for [ABS(PE)] and [ABS(PE+A)] are statistically significant at 5% level. The coefficient values are -0.0066 and -0.0063 respectively indicating that bondholders in this category are sensitive to asset volatility and they may sell off the bonds if they perceive any significant changes in asset values are detrimental.

37

Transposition All firms where change in ratio is negative

1

2

3

PE

PE+A

-0.0001 -1.2596

-0.0001 -1.3853

0.0001 -0.0615

BINDRET

0.2402 3.4782

0.2467 3.5802

BBSW3Y

-0.0281 -4.7268

S

4

6

MAR(180)

MAR(270)

0.0001 -0.1830

-0.0001 -1.3741

-0.0001 -1.4198

0.2475 3.5886

0.2475 3.5922

0.2466 3.5935

0.2459 3.5787

-0.0294 -4.9259

-0.0295 -4.9987

-0.0296 -5.0018

-0.0292 -4.9202

-0.0285 -4.7342

0.0014 1.0009

0.0014 0.9585

-0.0066 -2.2491

-0.0063 -2.2278

-0.0011 -0.6189

-0.0011 -0.6359

S-1

0.0021 0.8691

0.0019 0.7879

-0.0018 -0.8556

-0.0017 -0.8167

0.0016 1.0636

0.0016 1.0529

S-2

-0.0009 -0.3620

-0.0010 -0.4037

0.0025 0.9298

0.0024 0.8415

-0.0001 -0.0621

-0.0002 -0.0736

0.3949 0.3868 21 12828 7

0.4224 0.4149

0.3924 0.3851

0.3946 0.3872

0.3973 0.3899

0.3922 0.3848

C


ABS(PE) ABS(PE+A)

5

Table 18. Decreasing market-to-book ratio for full sample: bin 1. This table presents the coefficient estimates and the associated t-statistics (immediately beneath) of all firms with decreasing market-to-book ratios from 1997 to 2003. Transposition 1 to 2 test for the information content, 3 and 4 test for asset deviation; 5 and 6 refer to the estimates of the mean-adjusted model. The leftmost column displays the regression equation (7) RBPj,t = β1 + β2 Bindret + β3 BBSW3 + β4 Sj,t + β5 Sj,t-1 + β6 Sj,t-2 + εt , where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate.

4.6.2 Bin 2: increasing market-to-book ratio by 0% to 10% for full sample

In general, increasing market-to-book ratios is an indication of good prospects of the firm, especially if this increase is consistent over a period and this phenomenon can be explained by the information content hypothesis. Indeed, regression results in Table 19 provide such evidence. The coefficient values of S for [PE] and [PE+A], are 0.0037 each, indicating that positive abnormal returns of stocks are strongly associated with increase in returns on bond prices when a firm experience consistent and increasing market-to-book values. This informational effect also induces lagged responses. This is shown by the coefficients of S(38

1) for [PE], [PE+A], [MAR(180)] and [MAR(270)], with a value of 0.0019, 0.0019, 0.0015 and 0.0015 respectively, which are also statistically significant at 1%.

Transposition

1

All firms where average change is between 0% to 10%

2

3

PE

PE+A

-0.0001 -6.5615

-0.0001 -6.6309

-0.0001 -3.5100

0.3034 6.9131

0.3035 6.9149

-0.0309 -11.5655

S

4

6

MAR(180)

MAR(270)

-0.0001 -3.5128

-0.0001 -6.4987

-0.0001 -6.6137

0.3047 6.9173

0.3049 6.9166

0.3044 6.9289

0.2994 6.7896

-0.0313 -11.3621

-0.0308 -11.5626

-0.0308 -11.5227

-0.0307 -11.5725

-0.0301 -11.3524

0.0037 4.5250

0.0037 4.5450

0.0004 0.4084

0.0005 0.5856

0.0006 0.8667

0.0004 0.6806

S-1

0.0019 3.8815

0.0019 3.8547

-0.0009 -0.8461

-0.0006 -0.6129

0.0015 3.8173

0.0015 3.4534

S-2

-0.0005 -0.8597

-0.0005 -0.8303

-0.0001 -0.0748

-0.0003 -0.2966

-0.0009 -1.7209

-0.0010 -1.8987

0.4376 0.4306 63 42420 12

0.4382 0.4312

0.4363 0.4292

0.4379 0.4309

0.4377 0.4306

0.4345 0.4259

C BINDRET BBSW3Y


ABS(PE) ABS(PE+A)

5

Table 19. Increasing market-to-book ratio by 0% to 10% for full sample: bin 2. This table presents the coefficient estimates and the associated t-statistics (immediately beneath) of all firms with increasing marketto-book ratios by 0% to 10% from 1997 to 2003. Transposition 1 to 2 test for the information content, 3 and 4 test for asset deviation; 5 and 6 refer to the estimates of the mean-adjusted model. The leftmost column displays the regression equation (7) RBPj,t = β1 + β2 Bindret + β3 BBSW3 + β4 Sj,t + β5 Sj,t-1 + β6 Sj,t-2 + εt , where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate.

4.6.3 Bin 3: financial institutions

While increasing market-to-book values are favorable to investors, an increase in leverage is detrimental. This presents a contradiction to both equity and debt investors. Thus, it will be interesting to understand if bond investors will respond symmetrically to the undertakings of the firm when these ratios commensurate especially in the financial sector.

39

Panel 1 1

2

3

4

5

6

Average change is negative

PE

PE+A

ABS(PE)

ABS(PE+A)

MAR(180)

MAR(270)

C

-0.0001 -2.0412

-0.0001 -2.0169

-0.0001 -2.3476

-0.0001 -2.2280

-0.0001 -1.9943

-0.0001 -2.4430

BINDRET

0.1981 3.2134

0.1982 3.2156

0.1982 3.2609

0.1983 3.2601

0.1977 3.2183

0.1956 3.1477

BBSW3Y

-0.0291 -4.0961

-0.0291 -4.0974

-0.0292 -4.0604

-0.0293 -4.0678

-0.0290 -4.1122

-0.0267 -3.5095

S

-0.0007 -0.4703

-0.0006 -0.4383

0.0013 0.5316

0.0012 0.5058

-0.0010 -0.7025

-0.0011 -0.8223

S-1

0.0023 2.9407

0.0023 2.9081

0.0006 0.5567

0.0003 0.2487

0.0018 2.6284

0.0015 2.6160

S-2

0.0015 1.9267

0.0015 1.9150

0.0005 0.3982

0.0004 0.3323

0.0011 1.4980

0.0008 1.2184

R-squared

0.4075

0.4848

0.4052

0.4056

0.4090

0.3939

Adjusted R-squared

0.4009

0.4780

0.3985

0.3995

0.4026

0.3877

Only financial insitutions

Number of bonds

7

Number of Observations

4073

Number of firms

2 Panel 2

Average change is between 0% and 10% C

1

2

3

4

5

6

PE

PE+A

ABS(PE)

ABS(PE+A)

MAR(180)

MAR(270)

-0.0001 -5.3072

-0.0001 -5.3776

-0.0001 -6.1576

-0.0001 -6.1392

-0.0001 -5.2202

-0.0001 -5.3452

BINDRET

0.2553 4.2863

0.2553 4.2885

0.2562 4.2906

0.2563 4.2900

0.2568 4.3016

0.2491 4.1646

BBSW3Y

-0.0270 -7.8869

-0.0269 -7.8534

-0.0270 -7.8673

-0.0269 -7.8307

-0.0268 -7.8886

-0.0259 -7.6885

S

0.0040 4.4691

0.0040 4.5017

0.0004 0.3654

0.0007 0.5621

0.0015 2.5081

0.0014 2.3672

S-1

0.0028 4.9855

0.0029 4.8949

0.0008 0.8020

0.0010 1.1566

0.0022 4.8168

0.0022 4.7801

S-2

-0.0007 -1.0071

-0.0006 -0.9047

0.0009 0.7221

0.0006 0.5157

-0.0012 -1.9907

-0.0012 -2.2884

R-squared

0.3905

0.3915

0.3896

0.3906

0.3908

0.3875

Adjusted R-squared

0.3832

0.3841

0.3822

0.3831

0.3833

0.3777

Number of bonds Number of Observations Number of firms

42 28661 7

Table 20 Financial institutions with changes in market-to-book ratio: bin 4. This table presents the coefficient estimates and the associated t-statistics (immediately beneath) of the firms excluding non-financial institutions, grouped into negative market-to-book ratio changes and positive ratio changes. Transposition 1 to 2 test for the information content, 3 and 4 test for asset deviation; 5 and 6 refer to the estimates of the mean-adjusted model. The leftmost column displays the regression equation (7) RBPj,t = β1 + β2 Bindret + β3 BBSW3 + β4 Sj,t + β5 Sj,t-1 + β6 Sj,t-2 + εt , where Bindret refers to the bond return index (ABI) and BBSW3 denotes bank-bill swap rate.

40

The statistical tests presented in Table 20 provide substantial evidence to the conjecture that rising market-to-book ratios induces higher bond price sensitivity. In Panel 1, only the coefficients of S(-1) for [PE], [PE+A], [MAR(180)] and [MAR(270)] are statistically significant indicating that bondholders in this category are slow to respond to firm specific information. In comparison to the results in panel 2, the coefficients of S for [PE], [PE+A], [MAR(180)] and [MAR(270)] are statistically significant at 1%. Therefore, it can be concluded that increasing market-to-book ratios induces a stronger and more immediate response from bond holders to information content

5.0

Conclusion

Ascertaining the difference between two types of information, as perceived by bond holders, is the most important finding of this study. These two information types are namely, information content and asset volatility. Our study finds that information content induces a symmetric reaction from bond holders and equity holders. On the other hand, asset volatility posits that bond holders regard all information as negative information. The distinction between both is that the former predicts the price behavior of bond is similar to stock and the latter regards all information as asset volatility according to Merton (1974). To further explain the validity of these two hypotheses, tests are conducted according to various bond and firm characteristics. The empirical results strongly suggest that bond holders indeed respond in two manners, with regards to the specific firm and bond characteristics. Furthermore, in comparison to the mean-adjusted model, our approach provides more logical inferences in all of the analysis presented. Finally, the test results obtained from [PE+A] is consistently indifferent from those of PE. This confirms that αˆ j (which is an estimator of A) does not have an impact on the inferences drawn from εˆ j (the estimator of prediction error (PE)).

41

Appendix A The table presents the descriptive statistics of selected corporate bonds from each firm in the sample. These statistics are calculated based on the logarithmic returns of the daily traded (mid) price obtained from Datastream. Mean and Median of each bond is approximately 0. Skewness is generally negative with the exception of several bonds and the distribution is leptokurtic. Panel 1

Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis Jarque-Bera Start / End of Sample Date Number of Observations

AGL 0.000066 0.000000 0.008758 -0.009327 0.002725 -0.238503 3.512069

AMP 0.000000 0.000001 0.004008 -0.029820 0.001452 -15.846260 327.250500

ANZ 0.000109 0.000000 0.015607 -0.017090 0.004747 -0.200675 3.463052

BHP 0.000043 0.000000 0.013275 -0.012499 0.003425 -0.196618 3.651324

Boral 0.000000 -0.000029 0.017248 -0.048781 0.002348 -8.922883 203.506400

Caltex 0.000017 0.000024 0.002513 -0.001778 0.000401 0.212752 9.657156

CBA -0.000016 0.000000 0.005272 -0.010908 0.001383 -0.969874 10.873730

Coke -0.000169 -0.000096 0.007841 -0.008229 0.000872 0.089935 40.149040

10

2419171

8

15

1691764

630

1707

22484

5/16/2002 3/31/2004

10/13/1999 11/15/2001

3/21/2002 3/31/2004

11/22/2001 3/31/2004

10/14/1999 8/15/2003

10/13/1999 1/30/2001

11/12/2001 3/31/2004

7/01/1998 12/29/1999

490

547

530

615

1002

340

623

391

NAB -0.000001 0.000000 0.009867 -0.010909 0.001539 -0.071089 11.652440

Origin 0.000072 0.000000 0.008494 -0.009002 0.002535 -0.184675 3.705419

PUB -0.000041 -0.000029 0.011016 -0.006483 0.001725 0.436870 8.472146

Qantas 0.000038 0.000000 0.011302 -0.010367 0.003013 -0.206517 3.883069

SouthCorp -0.000045 0.000000 0.060346 -0.088031 0.006510 -3.022262 69.624550

Panel 2

Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis Jarque-Bera Start / End of Sample Date Number of Observations

Coles 0.000010 0.000000 0.029749 -0.063931 0.003555 -4.533706 105.948300

JFF 0.000027 -0.000010 0.031586 -0.007561 0.002331 3.411690 49.428060

MACQ -0.000037 0.000000 0.004931 -0.005715 0.001200 -0.303060 6.260965

518453

67807

283

2303

13

958

24

137804

10/14/1999 3/31/2004

6/1/2001 3/31/2004

11/19/2001 3/31/2004

6/4/2001 3/31/2004

4/24/2002 3/31/2004

5/18/2001 3/31/2004

11/19/2001 3/31/2004

6/1/2001 3/31/2004

1165

739

618

738

506

749

618

739

Stockland 0.000009 0.000000 0.007558 -0.008246 0.002106 -0.178612 4.096137

Suncorp 0.000001 0.000000 0.010369 -0.006737 0.001337 0.319194 11.722720

Telstra 0.000071 0.000000 0.065737 -0.057110 0.005365 0.770497 49.555160

WestField 0.000004 0.000000 0.013473 -0.019988 0.002476 -0.603910 10.178510

WestPac 0.000012 0.000000 0.006559 -0.007744 0.002217 -0.295132 3.709001

Woolies 0.000036 0.000000 0.026188 -0.031379 0.003057 -0.896472 30.967850

Panel 3 Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis Jarque-Bera Start / End of Sample Date Number of Observations

34

1970

66810

1629

16

18523

11/21/2001 3/31/2004

11/19/2001 3/31/2004

6/01/2001 3/31/2004

6/4/2001 3/31/2004

7/23/2002 3/31/2004

1/30/2002 3/31/2004

616

618

739

738

442

566

42

Appendix B The table below and on the following page displays the descriptive statistics of model estimates for 6 randomly selected firms from the sample. These estimates will be subjected to empirical tests to be discussed in a later section. Mean of all estimates are approximately zero. However, a closer examination reveals that PE, PE(2), PE(3) tend to be closer to zero as compared to PE+A, MAR(180) and MAR(270). ABS(PE) and ABS(PE+A) are absolutes of PE and PE+A, hence they are positively skewed with abnormally large descriptive statistics. Panel 1

CBA Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis Jarque-Bera Start / End of Sample Date Number of Observations

PE -0.000035 0.000004 0.040620 -0.070702 0.009653 -0.494714 6.525414

PE+A 0.000288 0.000226 0.039919 -0.070930 0.009623 -0.513661 6.593143

PE(2) -0.000071 0.000329 0.066456 -0.067670 0.014608 -0.257478 4.575136

PE(3) -0.000107 0.000553 0.089716 -0.075023 0.018236 -0.197646 4.283586

ABS(PE) 0.007967 0.006404 0.070702 0.000005 0.006820 1.836449 10.137280

ABS(PE+A) 0.007055 0.005512 0.070930 0.000000 0.006550 2.229960 12.540400

MAR(180) 0.000236 -0.000118 0.074829 -0.071806 0.011811 -0.211865 5.426443

MAR(270) 0.000249 -0.000164 0.073728 -0.071916 0.011803 -0.230391 5.397495

1574

1640

322

212

3909

13023

712

700

PE 0.000002 0.000034 0.038139 -0.043294 0.009017 -0.284571 4.908715

PE+A 0.000076 0.000000 0.038557 -0.042524 0.009006 -0.303940 4.930794

PE(2) 0.000004 0.000419 0.065580 -0.064171 0.012554 -0.276785 4.745827

PE(3) 0.000007 0.000477 0.074066 -0.079715 0.015264 -0.253996 4.736235

ABS(PE) 0.006622 0.004987 0.043294 0.000000 0.006119 1.730617 7.249320

ABS(PE+A) 0.006614 0.004975 0.042524 0.000000 0.006112 1.732279 7.298681

MAR(180) -0.000051 -0.000154 0.038301 -0.044855 0.009726 -0.211579 4.583978

MAR(270) -0.000038 -0.000202 0.038484 -0.044785 0.009716 -0.220691 4.592524

466

481

394

384

3527

3579

316

321

6/14/1997 3/31/2004 2070 Panel 2

Stockland Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis Jarque-Bera Start / End of Sample Date Number of Observations

6/14/1997 3/31/2004 2070 Panel 3

Coles Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis Jarque-Bera Start / End of Sample Date Number of Observations

PE 0.000025 0.000210 0.136351 -0.179084 0.013237 -1.589231 31.251940

PE+A -0.000018 0.000000 0.136123 -0.178358 0.013216 -1.594864 31.234570

PE(2) 0.000051 0.000307 0.154979 -0.219103 0.019173 -1.063579 18.036000

PE(3) 0.000073 0.000403 0.160989 -0.194975 0.023521 -0.748549 11.933890

ABS(PE) 0.008781 0.006310 0.179084 0.000008 0.009904 5.763321 73.900310

ABS(PE+A) 0.008755 0.006254 0.178358 0.000000 0.009899 5.749760 73.610880

MAR(180) -0.000083 -0.000058 0.128246 -0.181850 0.014347 -1.327812 23.405040

MAR(270) -0.000069 -0.000043 0.128212 -0.181923 0.014337 -1.340755 23.459780

94905

94798

27067

9628

605837

600954

49716

49995

6/14/1997 3/31/2004 2070

43

References

Ball, R. and Brown, P. (1968), “An empirical evaluation of accounting income numbers”, Journal of Accounting Research, Vol. 6, No. 2, pp. 159-178. Campbell, John Y., Glen B., Taksler, 2003, Equity volatility and corporate bond yields, Journal of Finance, Vol 58, No. 6, 2321-2349. Collin-Dufresne, P., Goldstein, R. and Martin, S. (2001), “The determinants of credit spread changes”, Journal of Finance, Vol. 56, No. 6, pp. 2177-2208. Dhillon, S. U. and Johnson, H. (1994), “The effect of dividend changes on stock prices”, Journal of Finance, 49, No. 1, pp. 281-289. Elton, E. (1999), “Exepcted return, realized return, and asset pricing tests”, Journal of Finance, Vol. 54, No. 4, pp. 1199-1220.

Eom, Y. H., Helwege, J. and Huang J. Z. (2004), “Structural models of corporate bond pricing: An empirical analysis” Review of Financial Studies, Vol. 17, No. 2, pp. 499-544. Fama, E. F., Fisher, L., Jensen, M. C. and Roll, R. (1969), “The adjustment of stock price to new information”, International Economics Review, Vol.10, No. 1, pp. 1-21.

Longstaff, F. and Schwartz E. (1995) “Valuing Risky debt: A new approach”, Journal of Finance, Vol. 50, No. 3, pp. 789-820. Grier, P. and Katz, S. (1976), “The differential effects of bond rating changes among industrial and public utility bonds by maturity”, Journal of Business, Vol. 49, pp. 226-239. Handjinicolaou, G. and Kalay, A. (1984), “Wealth redistributions or changes in firm value: An analysis of returns to bondholders and stockholders around dividend announcements”, Journal of Financial Economics, Vol. 13, No. 1, pp. 35-63.

44

2 Coupon

Hand, J. R. M., Holthausen, R. W. and Leftwich, R. W. (1992), “The effect of bond rating agency announcements on bond and stock prices”, Journal of Finance, Vol. 47, No. 2, pp. 733-752. Hettenhouse G. W. and Sartoris, W. L. (1976), “An analysis of the informational value of bond-rating changes”, Quarterly Review of Economics and Business, Vol. 16, pp. 65-78. Jones, E. P., Mason, S. and Rosenfeld, E. (1984), “The contingent claims analysis of corporate capital structures: An empirical investigation”, Journal of Finance, Vol. 39, No. 3, pp.611-625. Katz, S. (1974), “The price adjustment process of bonds to rating reclassifications: A test of bond market efficiency”, Journal of Finance, Vol. 29, No. 2, pp. 551-559. Kwan, S. H. (1996) “Firm-specific information and the correlation between individual stocks and bonds”, Journal of Financial Economics, Vol. 40, No. 1, pp.63-80. Merton, R. C. (1974) “On the pricing of corporate debt: The risk structure of interest rates”, Journal of Finance, Vol. 29, No. 2, 449-470. Weinstein, M. I. (1978), “The effect of a rating change announcement on bond price”, Journal of Financial Economics, Vol. 5, No. 3, pp. 329-350.

45