SEARCH

SEARCH BY CITATION

ABSTRACT

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Hypotheses Development
  5. 3. Data and Summary Statistics
  6. 4. Results
  7. 5. Untabulated Robustness Tests
  8. 6. Conclusion
  9. REFERENCES

I examine the previously unexplored relation between aggregate earnings changes and corporate bond market returns. I find that aggregate earnings changes have a negative relation to investment-grade corporate bond market returns and a positive relation to high-yield corporate bond market returns. The aggregate earnings-returns relation is lower (i.e., less positive or more negative) for bonds with higher credit ratings and longer maturities. Further, I show that the aggregate earnings-returns relation is driven by both the expected and the news component of aggregate earnings changes. The expected component is negatively related to expected returns, and the news component is positively related to cash flow news and changes in nominal interest rates, and negatively related to changes in default premia. My results contribute to the understanding of the role of earnings in corporate bond markets as well as the macroeconomic role of accounting information.

1. Introduction

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Hypotheses Development
  5. 3. Data and Summary Statistics
  6. 4. Results
  7. 5. Untabulated Robustness Tests
  8. 6. Conclusion
  9. REFERENCES

Aggregate earnings are a valuable source of information about the macroeconomy as they reflect the value-generating ability of all the firms in the economy. Little is known, however, about the relation between aggregate earnings and asset prices and the exact nature of the information contained in aggregate earnings. I examine the relation between aggregate earnings changes and corporate bond market returns as well as the determinants of the aggregate earnings-returns relation for corporate bonds.

Kothari, Lewellen, and Warner [2006] document a negative relation between quarterly aggregate earnings changes and stock market returns, and Cready and Gurun [2010] confirm this negative relation at a daily frequency. Findings on the aggregate earnings-returns relation for stocks may not generalize to corporate bonds. Corporate bonds have shorter maturities and predetermined and senior cash flows compared to stocks, and entail different risk premia (e.g., Fama and French [1993]). Consequently, the sensitivity of corporate bond prices to the information contained in aggregate earnings can be different from the sensitivity of stock prices.

The determinants of the negative aggregate earnings-returns relation for stocks are not well understood. Academic literature suggests that aggregate earnings changes are negatively related to stock market returns because aggregate earnings move with discount rates, but the exact nature of the relation between aggregate earnings and discount rates remains an open question. One stream of literature argues that aggregate earnings move with discount rates because unexpected aggregate earnings changes are positively related to discount rate news (e.g., Kothari, Lewellen, and Warner [2006]), whereas another stream of literature suggests that aggregate earnings move with discount rates because expected aggregate earnings changes are negatively related to expected returns (e.g., Sadka and Sadka [2009]).

The precise mechanism that links the unexpected component of aggregate earnings changes to discount rate news is also unclear. Shivakumar [2007] suggests that unexpected aggregate earnings changes are positively related to discount rate news because they are positively associated with changes in inflation expectations, Patatoukas and Yan [2010] propose a positive relation between unexpected aggregate earnings changes and changes in real interest rates, and Patatoukas [2013] provides evidence of a positive relation between unexpected aggregate earnings changes and changes in real interest rates, inflation expectations, and the equity risk premium.

The use of corporate bond prices instead of stock prices can provide additional insights into the relation between aggregate earnings and discount rates for two reasons. First, corporate bonds have predetermined cash flows and finite maturities. As a result, I can accurately match the duration of corporate bonds to the duration of risk-free assets, I can isolate the return component that is associated with changes in interest rates from the return component that is associated with changes in risk premia, and I can more precisely examine the relation between aggregate earnings changes, changes in interest rates, and changes in risk premia. Second, the determinants of stock and bond prices are different. Stock prices move primarily in response to changes in equity risk premia, whereas bond prices move primarily in response to changes in inflation expectations (e.g., Campbell and Ammer [1993]). If aggregate earnings move with asset prices because they contain information about inflation expectations (e.g., Shivakumar [2007]), then using corporate bond prices to examine the relation between aggregate earnings and discount rates will produce more robust evidence than using stock prices.

I use a sample of quarterly observations from January 1973 through December 2010 to examine the previously unexplored relation between aggregate earnings changes and corporate bond market returns. I find that aggregate earnings changes are negatively related to investment-grade corporate bond market returns, and positively related to high-yield corporate bond market returns. The aggregate earnings-returns relation for corporate bonds is statistically as well as economically significant; a two-standard-deviation positive shock to aggregate earnings changes corresponds to a 1.66% decrease in the quarterly investment-grade corporate bond market returns, and a 1.23% increase in the quarterly high-yield corporate bond market returns. Further, I find that the aggregate earnings-returns relation is lower (i.e., less positive or more negative) for bonds with higher credit ratings and longer maturities. My results differ from the firm-level finding of a positive relation between earnings changes and corporate bond returns, and from the finding that the earnings-returns relation for corporate bonds at the firm level depends only on corporate bond credit ratings (e.g., Datta and Dhillon [1993], Easton, Monahan, and Vasvari [2009]).

Next, I shed light on the exact nature of the information contained in aggregate earnings. In line with Sadka and Sadka [2009], I find that expected aggregate earnings changes are negatively related to expected returns. Further, I show that unexpected aggregate earnings changes are positively associated with changes in nominal interest rates and cash flow news, and negatively associated with changes in default premia. These latter findings are consistent with Kothari, Lewellen, and Warner [2006], who suggest that aggregate earnings changes contain new information about discount rates.

My findings contribute to two streams of literature. First, I add to the understanding of the role of earnings in corporate bond markets (e.g., Datta and Dhillon [1993], Easton, Monahan, and Vasvari [2009]). My results suggest that firm-level findings on the relation between earnings changes and corporate bond returns do not generalize to the aggregate level because the information contained in firm-specific earnings is different from the information contained in aggregate earnings. Firm-specific earnings changes are positively related to firm-specific corporate bond returns because higher than expected firm-specific earnings provide good news about future cash flows. Aggregate earnings, however, also contain information about the growth of the economy, and thus about discount rates. Hence, the relation between earnings changes and corporate bond returns can be weak or even negative at the aggregate level.

Second, my analysis contributes to a growing body of literature that examines the macroeconomic role of accounting information (e.g., Kothari, Lewellen, and Warner [2006], Anilowski, Feng, and Skinner [2007]).1 Macroeconomic information impacts investors' consumption and asset allocation decisions as well as governments' policies. The macroeconomic role of accounting information, however, is largely unexplored. I contribute to this stream of literature by reconciling prior findings on the relation between aggregate earnings changes, expected returns, and discount rate news (e.g., Kothari, Lewellen, and Warner [2006], Sadka and Sadka [2009]). In particular, I show that aggregate earnings changes contain both an expected and a news component, and both of these components drive the earnings-returns relation at the aggregate level down. Further, my analysis promotes our understanding of the exact link between unexpected aggregate earnings changes and discount rate news (e.g., Kothari, Lewellen, and Warner [2006]). My findings suggest that unexpected aggregate earnings changes are positively related to changes in nominal interest rates and negatively related to changes in default premia. I do not address, however, whether aggregate earnings move with nominal interest rates because they are associated with real interest rates, inflation expectations, or both. Finally, I contribute to the understanding of the aggregate earnings-returns relation for stocks (e.g., Kothari, Lewellen, and Warner [2006]). My results suggest that the negative relation between aggregate earnings changes and stock market returns is partially driven by a negative relation between expected aggregate earnings changes and expected returns, and a positive relation between unexpected aggregate earnings changes and changes in nominal interest rates. Nevertheless, my contribution to the understanding of the aggregate earnings-returns relation for stocks is incomplete as my analysis does not provide evidence on the relation between aggregate earnings changes and changes in equity risk premia.

2. Hypotheses Development

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Hypotheses Development
  5. 3. Data and Summary Statistics
  6. 4. Results
  7. 5. Untabulated Robustness Tests
  8. 6. Conclusion
  9. REFERENCES

To explain the relation between aggregate earnings changes and corporate bond market returns, I use the return decomposition proposed by Campbell [1991].2 Campbell decomposes market returns (Rt) into expected returns (Et−1[Rt]), news about cash flows (NCF), and news about discount rates (NDR):

  • display math(1)

Consequently, the relation between aggregate earnings changes (ΔXt) and market returns depends on the relation between aggregate earnings changes and expected returns, cash flow news, and discount rate news (Hecht and Vuolteenaho [2006]):

  • display math(2)

Aggregate earnings changes can be decomposed into expected aggregate earnings changes (Et−1Xt]) and unexpected aggregate earnings changes (UΔXt):

  • display math(3)

The expected component of aggregate earnings changes is by definition orthogonal to the news component of market returns, and the unexpected component of aggregate earnings changes has by definition no relation to the expected component of market returns. So, equation (2) can be rewritten as:

  • display math(4)

Hence, the relation between market returns and aggregate earnings changes depends on the relation between expected returns and expected aggregate earnings changes (Cov{Et−1[Rt], Et−1Xt]}), the relation between cash flow news and unexpected aggregate earnings changes (Cov{NCF, UΔXt}), and the relation between discount rate news and unexpected aggregate earnings changes (Cov{NDR, UΔXt}).

The cash flow effect of aggregate earnings is positive because higher than expected aggregate earnings provide good news about future cash flows and thus cause higher prices and higher returns (i.e., Cov{NCF, UΔXt} > 0).3 The relation between aggregate earnings changes and discount rates, however, remains an open question.

Some authors argue that aggregate earnings changes are negatively related to discount rates. Ball, Sadka, and Sadka [2009] and Sadka and Sadka [2009] suggest that aggregate earnings are highly predictable. High expected aggregate earnings are associated with high expected economic growth. Assuming countercyclical risk premia (e.g., Fama and French [1989]), high expected economic growth results in lower risk aversion or risk, and lower risk premia. As a result, the negative contemporaneous aggregate earnings-returns relation documented by Kothari, Lewellen, and Warner [2006] is attributed to a negative relation between expected aggregate earnings changes and expected returns (i.e., Cov{Et−1[Rt], Et−1Xt]} < 0).4

Others suggest that aggregate earnings changes are positively related to discount rates. Patatoukas and Yan [2010] argue that aggregate earnings incorporate new information. Higher than expected aggregate earnings provide good news about future cash flows. Investors who wish to smooth their consumption borrow against their future income, and thus bid up interest rates and raise discount rates. Higher discount rates result, in turn, in lower prices and lower returns. This argument is also in line with monetary policy interventions. Governments and central banks increase interest rates when earnings are higher than expected to avoid an overheating of the economy. Shivakumar [2007] also proposes a positive relation between unexpected aggregate earnings changes and discount rate news and, more specifically, a positive relation between unexpected aggregate earnings changes and changes in inflation expectations. The line of reasoning is that good news about aggregate earnings signals improvement in economic conditions, rise in aggregate demand, and ultimately higher inflation. Academic literature also suggests a positive relation between unexpected aggregate earnings changes and changes in risk premia (Patatoukas [2013]). The idea is that risk-averse investors wish to consume more during economic expansions and thus demand higher rates in return for their investment (Cochrane [2006]). Overall, the latter stream of literature attributes the negative aggregate earnings-returns relation documented by Kothari, Lewellen, and Warner [2006] to a positive relation between unexpected aggregate earnings changes and discount rate news (i.e., Cov{NDR, UΔXt} > 0).

Regardless of whether the former or the latter stream of literature is correct, the fact remains that the cash flow effect of aggregate earnings is positive, and the discount rate effect of aggregate earnings is negative. As a result, the cash flow effect and the discount rate effect of aggregate earnings move asset prices in opposite directions. In the case of stocks, the discount rate effect dominates the cash flow effect, and the aggregate earnings-returns relation is negative. The relation between aggregate earnings changes and aggregate bond returns, however, is not clear a priori.

There are three reasons the aggregate earnings-returns relation can be different for bonds as compared to stocks. First, payments to bondholders are predetermined and take priority over payments to stockholders. Thus, corporate bond prices are less sensitive to cash flow changes than are stock prices. Second, corporate bonds have shorter durations than stocks. Therefore, they are less sensitive to discount rate changes than are stocks. Third, stocks and bonds are subject to different risk premia. Bond prices move primarily in response to changes in term and default premia, and stock prices move primarily in response to changes in equity risk premia (e.g., Campbell and Ammer [1993], Fama and French [1993]). Hence, stocks and corporate bonds are differentially sensitive to the cash flow effect and the discount rate effect of aggregate earnings, and the relation between aggregate earnings changes and corporate bond market returns is an empirical question.

The aggregate earnings-returns relation can also vary with corporate bond characteristics and, more specifically, with corporate bond credit ratings and corporate bond maturities. At the firm level, Easton, Monahan, and Vasvari [2009] show a higher earnings-returns relation for low-rated than for high-rated bonds. The reason is that the payoff for low-rated bonds is similar to a call option that is close to out-of-the-money, whereas the payoff for high-rated bonds is similar to a call option that is deep in-the-money. Prices of out-of-the-money options are more sensitive to changes in the value of the underlying asset than prices of in-the-money options. Accordingly, low-rated bonds are more sensitive to changes in earnings than high-rated bonds. There is no evidence that corporate bond maturity has an impact on the earnings-returns relation at the firm level.

In the case of the aggregate earnings-returns relation, the impact of corporate bond credit ratings is not obvious ex ante. On the one hand, it is possible that the aggregate earnings-returns relation will be lower in the case of bonds with higher credit ratings. The reason is that high-rated bonds are less sensitive to changes in cash flows than low-rated bonds (Easton, Monahan, and Vasvari [2009]). Hence, there should be less of a cash flow effect for high-rated bonds. At the same time, Shivakumar [2007] and Patatoukas and Yan [2010] suggest that aggregate earnings relate positively to interest rates. Interest rates have a negative relation to credit spreads, with a more pronounced negative relation for low-rated than high-rated bonds (Duffee [1998]). The negative relation between interest rates and credit spreads attenuates the discount rate effect of aggregate earnings and more so for low-rated than for high-rated bonds. Hence, the discount rate effect of aggregate earnings should be stronger for high-rated bonds. The weaker cash flow effect and the stronger discount rate effect of aggregate earnings for high-rated than low-rated bonds can lead to a lower aggregate earnings-returns relation for bonds with higher credit ratings. On the other hand, it is possible that the aggregate earnings-returns relation will be higher for bonds with higher credit ratings because high-rated bonds are less sensitive to discount rate changes than low-rated bonds (Fama and French [1989]).

With regard to the impact of bond maturity on the aggregate earnings-returns relation, long-term bonds are by construction more sensitive to changes in discount rates than short-term bonds. In addition, long-term bonds are less sensitive to changes in cash flows than short-term bonds because cash flow shocks are less persistent than discount rate shocks (Campbell [1991]). Therefore, I expect a lower aggregate earnings-returns relation for long-term than short-term bonds.

3. Data and Summary Statistics

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Hypotheses Development
  5. 3. Data and Summary Statistics
  6. 4. Results
  7. 5. Untabulated Robustness Tests
  8. 6. Conclusion
  9. REFERENCES

My sample consists of all firms with data available in Compustat North America Fundamentals Quarterly from January 1973 through December 2010.5 I drop firms that are not listed on the NYSE, AMEX, or NASDAQ to make my results more comparable to other studies. I exclude firms with fiscal year-ends other than March, June, September, and December to better align quarterly aggregate earnings changes with quarterly aggregate bond returns. I drop firms with earnings announcement dates more than three months after a quarter's end to exclude the likelihood of stale earnings figures. I also delete the top and bottom 0.5% of firms ranked by earnings changes each quarter to mitigate the influence of outliers.

Similar to Kothari, Lewellen, and Warner [2006], I measure aggregate earnings changes (ΔE/A) as the value-weighted average of firm-specific quarterly earnings changes.6 Firm-specific earnings changes are calculated as earnings in the current quarter minus earnings four quarters ago scaled by lagged total assets. The scaling factor is lagged by four quarters. I use total assets as the scaling factor in the primary analysis because total assets are always positive, unlike earnings or book value of equity, which can take negative values. Nevertheless, the results are robust to using alternative deflators. I measure earnings as income before extraordinary items. These sample selection criteria and data requirements yield a sample of 401,822 firm-quarter observations.

To measure quarterly corporate bond market returns, I use the total returns of the value-weighted Bank of America Merrill Lynch U.S. Corporate Bond Indices downloaded from Bloomberg. Total return is the sum of the price return, the accrued interest return, and the coupon return. I use 13 corporate bond indices with different maturities and different credit ratings (R_1-3_AAA-AA, R_1-3_A-BBB, R_3-5_AAA-AA, R_3-5_A-BBB, R_5-10_AAA-AA, R_5-10_A-BBB, R_15+_AAA-AA, R_15+_A-BBB, R_all_invest_grade, R_all_BB, R_all_B, R_all_CCC, and R_all_high_yield).

The numbers in the indices represent the remaining maturities of the bonds that constitute the indices, and the letters represent the credit ratings. Each letter represents all subcategories within each credit rating category. For example, the index R_1-3_AAA-AA includes bonds with AAA, AA+, AA, and AA− credit ratings. Credit ratings are the average of the individual bond ratings provided by Moody's, Standard & Poor's, and Fitch. The indices of bonds with credit ratings BBB− and above are the investment-grade corporate bond indices, and those of bonds with credit ratings BB+ and below are the high-yield corporate bond indices. “All” stands for “all available maturities.” For example, R_all_invest_grade and R_all_high_yield are indices of bonds of all available maturities that are investment grade and high yield, respectively. Bank of America Merrill Lynch provides fewer high-yield than investment-grade indices, and it provides indices with varying maturities only in the case of the investment-grade bonds. Thus, I use nine investment-grade and only four high-yield indices for the analysis.

There are an average of 776 issues in the indices and each firm is included in each index, with three issues in the case of the investment-grade indices and fewer than two issues in the case of the high-yield indices, on average. Thus, the corporate bond indices used in the analysis are well diversified. The indices are rebalanced monthly, so the average remaining maturities and the average credit ratings remain fairly stable through time. Bond illiquidity is not an issue in the analysis because I use quarterly frequency data. Further, the indices cover bonds with large issue size, and bond issue size is positively associated with bond liquidity (e.g., Hong and Warga [2000]). In particular, the investment-grade (high-yield) bonds in the indices have a minimum outstanding value equal to $250 ($100) million.

For estimation of the models, I use ordinary least squares and Newey-West heteroscedasticity- and autocorrelation-consistent standard errors. I set the bandwidth of the Bartlett kernel to the integer value of inline image, where T is the number of observations used in the time-series regressions (Newey and West [1987, 1994]).

Table 1 reports summary statistics for the main variables used in the regressions. The mean quarterly returns of the corporate bond indices range from 1.60% to 2.28%. As expected, bonds with longer maturities and lower credit ratings earn, on average, higher returns, but the differences are not statistically significant (Cornell and Green [1991]). In line with Blume, Keim, and Patel [1991], low-rated investment-grade bond returns have lower standard deviations than high-rated investment-grade bond returns, but the differences are not statistically significant.

Table 1. Summary Statistics
Panel A: Univariate statistics
 MeanMedianStandard DeviationSkewnessKurtosisAutocorrelationN
R_1-3_AAA-AA1.90%1.59%1.96%1.87210.7160.056140
R_1-3_A-BBB1.99%1.78%1.86%1.1439.2590.190140
R_3-5_AAA-AA2.02%1.84%2.72%1.1908.645−0.079140
R_3-5_A-BBB2.07%1.93%2.69%1.0627.7410.069135
R_5-10_AAA-AA2.09%1.78%3.71%0.6165.308−0.086152
R_5-10_A-BBB2.14%1.82%3.63%0.6785.8560.066152
R_15+_AAA-AA2.23%1.77%5.45%0.5725.338−0.128152
R_15+_A-BBB2.28%1.97%5.17%0.5345.572−0.024152
R_all_invest_grade2.09%1.81%4.10%0.7576.889−0.004152
R_all_BB1.90%1.79%4.19%−0.2046.2320.32356
R_all_B1.60%1.86%5.54%−0.1817.6400.31356
R_all_CCC2.24%2.00%9.98%0.7467.0890.38756
R_all_high_yield2.26%2.38%4.90%0.2438.4520.34097
ΔE/A0.15%0.21%0.44%−1.91610.8040.697152
Panel B: Correlation coefficients
  R_all_ R_all_  
  invest_grade high_yield ΔE/A
  1. This table presents summary statistics of the main variables of the regression models. Panel A presents univariate statistics and panel B presents Pearson (Spearman) correlation coefficients below (above) the diagonal. R_1-3_AAA-AA, R_1-3_A-BBB, R_3-5_AAA-AA, R_3-5_A-BBB, R_5-10_AAA-AA, R_5-10_A-BBB, R_15+_AAA-AA, R_15+_A-BBB, R_all_invest_grade, R_all_BB, R_all_B, R_all_CCC, and R_all_high_yield are quarterly total returns of the various corporate bond indices. Total return is the sum of the price return, the accrued interest return, and the coupon return. The numbers in the names of the corporate bond indices represent the remaining maturities of the bonds in the indices, and the letters represent the credit ratings. “All” stands for “all available maturities.” ΔE/A is the quarterly aggregate earnings changes measured as the value-weighted average of firm-specific earnings changes. Firm-specific earnings change is the seasonally differenced income before extraordinary items scaled by lagged total assets. The sample extends from January 1973 through December 2010. *, **, and *** denote significance at the 1%, 5%, and 10% levels, respectively (two-tailed test).

R_all_invest_grade 1.00 0.46*** −0.19*
R_all_high_yield 0.50*** 1.00 0.01
ΔE/A −0.20** 0.14 1.00

The first-order autocorrelations of the high-yield corporate bond indices are higher than those of the investment-grade corporate bond indices, presumably due to less frequent trading in the high-yield corporate bond market. The number of observations for the various corporate bond indices ranges from 56 to 152 because different indices have different inception dates. The mean quarterly change in the aggregate earnings is equal to 0.15%, similar to Kothari, Lewellen, and Warner [2006]. The aggregate earnings changes exhibit large time-series variation. Further, some variables are persistent, but the augmented Dickey-Fuller test rejects the null hypothesis of a unit root for all the variables at the 1% level.

As panel B of table 1 shows, the quarterly returns of the investment-grade and the high-yield indices are positively and significantly correlated at the 1% level. Thus, there is a high degree of commonality in the returns of the various corporate bond indices. Aggregate earnings changes are negatively and significantly related to the returns of the investment-grade corporate bond index, and not related to the returns of the high-yield corporate bond index. These correlation coefficients contrast with the firm-level findings of a positive relation between earnings changes and corporate bond returns (Easton, Monahan, and Vasvari [2009]). Further, the correlation coefficients provide preliminary evidence that the aggregate earnings-returns relation depends on corporate bond characteristics.

4. Results

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Hypotheses Development
  5. 3. Data and Summary Statistics
  6. 4. Results
  7. 5. Untabulated Robustness Tests
  8. 6. Conclusion
  9. REFERENCES

I examine the relation between aggregate earnings changes and corporate bond market returns, as well as the impact of corporate bond characteristics on the aggregate earnings-returns relation in section 'corporate bond market returns and aggregate earnings changes'. I examine the drivers of the aggregate earnings-returns relation for corporate bonds in section 'aggregate earnings, cash flows, and discount rates'.

4.1 corporate bond market returns and aggregate earnings changes

Table 2 presents the results of regressions of quarterly corporate bond market returns on contemporaneous aggregate earnings changes. As panel A shows, aggregate earnings changes are significantly and negatively related to the returns of investment-grade corporate bond indices; the regression coefficients of aggregate earnings changes range from −0.79 to −3.09 with t-statistics between −1.75 and −2.82. Further, as panel B shows, aggregate earnings changes are significantly positively related to returns of the high-yield corporate bond indices in two of four regressions; the slopes on aggregate earnings changes range from 0.38 to 3.29 with t-statistics between 0.43 and 1.81. The adjusted R2 of the regressions range from −1.51% to 5.50% and are presumably low because of the confounding impact of the cash flow effect and the discount rate effect of aggregate earnings. These low adjusted R2 are also similar to those of earlier studies (e.g., Kothari, Lewellen, and Warner [2006]).

Table 2. Corporate Bond Market Returns and Aggregate Earnings Changes
Panel A: Investment-grade corporate bond market returns
 R_1-3_R_1-3_R_3-5_R_3-5_R_5-10_R_5-10_R_15+_R_15+_R_all_
 AAA-AAtA-BBBtAAA-AAtA-BBBtAAA-AAtA-BBBtAAA-AAtA-BBBtinvest_gradet
Intercept0.020***0.021***0.022***0.022***0.024***0.024***0.027***0.026***0.024***
 (9.921)(9.186)(8.813)(7.517)(7.155)(6.325)(5.599)(5.351)(5.970)
ΔE/At−0.965**−0.790*−1.519***−1.114*−1.897**−1.483*−3.087**−2.461**−1.902**
 (−2.453)(−1.748)(−2.820)(−1.852)(−2.577)(−1.950)(−2.570)(−2.447)(−2.301)
N140140140135152152152152152
Adjusted R23.94%2.75%5.30%2.61%4.38%2.54%5.50%3.69%3.47%
F-statistic6.02**3.05*7.95***3.43*6.64**3.80*6.60**5.99**5.29**
Panel B: High-yield corporate bond market returns
 R_all_R_all_R_all_R_all_
 BBtBtCCCthigh_yieldt
Intercept0.019***0.014*0.0200.021***
 (2.827)(1.801)(1.287)(3.253)
ΔE/At0.3832.175*3.291*1.407
 (0.428)(1.813)(1.720)(1.211)
N56565697
Adjusted R2−1.51%4.53%2.65%1.05%
F-statistic0.183.29*2.96*1.47
Panel C: Impact of bond credit ratings
 R_1-3_AAA-AAR_3-5_AAA-AAR_5-10_AAA-AAR_15+_AAA-AAR_all_BBR_all_invest_grade
 
 R_1-3_R_3-5_R_5-10_R_15+_R_all_R_all_
 A-BBBA-BBBA-BBBA-BBBCCChigh_yield
  1. This table presents the results of regressions of corporate bond market returns on contemporaneous aggregate earnings changes. Panel A presents the results of the regressions for the investment-grade corporate bond market returns. Panel B presents the results of the regressions for the high-yield corporate bond market returns. Panel C compares the coefficients of the aggregate earnings changes in panels A and B for pairs of corporate bond indices with the same maturity but different credit ratings. Panel D compares the coefficients of the aggregate earnings changes in panel A for pairs of corporate bond indices with the same credit rating but different maturities. R_1-3_AAA-AA, R_1-3_A-BBB, R_3-5_AAA-AA, R_3-5_A-BBB, R_5-10_AAA-AA, R_5-10_A-BBB, R_15+_AAA-AA, R_15+_A-BBB, R_all_invest_grade, R_all_BB, R_all_B, R_all_CCC, and R_all_high_yield are quarterly total returns of the various corporate bond indices. Total return is the sum of the price return, the accrued interest return, and the coupon return. The numbers in the names of the corporate bond indices represent the remaining maturities of the bonds in the indices, and the letters represent the credit ratings. “All” stands for “all available maturities.” ΔE/A is the quarterly aggregate earnings changes measured as the value-weighted average of firm-specific earnings changes. Firm-specific earnings change is the seasonally differenced income before extraordinary items scaled by lagged total assets. The sample extends from January 1973 through December 2010. I use ordinary least squares for the calculation of the regression coefficients and the Newey-West heteroscedasticity- and autocorrelation-consistent standard errors with four lags. t-statistics are in parentheses below the coefficient estimates. F-statistics test the null hypothesis that the regression coefficients are jointly equal to zero. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively (two-tailed test).

Difference in−0.175−0.405−0.414−0.626−2.908−3.309**
ΔE/A coefficients(−0.292)(−0.502)(−0.391)(−0.400)(−1.377)(−2.321)
Panel D: Impact of bond maturity
 R_1-3_R_1-3_    
 AAA-AAA-BBB    
     
 R_15+_R_15+_    
 AAA-AAA-BBB    
Difference in2.122*1.671    
ΔE/A coefficients(1.679)(1.516)    

The relation between aggregate earnings changes and corporate bond market returns is statistically as well as economically significant; a two-standard-deviation positive shock to the aggregate earnings changes corresponds to a 0.69–2.70% reduction in the quarterly returns of the investment-grade corporate bond indices, and a 0.33–2.88% increase in the quarterly returns of the high-yield corporate bond indices. The results of table 2, panels A and B suggest that aggregate earnings comove with cash flows and discount rates. The discount rate effect of aggregate earnings dominates the cash flow effect in the case of the investment-grade indices, and the earnings-returns relation is negative. But the cash flow effect dominates the discount rate effect in the case of the high-yield indices, and the earnings-returns relation is positive.7

In table 2, panels C and D, I examine the impact of corporate bond characteristics on the aggregate earnings-returns relation. To this end, I compare the coefficients of the aggregate earnings changes from panels A and B either for pairs of indices with the same maturity but different credit ratings or for pairs of indices with the same credit rating but different maturities. Table 2, panel C presents the differences in the coefficients of the aggregate earnings changes for the pairs with the same maturity but different credit ratings. All differences are negative, but only the difference in the coefficients of the aggregate earnings changes for the last pair of indices is statistically significant. The insignificant differences of the first five pairs are seen presumably because these pairs capture differences in the coefficients of the aggregate earnings changes among either investment-grade or high-yield indices, whereas the last pair captures the difference in the coefficients for the investment-grade versus the high-yield indices.

Figure 1 plots regression coefficients of the aggregate earnings changes from panels A and B of table 2 for corporate bond indices with similar maturities but different credit ratings. The regression coefficients decline monotonically as the corporate bond credit ratings increase. The findings of table 2, panel C and figure 1 provide support for the hypothesis that low-rated bonds are more sensitive to cash flow changes than high-rated bonds (Easton, Monahan, and Vasvari [2009]). Further, my findings are in line with the proposition that the discount rate effect of aggregate earnings is weaker for low-rated bonds, because there is a more pronounced negative relation between interest rates and credit spreads for bonds with low credit ratings (Duffee [1998]).

image

Figure 1. This figure plots the regression coefficients of aggregate earnings changes from panels A and B of table 2 for indices with similar maturities but different credit ratings. The horizontal axis presents the credit ratings of the bonds in the indices, and the vertical axis presents the regression coefficients of aggregate earnings changes. In the case of high-yield indices, I use bonds of all available maturities. In the case of investment-grade indices, I use bonds with remaining maturities between 5 and 10 years, because the duration of these indices is similar to the duration of the high-yield indices used in the figure.

Download figure to PowerPoint

Panel D of table 2 shows the differences in the coefficients of aggregate earnings changes for pairs of indices with the same credit rating but different maturities. The differences in the coefficients are, as expected, positive and statistically different from zero at the 15% level or lower. Figure 2 plots the coefficients of the aggregate earnings changes from table 2, panel A for corporate bond indices with the same credit ratings but different maturities. There is a lower aggregate earnings-returns relation for long-term than short-term bonds. These findings are in line with the proposition that long-term bonds are more sensitive to discount rate changes than short-term bonds.8

image

Figure 2. This figure plots the regression coefficients of aggregate earnings changes from panel A of table 2 for indices with the same credit ratings but different maturities. The horizontal axis presents the remaining maturities of the bonds in the indices, and the vertical axis presents the regression coefficients of aggregate earnings changes. The light (dark) gray line presents the coefficients of aggregate earnings changes for indices with credit ratings between A and BBB (AAA and AA).

Download figure to PowerPoint

To sum up, I find a negative aggregate earnings-returns relation for investment-grade bonds and a positive aggregate earnings-returns relation for high-yield bonds. The aggregate earnings-returns relation is lower for high-rated and long-term bonds. These results differ from firm-level findings of a positive relation between earnings changes and corporate bond returns, and from the finding that the earnings-returns relation for corporate bonds at the firm level depends only on the corporate bond credit ratings. My findings suggest that aggregate earnings move with cash flows and discount rates.

4.2 aggregate earnings, cash flows, and discount rates

Next, I examine the information content of aggregate earnings changes by conducting four sets of tests. First, I regress corporate bond market returns on aggregate earnings changes and lagged stock market returns to study the relation between aggregate earnings changes, expected returns, and discount rate news. In this and all subsequent analysis, I use lagged stock market returns to control for the expected component of aggregate earnings changes. If expected aggregate earnings changes are negatively related to expected returns (e.g., Sadka and Sadka [2009]), then the contemporaneous aggregate earnings-returns relation will increase after controlling for lagged equity market returns. That is, the regression coefficients of aggregate earnings changes will become less negative or more positive after controlling for the stock market returns. Further, if unexpected aggregate earnings changes are positively related to discount rate news (e.g., Kothari, Lewellen, and Warner [2006]), then the negative aggregate earnings-returns relation for investment-grade bonds will persist even after controlling for the expected component of aggregate earnings changes.

Second, I regress one-quarter-forward corporate bond market returns on aggregate earnings changes and lagged equity market returns to examine the relation between unexpected aggregate earnings changes and discount rate news. If unexpected aggregate earnings changes are positively related to discount rate news, then unexpected aggregate earnings changes should be positive predictors of corporate bond market returns.

Third, I study the relation between unexpected aggregate earnings changes, changes in nominal interest rates, and changes in risk premia. To this end, I regress excess corporate bond market returns on aggregate earnings changes and lagged equity market returns. If unexpected aggregate earnings changes are positively related to changes in nominal interest rates (e.g., Shivakumar [2007]), then the aggregate earnings-returns relation will increase after controlling for nominal interest rates. In addition, if unexpected aggregate earnings changes are positively related to risk premia (e.g., Patatoukas [2013]), then the negative aggregate earnings-returns relation for investment-grade bonds will persist even after controlling for nominal interest rates.

And fourth, I study the relation between unexpected aggregate earnings changes, cash flow news, and changes in default premia. To this end, I regress corporate bond market returns on aggregate earnings changes, changes in credit default swap (CDS) spreads, and lagged equity market returns. Changes in CDS spreads are negatively related to news about cash flows and positively related to changes in default premia. If aggregate earnings news are positively related to cash flow news and negatively related to changes in default premia (e.g., Callen, Livnat, and Segal [2009]), then the aggregate earnings-returns relation will decrease after controlling for changes in CDS spreads.

Table 3 presents the results of the regressions of corporate bond market returns on aggregate earnings changes and lagged stock market returns (R_equity). Stock market returns are the quarterly returns, including distributions, of the equity market index downloaded from CRSP. I use lagged stock market instead of lagged corporate bond market returns to control for expected aggregate earnings changes because Downing, Underwood, and Xing [2009], among others, suggest that the stock market is informationally more efficient than the corporate bond market. In this and all subsequent analysis, I use t-statistics to compare the slopes on aggregate earnings changes across models. Controlling for lagged stock market returns raises the regression coefficients of aggregate earnings changes in table 3 compared to table 2 in 12 of the 13 regressions, and these increases are statistically significant at the 15% level or lower in 2 of the 13 regressions. Nevertheless, even after controlling for lagged equity market returns, the aggregate earnings changes remain negatively and significantly related to investment-grade corporate bond market returns and positively and significantly related to high-yield corporate bond market returns. For investment-grade bonds, the coefficients on aggregate earnings changes range from −0.67 to −2.96 with t-statistics between −1.44 and −3.09, and for high-yield bonds, the slopes on aggregate earnings changes vary considerably from 2.15 to 9.31 with t-statistics between 1.46 and 4.05.

Table 3. Corporate Bond Market Returns, Aggregate Earnings Changes, and Equity Market Returns
Panel A: Investment-grade corporate bond market returns
 R_1-3_R_1-3_R_3-5_R_3-5_R_5-10_R_5-10_R_15+_R_15+_R_all_
 AAA-AAtA-BBBtAAA-AAtA-BBBtAAA-AAtA-BBBtAAA-AAtA-BBBtinvest_gradet
Intercept0.020***0.022***0.023***0.024***0.024***0.025***0.028***0.028***0.025***
 (7.087)(6.861)(6.190)(5.857)(5.349)(5.225)(4.477)(4.470)(4.970)
ΔE/At−1.016***−0.672−1.409***−0.833−1.788**−1.149−2.957**−2.133*−1.600*
 (−3.087)(−1.629)(−2.795)(−1.440)(−2.325)(−1.454)(−2.185)(−1.961)(−1.828)
R_equityt−1−0.007−0.011−0.024−0.024−0.031−0.023−0.035−0.024−0.028
 (−0.328)(−0.555)(−0.701)(−0.777)(−0.721)(−0.640)(−0.559)(−0.414)(−0.664)
R_equityt−2−0.004−0.013−0.006−0.026−0.013−0.042−0.022−0.045−0.039
 (−0.262)(−0.710)(−0.284)(−1.117)(−0.442)(−1.503)(−0.519)(−1.125)(−1.296)
R_equityt−30.033**0.0140.0280.0150.045*0.0320.0460.0310.037
 (2.021)(0.903)(1.546)(0.702)(1.823)(1.096)(1.238)(0.759)(1.244)
R_equityt−4−0.010−0.017−0.026−0.035−0.027−0.036−0.028−0.035−0.035
 (−0.671)(−1.089)(−1.215)(−1.436)(−0.948)(−1.275)(−0.650)(−0.840)(−1.072)
N140140140135148148148148148
Adjusted R23.49%1.40%4.49%1.98%3.52%1.99%3.81%1.98%2.54%
F-statistic2.40**1.171.750.981.401.371.311.191.27
Panel B: High-yield corporate bond market returns
 R_all_R_all_R_all_R_all_
 BBtBtCCCthigh_yieldt
  1. This table presents the results of regressions of corporate bond market returns on contemporaneous aggregate earnings changes and lagged equity market returns. Panel A presents the results of the regressions for the investment-grade corporate bond market returns. Panel B presents the results of the regressions for the high-yield corporate bond market returns. R_1-3_AAA-AA, R_1-3_A-BBB, R_3-5_AAA-AA, R_3-5_A-BBB, R_5-10_AAA-AA, R_5-10_A-BBB, R_15+_AAA-AA, R_15+_A-BBB, R_all_invest_grade, R_all_BB, R_all_B, R_all_CCC, and R_all_high_yield are quarterly total returns of the various corporate bond indices. Total return is the sum of the price return, the accrued interest return, and the coupon return. The numbers in the names of the corporate bond indices represent the remaining maturities of the bonds in the indices, and the letters represent the credit ratings. “All” stands for “all available maturities.” ΔE/A is the quarterly aggregate earnings changes measured as the value-weighted average of firm-specific earnings changes. Firm-specific earnings change is the seasonally differenced income before extraordinary items scaled by lagged total assets. R_equity is the quarterly return of the equity market index. The sample extends from January 1973 through December 2010. I use ordinary least squares for the calculation of the regression coefficients and the Newey-West heteroscedasticity- and autocorrelation-consistent standard errors with four lags. t-statistics are in parentheses below the coefficient estimates. F-statistics test the null hypothesis that the regression coefficients are jointly equal to zero. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively (two-tailed test).

Intercept0.024***0.023**0.037**0.029***
 (2.765)(2.350)(2.238)(3.394)
ΔE/At2.1475.291***9.313***3.109**
 (1.464)(3.391)(4.054)(2.320)
R_equityt−1−0.080−0.139**−0.134−0.055
 (−1.493)(−2.178)(−1.312)(−0.985)
R_equityt−2−0.119−0.206*−0.461**−0.158*
 (−1.306)(−1.678)(−2.292)(−1.830)
R_equityt−3−0.076−0.150**−0.377***−0.088
 (−1.272)(−2.195)(−3.030)(−1.428)
R_equityt−4−0.056−0.065**−0.160**−0.068*
 (−1.490)(−2.405)(−2.477)(−1.710)
N56565697
Adjusted R22.57%17.38%26.13%7.62%
F-statistic0.713.93***5.20***1.49

The finding that the aggregate earnings-returns relation increases after controlling for expected aggregate earnings changes suggests that the aggregate earnings-returns relation is partially driven by a negative relation between expected aggregate earnings changes and expected returns (Sadka and Sadka [2009]). The negative relation between expected aggregate earnings changes and expected returns, however, cannot fully explain the contemporaneous negative aggregate earnings-returns relation for investment-grade bonds. Aggregate earnings changes remain significantly negatively related to investment-grade corporate bond market returns even after controlling for the expected component of aggregate earnings changes. This latter finding suggests that the earnings-returns relation at the aggregate level is also driven by the news component of aggregate earnings changes and, more specifically, by a positive relation between unexpected aggregate earnings changes and discount rate news (Kothari, Lewellen, and Warner [2006]). My results reconcile the findings of Sadka and Sadka [2009] and Kothari, Lewellen, and Warner [2006] as they suggest that both the expected and the news component of aggregate earnings changes drive the aggregate earnings-returns relation down.

Next, I focus on the information content of the unexpected component of aggregate earnings changes. Table 4 presents the results of the regressions of one-quarter-forward corporate bond market returns on aggregate earnings changes and lagged equity market returns. The regression coefficients of the aggregate earnings changes are not statistically distinguishable from zero in 12 of the 13 regressions, and the overall F-statistics are insignificant in all the 13 regressions. These findings cast doubt on the idea that aggregate earnings convey new information and are consistent with the theory put forward by Ball, Sadka, and Sadka [2009] and Sadka and Sadka [2009] that aggregate earnings changes are highly predictable. Still, the results of table 4 should be interpreted with caution because realized returns are a poor proxy for expected returns; news about cash flows and news about discount rates might not cancel out on average (Elton [1999]).

Table 4. Forward Corporate Bond Market Returns and Aggregate Earnings Changes
Panel A: Investment-grade corporate bond market returns
 R_1-3_R_1-3_R_3-5_R_3-5_R_5-10_R_5-10_R_15+_R_15+_R_all_
 AAA-A-AAA-A-AAA-A-AAA-A-invest_
 AAt+1BBBt+1AAt+1BBBt+1AAt+1BBBt+1AAt+1BBBt+1gradet+1
Intercept0.020***0.021***0.022***0.023***0.023***0.025***0.026***0.028***0.025***
 (7.433)(7.087)(6.361)(5.928)(5.683)(5.400)(4.597)(4.533)(5.102)
ΔE/At−0.516−0.672*−0.507−0.691−0.211−0.7280.271−0.604−0.732
 (−1.636)(−1.897)(−1.230)(−1.468)(−0.330)(−1.207)(0.235)(−0.652)(−1.116)
R_equityt−1−0.013−0.015−0.022−0.033−0.035−0.051*−0.063−0.067−0.053
 (−0.821)(−0.789)(−0.935)(−1.287)(−1.167)(−1.678)(−1.461)(−1.501)(−1.592)
R_equityt−20.031*0.0180.0220.0180.0330.0290.0170.0180.031
 (1.946)(1.170)(1.146)(0.842)(1.189)(1.044)(0.340)(0.415)(1.022)
R_equityt−3−0.010−0.014−0.029−0.030−0.036−0.036−0.047−0.042−0.038
 (−0.682)(−0.933)(−1.331)(−1.243)(−1.195)(−1.273)(−1.067)(−0.977)(−1.164)
R_equityt−4−0.010−0.0010.001−0.012−0.007−0.019−0.017−0.026−0.016
 (−0.514)(−0.049)(0.032)(−0.565)(−0.246)(−0.716)(−0.434)(−0.677)(−0.546)
N140140140135147147147147147
Adjusted R2−0.13%0.57%−1.04%0.51%−1.51%0.58%−1.88%−0.86%−0.15%
F-statistic1.261.200.830.930.711.190.640.700.97
Panel B: High-yield corporate bond market returns
    R_all_
 R_all_R_all_R_all_high_
 BBt+1Bt+1CCCt+1yieldt+1
  1. This table presents the results of regressions of one-quarter-forward corporate bond market returns on aggregate earnings changes and lagged equity market returns. Panel A presents the results of the regressions for the investment-grade corporate bond market returns. Panel B presents the results of the regressions for the high-yield corporate bond market returns. R_1-3_AAA-AA, R_1-3_A-BBB, R_3-5_AAA-AA, R_3-5_A-BBB, R_5-10_AAA-AA, R_5-10_A-BBB, R_15+_AAA-AA, R_15+_A-BBB, R_all_invest_grade, R_all_BB, R_all_B, R_all_CCC, and R_all_high_yield are quarterly total returns of the various corporate bond indices. Total return is the sum of the price return, the accrued interest return, and the coupon return. The numbers in the names of the corporate bond indices represent the remaining maturities of the bonds in the indices, and the letters represent the credit ratings. “All” stands for “all available maturities.” ΔE/A is the quarterly aggregate earnings changes measured as the value-weighted average of firm-specific earnings changes. Firm-specific earnings change is the seasonally differenced income before extraordinary items scaled by lagged total assets. R_equity is the quarterly return of the equity market index. The sample extends from January 1973 through December 2010. I use ordinary least squares for the calculation of the regression coefficients and the Newey-West heteroscedasticity- and autocorrelation-consistent standard errors with four lags. t-statistics are in parentheses below the coefficient estimates. F-statistics test the null hypothesis that the regression coefficients are jointly equal to zero. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively (two-tailed test).

Intercept0.022**0.021*0.038*0.030***
 (2.468)(1.967)(1.996)(3.101)
ΔE/At−0.4600.9424.3640.394
 (−0.275)(0.472)(1.536)(0.259)
R_equityt−1−0.050−0.082−0.303*−0.107
 (−0.750)(−0.827)(−1.840)(−1.498)
R_equityt−2−0.029−0.093−0.349*−0.061
 (−0.341)(−0.845)(−1.894)(−0.880)
R_equityt−3−0.042−0.065−0.205*−0.057
 (−0.991)(−1.156)(−1.923)(−1.220)
R_equityt−4−0.028−0.057−0.107−0.063
 (−0.473)(−0.746)(−0.807)(−1.141)
N56565697
Adjusted R2−3.71%−3.99%8.89%1.29%
F-statistic0.560.711.580.61

Table 5 presents the results of the regressions of quarterly excess corporate bond market returns on aggregate earnings changes and lagged equity market returns. The excess bond market returns (Excess_R_1-3_AAA-AA, Excess_R_1-3_A-BBB, Excess_R_3-5_AAA-AA, Excess_R_3-5_A-BBB, Excess_R_5-10_AAA-AA, Excess_R_5-10_A-BBB, Excess_R_15+_AAA-AA, Excess_R_15+_A-BBB, Excess_R_all_invest_grade, Excess_R_all_BB, Excess_R_all_B, Excess_R_all_CCC, and Excess_R_all_high_yield) are equal to the difference between the total returns of the corporate bond indices and the total returns of a key rate duration-matched basket of U.S. government securities. The advantage of using excess corporate bond returns to study the relation between unexpected aggregate earnings, nominal interest rates, and risk premia is that corporate bonds have finite maturities and predetermined payments. As a result, I can accurately match the duration of the corporate bonds to the duration of the risk-free assets, and isolate the component of corporate bond market returns that is related to changes in interest rates from the return component that is related to changes in risk premia.

Table 5. Excess Corporate Bond Market Returns and Aggregate Earnings Changes
Panel A: Investment-grade corporate bond market returns
 Excess_Excess_Excess_Excess_Excess_Excess_Excess_Excess_Excess_
 R_1-3_R_1-3_R_3-5_R_3-5_R_5-10_R_5-10_R_15+_R_15+_R_all_
 AAA-AAtA-BBBtAAA-AAtA-BBBtAAA-AAtA-BBBtAAA-AAtA-BBBtinvest_gradet
Intercept0.0030.0040.0030.0040.0010.0040.0040.0060.004
 (1.375)(1.049)(0.785)(0.718)(0.232)(0.532)(0.560)(0.645)(0.659)
ΔE/At0.2380.889**0.2651.464*0.5992.126*0.8072.457*1.577*
 (1.274)(2.042)(0.994)(1.919)(1.377)(1.806)(1.124)(1.789)(1.889)
R_equityt−1−0.009−0.023−0.009−0.036−0.005−0.047−0.027−0.077−0.042
 (−1.133)(−1.429)(−0.610)(−1.443)(−0.166)(−1.340)(−0.539)(−1.455)(−1.407)
R_equityt−2−0.025−0.060−0.033−0.093−0.046−0.137−0.092−0.180*−0.111*
 (−1.298)(−1.469)(−1.080)(−1.654)(−1.013)(−1.661)(−1.349)(−1.738)(−1.683)
R_equityt−3−0.019−0.039*−0.032−0.052−0.047−0.083*−0.081−0.126*−0.072*
 (−1.448)(−1.688)(−1.520)(−1.672)(−1.438)(−1.711)(−1.577)(−1.885)(−1.747)
R_equityt−4−0.011−0.021−0.018−0.025−0.020−0.036−0.039−0.061−0.036
 (−1.478)(−1.399)(−1.620)(−1.202)(−1.134)(−1.148)(−1.162)(−1.300)(−1.328)
N565656565656565656
Adjusted R20.93%4.07%−0.88%5.88%−3.37%6.58%1.21%7.90%6.32%
F-statistic0.581.350.711.000.670.980.621.071.04
Panel B: High-yield corporate bond market returns
 Excess_R_all_Excess_R_all_Excess_R_all_Excess_R_all_
 BBtBtCCCthigh_yieldt
  1. This table presents the results of regressions of excess corporate bond market returns on contemporaneous aggregate earnings changes and lagged equity market returns. Panel A presents the results of the regressions for the investment-grade corporate bond market returns. Panel B presents the results of the regressions for the high-yield corporate bond market returns. Excess_R_1-3_AAA-AA, Excess_R_1-3_A-BBB, Excess_R_3-5_AAA-AA, Excess_R_3-5_A-BBB, Excess_R_5-10_AAA-AA, Excess_R_5-10_A-BBB, Excess_R_15+_AAA-AA, Excess_R_15+_A-BBB, Excess_R_all_invest_grade, Excess_R_all_BB, Excess_R_all_B, Excess_R_all_CCC, and Excess_R_all_high_yield are quarterly excess returns of the various corporate bond indices. Excess return is the difference between the total return of the corporate bond index and the total return of a key rate duration-matched basket of government securities. The numbers in the names of the corporate bond indices represent the remaining maturities of the bonds in the indices, and the letters represent the credit ratings. “All” stands for “all available maturities.” ΔE/A is the quarterly aggregate earnings changes measured as the value-weighted average of firm-specific earnings changes. Firm-specific earnings change is the seasonally differenced income before extraordinary items scaled by lagged total assets. R_equity is the quarterly return of the equity market index. The sample extends from January 1997 through December 2010. I use ordinary least squares for the calculation of the regression coefficients and the Newey-West heteroscedasticity- and autocorrelation-consistent standard errors with four lags. t-statistics are in parentheses below the coefficient estimates. F-statistics test the null hypothesis that the regression coefficients are jointly equal to zero. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively (two-tailed test).

Intercept0.0100.0110.0280.014
 (0.945)(0.884)(1.487)(1.055)
ΔE/At4.074*6.918***10.534***6.381***
 (1.972)(3.334)(3.879)(2.929)
R_equityt−1−0.111−0.160*−0.138−0.134
 (−1.312)(−1.756)(−1.138)(−1.482)
R_equityt−2−0.206*−0.277*−0.513**−0.288*
 (−1.824)(−1.917)(−2.350)(−1.951)
R_equityt−3−0.111*−0.174**−0.385***−0.192**
 (−1.741)(−2.354)(−2.946)(−2.335)
R_equityt−4−0.072−0.069−0.153*−0.088
 (−1.239)(−1.401)(−1.883)(−1.564)
N56565656
Adjusted R29.68%19.35%24.92%17.83%
F-statistic0.963.72***4.74***2.61**

As table 5 shows, aggregate earnings changes become positively and significantly related to the investment-grade corporate bond market returns in five of the nine regressions, and remain positively and significantly related to the high-yield corporate bond market returns in all four regressions. In the case of investment-grade bonds, the slopes on aggregate earnings changes range from 0.24 to 2.46 with t-statistics between 0.99 and 2.04, and in the case of high-yield bonds, the slopes on aggregate earnings changes range from 4.07 to 10.53 with t-statistics between 1.97 and 3.88. Controlling for nominal interest rates increases the regression coefficients of the aggregate earnings changes in all 13 regressions and the increases are statistically significant at the 5% level or lower in 9 of the 13 regressions.9 In the case of investment-grade bonds, the overall F-statistics are insignificant in all nine regressions, and in the case of high-yield bonds, they are significant at the 5% level or lower in three of the four regressions. The number of available observations decreases to 56 because Bank of America Merrill Lynch provides data on excess returns only from January 1997 to the present.

The results of table 5 are in line with Kothari, Lewellen, and Warner [2006], who find a positive relation between aggregate earnings changes and changes in the one-year T-bill rate. My results provide support for the idea that unexpected aggregate earnings changes are positively related to changes in nominal interest rates (e.g., Shivakumar [2007]), but they do not address whether aggregate earnings move with nominal interest rates because they comove with real interest rates, inflation expectations, or both. Further, my results do not support the idea of a positive relation between unexpected aggregate earnings changes and changes in risk premia (e.g., Patatoukas [2013]), as the relation between aggregate earnings changes and investment-grade corporate bond market returns becomes positive after controlling for nominal interest rates.

Table 6 presents the results of the regressions of corporate bond market returns on aggregate earnings changes, contemporaneous changes in the spreads of a CDS index, and lagged stock market returns. The changes in CDS spreads (ΔCDS) are quarterly changes in the spreads of the Markit North America Investment Grade five-year CDS index downloaded from Bloomberg. I use the spreads of the five-year CDS index because the five-year issues are the most liquid issues. The number of available observations drops to 25 because the data for the CDS index spreads are available only since October 2004.

Table 6. Corporate Bond Market Returns, Aggregate Earnings Changes, and Changes in CDS Spreads
Panel A: Investment-grade corporate bond market returns
 R_1-3_R_1-3_R_3-5_R_3-5_R_5-10_R_5-10_R_15+_R_15+_R_all_
 AAA-AAtA-BBBtAAA-AAtA-BBBtAAA-AAtA-BBBtAAA-AAtA-BBBtinvest_gradet
Intercept0.012***0.013***0.015***0.015**0.016**0.017*0.023**0.021*0.017**
 (4.879)(3.198)(3.732)(2.434)(2.509)(1.973)(2.237)(2.027)(2.412)
ΔE/At−1.576**−1.341−3.763***−1.638−6.412***−2.901−12.190***−7.663**−3.989**
 (−2.357)(−1.207)(−3.101)(−1.264)(−3.045)(−1.451)(−3.494)(−2.779)(−2.291)
ΔCDSt−0.024−0.065−0.046−0.085*−0.081−0.126*−0.114*−0.163*−0.105*
 (−0.920)(−1.517)(−1.137)(−1.803)(−1.460)(−1.897)(−1.822)(−2.053)(−1.819)
R_equityt−10.0380.0190.0740.0110.1840.0550.403**0.2320.099
 (1.039)(0.474)(1.160)(0.179)(1.706)(0.567)(2.119)(1.498)(1.120)
R_equityt−20.083*0.0750.229**0.1320.415***0.2320.638***0.443**0.256**
 (1.799)(0.958)(2.756)(1.420)(3.115)(1.695)(3.454)(2.595)(2.161)
R_equityt−3−0.0110.008−0.035−0.015−0.056−0.057−0.114−0.130−0.054
 (−0.610)(0.165)(−1.019)(−0.227)(−0.968)(−0.580)(−1.075)(−1.029)(−0.689)
R_equityt−40.0280.0110.0850.0460.205**0.124*0.416***0.333***0.144**
 (0.932)(0.340)(1.717)(1.105)(2.575)(2.052)(3.057)(3.379)(2.459)
N252525252525252525
Adjusted R21.22%18.45%22.30%16.87%31.81%20.19%43.20%32.00%22.01%
F-statistic11.05***1.1124.66***0.8810.20***1.394.22***9.11***3.05**
Panel B: High-yield corporate bond market returns
 R_all_R_all_R_all_R_all_
 BBtBtCCCthigh_yieldt
  1. This table presents the results of regressions of corporate bond market returns on contemporaneous aggregate earnings changes, contemporaneous changes in spreads of a credit default swap (CDS) index, and lagged equity market returns. Panel A presents the results of the regressions for the investment-grade corporate bond market returns. Panel B presents the results of the regressions for the high-yield corporate bond market returns. R_1-3_AAA-AA, R_1-3_A-BBB, R_3-5_AAA-AA, R_3-5_A-BBB, R_5-10_AAA-AA, R_5-10_A-BBB, R_15+_AAA-AA, R_15+_A-BBB, R_all_invest_grade, R_all_BB, R_all_B, R_all_CCC, and R_all_high_yield are quarterly total returns of the various corporate bond indices. Total return is the sum of the price return, the accrued interest return, and the coupon return. The numbers in the names of the corporate bond indices represent the remaining maturities of the bonds in the indices, and the letters represent the credit ratings. “All” stands for “all available maturities.” ΔE/A is the quarterly aggregate earnings changes measured as the value-weighted average of firm-specific earnings changes. Firm-specific earnings change is the seasonally differenced income before extraordinary items scaled by lagged total assets. ΔCDS is the quarterly change in the spread of the five-year CDS index. R_equity is the quarterly return of the equity market index. The sample extends from October 2004 through December 2010. I use ordinary least squares for the calculation of the regression coefficients and the Newey-West heteroscedasticity- and autocorrelation-consistent standard errors with three lags. t-statistics are in parentheses below the coefficient estimates. F-statistics test the null hypothesis that the regression coefficients are jointly equal to zero. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively (two-tailed test).

Intercept0.021**0.0170.032*0.022*
 (2.395)(1.724)(1.946)(2.066)
ΔE/At2.3393.0543.7492.794
 (0.971)(1.142)(0.777)(0.940)
ΔCDSt−0.154**−0.212**−0.355**−0.216**
 (−2.676)(−2.530)(−2.486)(−2.559)
R_equityt−1−0.136−0.0760.078−0.065
 (−1.230)(−0.755)(0.375)(−0.533)
R_equityt−2−0.007−0.060−0.196−0.061
 (−0.051)(−0.364)(−0.706)(−0.348)
R_equityt−3−0.048−0.026−0.119−0.064
 (−0.387)(−0.211)(−0.559)(−0.453)
R_equityt−4−0.0250.0400.0190.006
 (−0.414)(0.646)(0.152)(0.089)
N25252525
Adjusted R247.43%58.27%63.38%57.28%
F-statistic1.922.125.28***2.21*

As table 6 shows, controlling for changes in CDS spreads reduces the size of the coefficients on aggregate earnings changes in 12 of the 13 regressions, and these declines are statistically different from zero in 4 of the 13 regressions at the 10% level or lower.10 For investment-grade (high-yield) bonds, the coefficients on aggregate earnings changes range from −1.34 to −12.19 (2.34 to 3.75) with t-statistics between −1.21 and −3.49 (0.78 and 1.14). The explanatory power of the models increases when adding changes in the CDS spreads as independent variable; the adjusted R2 range from 1.22% to 43.20% for the investment-grade indices and from 47.43% to 63.38% for the high-yield indices. The findings of table 6 provide evidence of a positive relation between unexpected aggregate earnings changes and cash flow news, and a negative relation between unexpected aggregate earnings changes and changes in default premia.

My results are consistent with Kothari, Lewellen, and Warner [2006], who document a negative relation between aggregate earnings changes and changes in default spreads, and they are contrary to the assertion of Bali, Demirtas, and Tehranian [2008] that the cash flow component of firm-specific earnings is diversified away with aggregation. Further, my findings are in line with Callen, Livnat, and Segal [2009], who document a negative relation between firm-specific earnings changes and firm-specific changes in CDS spreads.

To summarize, the results in tables 36 suggest that aggregate earnings changes have an expected component and a news component. The expected component of aggregate earnings changes is negatively related to expected returns, and the news component is positively related to cash flow news and to changes in nominal interest rates, and negatively related to changes in default premia. My results partially explain the finding in Kothari, Lewellen, and Warner [2006] of a negative relation between aggregate earnings changes and stock market returns. My tests, however, do not provide evidence on the relation between aggregate earnings changes and changes in equity risk premia, and so further analysis is necessary to fully understand the aggregate earnings-returns relation for stocks.

5. Untabulated Robustness Tests

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Hypotheses Development
  5. 3. Data and Summary Statistics
  6. 4. Results
  7. 5. Untabulated Robustness Tests
  8. 6. Conclusion
  9. REFERENCES

My findings are robust to several alternative specifications for the aggregate earnings changes. First, the findings are similar when I scale the firm-specific earnings changes by lagged market value of equity, lagged absolute earnings, lagged absolute book value of equity, and lagged enterprise value instead of lagged total assets. I define enterprise value as the sum of the market value of equity plus the book value of debt.

Second, my results are robust to the measurement of earnings as operating income after depreciation instead of earnings before extraordinary items. I use earnings before extraordinary items in the main analysis because operating income after depreciation does not account for taxes, nonoperating income, special items, and minority interests. Moreover, only a percentage of interest expense, which is included in operating income but is excluded from net income, accrues to corporate bondholders, whereas the rest relates to repayments of bank loans, charges relating to leases, expenses related to the issuance of debt, factoring charges, and interest expenses on deferred compensation.

Third, the findings are robust to defining aggregate earnings changes as the equal-weighted instead of the value-weighted average of firm-specific earnings changes. I use value-weighted aggregate earnings changes in the main analysis because the corporate bond indices provided by Bank of America Merrill Lynch are also value weighted.

Fourth, the results are similar when I calculate firm-specific earnings changes as firm-specific errors in analysts' forecasts instead of seasonally differenced firm-specific earnings. I define analyst forecast errors as reported earnings minus the median of analyst earnings forecasts announced at the end of the previous quarter scaled by lagged total assets. I downloaded analyst forecasts from IBES.

Fifth, the results are robust to measuring aggregate earnings changes as the seasonal difference in the sum of quarterly firm-specific earnings. More specifically, aggregate earnings changes are defined as the sum of firm-specific earnings in the current quarter minus the sum of firm-specific earnings four quarters ago. I divide the sum of firm-specific earnings each quarter by the number of available observations to control for the fact that data coverage changes through time. And sixth, my results are robust to shifting the corporate bond return window one and two months forward to include the earnings announcement date in the calculation of the quarterly returns.

My findings are also robust to measuring the aggregate bond returns using the Barclays Capital U.S. Corporate Bond Indices downloaded from Datastream. Further, the main inferences remain unchanged when I use annual instead of quarterly data, when I impose no sample selection criteria, when I match the firms of the earnings sample to the firms of the corporate bond sample, and when I only include firms with publicly traded bonds in my analysis. Following Faulkender and Petersen [2006], I define firms with publicly traded bonds as firms with credit ratings.

Moreover, the relation between aggregate earnings changes and corporate bond market returns is independent of the state of the economy. I measure the state of the economy using the recession indicator downloaded from the Federal Reserve Bank of St. Louis Economic Data database. Finally, Jorgensen, Li, and Sadka [2011] show that aggregate earnings changes are unusually low in 2001 and unusually high in 2003 because of the implementation of SFAS 142. These unusual levels of aggregate earnings changes might conceivably distort the aggregate earnings-returns relation. Nevertheless, the results of my analysis are not sensitive to the exclusion of the period 2001–2003.

6. Conclusion

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Hypotheses Development
  5. 3. Data and Summary Statistics
  6. 4. Results
  7. 5. Untabulated Robustness Tests
  8. 6. Conclusion
  9. REFERENCES

I study the relation between aggregate earnings changes and corporate bond market returns, and I find that aggregate earnings changes have a negative relation with investment-grade corporate bond market returns and a positive relation with high-yield corporate bond market returns. The aggregate earnings-returns relation is lower for bonds with higher credit ratings and longer maturities. I also examine the relation between aggregate earnings changes and the various components of corporate bond market returns. I find that expected aggregate earnings changes are negatively related to expected returns, and unexpected aggregate earnings changes are positively related to cash flow news and to changes in nominal interest rates, and negatively related to changes in default premia.

My findings contribute to two streams of literature. First, I contribute to the literature that examines the relation between accounting earnings and asset prices from the corporate bondholders' perspective (e.g., Easton, Monahan, and Vasvari [2009]). My findings suggest that aggregate earnings incorporate a different type of information from firm-specific earnings. Hence, firm-level findings on the relation between earnings and asset prices are not generalizable to the aggregate level. Further, I add to the literature that examines the relation between accounting information and the macroeconomy (e.g., Kothari, Lewellen, and Warner [2006]) in the following ways. First, I reconcile the findings in Kothari, Lewellen, and Warner [2006] and Sadka and Sadka [2009] by showing that the earnings-returns relation at the aggregate level is driven by both the expected and the news component of aggregate earnings changes. Second, I shed light on the relation between aggregate earnings news and discount rate news by showing that nominal interest rates increase and default premia decrease when aggregate earnings are higher than expected. Finally, my findings help explain the negative aggregate earnings-returns relation for stocks (e.g., Kothari, Lewellen, and Warner [2006]). My results suggest that the negative relation between aggregate earnings changes and stock market returns is partially driven by a negative relation between expected aggregate earnings changes and expected returns, and a positive relation between unexpected aggregate earnings changes and changes in nominal interest rates.

  1. 1

    Studies that examine the macroeconomic role of aggregate accounting numbers include Anilowski, Feng, and Skinner [2007] and Bonsall, Bozanic, and Fischer [2013], who examine the relation between aggregate earnings guidance and stock market returns; Hirshleifer, Hou, and Teoh [2009], who examine the relation of aggregate accruals and aggregate cash flows to stock market returns; Kang, Liu, and Qi [2010], who study the association of aggregate discretionary accruals and aggregate normal accruals with stock market returns; and Guo and Jiang [2011], who examine the relation between aggregate accruals and the conditional equity premium.

  2. 2

    Even though the return decomposition proposed by Campbell [1991] was developed for stock market returns, it can also be applied to the corporate bond markets (e.g., Campbell and Ammer [1993], Chen and Zhao [2009]).

  3. 3

    The positive autocorrelations of aggregate earnings changes for several quarters support the proposition of a positive covariance between cash flow news and unexpected aggregate earnings changes.

  4. 4

    This theory is also in line with the findings of Chen [1991], who documents a negative relation between expected growth in gross national product and stock market returns.

  5. 5

    I include all firms with data available in Compustat North America Fundamentals Quarterly—regardless of whether the firms have publicly traded bonds or not—to capture the earnings-generating ability of all the firms in the economy.

  6. 6

    The only difference in the definition of aggregate earnings changes is that Kothari, Lewellen, and Warner [2006] scale aggregate earnings changes by lagged earnings, book value of equity, or price, whereas I scale aggregate earnings changes by lagged total assets in my main analysis.

  7. 7

    I perform three additional tests. First, I examine whether the aggregate earnings-returns relation depends on the type of information incorporated into earnings. Easton, Monahan, and Vasvari [2009] show a lower firm-specific earnings-returns relation when earnings convey good news. The reason is that bonds have limited upside potential, so the cash flow effect of earnings is less pronounced when earnings news is good. I also expect a lower aggregate earnings-returns relation when earnings news is good. In line with my expectations, I find a lower relation in the case of investment-grade but not high-yield bonds. Second, I examine the relation between aggregate earnings changes and corporate bond market returns over time. Francis and Schipper [1999], Lev and Zarowin [1999], and Ryan and Zarowin [2003] document a decline in the contemporaneous firm-specific earnings-returns relation over time. My tests also provide some evidence of a weakening earnings-returns relation over time at the aggregate level. Third, I decompose aggregate earnings changes into aggregate accruals changes and aggregate cash flow changes, and examine their relation to aggregate bond returns. I find that both aggregate accruals and aggregate cash flow changes are negatively related to investment-grade corporate bond market returns. I also find that aggregate accruals changes have no relation and aggregate cash flow changes have a positive relation to high-yield corporate bond market returns. My findings are contrary to the findings of Hirshleifer, Hou, and Teoh [2009], who show that aggregate accruals changes are negatively related and aggregate cash flow changes are positively related to stock market returns.

  8. 8

    The indices in my analysis include callable bonds that are at least one year from the first call date. Even though call options are more common for long-term than short-term bonds (Kish and Livingston [1992]), it is not the differences in the optionality, but the differences in the maturity of the corporate bond indices that drive the results of table 2, panel D. The reason is that embedded call options make corporate bond prices less sensitive to discount rate changes. Therefore, the higher percentage of callable bonds in the long-term indices should weaken the discount rate effect of aggregate earnings and should ultimately have a positive impact on the aggregate earnings-returns relation.

  9. 9

    The increases in the coefficients of the aggregate earnings changes in table 5 compared to table 3 are not due to the different sample periods used in the regressions. My conclusion is the same when I use the same sample period for the regressions with excess returns and the regressions with total returns as dependent variables.

  10. 10

    The declines in the coefficients of the aggregate earnings changes in table 6 compared to table 3 are not due to the different sample periods used in the regressions. My conclusion is similar when I use the same sample period for the regressions with and without changes in CDS spreads as control variables.

REFERENCES

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Hypotheses Development
  5. 3. Data and Summary Statistics
  6. 4. Results
  7. 5. Untabulated Robustness Tests
  8. 6. Conclusion
  9. REFERENCES