Do bond rating changes affect the information asymmetry of stock trading?

Journal of Empirical Finance 18 (2011) 103–116

Contents lists available at ScienceDirect

Journal of Empirical Finance j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / e c o n b a s e

Do bond rating changes affect the information asymmetry of stock trading? Yan He a, Junbo Wang b,1, K.C. John Wei c,⁎ a b c

School of Business, Indiana University Southeast, New Albany, IN, United States Department of Economics and Finance, City University of Hong Kong, Tat Chee Avenue, Kowloon, Hong Kong Department of Finance, Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong

a r t i c l e

i n f o

Article history: Received 4 November 2009 Received in revised form 28 May 2010 Accepted 1 June 2010 Available online 9 June 2010

JEL classification: G10

a b s t r a c t Using a sample of 279 upgrades and 310 downgrades from 1996 to 2004, we find that bond rating changes affect the information asymmetry of stock trading and other measures of information risk. Specifically, when a firm's bond rating is upgraded, its stock information asymmetry and its analysts' earnings forecast dispersion are significantly reduced, while the institutional equity holdings of its shares are significantly increased. The reverse is true for a downgrade. In addition, the degree of change in stock information asymmetry is positively associated with the magnitude of the bond rating changes. © 2010 Elsevier B.V. All rights reserved.

Keywords: Credit rating changes Information asymmetry Probability of information-based trades (PIN)

1. Introduction The credit crisis that originated from the U.S. sub-prime and real estate loans during 2007–2009 has been causing a lot of attention and has exerted a significant and negative impact on financial markets. As a result, the U.S. Federal Reserve has moved to tighten the rules on mortgage lending, seeking to promote greater transparency in loan terms and to curtail excessively risky lending practices. Similar to the real estate and sub-prime loans, a company's debt should also be continuously monitored and evaluated for its default probability. The current literature suggests that the uncertainty of debt creditworthiness may affect both the debt and stock values of the firm. Ederington and Goh (1998) study the relation between bond ratings (reported by bond rating agencies) and future earnings forecasts (reported by stock analysts). They find that the Granger causality flows both ways. Most bond downgrades are preceded by declines in forecast earnings, and forecast earnings tend to fall following downgrades, and vice versa for upgrades. Odders-White and Ready (2006) examine the relation between equity-market adverse selection measures and debt-market credit ratings, and point out that the private shocks of a firm lead to a linkage between the adverse selection of its equity trading and the credit rating of its debt. Therefore, the connection between debt value uncertainty and stock value uncertainty does exist, and it goes both ways, leading to two lines of studies. One line focuses on the impact of stock uncertainty measures on debt, and the other line focuses on the impact of debt uncertainty measures on stock. The first line of studies has demonstrated that a firm's stock uncertainty measures (such as information asymmetry, bid–ask spread, earnings forecast dispersion, institutional ownership, and disclosure) affect its debt.2 Odders-White and Ready (2006) report that firms with a greater risk of private shocks, and therefore higher levels of equity-market adverse selection, tend to have

⁎ Corresponding author. Tel.: +852 2358 7676; fax: +852 2358 1749. E-mail addresses: [email protected] (Y. He), [email protected] (J. Wang), [email protected] (K.C.J. Wei). 1 Tel.: +852 2788 7243; fax: +852 2788 8806. 2 A firm's stock uncertainty measures also affect its cost of capital. See, for example, Easley et al. (2002), Easley and O'Hara (2004), Coles et al. (1995), Botosan (1997), Botosan and Plumlee (2002), Sengupta (1998), and Yu (2005). 0927-5398/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jempfin.2010.06.001

104

Y. He et al. / Journal of Empirical Finance 18 (2011) 103–116

lower credit ratings. In addition, periods with higher adverse selection are likely to be followed by ratings downgrades in the near future. Agarwal and O'Hara (2007) show that firms with higher information risk (measured by the probability of informationbased trades (PIN) and bid–ask spread) are more likely to have higher leverage. Mansi et al. (2005) point out that a wider analyst dispersion of earnings forecasts signals a higher degree of uncertainty regarding future cash flows. They find that firms with higher earnings forecast dispersion have lower credit ratings and higher credit spreads. Bhojraj and Sengupta (2003) argue that governance mechanisms can reduce agency risk, information risk, and default risk. They show that firms with greater institutional ownership and stronger outside control of the board enjoy lower bond yields and higher ratings on their new bond issues. Wang and Zhang (2006) document that increases in institutional equity holdings lead to decreases in credit spreads, and the relation is strongest for bonds with shorter maturities, lower ratings, higher leverage, and higher equity return volatilities. Sengupta (1998) and Yu (2005) show that companies with lower accounting disclosure scores have poorer credit ratings and higher credit spreads. The second line of studies has shown that a firm's debt uncertainty measures (such as credit ratings) affect its stock.3 According to Ederington and Yawitz (1987), credit rating agencies evaluate publicly traded companies and communicate their findings and opinions to investors. The agencies claim to receive inside information unavailable even to stock analysts such as the minutes of board meetings, profit breakdowns by products, and new product plans. In the U.S. markets, empirical studies generally provide evidence supporting such information asymmetry and signaling hypotheses. Negative effects on abnormal stock returns are reported in the case of bond rating downgrades, but no such effects are reported for upgrades (see, for instance, Holthausen and Leftwich (1986), Cornell et al. (1989), Hand et al. (1992), and Dichev and Piotroski (2001)).4 Thus, bond rating changes may provide information not already incorporated in security prices. Other than the effect on stock prices and returns, credit ratings also influence the scale of underpricing during equity issuance. Liu and Malatesta (2005) find that in seasoned equity offerings, firms with credit ratings are underpriced to a lesser extent and they have reduced negative announcement-period abnormal stock returns than those without credit ratings. Additionally, firms with better credit ratings also have reduced negative announcementperiod abnormal stock returns than firms with worse credit ratings. The results appear to suggest that the better the bond rating levels, the lower the level of informational asymmetry. Along the second line of studies, it seems that bond upgrades and downgrades exert different effects on stocks. Such a difference between good and bad news can be explained by the discretionary disclosure hypothesis; that is, managers have some discretion over the disclosure of information, and they prefer to announce good news immediately, while allowing bad news to dribble out slowly (see, for example, Chen et al. (2001) and Bae et al. (2006)).5 Hence, good news is associated with greater disclosure and reduced information asymmetry between insiders and outside investors; and bad news is associated with reduced disclosure and greater information asymmetry.6 The models proposed by Merton (1987) and Fishman and Hagerty (1989) imply that greater disclosure reduces the costs associated with processing and assimilating public information, and leads to more trading by otherwise uninformed liquidity traders. Brown et al. (2004) document that disclosure is negatively associated with information asymmetry. They find that firms that conduct conference calls enhance their disclosure and reduce their information asymmetry of stock trading. Ajinkya et al. (1999) find that financial analysts' ratings of overall corporate disclosure practices are positively related to institutional stock ownership and the proportion of the board that is composed of outsiders. Healy et al. (1999) report that sustained increases in disclosure ratings result in higher levels of institutional ownership. Following the second line of studies, we investigate the effects of debt-market credit rating changes on the equity-market information asymmetry. Our sample contains 279 upgrades and 310 downgrades from 1996 to 2004. Our contribution to the literature mainly lies in providing evidence to support both the signaling and the discretionary disclosure hypotheses. First, based on the signaling hypothesis, credit ratings reveal the private information of firms, resulting in the link between debt and stock value uncertainties. Specifically, on the one hand, a better bond rating or a bond rating upgrade reflects lower debt value uncertainty. On the other hand, a lower level of information asymmetry reflects lower stock value uncertainty. Thus, the relation between debt credit ratings and stock information asymmetry reveals the link between debt and stock value uncertainties. Second, as suggested by the discretionary disclosure hypothesis, good and bad news are revealed in different ways, and therefore bond upgrades and downgrades exert different effects on stocks. More specifically, a better bond rating or a bond rating upgrade signals good news about a firm's financial situation, and such good news is usually released quickly, giving rise to greater disclosure and reduced information asymmetry. In contrast, a worse bond rating or a bond rating downgrade signals bad news about a firm's financial situation, and such bad news is usually released slowly, giving rise to reduced disclosure and greater information asymmetry.

3 A firm's credit ratings also affect its debt in terms of debt price, return, yield, and leverage. See, for example, Katz (1974), Grier and Katz (1976), Ingram et al. (1983), Ederington et al. (1987), Hand et al. (1992), Hite and Warga (1997), Kliger and Sarig (2000), and Kisgen (2006). 4 In addition, Goh and Ederington (1993) point out that not all downgrades are bad news for equity holders. Bond downgrades associated with deteriorating financial prospects convey new negative information to the stock market and are bad news for equity holders, but downgrades due to changes in a firm's leverage (i.e., the transfer of wealth from bondholders to stockholders) are not necessarily bad news for equity holders. Abad-Romero and Robles-Fernandez (2006) document significantly negative excess returns for upgraded firms but no significantly excess returns for downgraded firms in the Spanish market, which suggests a wealth transfer from bondholders to shareholders. 5 Chen et al. (2001) report that the returns of small stocks are more positively skewed than the returns of large stocks. Bae et al. (2006) find that stock returns in emerging markets are more positively skewed than those in developed markets. These authors argue that their findings are consistent with the discretionary disclosure hypothesis. 6 Investors do understand this and are compensated accordingly. Empirical evidence has shown that disclosure, as a measure of stock uncertainty, affects a firm's cost of capital. For example, Botosan (1997) and Botosan and Plumlee (2002) find that greater accounting disclosure lowers the cost of equity capital. Sengupta (1998) and Yu (2005) document that greater disclosure lowers the cost of debt capital.

Y. He et al. / Journal of Empirical Finance 18 (2011) 103–116

105

Our findings verify both the signaling and the discretionary disclosure explanations. First, our results show that the level of debt-market credit ratings is related to the level of equity-market information asymmetry; that is, the better the bond rating, the lower the information asymmetry, and vice versa. Second, when a firm's bond rating is upgraded, the information asymmetry of its stock trading is significantly reduced. The reverse is true for a downgrade. Furthermore, the magnitude of change in asymmetric information is positively associated with the magnitude of bond rating changes. Third, bond rating changes also affect other measures of information risk, such as quoted and effective bid–ask spreads, the dispersion of analysts' earnings forecasts, and institutional equity holdings. When a firm's bond rating is upgraded, the bid–ask spreads and the dispersion of analysts' earnings forecasts tend to decrease, and the institutional equity holdings tend to increase, implying lower information risk, and vice versa for a downgrade. Overall, the results confirm the significant relation between credit ratings and information risk measures as well as the different effects of good and bad news on information risk measures. The remainder of this paper is organized as follows. Section 2 describes the measures of information risk, including the information asymmetry measure based on trades and several other measures. Section 3 discusses the data and sample selection. Section 4 presents empirical results on bond rating changes and the information asymmetry measure. Section 5 reports robustness checks on other measures of information risk. Finally, Section 6 concludes the paper. 2. Measures of information risk 2.1. The information asymmetry measure The traditional asset-pricing models with symmetric information assume that prices are always fully revealing. In contrast, the microstructure models explicitly account for the process of price discovery. That is, the microstructure models study how private information is incorporated into prices through trading. Information asymmetry manifests itself when investors trade on the basis of their private information. While it is not possible to identify which trades are based on private information, the presence of privately informed traders in the market can be inferred from large imbalances between the number of buy orders and the number of sell orders. This operability provides the intuition behind the microstructure model developed by Easley et al. (1997a,b), called the EKOH model of information asymmetry. The EKOH model is a learning model in which the market maker draws inferences about the presence and the type of private information-based on the observed order flow. Over a trading day, prices converge to their full information levels as private information is fully revealed through the trading activities of informed investors. Thus, one can estimate the probability of information-based trades (PIN) for a given stock over a particular period based on the daily order flow during the period. We employ the EKOH model to estimate the PIN. This parsimonious structural model has been shown to be very effective in capturing the process of a market maker's learning about private information from trades. The model has been successfully implemented in previous studies to address many important microstructure issues.7 As demonstrated in these studies, the great advantage of this model is that its parameters can be easily estimated from the trade data by using a maximum likelihood method. We briefly describe the structure of the EKOH model below. The model assumes that, at the beginning of each day, there is a probability, α, for the arrival of new information. This new information signals the value of the underlying asset that is being traded. Good news and bad news occur with the probabilities of 1 − δ and δ, respectively. On each trading day, traders arrive according to independent Poisson processes throughout the day. The market maker sets prices as traders arrive, conditional on the information at the time of trade. Orders from risk neutral and competitive informed traders arrive randomly at the daily arrival rate of μ only on good- and bad-news event days. Orders from uninformed buyers and sellers arrive randomly at the daily rates of εb and εs, respectively, on every trading day. Informed traders buy if they have learned of good news and sell if they have learned of bad news. All of the arrival processes are assumed to be independent and their parameters are common knowledge across all traders and the market maker. The trading process described above leads to one of the three general patterns (which are similar to branches in a decision tree) of trade orders. On a no-news day, the model predicts roughly equal numbers of buyer- and seller-initiated trade orders. On a good news day, there will be a large imbalance in the order flow with buyer-initiated trades predominating. On a bad news day, there will be more seller-initiated trades. Although unaware of the branch chosen by nature on any given day, the market maker knows the probability of each branch and the expected order process associated with each branch. He/she uses the observed numbers of buys and sells to update his/her beliefs throughout the trading day via the Bayes rule. The structural parameters of the EKOH model can be easily estimated using trade data for each individual stock. EKOH show that the likelihood function of the model for a single trading day for a particular stock, k (for simplicity, k is omitted), can be written as: LðθjB; SÞ = ð1−αÞe−εb

εBb −εs εSs −ðμ e + αð1−δÞe B! S!

+ εb Þ

B ðμ + εb ÞB −εs εSs −ε ε −ðμ e + αδe b b e B! S! B!

+ εs Þ

ðμ + εs ÞS ; S!

ð1Þ

where B and S are the total numbers of buy and sell orders for that day, respectively. θ = (α, μ, εb, εs, δ) is the vector of model parameters described above. The likelihood equation shows that buys and sells arrive according to independent Poisson distributions. When B (or S) is large, the direct computation of this likelihood function may result in numerical overflow since εBb 7 For example, the PIN methodology has been employed to investigate how informed trading varies across different stock exchanges (Easley et al., 1996a,b) and across different types of securities (Easley et al., 1998a,b) and to examine how stock splits affect informed and uninformed trading (Easley et al., 2001).

106

Y. He et al. / Journal of Empirical Finance 18 (2011) 103–116

B e−εb εBb as e[− εb + B ln(εb) − ∑i = 1 ln(i)]. We also and B! (or, εSs and S!) become very large numbers. To avoid this problem we compute B! e−εs εSs e−ðεb + μÞ ðμ + εb ÞB e−ðεs + μÞ ðμ + εs ÞS , , and similarly. compute the other three ratios, S! B! S! By imposing an independent structure across trading days, we have the following likelihood function for observations over I days:

I

V = LðθjM Þ = ∏ LðθjBi ; Si Þ;

ð2Þ

i=1

where (Bi, Si) is the trade data for day i = 1,…, I. As can be seen from the above two equations, the daily numbers of buy and sell orders are sufficient statistics for the data. As one observes the order flow over an increasing number of days, one can estimate the parameter vector θ = (α, μ, εb, εs, δ) with increasing precision, assuming that the parameter vector θ = (α, μ, εb, εs, δ) is stationary. Essentially, the model uses the normal levels of buy orders and sell orders to identify εb and εs. An abnormal buy or sell order volume is interpreted as information-based and identifies μ. The number of days on which there is an abnormal buy or sell volume identifies α and δ. The maximum likelihood that estimates each firm's parameter vector θ = (α, μ, εb, εs, δ) over a particular period allows us to calculate firm- and period-specific PINs. From the estimated model parameters, we can infer the unobservable information events through observed trade data. One variable of particular interest is PIN, which represents a measure of information risk. The PIN is defined as PIN =

αμ ; αμ + εs + εb

ð3Þ

where αμ + εs + εb is the arrival rate of all trades and αμ is the arrival rate of information-based trades. Compared with other measures, PIN is a direct measure of information asymmetry, and it avoids numerous econometric problems and interpretation difficulties (see O'Hara, 1995; Callahan et al., 1997). An additional advantage of the PIN methodology is that it enables us to analyze the sources of the underlying relation between the changes in bond ratings and the changes in stock information asymmetry. 2.2. Other information risk measures In addition to the PIN measure, several other proxies for information risk are employed in prior studies. First, bid–ask spread based proxies are used in studies such as Welker (1995), Leuz and Verrecchia (2000), and Odders-White and Ready (2006). Second, the dispersion of earnings forecasts is used by Mansi et al. (2005). Third, institutional share holding is used as a measure of information risk in Wang and Zhang (2006).8 Therefore, in our empirical studies, we conduct robustness checks on these three additional information risk measures, i.e., bid–ask spread, forecast dispersion, and institutional equity holding. 3. Data description Our sample period spans from January 1996 to April 2004.9 Our original data consists of bond rating changes for companies listed on the NYSE, AMEX, and Nasdaq stock exchanges. The bond rating categories and the rating changes are based on the Moody's, Standard & Poor's, and Fitch ratings in that sequence. That is, we use bond rating information from Moody's first; if Moody's bond rating is not available for some bonds, we use S&P's rating; if S&P's rating is not available either, we will use Fitch's rating information. The sample of bond rating changes is retrieved from FISD and Bloomberg; the accounting information is retrieved from Compustat; and the intraday transaction information is retrieved from TAQ. To create our final sample, we use several criteria to select bond rating changes from the original sample. First, we eliminate observations of bond rating changes associated with American Deposit Receipts (ADRs). Second, we require that the firms must be present in the Compustat for the fiscal year of the bond rating changes. This is to make sure that we can perform robustness checks by matching the sample firms with firms that did not experience bond rating changes. Third, the firms must have been on the TAQ for at least three months before and three months after the rating change date. Finally, following Healy and Palepu (1990), we exclude bond rating changes for the same firm during the three months before or after a rating change. We exclude these changes in order to reduce dependence in the statistical tests. Thus, when a firm has two or more bond rating changes within a six-month period, the bond rating changes after the first one are not included in our final sample. 8 Shleifer and Vishny (1986) argue that institutional shareholders have incentives to monitor corporate performance. Jarrell and Poulsen (1987) and Brickley et al. (1988) document that institutional shareholders are more likely to vote against harmful amendments. Agrawal and Mandelker (1990) find a positive relationship between institutional ownership and shareholder wealth effects of various antitakeover charter amendments. McConnell and Servaes (1990) find a positive relationship between institutional ownership and productivity, as measured by Tobin's q. 9 Our sample stops in 2004 because there was a structural change in the proportion of NYSE-listed stocks traded on the NYSE after 2004. The percentage of NYSE-listed stocks traded on the NYSE as the primary market has dropped significantly from about 80% in 2005 to about 35% in 2010. See, for example, the article, titled, “Faster, Faster: High Tech Hits the NYSE Floor” which appeared in the Wall Street Journal, March 29, 2010. Since the estimation of PIN uses the trades in the primary market, the dramatic change in the shares traded on the NYSE in recent years may affect the results of estimated PINs.

Y. He et al. / Journal of Empirical Finance 18 (2011) 103–116

107

Table 1 Summary of bond rating changes. This table presents the number of bond rating changes by year, by the pre-change rating category, and by the number of jumps. The pre-change rating category (AA and above, A, BBB, BB, B, and CCC and below) is the bond rating category before a rating change and it is based on the Fitch, Moody's, or S&P's ratings. The number of jumps is the number of categories that a bond rating change has crossed. We convert the bond rating categories to 22 numerical integers (i.e., AAA = 1, AA+ = 2, …, CCC = 19, CC = 20, C = 21, and D = 22). Panel A: Number of bond rating changes by year Year

Upgrades

Downgrades

1996 1997 1998 1999 2000 2001 2002 2003 2004 Total

8 29 46 38 37 38 36 34 13 279

2 14 29 49 51 52 52 53 8 310

Panel B: Number of bond rating changes by the pre-change rating category Rating category

Upgrades

Downgrades

AA and above A BBB BB B CCC and below Total

21 36 56 73 54 39 279

21 54 104 67 58 6 310

Panel C: Number of bond rating changes by the number of jumps Jumps

Upgrades

Downgrades

1 2 3 4 Total

180 66 33 0 279

264 30 12 4 310

Easley et al. (1997b) point out that a sixty-day trading window is sufficient to allow reasonably precise estimation of the parameters specified in Eqs. (1) and (2). As evidenced by Easley et al. (1997b) and others, the assumption of the time-invariant EKOH model parameters is appropriate if the estimation time period is short. We therefore follow the EKOH approach of limiting the estimation time period to three months (about 60 trading days) before and three months after a bond rating change. It is also short enough so that the stationarity built into the trade model is not too unreasonable.10 We estimate the parameters in Eqs. (1) and (2) separately for each stock and each period. To compute the likelihood function given in the EKOH model, we need to estimate the numbers of buys and sells on each day for each of our sample stocks. We can determine all these numbers of buys and sells from the TAQ data. First, we know that large trades sometimes have multiple participants on one side of the trades. Reporting conventions may treat such a transaction as multiple trades. To mitigate this problem, all trades occurring within 5 seconds of each other at the same price and with no intervening quote revisions are collapsed into one trade. Second, trades are classified into buys and sells using the technique developed by Lee and Ready (1991). Trades at prices above the midpoint of the bid and ask are called buys; those below the midpoint are called sells. The rationale for this classification is that trades originating from buyers are most likely to be executed at or near the ask price, while sell orders trade at or near the bid price. This scheme classifies all trades except those that occur at the midpoint of the bid and ask prices. These trades are classified using the “tick test.” Trades executed at a price higher than the previous trade are called buys, and those executed at a lower price are called sells. If the trade occurs at the midpoint and is at the same price as the last trade, its price is compared to the next most recent trade. This is continued until the trade is classified. This procedure undoubtedly misclassifies some trades, but it is standard and it has been proved to work reasonably well (Lee and Ready, 1991). To examine if bond rating changes affect the information asymmetry of stock trading, we first estimate the parameters (α, δ, μ, εb, and εs) of the asymmetric information structural model by using three months worth of trading transaction information both before and after a rating change. Second, we eliminate observations with extreme parameter estimates using the following filters: 10 It is noted that the bond rating changes are the only source of time variation in the model. It is true that the parameters might change in the long term due to the business cycle or other factors. This is the reason why we compare the model parameters for each individual stock during the three months before and the three months after a bond rating change. In addition, we also conduct robustness tests by pairing each test stock with a control stock which did not have any bond rating change during the same period.

108

Y. He et al. / Journal of Empirical Finance 18 (2011) 103–116

(1) 50 μ N ε or 50ε N μ, where ε = εb + εs; (2) α b 0.02 or α N 0.98; (3) δ b 0.02 or δ N 0.98; and (4) Min (μ,ε) b 1 (Brown et al., 2004). Third, we calculate the PIN measure based on the selected estimates of model parameters. Our final sample of bond rating changes includes 279 upgrade and 310 downgrade events. To check the robustness, we replicate our tests using all available bond rating changes without applying the above four filters to model parameters. For the estimates of the PIN parameter, it turns out that all the results remain virtually the same. More specifically, all of the coefficients on the bond rating changes retain the same signs and remain significant at the 1% level. We obtain the analysts' earnings forecast data from the Institutional Broker Estimation System (I/B/E/S). The analysts' earnings forecast dispersion (DISP) is the standard deviation of forecasts across analysts on earnings per share (EPS) divided by the absolute value of forecasted EPS 90 days before the bond rating changes or 90 days after the bond rating changes.11 Five different dispersion measures (Q1, Q2, Q3, Q4, and Y1) are used to represent the dispersion of earnings forecasts for the next one to four quarters and for the next fiscal year. Institutional equity holding data are collected from the Thomson Financial Ownership database. Institutional equity holders can be categorized into three groups: the transient group (TRA), the quasi-indexing group (QIX), and the dedicated group (DED). The transient group (TRA) has high turnover and holds highly diversified positions. The quasi-indexing group (QIX) has low turnover and holds highly diversified portfolios. The dedicated group (DED) has low turnover and are highly concentrated in their investments. We expect the TRA group to be the most sensitive to changes in information risk, and the DED group to be the least sensitive since it is more likely to possess superior information about their investments (Bushee, 1998, 2001; Wang and Zhang, 2006). The QIX group lies in between. For each firm, we follow Bushee (1998, 2001) to calculate four measures of institutional equity holding: the percentage of outstanding shares held by all types of institutional investors (Total%), the percentage of outstanding shares held by transient institutional investors (TRA%), the percentage of outstanding shares held by quasi-indexing institutional investors (QIX%), and the percentage of outstanding shares held by dedicated institutional investors (DED%). 4. Empirical results 4.1. Summary of bond rating changes Table 1 presents the number of bond rating changes from January 1996 to December 2004. Panel A of Table 1 shows that the number of bond rating changes is evenly distributed throughout the sample years for both the upgraded and downgraded groups. Panel B of Table 1 summarizes the number of bond rating changes classified by the pre-change bond rating category. The prechange bond rating category (AA and above, A, BBB, BB, B, and CCC and below) is the rating category before the change and is based on the Fitch, Moody's, or Standard & Poor's ratings. We notice that the number of rating changes is distributed evenly for the upgraded group. For the downgraded group, firms with a rating category of BBB are more likely to be downgraded, and firms with a rating category of CCC and below are less likely to be downgraded than firms in other categories. Panel C of Table 1 provides the number of rating changes by the number of jumps. The number of jumps is the number of rating categories that a bond rating change has crossed. We convert the alphabetical bond rating grades into 22 numerical integers. More specifically, we assign 1 to the rating grade of AAA, 2 to AA+, 3 to AA, …, 19 to CCC–, 20 to CC, 21 to C, and 22 to D. For example, a rating change from AA to AAA is equal to two jumps (i.e., crossing two categories, from AA to AA+ and then to AAA) for the upgraded group, and a rating change from AAA to AA is equal to two jumps (i.e., from AAA to AA+ and then to AA) for the downgraded group. The results show that most bond rating changes have one jump, that is, cross one rating category. 4.2. Estimates of model parameters and PIN Table 2 presents the mean and median estimates of the model parameters and PIN for firms whose bond ratings have been upgraded or downgraded during the sample period. The difference in means is equal to the mean after the bond rating changes minus the mean before the bond rating changes. Panel A reports the results for the upgraded group. We observe that after a bond rating upgrade, the probability of an information event (α) reduces significantly, the arrival rates of uninformed trades (εs and εb) increase significantly, and the arrival rate of informed trades (μ) does not change at all. In addition, we also check the percentages of the arrival rates for informed and uninformed trades among all the trades. It is noticed that the percentage of the arrival rate of informed trades (μ(%)) reduces significantly, while the percentages of the arrival rates of uninformed sell orders (εs(%)) and uninformed buy orders (εb(%)) increase significantly.12 All the changes in the model parameters point to a lower PIN. Thus, the improvement of a firm's bond rating sends a positive signal to the stock market, leading to a reduction in asymmetric information between the informed and uninformed stock investors. Panel B reports the estimation results for the downgraded group. As expected, the probability of an information event (α) increases significantly, the arrival rates of uninformed trades (εs and εb) decrease significantly, and the arrival rate of informed 11 When constructing the earnings forecast dispersion, we adopt different release windows as well as the stock price-scaled, the absolute mean forecast-scaled, and the absolute actual EPS-scaled dispersions. All results are very similar. 12 The percentage of the arrival rate of informed trades is equal to the arrival rate of informed trades divided by the sum of μ, εs, and εb, i.e., μ/(μ + εs + εb). We also compute the percentages of the arrival rates of uninformed buys and uninformed sells. The results indicate that the pattern of the percentages of the arrival rates has the similar trends as the original arrival rates. To save space, we do not report the percentages of the arrival rates in Table 2, but they are available upon request.

Y. He et al. / Journal of Empirical Finance 18 (2011) 103–116

109

Table 2 Estimates of model parameters and PIN for firms with bond rating changes. This table presents the mean and median estimates of model parameters and PIN for firms whose bond ratings have been upgraded or downgraded during the sample period. α is the probability of an information event, μ is the arrival rate of informed trades, εb is the arrival rate of uninformed buy orders, εs is the arrival rate of uninformed sell orders, and δ is the probability of a low signal (i.e., bad news). PIN is the probability of informed trades and is defined as PIN = αμ/(αμ + εb + εs). Diff is the difference between the estimate before and the estimate after the bond rating change. The t-statistic (t-stat) is used to test the null hypothesis that Diff = 0. ⁎, ⁎⁎, and ⁎⁎⁎ indicate significance at the 10%, 5%, and 1% levels, respectively. Before (1) Mean Panel A: Upgrades α 0.6459 μ 56.2087 63.3812 εb 58.5676 εs δ 0.7038 PIN 0.2408 Panel B: Downgrades α 0.5639 μ 54.6879 65.7084 εb 64.1328 εs δ 0.6370 PIN 0.2015

Diff = (2) − (1)

After (2) Median

Median std error

Mean

Median

Median std error

Mean

t-stat

0.6272 32.8724 26.6489 25.4067 0.7946 0.2515

0.0662 4.9320 2.2820 2.6652 0.0808

0.6025 56.7112 67.7088 65.9952 0.6846 0.2110

0.5278 34.7687 30.2988 30.1143 0.7233 0.2124

0.0617 5.0248 2.4485 2.8036 0.1003

− 0.0434 0.5025 4.3276 7.4296 − 0.0192 − 0.0298

− 3.05⁎⁎⁎ 0.23 1.98⁎⁎ 2.86⁎⁎⁎ − 1.01 − 6.89⁎⁎⁎

0.4740 39.2549 40.2420 41.2620 0.6785 0.1986

0.0654 5.5444 2.8774 3.0966 0.1062

0.6556 54.0850 64.1318 60.1074 0.6694 0.2369

0.6313 34.6827 34.6405 34.4605 0.7455 0.2521

0.06905 4.4808 2.5464 2.7828 0.0851

0.0917 − 0.6028 − 1.5766 − 4.0254 0.0324 0.0354

6.26⁎⁎⁎ − 0.28 − 1.69⁎ − 1.97⁎⁎ 1.69⁎ 7.77⁎⁎⁎

trades (μ) does not change at all. Additionally, the percentage of the arrival rate of informed trades (μ(%)) increases significantly, while the percentages of the arrival rates of uninformed sell orders (εs(%)) and uninformed buy orders (εb(%)) reduce significantly.13 All the changes in the model parameters indicate a higher PIN. Therefore, a downgrade in a firm's bond rating sends a warning signal (or bad news) to the stock market, leading to an increase in asymmetric information. In summary, the debt-market credit rating changes are significantly linked to the equity-market information asymmetry changes, verifying the link between debt and stock value uncertainties. Moreover, the credit rating upgrades give rise to a decrease in asymmetric information and vice versa for downgrades, confirming both the signaling and the discretionary disclosure hypotheses. More specifically, the occurrence of a bond upgrade signals a decrease in the debt value uncertainty, and so the stock value uncertainty and information asymmetry reduce accordingly. Meanwhile, as the good news due to upgrades is revealed quickly, the disclosure increases, and therefore the information asymmetry decreases. In contrast, as the bad news due to downgrades is revealed slowly, the disclosure decreases, and the asymmetric information increases. Table 3 reports the mean estimates of the model parameters and PIN by the level of the pre-change bond rating category. Panel A provides the estimation results for the upgraded group. For firms with a pre-change rating category of AA or A, the PIN measure does not change much after a bond rating upgrade. For firms with a pre-change rating category of BBB, BB, B, or CCC and below, the arrival rates of uninformed buy and sell trades increase significantly, and the PIN measure decreases significantly. These results indicate that the effect of a rating upgrade on the reduction of information asymmetry is significant only for firms with a relatively bad pre-change rating category (i.e., BBB or below). Panel B reports the estimation results by the pre-change bond rating category for the downgraded group. For firms with a pre-change rating category of AA, A, BBB, BB, or B, the PIN measure increases significantly after a bond rating downgrade. For firms with a pre-change rating category of CCC and below, there is no significant change in the PIN. These results imply that the effect of a rating downgrade on the increase in information asymmetry is significant only for firms with a relatively good pre-change rating category (i.e., B or above). Finally, in both Panels A and B, by comparing the PIN estimates across different pre-change bond rating categories, we notice that the better the pre-change bond rating, the lower the PIN. For example, the PIN is 0.11 in Panel A and 0.10 in Panel B for the prechange rating category of AA and above, and 0.30 in Panel A and 0.31 in Panel B for the pre-change rating category of CCC and below. Hence, a firm's bond rating category is related to its stock information asymmetry in the following way: the better the debtmarket credit rating, the lower the equity-market information asymmetry. Table 4 presents the mean estimates of the model parameters and the PIN by the number of rating jumps. Panel A reports the estimation results for the upgraded group, with the number of jumps ranging from one to three. We note that when a firm's bond rating is upgraded by two or three jumps toward the AAA direction, the probability of an information event (α) reduces significantly, the arrival rate of uninformed sell trades (εs) increases significantly, and the PIN measure reduces significantly. In addition, the results indicate that the greater the number of jumps toward the AAA direction, the larger the reduction in the PIN. For example, the PIN is reduced by only 0.002 for one jump, 0.052 for two jumps, and 0.140 for three jumps. Therefore, firms with their bond ratings more dramatically upgraded tend to see more dramatic reduction in the information asymmetry of their stock trading. Panel B provides the estimation results for the downgraded group, with the number of jumps ranging from one to four. The results indicate that when a firm's bond rating is downgraded by 1, 2, 3, or 4 jumps toward the CCC direction, both the probability

13

To save space, we do not report the percentages of the arrival rates in Table 2, but they are available upon request.

110

AA Before

A After

Panel A: Upgrades α 0.36 0.47 μ 37.31 44.77 52.16 70.24 εb 59.53 75.66 εs δ 0.61 0.66 PIN 0.11 0.13 Panel B: Downgrades α 0.31 0.59 μ 49.10 59.87 61.69 73.79 εb εs 74.59 83.16 δ 0.57 0.66 PIN 0.10 0.18

BBB

Diff

Before

After

0.11⁎ 7.47 18.18⁎⁎ 16.13⁎⁎

0.50 53.06 58.48 58.81 0.65 0.19

0.50 45.83 55.22 51.81 0.71 0.18

0.46 51.97 67.62 74.66 0.58 0.14

0.66 51.81 66.33 62.54 0.74 0.22

0.05 0.02 0.28⁎⁎⁎ 10.77 12.11 6.61 0.09 0.08⁎⁎⁎

Diff 0.00 − 7.23 − 3.26 − 7.00 0.06 − 0.01 0.19⁎⁎⁎ − 0.16 − 1.29⁎ − 12.12⁎⁎⁎ 0.16⁎⁎⁎ 0.07⁎⁎⁎

BB

B

Before

After

Diff

Before

After

Diff

Before

0.53 37.10 39.47 37.54 0.67 0.22

0.46 39.77 44.52 46.88 0.67 0.18

− 0.07 2.67 5.05⁎ 9.35⁎⁎

0.66 59.57 70.85 69.73 0.67 0.23

− 0.09⁎⁎⁎ 3.33 3.36⁎ 10.01⁎⁎

− 0.00 − 0.04⁎⁎⁎

0.74 56.24 67.26 59.72 0.72 0.27

− 0.05 − 0.04⁎⁎⁎

0.82 71.78 84.85 72.05 0.79 0.28

0.49 43.29 48.08 46.96 0.63 0.19

0.60 37.81 42.24 42.11 0.62 0.22

0.12⁎⁎ − 5.48 − 5.84 − 4.85 − 0.01 0.04⁎⁎⁎

0.66 53.06 65.52 59.38 0.66 0.24

0.67 55.28 63.59 58.19 0.64 0.26

0.01 2.22 − 1.94 − 1.19 − 0.02 0.02⁎⁎⁎

0.77 81.60 99.67 89.77 0.69 0.26

CCC and below After

Diff

Before

After

Diff

0.76 74.08 89.80 83.77 0.71 0.25

− 0.06⁎⁎ 2.33 4.45⁎ 11.72⁎⁎⁎ − 0.08⁎⁎ − 0.03⁎⁎⁎

0.68 75.12 71.30 67.20 0.70 0.30

0.65 68.12 75.39 69.73 0.68 0.25

− 0.03 − 7.00⁎⁎ 4.08⁎

0.79 83.31 100.98 86.06 0.74 0.28

0.01 1.71 1.30 − 3.78⁎ 0.05 0.01⁎⁎

0.63 54.33 41.90 28.92 0.75 0.31

0.36 40.61 39.93 40.76 0.50 0.20

− 0.27⁎⁎⁎ − 13.72⁎ − 1.98 11.84⁎

2.53 − 0.02 − 0.05⁎⁎⁎

− 0.25 − 0.12

Y. He et al. / Journal of Empirical Finance 18 (2011) 103–116

Table 3 Mean estimates of model parameters and PIN by the pre-change bond rating category. This table presents the mean estimates of model parameters and PIN by the pre-change bond rating category for firms whose bonds have been upgraded or downgraded during the sample period. The pre-change rating category (AA and above, A, BBB, BB, B, and CCC and below) is the bond rating category before the rating changes and it is based on the Fitch, Moody's, or S&P's ratings. α is the probability of an information event, μ is the arrival rate of informed trades, εb is the arrival rate of uninformed buy orders, εs is the arrival rate of uninformed sell orders, and δ is the probability of a low signal (i.e., bad news). PIN is the probability of informed trades and is defined as PIN = αμ/(αμ + εb + εs). Diff is the difference between the estimate before and the estimate after the bond rating change. ⁎, ⁎⁎, and ⁎⁎⁎ indicate significance at the 10%, 5%, and 1% levels, respectively.

Y. He et al. / Journal of Empirical Finance 18 (2011) 103–116

111

Table 4 Mean estimates of model parameters and PIN by the number of jumps. This table reports the mean estimates of model parameters and PIN by the number of jumps for firms whose bond ratings have been upgraded or downgraded during the sample period. α is the probability of an information event, μ is the arrival rate of informed trades, εb is the arrival rate of uninformed buy orders, εs is the arrival rate of uninformed sell orders, and δ is the probability of a low signal (i.e., bad news). PIN is the probability of informed trades and is defined as PIN = αμ/(αμ + εb + εs). Diff is the difference between the estimate before and the estimate after the bond rating change. ⁎, ⁎⁎, and ⁎⁎⁎ indicate significance at the 10%, 5%, and 1% levels, respectively. Panel A: Upgrades Number of jumps 1

α μ εb εs δ PIN

2

3

Before

After

Diff

Before

After

Diff

Before

After

Diff

0.674 67.394 78.833 71.905 0.734 0.238

0.702 70.926 85.960 78.866 0.732 0.236

0.027⁎ 3.532⁎ 7.128⁎⁎ 6.961⁎⁎⁎ − 0.002 − 0.002

0.556 31.807 34.329 35.543 0.630 0.221

0.428 30.672 33.761 40.102 0.595 0.169

− 0.128⁎⁎⁎ − 1.135 − 0.568 4.559⁎⁎ − 0.035 − 0.052⁎⁎⁎

0.672 43.999 37.205 31.868 0.686 0.295

0.410 31.251 36.051 47.580 0.606 0.155

− 0.261⁎⁎⁎ − 12.748⁎⁎ − 1.155 15.712⁎⁎⁎ − 0.079 − 0.140⁎⁎⁎

Panel B: Downgrades Number of jumps 1

α μ εb εs δ PIN

2

3

4

Before

After

Diff

Before

After

Diff

Before

After

Diff

Before

After

Diff

0.583 55.627 66.131 63.426 0.640 0.210

0.626 54.613 64.532 60.573 0.663 0.230

0.044⁎⁎⁎ − 1.014 − 1.599 − 2.853 0.023 0.020⁎⁎⁎

0.461 49.173 61.039 66.285 0.649 0.156

0.786 43.615 52.711 51.524 0.745 0.263

0.325⁎⁎⁎ − 5.559 − 8.328⁎ − 14.760⁎

0.455 44.601 64.710 71.828 0.480 0.140

0.887 52.840 66.052 63.107 0.5807 0.290

0.432⁎⁎⁎ 8.239⁎⁎ 1.342 − 8.722⁎⁎ 0.099 0.150⁎⁎⁎

0.428 64.340 75.833 71.553 0.789 0.154

0.920 101.491 117.623 84.753 0.781 0.330

0.492⁎⁎⁎ 37.151⁎ 41.786 13.200 − 0.008 0.176⁎

0.095 0.107⁎⁎⁎

of an information event (α) and the PIN measure increase significantly. In addition, the greater the number of jumps in the CCC direction, the greater the increase in the PIN. For example, the PIN increases by 0.020 for one jump, 0.107 for two jumps, 0.150 for three jumps, and 0.176 for four jumps. Hence, firms with their bond ratings more dramatically downgraded tend to see more dramatic increases in the information asymmetry of their stock trading. Overall, the magnitude of a firm's bond rating change is positively related to the magnitude of the change in its stock information asymmetry. 4.3. Cross-sectional regressions of the change in PIN To strengthen our analysis, the change in PIN is regressed on whether the bond rating is upgraded or downgraded, the prechange bond rating category, and the number of jumps as specified below: ΔPIN = α0 + α1 D + β0 Rating + β1 D × Rating + γ0 Jump + γ1 D × Jump;

ð4Þ

where ΔPIN is the difference in the PIN estimates between the post- and the pre-change periods. D is a dummy variable, which is equal to one for an upgrade and zero for a downgrade. “Rating” is the pre-change rating category and is classified into six broad categories with AA and above = 1, A = 2, BBB = 3, BB = 4, B = 5, and CCC and below = 6. “Jump” denotes the number of jumps in absolute value. As discussed earlier, when we compute the number of jumps, we classify the bond ratings into 22 grades, each with a corresponding integer. In Eq. (4), α0, β0, and γ0 represent coefficients for downgrades, α1, β1, and γ1 represent coefficients for the difference between upgrades and downgrades, and α0 + α1, β0 + β1, and γ0 + γ1 represent coefficients for upgrades. Table 5 presents the results of cross-sectional regressions of ΔPIN. As the first regression shows, the change in PIN is significant and positive (α0 = 0.035) for a bond downgrade, but is negative (α0 + α1 = 0.035–0.065 = − 0.030) for a bond upgrade. The results validate that following a firm's bond upgrade, the amount of asymmetric information of its stock trading reduces significantly. The vice versa holds for a downgrade. The second regression indicates that the pre-change bond rating category has a significant and negative effect on the change in PIN when a bond rating is downgraded. However, this negative effect is reduced when a bond rating is upgraded. Specifically, one rating category better is accompanied by a decrease in PIN by 0.023 (β0 = −0.023) for a downgrade, and a decrease in PIN by 0.009 (β0 + β1 = − 0.023 + 0.014 = −0.009) for an upgrade. Hence, when a bond rating is changed, the better the pre-change bond rating category, the greater the decrease in stock information asymmetry, especially for a downgrade. The third regression suggests that the absolute number of jumps is positively and significantly linked to the change in PIN, with the control for pre-change rating categories. In other words, when a bond is upgraded, the greater the number of jumps in absolute value, the greater the decrease in stock information asymmetry. The reverse is true for a downgrade. More specifically, one upgrade jump is accompanied by a decrease in PIN by 0.064 (γ0 + γ1 = 0.058–0.122 = −0.064), while one downgrade jump is

112

Y. He et al. / Journal of Empirical Finance 18 (2011) 103–116

Table 5 Cross-sectional regressions of changes in PIN. This table presents the estimates for the following regression: ΔPIN = α0 + α1D + β0Rating + β1D × Rating + γ0Jump + γ1D × Jump, where ΔPIN is the change in the PIN estimate, calculated as the difference between the post- and pre-change periods. D is a dummy variable that has a value of 1 for an upgrade and 0 for a downgrade. Rating is the bond rating category before the rating change and is classified into six categories with AA and above = 1, A = 2, BBB = 3, BB = 4, B = 5, and CCC and below = 6. Jump is the number of jumps in absolute value. To calculate Jump, we classify bond ratings into 22 grades each with a corresponding integer. More specifically, AAA = 1, AA+ = 2, AA = 3, …, CCC− = 19, CC = 20, C = 21, and D = 22. The t-statistics are in parentheses. ⁎, ⁎⁎, and ⁎⁎⁎ indicate significance at the 10%, 5%, and 1% levels, respectively. Model

Intercept

D

Rating

D × Rating

Jump

D × Jump

1

(α0)

(α1)

(β0)

(β1)

(γ0)

(γ1)

2

0.035⁎⁎⁎ (8.15) 0.113⁎⁎⁎

− 0.065⁎⁎⁎ (− 10.33) − 0.106⁎⁎⁎

3

(9.25) 0.024⁎

(− 6.13) 0.068⁎⁎⁎ (3.57)

(1.68)

Adj. R2

0.15 − 0.023⁎⁎⁎ (− 6.76) − 0.018⁎⁎⁎

0.014⁎⁎⁎ (2.96) 0.010⁎⁎⁎

(− 5.91)

(2.54)

0.23 0.058⁎⁎⁎ (9.10)

− 0.122⁎⁎⁎ (− 14.53)

0.44

associated with an increase in PIN by 0.058 (γ0 = 0.058). The regression results are all consistent with those in Tables 2, 3, and 4, and support both the signaling and the discretionary disclosure hypotheses. 5. Robustness checks 5.1. Matched firms without bond rating changes In order to examine whether the change in PIN is actually caused by the bond rating changes, we conduct a robustness test by matching the sample firms with firms whose bond ratings did not experience any changes during the same period, based on the pre-change bond rating and trading volume.14 Table 6 shows the mean and median estimates of model parameters and PIN for matched stocks whose bond ratings did not change during the sample period. The results from Panels A and B of Table 6 suggest that firms without bond rating changes do not show any significant changes in the model parameters (α, μ, εb, εs, and δ) or in the PIN measure. These contrast those results reported in Table 2, and strengthen our argument that the decrease in a firm's equitymarket information asymmetry can be attributed to its debt-market credit rating upgrade, and the increase to a downgrade. In the following, we examine whether our results are robust to other information risk measures, including traditional measures of asymmetric information (such as buy/sell numbers, buy/sell sizes, and bid–ask spreads), analysts' earnings forecast dispersion, and institutional equity holdings. We provide summary statistics on the above measures for firms whose bonds have been upgraded or downgraded.15 5.2. Trading activities around bond rating changes To investigate trading activities before and after bond rating changes, we focus on the traditional measures of asymmetric information. The first measure is Buy-Num (Sell-Num), which is defined as the number of stock transactions initiated by bid orders (ask orders). This measure reflects the change in the whole market environment, and we use the following procedures to estimate it. First, we obtain the daily Buy-Num (Sell-Num) with each bond rating change. We then compute the average of the daily BuyNum (Sell-Num) with each bond rating change in the three months before or after the rating change. Finally, we compute the mean and median for all stocks in the upgraded and downgraded groups in both the pre- and post-change periods. The second measure is Buy-Size (Sell-Size), which is the sum of the trading volume for bid-initiated (ask-initiated) stock transactions. We follow the same procedures above to obtain the mean and median for all stocks in the upgraded and downgraded groups in both the pre- and post-change periods. The last four measures (E-spread, E-spread_bp, Q-spread, and Q-spread_bp) represent the stock transaction costs for investors. E-spread is the effective bid–ask spread in dollars and E-spread_bp is the effective bid–ask spread in basis points. Q-spread is the quoted bid–ask spread in dollars and Q-spread_bp is the quoted bid–ask spread in basis points. E-spread is calculated as two times the difference between the transaction price and the midquote. Q-spread is calculated as the difference between the ask and bid prices. E-spread_bp is calculated as the effective spread (E-spread) divided by the transaction price and then multiplied by 100. Qspread_bp is calculated as the quoted bid–ask spread (Q-spread) divided by the transaction price and then multiplied by 100. Panel A of Table 7 shows that a firm's stock trading characteristics are affected in several ways when its bond rating is upgraded. The numbers of buy orders (Buy-Num) and sell orders (Sell-Num) increase significantly, and so do the sizes of buy transactions (Buy-Size) and sell transactions (Sell-Size). The bid–ask spreads (including the effective bid–ask spreads in dollars or in basis points 14 We initially match firms based on trading volume, size, industry, and/or the book-to-market ratio. Since the results for matched firms with different firm characteristics are very similar, to save space, we only report the result for the matched firms based on bond ratings and trading volume. In addition, Easley et al. (1997b) show that the trading frequency affects the level of information asymmetry. Therefore, our sample stocks are matched with the stocks belonging to firms whose bond ratings have been upgraded or downgraded during the sample period, based on both the pre-change bond ratings (the same ratings) and the closest stock trading volume. 15 For matched firms without bond rating changes, the summary statistics show that there are no significant changes in the information risk measures and there are also no significant abnormal returns surrounding the announcement dates. Results are available upon request.

Y. He et al. / Journal of Empirical Finance 18 (2011) 103–116

113

Table 6 Estimates of model parameters and PIN for the matched firms. This table reports the mean and median estimates of model parameters and PIN for the matched firms whose bond ratings do not experience any changes during the sample period. The matching process is based on both pre-change bond ratings and stock trading volume. α is the probability of an information event, μ is the arrival rate of informed trades, εb is the arrival rate of uninformed buy orders, εs is the arrival rate of uninformed sell orders, and δ is the probability of a low signal (i.e., bad news). PIN is the probability of informed trades and is defined as PIN = αμ/(αμ + εb + εs). Diff is the difference between the estimate before and the estimate after the bond rating change. The t-statistic (t-stat) is used to test the null hypothesis that Diff = 0. ⁎, ⁎⁎, and ⁎⁎⁎ indicate significance at the 10%, 5%, and 1% levels, respectively. Before Mean

After Median

Panel A: Upgraded matched firms α 0.475 0.384 μ 62.473 54.182 εb 60.077 46.863 εs 56.727 38.379 δ 0.650 0.681 PIN 0.217 0.217 Panel B: Downgraded matched firms α 0.472 0.412 μ 62.604 42.106 εb 54.343 31.410 εs 48.974 25.905 δ 0.607 0.645 PIN 0.227 0.224

Diff

Median std error

Mean

Median

Median std error

Mean

t-stat

0.072 7.706 3.206 3.293 0.117

0.489 64.514 61.155 58.006 0.652 0.220

0.407 50.380 42.649 37.031 0.698 0.202

0.067 8.607 2.976 2.943 0.125

0.014 2.041 1.079 1.278 0.002 0.003

1.03 0.77 0.72 0.84 0.07 0.48

0.074 7.229 2.691 2.388 0.124

0.474 60.891 55.231 50.734 0.655 0.224

0.404 40.043 28.851 27.269 0.720 0.216

0.069 6.569 2.581 2.366 0.119

0.003 − 1.714 0.888 1.759 0.048 − 0.003

0.19 − 0.37 0.53 1.06 1.16 − 0.39

and the quoted bid–ask spreads in basis points) decrease significantly. These findings indicate that an upgrade in a firm's bond rating increases the liquidity and reduces the transaction costs of its stock trading. Panel B of Table 7 reports the results on trading characteristics of the downgraded group. We observe that the numbers of buy and sell orders decrease significantly, the size of sell transactions also decreases significantly, and the bid–ask spreads (including the effective and quoted bid–ask spreads in dollars and in basis points) increase significantly. These results suggest that a downgrade in a firm's bond rating reduces the liquidity and increases the transaction costs of its stock trading. 5.3. Analysts' earnings forecast dispersion around bond rating changes Table 8 presents the means and medians of analysts' earnings forecast dispersion for firms whose bonds have been upgraded or downgraded. Panel A of Table 8 shows that the dispersions of analysts' forecasts on earnings for all five forecasted periods reduce significantly after the bond ratings are upgraded, implying that information risk is reduced significantly. Panel B of Table 8 shows that the dispersion of analysts' forecasts on earnings for the next one and four quarters and for the next fiscal year increase

Table 7 Summary statistics of trading activities around bond rating changes. This table reports the trading activities for firms whose bond ratings have been upgraded or downgraded during the sample period. Buy-Num (Sell-Num) is the number of transactions initiated by bid orders (ask orders) in a day. Buy-Size (Sell-Size) is the sum of trading volume (in thousand shares) of transactions initiated by bid orders (ask orders) in a day. E-spread is the effective bid–ask spread in dollars, and Espread (bp) is the effective bid–ask spread in the hundredth percentage. Q-spread is the quoted bid–ask spread in dollars and Q-spread (bp) is the quoted bid–ask spread in the hundredth percentage. Diff is the difference between the estimate before and the estimate after the bond rating change. The t-statistic (t-stat) is used to test the null hypothesis that Diff = 0. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Before

Panel A: Upgrades Buy-Num Sell-Num Buy-Size Sell-Size E-spread E-spread (bp) Q-spread Q-spread (bp) Panel B: Downgrades Buy-Num Sell-Num Buy-Size Sell-Size E-spread E-spread (bp) Q-spread Q-spread (bp)

After

Diff

Mean

Median

Mean

Median

Mean

t-stat

176.44 230.32 215.98 278.69 0.15 46.78 0.35 108.18

97.81 121.81 113.58 144.01 0.11 35.05 0.29 92.91

193.85 244.64 237.46 298.82 0.14 44.25 0.36 106.82

109.18 134.91 120.15 149.59 0.11 32.31 0.30 94.87

17.41 14.32 21.49 20.13 –0.01 –2.53 0.01 –1.36

4.07⁎⁎⁎ 3.75⁎⁎⁎ 2.71⁎⁎⁎ 1.97⁎⁎ –1.95⁎ –2.63⁎⁎⁎

183.68 229.30 216.95 283.42 0.11 51.84 0.28 119.25

100.56 123.21 103.55 125.82 0.09 36.67 0.23 106.98

176.24 211.52 211.59 257.09 0.12 54.21 0.29 121.86

91.54 113.77 96.99 110.24 0.09 37.96 0.24 111.23

–7.44 –17.78 –6.07 –26.33 0.01 2.36 0.01 2.61

–2.48⁎⁎ –4.55⁎⁎⁎ –0.80 –2.51⁎⁎⁎ 2.02⁎⁎ 1.86⁎ 1.93⁎ 1.66⁎

1.35 –2.25⁎⁎

114

Y. He et al. / Journal of Empirical Finance 18 (2011) 103–116

Table 8 Summary statistics of analysts' earnings forecast dispersion around bond rating changes. This table presents the means and medians of analysts' earnings forecast dispersion for firms whose bond ratings have been upgraded or downgraded. The dispersion is defined as the standard deviation of forecasted earnings per share (EPS) across all analysts divided by the absolute value of forecasted EPS 90 days before the rating changes or 90 days after the rating changes. Five different dispersion measures (Q1, Q2, Q3, Q4, and Y1) are used to represent the dispersion (in%) of analysts' earnings forecasts for the next one to four quarters and for the next fiscal year. Diff is the difference between the estimate before and the estimate after the bond rating change. The t-statistic (t-stat) is used to test the null hypothesis that Diff = 0. ⁎, ⁎⁎, and ⁎⁎⁎ indicate significance at the 10%, 5%, and 1% levels, respectively. Before

After

Mean

Median

Panel A: Upgrades Q1 Q2 Q3 Q4 Y1

23.71 29.29 40.76 61.25 15.94

5.10 6.19 6.27 6.09 3.34

Panel B: Downgrades Q1 Q2 Q3 Q4 Y1

47.07 52.07 68.08 80.32 34.88

12.75 15.89 13.95 17.03 6.73

Mean

Diff Median

Mean

t-stat

22.24 19.69 35.13 44.11 9.89

6.00 6.57 5.99 6.61 3.37

–1.47 –9.59 –5.64 –17.14 –6.04

–1.80⁎ –4.00⁎⁎⁎ –3.56⁎⁎⁎ –4.51⁎⁎⁎ –6.61⁎⁎⁎

48.53 68.78 95.46 101.44 77.22

14.55 17.39 19.31 19.66 9.51

1.47 16.71 27.38 21.12 42.34

1.99⁎⁎ 3.86⁎⁎⁎ 10.28⁎⁎⁎ 4.18⁎⁎⁎ 6.13⁎⁎⁎

significantly after the bond ratings are downgraded, implying that information risk increases significantly. Overall, the results based on the forecast dispersion measure are consistent with but not as strong as the results based on the PIN measure. 5.4. Institutional equity holdings around bond rating changes Table 9 provides the means and medians of institutional equity ownership around bond rating changes. Panel A shows that when bond ratings are upgraded, the percentages of outstanding shares held by all types (Total%) of institutional investors and transient institutional investors (TRA%) increase significantly, indicating that information risk is reduced. In Panel B, when bond ratings are downgraded, the percentages of outstanding shares held by all types of institutional investors (Total%), the transient group (TRA%), and the quasi-indexing group (QIX%) decrease significantly, indicating that information risk increases. It is expected that the transient group (TRA) experiences more significant changes than the dedicated group (DED). This is because the TRA group has high turnover and holds highly diversified positions, and they are most sensitive to information risk. The DED group is the opposite, and the QIX group lies in between. Furthermore, the holdings of quasi-indexing and dedicated institutional investors are much smaller than those of transient institutional investors. In summary, the results based on institutional equity holdings are in line with those based on the PIN measure. That is, bond rating changes do exert an impact on information asymmetry. As a bond rating becomes better, information asymmetry decreases; conversely, as a bond rating worsens, information asymmetry increases. 6. Conclusions Current literature has demonstrated that a firm's debt value uncertainty and its stock value uncertainty are connected, and the connection goes both ways. Among empirical studies in this area, one line of research explores the impact of stock uncertainty on Table 9 Summary statistics of institutional equity holdings around bond rating changes. This table reports the means and medians of institutional equity holdings for firms whose bonds have been upgraded or downgraded. Total% represents the percentage of outstanding shares held by all types of institutional investors. TRA%, QIX%, and DED% represent the percentages of outstanding shares held by transient, quasi-indexing, and dedicated institutional investors, respectively. Diff is the difference between the estimate before and the estimate after the bond rating change. The t-statistic (t-stat) is used to test the null hypothesis that Diff = 0. ⁎, ⁎⁎, and ⁎⁎⁎ indicate significance at the 10%, 5%, and 1% levels, respectively. Before

After

Diff

Mean

Median

Mean

Median

Mean

t-stat

Panel A: Upgrades Total% TRA% QIX% DED%

68.46 60.38 0.91 6.82

64.87 57.58 0.38 3.64

72.61 65.22 0.72 6.11

66.33 58.17 0.39 4.08

4.15 4.84 –0.19 –0.71

1.80⁎ 1.78⁎ –1.32 –0.70

Panel B: Downgrades Total% TRA% QIX% DED%

54.28 48.16 0.67 5.69

58.62 51.54 0.34 2.68

50.91 44.24 1.04 5.28

58.20 48.86 0.34 2.75

–3.37 –3.92 0.37 –0.42

–2.05⁎⁎ –1.96⁎⁎ 2.76⁎⁎⁎ –1.34

Y. He et al. / Journal of Empirical Finance 18 (2011) 103–116

115

debt, while the other line investigates the impact of debt uncertainty on stock. The recent credit crisis has led to particular attention being paid to the default risk of debt as well as the impact of debt uncertainty on stock. Following the second line of research, we examine whether the debt-market credit rating changes affect the equity-market information asymmetry. Using data of bond rating changes during the period of 1996 to 2004, we find that a firm's bond upgrade leads to a significant decrease in the PIN measure, and a downgrade leads to a significant increase in the PIN measure. In addition, the magnitude of a bond rating change is positively related to the magnitude of change in the PIN. Other measures of information risk, including bid–ask spreads, analysts' earnings forecast dispersion, and institutional equity holdings, are also affected by bond rating changes in a similar manner. Our results support both the signaling and the discretionary disclosure hypotheses. First, we verify the signaling explanation. The corporate bond market and the equity-market are interrelated. They both reveal information about a firm's expected future performance and uncertainty. An update on a firm's credit quality sends out a signal regarding a change in the firm's financial condition. Stock investors catch the upgrade or downgrade signal, interpret the signal as good or bad news, and revise their expectations of the firm's future performance and uncertainty. As a result, the distribution of informed and uninformed stock traders is affected accordingly, which, in turn, affects the information asymmetry of its stock trading. Second, we validate the discretionary disclosure explanation. Bond rating upgrades or downgrades reveal good or bad news about a firm's financial condition. Since good news is revealed quickly and bad news is revealed slowly, it gives rise to an increase in disclosure for good news and a decrease for bad news, and therefore a decrease in information asymmetry for good news and an increase for bad news. Acknowledgments The authors appreciate helpful comments from Honghui Chen, Mathijs Van Dijk, and seminar participants at the City University of Hong Kong, the Hong Kong University of Science and Technology, the 2007 Financial Management Association meeting and the 2007 China International Conference in Finance. The authors thank the editor (Franz Palm) and two anonymous reviewers for their constructive comments and suggestions and Dr. Virginia Unkefer and Alice Cheung for editorial assistance. Junbo Wang acknowledges financial support from the RGC Research Grant of the Hong Kong Special Administration Region, China (CityU1423/ 06H). John Wei acknowledges financial support from the RGC Research Infrastructure Grant of the Hong Kong Special Administration Region, China (RI93/94.BM02). References Abad-Romero, P., Robles-Fernandez, M.D., 2006. Risk and return around bond rating changes: new evidence from the Spanish stock market. J. Bus. Finance Acc. 33, 885–908. Agarwal, P., O'Hara, M., 2007. Information Risk and Capital Structure, Working paper. Cornell University. Agrawal, A., Mandelker, G., 1990. Large shareholders and the monitoring of managers: the case of antitakeover charter amendments. J. Financ. Quant. Anal. 25, 143–161. Ajinkya, B., Bhojraj, S., Sengupta, P., 1999. The Effect of Corporate Governance on Disclosure, Working paper. University of Florida. Bae, K.H., Lim, C., Wei, K.C.J., 2006. Corporate governance and conditional skewness in the world's stock markets. J. Bus. 79, 2999–3028. Bhojraj, S., Sengupta, P., 2003. Effect of corporate governance on bond ratings and yields: the role of institutional investors and outside directors. J. Bus. 76, 455–475. Botosan, C.A., 1997. Disclosure level and the cost of equity capital. Acc. Rev. 72, 323–349. Botosan, C.A., Plumlee, M.A., 2002. A re-examination of disclosure level and the expected cost of equity capital. J.Acc. Res. 40, 21–40. Brickley, J., Lease, R., Smith, C., 1988. Ownership structure and voting on antitakover amendments. J. Financ. Econ. 20, 267–291. Brown, S., Hillegeists, S., Lo, K., 2004. Conference calls and information asymmetry. J. Acc. Econ. 37, 343–366. Bushee, B., 1998. The influence of institutional investors on myopic R&D investment behavior. Acc. Rev. 73, 305–333. Bushee, B., 2001. Do institutional investors prefer near-term earnings over the long-run value? Contemp. Acc. Res. 18, 207–246. Callahan, C.M., Lee, C.M.C., Yohn, T.L., 1997. Accounting information and bid–ask spreads. Acc. Horiz. 11, 50–60. Chen, J., Hong, H., Stein, J., 2001. Forecasting crashes: trading volume, past returns, and conditional skewness in stock prices. J. Financ. Econ. 61, 345–381. Coles, J.L., Loewenstein, U., Suay, J., 1995. On equilibrium pricing under parameter uncertainty. J. Financ. Quant. Anal. 30, 347–364. Cornell, B., Landsman, W., Shapiro, A., 1989. Cross-sectional regularities in the response of stock prices to bond rating changes. J. Acc. Auditing Finance 4, 460–479. Dichev, I.D., Piotroski, J.D., 2001. The long-run stock returns following bond ratings changes. J. Finance 56, 173–203. Easley, D., O'Hara, M., 2004. Information and the cost of capital. J. Finance 59, 1553–1583. Easley, D., Kiefer, N.M., O'Hara, M., 1996a. Cream-skimming or profit-sharing? The curious role of purchased order flow. J. Finance 51, 811–833. Easley, D., Kiefer, N.M., O'Hara, M., Paperman, J., 1996b. Liquidity, information, and less-frequently traded stocks. J. Finance 51, 1405–1436. Easley, D., Kiefer, N.M., O'Hara, M., 1997a. The information content of the trading process. J. Empirical Finance 4, 159–186. Easley, D., Kiefer, N.M., O'Hara, M., 1997b. One day in the life of a very common stock. Rev. Financ. Stud. 10, 805–835. Easley, D., O'Hara, M., Paperman, J., 1998a. Financial analysts and information-based trade. J. Financ. Mark. 1, 175–201. Easley, D., O'Hara, M., Srinivas, P.S., 1998b. Option volume and stock prices: evidence on where informed traders trade. J. Finance 53, 431–465. Easley, D., O'Hara, M., Saar, G., 2001. How stock splits affect trading: a microstructure approach. J. Financ. Quant. Anal. 36, 25–51. Easley, D., Hvidkjaer, S., O'Hara, M., 2002. Is information risk a determinant of asset returns? J. Finance 57, 2185–2221. Ederington, L., Goh, J., 1998. Bond rating agencies and stock analysts: who knows what and when? J. Financ. Quant. Anal. 33, 569–585. Ederington, L., Yawitz, J., 1987. In: Altman, E. (Ed.), The Bond Rating Process, Handbook of financial Markets and Institutions, 6th ed. John Wiley and Sons, New York, NY. Ederington, L., Yawitz, J., Roberts, B., 1987. The informational content of bond ratings. J. Financ. Res. 10, 211–226. Fishman, M.J., Hagerty, K.M., 1989. Disclosure decisions by firms and the competition for price efficiency. J. Finance 44, 633–646. Goh, J., Ederington, L., 1993. Is a bond rating downgrade bad news, good news, or no news for stockholders? J. Finance 48, 2001–2008. Grier, P., Katz, S., 1976. The differential effects of bond rating changes on industrial and public utility bonds by maturity. J. Bus. 49, 226–239. Hand, J., Holthausen, R., Leftwich, R., 1992. The effect of bond rating agency announcements on bond and stock prices. J. Finance 47, 733–752. Healy, P., Palepu, K.G., 1990. Earnings and risk changes surrounding primary stock offers. J. Acc. Res. 28, 25–48.

116

Y. He et al. / Journal of Empirical Finance 18 (2011) 103–116

Healy, P., Hutton, A., Palepu, K., 1999. Stock performance and intermediation changes surrounding sustained increases in disclosure. Contemp. Acc. Res. 16, 485–520. Hite, G., Warga, A., 1997. The effect of bond-rating changes on bond price performance. Financ. Anal. J. 53, 35–51. Holthausen, R., Leftwich, R., 1986. The effect of bond rating changes on common stock prices. J. Financ. Econ. 17, 57–89. Ingram, R., Brooks, L., Copeland, R., 1983. The information content of municipal bond rating changes: a note. J. Finance 38, 997–1003. Jarrell, G., Poulsen, A., 1987. Shark repellents and stock prices: the effects of antitakeover amendments since 1980. J. Financ. Econ. 19, 127–168. Katz, S., 1974. The price adjustment process of bonds to rating reclassifications: a test of bond market efficiency. J. Finance 29, 551–559. Kisgen, D., 2006. Credit ratings and capital structure. J. Finance 61, 1035–1072. Kliger, D., Sarig, O., 2000. The information value of bond ratings. J. Finance 55, 2879–2902. Lee, C.M.C., Ready, M.J., 1991. Inferring trade direction from intraday data. J. Finance 46, 733–747. Leuz, C., Verrecchia, R., 2000. The economic consequences of increased disclosure. J. Acc. Res. 38, 91–124. Liu, Y., Malatesta, P.H., 2005. Credit Ratings and the Pricing of Seasoned Equity Offerings, Working Paper. University of Washington, Seattle. Mansi, S., Maxwell, W., Miller, D., 2005. Information Risk and the Cost of Debt Capital, Working Paper. University of Arizona. McConnell, J., Servaes, H., 1990. Additional evidence on equity ownership and corporate value. J. Financ. Econ. 27, 595–613. Merton, R.C., 1987. A simple model of capital market equilibrium with incomplete information. J. Finance 42, 483–510. O'Hara, M., 1995. Market Microstructure Theory. Blackwell, Malden, MA. Odders-White, E., Ready, M., 2006. Credit rating and stock liquidity. Rev. Financ. Stud. 19, 119–157. Sengupta, P., 1998. Corporate disclosure quality and the cost of debt. Acc. Rev. 73, 459–474. Shleifer, A., Vishny, R., 1986. Large shareholders and corporate control. J. Polit. Econ. 94, 461–488. Wang, A., Zhang, G., 2006. Institutional Equity Investment, Asymmetric Information and Credit Spreads, Working Paper. University of California, Irvine. Welker, M., 1995. Disclosure policy, information asymmetry, and liquidity in equity markets. Contemp. Acc. Res. 11, 801–827. Yu, F., 2005. Accounting transparency and the term structure of credit spreads. J. Financ. Econ. 73, 53–84.