Does Sell-Side Debt Research Have Investment Value?

Transcription

1 Does Sell-Side Debt Research Have Investment Value? Sunhwa Choi* Lancaster University and Sungkyunkwan University Robert Kim University of Massachusetts Boston January 2018 *Corresponding author: Lancaster University Management School, Bailrigg, Lancaster, LA14YX, UK. Tel: +44 (0)

2 Does Sell-Side Debt Research Have Investment Value? Abstract We examine whether debt research has investment value for debt investors. Specifically, we examine the event-time and post-event bond price reactions around the issuance of debt analysts recommendations and find that both the levels of and changes in recommendations are associated with the event time abnormal bond return, while the price reaction is stronger for changes in recommendations. We also find that changes in recommendations are associated with a significant post-event bond price drift, suggesting that the initial bond market reaction is incomplete. The calendar time portfolio approach shows that buying (selling) bonds following upgrades (downgrades) generates significant abnormal bond returns. We also document that debt analysts coverage of debt securities is non-random and that the information in their bond-specific recommendations is incremental to that in the firm-level recommendations. Overall, our results suggest that debt analysts reports have investment value. Keywords: debt analysts; recommendations; bond returns; market efficiency JEL Classification: G12; G14; G24; G29

3 1. Introduction Sell-side debt analysts provide forecasts of firm s financial performance and investment recommendations for firms or debt securities based on their analysis of firm fundamentals, recent corporate events, and bond price movements. The primary goal of debt research is to identify mispriced debt securities and to communicate this information to debt investors to help their investment decisions. Recent studies provide evidence that the bond market reacts to debt reports. For example, De Franco et al. (2009) show that the publication of debt analysts reports generates abnormal bond trading and price reactions, and De Franco et al. (2014) find that the negative tone of debt analysts discussion about debt-equity conflict events is associated with increases in credit spreads and bond trading volume. While these findings based on the event-period market reactions suggest that debt research is informative, it is unclear whether bond prices fully incorporate the information in debt research in a timely manner. For example, if bond prices continue to move in the directions predicted by debt analysts, debt research can provide debt investors with opportunities for profitable trading. In this paper, we examine the event-time and post-event bond price reactions around the issuance of debt analysts recommendations to test whether debt research has investment value. We focus on the levels of and changes in investment recommendations for bond securities because recommendations are derived from all available sources of information (e.g., public and private, quantitative and qualitative) and they are the primary output of analysts reports (Shipper 1991; Francis and Philbrick 1993). In addition, while other information in debt reports, such as forecast estimates or textual discussions, requires the interpretation from the user side, the recommendations are clear investment signals directly communicated to investors (Elton et al. 1986). 1

4 We hand collect investment recommendations for bond issues from 3,877 debt reports issued by sell-side debt analysts for US firms over the period July December We compute the abnormal bond return either by adjusting for returns on 30-day treasury bills (i.e., T- bill-adjusted returns) or by adjusting for returns on bonds with similar credit ratings and maturities (i.e., benchmark-adjusted returns). We first examine the bond price reactions at the event time (measured by a five-day window) to the issuance of different types of recommendations and their changes. We find significantly positive but moderate bond price reactions with 0.19% and 0.05% benchmark-adjusted returns around buy and hold recommendations, respectively. The bond price reacts negatively for sell recommendations with -0.18% benchmark-adjusted returns. The mean five-day benchmark-adjusted returns for upgrades and downgrades in recommendations are +0.65% and -0.65%, respectively, suggesting that the price reactions around changes in recommendations are greater than those associated with the level of recommendations. This is consistent with the idea that changes in recommendations contain more direct investment signals than their levels (Elton et al. 1986; Womack 1996). We also find that upgrades or downgrades in debt analysts recommendations is associated with a significant post-event bond price drift in the same direction as the initial reactions. The postevent drift following upgrades is significant but appears to be short-lived: the mean average benchmark-adjusted return for the (+6, +15) window is a significant 0.34% but insignificant thereafter. For downgrades in recommendations, the mean post-event benchmark-adjusted returns are -0.35% over the (+6, +15) window, -0.32% for the (+16, +30) window, and -0.41% for the (+31, +45) window, suggesting that the drift following downgrades continues in the subsequent period. Overall, our findings based on event-period and post-event period price reactions suggest 2

5 that the initial reactions to changes in debt analysts recommendations are incomplete and they are followed by the subsequent price drift. 1 We then examine whether debt investors can use the information in changes in debt analysts recommendations for an implementable trading strategy to generate abnormal profit. Specifically, at the end of each calendar month, we form the upgrade and downgrade portfolios based on the recent change in recommendations and estimate the buy-and-hold returns for each portfolio for the following one-, two-, three-, and six-month periods. The trading strategy of buying recently upgraded bonds and selling recently downgraded bonds generates 0.38%, 0.81%, 1.21%, and 1.39% benchmark-adjusted returns over the one-, two-, three-, and six-month holding periods, respectively. The profitability is still significant even after considering the transaction cost estimates in Edwards et al. (2007), where they report a 0.48% round-trip cost for institutional order size of $200,000. The results suggest that debt investors can exploit the slow incorporation of debt analysts recommendation changes to generate abnormal returns. We also provide new evidence that, when a firm has multiple bond issues outstanding, debt analysts often provide their recommendations only for a subset of bond issues. We find that debt analysts are more likely to provide recommendations for bonds that are recently issued, have large outstanding values, and are more frequently traded in the secondary market. We also find that the initial reactions for the bonds that are not included in the reports are weaker than those for bonds included in the reports. This finding suggests that the recommendations for the bonds included in the debt report do not necessarily carry over to other bonds that are issued by the same firm but are dropped from the report. 1 Womack (1996) and Barber et al. (2001) document a price drift after the issuance of equity analysts recommendations. They also find evidence consistent with the drift being short-lived (long-lived) for upgrades (downgrades) in equity analysts recommendations. 3

6 Finally, we document that, for about 9.3% of the reports that have two or more bonds, debt analysts provide different recommendations for the bonds issued by a firm (e.g., buy recommendation for a medium-term bond and hold recommendation for a long-term bond), reflecting different bond-specific characteristics such as maturity, seniority, and recent price movement. Our test results based on the sample of reports that include upgrades or downgrades for some bonds and reiteration for other bonds suggest that the market reaction to changes in recommendations is mostly bond-specific, although there is a weak spill over to other bonds whose recommendations are not changed. Our study makes several important contributions to the literature. First, our finding that debt analysts recommendations have investment value contributes to the small but growing body of research on debt analysts. By documenting the information content of investment recommendations in debt analysts reports, this study answers the calls from Beyer et al. (2010) and Berger (2011) for more research on debt analysts. Our use of long-window tests complements other studies based on short-window tests around the events and provides important evidence supporting the information role of debt analysts. 2 Furthermore, by focusing on debt analysts bondspecific recommendations rather than firm-level recommendations, we provide new evidence that debt analysts choose which bonds to include in their reports and that their bond-specific recommendations are sometimes different across the bonds issued by a firm. Our analyses suggest that bond-specific recommendations can provide incremental information over firm-level recommendations. 2 Whether debt analysts recommendations generate a profitable trading strategy to investors is not self-evident. Corporate bond investors are mostly institutional investors, who are less likely to naively follow analysts recommendations (De Franco et al. 2007; Mikhail et al. 2007; Malmendier and Shantikumar 2014). In addition, Gillette (2017) argues that debt analysts play a limited information role in the bond market. 4

7 Second, our results on the profitable investment strategies based on changes in debt analysts recommendations provide insight into the debates on the efficiency of debt markets. While less mispricing is expected in the debt markets than in the stock market because of the fixed cash flows of debt securities and the dominance of institutional investors in the debt markets (Chordia et al. 2016), our evidence suggests that information in debt reports is not fully reflected into bond prices in a timely fashion, creating an opportunity for profitable trading strategies. Therefore, we provide evidence for inefficiency in debt markets (Bhojraj and Swaminathan 2009; Jostova et al. 2013). 3 The next section discusses institutional background and reviews the related literature. Section 3 describes our data collection and sample. Section 4 discusses the results of our tests, and Section 5 concludes. 2. Institutional background and literature review Sell-side debt analysts, like sell-side equity analysts, work for investment banks or independent research firms. They collect and analyze information about firms and debt securities they follow and provide research reports to their clients, who are usually institutional investors. The objective of their research is to identify undervalued or overvalued debt securities and to communicate this information to their clients. Therefore, debt analysts play a distinct role from other information intermediaries, such as credit rating agencies who assess the credit quality of issuers or securities, or equity analysts who focus on the equity side. 3 Relatedly, our finding of incomplete event reactions also supplements the literature that documents investors underreaction to major announcements, such as earnings release (Bernard and Thomas 1989), dividend initiation and omissions (Michaely et al. 1995), management earnings forecast (Ng et al. 2013), and issuance of equity analysts recommendations (Womack 1996). 5

8 Debt analysts discuss firms financial performance and relevant corporate events, such as earnings announcements and M&A, and predict the likelihood of credit rating changes, bankruptcies, or covenant violations (De Franco et al. 2014; Gillette 2017). They also often provide forecasts of future performance such as earnings and cash flows. Kim et al. (2016) show that debt analysts are more likely to issue cash flow forecasts than equity analysts and that debt analysts cash flow forecasts are more accurate than those of equity analysts. Most debt reports also include bond issue specific information such as maturity, coupon rate, credit rating, transaction price, yield to maturity, and spreads, and bond-specific recommendations. The key piece of information in debt reports is debt analysts investment recommendations (i.e., buy, hold, or sell) for debt securities or firms. 4 The recommendations reflect debt analysts expectations about whether the securities are likely to outperform or underperform relative to bonds of comparable risk over an approximate 90-day horizon (De Franco et al. 2014). While debt analysts usually provide the same recommendations for the bond securities issued by a firm, they may provide different debt-issue specific recommendations, reflecting different characteristics of the bonds, such as their maturity, seniority, collateral, and recent transaction prices. 5 In addition, debt reports often include only a subset of bonds issued by a given firm. Similarly, debt analysts can revise their prior recommendations upward (i.e., upgrade) or downward (i.e., downgrade) for all bonds or for only a subset of bonds. Compared to a large amount of research on equity analysts (e.g., Bradshaw 2011), there are only a handful of studies on debt analysts in the literature. Johnston et al. (2009) examine the 4 While the recommendations by equity analysts are usually on a five-point scale (e.g., strong buy, buy, hold, sell, or strong sell), we standardize debt analysts recommendations into three categories of buy, hold, or sell (see De Franco et al. 2009, 2014). For example, if the recommendation is stated as buy, attractive, overweight, overperform, we code them as buy. If the recommendation is stated as sell, unattractive, underweight, underperform, we code them as sell. If the recommendation is stated as hold, neutral, marketweight, marketperform, core hold, we code them as hold. 5 See Appendix A for more details. 6

9 determinants of debt analyst coverage and document that debt analysts are more likely to follow firms with a higher probability of financial distress, larger debt, and higher leverage, reflecting greater demand for debt research of such firms. They also show that the equity market responds to the publication of debt reports, suggesting that the market reacts to debt research. Consistent with the informativeness of debt research, Gurun et al. (2016) find that the presence of debt reports reduces the lag between equity returns and bond returns, suggesting that debt analysts enhance the efficiency of bond markets. Closely related to our study, De Franco et al. (2009) examine the level of debt analysts recommendations and find that the issuance of debt reports, particularly those with sell recommendations, is associated with significant increases in trading volume in the bond markets, suggesting that debt reports provide information to the debt market. Their findings on the bond price reactions to the level of recommendations, however, are inconclusive. They report that T- bill-adjusted returns over the five-trading-day period for sell recommendations are not significantly different from zero, while the returns are significantly positive for buy and hold recommendations. 6 Their weak finding on the bond price reactions to the issuance of recommendations may be due to a low power of the bond return tests, or it could reflect the fact that the level of investment recommendations is less informative than changes in recommendations (Jegadeesh et al. 2004). Two recent studies focus on the qualitative discussion in debt reports. De Franco et al. (2014) analyze the tone of debt analysts discussions about debt-equity conflict events, such as M&A and share repurchases, and find that the negative tone in such discussions is associated with increases in the credit default swap (CDS) spreads and increases in offering yields. On the other 6 On the relative basis, the returns over the five-day window around sell recommendations are 20 basis points lower than the returns associated with hold recommendations. 7

10 hand, based on the reading of individual debt reports for financially distressed firms, Gillette (2017) argues that the majority of debt reports do not provide accurate predictions about firms bankruptcy or covenant violations and instead simply reiterate the information already available in other sources. Taken together, the existing research on debt analysts generally supports the idea that debt analysts are an important information intermediary in the debt market and that the bond market reacts to their reports. However, there has been little research on the investment value of debt analysts recommendations or changes in recommendations. Specifically, it is unclear whether the price reaction to debt reports is concentrated in the short event-time period only or the bond prices continue to move in the direction predicted by debt analysts for an extended period, providing debt investors with opportunities for profitable trading. This question is important because it directly speaks to the ultimate goal of debt research (i.e., identifying mispriced securities). Regarding the investment value of equity research, several studies provide empirical evidence on the profitability of equity analysts recommendations. Stickel (1995) and Womack (1996) find that the initial stock price reactions associated with changes in sell-side equity analysts recommendations are followed by a subsequent drift in stock prices up to six months. Barber et al. (2001) examine the profitability of levels of sell-side equity analysts recommendations and find that a strategy of purchasing (selling short) stocks with the most (least) favorable recommendations with daily portfolio rebalancing can generate significantly positive annual abnormal stock returns. However, they also document that the abnormal return disappears when transaction costs are accounted for. Jegadeesh et al. (2004) show that changes in equity analysts recommendations, but not the level of their recommendations, have a predictive power for future returns over the subsequent six-month period. Consistent with the weak predictability of recommendation levels, 8

11 Bradshaw (2004) and Barniv et al. (2009) report that the consensus recommendations of equity analysts do not predict one-year-ahead stock returns. Rebello and Wei (2014) examine the investment value of the research by buy-side equity analysts, who work for fund management firms to provide in-house research. They find that the excess return predicted by buy-side equity analysts has investment value for a one-year horizon. Overall, the results on equity analysts generally support that their recommendations, particularly changes in recommendations, has a predictive ability for future stock returns. For debt research to have investment value, at least three conditions should be met. First, mispricing should exist in the debt market, and secondly, debt analysts should be able to identify the mispricing. Third, the information in debt research should be slowly impounded into bond prices so that investors can act on it with some lags. Regarding the first condition of mispricing, we expect that the bond market is less likely to be subject to mispricing than the equity market because of the investor base in the bond market and the characteristics of bond valuations. Specifically, the bond market is dominated by large and sophisticated institutional investors while the equity market has diverse clienteles, including less sophisticated individual investors. In addition, because the cash flows of bonds are pre-determined, the valuation of debt securities is more straightforward than the valuation of equity. Consistent with this argument, Chordia et al. (2016) examine whether several variables that predict equity returns also have predictability in the bond market and conclude that bonds are efficiently priced, after accounting for transaction costs. In terms of the ability of debt analysts to identify mispricing, the ways that equity and debt research is funded can affect the size of available resources for each type of research and potentially affect their ability to identify mispricing. While equity research is usually funded by a hefty subsidy from soft dollar commissions, there is no explicit commission for fixed income 9

12 trading (i.e., no soft dollar subsidy to debt research). 7 Instead, broker-dealers in the debt market are compensated by the spread between bid and ask. As a result, the size of resources available to debt analysts is much smaller than those available for equity analysts. For example, a survey estimates that the spending on debt research is $1.5 billion, which is about one-fifth of the spending on equity research of $7.8 billion. 8 Therefore, to the extent that the amount of resources available to debt and equity analysts determines the informativeness of their respective research, debt research may be less likely to be informative than equity research. On the other hand, there is evidence that bond prices incorporate new information with a delay relative to equity prices (Kwan 1996; Downing et al. 2009; Gurun et al. 2016), presumably because the bond market is less liquid than the stock market. If this is the case, it might be easier for debt analysts, compared to their equity counterpart, to identify information that has not been fully incorporated into the prices. Thus, this line of reasoning provides an alternative view on the presence of mispricing and the ability of debt analysts to identify mispricing. To summarize, whether debt investors can profit from debt analysts recommendations or changes in recommendations is an empirical question, and the answer to this question directly relates to the efficiency of debt markets. 3. Data and sample We download debt analysts reports issued between July 2002 and December 2010 from Investext, a provider of full-text debt analysts reports. 9 We exclude industrial, geographic, or 7 Soft dollars refer to the payment made by investment managers to their brokerage-dealer firms for their services (e.g., research) through commission revenues from directing trades to the brokerage-dealer firms, rather than through direct payment in cash (i.e., hard dollars) Investext is now available from Thomson ONE Banker. Our sample period begins in July 2002 because the coverage of the Trade Reporting and Compliance Engine (TRACE) database begins in July

13 macroeconomic research reports, debt reports for non-us firms, close-end funds, or derivatives, and the reports issued by credit rating agencies. For each report, we manually collect information about report date, the name of the company, the name of brokerage firm who issues the report, analysts current and previous recommendations (i.e., buy, hold, or sell) for firm and bond issues, bond-specific information including coupon rate, maturity date, principal and outstanding amounts, and seniority. We delete reports that do not provide any investment recommendations or those without bond-specific information. We identify issuer-level CUSIP using company names and other firm identification information available from the report, such as exchange ticker and CUSIP. We then match each bond in the report to the bond data from Mergent-FISD based on issuer-level CUSIP, maturity date, coupon rate, and other bond-specific information available. 10 We delete convertible bonds, floating-rate bonds, and bonds without coupon rate and maturity information. We further delete bonds with maturity less than 12 months, bonds without credit rating information available from Mergent-FISD, and bonds for which we cannot identify prior recommendations. Table 1 summarizes the sample selection process for debt reports. We use 3,877 debt reports covering 7,394 bonds issued by 567 unique firms for our analyses. 11 Panel A of Table 2 presents the transition matrix of debt analysts recommendations. 36.6% of the total recommendations are buys, while sell recommendations account for only 11.0%. The 10 To match bond issues from debt reports with bond data from Mergent-FISD, we require that the bonds have the same issuer-level CUSIP, maturity date, and coupon rate. When multiple bonds from Mergent-FISD are matched to a given bond based on the criteria above, we use other bond-specific information (i.e., principal value, outstanding value, seniority) whenever it is available. Specifically, we use a scoring system in which one point is given to each of the followings: same principal value, same outstanding value, and same seniority, and then we match the bond with the highest score % of our sample reports are from 8 brokerage firms including Deutsche Bank (45.2%), Bear Stearns (19.8%), UBS (9.4%), Morgan Stanley (6.6%), Keybanc Capital (6.2%), JP Morgan (5.0%), RBC Capital Markets (4.9%), and CIBC Worked Markets (2.3%). Our sample size of 3,877 debt reports is relatively small compared to the sample size used in prior studies. For example, the number of debt reports used in Johnston et al. (2009) is 8,009 from 15 brokers during the period, and those used in De Franco et al. (2009) is 28,378 from 10 brokers for the Our correspondence with the authors of these papers and the researchers of Thomson ONE Banker indicates that a number of banks removed their reports from the database. However, we do not expect that this issue would systematically bias our empirical findings. 11

14 remaining 52.4% is hold recommendations. This positively skewed distribution of debt analysts recommendations is consistent with the finding in De Franco et al. (2009). 12 The diagonal elements of the table (87.9%) reflect the reiterations of prior recommendations. The elements above (below) the diagonal are downgrades (upgrades): 6.5% of the recommendations (i.e., 477 out of 7394) are downgrades, while 5.7% (i.e., 418 out of 7394) of them are upgrades. As presented in Panel B, the majority of the sample bonds (81.4%) are non-investment grade bonds, reflecting higher demands for debt research when the probability of financial distress is high (Johnston et al. 2009). We use Moody s credit ratings and then S&P s credit ratings when Moody s ratings are not available. Panel C reports the distribution of years to maturity. 29.7% of the bonds have a remaining time to maturity of 5 years or less, while 10.3% of them have more than 10 years to maturity. Panel D presents the number of bonds included in debt reports. A typical debt report includes 2.8 bonds among which we successfully match 1.9 bonds with the Mergent-FISD data. Interestingly, Mergent-FISD records on average 3.9 bonds that are issued by our sample firms and are outstanding at the time of debt reports. This indicates that debt analysts include only a subset of bonds that are issued by the firm. Based on the number of bonds matched to the Mergent-FISD data, 2,041 reports include one bond and 1,836 reports include two or more bonds (untabulated). Among 1,836 reports that include two or more bonds, 171 reports (9.3%) provide different recommendations for bonds issued by a firm, while the remaining 1,665 reports (90.7%) provide the same recommendations for all bonds (untabulated). 12 De Franco et al. (2009) report that the percentage of recommendations for buy, hold, and sell is 39.1%, 47.3%, and 13.6%, respectively, which is close to the distribution of recommendations reported above. Note that while De Franco collect the firm-level recommendations, we report the bond-level recommendations. 12

15 In Panel E, we report the trading activity of sample bonds in terms of the number and the amount of transactions made in the month of the report. For comparison, we also report the corresponding figures using all trades from TRACE during the sample period. 96% of the bonds in our sample have transaction data in TRACE, and the mean (median) number of transactions during the month is (47), while the corresponding number for all TRACE trades is 38.4 (10). The amount of trades (in million $) for our debt report sample is about five times greater than a typical bond in TRACE. These findings suggest that the bonds in our debt report sample are actively traded and the trade size is large. The results are similar when we use only trades over $100,000 (Bessembinder et al. 2009). 4. Empirical results and robustness checks 4.1 The event-period bond price reactions to debt analysts recommendations To examine short-window bond price reactions to debt analysts recommendations, we follow Bessembinder et al. (2009) and compute the holding period bond return over the period between day t and day t+k for a bond issue as follows: Rett, t+k =, (1) where Pt+k, and Pt is the daily price on day t+k and t, respectively. AIt (AIt+k) is the accrued interest on day t (t+k), and Ct, t+k is the coupon payment made between day t and t+k, if any. As recommended by Bessembinder et al. (2009), we use the volume-weighted average price using all trades from the TRACE Enhanced database, rather than the last price of the day, as the daily price, because the last price of the day can be subject to bid-ask bounce if the last trade size is small The enhanced version of TRACE reports uncapped transaction volume data, while the trade sizes in the standard TRACE are capped at $5 million for investment-grade bonds (i.e., reported as 5MM+) and $1 million for speculative- 13

16 We calculate the coupon payment and accrued interest using information from the Mergent-FISD database. To mitigate the effect of outliers, we delete observations in the top or bottom one percent of the distribution of the returns. We use two approaches to calculate the abnormal bond return. To adjust for the interest rate movement, we measure the T-bill-adjusted return by subtracting the return of US 30-day Treasury bill from the raw return. To control for risk related to default risk and maturity, we measure the benchmark-adjusted return by subtracting the value-weighted return of matching portfolio based on credit ratings and time to maturity. Specifically, we form 6 groups based on credit ratings (Aaa, Aa/ A/ Baa/ Ba/ B/ Caa and Ca) and 4 groups based on time to maturity (1-3 years/ 3-5 years/ 5-10 years/ over 10 years) and then calculate the value-weighted return for each of 24 portfolios. We delete observations with Moody s ratings of C or below (i.e., bonds in default). The benchmark-adjusted bond return is then computed as the difference between the raw bond return and the return of the corresponding benchmark portfolio over the same holding period. We first examine the bond return over the (-2, +2) window around the issuance of debt analysts reports (i.e., day=0). If the bond transaction price is not available on day -2 (+2), we use the daily bond price of the nearest day before day -2 but on or after day -5 (after day +2 but on or before day +5) (De Franco et al. 2009). 14 We report the results in Panel A of Table 3. The mean T-bill-adjusted return and benchmark-adjusted return around the issuance of buy recommendations are 0.20% and 0.19%, respectively, suggesting that bond market reacts positively to buy recommendations over the five-day windows. The bond price reaction for sell recommendations grade bonds (i.e., reported as 1MM+). We follow Dick-Nielsen (2014) to eliminate transactions that are known to be errors, agency transactions, and interdealer double counted transactions. 14 Therefore, our returns over the (-2, +2) window measure bond price reactions over a period of up to 11 days around the issuance date of debt reports. If we require that the bond prices are available on day -2 and day +2, we find similar results, although the sample size is reduced by 50%. 14

17 is negative and significant and its magnitudes of the returns are similar to those for buy recommendations. Hold recommendations are accompanied with smaller market reactions that are insignificant for T-bill-adjusted returns and marginally significant at the 10% level for benchmarkadjusted returns. Changes in recommendations are associated with stronger market reactions around the events. The average of T-bill-adjusted return for the five-day window around upgrades in debt analysts recommendations is 0.81% and the corresponding mean return for downgrades is -0.83%. The mean benchmark-adjusted return is 0.65% for upgrades and -0.65% for downgrades. The bond price reactions to upgrades or downgrades are all significantly different from zero at the 1% level. Reiterations of recommendations are on average associated with positive market reactions (0.07% T-bill-adjusted returns and 0.09% benchmark-adjusted returns) but their magnitude is much smaller in absolute value compared to the reactions to upgrades or downgrades in recommendations. When the level of recommendations is conditioned by whether it is an upgrade, downgrade, or reiteration, the bond market reactions are stronger for upgrades and downgrades than for reiterations across all levels of recommendations. For example, the magnitudes of bond price reactions over the five-day window are particularly great for Hold-Upgrade (i.e., upgrades to hold) (0.90% for benchmark-adjusted return) and for Sell-Downgrade (i.e., downgrades to sell) (-1.17% for benchmark-adjusted return). The magnitudes of mean returns around Buy-Reiterate, Hold-Reiterate, and Sell-Reiterate are relatively small (0.15%, 0.05%, and 0.07%, respectively, for benchmark-adjusted returns). These findings suggest that changes in debt analysts recommendations (i.e., upgrades/ downgrades) convey relevant information to the bond market than the levels of recommendations We also examine the bond returns over the (0, +2) period and find similar results. 15

18 For comparison, in Panel B, we present the bond return over the five-day window around changes in credit ratings. To facilitate the comparison to the results in Panel A, we identify Moody s rating upgrades or downgrades for the bonds in our debt analyst sample during the sample period. Consistent with the finding in the prior study (May 2010), the bond prices react strongly to changes in credit ratings. The benchmark-adjusted return is 0.26% for upgrades in credit ratings and -0.84% for downgrades in credit ratings. When the results in Panel A and B are compared, the absolute bond returns around upgrades in debt analysts recommendations (0.81% for T-billadjusted return; 0.65% for benchmark-adjusted return) are about twice as large as those around the credit rating upgrades (0.49% for T-bill-adjusted return; 0.26% for benchmark-adjusted return). The magnitude of bond market reactions to downgrades of debt analysts recommendations is similar to those associated with downgrades of credit ratings. Overall, the results in Table 3 confirm findings in prior studies (De Franco et al. 2009) that debt analysts recommendations are informative. More importantly, we provide further evidence that the market reactions are greater for changes in debt analysts recommendations than for the level of recommendations. 4.2 Post-event drift While the preceding analysis of the event-period return suggests that debt analysts recommendations have valuable information content as evidenced by immediate bond price reactions, it does not provide evidence on whether the initial market reactions are quick and complete. To test whether the information in debt analysts recommendations is fully incorporated into bond prices over the event period, we examine post-event bond returns after the issuance of debt reports. Non-zero price movement in the post-event period would indicate mean reverting (i.e., initial overreaction) or price drifts (i.e., initial underreaction). 16

19 Specifically, we examine the bond return over the 10-trading day period starting from day +6 from the issuance of debt reports (i.e., (+6, +15)). If the bond transaction price is not available on day +6 (+15), we use the daily bond price of the nearest day after day +6 but on or before day +9 (after day +15 but on or before day +18). 16 In Panel A of Table 4, we report the T-bill-adjusted and benchmark-adjusted bond returns over the (+6, +15) window. While all of buy, hold, and sell recommendations are followed by significantly positive T-bill-adjusted returns over the (+6, +15) window, corresponding benchmark-adjusted returns are insignificant. The findings suggest that there is no systematic price reversals or drift associated with the level of debt analysts recommendations, after controlling for risk associated with credit quality and maturity. In contrast, the average post-event T-bill-adjusted (benchmark-adjusted) return after upgrades in debt analysts recommendations is a significant 0.66% (0.34%). While the average T-bill-adjusted bond return after downgrades (0.03%) is not significantly different from zero, the average benchmark-adjusted return is a significantly negative 0.35%. These findings suggest that after the issuance of upgrades or downgrades in debt analysts recommendations, bond prices continue to move in the same direction as the initial reactions. Reiterations of prior recommendations are not associated with significant post-event benchmarkadjusted returns. In terms of economic significance, the benchmark-adjusted returns for the (+6, +15) window following upgrades and downgrades is 52% and 54% of the five-day event time returns around the issuance of upgrades and downgrades, respectively. Overall, the results from the post-event returns after changes in recommendations indicate that upgrades or downgrades are followed by the subsequent price drift. 16 Note that our extended period to calculate post-event bond returns does not include the extended event-period that we use (i.e., -5, +5) to calculate the five-day event-period bond returns. 17

20 When both the level of and changes in recommendations are considered in the lower part of the table, Buy-Upgrade, Hold-Upgrade, and Sell-Downgrade are followed by the subsequent price drift. There is no significant price drift after reiterations, except for reiteration of sell recommendations. This result confirms our findings that changes in debt analysts recommendations are not fully incorporated in bond prices over the short event period and are followed by bond price changes in the same direction as in the short-term windows. In Panel B, we further examine the bond returns over the (+16, +30) and (+31, +45) windows to examine whether the price drift after upgrades or downgrades in debt analysts recommendations continues over longer intervals. For brevity, we only present the results based on benchmark-adjusted returns, because they better control for risk than T-bill-adjusted returns. The mean bond returns for the (+16, +30) and (+31, +45) windows after upgrades are not significantly different from zero, suggesting that the drift following upgrades is short-lived. However, the bond returns after downgrades in debt analysts recommendations remain significantly negative over the (+16, +30) and (+31, +45) windows, suggesting that it takes longer for downgrades to be fully incorporated into bond prices. For comparison, in Panel C, we present the bond returns over the (+6, +15) window after changes in credit ratings. While the mean T-bill-adjusted return is positive after both upgrades and downgrades in credit ratings, results based on benchmark-adjusted returns suggest that no significant price drift exists for credit rating changes. 4.3 Portfolio approach Given the evidence of price drifts documented in Section 4.2, we examine whether investors can use changes in debt analysts recommendations to construct a profitable trading strategy. We focus on upgrades and downgrades because the results in Table 4 suggest that there 18

21 is no price drift associated with the level of recommendations. To take an investor-oriented perspective, we adopt a calendar time portfolio approach rather than even-time approach because it is difficult for investors to expect when debt reports will be issued. We use a buy-and-hold strategy because an active rebalancing strategy such as daily rebalancing is not feasible and involves excessive transaction costs for bond trading (Barber et al. 2001; Jegadeesh et al. 2004). Specifically, at the end of each month, a bond is placed into an upgrade portfolio (downgrade portfolio) if any of the outstanding recommendations is upgrades (downgrades). Recommendations are assumed to be outstanding (i.e., effective) for the next three months from the issuance or until the next debt report by the same brokerage firms is issued for the same covered firm, whichever earlier. 17 Then the portfolios are held over one, two, three, and six calendar months. To calculate the return of each portfolio over the holding periods, we use the volume-weighted transaction price on the last trading day of the month and calculate the monthly returns for each bond, after taking account of accrued interest and coupon payments during the month. By compounding monthly returns, we calculate the buy-and-hold returns for each bond over the different horizons. Then the portfolio return is calculated as the value weighted returns using the outstanding amount of each bond as weight. This procedure is repeated every month during the period from July 2002 to December 2010 (i.e., 102 months). Table 5 reports the mean of portfolio returns from this monthly strategy and the t-values based on the distribution of monthly returns. When the T-bill-adjusted return is used, the portfolio returns for upgrades are all positive and significant over one-, two-, three-, and six-month holding periods, while the portfolio returns for downgrades are also positive and significant for all holding 17 Because our approach uses recommendations issued up to three months earlier, we expect that this portfolio approach would yield conservative estimates of portfolio returns, compared to a strategy of using only recommendations that are very recently issued (e.g., within a month). 19

22 periods. The differences in bond returns between the upgrade and downgrade portfolios are positive for all holding periods and it is significant for two-, three-, and six-month holding periods. When benchmark-adjusted returns are used, the portfolio returns for upgrades are not significantly different from zero, consistent with the short-lived price drift for upgrades reported in Table 4. However, the portfolio returns for downgrades are significant and negative for two-, three-, and six-month holding periods. As a result, the differences in bond returns between the two portfolios are positive for two-, three-, and six-month holding periods. For example, buying (selling) bonds that experienced recent upgrades (downgrades) in recommendations by debt analysts generates 0.81%, 1.21%, and 1.39% over two-, three-, and six-month holding periods, even after controlling for risk of bonds with similar credit ratings and time to maturity. This result suggests that changes in debt analysts recommendations have a predictive value and this can be used to generate abnormal returns in the bond market. For comparison purposes, we also form portfolios based on changes in Moody s credit ratings and estimate the portfolio returns for upgrades, downgrades, and the difference between the two groups. In Panel B, the portfolio returns for upgrades in credit ratings are significantly negative for all holding periods, while the portfolio returns for downgrades are insignificant. As a result, the differences between the two portfolios are not statistically different from zero. This finding suggests that the predictability of changes in recommendations exists only for debt analysts report, but not for credit ratings. We do not directly examine whether the return from the portfolio approach discussed above is still profitable after transaction costs because transaction costs vary depending on trade sizes and bond characteristics, such as credit quality. Edwards et al. (2007) estimate a mean round-trip transaction cost for a trade of $200,000 is 0.48%, and it decreases to 0.18% for a trade of $1 million, 20

23 a median trade size for institutional-sized trades (i.e., more than $100,000). Given that our portfolio strategy involves one round-trip transaction, it appears that the abnormal return from this approach is economically significant even after transaction costs, at least for institutional investors with large trades sizes. 4.4 The implication of bond-specific recommendations The choice of bonds to include in the reports As we briefly discuss in Section 3, debt analysts often include only a subset of bonds issued by a firm in their reports. For example, even when a firm has five bonds outstanding, a debt analyst can provide recommendations only for three bonds. In this section, we examine the implications of this choice of bonds to include in the reports. To be included in our sample, we require that debt analysts provide bond-specific information and the bond is matched with the Mergent-FISD data by our matching process (N=7,394 bonds). We note that Mergent-FISD has another 7,624 outstanding bonds issued by the sample firms, for which debt analysts do not provide recommendations, or for which we cannot match with the Mergent-FISD data. Our untabulated analysis suggests that the bonds for which debt analysts provide recommendations (i.e., covered bonds) are more likely to be recently issued, to have larger outstanding values, and to be more actively traded in the secondary market, than bonds for which debt analysts do not provide recommendations (i.e., uncovered bonds). 18 Thus, in addition to the explicit recommendations for bonds in debt reports, debt analysts choice to include a given bond issue in the reports may provide incremental information. For example, the recommendations given to the bonds in the report may not carry over to other outstanding bonds that are issued by the same firm but are not included in the report. To examine this possibility, we 18 Debt analysts may have greater incentives to issue reports for these bonds in which the information demand is likely to be higher (Johnston et al. 2009). 21

24 repeat our event-time and post-event analyses for uncovered bonds (i.e., bonds that are issued by the sample firms but are not included in the bond reports). Because there is no explicit recommendations for these uncovered bonds, we use the average of recommendations for the covered bonds as the recommendations for uncovered bonds. Specifically, with a scoring system in which buy=3, hold=2, and sell=1, if the average of the recommendations of covered bonds in the report is greater (less) than 2, we classify the uncovered bonds as buy (sell) recommendations, otherwise we classify them as hold recommendations. Similarly, if the average of the current recommendations for covered bonds is greater (less) than the average of the previous recommendations for covered bonds, we code them as upgrades (downgrades) for uncovered bonds. In Table 6, we report benchmark-adjusted event-time returns and post-event returns for uncovered bonds in Panel A and the portfolio approach results in Panel B. In Panel A, we find an insignificant five-day period return around the issuance of buy recommendations, while the initial return for sell recommendations is significantly negative at -0.28%. The mean event-time return for the (-2, +2) window around upgrades is not statistically different from zero, and the corresponding return around downgrades is a significant -0.63%. Note that when we use covered bonds in Table 3, the event-time returns for buy recommendations and upgrades are all significantly positive. The results for the (+6, +15) window are similar to those reported in Table 3: the price drift exists following upgrades or downgrades, but not following the levels of recommendations or reiterations of prior recommendations. In Panel B, the portfolio approach based on uncovered bonds generally does not yield significant returns. For one-, two-, and threemonth holding periods, benchmark-adjusted returns for the upgrade and downgrade portfolios, and the differences between the two portfolios are all insignificant. For six-month holding period, the 22

25 mean benchmark-adjusted return for the downgrade portfolio is -1.60% and the difference between the downgrade and upgrade portfolios is 1.35%, which is marginally significant at the 10% level. Overall, the analyses in Table 6 for uncovered bonds generally suggest that debt analysts bondspecific recommendations do not necessary hold for other bonds that are issued by the same firm but are not included in the reports. Different recommendations for bonds As our main analyses focus on bond-specific recommendations, one might wonder whether bond-specific recommendations really matter given that the bonds are issued by the same firm. As discussed in Section 3, among our sample, 171 debt reports provide different recommendations for bonds issued by a firm. The cited reasons for issuing different recommendations include different maturity, collateral, and seniority. Given this practice of providing different recommendations, we test whether the bond market reactions are consistent with the bond-specific recommendations. For example, assume a debt analyst upgrades her recommendation for one bond but maintains prior recommendations for other bonds of the firm. If the bond market interprets this as good news for all bonds of the firm, the bond prices of all bonds will react positively. On the other hand, if the bond market views that this upgrade is specific to only one bond, the price of upgraded bond will increase but the price of other bonds will not change. Specifically, among 171 debt reports containing different bond-specific recommendations, we identify reports with upgrades or downgrades of at least one bond. Then for a given report, we identify (i) bonds that are upgraded or downgraded and (ii) bonds whose recommendations are not changed in the report. We examine the event-time bond price reaction over the (-2, +2) window for these two groups separately. If the upgrade or downgrade of one bond carries over to other bonds whose recommendations are not changed, we expect that group (ii) will experience similar market 23

26 reaction to group (i). On the other hand, if the upgrade or downgrade of one bond is relevant only for the bond in question and does not carry over to other bonds, we expect no significant market reaction for group (ii). Panel B of Table 7 reports the results. For the bonds whose recommendations are upgraded, the mean benchmark-adjusted return is 1.13%, and for the bonds whose recommendations are downgraded, the mean return is -1.67%. For the other bonds whose recommendations are not changed, the bond price reactions are generally weak. Specifically, the mean reaction for upgrades is not different from zero, and the mean reaction for downgrades is significant -0.93% but its magnitude is smaller than the reaction to downgrades of the first group that are downgraded (i.e., -1.67%). Overall, the analyses in this section suggest that debt analysts bond-specific recommendations convey meaningful news for the specific bonds and that they partly (but not fully) spill over to other bonds in the same reports. 5. Summary and conclusions This study explores whether debt investors can profit from recommendations provided by sell-side debt analysts. The main objective of debt research is to identify overvalued/undervalued debt securities and help investors to make better investment decisions. Our sample includes 3,877 debt reports covering 7,394 debt securities and 567 unique firms for the period between July 2002 and December The results are summarized as follows. First, we find that the five-day abnormal bond return around the issuance of debt report is positive (negative) for buy (sell) recommendations and upgrades (downgrades) in recommendations. Further, the bond market reactions to upgrades (downgrades) are greater than those associated with the level of recommendations, indicating that the changes in 24

27 recommendations contain more relevant information than the level of recommendations. Second, the post-recommendation analyses reveal that the bond prices following upgrades and downgrades continue to move in the direction in line with the initial reactions, suggesting that the information in changes in recommendations is not fully incorporated into bond prices over the event window. While the drift following upgrades is short-lived (i.e., up to +15 days), the drift subsequent to downgrades persists over a longer horizon, i.e., up to +45 days. Third, our investor-oriented calendar time portfolio approach shows that buying (selling) bonds that received the upgrade (downgrade) signals generates 0.81%, 1.21%, and 1.39% abnormal bond returns over the two-, three-, and six-month holding periods. These magnitudes of abnormal bond returns suggest a profitable trading strategy even after considering the transaction cost estimates documented in Edwards et al. (2007). Finally, we find that debt-specific recommendations issued by debt analysts provide information, distinct from the firm-level recommendations, to debt investors. Collectively, our evidence is consistent with debt analysts role in providing investment value to the debt investors. We contribute to the literature on sell-side debt analysts. Our results of debt analysts recommendations providing profitable trading opportunities to debt investors support the view that debt analysts play an important informational role in the corporate bond market. In addition, we extend the literature by documenting evidence that debt analysts coverage of debt securities is non-random and that the information in their bond-specific recommendations is incremental to that in the firm-level recommendations. Our paper is also related to the literature on the debt market efficiency. Specifically, our findings that bond prices do not incorporate the information in changes in debt analysts recommendations in a timely manner provide insight into the debates on efficiency in the debt market. 25

28 Appendix A. Sample of debt analysts reports: Inconsistent recommendations within a report Example 1: We are initiating coverage of Petrohawk Energy ( HK ) with a BUY rating on its 7.125% Snr Notes due 2012 and 9.125% Snr Notes due 2013 and with a SELL rating on its 10.5% Snr Notes due 2014 and 7.875% Snr Notes due 2015 We would recommend owning shorter dated HK bonds. We believe investors could participate in credit improvement from a possible takeover or benefit from rolling down the curve as maturities near... Further out the curve, we prefer owning Quicksilver bonds over Petrohawk bonds. Over the next year, we believe Quicksilver will have less variability and benefit from its oilier assets.. (excerpt from O Connor and Weil (2010), Deutsche Bank). Example 2: We are changing our credit recommendation to Neutral from Overweight on AIG..We would be better sellers of the credit in general at these levels, and prefer up-incapital-structure trades to senior issues with higher potential recovery rates. (excerpt from Frost (2009), HSBC). 26