Determinants of Corporate Bond Returns in Korea: Characteristics or Betas? *

Transcription

1 Asia-Pacific Journal of Financial Studies (2009) v38 n3 pp Determinants of Corporate Bond Returns in Korea: Characteristics or Betas? * Woosun Hong KIS Pricing, INC., Seoul, Korea Seong-Hyo Lee ** MyongJi University, Seoul, Korea Young S. Park Sogang University, Seoul, Korea Received 25 August 2008; Accepted 13 October 2008 Abstract This study investigates how the corporate bond s characteristics and Betas affect bond returns by using extensive Korean corporate bonds data from 2001 to the first half of Overall, our results indicate that bond characteristics provide significant explanations to excess returns while market factors (i.e., Betas) do not. It is strikingly different from the U.S. study of Gebhardt et al. (2005), which showed that market factors notably affect the excess returns of U.S. corporate bonds. 1) Keywords: Bond investment; Bond characteristics; Bond returns; Credit rating; Duration JEL Classification: G11, G12, G24 * The data was obtained from KIS Pricing, Inc. And we thank to Sungjin Lee of KIS Pricing, Inc. for his help in data processing. ** Corresponding Author, Address: MyongJi University, Department of Business Administration, 50-3 Namgajwa-Dong, Seodaemun-Gu, Seoul, Korea ; hyo@mju.ac.kr; Tel: ; Fax:

2 Determinants of Corporate Bond Returns in Korea 1. Introduction Bonds are as important financial products as stocks in the Korean financial market. As of the end of 2006, the total outstanding amount of bonds in Korea recorded 813 trillion won, approximately 5 percent higher than the total stock market value including that of KOSDAQ. In spite of their significant volume, bonds have been regarded as buy-and-hold savings tools rather than as investment instruments for active trading. Until early 2000, bonds held by financial institutions were not priced at a market value, but at a book value. Therefore, the profitability of the bond investment had been easily pre-determined. After the Asian Crisis in the late 1990s, in an attempt to stabilize the financial system, the government started to promote the development of fixed-income markets since With the introduction of mark-to-market pricing of fixed-income securities held by financial institutions, bonds started to be recognized as an investment tool, instead of a simple savings tool. Furthermore, corporate bonds began to be issued as uninsured debentures and, thus, corporate bonds have become more popular as they are exposed to default risks with corresponding high returns. Accordingly, recent analysis on bond investment has focused mainly on corporate bonds. However, most of the existing studies on corporate bond profitability use the spreads relative to specific bonds such as those of risk-free bonds while studies based on bond returns have been rare. Up until 2001, return data had been difficult to be obtained compared to the spread data; the accumulation of such data in Korea had not been available before then. However, investment returns based on investment period consideration is a more important factor in making bond investment decisions than the spread. Furthermore, existing studies using bond spreads have been limited to investigation on how bond characteristics affect the spreads or interest rate decisions. In this study, however, we investigate the need for the consideration of Beta, a systematic risk indicator, in bond analysis just as in stock analysis. While some studies on the U.S. bond market have reflected Beta, Korean studies have not. In this study we investigate how bond characteristics and Beta (factor loadings) impact the cross-sectional investment performance of corporate bonds. We perform an empirical analysis based on Korean corporate bond data from 2001, when mark-to-market valuation system was first introduced, to first half of 2007 to test which of those factors actually contribute to excess bond returns. 418

3 Asia-Pacific Journal of Financial Studies (2009) v38 n3 2. Literature Review The most representative study on the expected return determinant factor model on risky assets such as stocks and bonds is the Capital Asset Pricing Model (CAPM) developled by Sharpe (1964), Lintner (1966) and Mossin (1966), which is based on the portfolio theory pioneered by Markowits (1952). This model is a one-factor (market Beta) model explaining how a risk asset price is determined, in which the expected return of a risky asset varies positively and linearly with the asset s market Beta. That is, only the market Beta, a systematic variable, can explain the expected return. However, in numerous empirical studies, the Beta fails to explain the cross-sectional returns of risky assets. In the case of stocks, the effect of characteristics, such as size and book-to-market ratio, are found to explain more of excess returns (Fama and French (1993)). Thereafter, much debate has continued on which, between the characteristic variable and the market Beta, is the determining factor to cross-sectional return of risky assets. On the other hand, Merton (1973) used a multi factor model called Inter-temporal CAPM (ICAPM), in which he claims that in the case of a long investment period the relationship between the bond return and beta in CAPM may not hold. He asserts that other factors affect the returns such as labor income, changes in major raw material prices, and future investment opportunities. Debates over the significance of the issuing firm specific characteristics and betas have been continued in determining the cross-sectional return until now. The studies are more active in stocks than in bonds. Fama and French (1993) and Daniel and Titman (1997) analyze that characteristics, such as the market size and book-to-market ratio, contribute more in excess returns than the beta, whereas Davis et al. (2000) argue that Beta explains more than firm-specific characteristics. 1) Also, Fama and French (1993) argue that characteristics, such as bond maturities and default rates, are stronger determinants of bond returns than Beta itself, whereas Gebhardt et 1) Davis et al. (2000) used the same methodology as of Daniel and Titman (1997) on stocks and derived an opposing result based on a longer-term analysis. They used the market size, book-to-market ratio, and market risk premium (Rm-Rf) as independent variables and the stock return rate as the dependent variable in their time series regression analysis using monthly trading data from 1929 to 1997 (stock return rates, size and book-to-market ratio). 419

4 Determinants of Corporate Bond Returns in Korea al. (2005) show that Beta is a stronger determinant than characteristics. 2) Numerous studies in Korea, including Kook and Han (1999), argue Beta (CAPM) as having a lower effect. Although Kim (2004) and Ahn (1999) recognize some effect of Beta, few studies compare the effectiveness of bond characteristics versus Beta as determinants of corporate bond excess returns. Most Korean studies that compare firm-specific characteristics against Beta have been focused on stocks only. Kim and Koh (2002), for instance, analyze that the return curve, estimated from a contract model, cannot explain the market price movement of individual bonds and stated that characteristics, such as fixed asset ratio, debt ratio, and volatility of stock return, are more effective determinants of bond spread than the return curve. However, some studies, such as Kim and Yang (2004), report that the returns of corporate bonds react sensitively to the government s short-term monetary policies. Most studies on corporate bonds have used the spread as the dependent variable when analyzing the impact of credit rating, duration, and liquidity of corporate bonds excess returns. That is, the analysis uses the spread between corporate bonds and risk-free government bonds. According to Chakravarty and Sarkar (1993), the spread has a positive relationship with remaining maturity and holding period while showing a negative relationship with credit rating. On the other hand, Fama and French (1993) and Gebhardt et al. (2005) use the return as the dependent variable to analyze the return of corporate bonds in terms of monthly excess returns (long-term corporate bond return-short-term government bond return, TERM+DEF). Most Korean studies on corporate bond returns also use the spread as the dependent variable. Byun and Jang (2004) use corporate characteristics such as credit ratings to analyze the spread of corporate bonds. However, few studies use the return as the dependent variable for analysis of corporate bond profitability. Uniquely, Koh (2006) uses weekly returns for the analysis of the Korean bond market momentum. Furthermore, no Korean studies use returns while most studies on corporate bond returns use spreads. In fact, most studies using spreads focus on the credit spreads related to credit rating changes. However, the return is inherently the most important 2) Gebhardt et al. (2005) analyzed that beta (factor loading) is a stronger determinant of cross-sectional return rate than characteristics. In particular, default beta has been found to have a strong effect even when characteristics (Duration and credit rating) are controlled. Their study analyzed the monthly data on straight bonds and investment grade bonds from 1973 to 1996 and argued that the systematic risks the determinant of corporate bond returns, and return-to-maturity (YTM), one of the characteristics that impacts their returns even when Betas are controlled. 420

5 Asia-Pacific Journal of Financial Studies (2009) v38 n3 consideration in making bond investment decisions. Accordingly, it is necessary to analyze market movements (i.e., movement of term spread) and changes in individual bond return following respective changes in the spreads and the credit ratings, but such research has not been carried out in Korea. Existing studies on spreads mostly focus on how bond characteristics affect spreads against government bonds or how they determine return rates while insufficient attention has been payed to the market movements (Beta). Studies on Betas as a determinant of bond returns did not begin until 2005 by Gebhardt et al. (2005). Therefore, our study can be regarded as the first attempt in Korea that considers both bond characteristics and Betas for corporate bond returns. 3. Methodology 3.1 Components of Corporate Bond Return In general, corporate bonds holding period return can be defined as Eq. (1): (1) in which : holding period return over time 0 to t;,, if coupon payment exists for analysis period, else;, : Dirty price at time 0 and time t; Dirty price = Clean price + Accrued interest;, : Clean price at time 0 and time t;, : Accrued interest at time 0 and time t; and : Coupon received at time t. Therefore, the corporate bond holding period return is determined by the change in its clean price and the accrued interest (including coupon payments). Here, the former is called capital income and the latter, interest income. Interest income is 421

6 Determinants of Corporate Bond Returns in Korea determined by the coupon rate and the holding period, 3) where the coupon rate is determined by the market valuation of the issuing firm s credit rating at the time of issuance, while the capital income is determined by changes in the bond s return (return-to-maturity: YTM) during a holding period and the bond maturity. Thus, a corporate bond s holding period return is determined by the coupon rate at the time of its issuance, interest rate factors such as changes in YTM during the holding period, and characteristics of individual bond such as duration. Here, since the coupon rate and YTM are affected by the issuing firm s credit rating, a corporate bond s holding period return is determined by the issuing firm s credit level and the bond s characteristics such as the duration. Considering that return is, in general, higher to compensate for higher risk levels, a bond s return should be higher also if an inherent risk is higher in terms of the issuing firm s credit level or duration. In short, a bond s return should be increased for a lower credit rating with a longer duration. On the other hand, when we measure a bond s investment risk in terms of its sensitivity to changes in the market rate, more sensitive bonds (higher beta) should be compensated by higher returns. According to the argument of Gebhardt et al. (2005), a corporate bond s holding period return is determined not by the bond s characteristics but by its sensitivity to changes in the market rate. This means that when Beta is considered as one of indicators of exposure to risks, the higher the Beta is, the higher the holding period return gets. In this study, therefore, we design two analytical models for empirical analysis of corporate bond holding period return: one for bond characteristics and the other for sensitivity to changes in the market rate. 3.2 Corporate Bond Return Model: Bond Characteristics If we express the corporate bond holding period return of Eq. (1) using the effective duration formula, we obtain Eq. (2), which shows the holding period return of a corporate bond determined by the holding period duration ( ) and YTM change ( ). Also, we note the changes in YTM of the bond consisting of the changes in risk-free bonds which reflect the entire bond market movement with changes in the spread 3) Since a holding period is determined by the analysis period, it is not a factor of bond profitability. Therefore, the only factor of a bond s interest income is its coupon rate. 422

7 Asia-Pacific Journal of Financial Studies (2009) v38 n3 of the corporate bond and changes in the term spread of risk-free bond (Spread between long-term and short-term risk-free bond): ㆍ (2) in which : Effective Duration of the bond at time 0; ㆍ ; and : Changes in YTM of the bond over period from time 0 to time t. In this study, in order to control the factors of rate changes in risk-free bonds which affect all type of bonds, we define the holding period return of a corporate bond as an excess return over the holding period return of risk-free bonds. As in Gebhardt et al. (2005), we analyze the corporate bond excess return, which is the spread between short-term risk-free bond and the corporate holding period return in order to identify the bond characteristics, as in Eq. (3): (3) in which : Excess returns of a corporate bond from time 0 to time t; : Holding period return of a corporate bond during time 0 to time t; : Holding period return of short-term risk-free bond during time 0 to time t. By substituting Eq. (2) into Eq. (3) and rearranging, we obtain Eq. (4) as the corporate bond excess return: ㆍ ㆍ (4) ㆍ ㆍ ㆍ ㆍ ㆍ ㆍ ㆍ ; in which, : Durations of corporate bond and risk-free bond at time 0, respectively; 423

8 Determinants of Corporate Bond Returns in Korea, : Changes in YTM of corporate bond and risk-free bond during time 0 to t, respectively; and : Change in Spread of corporate bond against risk-free bond during time 0 to t. As can be seen from Eq. (4), the excess returns of corporate bonds are determined by the changes in YTM of short-term risk-free bonds and the corporate bond as well as the Duration, where the difference between changes in YTM of short-term risk-free bond and a corporate bond is expressed in terms of changes in spread, which reflects the market valuation of credit level of the corporate bond. Accordingly, we choose credit rating and the duration as two key characteristics that affect bond returns. 3.3 Analytical Model for Corporate Bond Return: Beta Factors Fama and French (1993) model the corporate bond excess returns in terms of the market s systematic factors; pre-ranking DEF Beta and pre-ranking TERM Beta. Gebhardt et al. (2005) apply the same methodology to analyze corporate bonds holding period return. According to their definition, DEF Beta and TERM Beta can be estimated by using the regression equation of Eq. (5) below. Here, DEF Beta refers to the difference in holding period return between long-term corporate bonds and long-term risk-free bonds, while TERM Beta refers to the difference in holding period return between long-term risk-free bonds and short-term risk-free bonds: ㆍ ㆍ (5) in which : Excess return time series data prior to time t of bond i;, : DEF Factor and TERM Factor prior to time t; and, : DEF Beta and TERM Beta of bond i applicable to time t. Here, DEF Factor and TERM Factor are derived from holding period returns of a representative bond group, not of individual bonds, as defined in Eq. (6):, (6) in which, : DEF Factor and TERM Factor at time t, respectively; 424

9 Asia-Pacific Journal of Financial Studies (2009) v38 n3 : Corporate bond s holding period return; and, : Holding period returns of risk-free long-term bonds and short-term bonds, respectively. In this respect, DEF Beta and TERM Beta represent the sensitivity of a bond s DEF Factor and TERM Factor, respectively, in which the sensitivity values differ for each individual corporate bond. Fama and French (1993) and Gebhardt et al. (2005) use DEF Beta ( ) and TERM Beta ( ), which are estimated to use data up to time (t-1), as indicators to explain the excess return of a specific bond at the time of t. The following correlation can be derived between the assumed Beta and the excess return of a bond: The higher the Beta of a bond, the greater its sensitivity to DEF factor and TERM factor, which are market factors, and the greater exposure to market movements Therefore, from the viewpoint of risk-return trade off, the higher the Beta of a bond is, the higher its excess return should be: (7) ㆍ ㆍ ㆍ ; in which : DEF Factor at time t;, : Market average holding period return of corporate bonds and risk-free long-term bonds, respectively, at time t;,, : Durations of corporate bonds, risk-free long-term bonds and short-term bonds, respectively, at time 0;, : Changes in YTM of corporate bonds and risk-free long-term bond during time 0 to t; and : Changes in credit spread during time 0 to t. Considering the significance of the market factors such as DEF and TERM factors using our model of Eq. (2)-Eq. (4), we observe the followings: DEF Factor definition 425

10 Determinants of Corporate Bond Returns in Korea of Eq. (6) can be rearranged from the viewpoints of Eq. (2), Eq. (3), and Eq. (4), thereby Eq. (7) can be obtained. According to Eq. (7), DEF Factor is determined by changes in credit spread ( ). Here, the spread is not spread of each bond, as in Fama and French (1993) and Gebhardt et al. (2005), but refers to the representative or average spread of the market. On the other hand, TERM Factor is determined as in Eq. (8) by the term premium ( ), which is the spread between risk-free long-term bonds and short-term bonds 4) : (8) ㆍ ㆍ ㆍ ㆍ ㆍ ㆍ ; in which, : Market average holding period return of risk-free long-term bonds and short-term bonds, respectively, at time t;, : Durations of risk-free long-term bonds and short-term bonds at time 0, respectively;, : Changes in YTM of risk-free long-term bonds and short-term bonds at time t, respectively; and : Changes in Term premium between risk-free long and short-term bonds. Thus, we see that DEF Factor and TERM Factor are determined by credit spread and term premium, which are indicators of the market rate. Accordingly, DEF Beta and TERM Beta are risk measures, which represent sensitivity to credit spread and term premium movements. By looking at the YTM structure of corporate bonds in Figure 1, we can clearly understand the significance of DEF factor and TERM factor. Here, we note the YTM, 4) Gebhardt et al. (2005) use average or representative values for risk-free bonds. Therefore, the Duration difference between short-term bonds and long-term bonds is close to a constant. Therefore, the duration gap in TERM factor is meaningless as a determinant variable. In particular, since three-month and three-year hypothetical bonds are used as short-term and long-term bonds, respectively, the Duration gap is constant at all analysis time points. 426

11 Asia-Pacific Journal of Financial Studies (2009) v38 n3 which determines a corporate bond s holding period return, is a composite of credit spread (DEF) and term premium (TERM). Figure. 1 Structure of corporate bond s return return Corporate bond return curve <DEF> Risk-free bond return curve <TERM> Maturity 4. Test Hypotheses From our discussion above we posit that a corporate bond s excess return is determined by the characteristics of the bond (credit rating and Duration) and by Betas (DEF Beta and TERM Beta). In this study we test the following four hypotheses to identify which factors determine the excess return of corporate bonds in the Korean bond market. In the analytical model above, we observe that a corporate bond s holding period return consists of interest income and capital income, which are determined by the Duration and spread. (Refer to Eq. (4)). Here, the spread is affected by credit rating. Changes in credit ratings affect the spread. Even in the absence of changes in credit rating, any change in market valuation on a corporate bond s credit quality will also result in discrimination of the spread within a credit rating grade. Therefore, we have shown a theoretical model that a corporate bond s excess return is determined by the bond s characteristics, in particular, the duration and credit rating. According to this model, as credit grade is lower and the duration is longer, the higher excess return will be received from the corporate bond. As such, we establish the following Hypothesis 1 to empirically analyze the prediction of our theoretical model using the excess return data of corporate bonds. Hypothesis 1 is tested through 427

12 Determinants of Corporate Bond Returns in Korea the regression analysis and a univariate portfolio approach on the bond s excess return in terms of the bond characteristics. Hypothesis 1: Excess return has a negative correlation with credit rating grade of a bond and a positive correlation with the duration. Since corporate bonds are linked to YTM, just as straight bonds are, an excess return over holding period is also determined by its sensitivity (risk) to the average market rate movements. Here, the market rate movement is a composite of credit spread movement and term premium movement. Based on this logic, it will be possible to explain the excess return of corporate bonds using Betas (DEF Beta and Term Beta), which are indicators of a bond s sensitivity to market rate. From the analytical model above, we estimate that, the higher a bond s DEF and TERM Betas are, which are sensitivity measures for credit spread and term premium, respectively, the higher its excess return will be. We define Hypothesis 2 as follows for the purpose of empirical analysis of Betas significance using the Korean bond market data of corporate bond excess returns. And we will test this hypothesis using regression and univariate portfolio approach on the Betas as determinants of the excess return. Hypothesis 2: DEF Beta and TERM Beta are positively correlated with the excess returns. Although corporate bonds excess returns can be affected by characteristics, such as the duration and credit rating, as well as by Betas, trading and risk management processes employed by Korean asset management companies (AMCs hereinafter) focus mainly on credit rating and Duration. On the other hand, in the U.S., Betas have explanatory power on bonds excess returns while bond characteristics do not, unlike in Korea (Gebhardt et al. (2005)). Characteristics and Betas can affect the bond return through mutual interaction. Therefore, it will be meaningful to analyze the impact of either characteristics or Betas on the return while controlling the other variable. In our study, we perform an empirical analysis to identify which variable, between bond characteristics or Betas, maintains the explanatory power as determinants of the corporate bonds excess return when the other effect is under control. Our analysis will verify whether the U.S. findings of the higher explanatory power 428

13 Asia-Pacific Journal of Financial Studies (2009) v38 n3 of Betas over bond characteristics applies to the Korean bond market also. At the same time, we wish to examine the significance of the current trading and risk-management processes of the Korean AMCs based on this analysis. For the purpose of performing such analysis, we define Hypothesis 3 and Hypothesis 4 as follows, and employ regression and multi-variate portfolio approach, which consider both bond characteristics and Betas at the same time, to test these hypotheses. Hypothesis 3: Beta has a significant impact on the corporate bonds excess return even when bond characteristics are under control; DEF Beta and TERM Beta are positively correlated with excess returns. Hypothesis 4: Bond characteristics have a significant impact on excess returns even when Beta is under control; credit rating is negatively correlated and the duration is positively correlated with excess returns. 5. Sample Data 5.1 Design of Sample Data For our analysis of corporate bonds excess returns, we use the samples of corporate bonds issued by listed firms, whose credit ratings and non-guarantee financial bonds issued by the non-banking institutions including credit card and capital companies (See Table 1). Bonds issued by non-banking institutions have the same attributes as corporate bonds in the market and thus present no difficulty in categorizing them together with corporate bonds. On the other hand, bank bonds are regarded as free of default risks; thus, most of them have credit ratings of AA or higher. They are excluded because their inclusion may result in bias in bond credit rating. Also, in the case of ABS bonds, although they have the same rating as corporate bonds, their returns are different from the straight corporate bonds in the same credit rating category. For this reason, ABS bonds are excluded. That is, as seen in the recent sub-prime mortgage crisis, ABS bonds are traded with a certain level of spread above that of straight corporate bonds in the same credit rating. Most ABS bonds of AAA rating, although their treatment varies by the market perception on the current economic and credit situation, were traded at prices of 5-15 bp added to the spread 429

14 Determinants of Corporate Bond Returns in Korea of straight corporate bonds during the analysis period of our study. Table 1. Corporate bonds for analysis Category Bond Group credit rating Guarantee Y/N Option Y/N Time of Issue Subject Bonds Corporate bonds, credit card bonds, capital bonds credit rating BBB- or higher Publicly sold non-guarantee bonds Straight bonds with no option (Convertible bonds, option bonds and floating rate notes are excluded.) Bonds which passed 52 weeks from the time of issue All bonds included in our sample have credit ratings of BBB or higher. Bonds are mostly traded by institutional investors rather than by individuals, and they are traded not as frequently as stocks are. Therefore, to minimize the bias arisen from infrequent trading, the samples consist of investment grade corporate bonds that have high liquidity. Bonds with speculative investment grades below BB are excluded from the sample for two reasons: they cannot be bought by AMCs due to their compliance requirements; and the pricing of such bonds are not accurate because they are not frequently traded. Credit rating of any bond can vary depending on the time of analysis and is incorporated into its valuation. Only publicly traded non-guarantee corporate bonds are included while convertible bonds (CB), bonds with embedded call and put options, and floating rate notes (FRN) are excluded because they, unlike straight bonds, do not have the general negative correlation between the market rate and the price, thus are not suitable for the intent of our analysis of corporate bonds holding period return in terms of bond characteristics or the differences in credit spread and long and short-term rates. The analysis is based on the sample data from January 1, 2001, when accumulation of corporate bond data such as return began systematically till June 30, Since historical return data is required to compute the pre-ranking Betas (DEF Beta and TERM Beta) for the market factors such as DEF and TERM factors, we use past 52 weeks return data to compute Betas. For this reason, we only include bonds which passed 52 weeks from the time of issuane. Table 2 above shows the number of sample bonds by year and credit rating. The average number of corporate bonds and other financial bonds satisfying the conditions 430

15 Asia-Pacific Journal of Financial Studies (2009) v38 n3 per year is 686, consisting of 51 AAA-rated bonds, 204 AA-rated bonds, 239 A-rated bonds, and 190 BBB-rated bonds. Table 2. No. of Sample bonds by year Year AAA AA A BBB Total Average The credit rating distribution of sample bonds by YTM and the duration is shown in Table 3. From this table we can see that the duration is longer for bonds with higher credit ratings. While AAA-rated bonds have an average of 1.88 duration, the durations are shorter for lower credit-rated bonds, i.e. BBB-rated bonds average duration is This is due to the fact that firms with high credit ratings can issue long-term bonds, whereas firms with lower credit ratings are considered default-prone and thus cannot easily issue long-term bonds; even if they can, they have to pay high risk premiums. Therefore, such firms tend to issue relatively short-term corporate bonds. So far, the duration increases each year across the credit rating groups. This means that newer issues have longer maturities than the bonds issued in the past. The YTM trend of sample bonds by credit rating shows that YTM increases as the credit rating falls, which is as expected. AAA bonds have an average YTM of 4.94 percent, while BBB bonds YTM is 180bp higher at 6.74 percent. The YTM trend over the years indicates a continued trend of declining interest rate since YTM continued to fall until 2004 for all ratings except BBB and has changed to either a static or rising trend thereafter. 5) In case of BBB rated bonds, there is no apparent trend of YTM since those bonds have a strong discriminatory effect among individual 5) From the 3-year government bonds and A + corporate bonds, which are representative bonds in the Korean market, continued interest rate fall occurred until 3Q of 2004 and a rising trend during

16 Determinants of Corporate Bond Returns in Korea Table 3. Duration and YTM of sample bonds Category Duration (Yr) YTM (%) AAA AA A BBB AAA AA A BBB Average Note: Values are weighted averages based on the total market price. bonds. In fact, BBB rating shows the highest YTM differences among bonds. Table 4 summarizes the coupon rate and YTM spread by credit ratings of sample bonds. Since coupon rate embodies the market valuation of a bond s credit level at the time of issuance, the lower the credit rating, the higher the coupon rate should be. This fact can be verified in Table 4. However, in the case of years 2002 and 2004, AA bonds have a lower coupon rate than AAA bonds of higher rating. This can be accounted for the discriminatory effect among individual bonds, which is derived from different market valuation of firms while they have the same credit ratings. In this way different coupon rates result within the same credit rating. 6) On the other hand, the right columns of Table 4 show YTM spread by credit rating, which is the difference between the AAA rating YTM and respective category YTM. The lower the credit rating, the higher the YTM spread should be. This fact can be verified in Table 4. However, in the case of AA bonds, we note the opposite is true in 2005 and Also, in the case of A rating, the spread decreased rapidly in 2005 and YTM by credit rating also shows discrimination among bonds of the same rating just as in the case of the coupon rate. 6) Rate discrimination occurs between bond issuers within the same credit rating since introduction of the mark-to-market valuation valuation system. Prior to this, bonds were issued and traded based on YTM matrix by credit rating, in general. However, since the system s introduction, bond pricing agencies incorporate, on top of credit rating, the issuing firm s credit level and liquidity premium. Consequently, the trading practices reflect the market rate of each issuing firm. As a result, rate discrimination occurs by issuers within the same credit rating. That is, some bonds of AA rating are issued at higher coupon rates than those of AAA rating. 432

17 Asia-Pacific Journal of Financial Studies (2009) v38 n3 Table 4. Coupon rate and spread of sample bonds (Unit: %) Category Coupon Rate Spread (YTM) AAA AA A BBB AAA AA A BBB Average Note: Values are weighted averages based on the total market price. The spread trend shows a steady fall since This is consistent with the steady YTM decrease seen in Table 3. The steady decline of YTM since 2000 has also brought down the spread in credit rating. And we observe from Table 4 that this spread has started to increase in response to the trend of market rate increase since Here, we note that the spread in credit rating decreases as the market rate (YTM) falls in the Korean bond market. 5.2 Descriptions of Variables The sample data on the analysis variables of return, duration, and credit rating, as well as outstanding issue amount and remaining maturity, are obtained from the pricing data serviced by KIS Pricing Inc. (KIS). The KIS pricing data is in the form of daily ending prices. For days without transactions, they compute theoretical prices using in-house developed pricing models. Since such pricing data is provided for all bonds held by institutional investors on a daily basis to facilitate mark-to-market pricing, the use of the pricing data should not pose a problem for our analysis purpose. Here, durations are in year unit and returns in annual percentages Return Excess returns of the corporate bonds in this study are computed by subtracting weekly returns of Treasury bills with 3-month remaining maturity, which are risk-free bonds, from the weekly returns of corporate bonds. Fama and French (1993) 433

18 Determinants of Corporate Bond Returns in Korea and Gebhardt et al. (2005) on the U.S. bond market use monthly returns for their analysis, however, to obtain the weekly returns while sample bonds must be those issued 24 months ahead of time to be included in computing the pre-ranking Beta for the monthly analysis. However, most Korean corporate bonds have only 3 years maturity, so only a few Korean corporate bonds satisfy this requirement. Although monthly data analysis would be more stable, we choose to use weekly data for our analysis. (9) in which : Excess return of the corporate bond at time t; : Weekly return of corporate bond at time t; and : Weekly return of a 3-months Treasury bill at time t. Corporate bond weekly return ( ) is computed using the bond price data as in Eq. (10). Bond prices were obtained from the daily bond pricing data provided by KIS. Price information used for return analysis include (1) traded price, (2) quoted price, and (3) theoretical (evaluated) price. Since the purpose of this study is to identify determinants of corporate bonds holding period return and provide insightful valuation for trading and risk management purpose for AMCs, the use of traded prices or quoted prices in order to calculate weekly returns is more desirable than the use of theoretical prices computed by pricing agencies. It is because traded prices reflect the market better, and it is more helpful to use them in making investment decisions. Since the corporate bond trading in Korea is not very active, however, only few bonds are applicable for the computation of weekly returns using market prices. Also, there is no systematic way of collecting quotation price information. 7) 7) In December 2007 Korea Securities Dealers Association (KSDA) launched a service for systematic collection, providing bond quotation information directly from the market to market participants. In the future analysis it will be possible to use this quotation price data. Because it is still difficult to assume that bonds are traded at their quotation prices, it is not possible to base our analysis on actual transaction price or quotation price data. Since the valuation price data by a bond pricing agency is directly used for computing fund return rates and it reflects the market rate by faithfully incorporating the actual price and quotation price, we think it might be reasonable to use the assessed pricing data by a bond pricing agency for our analysis. Apparently, on the other hand, the assessed pricing data cannot be easily used in actual investment decisions. 434

19 Asia-Pacific Journal of Financial Studies (2009) v38 n3 (10) in which : Corporate bond s weekly return;, : price of individual corporate bond weekly on Saturday (dirty price); and : Coupon accrued from a corporate bond between Sunday of the previous week and Saturday of current week. Weekly returns of short-term risk-free bonds are estimated from the Treasury bill prices of 3-month maturity. Certificate of Deposits (CDs hereinafter) and call rates do not reflect the Korean market effectively and one-month risk-free bond returns are not disclosed and thus are not used. 8) Further, treasury bills of 3-month maturity are not sufficient in the market and their maturity keeps decreasing over time. Therefore, we used hypothetical bonds of 3-month maturity to compute returns 9) by assuming a hypothetical bond of 3-month residual maturity. Moreover, we use the 3-month YTM from the respective pricing YTM matrix to compute the bond s Saturday price of the previous week and the Saturday price of the current week, and compute the weekly return. In this way the residual maturity of this Treasury bill is always maintained at 3-months for different analysis time periods. And the weekly return is always linked to the rate in the YTM matrix, which represents the market rates (refer to Eq. (11)). For the YTM matrix, we used the daily pricing data provided by KIS. (11) 8) Since a call rate is largely determined by a policy decision of the government monetary control agency, it cannot be considered as reflecting the market rate. Although recent CDs have market priced return data, they cannot be included in our sample because CD return data from 2001 to 2004 does not exist. 9) Gebhardt et al. (2005) used monthly returns of treasury bills. However, there are not many short-term bonds in the Korean bond market. Therefore, the use of actual bond price data will pose difficulties. Accordingly, we used hypothetical bond price data to compute weekly returns. Although we could have used weekly return of bond index, the residual maturity of the average of 3-month bond index will continue to change, and the bond index is not for a specific period. Thus, bond index was not suitable for our analysis which focuses on specific time periods. In the case of DEF Factor and TERM Factor computation, hypothetical bonds were used to compute weekly return for the same reasons. 435

20 Determinants of Corporate Bond Returns in Korea in which : Weekly return of hypothetical Treasury bill with 3-month maturity;, : Saturday price of hypothetical Treasury bill with 3-month maturity;, : YTM of 3-month bond in the YTM matrix; and : Coupon value occurring between Sunday of the previous week to Friday of the current week Characteristics: duration and credit rating Durations of sample corporate bonds are obtained from the corporate bond duration data disclosed by KIS. This duration data is the effective duration computed by Eq. (2). Credit ratings of corporate bonds are obtained from credit rating agency data. Since each bond issuance in Korea requires credit ratings from two or more credit rating agencies, multiple ratings exist for each bond. Based on the conservative practice of the market, we also use the lowest rating for each bond. We limit credit ratings which are less than 1.5 years old since their assignment. Bonds with no updated credit ratings in the recent 1.5 years are regarded as non-rated bonds and, therefore, excluded from our sample. This is in line with the market practice in which market participants recognize only credit ratings of the past one and half years Beta Variables: Pre-ranking DEF Beta and Pre-ranking TERM Beta As in Gebhardt et al. (2005), we estimate pre-ranking DEF Beta (hereinafter DEF Beta ) and pre-ranking TERM Beta (hereinafter TERM Beta ) using Eq. (12) for each bond. For the weekly return data necessary for DEF factor computation, we used YTM data of 3-year corporate bonds of A+ credit rating, which is the representative rate of the corporate bond market. 10) Here, as in the case of 3-year Treasury bonds, there are not many bonds with residual maturity of 3 years. Moreover, their residual maturities become shorter over time. Therefore, we use hypothetical bonds to compute the return. In doing so, we first generate a hypothetical bond of 3-year residual maturity, use 3-year YTM in the YTM matrix to compute the bond s Saturday price of the previous week and the Saturday price of the current week, and, finally, compute the weekly return of this bond. 10) Although the market s representative rate criterion was changed from A+ to AA during our analysis period, we used A+ return rates for continuity of our analysis. Also, bonds of A- rating have the largest share in our sample and take the position of the median rating in the sample. 436

21 Asia-Pacific Journal of Financial Studies (2009) v38 n3 (12) in which, : Weekly returns of corporate bond and risk-free short-term bonds prior to time t, respectively; : Default factor of prior to time t, which is the difference between corporate bond weekly return and Treasury bond 3-year return; : Term factor of prior to time t, which is the difference in return between Treasury 3-year bonds and 3-month bonds; and, : DEF Beta and TERM Beta at time t. Thus, hypothetical bonds are used for computation of DEF factor and TERM factor. Market prices of these bonds are computed using the market s representative return Average Category Table 5. TERM Beta distribution by Duration Duration Low Duration Medium Duration High Total TERM DUR TERM DUR TERM DUR TERM DUR TERM DUR TERM DUR TERM DUR TERM DUR Note: TERM refers to TERM Beta and DUR to Duration. Values are weighted averages based on the market price total. 437

22 Determinants of Corporate Bond Returns in Korea Accordingly, DEF factor and TERM factor can be regarded as market rates that reflect the average movement of the risk-free Treasury bond market and the corporate bond market. DEF Beta and TERM Beta are computed weekly using the regression equation of Eq. (12). For each corporate bond, the excess return time series data is used as the dependent variable, and the time series data of DEF Factor and TERM Factor as independent variables for a regression analysis to estimate Beta on a weekly basis. The time series data used for the estimation is obtained from at least 52 weeks of the immediate past from each analysis time point. Thus, the sample corporate bonds have at least 52 weeks elapsed since their issuance. Table 5 divides the duration into three groups and shows the TERM Beta distribution by year and by the duration group. As for 7-year averages, the TERM Beta for low Duration bonds is while that of the high Duration group is This shows that, the higher the Duration, the higher the TERM Beta is. We also note that TERM Beta has been increasing recently. This is due to the increasing trend of recent bond durations as seen in Table Table 6. DEF Beta distribution by credit rating Category AAA AA A BBB Total DEF DUR DEF DUR DEF DUR DEF DUR DEF DUR DEF DUR DEF DUR DEF Average DUR Note: TERM refers to TERM Beta and DUR to Duration. Values are weighted averages based on the market price total.

23 Asia-Pacific Journal of Financial Studies (2009) v38 n3 Table 6 summarizes DEF Beta distribution by year and by credit rating. DEF Beta showed an increasing trend since It was then reduced during 2005 and 2006 when credit spread was reduced, and began to increase again in 2007 with the rise of credit spread. There is no clear trend of DEF Beta over credit rating. This, we believe, is because TERM Beta effect was already incorporated in DEF Beta estimation. That is, Table 3 shows that, the higher the credit rating, the higher is the bond Duration. This fact is also verified in Table 6. Here, although high Duration results in higher TERM Beta values, DEF Beta does not show any systematic relation with credit rating due to the strong correlation between credit rating and TERM Beta. Table 7 shows the correlation between bond characteristics and Betas used as independent variables for excess return of the corporate bond in our study. Here, we can see a positive correlation coefficient between DEF Beta and TERM Beta at On the other hand, the correlation coefficient between credit rating and Duration, which are bond characteristics, is , showing a negative correlation. We note that, the higher the credit rating, the longer the duration gets. The correlation coefficient between credit rating and TERM Beta is , a similar correlation as with the duration, whereas credit rating and DEF Beta are negatively correlated with a low correlation coefficient. Correlations between Duration, a bond characteristic, and DEF Beta and TERM Beta are positive with coefficients and 0.946, respectively. This result implies that we may obtain more reliable results through the portfolio approach rather than the categorized approach, in which two variable groups are considered simultaneously as independent variables. Table 7. Correlations between independent variables DEF Beta TERM Beta credit rating Duration DEF Beta TERM Beta credit rating Duration Note: credit rating was assigned numerical values with AAA as 1 and CCC as Empirical Models We perform an empirical analysis to test which, bond characteristics or Betas, are 439

24 Determinants of Corporate Bond Returns in Korea a stronger determinant for the excess returns of corporate bonds. We use the following two approaches: the first approach is the portfolio approach used in Daniel and Titman (1997) and Gebhardt et al. (2005). In this approach, portfolios are formed based on a bond s characteristics (Duration and credit rating) and Betas (DEF Beta and TERM Beta). Then, we identify the determinant of excess return between portfolios. The second approach is regression on all sample bonds to identify the determinant. This method was used in Fama and Macbeth (1973). 6.1 Portfolio Approach In the portfolio approach, we form portfolios consisting of sample bonds based on bond characteristics and Betas and conduct analysis on them. We first categorize credit rating and duration, which are bond characteristics, into High, Medium and Low groups. We then use these groups to form nine portfolios from three groups according to each credit rating and duration. In these portfolios, the characteristic effect among bonds in a portfolio is assumed to be removed. Here, credit ratings are grouped so that AAA and AA are grouped into the High group, A rating to the Medium group, and BBB rating into the Low group. Duration is grouped so that the highest one-third is grouped to High group, the medium one-third to Medium group, and the bottom one-third to Low group. The composition and attributes of the nine portfolios are as follows. Table 8. Portfolio organization based on bond characteristics Category credit rating AAA AA A BBB High (GH, DH) (GM, DH) (GL, DH) Duration Medium (GH, DM) (GM, DM) (GL, DM) Low (GH, DL) (GM, DL) (GL, DL) Next, we form three groups based on individual bond s DEF Beta and TERM Beta, then form 9 portfolio corresponding to 9 ordered pairs of such three groups each. Here, DEF Beta and TERM Beta are grouped to High 1/3, Medium 1/3 and Low 1/3, as in the case of Duration grouping. Among bonds in each of these nine portfolios, the Beta effect is considered removed. Composition and attributes of thus formed nine portfolios are as follows. 440