The Pricing Impact of Debt IPO Prospectus: A Text-Based Study. Abstract

Transcription

1 The Pricing Impact of Debt IPO Prospectus: A Text-Based Study Abstract This paper provides the first empirical evidence on the significant pricing impact of the textual content of DIPO prospectus. We find that DIPOs with less readable and more ambiguous prospectuses are associated with worse initial credit ratings and higher initial offering yields. As time passes by, these bonds also experience more downgrades and are also associated with higher secondary yields. Overall, our evidence is consistent with firms successfully engaging in textual management and investors incomplete adjustment to the hidden adverse information. 1

2 The Pricing Impact of Debt IPO Prospectus: A Text-Based Study I. Introduction Security offering is an important process that connects an operating company to its investors. The offering documents used in this process convey not only crucial information about the underlying securities, but also important motives of the issuers. Previous studies on offering documents focused mostly on the textual content of equity IPOs (see, for example, Hanley and Hoberg 2010, 2012; Loughran and McDonald 2013, among others). Although debt financing is also important for corporations, very little is known about whether (and how) the textual content of debt IPO offering documents affects the pricing of debt. And if the prospectus matters, what motives does its textual content convey about the issuers? This paper makes the first attempt to understand these questions using corporate debts by first time issuers in the public bond market (debt IPOs, or DIPOs). Like equity IPOs (EIPOs), one of the first steps in issuing a DIPO is to file an initial prospectus with the Securities and Exchange Commission (SEC). DIPO issuers have no track record in the bond market and therefore investors rely on the prospectus filings to obtain information regarding the details of the bond offers. In particular, for firms that are private at the time of the DIPOs, the prospectus is the first public document regarding the firms operations, future business plans and cash flows. To examine the impact of DIPO prospectuses, we hand-collect prospectuses for 1,164 DIPOs between 1995 and We apply the methodology developed from other corporate disclosures (such as EIPOs and 10-Ks) to analyze the textual content of the DIPO prospectuses. We find that the textual content matters significantly for the pricing of the DIPOs. Specifically, DIPOs with larger (measured by more word count or larger file size) and more ambiguous 2

3 prospectuses (measured by larger count of uncertain words) are associated with higher initial offering yield spreads and lower initial credit ratings, after controlling for the bond-, firm- and market- specific variables. One standard deviation increase in the log of the word counts or one standard deviation increase in the log of the number of the uncertain words is associated with 21 bps increases in initial yield spreads; and one standard deviation increase in the log of the file size is associated with 17 bps increases in initial yield spreads. Our findings suggest that investors and credit rating agencies perceive a higher risk when facing a longer or more ambiguous prospectus. This is in line with the literature that studies the complexity and readability of corporate disclosure. When fundamental information is negative, managers have incentives to hide some adverse information and make the disclosure less readable or more ambiguous so that it is harder for investors to fully extract such information (e.g., Li 2008, Loughran and McDonald 2013, 2014). These results may appear surprising if one views larger and more ambiguous prospectuses as more complete and transparent disclosures, because more transparent disclosures should reduce the noise in the quantitative information and correspond to lower bond yields (e.g., Duffie and Lando 2001, Sengupta 1998, and Yu 2005). Further, higher initial offering yield spreads also suggest that investors to some extent are able to detect managers motive to hide bad news in the lengthier and more ambiguous prospectuses and incorporate it into the initial bond prices. The remaining key question is whether investors can fully extract the adverse information and completely incorporate it into the initial bond prices. To answer this question, we examine the predictability of the textual content of DIPO prospectuses on the prices of the same bonds in the secondary market. We find that less readable and more ambiguous prospectuses are associated with more future downgrades of the bond 3

4 ratings and more increases in the yield spreads (relative to the initial yield spreads). These results lend support to the idea that investors are not able to fully incorporate the adverse information embedded in the prospectuses at bond issuance when it is made harder to extract, consistent with the incomplete information revelation or information obfuscation hypothesis by Bloomfield (2002). When adverse information cannot be withheld anymore and has to be released at some point in the future, it gets reflected in the bond prices and credit ratings. In addition, we find that less readable and more ambiguous prospectuses are associated with higher idiosyncratic risks after bond issuance. Overall our findings indicate that managers have incentives to withhold certain adverse information in the bond issuance and make the prospectuses less readable and more ambiguous so that investors are not able to fully extract such information and incorporate it into the initial bond prices. Our paper contributes to the broad literature of corporate bond pricing by identifying an important factor that contributes to the prices of newly issued bonds. 1 The key distinction of our study from prior literature is the focus on the textual content of the prospectuses. Our paper suggests that in the setting of high information asymmetry such as the DIPOs, the qualitative information embedded in the corporate disclosures plays an important role of the price discovery and investors should pay special attention to it. Our paper also contributes to the growing literature that examines how managerial incentives might be reflected in the ways that the corporate disclosures are written. Most of existing studies in this literature have focused on disclosures such as 10-Ks and prospectuses of EIPOs. In the context of 10-Ks, Li (2008) finds that 10-K reports are harder to read when 1 For example, Ederington (1974), Lindvall (1977), and Weinstein (1978) find that the yield to maturity (YTM) on a new bond offering (can be a DIPO or an SBO) is usually higher than the YTM on an existing corporate bond that trades in the secondary market. Sorenson (1982) concludes that the seasoning process observed in previous papers is mostly an underpricing phenomenon. Cai, Helwege and Warga (2007) examine DIPOs and SBOs in the corporate bond market. 4

5 earnings are lower. You and Zhang (2009) find that more complex 10-Ks are associated with a delayed investor reaction. Hoberg and Lewis (2015) find that fraudulent firms produce abnormal verbal disclosures in the MD&A section of 10-Ks and find evidence consistent with the hypothesis that some managers commit fraud in order to improve the likelihood of raising financing at a low cost. In the context of EIPOs, Hanley and Hoberg (2010) find that the informative content, as opposed to standard content for industry or market peer firm IPOs, positively reflects more premarket due diligence which results in more accurate offer prices and less underpricing. Hanley and Hoberg (2012) show that issuers tradeoff underpricing and strategic disclosure as potential hedges against litigation risk. Loughran and McDonald (2013) examine the tone of initial IPO prospectus and find the uncertain language makes it more difficult for investors to precisely assimilate the value-relevant information; as such EIPOs with high levels of uncertain text have higher initial returns, absolute price revisions and subsequent volatility. Our paper represents the first such study in the DIPO market. The debt IPO market provides an ideal setting to test the information obfuscation hypothesis for two reasons. First, the investor base is relatively homogeneous. Bond prices are likely to respond more to information than other non-information factors such as investor sentiment or investor attention driven by retailed investors. Second, due to the potential conflict of interests between bondholders and shareholders, the management may have incentives to withhold and obfuscate adverse information related to bond issuance which could make the detection of such behavior possible. Our results suggest that even though most investors are institutional investors in the bond market, the obfuscation of information in the textual content could still make the adverse information hard to extract and affect future bond prices significantly. 5

6 The remainder of the article is organized as follows. Section 2 develops hypothesis. Section 3 describes the data and methodologies. Section 4 discusses the empirical analysis of pricing differentials between DIPOs and SBOs. Subsample analysis of DIPOs based on S-1 tone appears in Section 5. Section 6 concludes the study and summarizes the results. II. Literature Review and Hypotheses Existing research on information management has focused mostly on the equity market. For example, earlier studies argue that firms disclose adverse information to avoid lawsuits in the future (Skinner 1994). More recent evidence suggests firms disclosing more also have more frequent litigation (Skinner 1997), and firms withhold bad news up to a certain threshold (Kothori, Shu, and Wysocki 2009). A growing literature examines how this informationmanagement tendency might be reflected in the ways that firms disclose the information in their financial reports such as 10-Ks. For example, Li (2008) finds that 10-K reports are harder to read when earnings are lower and his evidence supports the incomplete information revealing hypothesis (Bloomfield 2002). You and Zhang (2009) find that more complex 10-Ks have a delayed investor reaction to 10-K filings over the 12 months following the filing date. Other studies find that 10-K readability positively relates to small investor trading (Miller (2010)), stock holdings of retail investors (Lawrence (2013)), as well as analyst coverage (Lehavy, Li, and Merkley (2011)). Huang, Teoh, and Zhang (2014) find that firms manage the tone in earnings press releases and the abnormal tone can predict negative future earnings and cash flows, consistent with the hypothesis that managers use strategic tone management to mislead investors about firm fundamentals. In a recent paper, Ertugrul et al (2015) investigate the impact of a firm s 10-K readability and ambiguous tone on its borrowing costs of bank loans. They find 6

7 that firms with larger 10-K file sizes and a higher proportion of uncertain and weak modal words in 10-Ks have stricter loan contract terms and greater future stock price crash risk. In the equity IPO setting, Loughran and McDonald (2013) examine the tone of initial IPO prospectus and find the uncertain language makes it more difficult for investors to precisely assimilate the valuerelevant information. In this paper, we extend the literature by examining the possible information management/withholding in the DIPO market. Previous literature document that the information asymmetry is likely to be high in the DIPO market. For example, Cai, Helwege and Warga (2007) find that there is more information problems in the DIPO market than in the SBO market. Because for SBOs, learning about the bond can be more effective as investors can use other bonds of the same firm to infer more information, whereas for DIPOs, market participants have less publicly available information regarding the company s creditworthiness and profitability. In the high asymmetric information setting, for firms with negative fundamentals, the managers have incentives to hide certain adverse information and obfuscate it in the prospectuses. This is in line with the management obfuscation hypothesis (Bloomfield 2002) which states that mangers have incentives to obfuscate information when the fundamentals are bad. Following aforementioned empirical literature, we argue that managers attempt to hide the adverse information by either making the prospectus harder to read (low readability through long and large disclosure documents) or more ambiguous (more uncertain words). On the other hand, lengthier and larger disclosures could also represent more transparent and complete disclosures. Therefore, it is an empirical question which managerial incentives the readability measures can capture. Since most investors in the public bond market are institutional investors, they will be able to read through the lines in the prospectus and price it in the initial offering yields. If 7

8 lengthier disclosures represent more transparent ones, we expect that investors would be willing to pay for higher prices or lower yields for the bonds. In contrast, if lengthier disclosures represent harder-to-read ones, we expect investors to discount the bonds for low readability. Given that our measures are constructed from the readability literature, we state our hypotheses that are consistent with this line of the literature. Hypothesis 1a: DIPOs that have a less readable or more ambiguous prospectus are associated with worse initial credit ratings. Hypothesis 1b: DIPOs that have a less readable or more ambiguous prospectus are associated with higher cost of debt after controlling for ratings and other bond-, firm- and market- characteristics. Consistent with the management obfuscation (or incomplete information revelation) hypothesis by Bloomfield (2002), we expect that managers have incentives to withhold some adverse information and make the DIPO prospectuses hard to read and extract information. Therefore, credit agencies and investors should not be able to fully incorporate it in the initial bond ratings and prices. In the meanwhile, although some adverse information may be withheld at the bond issuance time, it cannot be withheld forever. In other words, the firms will need to release such information to the market at some point in the future. As such information starts to get revealed in the market, the market would respond to the new (adverse) information correspondingly. Thus, the following hypothesis is proposed. Hypothesis 2a: DIPOs with less readable or more ambiguous prospectuses will be associated with more downgrades in the future; and when these downgrades happen, firms will experience higher level of idiosyncratic risks. 8

9 Similarly, investors cannot fully incorporate the adverse information into initial bond prices because it is made harder to extract from the public disclosures. And when the adverse information is released after the bond issuance, bond prices would also react accordingly. Thus, our final hypothesis is: Hypothesis 2b: DIPOs that have a less readable or more ambiguous prospectus are associated with an increasing trading spread in the secondary bond market after controlling for rating and other bond-, firm- and market- characteristics. III. Data 3.1 Sample Construction We identify first-time issuers in the U.S. corporate bond market through the Fixed Income Securities Database (FISD) from Mergent, Inc. This database maintains issuance information on all fixed income securities that are assigned CUSIPs, or instruments that are likely to receive one in the near future. FISD represents one of the most comprehensive bondissuance databases extant, and serves as the issuance database engine for a variety of data resellers. Using FISD, we first identify all firms that completed new fixed rate bond offerings between January 1995 and June 2013, but have never issued any public debt prior to This sample of potential bond IPOs is then screened to further ascertain that the first offering observed in FISD is indeed a DIPO. We use data on outstanding public bonds reported in SDC to eliminate firms that had outstanding public straight bonds before 1995, when the FISD starts. 2 2 FISD includes all public bonds that mature in 1990 or later. Thus, a firm that issued a bond prior to 1990 that had already matured would be mistakenly identified as having a bond IPO were it not for the crosscheck in the other databases. 9

10 One of the prominent features in the DIPO market is that many IPO bonds are exchange offers of previously-issued private debt to the so-called qualified institutional buyers based on the SEC Rule 144A (144A bond), which exempts the issuer from registering the bond with the SEC. As documented in Fenn (2000), most of the Rule 144A bonds are subsequently registered with identical terms as exchange offers within a short period. DIPOs that are exchange offers typically do not have pricing information in FISD. We trace their Rule 144A predecessors to get the pricing information. 3 We exclude all bonds, whether they are IPOs or not, that have unusual features that could affect yield to maturity. These include foreign issues, pay-in-kind bonds, corporate pass-thru trusts and bonds issued by financial firms. We also exclude all offerings below $5 million. In total, 1,447 bond offerings meet our criteria. We obtained secondary bond prices from the National Association of Insurance Commissioners (NAIC) secondary market pricing database. We use the NAIC data primarily to obtain the longest possible sample of bond transaction. 4 NAIC records all bond transactions by life insurance companies, property and casualty insurance companies, and health maintenance organizations starting from Campbell and Taksler (2003) note that these institutions hold about one-third of outstanding corporate bonds. 3.2 Textual Content of Offering Documents 3 For DIPOs with missing offering price and yield, we assume the offer is a Rule 144A exchanged offer. To get the predecessor 144A bond, we require DIPO and its predecessor Rule 144A offer have the same coupon rate, maturity date and offering amount, and also require that the Rule144A issue is offered within 1 year of the subsequent DIPO s offering date. 4 An alternative would be to use the Trade Reporting and Compliance Engine (TRACE) data. However, TRACE data starts only in In addition, two-thirds of the debt IPOs are exchange offers of previously-issued 144A bonds, and transactions of 144A bonds are only partially available in TRACE. We compare the number of transactions of Rule-144A-converted DIPOs in both databases, and find that NAIC has a substantially larger number of transactions than TRACE before

11 We download the bond prospectuses from SEC s EDGAR website. These forms come in the forms of S-1, S-4, and Out of 1,447 DIPOs, we are able to locate the prospectuses for 1,164 DIPOs. We extract the filing firm s CIK and filing date from these prospectuses. Then we link them to the textual information provided by the WRDS SEC Analytics Suite by matching the CIK, form type, and filing date. For validation purposes, we also parse the prospectuses that we download and calculate variables provided by WRDS SEC Analytics Suite, and find that our measures are highly correlated with theirs. The textual information we use in our paper includes (1) measures of readability/complexity of prospectuses such as file size and number of words 6,7 ; (2) measures of uncertainty by the number of uncertain words as in Loughran and McDonald (2013). 3.3 Summary Statistics and Correlations Table 1 provides the summary statistics of DIPOs for which we are able to download the prospectuses. We note that the average maturity for DIPOs is 8.6 years. There are about 21% of DIPOs are rated investment grade. The average credit rating of DIPOs is 5.53, which is between B+ to BB-. There are about 55% of DIPOs are issued by private firms and 73% of bonds are Rule 144A exchanged offers. For DIPOs that are issued by public firms, the average (median) long-term debt ratio is 27% (22%), and the daily idiosyncratic volatility for stocks is 2% (2%). Our sample is similar to DIPO samples in previous studies. For example, Datta, Iskandar-Datta and Patel (2000) find that 62% of DIPOs have a maturity between 5 to 10 years, and Cai, 5 IPO offers use Form S-1 to file preliminary prospectuses; however, exchange offers, which in our case are 144Aconverted IPO bonds, use Form S-4 to file prospectuses. There are also cases where we cannot find either Form S-4 or S-1, but instead find Form 424, which is generally the final copy of prospectus with determined offering price. The textual content of Form 424 is typically very similar to its preliminary counterpart. 6 The word list that we use to measure the number of words is from Loughran and McDonald (2011). 7 We also used Fog index as a measure of the readability. However, consistent with Loughran and McDonald (2014), the results using Fog index are not significant. These results are available upon request. 11

12 Helwege and Warga (2007) document 67% of DIPOs are rated BB or B and 52% of DIPOs are issued by private firms. Table 1 also lists information on the price and variables for textual content. On average, our sample of DIPOs have an average offer yield and offer yield spread of 9.37% and 4.43%, respectively. In terms of readability, the prospectus has a mean (median) of 74,000 (71,000)words, which is longer than EIPOs documented in the literature. 8 The average fox index is (22.37), as compared to the average Fog index of 19 for 10-Ks reported in Loughran and McDonald (2014). In terms of uncertain tone, the average number of (median) uncertain words is 1,604 (1,526), which accounts for 2% of the entire prospectus. [Table 1 about here.] We present the degree of correlation between these variables in Table 2. Consistent with our expectations, yield spread is strongly positively correlated with textual content measures. The correlation between DIPO offering spread and log_wc, our primary measure of readability, is 0.5. This correlation provides first evidence that harder-to-read prospectuses are associated with higher cost of debt. In addition, we note the presence of collinearity among the textual content variables, which precludes us from including all these variables in the same model. [Table 2 about here.] IV. Empirical Results 4.1 Immediate Market Reaction to the Textual Content of Prospectus We first examine the relation between textual content measures (readability and ambiguity) and DIPOs initial credit ratings. Table 3 presents the results of regressing the initial 8 For EIPOs, the average word count is thirty four thousand in Hanley and Hoberg (2010), and the mean and median are, respectively, forty-eight and forty-two thousand words in Loughran and McDonald (2013). 12

13 credit rating on each of the three textual content measures. The control variables consist of issuespecific attributes (maturity, offering amount and whether a bond is callable), issuer-specific attributes (whether the issuer is a public firm, long term debt ratio (ltdebt_ratio), stock s idiosyncratic return volatility (iv)), and aggregate bond market credit and liquidity risk factors (default, term-structure slope, funding liquidity spreads (TED) and the VIX index ). All of the variables are defined in Appendix A. For all dependent variables, we run two sets of regressions. In the first set, we use only issue-, and market- specific variables along with the public dummy. In the second set, we include long term debt ratio and stock s idiosyncratic return volatility. For private firms where these two variables are not available, we set them to zero. The results for these two sets of regressions are of similar economic magnitude and statistical significance, and hence we focus on the first set in the discussion. We control for year fixed-effects, cluster the standard errors by issuer, and also adjust the standard errors for heteroscedasticy in all regressions. [Table 3 about here.] Models 1 to 4 show the effect of readability on the initial ratings, and models 5 to 6 show the effect of ambiguity on the initial ratings. We find that the coefficients on word count (log_wc) and file size (log_fsize) are both positive (coefficient estimates of 0.85 and 0.43 respectively) and significant at the 1% level. Since higher values of the dependent variable indicate worse ratings, these results suggest poorer initial ratings for DIPOs with lower readability. In terms of economic significance, we note that one standard increase in the word count measure (log_wc) will lead to a credit rating downgrade of In such a system, a credit rating change of, for example, BBB to BB, will counted as a value of one; and hence, one 13

14 standard deviation increase in log_wc will result in more than a notch of downgrade in credit rating. For ambiguous measures, the coefficient on uncertainty word count measure (log_wc_unc)is also positive (coefficient estimates of 0.77 and 0.58 in Models 5 and 6, respectively) and significant at 1% level, suggesting that DIPOs with more uncertain words in the prospectus are associated with significantly worse initial credit ratings. The economic significance of log_wc_unc is similar to that of log_wc. Overall, the results in Table 3 are consistent with Hypothesis 1a, that rating agencies assign worse initial credit ratings to DIPOs that have less readable or more ambiguous prospectuses. Next, we examine whether investors discount DIPOs that have lengthy or uncertain prospectuses. Table 4 presents the results of regressing the offer yield spread on the textual content variables. The dependent variable is the yield spread (cost of debt), which is computed as the bond s offer yield minus interpolated Treasury yield benchmarks. [Table 4 about here.] Models 1 to 4 in Table 4 show the effect of readability on the cost of debt. We find that the coefficients on word count (log_wc) and file size (log_fsize) are positive (coefficient estimates of 0.36 and 0.15 respectively) and statistically significant at 1% level. These results suggest that investors pay a lower price for a DIPO with a less readable prospectus. Based on model 1, a one standard deviation increase (0.58) in log_wc implies 21basis points (bps) increase in the yield spread. The effect of file size is not as strong as word count; however, it is also statistically and economically significant. A one stand deviation increase (1.16) in log_fsize increases the yield spread by 17 bps. The results for the ambiguous measures are shown in models 5 to 6. The significant positive coefficients on the count of uncertain words indicate that 14

15 DIPOs with more uncertain tones in the prospectuses are associated with higher costs of debt. The economic significance of log_wc_unc is identical to that of log_wc. The results in Tables 3 and 4 show that rating agencies assign a worse initial credit rating and that investors pay a discounted price to a DIPO whose prospectus is lengthier or more uncertain. These results are consistent with the previous literature on EIPO and suggest that managers use prospectus readability and lexical features to strategically hide the adverse information. Rating agencies and investors in the DIPO market, however, understand, at least to some extent, managers motivation to hide adverse information and react accordingly. 4.2 The Implications of Textual Content for the Future Performance of the Issuing Firms Our results so far support Hypothesis 1 that rating agencies assign a worse crediting rating and investors pays a discount price for bonds with low readable and high uncertain prospectuses. One potential source for such discounting to take place is that low readability and high uncertainty represent complexity from management s information obfuscation behaviors. A natural question that arises is whether credit agencies and investors are able to fully incorporate such information at issuance. If management indeed withholds adverse information and investors are not able to fully detect the information withheld by the management, then as time passes by, the adverse information tends to get revealed to the market and the investors would react to it eventually. In this section we test our Hypothesis 2 of whether such information withholding exists by examining time-series changes in credit rating and yield spread of DIPOs. In Panel A of Table 5, we first examine the relation between textual content measures and the change of DIPOs credit ratings by regressing the change of the DIPO s credit rating in the four years following debt issuance on the textual management variables. As with before, for the 15

16 textual content measures we use log_wc, log_fsize, and log_wc_unc. The coefficients of these variables on credit rating changes are respectively 0.13, 0.07, and 0.14, all significant at the 1% level. These results suggest that DIPO that have a lower readability (measured by higher log_wc or log_fsize) or higher uncertainty (measured by higher log_wc_unc) tends to be downgraded by rating agencies as the bond seasons. One direct reason for this could be that adverse information withheld at time of issuance gets gradually revealed in the market, and the rating agencies react accordingly. Hence, the rating change results in Panel A of Table 5 are consistent with the information withholding story. [Table 5 about here.] Adding support to the information withholding hypothesis, we also find that textual content of debt IPO prospectus is related to firm fundamentals changes at the time of credit rating changes. Since stock returns can be considered as the amalgamation of firm fundamentals, we examine whether firms with less-readable or more uncertain prospectuses are associated with higher idiosyncratic volatility of stock returns at the time of the rating change. 9 Panel B of Table 5 presents the results of regressing the change of the firm s idiosyncratic volatility (relative to its idiosyncratic volatility at the time of DIPO) on the textual content variables at the time of credit rating changes. The coefficient estimates of log_wc, log_fsize, and log_wc_unc are 0.26, 0.18 and 0.24, respectively, all statistically significant at 1% level. The results confirm that issuing firms with less readable or more ambiguous prospectuses experience higher idiosyncratic risks at the time of credit rating change. 9 Availability of stock returns for the calculation of idiosyncratic volatility constrains us to only public firms in this test. 16

17 Overall, the results in Table 5 are consistent with the hypothesis that in DIPOs, firms strategically withhold information through lengthier or more uncertain prospectuses. One manifestation of information withholding is that information withheld will be revealed sooner or later after the bond issuance; and market will react accordingly. We show that DIPOs with lengthier or more uncertain prospectuses are more likely to experience credit rating downgrades and worsening firm fundamentals after the bond issuance. These results lend support to DIPOs withhold information in prospectus. 4.3 Delayed Market Reaction to the Textual Content of Prospectuses In the section above, we show the both rating agencies and bond investors cannot fully incorporate the adverse information into initial bond rating and pricing because it is made harder to extract from the public disclosures. And when the adverse information is released after the bond issuance, bond prices would also react accordingly. Thus, our final hypothesis is that secondary yield would increase more for firms with more severe information withholding. We test this hypothesis (Hypothesis 2b) in Table 6, where we regress the spread-difference on the textual content measures. The dependent variable is the difference between the secondary-market trade spread (computed as secondary market bond s trading yield minus interpolated Treasury yield benchmark) and the offering spread. We use all of the transactions within 30 days before and after the credit rating change. We choose credit rating change as the potential information revelation event for the information withheld previously, consistent with our analysis above. Our results are robust to using 60 and 120 days around credit rating changes. [Table 6 about here.] Models 1 to 2 in Table 6 show the effect of readability on the change of spreads in the secondary market. We find that the coefficients on log_wc and log_fsize are positive and significant at the 17

18 1% level. More importantly, the economic significance is large. A one standard deviation increase in log_wc will lead to 78 bps spread increase around a credit rating change event. Similarly, Model 3 of Table 6 shows a significant and positive relation between prospectus uncertainty (log_wc_unc) and yield spread change around credit rating changes, indicating that secondary market prices continue to go down for DIPOs with more uncertain prospectuses. The economic significance is comparable to that of log_wc. Overall results from Table 6 is consistent with the incomplete information revealing hypothesis that even though investors may understand that mangers manage textual content to hide adverse information, their immediate reaction is not complete (Bloomfield, 2002). And when information withheld gets gradually revealed, the market responds with lower prices (higher yields). These results are consistent with Hypothesis 2b. V. Robustness Tests 5.1 Using Comparables for Identification One potential difficulty of interpreting our results lies in the possibility that the textual content measures may also capture some aspects of information content in the offering documents. To achieve better identification, we use a matching sample approach. We take advantage of the fact that information problems are much less severe in seasoned debt offerings (Cai, Helwege and Warge (1997)) and its textual content is more likely to reflect the information content of offering documents. We match each DIPO to a seasoned offering bond (SDO) with very similar offering attribute. The difference of offering documents and the difference of offering yields between the DIPOs and the SDOs are most likely to reflect the extent of information withholding in the DIPOs. 18

19 We follow the procedure below to create our match sample. For each debt IPO issue in the primary market, we identify a matched issue from the pool of SDOs (i.e. non-dipos) based on the following criteria: i) the two bonds must be issued in the same year; ii) they must fall into the same credit rating group (10 credit rating groups, from AAA, AA, to D); iii) they must have the same callability; 10 iv) they must belong to the same industry group (four industry groups as defined by FISD); and v) they must then have the closest offering size and maturity (as defined by the sum of absolute distance in decile ranks). We use a non-repetitive matching scheme that a bond can be used only once for the matching. We are able to matched SDO issue in the primary market for the 996 of 1,164 DIPOs. In Panel A of Table 7, we present univariate analysis of the textual content difference between the offering documents of DIPOs and the matched SDOs. It is clear that all three textual content measures are significantly higher in DIPOs. For example, the total word count is 65% higher for DIPOs, the file size is 107% larger for DIPOs, and the uncertain word count is 79% higher for DIPOs. We then regress the differences in yield spreads between DIPO and its match SDO on the textual content differences of DIPO and SDO offering documents. If our results are driven by omitted (bond- or market-specific) variables that are captured in the matching procedure, then we should not observe any relation between the yield spread difference and textual measure difference. Models 1 to 4 in Panel B show the effect of difference in readability on the difference in yield spreads between DIPOs and SDOs. We find that the coefficients on difference in word count (log_wc_diff) and difference in file size (log_fsize_diff) are positive (coefficient estimates of 0.67 and 0.45 respectively) and statistically significant at 1% level. Models 5 and 6 in Panel 10 We impose this condition because callability is the most important option feature in debt IPOs. In our sample, 95% of debt IPOs are callable; and other option features, such as puttable, are much less common. 19

20 B show the effect of difference in the ambiguous measures. The coefficients are also positive and statistically significant at 1% level. The results presented in Panel B of Table 7 show that the difference in yield spreads is positively related to the difference in all three sets of the textual content variables. Overall, these results ameliorate the identification issue of our main findings. [Table 7 about here.] 5.2 Additional Robustness and Sub-Sample Analysis We conduct additional robustness tests and sub-sample analysis to sharpen our findings and the results are reported in Table 8. Fog index is widely used in the literature, we add the Fog index as a control variable in all regressions in Panel A. Ertugrul et al (2015) find that 10-K file size affects the terms and cost of firm s bank loans. We control for the textual contents of 10-Ks around the DIPOs for public firms in all regressions in Panel B. To alleviate the concern that our results are driven by different subsamples, in Panel C, re-run all the tests for different subsamples, including DIPOs issued by private or public firms and whether DIPO is a Rule 144A exchanged offer. All the regression results are intact and consistent with our main findings. [Table 8 about here.] VI. Conclusions This paper provides the first empirical evidence on the significant pricing impact of the textual content of DIPO prospectus. Our central findings are twofold. First, we find that DIPOs with less readable and more ambiguous prospectuses are associated with worse initial credit ratings and higher initial offering yields. Second, we find that as time passes by, these bonds also experience more downgrades and are also associated with higher secondary yields. 20

21 These novel findings support the information obfuscation hypothesis (Bloomfield (2002)). Managers strategically manage the textual content of DIPO prospectus to hide adverse information for the bond issuance. Although rating agencies and investors can detect this tendency to some extent, they cannot fully incorporate the hidden adverse information into the initial bond ratings and prices. However, as time passes by and the adverse information eventually gets revealed to the market after bond issuance, the bonds subject to more severe information withholding get downgraded more and experience adverse price changes. Overall, our evidence is consistent with firms successfully engaging in textual management and investors incomplete adjustment to the hidden adverse information. 21

22 RERFERENCE: Bloomfield, R.J., The incomplete revelation hypothesis and financial reporting. Accounting Horizons 16, Cai, N., J. Helwege, and A. Warga, Underpricing in the corporate bond market. Review of Financial Studies 20, Ederington, L. H., The Yield Spread On New Issues of Corporate Bonds. Journal of Finance 29, Ertugrul, M, J. Lei, J. Qiu and C. Wan, Annual Report Readability, Tone Ambiguitiy, and the Cost of Borrowing. forthcoming, Journal of Financial and Quantitative Analysis. Hanley, K. and G. Hoberg, The information content of IPO prospectuses. Review of Financial Studies 23, Hanley, K. and G. Hoberg, Litigation risk, strategic disclosure and the underpricing of Initial Public Offerings. Journal of Financial Economics 103, Huang, X., S. H Teoh, and Y. Zhang, Tone Management. The Accounting Review 89, Kothari, S. P., S. Shu, and P. D. Wysocki, Do Managers Withhold Bad News? Journal of Accounting Research 47, Lawrence, A., 2013, Individual investors and financial disclosure. Journal of Accounting and Economics 56, Lehavy, R., F. Li, and K. Merkley, 2011.The Effect of Annual Report Readability on Analyst Following and the Properties of Their Earnings Forecasts. The Accounting Review 86, Li, F., Annual report readability, current earnings, and earnings persistence. Journal of Accounting and Economics 45, Lindvall, J., New Issue Corporate Bonds, Seasoned Market Efficiency and Yield Spreads. Journal of Finance 32, Loughran, T. and B. McDonald, When is a liability not a liability? Textual analysis, dictionaries, and 10-Ks. Journal of Finance 66, Loughran, T and B. McDonald, IPO First-Day Returns, Offer Price Revisions, Volatility, and Form S-1 Language, Journal of Financial Economics, 109:2,

23 Loughran, T and B. McDonald, 2014.Measuring Readability in Financial Disclosures. Journal of Finance 69, Miller, B. P., 2010, The Effects of Reporting Complexity on Small and Large Investor Trading, Accounting Review 85, Skinner, D., Why firms voluntarily disclose bad news. Journal of Accounting Research 32, Skinner, D., Earnings disclosures and stockholder lawsuits. Journal of Accounting and Economics 23, Sorenson, E. H On the Seasoning Process of New Bonds: Some are More Seasoned than Others. Journal of Financial and Quantitative Analysis 42, Weinstein, M. I The Seasoning Process of New Corporate Bond Issues. Journal of Finance 33, You, H., and X. Zhang, Financial reporting complexity and investor underreaction to 10-K information. Review of Accounting Studies 14,

24 Appendix A. Variable Definitions This table describes all the variables used in the paper. Textual management variables (Sources: SEC s EDGAR website, WRDS SEC Analytics Suite) log_wc Measure of readability. Number of words in the prospectus. In regressions, the variable is transformed into logarithm form for scaling. log_fsize Measure of readability. The file size of the prospectus. In regressions, the variable is transformed into logarithm form for scaling gunning_fog_index Measure of readability. The Fog index from the computational linguistics literature, which statistic combines the number of words per sentence and the number of syllables per word to create a measure of readability. log_wc_unc Measure of ambiguity. Number of uncertain words (Loughran and McDonald (2013)) in the prospectus. In regressions, the variable is transformed into logarithm form for scaling. Issue-specific characteristics (Sources: FISD, Datastream) maturity Maturity of the bond, in years, either on bond issuance date or on transaction date. vintage Bond age, in years from issuance date, on the transaction date. credit rating The numerical value of first non-nr credit rating by Moody s, Standard and Poor s, or Fitch. Credit ratings are converted to numerical values using the following coding: AAA = 1, AA= 2, A = 3, BBB = 4, BB = 5, B = 6, CCC = 7, CC = 8, C = 9, DDD and below = 10. bb A dummy variable that equals to 1 if the bond rating is BB. b A dummy variable that equals to 1 if the bond rating is B. ccc_n_below A dummy variable that equals to 1 if the bond rating is CCC and below. offer_amt Offer (issue) amount of the bond, in millions. In regressions, the variable is transformed into logarithm form for scaling. offer_yield Primary market yield of the bond, in %, on issuance date. offer_spread Primary market yield spread of the bond benchmarked to interpolated swap yield, in %, on issuance date callable A dummy variable that equals one if the bond is callable. rule144a A dummy variable that equals one if the bond is a Rule 144A exchanged issue. Issuer-specific characteristics (Sources: COMPUSTAT, COMPUSTAT global, FISD) public A dummy variable that equals one if the bond is issued by a public firm. ltdebt_ratio Ratio of long-term debt to total book value of assets of the issuing firm. iv Idiosyncratic return volatility, computed as the standard deviation of market-adjusted excess return on six months of daily stock returns preceding the transaction date. Secondary market variables (Sources: NAIC, Datastream) trading yield Secondary market yield of the bond, in %, on transaction date. trading spread Secondary market yield spread of the bond benchmarked to interpolated 24

25 spread diff delta_cr swap yield, in %, on transaction date. The difference of the trading spread and the offering yield spread. Aggregate market variables (Source: Datastream) def Default factor, obtained as Moody s BAA yield minus 10-year swap rate. term Term-structure factor, obtained as 10-year swap rate minus 2-year swap rate. vix Equity market volatility factor, obtained as VIX index. ted Aggregate liquidity factor, obtained as 30-day LIBOR rate minus 3- month Treasury-Bill rate. 25

26 Table 1. Summary Statistics of DIPO Sample This table provides the summary statistics of DIPOs for which we are able to download the prospectuses. Appendix outlines the definitions for all the variables. N Mean Median Std P1 P99 Offering amount ($1,000) 1, , , ,345 22, ,000 Offering yield (%) 1, Offering spread (%) 1, Maturity (years) 1, Credit rating 1, Investment grade (%) 1, Convertible 1, Callable 1, Rule144A 1, log_wc 1, log_fsize 1, log_wc_unc 1, gunning_fog_index 1, Long-term debt ratio Idiosyncratic volatility

27 Table 2. Correlations This table shows the degree of correlation between the key variables. The Appendix outlines the definitions for all the variables. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (1) Offering yield 1 (2) Offering spread (3) Maturity (4) Credit rating (5) Callable (6) Rule144A (7) log_wc (8) log_fsize (9) log_wc_unc (10) gunning_fog_index (11) Long-term debt ratio (12) Idiosyncratic volatility

28 Table 3. The Relation between Textual Management and DIPO s Initial Credit Rating This table shows the results of regressing the initial credit rating on the textual management measure. Appendix outlines the definitions for all the variables. *** Significant at the 1% level. ** Significant at the 5% level. * Significant at the 10% level. (1) (2) (3) (4) (5) (6) log_wc 0.85*** 0.65*** (13.19) (10.07) log_fsize 0.43*** 0.32*** (13.26) (9.96) log_wc_unc 0.77*** 0.58*** (12.87) (9.88) offering_mature -0.03*** -0.02*** -0.03*** -0.03*** -0.03*** -0.03*** (-3.42) (-3.04) (-4.17) (-3.46) (-3.29) (-2.93) offer_amt -0.38*** -0.34*** -0.35*** -0.31*** -0.37*** -0.33*** (-6.92) (-7.10) (-6.35) (-6.57) (-6.49) (-6.68) callable 0.87*** 0.70*** 1.13*** 0.86*** 0.94*** 0.75*** (7.20) (6.11) (10.00) (7.73) (7.57) (6.25) def (-0.49) (-0.66) (-0.95) (-1.04) (-0.46) (-0.63) term (-0.63) (-0.37) (-0.37) (-0.13) (-1.04) (-0.68) ted (-0.45) (-0.40) (-0.44) (-0.42) (-0.62) (-0.55) vix (-1.25) (-1.03) (-1.18) (-0.98) (-1.09) (-0.90) public -0.21*** -1.40*** -0.17*** -1.52*** -0.24*** -1.48*** (-3.91) (-10.87) (-3.02) (-11.73) (-4.54) (-11.58) ltdebt_ratio 0.66*** 0.80*** 0.74*** (4.12) (4.92) (4.48) iv 38.98*** 43.52*** 40.37*** (10.70) (11.97) (11.01) Constant *** 3.36*** 4.67*** 4.23*** 5.30*** (0.53) (2.61) (4.04) (6.11) (5.06) (7.07) Observations 1,123 1,123 1,123 1,123 1,123 1,123 Adjusted R-squared

29 Table 4. The Relation between Offer Yield Spread and Textual Management This table presents the results of regressing the offer yield spread on the textual management variable. The dependent variable is the yield spread (cost of debt), which is computed as the bond s offer yield minus interpolated Treasury yield benchmarks. Appendix outlines the definitions for all the variables. *** Significant at the 1% level. ** Significant at the 5% level. * Significant at the 10% level. (1) (2) (3) (4) (5) (6) log_wc 0.36*** 0.30*** (4.03) (3.39) log_fsize 0.15*** 0.12** (2.88) (2.31) log_wc_unc 0.36*** 0.30*** (4.31) (3.67) offering_mature (-1.13) (-1.09) (-1.18) (-1.13) (-1.11) (-1.07) offer_amt -0.50*** -0.51*** -0.47*** -0.49*** -0.49*** -0.51*** (-6.34) (-6.65) (-5.92) (-6.26) (-6.29) (-6.59) callable (0.77) (0.66) (1.22) (1.03) (0.78) (0.66) bb 1.89*** 1.83*** 1.99*** 1.91*** 1.89*** 1.83*** (11.79) (11.76) (12.48) (12.26) (12.09) (12.05) b 3.19*** 3.03*** 3.30*** 3.11*** 3.19*** 3.02*** (25.43) (23.26) (27.39) (24.39) (26.02) (23.75) ccc_n_below 4.19*** 3.95*** 4.31*** 4.04*** 4.19*** 3.95*** (21.07) (19.02) (22.03) (19.53) (21.37) (19.24) def (1.49) (1.38) (1.37) (1.28) (1.49) (1.39) term 0.47** 0.47** 0.47** 0.48** 0.45** 0.46** (2.44) (2.52) (2.45) (2.54) (2.36) (2.45) ted (0.16) (0.25) (0.17) (0.26) (0.09) (0.18) vix (-1.51) (-1.40) (-1.43) (-1.34) (-1.48) (-1.38) public -0.41*** -1.12*** -0.41*** -1.17*** -0.41*** -1.12*** (-4.94) (-5.68) (-4.62) (-5.92) (-4.94) (-5.71) ltdebt_ratio (-1.49) (-1.30) (-1.41) iv 29.13*** 30.71*** 28.97*** (4.79) (5.07) (4.78) Constant 3.52*** 4.52*** 5.00*** 5.85*** 4.86*** 5.64*** (2.60) (3.39) (4.15) (4.94) (4.24) (4.99) Observations 1,123 1,123 1,123 1,123 1,123 1,123 Adjusted R- squared