Journal of accounting, auditing finance tập 28, số 01, 2013 1

Taken together, the results suggest that reputableauditors and underwriters have integral, but different, roles in the bond-issuing process.Keywords reputable auditor, reputable underwri

http://jaf.sagepub.com/ Finance Journal of Accounting, Auditing &

can be found at:

Journal of Accounting, Auditing & Finance

Additional services and information for

The Author(s) 2013 Reprints and permission: sagepub.com/journalsPermissions.nav DOI: 10.1177/0148558X12472129

http://jaaf.sagepub.com

Dedication

This issue is dedicated to Professor Lee J Yao, PhD who passed away on November 14,

2012 due to complications from cancer at the age of 54 He was an expert in ary research on forensic accounting, finance and information systems He published widely

interdisciplin-on issues in these diverse topics, resulting in three books and two book chapters, and morethan forty articles in leading refereed journals His papers have appeared in several leadingjournals on Accounting, Finance, Electronic Markets, Accounting Information Systems,Accounting Education, International Accounting, Information Management, ComputerInformation Systems, and Quantitative Finance

28(1) 4–19 ÓThe Author(s) 2013 Reprints and permission: sagepub.com/journalsPermissions.nav DOI: 10.1177/0148558X12459606

http://jaaf.sagepub.com

Not All That Glitters Is Gold:

The Effect of Attention and

Blogs on Investors’ Investing

on the rational-agent framework Instead, our results suggest that blog effect can be uted to the limited attention theory and cannot be arbitraged due to investors’ self-attribution and short-sale constraints Our research points out the importance of blogs ininformation dissemination, especially for the stocks with limited attention

1 University of Wisconsin, Eau Claire, USA

displaying a long-term return reverse pattern In addition, the abnormal returns of a lio composed of stocks with high institutional ownership meander around zero over thenext 12 months after formation of the portfolio We examine three plausible explanationsfor this return premium, two from the rational-agent framework (liquidity and investor rec-ognition) and one from the behavior finance framework (limited attention with short-saleconstraints) Surprisingly, contrary to media coverage, for which the return premium can beexplained by liquidity and investor recognition based on the rational-agent framework(Merton, 1987), the return premium of blog exposure cannot be explained by that frame-work Instead, our findings show that a blog exposure premium originates from the jointforces of biased disclosure of bloggers and limited attention of consumers Such overvalua-tion becomes larger over time due to consumers’ biased self-attribution, while the short-sale constraints prevent such a premium from being arbitraged Our results are most consis-tent with the notion in Hirshleifer, Lim, and Teoh (2009) In addition, our results show thatthe blog exposure effect is not subsumed by traditional media exposure, as documented byFang and Peress (2009) However, without mainstream media (e.g., news) planting a seed

portfo-of discussion, blog exposure actually has no impact on security returns

Understanding the relation between blog exposure and stock returns is very importantfor the marketing community and the finance community because it demonstrates how afirm’s marketing strategy within blogspaces can influence elements of its finance strategy,such as cost of capital As a form of word of mouth (WOM)1, blogs represent the fastestgrowing medium of personal publishing and the newest method of individual expressionand opinion on the Internet.2 In 2004, blogs were a fairly new phenomenon with only 5million bloggers worldwide (Wright, 2006) However, at the time of writing this article,according to www.BlogPulse.com, there were more than 126 million blogs on the WorldWide Web We believe blogs are playing a role that is as important as that of newspapersbecause (a) information in a blog is not a simple reflection of what is covered by traditionalnews In fact, many blogs address topics that are not covered by the mainstream media atall Blogs might either lead or follow traditional news and (b) blogs disseminate informa-tion to a much broader audience faster and with in-depth analysis In fact, compared withother online media, blogs are viewed as more credible In addition, compared with tradi-tional sources, more than three quarters of respondents view blogs as moderate to verycredible

This article adds to the growing body of studies of the valuation of online WOM ture (Antweiler & Frank, 2004; Tumarkin & Whitelaw, 2001; Wysocki, 1999) Our article

litera-is also closely related to but dlitera-istinct from Fang and Peress (2009)’s media coverage studythat shows that, by helping to reach a broad population of investors, mass media can allevi-ate information frictions and affect stock price even if it does not contain authentic news.However, Fang and Peress focus on studying the impact of traditional media coverage,measured by the number of newspaper articles about a firm on its stock returns We con-centrate on examining the influence of nontraditional media coverage—blog exposure—onsecurity pricing, while controlling for the traditional media coverage Furthermore, ourresults are distinct from Fang and Peress’ conclusions The return premium of news cover-age can be explained by the rational-agent framework (Fang & Peress, 2009) However,such a return premium of blog exposure can only be explained by the joint forces of inves-tors’ behavior and short-sale constraints

Another branch of literature related to our study is the behavior finance literature thatrecognizes that attention influences investors’ selling and purchasing behavior, and causesasset pricing deviation from its fundamental value The underlying reason is that investors

face a formidable search problem when buying a stock Investors address such an issue bylimiting their choice sets (Barber & Odean, 2008) to those stocks that have recently caughttheir attention (Odean, 1999) Because investors are overconfident (Daniel, Hirshleifer, &Subrahmanyam, 1998; Odean, 1998) and biased toward self-attribution (Daniel et al.,1998), stocks over bought heavily by individual investors will enjoy short-term positivecontemporaneous returns These results emphasize the attention driven by traditionalmedia, such as securities mentioned in the news or securities that have gone through largevolume or price changes Our study, however, focuses on the attention driven by nontradi-tional media, blogs.

One practical implication of our results is that a firm’s visibility within blogspaces,regardless of whether the blog discussion is positive or negative, influences investors’ pur-chase decision As a nontraditional media, blog discussions serve the role of disseminatinginformation that was traditionally covered by conventional information channels such asanalysts’ forecasts or newspapers Such a role is especially important for stocks with lim-ited attention In fact, as we documented, because blog discussions do not affect theexpected stock returns when there is a lack of echo from the mainstream media, companiesshould plant the seed of a discussion to foster the conversation within blogspaces.Marketing managers of a firm can use blogs not only to communicate more efficiently withits customers, partners, suppliers, and other stakeholders but also to work closely withfinance managers to lower the cost of capital by delivering information to a broader audi-ence in a faster manner

The rest of this article is organized as follows In the next section, we discuss our datacollection processes, elaborate our variable definitions, and present our empirical results Inthe section ‘‘Explaining the Blog Visibility Effect,’’ we discuss three possible causes of theblog exposure effect, and in the section ‘‘Conclusion,’’ we present our concluding remarks.Data Collection and Empirical Results

Data Collection

In this article, blog visibility/exposure is defined as the extent to which a company’s ucts or services are discussed in blogspaces We collect such a measurement fromwww.BlogPulse.com using its conversation track tool Our data collection is composed oftwo steps In Step 1, we identify the brand names of products or services associated with acompany by searching a company’s web site, reading its financial reports, or usingresearchers’ domain knowledge In Step 2, using the names identified in Step 1 as key-words, for each company, we retrieve its blog visibility over time using the conversationtracker tool provided by www.BlogPulse.com Of the total daily blogging activities traced

prod-by www.BlogPulse.com, this measurement represents the percentage of the total blog versation related to a particular firm, its products, or its services It measures how (andhow much) a firm’s current customers, potential customers, competitors, industry peers,and so on are talking about the products or services of the firm Hence, it represents thevisibility of a firm within the overall blogspaces Figure 1 shows one example of the blogtrend for Advance Auto Parts Inc.3

con-The data on a firm’s media exposure were collected from Factiva Factiva is a database

of the Dow Jones and Reuters companies It provides timely, domestic, and internationalinformation, such as articles from the Dow Jones and Reuter’s newswires and The WallStreet Journal This information covers market data, firm and industry news, financial

quotes, and newspaper articles We collected and counted all the nonredundant news itemseach day for the fiscal year of each company Our analysts’ forecast data were collectedfrom the Institutional Brokers Estimates System The company accounting data wereobtained from CompuStat, and the stock return data were from the Center for Research inSecurity Prices (CRSP) For detailed variable definitions, please refer to the appendix.Following Amihud (2002), we delete one firm with extreme illiquidity value The finalsample contains 404 Standard & Poor (S&P) 500 firms with daily blog visibility and tradi-tional media coverage information from March 1, 2006, to August 22, 2006 We givedetailed variable definitions in the appendix Following Fang and Peress (2009), to mini-mize the noise of daily data, for each firm, we aggregate its daily blog visibility andFactiva news to a monthly level to represent its blog visibility and media coverage, respec-tively Table 1 shows the summary statistics of our key variables, including blog coverageand media coverage As we can see, only 1% of the stocks in our sample do not havemedia coverage, but more than 25% of the stocks in our sample do not have blog coverage.

It seems that for the S&P 500 firms, and for this sample period, traditional media hasbroader coverage in terms of the number of stocks discussed than blog conversations.However, this does not necessarily mean that, in general, traditional news media hasbroader coverage than blogs

Table 2 shows the Pearson correlation among blog coverage, media coverage, and otherfirm characteristics We find blog coverage and media coverage are positively correlated,and analysts, news, and blogs have the tendency to feature the same set of stocks This isreasonable because analysts, news, and blogs pay attention to large firms and well-knownfirms Furthermore, it seems that traditional media cares more about firms with a lower

Figure 1 Blog coverage data collection

Note: In this figure, we use Advance Auto Parts Inc as an example to show how we collect the sure of a firm within blogspaces using www.blogpulse.com

expo-institutional ownership, whereas bloggers do not differentiate between firms with high tutional ownership and those with low institutional ownership.

insti-Blog Coverage and Short-Term Cross-Section of Stock Return

In this section, we study whether a firm’s exposure within blogspaces affects its securitypricing by regressing stock returns on blog coverage, with media coverage, beta, size, BMratio, and other risk factors controlled (Table 3, Model 1) For a robustness check, we take

an incremental approach by adding momentum, media coverage, percentage of institutionalownership, and illiquidity one by one, and present the results in Models 2, 3, and 4 ofTable 3, respectively To control for the potential confounding effect caused by the differ-ence across months and industries, we also include month fixed effect and industry fixedeffect (two-digital standard industrial classification from CRSP) in all our models

For Model 1, the coefficient of blog exposure is 20.0026 with t-value at 21.93, which

is negatively associated with the following month’s stock return Therefore, stocks withhigh blog exposure tend to have lower stock returns compared with the stocks with lowblog coverage Such an impact is not subsumed when we add the traditional media expo-sure and other risk factors (Models 3 and 4) Furthermore, using Factiva as proxy formedia coverage, we found that media coverage has an insignificant coefficient regardless

of whether we exclude the blog exposure The untabulated table shows that our results arequalitatively the same if we exclude those firm-month observations in which there are earn-ing announcements for the firm in that month Therefore, we conclude that the blog effect

is not driven by month effect, industry effect, or earnings announcements

Our media coverage results are different from those of Fang and Peress (2009).However, in their article, they focus on four influential newspapers with large subscrip-tions Our media coverage includes all the newspapers included in the Factiva database.Another reason for the different results might lie in the sample selection In their article,

Table 1 Summary Statistics of Blog Exposure and Media (Factiva) Coverage

they argue that the most significance of high media coverage, low return effect comes fromthe small and illiquid firms However, in our article, we use only S&P 500 firms, which areall big firms.

Blog Coverage and Long-Term Cross-Section of Stock Returns

In the previous section, we documented the short-term return premium associated with lowblog exposure firms In this section, we study the long-term return impact of blog exposure.Figure 2 represents the cumulative abnormal returns of stocks with different blog coverage,starting from Month 1 after the portfolio formation Each month, we sort our sample intothree groups according to their monthly blog exposure Then, based on capital asset pricingmodel, for each stock, we estimate its abnormal return in the subsequent 13 months Lowblog curve (Figure 2) represents the average abnormal returns for the lowest blog coveragegroup, whereas high blog curve represents those of the highest blog coverage group Figure

2 shows that the highest blog coverage portfolio consistently has insignificant abnormalreturns starting from the 1st month after the formation of the portfolio to the next 13

Table 2 Pearson Correlation

Blog Factiva Size BM Momentum Dispersion IO Coverage Illiquidity Idiorisk

months, and the low blog coverage portfolio enjoys a positive coverage abnormal return inthe following 13 months In addition, there is a return reversal when the cumulative abnor-mal returns peak at the 10th month.

Figure 2 Cumulative abnormal return

Note: CAPM = capital asset pricing model; CAR = coverage abnormal return Figure 2 presents thecumulative abnormal return of a high blog coverage sample and a low blog coverage sample startingfrom Month 1 after the portfolio formation Each month we sort our sample into two groups accord-ing to its monthly blog coverage Then, based on CAPM model, for each stock, we estimate its cumu-lative abnormal return in the subsequent 13 months The blue line represents the average CAR forthe low blog coverage group, whereas the red line plots that of the high blog coverage group

Table 3 Blog Coverage, Media Coverage, and Stock Returns

Nextmonthre5a 1 bbeta3 beta 1 bsize3 size 1 bBM3 BM 1 bmomentum3 momentum 1 bIO3 IO 1 bcoverage3 coverage 1 billiquility3 illiquility 3 bblog3 blog 1 bFactiva3 Factiva:

Our results are not driven by the return reversals among no-coverage stocks because theabnormal returns associated with low blog visibility firms do not reverse until 10 monthsafter the portfolio formation, while a typical return reversal pattern among no-news losers

is short lived (Chan, 2003) Given that our blog coverage effect represents a long-termeffect, there must be a force to prevent traders from such arbitrage So what are the drivingforces?

Explaining the Blog Visibility Effect

In this section, we examine three potential explanations of the return premium associatedwith blog exposure: illiquidity, investors’ recognition, and bloggers’ limited attention andshort-sale constraints The first two are based on the rational-agent framework, whereas thethird explanation originates from behavior finance literature

Illiquidity Hypothesis

Fang and Peress (2009) document that stocks with no media coverage enjoy higher returnsthan stocks with high media coverage Furthermore, such a return premium is very stableover time They believe that the lack of liquidity explains why such an abnormal returncannot be arbitraged Hence, we first test whether blog visibility is similar to media cover-age aroused due to illiquidity According to the rational-agent framework, if the blog effectrepresents an arbitrage opportunity, it can only be persistent in the situation where somekind of impediment to trade prevents arbitrage Hence, if the blog visibility effect is alsocaused by illiquidity, we expect the blog effect to be most significant in the portfolios thatare composed of the most illiquid stocks

We use multivariate regressions (either separate regressions for each portfolio or pooledtogether) to study whether the illiquidity can explain the cross-sectional return differences

we document Each month we sort our sample into three portfolios based on various quidity proxies proposed by previous literature including the Amihud (2002) illiquidityratio, bid/ask spread, trading price, and firm size Stocks with the highest illiquidity ratio orspread, or stocks with the lowest price or size, are the most illiquid ones For each illiquid-ity proxy and each portfolio, the following month’s stock returns are regressed on blog cov-erage with other controlled factors that are known to affect the cross-section of returns,such as beta, size, BM ratio, and momentum To control for the potential heterogeneityacross months as well as industries, we run our regression by controlling month and indus-try effects

illi-Table 4 reports the blog effect of stocks sorted by different illiquidity proxies Due tospace constraint, for each portfolio under different illiquidity measures, we report only thecoefficient before blog coverage and the corresponding t statistic If the blog visibilityeffect is caused by illiquidity, we expect the blog exposure effect to be most significant inthe portfolios composed of the most illiquid stocks, such as stocks with the highest illiquid-ity and spread or those with the lowest price or size However, the results shown in Table 4fail to support the illiquidity hypothesis For example, with respect to the Amihud illiquid-ity ratio, blog effect is significant in the low illiquidity portfolio (para = 20.0068 andt-value = 23.04) and the medium illiquidity portfolio (para = 20.0058 and t-value =22.9), but not in the high illiquidity portfolio (para = 0.0022 and t-value = 0.67) This con-tradicts the illiquidity hypothesis that high impediments of trade should result in the mostsignificant blog exposure effect Similar examples of evidence are found when we use

alternative illiquidity measurements Using price or firm size to proxy for liquidity, weobserve that the blog exposure associated return premium is most significant in the portfo-lios composed of the most liquid stocks, for example, stocks with highest trading prices(para = 20.0065 and t-value = 21.68) or stocks with the biggest firm sizes (para =20.0072 and t-value = 23.57) In addition, we also use bid/ask spread to proxy for illiquid-ity because highly illiquid stocks often have a large bid/ask spread The coefficients ofblog visibility for low and medium bid/ask spread portfolios are negative and significantbut not for the high bid/ask spread portfolio (para = 20.0015 and t-value = 20.55)—themost illiquid stocks Overall, our results show that the blog visibility effect disappearsamong the most illiquid stocks, hence are contrary to the illiquidity theory that suggeststhat the blog visibility effect should be the strongest for most illiquid securities.

To check the robustness of the above results, for each illiquidity proxy, we run a pooledregression using all three portfolios To be more specific, for each illiquidity proxy, wefirst define one dummy variable to distinguish high illiquidity firms from low illiquidityfirms For example, the dummy variable Rank for Amihud illiquidity is set to one if the illi-quidity ratio of that firm is higher than the median Amihud illiquidity ratio, and zero other-wise Then, in addition to the original independent variables, we also include an interactionterm between the blog coverage variable and the illiquidity dummy variable Rank to testwhether the blog exposure effect is more significant for stocks with higher illiquidity Ouruntabulated results show that consistent with the conclusions from running separate regres-sions, results using pooled regressions also demonstrate that illiquidity cannot explain theblog exposure effect that we have documented

Investors’ Recognition Hypothesis

Fang and Peress (2009) posit that, in addition to illiquidity, the investor recognition theoryproposed by Merton (1987) can also explain the media coverage effect they observe Underthis theory, investors are assumed to have incomplete information and are aware of only asubset of the available stocks For stocks with less investor recognition, investors withincomplete information will require higher returns as a compensation for the undiversifiedrisk and market clearing Therefore, we hypothesize that if blog coverage can increase astock’s recognition, then the blog effect should be more prominent for stocks with a lowdegree of investor recognition Several measures, including analyst coverage, idiosyncratic

Table 4 Illiquidity and the Blog Effect

illiquid-risk, advertising expenditure, and the number of investors of a stock, are adopted as proxiesfor investors’ recognition Analyst coverage is selected because investors listen to analysts

to make investment decisions, and stocks with high analyst coverage are assumed to have ahigh degree of investor recognition Idiosyncratic risk is adopted as another investor recog-nition measure because it represents the imperfect diversification driven by a lack of inves-tors’ recognition (AHXZ 2006) Furthermore, following previous literature (Grullon,Kanatas, & Weston, 2004; Singh, Faircloth, & Nejadmalayeri, 2005), we develop two addi-tional investor recognition measures: advertising expenditure and the number of commonshareholders of a security

Following the methods in the illiquidity hypothesis test section, we examine whether theblog effect can be explained by investors’ recognition using a separate regression approachand a pooled regression approach For the separate regression methods, again we sort oursample into three portfolios in each month according to different investor recognition mea-sures Then, for each investor recognition proxy and each portfolio, following months’returns are regressed on blog visibility while controlling for other factors that are known toaffect the cross-section of returns such as beta, size, BM ratio, and momentum (Table 5).Similarly, as in the section ‘‘Illiquidity Hypothesis,’’ we also ran pooled regressions tostudy the impact of investor recognition measures on stock returns and reach very similarconclusions as when we followed a separate regression approach

Table 5 presents the results of separate regressions for portfolios with different degrees

of investor recognition Following Table 4, for each investor recognition proxy, we reportonly the coefficient and t-value of blog coverage in each portfolio As we can see, wheninvestor recognition is measured by idiosyncratic risk, the coefficient of blog visibility issignificant only for the low and medium idiosyncratic risk portfolios, but not for the highidiosyncratic risk portfolio In addition, the absolute value of the coefficient of the lowidiosyncratic risk portfolio is bigger than that of the medium idiosyncratic risk portfolio.This indicates that the blog effect is more significant in portfolios with high investor recog-nition, failing to support the hypothesis that the blog exposure effect can be explained byinvestors’ recognition We reach a similar conclusion when we use analyst coverage, adver-tisement expenditure, and number of investors as proxies for investor recognition The blogeffect is much more prominent for stocks with higher analyst coverage, medium advertise-ment expenditure, and a medium investor base None of these evidences support the inves-tor recognition hypothesis

Table 5 Investor Recognition and the Blog Effect

recogni-a stock The following months’ stock returns recogni-are regressed on blog visibility, betrecogni-a, size, BM rrecogni-atio, recogni-and momentum

as well as month fixed effect and industry fixed effect For brevity, for each portfolio under different illiquidity sures, we report only the coefficient before blog coverage and the corresponding t statistic.

mea-Limited Attention Hypothesis and Short-Sale Constraints

Based on the rational-agent framework, the illiquidity hypothesis and the investors’ nition hypothesis are able to explain how news media coverage drives the cross-section ofstock returns (Fang & Peress, 2009) However, such a rational-agent framework cannot jus-tify the blog exposure effect we have documented in the section ‘‘Data Collection andEmpirical Results.’’ Given that the blog exposure effect persists over time, there must beother forces driving this return premium and preventing it from being arbitraged

recog-In this section, we propose that the limited attention theory (Hirshleifer et al., 2009)offers an explanation for the cross-sectional return differences we document, while theshort-sale constraints sustain such return premium over time From a disclosure prospec-tive, blogs serve as a positively biased disclosure channel Investors with limited attentionwill selectively interpret the biased disclosure by assuming that no news is good news Thiswill make investors net buyers of low blog exposure stocks The biased interpretation beha-vior of investors on blog disclosure is consistent with the framework proposed byHirshleifer et al (2009) about how investors interpret the disclosures of those firms withlimited attention To make matters worse, investors with limited attention are overconfident(Daniel et al., 1998) and biased toward self-attribution (Daniel et al., 1998) In other words,individual investors believe they are better in assessing blog information (i.e., overconfi-dent) and they selectively trust the messages in WOM and selectively validate their beliefs(i.e., biased toward self-attribution) The overall net buying behavior will be sustained over

an even longer period of time

Are investors net buyers of low blog exposure securities? To prove that consumers are morelikely to be net buyers of securities with low blog exposure, we estimate the percentages ofbuy transactions out of the daily total number of transactions for a high blog exposure portfo-lio and a low blog exposure portfolio, respectively (Figure 3) We follow the buy and sellclassification algorithm proposed by Lee and Ready (1991), which is commonly used inearly literature, such as Easley, Kiefer, O’Hara, and Paperman (1996), and Easley, Hvidkjaer,and O’Hara (2002) Following Lee and Ready (1991), a transaction is defined as a buy (sell)

if it is executed above (below) the midpoint of the bid and ask price For trades on the bid/ask midpoints, we use a ‘‘tick test’’ to determine whether it is a buy or sell To be more spe-cific, a trade is a buy (sell) if it is executed at a higher (lower) price than the previous trade.For those trades that have the same price as the previous trade, we look at the historical price

Figure 3 Percentage of buy over time

Note: This figure presents the percentage of buy transaction for firms with the low daily blog age during the subsequent 30 days after portfolio formation We sort our sample into 2 portfoliosbased on their daily blog coverage, and low (high) daily blog coverage firms include those whose dailycoverage is lower (higher) than the median blog coverage on that day We also conduct a similar anal-ysis by sorting our sample into 4 or 10 portfolios and reach a similar conclusion

cover-until we find a change in the trade price Following Lee and Ready, for the computation, wematch our trade prices with 5-s-old quotes After classifying the buy and sell for each trade,

we cumulate the number of buys and sells in a day to get the aggregated daily number ofbuy and sell Then the daily percentage buy is defined as the total number of buy transactionsdivided by the total number of all transactions on that day

Figure 3 presents the percentage of buy transactions for firms with low daily blog age for the 30 days subsequent to portfolio formation We sort our sample into 2 portfoliosbased on their daily blog coverage Low (high) daily blog coverage firms include thosewhose daily coverage is lower (higher) than the median blog coverage on that day We alsoconduct a similar analysis by sorting our sample into 4 or 10 portfolios and reach a similarconclusion Our results show that, regardless of the magnitude of blog exposure, on aver-age, there is an increase in the percentage of buy transactions after the portfolio formation,which might be driven by the overall market situation Furthermore, Figure 2 shows that,compared with the high blog exposure portfolio, the low blog exposure portfolio has ahigher percentage of buy transactions.4The untabulated mean difference comparison alsovalidates this conclusion (difference = 0.0022 and t-value = 3.99) In addition, this differ-ence remains relatively stable over time after portfolio formation In addition, the untabu-lated results show that most of the increase in the percentage of buy transactionsconcentrates on small investors

cover-Short-sale constraints In our previous section, we documented that because investors withlimited attention become net buyers of low blog exposure securities, stock prices go up.How about institutional investors or other rational investors? Why do they not come in andfix this ‘‘irrational’’ behavior? We believe that the short-sale constraints might be one ofthe answers Short-sale constraints can prevent the arbitrage from happening, hence makingthe return pattern we observe last longer.5We expect, if short-sale constraints can explainthe blog exposure effect, that effect should be concentrated in a portfolio composed ofstocks with the highest short-sale constraints Following Asquith, Pathak, and Ritter (2005),

we use institutional ownership of a firm as a proxy for its short-sale constraints, and a firmwith high institutional ownership is treated as the one with low short-sale constraints.Table 6 shows that the blog effect is significant only in low institutional ownershipgroups (para = 20.0067 and t-value = 22.58), representing a high sale constraints situa-tion Such evidence supports our hypothesis that the blog effect we documented can beexplained by short-sale constraints

Relationship Between Blog Exposure and Media Coverage

Our readers might question how securities with low blog exposure get attention If nobodyever talks about a stock, how can investors be aware of it? Hence, in this section, we study

Table 6 Short-Sale Constraints and the Blog Effect

Note: IO = institutional ownership; BM = book to market In this table, in each month, we sort our sample into three portfolios according to their short-sale constraints proxy, the institutional ownership Then the following month’s stock returns are regressed on blog visibility, beta, size, BM ratio, and momentum, as well as month fixed effect and industry fixed effect For brevity, for each portfolio, we report only the coefficient and t-value of blog coverage

where the attention comes from We believe that blog exposure needs to be built on newsattention Hence, we investigate whether our blog effect will be subsumed under therecently documented media effect (Fang & Peress, 2009) We first sort our sample intothree portfolios according to their monthly media coverage, proxied by Factiva Then foreach portfolio, we regress the following month’s stock returns on blog coverage, size, beta,

BM ratio, and momentum Table 7 shows that, with traditional media coverage controlled,the blog effect is significant only in high media coverage portfolios and is insignificant inlow and medium media coverage portfolios Our interpretation is that blog conversationsare not a simple reflection of the information content of traditional media; hence, the blogeffect cannot be fully subsumed by it As a result, to really force the market to listen toblog conversations, mainstream media need to plant seeds to spark the discussion; other-wise, a simple blog conversation without a mainstream echo will have no market response.Furthermore, even with heavy traditional media coverage, blog conversations will not beburied and will still stimulate market responses

Conclusion

In this article, we investigate the relation between blog coverage and the cross-sectionalstock returns We show that blog coverage is different from the traditional media coveragedocumented by previous literature (Fang & Peress, 2009) We find that high blog coverage

is associated with low stock returns, even when controlling for other risk factors and tional media coverage We further illustrate that such an effect is more prominent forstocks with low institutional ownership and cannot be explained by either the illiquidityhypothesis or the investor recognition hypothesis, which have been shown in explaining thecross-sectional relation between media coverage and expected stock returns Our interpreta-tion is that the blog coverage effect is caused by the selective interpretation of investorswith limited attention on the blog posting The abnormal returns associated with the blogexposure effect are sustained over time and cannot be arbitraged within a short period oftime due to short-sale constraints All these things make blogs an important informationdissemination channel

tradi-However, we should carefully interpret our results because there may be some othermechanism that might cause the same phenomena we documented For example, if overallblog contents are negative instead of positive, then ‘‘no news is good news’’ might actually

be a rational response in the blog coverage context If investors with limited attention fail

to understand such a relation, then investors with limited attention are likely to be tic about firms with low blog coverage compared with rational investors, which will lead toundervaluation in the current period If stocks with low blog coverage are undervalued inthe current period, subsequently they outperform stocks with high blog exposure Even

pessimis-Table 7 Relationship Between Factiva and Blog

Note: We sort our sample into three portfolios in each month according to their media coverage, proxy by Factiva Then the following month’s stock returns are regressed on blog visibility, beta, size, BM ratio, and momen- tum as well as month fixed effect and industry fixed effect For brevity, for each portfolio, we report only the coef- ficient and t-value of blog coverage.

though for our data set, our untabulated sentiment mining results rule out such an tion,6 future research might look deep into this issue with new data set covering big,medium, and small firms, and across a longer period of time to draw more insights.Our study offers great insights as to the importance of the marketing activities of a firm

explana-on its expected return Our results justify the buzz value creatiexplana-on of marketing strategy explana-onfirm’s valuation Our study also has policy implications for government agencies, such asthe Securities and Exchange Commission because it brings to the forefront the effect andimportance of blog information in the market valuation of firms This is especially impor-tant for those firms with many small and naı¨ve investors, who have limited channels toaccess and limited capabilities to process/digest value-relevant information

Appendix

Variable Definitions

Brokers Estimates System (I/B/E/S) It is based on the variable Numest(number of estimates), which represents the total number of estimatorscovering the company for the fiscal period (annual forecast only)

previous 60-month data

database It is based on stdev/medest, where stdev represents the standarddeviation of the forecast and medest represents the median estimation offorecast for the fiscal period (annual forecast only)

based on the variable XAD

Factiva

It is based on CSHR collected from CompuStat

Research in Security Prices’ (CRSP) monthly file, where PRC is the closingprice and SHROUT is the common shares outstanding

trading volumes during that month

We would like to thank the Editor-in-Chief Kashi R Balachandran and an anonymous reviewer aswell as workshop participants at the 2010 American Accounting Association Annual Meeting forhelpful suggestions

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/

or publication of this article

applies machine-learning and natural-language processing techniques to discover trends in thehighly dynamic world of blogs

3 We admit that there might be some noise in the blog exposure proxy, depending on the typesand the number of blogs www.BlogPulse.com covered However, given the large number of

www.BlogPulse.com is representative of the overall discussion on blogspaces Furthermore, eachsecurity’s blog exposure as retrieved from www.BlogPulse.com might either underestimate oroverestimate the true volume of blog conversation, depending on the keywords we specify aswell the algorithm www.BlogPulse.com uses to identify the information related to a firm.However, we believe the results of such an estimate noise are more likely to be biased againstour findings

4 We should be aware that there is another mechanism that might result in the return differencebetween the high blog exposure stocks and the low blog exposure stocks Investors are net sellers

of the securities with high blog exposure However, Figure 2 results rule out such a possibility

If that were true, the portfolio composed of high blog exposure stocks should have big and longed negative returns, but that is not the case The portfolio return of high exposure stocks is,

pro-in fact, meanderpro-ing around zero

the model of Merton (1987) under the more traditional rational-agent framework, whereas thelimited attention refers to the behavior financial model (e.g., Barber & Odean, 2008).Even though one of the key assumptions of the investor recognition hypothesis (tested in thesection ‘‘Investors’ Recognition Hypothesis’’) is that investors know about only the subset

of the available stocks, Merton’s (1987) model is built on the rational-agent framework and isdifferent from the limited attention hypothesis we study here In Merton’s model, attentiongrabbing by itself will not influence an investor’s purchase decision, whereas in our case, itdoes

6 Our unreported sentiment mining results of the blog contents collected from LexisNexis databaseshow that overall blog contents are dominated by positive sentiments

of the Academy of Marketing Science, 33, 432-444.

Tumarkin, R., & Whitelaw, R (2001) News or noise? Internet postings and stock prices FinancialAnalyst Journal, 57, 41-51

Wright, J (2006) Blog marketing New York, NY: McGraw-Hill

Wysocki, P D (1999) Cheap talk on the web: The determinants of postings on stock message boards(Working Paper) Ann Arbor: University of Michigan

28(1) 20–52

Ó The Author(s) 2013 Reprints and permission: sagepub.com/journalsPermissions.nav DOI: 10.1177/0148558X11421673

http://jaaf.sagepub.com

The Role of Reputable

Auditors and Underwriters in

the Design of Bond Contracts

Yun Lou1 and Florin P Vasvari1

Abstract

The authors empirically test the certification hypothesis by studying the roles of reputableauditors and bank underwriters in the design of bond contracts The certification hypoth-esis suggests that reputable capital market intermediaries can credibly communicate insideinformation to outside investors, thereby helping improve financing terms for firms thatraise external funding Consistent with this hypothesis, the authors provide evidence thatreputable auditors and underwriters help corporate bond issuers obtain lower bond yields.The effect of reputable auditors on the yields is greater than that of reputable underwriters

in terms of economic magnitude and significance, consistent with auditors’ multiple roles asinformation intermediaries, monitors, and insurance providers The authors also find thatthe presence of reputable auditors and underwriters affects bonds’ nonpricing terms Firmsthat hire reputable auditors obtain longer term bonds, whereas those that engage reputableunderwriters can issue larger bonds Taken together, the results suggest that reputableauditors and underwriters have integral, but different, roles in the bond-issuing process.Keywords

reputable auditor, reputable underwriter, bond terms, certification hypothesis

Several theoretical articles suggest that third-party information intermediaries can certifythe quality of security-issuing firms that face significant information asymmetries in capitalmarkets (i.e., the Certification Hypothesis) For instance, the models of DeAngelo (1981),Beatty and Ritter (1986), Booth and Smith (1986), and Titman and Trueman (1986) exam-ine how bank underwriters and auditors help resolve information asymmetries of issuingfirms These theories argue that underwriters and auditors use their reputation capital as abonding mechanism to credibly certify the information about the future prospects of theissuing firms, thereby helping improve firms’ financing terms when raising external financ-ing In this article, we empirically investigate the certification hypothesis in the primarybond market by combining the role of auditors and underwriters Specifically, we study theroles of reputable auditors and bank underwriters in the design of bond contracts

Auditors and underwriters are important information intermediaries in the bond market,

a market that has received little attention so far despite the fact that it provides the mostsignificant source of external financing to U.S firms.1 Auditors play a role in certifyingthat the accounting information provided in the bond prospectuses by issuing firms is accu-rate and prepared in accordance with Generally Accepted Accounting Principles In addi-tion to the certification function, auditors have a monitoring role, which they fulfill byreporting potential errors in financial statements and violations of covenants set in bondcontracts Auditors also bear legal liability for accounting irregularities that occur in thereports of the firms they audit and, under certain conditions, provide bond investors with ameans to indemnify their losses (e.g., Dye, 1993; Mansi, Maxwell, & Miller, 2004).2Thesemonitoring and insurance roles complement auditors’ certification role and potentiallymake auditors more relevant information intermediaries to bondholders than underwriters.Although underwriters certify information about the future prospects of issuing firms anduse their extensive distribution networks and selling expertise to help issuing firms placebond securities, their liability is limited to situations where negligence is proven Hence,underwriters typically do not provide insurance against investment losses Furthermore,they have a limited monitoring role after a bond is issued.3

High-quality bond issuers are likely to signal their type by seeking certification fromreputable auditors and underwriters Auditors and underwriters develop reputation capital

by repeatedly entering into the market and providing credible information about the issuingfirms As a result, the value of their reputation capital likely exceeds even the largest possi-ble one-time gain that could be obtained from certifying falsely Rational investors shouldunderstand these incentives and thus provide capital under more favorable terms to thefirms certified by intermediaries with reputation capital at stake

To test these arguments in the bond market, we first construct reputation proxies forauditors and underwriters We designate an auditor as a reputable auditor if its marketshare based on the clients’ sales is the largest in the industry and outpaces the rest of audi-tors by at least 10% (Dunn & Mayhew, 2004; Palmrose, 1986) We define reputable audi-tors at the industry level because the prior literature shows that industry expertise possessed

by auditors can affect managers’ earnings management behavior and reduce informationasymmetry between firms and investors (e.g., Almutairi, Dunn, & Skantz, 2009; Balsam,Krishnan, & Yang, 2003) As most issuers in the bond market hire large auditors, our focus

is only on companies audited by the big four/five auditors We define an underwriter asreputable if its market share, as captured by the bond volume advised in the whole bondmarket, persistently ranks among the top five underwriters in the past 3 years

Consistent with the certification hypothesis, we find that hiring reputable auditorsreduces bond issuance yields by 35 basis points, which is both statistically and economi-cally significant This decrease in bond yields translates into annual interest savings ofUS$65,450 for the average bond issue in our sample Reputable underwriters also helpissuers lower the yields by 19 basis points, a significantly weaker effect than that of reputa-ble auditors The greater impact of hiring reputable auditors is consistent with auditors’multiple roles as information intermediaries, monitors, and insurance providers

We further examine whether reputable auditors and underwriters provide value withrespect to nonpricing terms of bond contracts, such as bond maturity and size.4Debt matu-rity plays an important role in reducing agency costs associated with asset substitution andimproving the efficiency of monitoring by lenders (Leland & Toft, 1996; Stulz, 2000).Short bond maturities may reduce agency costs by subjecting managers to more frequentmonitoring by investors and rating agencies (e.g., Datta, Iskandar-Datta, & Raman, 2005)

However, hiring a reputable auditor may provide an alternative monitoring mechanism toreduce these costs because reputable auditors are incentivized to monitor issuing firms’financial reporting continuously to maintain their reputation in the industry As a result,issuing firms with reputable auditors may potentially borrow from bondholders for a longerperiod compared with those with ordinary auditors Consistent with our conjecture, we findevidence that hiring reputable auditors, on average, lengthens bond maturities by 2.54years, a statistically significant effect We do not find a strong substitution effect betweenthe presence of reputable underwriters and bond maturity This could be explained by thenature of the underwriters’ job Underwriters are responsible for marketing and sellingbonds; however, once the bonds are issued, they do not have any monitoring role.

Finally, we examine the impact of reputable auditors and underwriters on bond size,which is an indicator of the issuing firm’s repayment ability If an issuing firm has a higherlevel of tangible assets and/or is able to generate larger future cash flows, it can borrow moredebt To the extent that reputable auditors and underwriters can certify the accuracy of tangi-ble assets and the ability to generate cash flows to pay the debt back, issuing firms with repu-table auditors and underwriters may be able to borrow a larger amount than those withordinary auditors and underwriters In addition to the certification effect, the size of a bondissue also depends on underwriters’ marketing and selling abilities Reputable underwritershave extensive distributional networks and superior selling power, which allow them to placelarger bond issues Consistent with these arguments, we find evidence that reputable under-writers have a strong positive effect on the size of the bond Specifically, hiring reputableunderwriters increases the actual offering amount by 13.73% relative to the average offeringamount in our sample We do not find a similar result for reputable auditors, suggesting thatthese information intermediaries play a different role in the bond-issuing process

Our article makes two significant contributions to existing knowledge on the value ofauditors and underwriters in the bond market First, we bridge two disconnected strands ofliterature by testing the certification hypothesis simultaneously for reputable auditors andunderwriters in the bond market Although Mansi et al (2004) and Ahmed, Rasmussen,and Tse (2008) have explored the role of auditors in reducing bondholder–shareholder con-flicts, reputable underwriters were not considered as additional intermediaries in the analy-sis By combining the certification role of reputable auditors with that of reputableunderwriters, we highlight different roles played by these important capital market interme-diaries with respect to the structuring of public debt financing

Second, our article contributes to the growing body of literature that examines moredetailed aspects of debt contracts (e.g., Brockman, Martin, & Unlu, 2010; Qian & Strahan,2007) Previous studies in the bond or syndicated loan market attempt to understand thedrivers of a single, contractual dimension (typically the bond yield or the loan spread).5Weprovide unique evidence on the effects of reputable auditors and underwriters on the bondmaturity and size As a result, this study sheds light on the role of information intermedi-aries with reputation capital on the nonpricing terms of bond contracts and examines thejoint role of auditors and underwriters in a richer setting than other articles that focused onthe equity market

The remainder of the article proceeds as follows: Section titled ‘‘Related Literature andHypotheses’’ provides a discussion of related literature and formalizes our hypotheses, sec-tion titled ‘‘Data and Research Design’’ describes empirical strategies and data and is fol-lowed by the ‘‘Results’’ section that presents the main results, section titled ‘‘SensitivityAnalyses and Additional Tests’’ offers some robustness checks, and the final section titled

‘‘Conclusion’’ concludes the article

Related Literature and Hypotheses

The certification hypothesis is derived from the literature on the use of reputation capital toguarantee product quality (Klein & Leffler, 1981) As an extension to this theoretical litera-ture, DeAngelo (1981) shows that when incumbent auditors earn client-specific quasi rents,auditors with a greater number of clients have more to lose by failing to report a discoveredbreach in a client’s accounting system The higher the value placed by large auditors ontheir reputation, the better is the quality of their audits Consistent with this argument, themodel of Titman and Trueman (1986) finds that firms that hire high-quality auditorsreceive greater valuations when securities are issued Similarly, Booth and Smith (1986)model underwriter reputation as a bonding mechanism to solve the information problemsbetween issuing firms and investors, and find that underwriter reputation is formed eitherthrough a premium price charged for quality assurance or the objective of maintaininglong-term profits through repeated entries into the market The models of Chemmanur andFulghieri (1994) or Beatty and Ritter (1986) provide similar arguments by showing thatinvestment banks’ reputations are achieved by adopting stringent evaluation standards.Taken together, these theories imply that reputation capital can provide capital marketintermediaries such as auditors and underwriters with incentives to commit to honest infor-mation production on the firms they serve

By providing more accurate information, these information intermediaries allow outsideinvestors to make more precise estimates of firm values and better investment decisions

As intermediaries with reputation capital at stake can be adversely and materially affected

if their information certification proves false, investors may accept less protection on thesecurities issued by firms hiring these intermediaries Therefore, we hypothesize that bothauditors and underwriters with reputation concerns play certification roles that help reduceissuers’ cost of debt or relax the nonpricing terms of their debt contracts

Empirical studies have examined the certification roles of auditors and underwriters arately Pittman and Fortin (2004) and Mansi et al (2004) find that the cost of debt islower for firms with larger auditors Ahmed et al (2008) show that industry audit special-ists help firms reduce the cost of capital, both equity and debt Empirical evidence infinance, however, finds that reputable underwriters obtain lower yields and charge higherfees (e.g., Fang, 2005) However, auditors and underwriters have integral, but different,roles in the bond-issuing process, and ignoring either in empirical analyses can lead toimprecise inferences of their respective contributions

sep-The theoretical model developed by Balvers, McDonald, and Miller (1988) provides dance for our empirical analysis of auditors and underwriters in the bond market Themodel shows that investment banks with reputation concerns are more likely to select high-reputation auditors as a signal of their own quality, and together, they reduce the underpri-cing of initial public offerings of equity issues The model also predicts that highly reputa-ble auditors and underwriters have divergent effects on underpricing—as the reputationeffect of one intermediary increases, the effect of the other diminishes We expect thesefindings to apply to the bond market, as well, for several reasons First, auditors provideassurance that firms’ financial statements are prepared in accordance with GenerallyAccepted Accounting Principles, whereas underwriters assist firms in documenting, market-ing, and selling securities Hence, the information content of both certification roles candiffer, with auditors verifying accounting information before and after a bond is issued, andunderwriters affirming to general future prospects about bond issuers

gui-Second, auditors are at high risk for litigation and play an additional insurance role byindemnifying investors against disclosures of false accounting information In recent years,the litigation against auditors has grown dramatically, both in frequency and cost.6The pas-sage of the Sarbanes-Oxley Act further expanded the legal responsibility of auditors, requir-ing them to report on the adequacy of client firms’ internal control over financial reporting.

In addition, auditors—especially those with reputation capital at stake—incur indirect costsfrom litigation, such as loss of reputation capital If investors recognize the relatively highlitigation costs associated with reputable auditors in the event of failure to detect account-ing irregularities, they may place more value on the certification role of reputable auditorsthan on that of reputable underwriters Therefore, given the differences in the certifiedinformation content and exposed litigation costs that exist between auditors and underwri-ters, we expect reputable auditors to play a stronger role in reducing the cost of debt thanreputable underwriters

Although the theoretical predictions about the certification effect on the credit spreadsare clear, inferences about certification’s role on the negotiated nonpricing bond terms areless straightforward Debt maturity is one of the main nonpricing terms of a bond contractand is well regarded as an ex-post monitoring device For example, Leland and Toft (1996)argue that short-term debt reduces or even eliminates the agency costs associated with assetsubstitution Also, Stulz (2000) illustrates that short-term debt provides creditors with anextremely powerful tool to monitor the borrowing firm’s management Managers withhigher stock ownership choose a larger proportion of short-maturity debt, thereby commit-ting to more frequent monitoring (Datta et al., 2005) Auditors, too, play a role in monitor-ing issuing firms’ financial reporting systems In particular, auditors with a reputationconcern have stronger incentives to assure the quality of financial reporting throughout theperiod when debt is outstanding Given reputable auditors’ incentives to facilitate ex-postmonitoring of issuers, one could expect either a substitution or a complementary effectbetween the presence of reputable auditors and the negotiation of a shorter debt maturity

In contrast, the underwriters’ main role is to assist borrowers only at issuance; they have

no responsibility to monitor borrowers after issuance As a result, we do not expect an ciation between the presence of reputable underwriters and bond maturity

asso-Another important nonpricing term of a bond contract is the size of the bond issue Thesize is associated with default risk—the larger the bond, the greater the pressure on itsissuer’s repayment ability To the extent that reputable auditors and underwriters reduce theinherent uncertainty associated with the measurement of default risk at issuance, one wouldexpect an increase in the bond sizes of issuers with these types of intermediaries The size

of a bond issue is also a function of the distributional networks and selling abilities of theunderwriter Reputable underwriters have extensive distributional channels, strong relation-ships with institutional and individual investors as well as superior marketing and sellingabilities, all of which facilitate the issuance of larger amounts of debt (Fang, 2005) Takingthis into account, we expect that reputable underwriters potentially play a more importantrole in increasing the size of the bonds issued when compared with reputable auditors.Data and Research Design

Proxies of Reputable Auditors and Underwriters

To capture the reputation concerns of auditors and underwriters, we measure their tion capital based on the magnitude of their respective market share This is consistent with

reputa-the reputa-theoretical argument that if an information intermediary, such as an auditor or writer, engages in quality cutting, this information disseminates faster if the intermediaryhas a large market share (e.g., Klein & Leffler, 1981) Furthermore, with a large marketshare, the expected long-term fee premium from information intermediaries’ reputation isalso likely to exceed the short-term benefits that could be obtained by misinforming inves-tors Therefore, the market share reflects the income stream at stake, and larger auditors orunderwriters have more to lose from a damaged reputation.

under-We measure a reputable auditor’s market share using the total sales audited by an tor within an industry (Dunn & Mayhew, 2004; Palmrose, 1986) We focus on the certifica-tion role of auditors specializing in a particular industry because they are associated withhigh-quality audits (Craswell, Francis, & Taylor, 1995; Krishnan, 2003) Becoming anindustry specialist requires a significant investment in training and time to establish a solidreputation Also, industry audit specialists have a large market share, as their expertise isrecognized and they are sought out within the industry As a result, consistent withDeAngelo’s (1981) argument, they have more to lose if they fail to detect frauds in theirclients’ audits

audi-We define an industry as all firms with the same two-digit primary Standard IndustryClassification (SIC) code in the Compustat universe.7We designate an auditor as a reputa-ble auditor if its market share is the largest in the industry and outpaces the rest of auditors

by at least 10% The 10% cutoff supports our inferences on the qualitative differences inauditors’ reputations in a particular industry In checking for robustness, we also confirmthat using a 15% or 5% cutoff does not alter the robustness of our results Furthermore, wevalidate this measure by investigating the association between the presence of industryaudit specialists and the accounting and governance risks of the firms that hire them.8Although auditors provide services for the universe of public firms and are pressed to dif-ferentiate themselves through industry specialization, underwriters in the debt market, which

is not as competitive as the equity market, tend to focus on multiple segments For instance,Yasuda (2005) documents that underwriters’ bank relationships with borrowers have a posi-tive and significant impact on their bond underwriting business.9 Therefore, we use themarket share based on the underwriter’s volume in the whole bond market to identify reputa-ble underwriters We define an underwriter as reputable if its market share persistently ranksamong the top five underwriters in the past 3 years.10 The intuition behind this measure isthat an underwriter with a large market share will not imperil its reputation for the sake ofshort-term profits Underwriters with a large market share extract economic rents on reputa-tion from their clients (Fang, 2005) Moreover, they are repeat players, and the poor perfor-mance of a bond not only damages their reputation in the bond market but could also affecttheir businesses in other areas, such as bank lending, equity underwritings, or Mergers &Acquisitions Advisory services In robustness checks (see section titled ‘‘Sensitivity Analysesand Additional Tests’’), we also present results using the top eight underwriters and classify-ing reputable underwriters based on the number of bonds they place

Regression Specifications

This section presents the regression specifications concerning the effects of reputable tors and underwriters on the bond terms To examine the certification roles of reputableauditors and underwriters on bonds’ spreads, we estimate the following regression (wepresent the computation of all variables in Appendix A):

audi-Credit Spread 5 a1b Reputable Auditor1r Reputable Underwriter

1uInteraction Term1p Firm Controls1y Bond Controls1sIndustry and Year  Fixed Effects1e:

ð1Þ

The dependent variable Credit Spread represents the risk premium that investors require

to hold the issuer’s bond, taking into consideration the effect of business cycles CreditSpread is a better proxy for the cost of debt than interest expense used by prior studies forseveral reasons First, the interest expense field pools together the cost of debt with differ-ent types of lenders, different maturities, and security features Banks rely less on auditorsthan bondholders because they have access to private information that is not reflected inpublic financial statements Even within public debt, debt securities are not homogeneous,and they cannot be pooled easily unless their distinctive features are controlled for in theempirical tests Second, the interest expense field from Compustat includes other items thatare irrelevant to the cost of debt, such as amortization of expenses associated with debtissuance.11 Third, the interest expense is not adjusted for treasuries; thus, the measuremoves with the interest rate environment or macroeconomic conditions, which may createspurious correlations

Reputable Auditor and Reputable Underwriter are indicator variables equal to ‘‘1,’’ if anissue is certified by reputable auditors or reputable underwriters, and ‘‘0’’ otherwise We alsoinclude an interaction term, Reputable Auditor and Reputable Underwriter to examine whetherthe effects of reputable auditors and underwriters on credit spreads vary with each other

We control for firm-specific variables to account for cross-sectional differences in creditspreads beyond the effect of hiring reputable auditors and underwriters We include auditortenure (Tenure) as it has been shown to be negatively related to credit spreads because ofits role in reducing the information asymmetry between the issuer and the investors (Mansi

et al., 2004) Firm size and leverage are proxies for the issuer’s financing risk We measurethem as the natural log of total assets of the issuer (Firm Size) and the ratio of long-termdebt to total assets (Leverage), respectively To control for the issuer’s risk of repaying thedebt and the coupons, we also include the asset tangibility computed as net property plantand equipment scaled by total assets (Tangibility) and the return on assets (ROA) in theregression

Furthermore, we control for bond variables in our analysis Although put options orsinking fund features add safety to bond issues and are expected to be negatively related tocredit spreads, call options or subordinated clauses put bondholders at a disadvantage and,therefore, are priced in the risk premium We include maturity and bond size in the regres-sion of credit spreads, as the potential losses to bond investors increase in the time horizons

of repayments and the offering amounts Furthermore, restrictive covenants mitigate thewealth transfer from bondholders to shareholders Investors demand more covenants if therisk of wealth expropriation or asset substitution is more severe Hence, we use the number

of covenants to account for the riskiness of bond issues due to a higher agency cost ofdebt

Credit-rating agencies, as information intermediaries, also provide independent ments of the issuer’s credit risk As they have access to the issuer’s private information,their opinions are valued by bondholders Therefore, we include credit-rating information

assess-as a control in our analysis of credit spread We designate an indicator variable SpeculativeGrade equal to ‘‘1’’ to the issues rated below a BBB rating by Moody’s or Standard &Poor’s (S&P’s), and ‘‘0’’ otherwise

We then examine the relationship between reputable auditors and underwriters and bondmaturities, controlling for the risk of bonds as well as other factors known to influencebond maturities The specification of the maturity regression is as follows:

Maturity 5 a1b Reputable Auditor1r Reputable Underwriter

1uInteraction Term1p Firm Controls1y Bond Controls1sIndustry and Year  Fixed Effects1e:

Finally, to explore the effects of reputable auditors and underwriters on bond sizes, weuse a similar regression specification as the regression of bond maturity:

Bond Size 5 a1b Reputable Auditor1r Reputable Underwriter

1uInteraction Term1p Firm Controls1y Bond Controls1sIndustry and Year  Fixed Effects1e:

ð3Þ

To account for the possibility that there are shifts in debt financing over time caused bychanges in general capital market conditions, we include year-fixed effects in all bondterms regressions Furthermore, we also add fixed effects at the industry level, given thatthe cost of debt and other bond terms vary with the industry membership (e.g., Jorion, Shi,

& Zhang, 2009) We account for the correlation of error terms across observations thatbelong to the same issuing firm by calculating robust standard errors that allow for cluster-ing at the firm level

Data and Descriptive Statistics

We use two data sources for our main analysis: the Compustat database and the MergentFixed Income Securities Database (FISD) The former provides us information to measurethe reputation of auditors and other firm-level variables, whereas the latter enables us toidentify our dependent variables, the reputation of underwriters, and other bond-specificcharacteristics that we use as controls in our empirical tests Information on corporate bondissues received reasonable coverage in 1995; therefore, we focus our analysis from 1995 to2006

To calculate the market shares of auditors in each industry, we begin with a sample offirm-year observations that have sales information in the Compustat database Based onthese market shares, we construct the reputable auditor measure for each of the years in oursample.12 We then manually match this Compustat sample with the Mergent FISD bonddata (based on company names, industry, and location), excluding observations withoutinformation on firm-level control variables (size, leverage, etc.) We do not consider corpo-rate bonds that are convertible, privately placed, issued in foreign currencies, or do nothave fixed coupon payments These filters allow us to select a more homogeneous group ofbond securities that facilitates better cross-sectional comparisons As the majority of

auditors hired by firms issuing bonds are Big Auditors (98% in our matched sample), wefurther eliminate firms not audited by Big Auditors and focus on the variation of reputationconcerns that pertains to industry expertise.

Finally, we exclude issues without underwriting information As a result, our finalsample consists of 9,517 bond issue-level observations We include financial institutions inour analysis because prior literature (e.g., Craswell et al., 1995) argues that the demand forauditor industry specialization is increasing in the complexity of auditing tasks Financialinstitutions have complicated contracts for financial instruments and derivatives, requiringauditors with sophisticated financial knowledge to prepare the audits

We present the details of our sample selection process in Panel A of Table 1 As firmsoften issue multiple bonds over the sample period or during a year, we report both firm-level and issue-level distributions by year in Panel B of Table 1 Our final sample com-prises 1,771 firm-year observations and 9,517 bond issue-year observations Panel A ofTable 2 presents summary statistics for all variables used in our tests grouped depending

Table 1 Sample Selection and Distribution

Panel A: Sample selection

Eliminate observations without information on firm-level

control variables

(790)

Panel B: Sample distribution by year

on whether they capture firm or bond characteristics On average, 27% of bond issues inour sample are certified by reputable auditors The standard deviation of the reputable audi-tor measure Reputable Auditor is 0.45 The average auditor–client relationship is about 6years (Tenure) Also, the average leverage ratio is 43%, suggesting that the firms in oursample rely more on debt financing than the average Compustat firm (25%) Firms in oursample are relatively large, consistent with the fact that large firms frequently access thecorporate bond market.

The average value of the indicator variable Reputable Underwriter is 0.09, implying that9% of bond issues are placed by reputable underwriters Credit Spread, measured as thedifference between a bond issue’s offering yield and the yield of a matching treasury issue,

is the pricing term of a bond and captures the direct cost of borrowing The match withtreasury bills’ yields integrates the influence of business cycles on the corporate bondmarket The average credit spread of bonds in our sample is about 130 basis points, with astandard deviation of about 146 basis points The mean maturity of bonds is approximately9.01 years, consistent with the fact that bonds are usually issued with long terms BondSize is calculated as the logarithm of an issue’s offering amount, and its mean value is9.83, with a standard deviation of 2.14 We assign numeric values to the credit ratings ofMoody’s or S&P’s.13The variable Credit Rating is an increasing function of the riskiness

of a bond issue The average firm in our sample receives a rating level of 6.75, well withinthe investment grade range We also report the Spearman correlations among all the vari-ables in Panel B of Table 2

be at odds Issues with reputable underwriters are associated with larger credit spreads Asthis univariate test does not control for other correlated variables that explain creditspreads, we rely on multivariate regression specifications to make proper inferences aboutthe role of reputable underwriters on the spreads However, the effect of hiring reputableunderwriters on bond maturity and size is consistent with the arguments that they are reduc-ing information asymmetries and have access to extensive distribution channels Firms that

hire reputable underwriters obtain bonds with a significantly longer term at the 1% leveland can issue larger bonds also at the 1% level.

Multivariate Evidence

Table 4 reports the results regarding the effects of reputable auditors and underwriters oncredit spreads, maturities, and bond sizes In Models 1, 3, and 5, we estimate the effects ofreputable auditors and underwriters on bond terms in a reduced form that only includesfirm, industry, and year-fixed effects (e.g., Qian & Strahan, 2007) This mitigates the con-cern that bond features such as offering yield or maturity likely are simultaneously deter-mined and therefore might produce biased estimates when included as explanatoryvariables Models 2, 4, and 6 correspond to the full regression specifications

Models 1 and 2 show the results of the credit spread regression In Model 1, withoutcontrolling for bond characteristics, the effect of reputable auditors is negative and statisti-cally significant at the 10% level Model 2 displays the results for the full regression ofcredit spread as specified in the previous section Consistent with the certification hypoth-esis, the coefficient estimates on both reputable auditors and reputable underwriters arenegative and statistically significant at the 1% and 10% level, respectively These resultscan be interpreted as evidence that intermediaries with reputation concerns help issuersobtain lower costs of debt Furthermore, in terms of economic magnitude, reputable audi-tors on average reduce credit spreads by 35 basis points, whereas reputable underwritersreceive only a 19 basis points decrease in spreads This effect suggests that for the averagebond issued by a firm hiring reputable auditors and underwriters, the cost of debt is lower

by US$65,597 and US$35,609, respectively

Table 3 Univariate Tests

Panel A: Reputable versus ordinary auditor

Note: This table presents the results for univariate tests of reputable auditors and reputable underwriters In Panel

A, we compare the differences in bond terms between firms hiring reputable auditors and those hiring ordinary auditors In Panel B, we compare the differences in bond terms between firms hiring reputable underwriters and those hiring ordinary underwriters M 1 is the average Credit Spread/Maturity/Bond Size for firms hiring reputable auditors or underwriters, whereas M 0 is the average Credit Spread/Maturity/Bond Size for firms hiring ordinary audi- tors or underwriters Difference in M calculates the differences in bond terms between firms hiring reputable audi- tors/underwriters and those hiring ordinary auditors/underwriters t statistics are from one-tailed t tests.

*,**, *** are 10%, 5%, and 1% significance levels, respectively.

The null that the coefficient of Reputable Auditor equals to that of ReputableUnderwriter is rejected at the 10% level The economic significance of the coefficient esti-mate on reputable auditors is also stronger than that on reputable underwriters Thus, theseresults imply that reputable auditors play a relatively more important role in certifying thequality of bond issues, consistent with the additional insurance role of reputable auditorswho provide effective legal protection for bondholders against potential losses arising fromfraud audits The coefficient estimate of the interaction term of reputable auditor andunderwriter is not significant, implying that the effect of hiring a reputable auditor oncredit spreads does not vary with that of hiring reputable underwriters.

The signs of the coefficient estimates for the control variables are as expected: Higherleverage is associated with higher credit spreads, reflecting greater credit risk Issues withput options or sinking fund features receive lower credit spreads, whereas callable or subor-dinated bonds receive higher credit spreads Speculative Grade is strongly positive and sig-nificant, indicating that bondholders incorporate the risk assessments of credit-ratingagencies into their decision of risk premiums on bonds

Models 3 and 4 show the results of the effects of reputable auditors and underwriters onthe maturities of bonds In both models, the coefficient estimates on reputable auditors arepositive and significant, whereas those on reputable underwriters are positive but not signif-icant Specifically, in Model 4, the coefficient on Reputable Auditor indicates that hiringreputable auditors on average lengthens bond maturities by 2.54 years These results implythat the monitoring role of reputable auditors substitutes for the ex-post monitoring role ofthe short-term debt Thus, issuers can issue bonds with longer maturities when they hirereputable auditors

Turning to bond size, in Models 5 and 6, we find that, in contrast to the maturity sion, the coefficient estimate on reputable underwriters is significantly positive, whereasthe estimate on reputable auditors is negative but not significant The F test rejects the nullthat the coefficient estimate on reputable underwriters equals to that on reputable auditors

regres-at the 1% significance level These results indicregres-ate thregres-at the role of reputable underwriters ismore valued in determining the issuing amounts of bonds, consistent with the extensive dis-tribution networks and superior marketing and selling skills that reputable underwritershave The point estimate of 1.35 for the coefficient on the variable Reputable Underwriterimplies that hiring reputable underwriters on average increases the actual offering amount

Sensitivity Analyses and Additional Tests

Self-Selection Bias

The previous literature suggests that the choices of auditors and underwriters may be jected to a selection bias (e.g., Chaney, Jeter, & Shivakumar, 2004; Fang, 2005) To theextent that unobservable determinants of the auditors’ and underwriters’ choices correlate

sub-with bond terms, our coefficient estimates are biased and our inferences are confounded.Hence, in this section, we address the selection-bias issue of reputable auditors and under-writers and demonstrate that the inferences drawn from ordinary least squares [OLS]regressions remain robust.

Endogenous Switching Model

We first use an endogenous switching model to address the selection-bias issue of reputableauditors and underwriters (Maddala, 1983) Conceptually, we want to hold the bond issueconstant and keep separate the treatment effects on bond terms due to reputable auditorsand underwriters More specifically, we are interested in the following counterfactual out-comes: For an issue certified by a reputable auditor or underwriter, what would the alterna-tive credit spread, maturity, and bond size be had it been certified by an ordinary auditor orunderwriter? Empirically, the endogenous switching model gives us a way to determinethese hypothetical outcomes The model consists of a regression of the choice of reputableauditors or underwriters and two outcome equations on the dependent variables ofinterest—one for issues with reputable auditors/underwriters, and one for issues with ordi-nary auditors/underwriters:

Reputable Auditor or Underwriteri 5 Z0ig1ei ð4Þ

by allowing the error term in the regression of the auditor/underwriter choice to correlatewith the error terms in the outcome equations (i.e., Regressions 5 and 6, where the depen-dent variables are the bond yields, maturities, or sizes) In this way, the unobserved factorsthat affect the choice of reputable auditors or underwriters are also allowed to influencebond terms

In addition to addressing the endogeneity concern, the switching regression also relaxesthe restriction that the estimates of the b parameters are identical for issues with reputableauditors/underwriters and those with ordinary auditors/underwriters By estimating the coef-ficients of variables in the vector X separately, the model provides the estimates needed tocalculate the hypothetical outcomes of bond terms

Table 5 presents the results of the endogenous switching regressions Panel A showsthat reputable auditors are more likely to certify smaller issuers with lower leverage andlarger tangible assets We also include Reputable Underwriter as a control variable becauseBalvers et al (1988) suggest that the underwriter influences the auditor choice Reputableunderwriters may more frequently select reputable auditors as a reflection of their high rep-utation Consistent with this argument, Panel A of Table 5 shows that the presence of

reputable auditors is positively and significantly associated with the presence of reputableunderwriters.

Table 5, Panel B displays the results of the outcome equations Although most variableshave the same sign in both equations corresponding to credit spreads, maturities, and bondsizes, their magnitudes are notably different for bonds issued by firms hiring reputable andordinary auditors This supports the relaxation of the assumption that the coefficient esti-mates of explanatory variables are identical across the two alternative regimes (ReputableAuditor = 1 vs Reputable Auditor = 0) Furthermore, in the outcome regressions of creditspreads, reputable underwriters appear to exert a negative effect only in the equation ofissues with ordinary auditors (Reputable Auditor = 0) This result suggests that, after con-trolling for the selection of reputable auditors, there is an interaction effect between reputa-ble auditors and reputable underwriters; namely, reputable underwriters play thecertification role in the absence of reputable auditors

The separate estimations of bond terms across issues with reputable auditors and thosewithout show that the effects of the control variables on bond terms vary with the presence/absence of reputable auditors More importantly, such estimation techniques also enable us

to calculate the hypothetical outcomes of bond terms in alternative regimes We do so byapplying the coefficient estimates of the regime of reputable auditors to issues with ordi-nary auditors, and vice versa Panel B of Table 5 compares the means of the actual values

of bond terms with their hypothetical counterparts, derived from the switching models Forissues certified by reputable auditors (Group 1), the actual credit spread and maturity are

118 bps and 11.94 years, respectively If certified by ordinary auditors, the hypotheticalspread and maturity would be 192 bps and 8.42 years, 74 bps more and 3.52 years shorterthan the actual case, these differences being statistically significant

In contrast, the average issue with ordinary auditors (Group 2) has a credit spread of

134 bps and a maturity of 7.91 years If certified by reputable auditors, the credit spreadwould decrease by 23 bps, and the maturity would lengthen by 1.90 years Again, the dif-ferences between these actual and hypothetical values are statistically significant In thecase of issues with ordinary auditors, their average bond size would have been larger, aswell

We apply the endogenous switching model again to account for the self-selection bias ofreputable underwriters We also calculate the differences in bond terms across the regimes

of reputable and ordinary underwriters Panel C of Table 5 presents these results For issuescertified by reputable underwriters, the actual credit spread and maturity are 178 bps and12.06 years, respectively If certified by ordinary underwriters, the hypothetical spreadwould increase by 10 bps and the bond maturity would decrease by 1.78 years For issuesplaced by ordinary underwriters, the actual credit spread and bond maturity are 125 bpsand 8.72 years, respectively Consistent with the main results, we also find that the bondsize decreases by approximately 17.39% relative to the average offering amount in oursample (computed as 1.71 divided by 9.83) If placed by reputable underwriters, theirspread would be 13 bps less, their maturities would be 3.79 years longer, and their sizewould increase by 24.72% relative to the average offering amount This evidence showsthat, after controlling for the self-selection bias of reputable underwriters, the presence ofreputable underwriters can help issuers obtain favorable credit spreads, bond maturities,and offering amounts In particular, the substitution effect between hiring reputable under-writers and the monitoring role of short-term debt turns out to be significant

Overall, these sensitivity tests document that the positive impacts of reputable auditorsand underwriters on bond terms shown in our main OLS regressions are robust after con-trolling for selection-bias problems.

Changes Analysis

We use a change specification of our regressions to further address the selection-bias issuesregarding both reputable auditors and underwriters This change approach can eliminatefirm-level unobserved factors that could potentially confound our results and that allows us

to establish a stronger causal link between the types of the two intermediaries and the bondterms of interest

Panel A of Table 6 presents descriptive statistics of the changes in bond terms caused

by changes of reputable auditors and underwriters, respectively Columns (1) and (2) reportthe statistics concerning the changes in credit spreads, maturities, and bond sizes after issu-ing firms switch from ordinary auditors to reputable auditors, and vice versa Among theissuing firms that change to reputable auditors from ordinary ones, 52.5% enjoy a decrease

in credit spreads, 40% have longer bond maturities, and 50% issue larger bonds This isindicative that the new hires of reputable auditors have positive impacts on bond terms Incontrast, when issuing firms dismiss reputable auditors, 60% of them have an increase intheir credit spreads and 45% of them experience a shortening in bond maturities These sta-tistics suggest that bondholders seem to punish issuers for switching from reputable audi-tors to ordinary ones

Columns (3) and (4) present descriptive statistics for the changes in bond terms due tothe changes from ordinary underwriters to reputable ones, and vice versa For the issuingfirms that experience the changes from ordinary to reputable underwriters, 41% of themhave longer bond maturities and 52% of them increase bond sizes The changes in the bondterms concerning issuers’ switches from reputable to ordinary underwriters are not veryinformative of any punishment by bond investors

We then turn to the multivariate changes analysis to examine the effects of reputableauditors and underwriters on the bond terms In our OLS regressions, we do not excludethe issuers that have multiple bonds in a given year In the change regressions, however, tomake the changes comparable, we only include the largest bond of an issuer in a givenyear Furthermore, we also eliminate bonds in the year of changes of reputable auditors/underwriters because it is unclear whether they were issued before or after the auditor/underwriter changes Finally, we halve the full-change sample as follows: The first sampleconsists of issuing firms that switch from ordinary auditors to reputable auditors and thosethat do not change auditors (DReputable Auditor = 1 or 0) and the second sample consists

of issuing firms that change from reputable auditors to ordinary auditors and those that donot change auditors (DReputable Auditor = 21 or 0) Both samples use issuing firms that

do not change auditors as the comparison group.15 Such separation allows us to determinewhether the univariate evidence in Panel A holds in the multivariate framework Weinclude the change in reputable underwriters in both regressions

Panel B of Table 6 presents the multivariate regression results for the first sample.Interestingly, the coefficient estimate of D Reputable Auditor in the regression of creditspread is negative, but not statistically significant, implying that bondholders do not react

to switches from ordinary to reputable auditors strongly, although this might also be due tothe limited power because we do not have many reputable auditor switches in the sample.Panel C of Table 6 displays the multivariate regression results for the second sample Thecoefficient estimate of D Reputable Auditor with respect to credit spread is negative andstatistically significant at 10%, indicating that the change from reputable auditors to