Firm-specific information,analysts’superiority and investment value

Lu Li,Erjia Yang,Tusheng X iao

aSchool of Economics and Finance,Shanghai International Studies University,China

bChina Industrial International Trust Lim ited,China

cSchool of Accountancy,Central University of Finance and Economics,China

Firm-specific information,analysts’superiority and investment value

Lu Lia,*,Erjia Yangb,Tusheng X iaoc

aSchool of Economics and Finance,Shanghai International Studies University,China

bChina Industrial International Trust Lim ited,China

cSchool of Accountancy,Central University of Finance and Economics,China

A R T IC L E IN F O

Article history:

Received 23 M ay 2013

Accep ted 14 October 2014

Availab le online 20 November 2014

Firm-specif c in formation

Using a sam p le o f Chinese security analysts’recommendations from 2005 to 2010,we exam ine the source o f analysts’superiority and the investment value of their recommendations.Using a calendar-time portfo lio app roach,we f nd that,on average,analysts’recommendationsare valuableand thatanalystsare better at analyzing and transferring f rm-specif c in formation than market-w ide or industry-level in formation.In addition,we show that the investment value o f recommendations increases as f rm-specif c information becom esm ore im po rtan t in stock p ricing.Ou r em pirical resu lts are usefu l in guiding investors and helping brokerage houses to evaluate the output of research departments.

?2014 Production and hosting by Elsevier B.V.on behalf o f China Journalo f Accounting Research.Founded by Sun Yat-sen University and City University o f Hong Kong.This is an open access article under the CC BY-NC-ND license(http://creativecomm ons.o rg/licenses/by-nc-nd/3.0/).

1.Introduction

The securities analyst industry has grown rapid ly w ith the developm en t of the Chinese cap italm arket.The number of p ractitioners,their salaries and the market in fuence o f the securities-consulting industry has undergone rapid changes over the past few years.M eanwhile,p rob lems related to security analysts,such as the value o f the securities analyst industry,the information content of analysts’research reports and the investment value of analysts’recommendations,have caused great concern among academ ics and p ractitioners.

The solutions to these issuesw ill inevitab ly invo lve studying analysts’expertise.A cco rding to the ef cien t market hypothesis(EMH),Ro ll(1988)decomposes the in formation incorporated into stock p rices into three types:market-level,industry-level and f rm-specif c in formation.However,the extent to which these three types o f information explain the variations in f rms’stock returns varies.If f rms’stock returns aremain ly explained by f rm-specif c in formation,investors have a greater need for f rm-specif c in formation than for market-or industry-level in formation.In this case,security analysts who are good at analyzing and transferring f rm-specif c information w ill be favored,as their research reports are better ab le to alleviate the information asymm etry between listed companies and investors.In contrast,if f rms’stock returns are mainly explained by industry-level information,then security analysts who are good at analyzing and transferring industry-level inform ation w ill perform better.Unfortunately,p revious stud ies still p rovide no consisten t conclusion on what m akes a superior securities analyst.Som e stud ies have shown that analysts’expertise lies in analyzing and transferring f rm-specif c in form ation(e.g.,G rossm an and Stiglitz,1980; D iamond and Verrecchia,1991;Bhushan,1989;Ram nath et al.,2008;Palmon and Yezegel,2012).Other scho lars suggest that analysts play an important role during the process o f searching,analyzing and transferring industry-level in formation(e.g.,Piotroskiand Roulstone,2004;Chan and Hameed,2006).The conclusionso f these studiesare inconsistent due to diferences in their research samp lesand designs.Asa securities analystmay be good at analyzing and transferring either f rm-specif c or industry-level in formation, which of these is superior is an empirical question.This study attemp ts to answer the question o f what constitutes security analysts’superiority and their ro le in the capitalm arket.

In this paper,we use 192,012 recomm endations issued by Chinese security analysts from 2005 to 2010 and use a calendar-tim e portfolio approach to study the follow ing two questions:(1)what constitutes Chinese secu rity analysts’superiority?and(2)how do the dem and and supp ly factors of analysts’research activities in fuence the investment value of recommendations?W e calcu late three estimates o f abnormal returns for each portfo lio,nam ely market-adjusted returns,the intercep t of the Capital Asset Pricing M odel (CAPM)and the intercept of the Fama and French(1993)three-factor model.The empirical resu lts indicate,f rst,that Chinese security analysts are better at analyzing and transferring f rm-specif c information than market-or industry-level inform ation.Specif cally,ceteris paribus,analysts’research reports increase the ability of f rm-specif c in formation to explain variations in stock returns,but reduce the ability ofmarket-and industry-level information to exp lain variations in f rms’stock returns.In addition,covering more f rms in the same industry does not imp rove security analysts’ability to capture the changes in industrylevel in form ation and hence im p rove the investm ent value of their recomm endations.Second,analysts’recomm endations have greater investm en t value when f rm-specif c info rm ation p lays a m ajo r ro le in stock pricing,bu t there is no signif cant d if erence in investm ent value when industry-level in form ation p lays a major ro le in stock pricing.

This paper helps us to understand the comparative advantages o f analysts and enriches the literature on the relationship between analyst behavior and R2.Assessing the investment value o f analysts’recomm endations is actually identical to identifying and con f rm ing the source of analysts’superiority.Loh and M ian (2006)suggest that the comparative advantages of superior analysts lie in their ability to accurately p redict accounting earnings and then convert them into stock recommendations.Hence,they exam ine the investment value of recommendations based on the accuracy o f accounting earnings p redictions.Palm on and Yezegel(2012)shows that the advantages of analysts lie in analyzing and transferring f rm-specif c in form ation,and thus uses the R&D expenditure ratio(as a p roxy o f the degree o f in form ation asymm etry between listed com panies and investo rs)to m easure the investm en t value o f analysts’recomm endations.As the investm ent value is roo ted in analysts’com parative advantages,any em p irical fndings regarding when and which research reports have greater investment value w ill also help to explain analysts’comparative advantages.Our study indicates that Chinese security analysts are better at processing f rm-specif c than industry information.Unlike Piotroski and Roulstone(2004),who on ly explore the relationship between the number of analysts follow ing and R2,this study combines the supp ly and demand factors o f analysts’research activities and p rovidesm ore direct and convincing empirical evidence for how analysts’recommendations inf uence stock prices,which enriches the literature on the relationship between analyst behavior and R2.

2.Literature view and hypotheses development

Roll(1988)decomposes in formation into market-level,industry-level and f rm-specif c in formation.He points out that f rms’stock returns shou ld be exp lained by these three kinds o f in formation under the EM H.The extent to which market,industry and f rm-level in formation exp lain variations in f rms’stock returnsare calculated as fo llows.

where Ri,j,tdenotes the stock return for f rm i in industry j on day t,Rm,tdenotes the value-weighted market return on day t and Rj,tdenotes the industry return for industry j on day t.The regression statistic formodel (1),R2,measures the percentage of thevariation in f rms’stock returns that isexplained bymarket-level inform ation.The regression statistic for model(2),R2,measures the percentage o f the variation in f rms’stock returns that isexplained bymarket-and industry-level in formation.Thus,the dif erence in R2betweenmodel (2)andmodel(1)rep resents the percentage o f the variation in f rms’stock returns that isexplained by industry-level info rm ation.1-R2m easures the percen tage o f the variation in f rm s’stock returns that is exp lained by f rm-specif c in form ation.Roll(1988)show s that on average,on ly 20–30%of the variation in stock returns can be exp lained bym arket-and industry-level info rm ation.M orck et al.(2000)fnd that R2is lower in developed than in emerging econom iesand conclude that thehigh R2in emerging econom ies isassociated w ith poor p rotection of investor property rights,thus reducing investors’incentives to use f rm fundamentals.They also p ropose the concep to f synchronicity to ref ect theextent to which stock returns tend tomove together.Based on thestudy by M orck etal.(2000),Durnev etal.(2003)further explore theeconom ic consequencesof R2and f nd that a lower R2indicatesmore in formation about futureearnings in current stock returns,and viceversa. They argue that stock marketsw ith more synchronous returns exhibit lower ef ciency,which means that the degree of stock price synchronicity is no longer a neutral phenomenon.

It shou ld benoted thatM orck et al.(2000)defne two stock price synchronicitymeasures:F,def ned as the fraction of stocks in a country whose prices rise(o r fall)and weigh ted R2.F rep resen ts the p roportion o f stock p rices thatm ove in the sam e d irection w ithin a coun try,a higher F ind icates that stock prices frequentlym ove together.R2rep resents the relationship between stock returnsandmarket returns(i.e.theextent towhichmarket returns explains variations in f rms’stock returns).In contrast to F,R2neither refects the relationship between two changes(in the same or the opposite direction),nor characterizes themagnitude o f the changes. In fact,it ishard to judgewhether a high R2isa good or bad phenomenon.The use of theword synchronicity seems to imp ly thata high R2isa bad phenomenon.For exam ple,Jin and M yers(2006)suggest that R2can be used as an indicator of a f rm’s transparency.Opaque stocksw ith a high R2are also more likely to crash.

However,other studies do not support the interp retation o fM orck etal.(2000),while agreeingw ith Ro ll’s (1988)classif cation of in formation.For examp le,Piotroski and Rou lstone(2004)fnd that R2is positively associated w ith analyst forecasting activities in the U.S.,consisten t w ith analysts increasing the am ount o f industry-level info rm ation in prices through in tra-industry in form ation transfers.Therefo re,a higher R2neither indicates a less ef cientm arket,no r greater opacity.Chan et al.(2013)show that a higher R2im p roves liquidity,contradicting the view that it isusually negatively related tomarketef ciency and f rm transparency. K elly(2005)also opposes the view of Durnev et al.(2003)that R2can beused asa p roxy for in formation efciency.Teoh et al.(2007)consider that a lower R2is the resu lt o f noisy trading and Hou et al.(2013)also doubt the conclusion that a lower R2is associated w ith higher p ricing ef ciency.

From this contradictory evidence,we can draw the follow ing two conclusions.First,the factors that in fuence R2arevaried and it ishard to judgewhether a high R2isgood or bad.Second,regardlesso f the causeof a high R2and whether it isa good or bad phenomenon,Ro ll(1988)interp rets R2as the extent to whichmarketand industry-level information explains the variation in f rms’stock returns.Brockman and Yan(2009)use 1-R2as a p roxy o f the percentage o f the variation in a f rm’s stock retu rns that is directly exp lained by f rm-specif c in form ation.

Feng et al.(2009)also justify that R2can be used as a proxy fo rm easuring private info rm ation arbitrage activities.However,weargue that thismay beopen to question.Thedirectextension of Ro ll’s interpretation isthat R2m easureswhether f rm-specif c info rm ation is valuable.Specif cally,a low R2indicates that the ability ofmarket-and industry-level in formation to explain the variation in f rms’stock returns isweak,thus f rmspecif c in formation p laysamore important ro le in predicting stock returns.In contrast,a high R2illustrates thatmarket-and industry-level in formation can easily p redict f rm performance,while f rm-specif c in formation is relatively less important.Here,f rm-specif c information is not necessarily private information.For examp le,announcements o f accounting earnings,m ergers and acquisitions,and m anagement turnovers are all typeso f f rm-specif c in formation,but are not necessarily private information.Ro ll(1988)excludes stock returns near the event day to investigate the ef ect o fm arket-and industry-level information on R2.Using a clean sample that isunaf ected by f rm-specif c in formation,the results show that R2doesnot imp rove significantly,con f rm ing the existence of private in form ation.However,due to the follow ing reasons,there are still som e p roblem sw ith Ro ll’s(1988)m ethod.First,for f rm sw ith diferen t R2,f rm-specif c in form ation doesno t have the sam e im portance,thus the exten t of the ef ect of such info rm ation on stock retu rns is distinct.Ro ll’s app roach underestimates the inf uence of events for f rmsw ith low R2and overestimates it for f rmsw ith high R2.As themagnitude of R2measured by Roll’s(1988)method is low,excluding daily stock returns near the event dayw illseriously underestimate the in fuence of f rm-specif c inform ation.Second,as f rm-specif c information is end less,it is dif cult to perfectly exclude the ef ect of events from two newspapers,thus underestimating the in fuence of f rm-specif c information.Therefore,a low R2doesnotnecessarily im ply the existence of private in formation,but itmust indicate that f rm-specif c information is very important.

In summary,we suggest that R2can be used as an indicator to measure the importance o f f rm-specif c in formation in stock pricing.The higher the value of R2,the less important f rm-specif c information is.

Yang et al.(2014)exam ine whether the research repo rts o f Chinese secu rity analysts have investm en t value and f nd that,on average,analysts’recomm endations are valuab le.Specif cally,the du ration o f the investment value is quite short(usually a coup le of days)when it comes to favorable recommendations, while the duration ismuch longer(usually severalmonths)when it comes to un favorab le recommendations. Furtherm ore,they also investigate the dif erence in investment value between star analysts’and non-star analysts’research reports.The empirical results show that the investment value of favorab le recomm endations issued by star analysts is greater than non-star analysts,while the dif erence in investment value is not signif cant for un favorab le recomm endations.Unlike Yang et al.(2014),we attemp t to answer the question of what constitutes the Chinese security analysts’superiority,which helps to understand the comparative advantages o f analysts.

In fact,the expertise o f security analysts isexam ined extensively in the literature and them ajority of studies investigatewhether analystsare ab le to identify the efect of a specif c accounting variab le o r econom ic event. Piotroskiand Roulstone(2004)investigate the relationship between the num ber o f analysts fo llow ing and R2, and the results con f rm that analysts are good at analyzing and transferring industry-level information.The advantage of our research is that it exam ines the relationship between analysts’recommendations and R2to provide a better understanding of the in fuence o f analysts’behavior,and thus p rovides more direct evidence on the source o f analysts’superiority.Therefore,we aim to answer the fo llowing three questions.

First,once research reports are issued,the extent to which market-and industry-level in formation can explain the variation in f rms’stock returns w ill increase if the security analysts aremain ly analyzing and transferring industry-level in formation,thus increasing R2.Therefore,we expect that R2should decrease if daily stock returns around the report announcement date are removed.Conversely,when analysts are good at analyzing and transferring f rm-specif c in fo rm ation,we expect that R2should increase if daily stock returns around the event day are excluded.Considering that them ain role of analysts is to im p rove the extent to which f rm-specif c in form ation exp lains the variation in f rm s’stock retu rns,R2shou ld decline.Based on the above analysis,we propose the fo llow ing two competing hypotheses.

H 1a.The release of research reports increases f rms’R2when security analysts are good at analyzing and transferring industry-level inform ation.

H 1b.The release o f research reports decreases f rms’R2when security analysts are good at analyzing and transferring f rm-specif c information.

Second,both C lem en t(1999)and Jacob et al.(1999)fnd that the accuracy o f an analyst’s earnings fo recasts is negatively associated w ith the number o f f rms and industries that the analyst covers(proxy for the degree of analyst expertise).To further test the in fuence o f analysts’superiority,we exam ine the relationship between the number o f f rms in the same industry that theanalyst coversand the investment value of research reports.Analysts covering a larger number of f rms in the same industry shou ld be able to obtain more timely and accurate industry-level information,thus improving the investment value of research reportswhen security analystsaregood atanalyzing and transferring such inform ation.Conversely,when analystsare notgood at analyzing and transferring industry-level in formation,coveringmore f rms in the same industry should not bring additional know ledge or improve the investment value o f their research reports.Based on the above analysis,we propose the fo llow ing two com peting hypo theses.

H 2a.The investment value o f research reports is positively associated w ith the number o f f rms in the same industry that a securitiesanalyst coverswhen the analyst isgood at analyzing and transferring industry-level in formation.

H 2b.The investm en t value o f research reports is un related to the num ber of f rm s in the sam e industry that a securities analyst covers when the analyst is good at analyzing and transferring f rm-specif c in fo rm ation.

The supply factors that inf uence analysts’activitiesare discussed above.Next,we analyze the demand factors that derive from the information asymmetry in the capitalmarket.However,the concep t of information asymm etry isused asa general term because diferent f rmshavevaried in formation asymm etry.For example, Bradshaw et al.(2001)and Barth et al.(2001)poin t ou t that accruals and in tangible assets are im po rtan t sources of in form ation asymm etry.Palm on and Yezegel(2012)argue that the R&D expenditure ratio is also an important sourceo f information asymmetry.A ll of these typesof in formation asymm etry afect the behavior o f security analysts.Lang and Lundho lm(1996),Healy and Palepu(2001)and Byard and Shaw(2003)use diferent disclosure indices to measure the degree o f inform ation asymmetry,and exam ine the in fuence o f these indices on analyst behavior.We can see that in formation asymmetry is varied and the key question is which typesof in formation aremost important.A lthough p revious studiesexam ine several typeso f in form ation asymmetry,none considerswhich type of in formation is them ost valuable overall.In fact,p rior studies only exam ine incremental information asymm etry caused by a particular account,which isnot necessarily the m ost important demand from the perspective o f analysts’activities.

By com bining the supp ly and dem and facto rso f analysts’research activities,this study attem p ts to p rovide a m o re com p rehensive fram ework to investigate these research questions.Asm entioned,Ro ll(1988)decomposes the inform ation inco rpo rated into stock prices in to three kinds of in fo rm ation:m arket-w ide,industryleveland f rm-specif c information.Specif cally,market-w ide inform ation such asmonetary policy,fscal policy and market shocks is value relevant for all stocks.Industry-level in formation such as industry policy and industry shocks is value relevant for all stocks in a particular industry.Firm-specif c in formation such as announcements o f accounting earnings,dividends and mergers and acquisitions is value relevance for a particu lar stock.

Theoretically,all of the company’s stock returns can be exp lained by these three typeso f in formation,but theextent towhichmarket-,industry-and f rm-level in formation explain thevariation in a f rm’sstock returns difers.Dechow et al.(2010)suggest that a f rm’s fundamental earnings p rocess is jointly determ ined by its operating cycle,m acro environm en t,investm en t oppo rtunities,m anagem ent and other f rm characteristics. These f rm characteristics not on ly af ect the pro f tability of the com pany directly,but also determ ine the im portance o f diferen t types of inform ation for stock pricing.Fo r som e com panies,industry-level in form ation ismore important,whereas for others,f rm-specif c information m ay bemore im portant.This study attempts to identifywhich type of information asymmetry is themost im portant,thus resulting in the demand for analysts’activities.If companies’stock returnsaremain ly exp lained by industry-level(f rm-specif c)inform ation,investorswill have a great need for analystswho are good at searching and analyzing industry-level (f rm-specif c)in formation.

Our research com bines thesupp ly and demand factorso f analysts’research activities to exam ine the investm ent value of analysts’recommendations.Table 1 summarizes the framework o f supply and demand factors.

Table 1 illustrates that if f rm s’stock retu rnsarem ain ly exp lained by industry-level info rm ation,the supp ly and demand of security analysts’activitiesw illmatch perfectly if the analystsaregood atanalyzing and transferring such information.In this case,their research reports have higher investment value because they are better able to alleviate the information asymmetry.In contrast,the supply and demand o f analysts’activities will bem ismatched if the analysts are good at analyzing and transferring f rm-specif c information.In this case,their research reportshave lower investment valuebecause they have lim ited ability to alleviate the information asymmetry.If f rms’stock returns aremainly explained by f rm-specif c in formation,the supp ly and demand of analysts’activitiesw illmatch perfectly if the analystsare good at analyzing and transferring such in formation.In this case,their research reportshave greater investment value because they are better able to alleviate the inform ation asymm etry.Conversely,the supp ly and dem and of analysts’activities w ill bem ism atched if the analysts are good at analyzing and transferring industry-level in form ation.In this case,their research repo rts have lower investm en t value because they have lim ited ability to alleviate the info rm ation asymmetry.Based on the above analysis,we propose the fo llow ing pair o f competing hypotheses.

H 3a.Analysts’research reportshave greater investment valuewhen industry-level information p laysamore im portan t role in stock pricing.

H 3b.Analysts’research reports have greater investment value when f rm-specif c information plays a more important role in stock pricing.

3.Research design

Fo llow ing p reviousstudies(Barber etal.,2001;Loh and M ian,2006),we construct calendar-timeportfolios to calculate the abnormal returnso f analysts’recommendations.Thismethodology was initially used by Jaf e (1974)and M andelker(1974),and was strongly supported by Fama(1998).Compared w ith buy-and-ho ld portfo lios,ourmethodo logy hasseveraladvantages.First,bad-modelproblemsaremoreacutew ith long-term buy-and-hold abnormal returns,which compound an expected-return model’s p rob lems in exp laining shortterm returns(Fama,1998).Second,it is dif cu lt to control for intra-portfo lio correlationsand easy to obtain signif cant resu lts when we estimate the long-term buy-and-ho ld abnormal returns.Conversely,using calendar-time portfolios to calcu late long-term abnormal returnsautomatically cancelsout the intra-portfo lio correlations.Last bu t not least,the calendar-tim e po rtfolio app roach ism ore investo r-oriented and m o re feasible as an investm en t strategy.

In this paper,we construct two k inds of portfolios,one based on analysts’consensus recomm endationsand ano ther based on revised or initial recomm endations.Aswe need a long period to calculate analysts’consensus recommendations,the former portfo lio is used to exam ine the long-term investment value of research reports,while the latter portfolio ismore suitab le to exam ine the short-term investment value because the exact recommendation dates are availab le.

3.1.Test of Hypothesis 1

To investigate the inf uenceo f research reports issued by security analystson a f rm’s R2(i.e.Hypothesis1), we use the fo llow ing procedure.

First,using daily stock returns from day T-2X to T-1(where X=30,90,180),we regressm odel(3)and model(4)by f rm to estimate the R2statistic,respectively.

where

Table 1 Analysis of supp ly and dem and factors in the investm ent value o f recomm endations.

where Ri,j,tdenotes the stock return for f rm i in industry j on day t.Rj,tdenotes the industry return for industry j on day t w ith Ri,j,tom itted.The industry classif cation criteria are based on the“Industry Classif cation Guidance for Listed Companies”published by the China Securities Regulatory Comm ission(CSRC)in 2001. W e adopt a three-digit code category for themanufacturing industry(C)and a two-digit code category for other industries.W e also restrict industriesw ith no less than 10 listed companies when calculating industry returns.Rm,tdenotes the value-weighted market return on day t w ith Rj,tom itted.Wk,j,t,Wi,j,tand Wj,trepresent theweight ofmarket capitalization on day t.

The R2regression statistic fo rm odel(3)m easures the percentage of the variation in f rm s’stock retu rns that is exp lained bym arket-level in form ation.The R2regression statistic form odel(4)m easures the percen tage o f the variation in f rm s’stock retu rns that isexp lained bym arket-and industry-level in form ation.Therefo re,the diference between the R2values for models(3)and(4)represents the percentage of the variation in f rms’stock returns that is explained by industry-level information.

Second,for each f rm i,we exclude daily stock returnson the day before,the day of and the day fo llow ing the recommendation date,and re-regressmodel(4)to estimate R2new.

Finally,we test thediference between R2and R2new.An R2value that ishigher(lower)than the R2newvalue indicates that security analysts are good at analyzing and transferring industry-level(f rm-specif c)in formation,thus increasing(decreasing)R2.

3.2.Test of Hypothesis 2

To exam ine whether the num ber o f f rm s that analysts cover in the sam e industry af ects the investm en t value o f research repo rts(i.e.H ypothesis 2),we adop t the fo llow ing procedu re.

We begin w ith a simple calculation o f the number of f rms covered by each analyst and for each industry. For each analyst,the number o f f rms in thesame industry covered iscalculated on a 180-day w indow before day T(i.e.from day T-180 to T-1).The industry classif cation is defned as described in Section 3.1.

Next,we divide the sample into low and high groups according to themedian num ber of f rms that the analysts cover in the same industry.

Finally,we test the dif erence in the investment value o f the two groups for each of our constructed portfo lios.W e calculate three estimates o f abnormal returns for each portfolio,namelymarket-ad justed returns, the intercept of the CAPM and the intercep to f the Fama-French three-factorm odel.A llportfo lio returnsare m onth ly returns.

For the revised or initial recommendations portfo lio,the portfo lio on date T is constructed as fo llows.W e purchase stocks depending on the revised or initial recommendationsduring the T-d to T-1 period(where d=1,5,7).1A ll recommendations of strong buy,buy,hold,sell and strong sell are defned as integer numbers between 1 and 5,respectively.In o ther wo rds,a rating of 1 ref ects a strong buy recomm endation,2a buy,3a hold,4a sell and 5a strong sell.Specif cally,we purchase stocks w ith initial recommendations no higher than 2,or upgrade ratingsw ith new recomm endations no higher than 2.2G iven that downgrade recomm endations are rare in our sam p le,we do no t construct sho rt portfolios.

Fo r the consensus recomm endations po rtfolio,the portfolio on date T is constructed as fo llow s.W e begin by calcu lating the consensus recommendationsof each covered f rm during the T-X to T-1 period(where X=30,90,180)according tomodel(5).Then,we purchase stocks in the portfo lio w ith consensus recommendations no higher than 2 and sell short stocks in the portfo lio w ith consensus recommendations higher than 2.5.Stocksw ith consensus recommendationsbetween 2 and 2.5 are excluded from the transactions to reduce the efect of analyst optim ism(Barber et al.,2001;Loh and M ian,2006).

where Ni,T-1,T-Xis thenumber of recommendations for f rm i during the T-X to T-1 period,Reci,j,T-1,T-Xis the standardized analyst recommendation of analyst j for f rm i.A ll recommendations of strong buy,buy, hold,sell and strong sell are def ned as integer numbers between 1 and 5,respectively.

3.3.Test of Hypothesis 3

To exam ine how the supp ly and demand factors af ect the investment value of research reports(i.e. Hypothesis 3),we sort the samp le by the extent to which f rm-specif c and industry-level in form ation exp lain the variation in f rms’stock returns,respectively.

(1)Sort by the extent to which f rm-specif c in formation exp lains the variation in stock returns.

First,we regressmodel(4)by f rm to calcu late the extent to which f rm-specif c information explains the variation in stock returns from day T-60 to T-1(i.e.1-R2),and classify all covered f rms into one o f fve groups.

Second,for each group,using the investment strategy in Section 3.2,we construct two kinds of portfolios based on consensus recommendationsand revised recommendations,respectively.A fter determ ining the composition of each portfolio on date T-1,the value-weighted portfolio returns are calculated.

Finally,we calculate three estimates o f abnormal returns for each portfolio,namely market-adjusted returns,the intercept o f CAPM and the intercept of the Fama-French three-factormodel.

(2)Sort by the extent to which industry-level info rm ation exp lains the variation in stock retu rns.

First,using daily returns from day T-60 to T-1,we regress models(3)and(4)for each f rm to calculate R2,respectively.Thedif erence in R2betweenmodels(3)and(4)measures the extent to which industry-level in formation explains thevariation in f rms’stock returns.We classify all covered f rms into oneof fve groups.

Second,for each group,using the investment strategy in Section 3.2 we construct two portfoliosbased on consensus recommendationsand revised recommendations,respectively.A fter determ ining the composition o f each portfo lio on date T-1,the value-weighted portfolio returns are calcu lated.

Finally,we calculate three estimates o f abnormal returns for each portfolio,namely market-adjusted returns,the intercept o f CAPM and the intercept of the Fama-French three-factormodel.

4.Empirical results

4.1.Sample selection and data sources

We obtain analysts’recomm endation data from the W IND database during the 2005–2010 period.The W IND database coversm ost o f the analysts’recomm endations,includ ing details such as the recomm endation date,the type of recomm endation(if the recomm endation is not an initially of ered one,the reco rd also includes the last recomm endation),the analysts’nam es and their af liated brokerage houses.One p rob lem is that for a certain period,theW IND database on ly allowsquerying the latest recommendation for each f rmand each analyst,thus it isdif cult to exportall recommendations including thehistory atone time.Therefore, for each brokerage-f rm-analyst keyword,we query the recomm endation recordsby week.Finally,we obtain 192,012 recommendations as the initial sample.

Table 2 D istribution o f analysts’recomm endations.

Both the fnancial and stock return data are obtained from the CSM AR database.To reduce the ef ect o f potential outliers,we d rop all observations w ith an absolute value of daily returns higher than 11%.The risk-free rate(measured by themonthly yield rate on treasury bills)and Fama-French three-factor data are collected from the RESSET database.

Table2 reports thedistribution of analysts’recommendations.W e f nd that two typesof recommendations, strong buy(1)and buy(2),account for almost three quartersof the total number,while nomore than 2%o f recommendations are lower than sell(4),consistentwith Loh and M ian(2006).The resu lts indicate that on average,security analysts are less w illing to issue unfavorab le than favorable recommendations,and tend to be optim istic.Follow ing Loh and M ian(2006),we purchase stocks w ith consensus recommendations no higher than 2 and sell short stocks w ith consensus recomm endations higher than 2.5 to control for analyst optim ism.

Next,we divide the sample into revised and initially o fered recommendations.Table 3 illustrates that for the initial recomm endations sam p le,strong buy(1)and buy(2)recomm endationsaccount fo rm ore than 70%, while sell(4)and strong sell(5)recomm endations accoun t for only about 2%.For revised recomm endations, m ost of the downgrade ratings are changed to buy(2)or ho ld(3)recomm endations,consisten tw ith the f nding that analysts rarely issue unfavorable recommendations even when they downgrade a f rm.M ost of the revised recommendations are upgrades or reiterations,which further supports the view that analysts tend to be optim istic.

4.2.Empirical results of Hypothesis 1:analysts’superiority

Hypothesis 1 exam ines whether the research reports issued by analysts increase the percentage o f the variation in f rm s’stock returns that is exp lained by f rm-specif c(or industry-level)info rm ation.

Table 4 reports the results using a 60-day w indow ending on date T-1 to estim ate f rm s’R2(i.e.X=30). W e f nd that them ean(m edian)percen tage of the variation in f rm s’stock returns that is exp lained by industry-level in form ation is 36.30%(35.33%)and the percentage of the variation in f rm s’stock retu rns that isexp lained bym arket-and industry-level in form ation is52.45%(52.88%).Them ean(m edian)percentage of the in fuence o f analysts’research reports on f rm s’R2is-3.80%(-1.41%),which ind icates that the extent to which f rm-specif c in form ation exp lains the variation in stock retu rns increases by 3.80%(1.41%).A lthough some of the research reports seem to increase the extent to which market-and industry-level information explain the variation in stock returns,themain ro le o f analysts is to imp rove the extent to which f rm-specif c in formation exp lains the variation in f rms’stock returns,and thus their superiority is in analyzing and transferring f rm-specif c information.

Table 3 Descriptive statistics of analysts’revised and initially ofered recomm endations.

Asa robustness test,we also use 180-day and 360-day w indowsending on date T-1 to estimate f rm s’R2(i.e.X=90,180).The resu lts are consistent.3Fo r sim p licity,we do not tabu late the results of the robustness tests,bu t they are available upon request.

From the above evidence,we can conclude that themain role of analysts is to imp rove the extent to which f rm-specif c in formation explains the variation in f rms’stock returns,which supports H 1b.Therefore,Chinese secu rity analystsare good at analyzing and transferring f rm-specif c inform ation.If the above conclusion is correct,we further expect that covering m ore f rm s in the sam e industry w ill no t im prove the investm en t value o f analysts’research reports(Hypothesis 2).

4.3.Empirical results of Hypothesis 2:the inf uence of the number of f rms covered

H ypothesis 2 exam ines whether the research reports issued by analysts who coverm o re f rm s in the sam e industry have greater investm ent value.

Table 5 p resents the resu lts based on the portfo lio of analysts’revised recomm endations.Specif cally,Panel A shows the investment value of favorable recomm endations issued by analystswho cover a low number o f f rms.Using the recomm endations issued on date T-1 to construct theportfo lio(thedaily portfolio contains 4.09 stocks on average),we f nd that the portfo lio raw and market-adjusted returns are 11.43%and 8.30%, respectively,while the intercept of the CAPM and Fama-French three-factor model is 7.76%and 7.78%, respectively.A llportfo lio returnsaresignif cantat the1 percent level.W eexpect that less frequent rebalancing will cause portfolio returns to dim inish inmagnitude.W ith a 5-day rebalancing period,for examp le,the portfo lio returns decline from 7.78%to 2.75%under the Fama-French three-factormodel(co lumn 4)and remain signif cant.Whenwe further expand the rebalancing period to 7 days,the portfo lio returnsdecline from 7.78% w ith 1-day rebalancing to 1.59%w ith 7-day rebalancing under the Fam a-French th ree-facto rm odel,bu t still w ith m arginal signif cance.These em pirical results suggest that the favo rab le recomm endations issued by analystswho cover a low num ber o f f rms are valuable.

Panel B o f Table 5 illustrates the investment value of favorable recommendations issued by analystswho cover a high number of f rm s.Sim ilarly,using the recomm endations issued on date T-1 to construct theportfo lio(the daily portfo lio contains 8.15 stocks on average),we f nd that the portfolio raw and marketadjusted returns are 9.03%and 5.90%,respectively,while the intercep t of the CAPM and Fama-French three-factorm odel is5.66%and 5.73%,respectively.A llportfolio returnsaresignif cantat the 1 percent level. The portfolio returns dim inish in magnitude as the rebalancing period is lengthened to 5 days,declining from 5.73%w ith 1-day rebalancing to 2.63%w ith 5-day rebalancing under the Fam a-French three-factor model (co lum n 4),which is signif can t at the 1 percen t level.Further expand ing the rebalancing period to 7 days, the portfolio retu rns decline to 2.08%under the Fam a-French th ree-facto r m odel,but rem ains signif cant.

These fnd ings suggest that the favo rab le recomm endations issued by analysts who cover a high num ber o f f rms also have signif cant investment value.

Table 4 The infuence of analysts’research repo rts on f rms’stock returns.

Table 5 The infuence of the number of f rm s covered on the investment value o f analysts’revised recommendations.

PanelC o f Table5 compares the diference in investment value for favorable recommendationsbetween the two types of analysts.A zero-investment portfo lio based on the recommendations issued on T-1 indicates that investors can earn positive abnormal returns.The portfo lio market-adjusted return is 2.40%(w ith a t-statistic of 1.36),whereas the intercep tso f the CAPM and Fama-French three-factormodelare 2.10%(w ith a t-statistic of 1.16)and 2.05%(w ith a t-statistic of 1.07),respectively.However,all o f the hedge returnsare insignif cantly diferent from zero.The hedge returns decrease signif cantly as the rebalancing period is lengthened to 5 days,declining from 2.05%to 0.12%(with a t-statistic o f 0.08)under the Fama-French th ree-facto r m odel(co lum n 4),and further decrease as the rebalancing period is lengthened to 7 days, declining to-0.49%(w ith a t-statistic o f-0.38)under the Fam a-French th ree-facto r m odel.Overall,the fnd ings show that the favorable recomm endations issued by analysts who cover a high num ber of f rm s do not have a greater investment value than those issued by analysts who cover a low number o f f rms. This f nding also suggests that covering more f rms does not mean that analysts havemore industry-level in formation.From the above evidence,we can conclude that Chinese security analystsare better at searching for and analyzing f rm-specif c information rather than industry-level information.

Table 6 presents the estimated coef cients for the Fama-French three-factor model.W e fnd that the coef cientson RM RF,SMB and HM L are not signif cantly dif erent between the portfolios of the two types of analysts,indicating that f rm characteristics such asmarket risk,grow th and book-to-market ratios are qualitatively the same for each portfolio.

Table 7 p resents the results based on the portfo lio of analysts’consensus recomm endations.Specif cally, Panel A show s the investm en t value of favo rab le recomm endations issued by the two types of analysts. For analysts covering a low number o f f rms,the portfo lio raw return o f 2.86%is signif cant at the 10 percent level,whereas the portfo lio abnormal returnsestimated bymarket-ad justed returns and the intercep tso f theCAPM and Fama-French three-factormodel are neither statistically nor econom ically signif cant.The f ndings suggest that investo rs who purchase stocks based on analysts’consensus recomm endations du ring the T-30 to T-1 period(i.e.X=30)do not earn positive abnorm al retu rns.Sim ilarly,fo r analysts covering a high num ber of f rm s,the po rtfolio abno rm al retu rns are neither statistically o r econom ically signif cant. The investment values o f favorable recommendations also show no signif cant dif erence between the two types o f analysts.The corresponding p-values of the abnormal returns estimated by market-adjusted returns and the intercep ts of the CAPM and Fama-French three-factor model are 65.9%,70.1%and 61.5%, respectively.

Table 6 Fama-French three-factor regressions based on the portfo lio of analysts’revised recommendations.

Table 7 The infuence of the number of f rm s covered on the investm ent value o f analysts’consensus recommendations.

Panel B illustrates the investment value of un favorab le recomm endations issued by the two types o f analysts.For analysts covering a low number of f rms,excep t for the portfolio raw return of 1.14%,which isnot signif cant,the portfo lio abnormal returns estimated by market-adjusted returns and the intercepts o f the CAPM and Fama-French three-factormodel are-1.99%,-2.23%and-1.96%,respectively,and all of them are signif cant at the 1 percen t level.The resu lts for analysts covering a high num ber of f rm s are qualitatively the sam e,w ith po rtfo lio abnorm al retu rns o f-1.75%,-1.93%and-1.95%,respectively.The investm en t values of unfavo rable recomm endations also show no signif cant d if erences between the two types of analysts. The co rrespond ing p-values o f the abno rm al returns estim ated by m arket-ad justed retu rns and the intercep ts o f the CAPM and Fama-French three-factormodel are 79.2%,73.9%and 99.2%,respectively.The fndings suggest that both types of analysts’un favorab le recommendations have signif cant investment value.

Table 8 Fam a-French three-factor model regressions based on the portfo lios of analysts’consensus recomm endations.

PanelC p resents thehedge returns for each portfolio.The resu lts show that thehedge returns for portfolios fo rm ed on the basis of analysts’consensus recomm endationsare no t on ly signif can t at the 5 percent level,bu t also do no t depend on the num ber of f rm s covered by analysts.

Table 8 reports the estimated coef cients for the Fama-French three-factormodel.The signif cant coefcients on SMB and HM L indicate that favorab le recommendations are associated w ith f rms of large size and lower book-to-market ratios,while unfavorable recommendations are associated w ith f rmso f small size and higher book-to-market ratios.

Asa robustness test,we also exam ine the abnormal returns for portfo lios formed on the basiso f analysts’consensus recomm endations during the T-90 to T-1 period(i.e.X=90)and the T-180 to T-1 period (i.e.X=180),respectively.The resu lts are qualitatively the same.

From the above evidence,we can conclude that the investment value o f neither favorable nor unfavorable recomm endations show s a signif cant d if erence between the two types of analysts.In other words,covering m ore f rm s in the sam e industry does no t help analysts to incorporate industry-level info rm ation in to their recomm endations,supporting H 2b.The resu ltsalso further conf rm the fndingsof H ypo thesis1,that Chinese analysts are good at analyzing and transferring f rm-specif c rather than industry-level information.

4.4.Empirical results of Hypothesis 3:the inf uence of supply and demand factors

Given that theaboveevidenceshows that Chineseanalystsare good atanalyzing and transferring f rm-specif c in formation,we expect the investment value o f analysts’recommendations to increase(decrease)as f rmspecif c(industry-level)in form ation p laysamore important ro le in stock pricing.Specif cally,weexam ine the fo llow ing four cases.

Case#1:Sort by the extent towhich f rm-specif c information explains thevariation in stock returnsand construct portfolios based on recommendation changes.Table 9 reports the investment value o f favorab le recomm endations that involve daily po rtfolio rebalancing.As shown in colum ns 1–5 o f Panel A,there is a m onotonic decrease in po rtfo lio returns.Taking the in tercept of the Fam a-French three-factor m odel as an exam p le,the abnorm al returns on portfo lios 1–5 are 11.15%,7.44%,6.91%,6.41%and 3.24%,respectively, and all of them are signif cant at the 1 percent level.The hedge returns that can be generated by a strategyo f purchasing stocks in portfolio 1 and selling short stocks in portfolio 5 are 7.91%(w ith a p-value of 0.2%). The portfo lio abnormal returns estimated bym arket-adjusted returnsand the CAPM intercept showsqualitative sim ilar patterns.Panel B p resents the estimated coef cients for the Fama-French three-factormodel. Overall,Tab le 9 provides strong evidence that the investment value o f analysts’favorable recommendations increases as f rm-specif c information p lays a more important role in stock pricing.

Table 9 The investmen t value of analysts’revised recomm endations by the im portance of f rm-specif c information in stock pricing.

Case#2:Sort by the extent to which industry-level information explains the variation in stock returns and construct portfolios based on recommendation changes.Table 10 repo rts the investm ent value o f favorable recomm endations that invo lve daily po rtfolio rebalancing.From portfo lios 1 to 5,the im po rtance of industrylevel in formation in stock pricing increases.Asshown in co lumns1–5 of Panel A,there isnomonotonic trend in the portfo lio returns.Taking the intercept o f the Fama-French three-factor model as an examp le,the abnormal returns range from a low o f 5.09%on portfo lio 3,to a high o f 9.54%on portfo lio 4.The portfolio abnormal returns estimated by market-adjusted returns and the CAPM intercept shows qualitative sim ilar patterns.Panel B presents the estimated coef cients for the Fama-French three-factor model.Overall,the above fndings suggest that security analysts are not good at analyzing and transferring industry-level in formation.

Asa robustness test,we f rst rank the fu llsample into f ve groupsby theextent towhich f rm-specif c inform ation exp lains the variation in stock returns and then re-construct long portfolios based on revisedrecommendationsw ith a frequency of portfo lio rebalancing of nomore than 7 days.Fig.1 illustrates the intercept of the Fama-French three-factormodel for each portfolio and each frequency of portfo lio rebalancing. The fgure shows that(1)the analysts’favorable recommendations are valuab le,(2)investors react quick ly (w ithin three days)to changes in analysts’favorable recommendations and(3)the portfo lio returns decrease signif cantly on po rtfolios1–5.These f ndings suppo rt that the sho rt-term investm en t value of analysts’repo rts increases as f rm-specif c inform ation p lays a m ore im po rtan t role in stock pricing.

Table 10 The investmen t value o f analysts’revised recommendations by the im portance of industry-level information in stock pricing.

Sim ilarly,we rank the fu ll sam p le in to f ve groups by the exten t to which industry-level info rm ation explains the variation in stock returns and then re-construct long portfolios based on recommendation changes with a frequency o f portfolio rebalancing of no m ore than 7 days.Fig.2 illustrates the intercept o f the Fama-French three-factor model for each portfolio and each frequency o f portfolio rebalancing.The fgure shows that there isno monotonic trend in portfo lio returns,consistentw ith the resu lts in Table 10.

Figure 1.The im portance o f f rm-specif c information and the investment value of analysts’revised recommendations by rebalancing frequency.

Figure 2.The importance o f industry-level information and the investment value of analysts’revised recommendations by rebalancing frequency.

Case#3:Sort by the extent towhich f rm-specif c information explains thevariation in stock returnsand construct portfoliosbased on consensus recommendations.Table11 reports the portfo lio returns.Specif cally,Panel A shows the investment value o f favorab le recommendations for each portfo lio.From portfolios 1 to 5,the importance o f f rm-specif c information in stock pricing decreases.As shown in column 1,the raw returnson portfo lios 1–5 are signif cantly positive at the 1 percent level,but the dif erence between portfolio 1 and portfo lio 5 isnot signif can t.In contrast,regard lessofwhether abnorm al returnsare estim ated bym arket-ad justed returns,the intercep to f the CAPM or the intercept of the Fam a-French three-factormodel,most of the portfo lios(except portfolio 1)abnormal returnsare neither statistically nor econom ically signif cant.These results suggest that the duration o f the investment value isquite shortwhen it comes to favorab le recommendations. W ith a one-month delay,the portfo lio abnormal returnsare not signif cantly greater than zero.It should be noted that the slightly positive abnormal returns on portfo lio 1 show,to some extent,that the investment value o f analysts’recomm endations increases as f rm-specif c in fo rm ation p laysam o re im po rtan t ro le in stock p ricing.

Panel B illustrates the investm en t value of unfavorable recomm endations for each po rtfolio.M ost of the portfo lio abnormal returns in columns2–4 are signif cantly negative and dim inish inmagnitude as the importanceo f f rm-specif c inform ation in explaining thevariation in stock returnsdecreases.Taking the intercepto f Fama-French three-factormodelasan examp le,theabnormal returnson portfo lios1–5 are-5.42%,-1.73%, -2.29%,-2.55%and-1.56%,respectively,which are allsignif cant at the 1 percent levelexcep t for portfolio 2.In addition,the diference between portfo lio 1 and portfo lio 5 is 3.86%(w ith a p-value of 0.1%).These results suggest that analysts’un favorable recommendations are valuab le and that the duration o f the investm ent value ismuch longer than that for favorab le recommendations.A lso asexpected,the investment valueo f analysts’un favorab le recomm endations increasesas f rm-specif c in formation p lays amore important ro le in stock p ricing.

Table 11 The investmen t value o f analysts’consensus recomm endations by the im portance of f rm-specif c in formation in stock p ricing.

Panel C presents the hedge returns for each portfolio.The results indicate that,except for portfo lio 4,the portfo lio hedge returns decreasem onotonically as in Panel B.The portfolio hedge returnsestimated bymarket-ad justed returnsand the interceptso f the CAPM and Fama-French three-factormodelare 4.90%,4.78% and 4.26%,respectively,and allof them are signif cantat the 1 percent level.These results further conf rm thatthe greater the importance of f rm-specif c in formation in stock p ricing,the greater the investment value o f analysts’research reports.

Table 12 Fama-French three-factorm odel regressions based on analysts’consensus recommendations portfoliosby the impo rtance of f rm-specif c information in stock pricing.

Table 12 reports the estimated coef cients for the Fama-French three-factormodel.The signif cant coefcientson SMB indicate that un favorab le recommendationsare associated w ith larger f rm size than favorable recommendations,while the coef cientson RM RF and HM L suggest that there are no signif cant dif erences in them arket risk and book-to-m arket ratios between the two types o f recomm endations.

Case#4:Sort by the extent to which industry-level information explains the variation in stock returns and construct portfolios based on consensus recommendations.Tab le 13 repo rts the portfo lio returns.Panel A show s that the abnorm al returns for portfolios form ed on the basiso f analysts’favorab le recommendationsare neither statistically nor econom ically signif cant.Panel B shows that although analysts’unfavorable recommendations are valuab le,there is no monotonic trend in portfo lio returns across portfolios 1–5,especially portfo lio 5 in which industry-level information plays the most important ro le in stock p ricing and whichobtains the lowestabnormal returns.Overall,the f ndingssuggest that the investmentvalueo f research reports is unrelated to the importance of industry-level in formation in stock pricing.

Table 13 The investmen t value o f analysts’consensus recomm endations by the im portance of industry-level information in stock pricing.

Table 14 reports the estimated coef cients for the Fama-French three-factor model.The coef cients on RM RF,SM B and HM L indicate that both themarket risk and book-to-market ratios show no signif cantdiference between long and short portfolios,while f rm size is larger for short portfolios,consistentw ith the f ndings in Table 12.

Table 14 Fama-French three-factorm odel regressionsbased on analysts’consensus recomm endationsportfoliosby the im portance of industry-level information in stock pricing.

Asa further robustness test,we exam ine the abnormal returns for portfolios formed on the basis o f analysts’consensus recommendations during the T-90 to T-1 period(i.e.X=90)and the T-180 to T-1 period(i.e.X=180),respectively.The resu lts are consistent.

Overall,the em p irical results in Section 4.4 show that(1)analysts’recomm endations are valuable;(2)the investm en t value of recomm endations increasesas f rm-specif c info rm ation becom esm o re im po rtan t in stock p ricing,while there is no signif cant relationship between the investm en t value o f recomm endations and the importanceo f industry-level in formation;(3)theduration o f the investment value isquiteshort(usually a coup le of days)when it comes to favorab le recommendations;and(4)the duration o f the investment value is m uch longer(usually severalmonths)when it comes to un favorable recommendations,which may be due to short-sale constraints and analyst optim ism.

In summ ary,we can conclude that(1)Chinese security analysts are better at analyzing and transferring f rm-specif c information than industry-level information.On theonehand,analysts’research reports increase the ability of f rm-specif c information to explain the variation in stock returns,while on the other hand,coveringmore f rms in the same industry doesnot help analysts incorporate industry-level information into their research reportsand thus imp rove the investment value of their recommendations.(2)Asexpected,the investment value of analysts’recomm endations increases as f rm-specif c in formation becomesmore important in stock p ricing,which conf rms the analysts’superiority.

5.Conclusion and limitations

5.1.Conclusion and implications

W ith the development o f the Chinese capitalmarket,the securities analyst industry is grow ing rapidly. Whether analysts’activities help to decrease information asymmetry and thus imp rove the ef ciency o f resource allocation in the capital market has caused great concern among academ ics and practitioners. However,the fndings in the literature are controversial.Our study exp lores this debate and p rovides a new form of evidence.

Using data on 192,012 recommendations issued by Chinese security analysts from 2005 to 2010,we f nd that on thewhole,analysts are better at analyzing and transferring f rm-specif c than industry-level in formation.Specif cally,ceteris paribus,analysts’research reports increase the ability of f rm-specif c in formation to exp lain the variation in stock returns.Fu rtherm ore,covering m o re f rm s in the sam e industry does not help analysts to inco rpo rate industry-level info rm ation into their research repo rtsand thus im prove the investm en t value of their recommendations.The investment value of analysts’recommendations increasesas f rm-specif c in formation p lays amore important role in stock p ricing,which also con f rms that analystsare good at analyzing and transferring f rm-specif c information.Our empirical results suggest that security analysts p lay an important ro le in alleviating the in formation asymmetry in the capitalmarket and that their research reports can guide investors.Our fndingsalso show that the investment value o f analysts’recommendations ismain ly derived from their research activities o f analyzing and transferring f rm-specif c rather than industry-level in formation.

The resu lts of this study also generate some important imp lications.First,the securities analyst industry shou ld fu rther enhance its ability to p rocess industry-level in form ation,so that itm ay p lay a m ore im portan t role in alleviating the in form ation asymm etry arising from industry-level in form ation.Second,listed com panies shou ld further im prove their info rm ation disclosu re environm ent.Our f nd ings suggest that the investment value o f analysts’research reports increases as f rm-specif c information becomesmore important in stock p ricing,whichmeans that f rm-specif c information isnotwellunderstood by investors,thus resu lting in the demand for information from intermediaries.Once the information environment of listed companies imp roves at the institutional level,a huge amount of transaction costsw ill be saved.

5.2.Limitations

First,our study shows that security analysts’superiority lies in analyzing and transferring f rm-specif c in form ation,which igno res the fact that som e analysts are good at p rocessing industry-level in form ation. Unfortunately,this paper does no t distinguish between analysts who are good at processing industry-level and f rm-specif c info rm ation.

Second,thedescriptivestatistics in Section 4 show that security analysts tend to beop tim istic.A lthough we fo llow Loh and M ian’s(2006)m ethod to construct our portfo lios,it is stillpossible that the reliability o f our conclusions isaf ected by analyst op tim ism.4W e thank the referee for pointing th is out.Therefore,readers should be aware that som e lim itationsexist in the reliability of our conclusions.We look forward to more academ ic research based on mature data in the future.

Acknow ledgments

The authors thank the executive editor and the anonym ous referees for their helpfu l suggestions.W e acknow ledge the f nancial support o f the research pro ject(13R21421500)sponsored by the Shanghai Science and Technology Comm ission and support from the National Natural Science Foundation o f China (71272221,71402197).This study was also supported by grants from the Shanghai College of f rst-class discip line in the Shanghai International Studies University,the Beijing M unicipal Comm ission of Education“Joint Construction Project”and the Beijing M unicipalComm ission of Education“Pilot Reform of Accounting D iscipline Clustering”.We appreciate valuable comments from ProfessorsCongW ang,Yong Yang,Zhen Sun,Zengquan Liand Qinglu Jin.Any errors are ou r own.

Barber,B.,Lehavy,R.,M cN ichols,M.,Trueman,B.,2001.Can investors p rof t from the prophets?Security analyst recommendations and stock returns.J.Finan.56(2),531–563.

Barth,M.E.,K asznik,R.,M cN ichols,M.F.,2001.Analyst coverage and intangible assets.J.A ccount.Res.39(1),1–34.

Bhushan,R.,1989.Firm characteristics and analyst follow ing.J.Account.Econ.11(2–3),255–274.

Bradshaw,M.T.,Richardson,S.A.,Sloan,R.G.,2001.Do analystsand auditorsuse information in accruals?J.Account.Res.39(1),45–74.

Brockman,P.,Yan,X.M.,2009.Block ownership and f rm specif c in formation.J.Bank.Finan.33,308–316.

Byard,D.,Shaw,K.W.,2003.Co rporate disclosu re quality and properties o f analysts’in formation environm ent.J.A ccount.,Audit., Finan.18(3),355–378.

Chan,K.,Hameed,A.,2006.Stock price synchronicity and analyst coverage in emergingmarkets.J.Finan.Econ.80,115–147.

Chan,K.,H am eed,A.,Kang,W.,2013.Stock price synchronicity and liqu idity.J.Finan.M arkets 16(3),414–438.

C lem ent,M.B.,1999.Analyst forecast accuracy:do ability,resources,and portfolio com p lexitym atter?J.A ccount.Econ.27,285–303.

Dechow,P.,Ge,W.,Schrand,C.,2010.Understanding earnings quality:a review o f the proxies,their determ inants and their consequences.J.Account.Econ.50(2–3),344–401.

D iam ond,D.W.,Verrecchia,R.E.,1991.D isclosu re,liquidity,and the cost of cap ital.J.Finan.46(4),1325–1359.

Durnev,A.,M orck,R.,Yeung,B.,Zarow in,P.,2003.D oes greater f rm-specif c return variation m ean m ore o r less in formed stock p ricing?J.Account.Res.41(5),797–836.

Fama,E.F.,1998.M arket ef ciency,long-term returns,and behavioral fnance.J.Finan.Econ.49,283–306.

Fam a,E.F.,French,K.R.,1993.Comm on risk facto rs in the retu rns on stocks and bonds.J.Finan.Econ.33,3–56.

Feng,Y.F.,D ong,Y.,Yuan,Z.B.,Yang,R.M.,2009.Private information arbitrage in Chinese stock m arket:a study based on R2.Econ. Res.J.8,50–59(in Chinese).

Grossman,S.J.,Stiglitz,J.E.,1980.On the im possibility of informationally ef cientmarkets.Am.Econ.Rev.70(3),393–408.

Healy,P.M.,Palepu,K.G.,2001.In formation asymm etry,co rporate disclosu re,and the cap ital m arkets:a review of the emp irical d isclosure literature.J.A ccount.Econ.31(1–3),405–440.

Hou,K.,Peng,L.,Xiong,W.,2013.Is R2a M easure of M arket Inef ciency?Working Paper.

Jacob,J.,Lys,T.,Neale,M.,1999.Expertise in forecasting performance of security analysts.J.Account.Econ.28,51–82.

Jafe,J.F.,1974.Special inform ation and insider trading.J.Business 47(3),410–428.

Jin,L.,M yers,S.C.,2006.R2around the world:new theory and new tests.J.Finan.Econ.79,257–292.

Kelly,P.J.,2005.Information Ef ciency and Firm-Specif c Return Variation.Working Paper.

Lang,M.H.,Lundholm,R.J.,1996.Corporate disclosure policy and analyst behavior.Account.Rev.71(4),467–492.

Loh,R.K.,M ian,G.M.,2006.Do accurate earnings forecasts facilitate superior investment recommendations?J.Finan.Econ.80,455–483.

M andelker,G.,1974.Risk and return:the case ofmerging f rm s.J.Finan.Econ.1,303–335.

M orck,R.,Yeung,B.,Yu,W.,2000.The information contentof stock markets:why do emergingmarketshave synchronous stock p rice movements?J.Finan.Econ.58,215–260.

Palm on,D.,Yezegel,A.,2012.R&D In tensity and the value of analysts’recommendations.Contem porary A ccoun t.Res.29(2),621–654.

Piotroski,J.D.,Roulstone,D.T.,2004.The inf uence of analysts,institutional investors,and insiders on the incorporation of m arket, industry,and f rm-specif c information into stock prices.Account.Rev.79(4),1119–1151.

Ramnath,S.,Rock,S.,Shane,P.,2008.The fnancial analyst forecasting literature:a taxonom y w ith suggestions for future research.Int. J.Forecasting 24,34–75.

Ro ll,R.,1988.R2.J.Finan.43(3),541–566.

Teoh,S.H.,Yang,Y.,Zhang,Y.,2007.R-Square:Noise or Firm-Specif c Information?Working Paper.

Yang,E.J.,Li,L.,Chen,Y.S.,2014.Analysts’recommendation revisionsand investment value.China Account.Finan.Rev.16(1),147–183.

*Corresponding author.

E-mail addresses:lilucf@163.com(L.Li),tsh.xiao@aliyun.com(T.X iao).

http://dx.doi.org/10.1016/j.cjar.2014.10.002

1755-3091/?2014 Production and hosting by Elsevier B.V.on behalf of China Journalof Accounting Research.Founded by Sun Yat-sen U niversity and City University of H ong K ong.

This is an open access article under the CC BY-NC-ND license(http://creativecommons.org/licenses/by-nc-nd/3.0/).

Analysts’superiority

Investment value