The Impac of Relaed Pary Sales by Lised Chinese Firms on Earnings Informaiveness and Earnings Forecass* JIWEI WANG School of Accounancy Singapore Managemen Universiy HONGQI YUAN School of Managemen Fudan Universiy Forhcoming Inernaional Journal of Business * We hank Invesoday Ld. (www.invesoday.com.cn) for providing Chinese earnings per share forecas daa for academic research. We very much appreciae he insighful commens and suggesions of Joseph Aharony, Tom Lechner, Barry Marks, Nancy Su, and seminar paricipans a AAA 2007 Chicago Annual Meeing, Ciy Universiy of Hong Kong, EAA 2007 Lisbon Annual Congress, and Hong Kong Polyechnic Universiy. Wang acknowledges he financial suppor (gran number C206/MSS5A001) a Singapore Managemen Universiy. Yuan acknowledges he suppor of he Naional Naural Science Foundaion of China (gran number 70872066). Conac auhor: Jiwei Wang, School of Accounancy, Singapore Managemen Universiy, 60 Samford Road, Singapore 178900. Tel: +65-68280616, Fax: +65-68280600, E-mail address: jwwang@smu.edu.sg.
The Impac of Relaed Pary Sales by Lised Chinese Firms on Earnings Informaiveness and Earnings Forecass ABSTRACT Using a random sample of 140 of China s lised firms, we show an adverse impac of relaed pary (RP) sales of goods and services on he usefulness of accouning earnings o invesors and on he qualiy of earnings forecass by financial analyss. Consisen wih he conenion ha RP sales may violae he arm s-lengh assumpion of regular ransacions and consequenly impair he represenaional faihfulness and verifiabiliy of accouning daa, we find ha earnings of firms engaged in RP sales are a leas 33% less informaive afer conrolling for facors known o affec earnings informaiveness. We also find ha financial analyss are overly credulous in heir accepance of earnings numbers ha are conaminaed by unreliable RP sales, and provide less accurae and more opimisic earnings forecass for firms wih more RP sales. Overall, our resuls provide srong empirical evidence on he negaive impac of RP ransacions on he usefulness of accouning earnings daa used by invesors and by financial analyss. JEL classificaion: M41; G14; G24 Keywords: Relaed Pary Sales; Earning Informaiveness; Earnings Qualiy; Earnings Forecass; China 1
The Impac of Relaed Pary Sales by Lised Chinese Firms on Earnings Informaiveness and Earnings Forecass 1. INTRODUCTION This paper invesigaes he impac of relaed pary (RP) ransacions on earnings informaiveness as perceived by invesors and he qualiy of earnings forecass made by financial analyss. RP ransacions may violae he arm s-lengh assumpion of regular ransacions and consequenly impair he represenaional faihfulness and verifiabiliy of accouning daa. Recen corporae scandals such as Enron and Adelphia, o name jus wo, exemplify how RP ransacions may reduce he reliabiliy of repored earnings. The Financial Accouning Sandards Board (FASB) has long been expressing concerns abou he possible negaive impac of RP ransacions on he reliabiliy of financial informaion, in erms of boh represenaional faihfulness and reliabiliy of repored amouns (SFAS 57, 15), while previous research provides empirical evidence of a significan posiive associaion beween RP ransacions and earnings manipulaion (e.g., Gordon and Henry 2003, in he US, and Aharony, Wang, and Yuan 2010, in China). The evidence suggess he possibiliy of an adverse impac of RP ransacions on he qualiy of accouning numbers. There are various ypes of RP ransacions. On a se of RP ransacions for 366 American public firms ha hey hand-colleced during he years 2000-2001, Gordon and Henry (2003) idenify several major ypes of RP ransacions, such as direc ransacions wih employees or wih board members, conrac services or legal services acquired from managemen, sales o (purchases from) relaed paries and loans o (from) relaed paries. Their saisics show ha 47% of heir sample firms provided loans o RPs, 20.2% engaged in direc service relaions involving an execuive and anoher relaed pary, while 19.2% made purchases from RPs, and 1
11.3% made sales o relaed paries. In conras, Aharony e al. (2010) show ha around 60% of Chinese iniial public offerings during he period 1999-2001 engaged in sales of goods and services o heir conrolling paren companies, in a mean dollar amoun of US$ 23.57 million, compared wih a mean of US$3.66 million for Gordon and Henry s (2003) American sample. Our sudy focuses on he impac of RP sales of goods and services on earnings informaiveness as perceived by capial marke invesors, and on he qualiy of analyss earnings forecass, he reason being ha RP sales have he mos direc impac on earnings number. We choose a Chinese seing for our analysis because China provides boh a unique insiuional seing and RPT daa. Our resuls provide an assessmen of he exen o which RP sales of goods and services conribue o impairing earnings informaiveness and he qualiy of analyss forecass. We randomly selec 140 firms ou of 383 firms ha were lised on he Shanghai Sock Exchange in 1997. We hen hand-collec all RP sales of goods and services daa for he 140 companies, from heir annual repors during he period of 1998 o 2004. Saisical analysis of he daa colleced for our sample firms indicae ha hey are no significanly differen in heir characerisics from he enire populaion of 383 firms. Boh our wo-way fixed effec panel daa analysis and our pooled ordinary leas square regressions sugges ha RP sales significanly reduce earnings informaiveness in China. The resuls are robus afer conrolling for facors ha may affec earnings informaiveness such as firm size, marke-o-book raio, and leverage. In addiion, we find ha Chinese financial analyss do no accoun for he adverse impac of RP sales on earnings. In paricular, heir forecass are considerably less accurae for firms wih more RP sales of goods and services, and he forecass hey provide become ever more excessively opimisic he more firms engage in RP sales. These resuls sugges ha financial analyss are 2
unduly willing o rus earnings numbers ha are conaminaed by unreliable RP sales, consisen wih he evidence in Teoh and Wong (2002) on he creduliy of financial analyss. In summary, our resuls sugges ha RP sales significanly reduce boh earnings informaiveness o invesors and he qualiy of analyss earnings forecass in China. The remainder of he paper is organized as follows. In secion II, we provide he insiuional background on China s disclosure of RP ransacions. Secion III discusses prior research and develops our hypoheses. We presen he research design, daa and empirical resuls of he impac of RP sales on earnings informaiveness and on he qualiy of analyss forecass in secions IV and V, respecively. Secion VI provides a conclusion. 2. INSTITUTIONAL BACKGROUND In 1997, he Chinese Minisry of Finance, which serves as he accouning sandards seer in China, promulgaed an accouning sandard for RP ransacions (hereafer, he RPT Sandard), which requires publicly lised companies o disclose all maerial RPTs in he form of noes o he financial saemens. The RPT Sandard defines RPTs o include ransacions occurring beween a lised firm and is paren company or wih anoher affiliaed firm. However, he relaed paries need no necessarily be oher lised companies. They may also be a lised firm and is managemen, board members, principal owners, or members of he immediae families of any of hese groups. Neverheless, as Chinese firms are largely owned by corporae shareholders, almos all RP ransacions are execued beween lised firms and heir respecive affiliaed firms such as paren companies or oher subsidiaries of hese paren companies (Aharony e al. 2010). The RPT Sandard has relaively deailed disclosure requiremens, similar o hose se forh by SFAS 57 in he USA (Gordon, Henry and Palia 2004). The disclosure mus include 3
informaion on he naure of he relaionship beween he paries involved, he core operaions of each relaed pary, a descripion of he naure of each ype of ransacion, and informaion on he amouns involved. The RPT Sandard also liss he various ypes of RPTs ha publicly lised firms mus disclose in he form of noes o he financial saemens: sales and purchases of goods and services, commission relaionships, overhead reimbursemens, ransfer of R&D, permis and franchises, rade of asses oher han goods, including exchange of fixed asses, capial and operaing leases and borrowing or lending, including ineres paymens. 3. PRIOR RESEARCH AND HYPOTHESES Concerns ha RP ransacions could adversely impac he reliabiliy of financial informaion boh in erms of represenaional faihfulness and reliabiliy of repored amouns dae back o 1982, when he FASB issued he Saemen of Financial Accouning Sandards No. 57. As poined ou in SFAS 57, ransacions involving relaed paries canno be presumed o be carried ou on an arm s-lengh basis, as he requisie condiions of compeiive, free-marke dealings may no exis. I furher elaboraes ha relaionships beween paries may enable one of he paries o exercise a degree of influence over he oher such ha he influenced pary may be favored or caused o subordinae i independen ineress (para. 13, FASB 1982). Such concerns did no draw much aenion unil he ouburs of he recen accouning scandals of Enron, Adelphia and ohers. Enron used special purpose eniies conrolled by is CEO o manipulae income and Adelphia guaraneed relaed pary deb and provided exensive loans o execuives. In boh cases, he resul was irrelevan and unreliable accouning informaion. Neverheless, here is limied empirical evidence on he impac of RP ransacions on he qualiy of key accouning informaion such as earnings. To he bes of our knowledge, Gordon and 4
Henry (2003) is he firs and only sudy o use US daa o show he impac of RP ransacions on earnings qualiy. They show ha absolue abnormal accruals are posiively associaed wih RP ransacions, suggesing ha his is due o he incenive of managers o manipulae earnings o mask heir expropriaion from RP ransacions. As abnormal accruals are well acceped as a measure of earnings qualiy (see, e.g., Dechow and Schrand 2004), Gordon and Henry s resuls imply ha RP ransacions may reduce earnings qualiy. The adverse impac of RP ransacions on earnings qualiy is also documened in a China seing. Jian and Wong (2010) show ha Chinese lised companies wih group srucure are more likely o engage in RP sales wih heir paren companies o manipulae earnings. Aharony e al. (2010) presen evidence of earnings managemen hrough RP sales during he IPO process in China. They also show ha abnormal RP sales are posiively associaed wih abnormal accruals. Overall, hese sudies provide empirical evidence of earnings managemen via or induced by RP ransacions. The adverse impac of RP ransacions on earnings qualiy implies an adverse impac on he usefulness of earnings o invesors insofar as earnings usefulness is highly dependen on earnings qualiy (Dechow and Schrand 2004). Following Francis, Schipper, and Vincen (2005), we use earnings informaiveness o measure he usefulness of accouning informaion. Accordingly, our firs hypohesis is saed as follows (in alernaive form): H 1 : Earnings informaiveness is negaively associaed wih he level of he firm s RP sales of goods and services. Prior sudies sugges ha financial analyss are credulous in he sense ha hey canno idenify earnings qualiy erosion due o facors such as opporunisic earnings managemen. For example, Teoh and Wong (2002) show ha financial analyss sysemaically underesimae he 5
exen of earnings managemen, and herefore are overly ready o believe ha high accruals reflec good news. As already noed, Aharony e al. (2010) show ha Chinese companies engage in RP sales of goods and services o manipulae heir earnings and ha here is a significan posiive associaion beween RP sales and absolue abnormal accruals. Consequenly, we expec o find ha Chinese financial analyss are also more credulous abou firms ha engage in higher level of RP sales of goods and services, and ha he resul will be a higher forecas error and more opimisic earnings forecass. Formally saed, our second hypohesis (in alernaive form) is: H 2 : Financial analyss earnings forecass are less accurae and more opimisic for firms ha engage in a higher level of RP sales of goods and services. 4. THE IMPACT OF RP SALES ON EARNINGS INFORMATIVENESS 4.1 Research design To es wheher earnings informaiveness is decreasing wih he level of RP sales of goods and services (H 1 ), we invesigae wheher differenial earnings informaiveness exiss beween firms wih RP sales and firms wihou RP sales. Following Eason and Harris (1991), Francis e al. (2005) and ohers, we assess informaiveness by he magniude of he earnings response coefficien (ERC), i.e., he slope coefficiens obained from regressions of annual reurns on boh he level of and change in annual earnings. We include ineracions beween earnings (and changes in earnings) and he level of RP sales and inerpre he respecive esimaed coefficiens as evidence of differences in earnings informaiveness or lack of such evidence. We also include in hese regressions four conrol variables ha have been shown in 6
prior research o affec earnings informaiveness (losses, firm size, marke o book raio and financial leverage). Accordingly, we run he following wo regression models: RET 4 k 0 EARN 2EARN * RPS k EARN * CON 1 k 1, (1a) RET 0 EARN 2EARN RPS EARN 4 EARN * * 1 3 RPS 4 k 1 4 k k k EARN * CON k EARN * CON k 1, (1b) where: RET = firm j s 12-monh cumulaive raw sock reurn over he period beginning in he fifh monh following he end of fiscal year -1 o he fourh monh afer fiscal year ; EARN = firm j s ne income in year, scaled by he marke value of equiy as of year-end - 1; ΔEARN = change in ne income beween year and -1, scaled by he marke value of equiy as of year-end -1; RPS = firm j s sales of goods and services o is relaed paries for year, scaled by oal ne sales for year ; CON, = a vecor of conrol variables, k = 1, 2, 3, and 4: k j 1 CON = LOSS = 1 if EARN < 0, and 0 oherwise; 2 CON = SIZE = he naural log of firm j s oal asses as of year-end ; 3 CON = MBE = he raio of firm j s marke value of equiy o book value of equiy as of year-end ; 7
4 CON = DEBT = firm j s leverage, measured as he raio of firm j s long-erm deb o oal book value of asses as of year-end. We include he four conrol variables in our regression analyses because prior sudies have shown ha each is associaed wih he magniude of he ERC, our measure of earnings informaiveness. Hayn (1995) documens ha he ERC is smaller for loss observaions han for profi observaions; Chaney and Jeer (1992) find he ERC is increasing in firm size while Freeman (1987) repors ha i is negaively relaed o firm size; Collins and Kohari (1989) show ha he marke-o-book raio, a common measure of he firm s growh, is posiively associaed wih earnings informaiveness; finally, Dhaliwal e al. (1991) repor ha he ERC is negaively associaed wih financial leverage. To es hypohesis H 1 we examine he magniude of he esimaed coefficien for he ineracion beween earnings and he level of RP sales ( 2 ) in model (1a). Similarly, for model (1b) we examine he magniude of ( 2 + 4 ), he sum of he esimaed coefficiens for he ineracions of level of RP sales wih earnings and change in earnings, respecively. An adverse impac of RP sales on earnings informaiveness implies 2 < 0 for he firs model and ( 2 + 4 ) < 0 for he second model. 4.2 Sample selecion and daa Our sample consiss of 140 companies ha were randomly seleced wihou replacemen from he enire populaion of 383 companies lised on he Shanghai Sock Exchange in 1997. As shown below, our sample firms characerisics well represen hose of he enire populaion. For each firm, we manually colleced all daa on RP sales of goods and services from he annual repors for he seven-year period 1998 o 2004, a oal of 980 firm-year observaions. Oher 8
accouning and financial informaion was obained from he China Sock Marke & Accouning Research (CSMAR) daabase, he leading supplier of boh financial accouning daa and sock prices for all lised companies in China. Table 1 presens he sample composiion classified by nine major indusries (wo-digi SIC code). The indusry classificaion was firs obained from he China Securiies Regulaory Commiee. We hen reclassified he indusries ino nine caegories based on Campbell (1996). As here are only a small number of firms in he peroleum indusries (SIC code 13, 29), we combine hem wih he basic indusries. The sample excludes he uiliy indusries (SIC code 46, 48, 49) and he financial services indusries (SIC code 60-69). As Table 1 shows, he oal sample is abou 37% of he enire populaion. Per indusry, we randomly seleced from 25% o 35% of he respecive populaion excep for he services indusry (76%) and he ransporaion indusry (53%). The las wo columns of Table 1 presen he median size (in erms of oal asses) of our sample firms per indusry as of December 31, 1998 and he corresponding figures for he enire populaion per indusry, respecively. A Wilcoxon signed-rank es (no abulaed) indicaes ha none of he paired medians are significanly differen from each oher a he 5 percen level or lower. [Inser Table 1 Here] Table 2 presens summary saisics for he variables in models (1a) and (1b). We winsorize he observaions of each coninuous independen variable by half a percen in each ail. As depiced in he able, on he average, he 140 sample firms earned an 11.78% annual reurn during he sample period 1998 o 2004, compared wih a 9.35% annual reurn for he 243 non- 9
sample firms in he same period. The mean of earnings scaled by marke value is abou 1.68% while he mean of change in earnings scaled by marke value is close o 0% (compared wih 1.95% and 0%, respecively, for he non-sample firms). Noably, on he average, RP sales of goods and services accoun for 11.56% of oal ne sales and more han half (51.47%) of he sample firms have RP sales wih heir paren companies. As for he conrol variables, 8.63 percen of he sample firms repor losses during he sample period; on he average, size measured by oal asses book value is 2,003 million RMB (oal equiy marke value is 3,110 million RMB), marke-o-book raio is 3.90 and financial leverage raio is 7.75%. In comparison, of he remaining 243 non-sample firms lised on he Shanghai Sock Exchange during he same period, 9.58 percen repor losses, and on he average, heir size, measured by oal asses, is 2,319 million RMB (oal equiy marke value is 3,310 million RMB), heir marke-o-book raio is 4.33, and heir financial leverage raio is 7.15%. Eiher a -es beween mean values or a Wilcoxon signed-rank es beween medians (no abulaed) indicaes ha here is no significan difference beween our randomly seleced sample and he remaining non-seleced firms on he Shanghai Sock Exchange for any of he summary saisics repored above. [Inser Table 2 here] 4.3 Empirical resuls We use pooled ordinary leas square (OLS) regressions o esimae models (1a) and (1b). To conrol for possible heeroskedasiciy and auocorrelaion problems, we also employ woway fixed effecs panel daa regressions. The disadvanage of he fixed effecs analysis is is relaively low power, due o he large number of dummy variables ha may sap he model of he 10
degrees of freedom required for adequaely powerful saisical ess. Moreover, a model wih a large number of such variables may be plagued wih mulicollineariy, which increases he sandard errors, hereby draining he model of saisical power o es parameers. Panel A of he Appendix shows he correlaion coefficiens among he variables in he earnings informaiveness analysis. As expeced, boh earnings level and changes are posiively associaed wih annual sock reurns. Among he independen variables, we observe large correlaion coefficiens, paricularly among earnings level, earnings changes, and losses. For example, he correlaion coefficien beween firm losses (LOSS) and earnings level (EARN) is - 0.684 wih p-value of 0.000. To alleviae concerns abou problems of mulicollineariy among he independen variables, we calculae he Variance Inflaion Facor (VIF) for all OLS regression variables. Neer, Wasserman and Kuner (1983) sugges ha a VIF level below 10 indicaes he absence of mulicollineariy problems. The resuls of hese ess indicae ha none of he independen variables has a VIF value ha exceeds 2. We conclude, herefore, ha no serious mulicollineariy problem is presen in he regression analysis. The regression resuls repored in Table 3 srongly suppor hypohesis H 1 ha earnings informaiveness is negaively associaed wih he firm s level of RP sales of goods and services. This is eviden from he esimaed coefficiens of our main variables of ineres, EARN*RPS and ΔEARN*RPS. In model (1a), he esimaed coefficien of EARN*RPS, ˆ 2, is -2.501 (p-value = 0.001) for he pooled regression and -1.640 (p-value = 0.008) for he wo-way fixed effec regression. In model (1b), he esimae of ( ˆ ˆ 2 4 ) is -3.933 (p-value of 0.030) for he pooled regression and -3.396 (p-value of 0.019) for he wo-way fixed effec regression. [Inser Table 3 here] 11
Consisen wih prior research (e.g., Eason and Harris 1991 and Francis e al. 2005 in he US, and Chen, Chen, and Su, 2001 in China) all regression resuls in Table 3 show ha he earnings response coefficiens (ERC), relaing sock reurns o he level and change in earnings, are posiive and significan. For example, for he pooled regression, ˆ1, he esimaed coefficien of EARN, is 5.747 (p-value of 0.001) in model (1a) and ˆ, he esimaed coefficien of ΔEARN 3 is 3.781 (p-value of 0.048) in model (1b). The effecs of RP sales on he magniude of deerioraion in earnings informaiveness may be assessed by he raio of ˆ 2 o ˆ1 in model (1a), or he raio of ( ˆ ˆ 2 4 ) o ( ˆ ˆ 1 3) in model (1b). The coefficien esimaes obained from he wo-way fixed effec regressions indicae ha earnings of firms wih RP sales are abou 45% - 48% less informaive ((-1.640/3.637) =-45% in model (1a) and -3.396/(4.749 + 2.400) = -48% in model (1b)). Based on pooled regressions, earnings of firms wih RP sales are abou 33% - 44% less informaive (-2.501/5.747= -44% in model (1a) and -3.933/(7.975 + 3.781) = -33% in model (1b)). The esimaed coefficiens for he conrol variables in boh regressions are generally consisen wih hose repored in prior lieraure. The loss (LOSS) observaions reduce he ERC, as evidenced by he negaive and highly significan coefficien esimaes ˆ 1 in model (1a) and ( ˆ 1 + ˆ 1 ) in model (1b). These resuls are consisen wih hose of Hayn (1995). Firm size (SIZE) increases earnings informaiveness, as evidenced by he posiive and highly significan coefficien esimaes ˆ 2 in model (1a) and ( ˆ 2 + ˆ 2 ) in model (1b) obained by he wo-way fixed effecs regressions (hough hey are posiive bu saisically insignifican in he pooled regressions). These resuls are consisen wih hose of Chaney and Jeer (1992). Consisen wih 12
Collins and Kohari (1989), he resuls for he pooled regression show ha he marke-o-book raio of equiy (MBE) is posiively associaed wih earnings informaiveness. This is evidenced by he posiive and highly significan coefficien esimaes ˆ 3 in model (1a) and ( ˆ 3 + ˆ 3 ) in model (1b). However, for he wo-way fixed effecs regressions he corresponding coefficien esimaes are negaive and highly significan. We have no explanaion for his. Lasly, consisen wih Dhaliwal e al. (1991), we find a negaive effec of financial leverage (DEBT) on earnings informaiveness. This is demonsraed by he negaive coefficien esimaes ˆ 4 in model (1a) and ( ˆ 4 + ˆ 4 ) in model (1b) in boh he pooled and wo-way fixed effecs regressions. However, while in he former he corresponding coefficien esimaes are highly saisically significan, in he laer hey are no. 5. THE IMPACT OF RP SALES ON THE QUALITY OF ANALYSTS EARNINGS FORECASTS 5.1 Research design To es wheher financial analyss earnings forecass are less accurae and more opimisic for firms ha engage in a higher level of RP sales of goods and services (H 2 ), we firs esimae wo imporan properies of earnings forecass: forecas accuracy and forecas bias. Following he radiion in he lieraure (e.g., Duru and Reeb 2002), we measure financial analyss earnings forecas accuracy, for each firm-year observaion, by he absolue value of he difference beween he consensus analyss earnings per share forecas and he acual earnings per share, scaled by he sock price a he laes forecas dae: 1 FORECAST EPS ACCU ( 1), (2) PRICE 1 13
where ACCU = he negaive of he consensus absolue forecas error for year ; 1 FORECAST = he consensus analyss earnings per share forecas for year made prior o acual earnings announcemen for year ; EPS = he acual earnings per share for year ; PRICE -1 = he sock price per share a he laes forecas dae. Muliplying his absolue forecas error by (-1) gives a measure ha increases wih greaer forecas accuracy. Thus, when a firm aribue is posiively associaed wih ACCU, such an aribue conribues o a more accurae analys forecas. We measure financial analyss earnings forecas bias (BIAS), for each firm-year observaion, as he signed forecas error, defined as he difference beween he consensus analyss earnings per share forecas and he acual earnings per share, scaled by he sock price a he laes forecas dae: 1 FORECAST EPS BIAS ( OPTIMISM). (3) PRICE 1 Forecas opimism increases as BIAS becomes larger. Thus, when a firm aribue is posiively associaed wih BIAS, such an aribue conribues o a more opimisic analys forecas. We nex examine he associaion beween RP sales of goods and services and each of he wo forecas properies, ACCU and BIAS, by regressing each measure separaely on he level of RP sales of goods and services and on six conrol variables ha have been shown in prior sudies o influence he wo properies of analyss forecasing: accuracy and bias. Accordingly, we run he following regression model: 14
ACCU ( BIAS ) RPS SIZE LOSS EARN HORZ 0 1 2 3 4 5 6FOL 7DISPj, TimeDummie s IndusryDummies, (4) where: ACCU = firm j s earnings forecas accuracy in year as defined by Eq (2); BIAS = firm j s earnings forecas bias in year as defined by Eq (3); RPS = firm j s sales of goods and services o is relaed paries for year, scaled by oal ne sales in year ; SIZE = he naural log of firm j s oal asses as of year-end ; LOSS = a dummy variable, equal o one when firm j s ne income for year is negaive, and zero oherwise; ΔEARN = change in ne income beween year and -1, scaled by he marke value of equiy as of year-end -1; HORZ = firm j s forecas horizon in year, expressed as he number of monhs beween he forecas s monh and he end of fiscal year ; FOL = number of financial analyss following firm j in year ; DISP = financial analyss forecas dispersion, measured as he sandard deviaion of firm j s analyss forecass in year deflaed by he sock price per share a he forecas dae. TimeDummies = six dummies for he seven-year period 1998-2004. IndusryDummies = eigh dummies for he nine indusries oulined in Table 1. Duru and Reeb (2002) find ha firm size (SIZE) likely indicaes more complexiy and hus greaer forecas error, bu, a he same ime, he availabiliy of more pre-disclosure informaion by larger firms may lead o lower forecas error. Hwang e al. (1996) documen ha 15
analyss forecass of losses (LOSS) are less accurae han heir forecass of profis, and Brown (2001) provides evidence ha analyss issue more opimisic forecass in loss periods. Lang and Lundholm (1996) find ha larger changes in earnings (ΔEARN) are associaed wih less accurae forecass. Prior sudies (e.g., Brown 1993) indicae ha longer forecas horizons (HORZ) are associaed wih less accurae analyss earning forecass. Lys and Soo (1995) find ha forecas accuracy increases wih he number of financial analyss following he firm (FOL), whereas Das e al. (1998) sugges less opimisic forecass for more heavily followed firms. Lang and Lundholm (1996) find ha forecas dispersion (DISP) is negaively relaed o analyss forecas accuracy. Finally, o conrol for he poenial impac of changes in macro economic condiions and indusry complexiy, hroughou he sample period, on analyss' forecasing behavior, we add ime-dummies and indusry-dummies o regression model (4). 5.2 Earnings per share forecas daa Chinese earnings forecas daa were obained from Invesoday Ld., a leading financial informaion vendor based in Shenzhen, China. The daa consis of annual earnings per share forecass for lised companies covered by Chinese sell-side analyss since 2002. We obained annual earnings per share forecas daa for he hree-year period 2002-2004 and merged i wih he daa described in secion IV for our 140 sample firms. This procedure resuled in a reduced sample consising of a oal of 361 firm-year observaions represening he 140 firms during he period 2002-2004. Due o he fac ha some sample companies were no covered by financial analyss in he early years, he final number of firm-year observaions is 361 raher han 420, i.e., 140 imes 3. 16
Panel B of he Appendix shows he correlaion coefficiens among he variables paricipaing in he analyss forecass analysis. As expeced, RP sales of goods and services as a percenage of oal ne sales are associaed wih less accurae and more opimisic forecass, consisen wih hypohesis H 2. We also find ha analyss issue less accurae and more opimisic forecass in loss firm-years, as does Brown (2001). Similarly o he approach aken in secion IV, we calculae he Variance Inflaion Facor (VIF) for all OLS regression variables o alleviae concerns abou mulicollineariy problems among he independen variables. The resuls of hese ess indicae ha none of he independen variables has a VIF value ha exceeds 3. We conclude, herefore, ha no serious mulicollineariy problem is presen in he regression analysis. Table 4 presens summary saisics for he variables in regression model (4). We winsorize he observaions of each coninuous independen variable by half a percen in each ail. The mean and median accuracy (ACCU) values are negaive by consrucion. The closer he accuracy value is o zero, he more accurae is he forecas. Thus, variables ha are posiively relaed o ACCU are associaed wih more accurae forecass. Consisen wih US analyss behavior (e.g., Richardson e al. 2004), he mean and median signed forecas errors (BIAS) are posiive, indicaing ha Chinese analyss are, on he average, opimisic. Variables ha are posiively relaed o BIAS are associaed wih more opimisic forecass. As mean and median consensus forecas daa give similar resuls, we repor only he resuls based on he mean forecass. Noably, compared wih he full sample used in secion IV, he reduced sample firms have a lower average raio of RP sales of goods and services over oal ne sales (7.49% versus 11.56%) bu a higher median raio (0.46% versus 0.03%). In addiion, 13.3% of he reduced sample firms repor losses, compared wih 8.63% of he full sample. Firm size measured by oal asses indicaes ha he reduced sample is, on he average, larger, probably because is saisics 17
is based on more recen years. As for he remaining analys forecass conrol variables, on he average, analyss make heir las forecass abou one monh (0.967) prior o he acual earnings announcemen dae; here are five analyss following each sample firm; and forecas dispersion is abou 0.5% of sock prices a he forecas dae. [Inser Table 4 here] 5.3 Empirical resuls We use pooled ordinary leas square (OLS) regressions o esimae model (4). The regression resuls repored in Table 5 srongly suppor hypohesis H 2 ha financial analyss earnings forecass are less accurae and more opimisic for firms wih a higher level of RP sales of goods and services. These resuls are obained afer conrolling for previously idenified deerminans of forecas accuracy and bias. Regression resuls using forecas accuracy as a dependen variable are presened in he second column and hose wih forecas bias as a dependen variable in he hird column. Our primary resul is ha he level of RP sales of goods and services (RPS), in he forecas accuracy model, is negaively and significanly associaed wih forecas accuracy ( ˆ1 is -0.003, significan a he 5% level), indicaing ha analyss provide less accurae forecass for firms ha have a higher level of sales of goods and services wih heir relaed paries. This may sugges ha Chinese financial analyss do no adequaely accoun for he adverse impac of RP sales on earnings. Consisen wih prior research, he resuls also show highly significan negaive coefficien esimaes (p-value = 0.001 or lower) for firm losses (LOSS), changes in earnings (ΔEARN), forecas horizon (HORZ) and forecas dispersion (DISP), suggesing ha 18
each of hese four conrol variables is associaed wih less accurae analys forecass. Only he coefficien esimaes for firm size (SIZE) and analys following (FOL) are insignificanly differen from zero. [Inser Table 5 here] The las column of Table 5 repors he resuls of model (4) using earnings forecas bias (BIAS) as a dependen variable. We find ha he level of RP sales (RPS) is posiively associaed wih earnings forecas bias ( ˆ1 = 0.004, significan a he 5% level), This resul is consisen wih our predicion ha financial analyss provide more opimisic earnings forecass for firms ha engage in a higher level of RP sales of goods and services wih heir relaed paries. Our inerpreaion of his finding is ha Chinese financial analyss are credulous abou RP sales and hus provide more opimisic earnings forecass. Consisen wih prior research, he resuls also show highly significan posiive coefficien esimaes (p-value = 0.001) for firm losses (LOSS), changes in earnings (ΔEARN), forecas horizon (HORZ) and forecas dispersion (DISP), suggesing ha each of hese four conrol variables is associaed wih more opimisic earnings forecass. Only he coefficien esimaes for firm size (SIZE) and analys following (FOL) are insignificanly differen from zero. These resuls are also consisen wih our earnings predicion accuracy model. 6. SUMMARY AND CONCLUSIONS Anecdoal evidence suggess ha RP ransacions impair he represenaional faihfulness and verifiabiliy of accouning daa. Thus, for example, he misreporing of profiabiliy via RP ransacions by a former CFO of Enron only became apparen as he firm began o collapse. In his paper we explore wheher RP sales of goods and services reduce earnings informaiveness o 19
capial marke invesors and consequenly affec he qualiy of earnings forecass by financial analyss. We focus on RP sales of goods and services because hey affec accouning earnings more direcly han oher ypes of RP ransacions such as direcors loans. Using a random sample of 140 firms from he Shanghai Sock Exchange, we find a significan negaive associaion beween he level of RP sales of goods and services and earnings informaiveness measured by he earnings response coefficien (ERC), afer conrolling for oher facors known o affec he ERC. Earnings of firms engaged in RP sales are a leas 33% less informaive han hose of oher firms. The resuls are consisen wih he conenion ha RP sales violae he arm s-lengh assumpion for regular ransacions and hence reduce he qualiy of accouning earnings. We also find ha financial analyss credulously accep misleading RP sales daa and provide less accurae and more opimisic earnings forecass for firms wih more RP sales of goods and services. Overall, our resuls show ha RP sales of goods and services reduce he usefulness of accouning earnings o capial marke invesors and make i more difficul for financial analyss o provide accurae earnings forecass. The paper has, a leas, wo implicaions. The resuls have policy implicaions for China and oher markes wih pervasive relaed pary ransacions. As oulined in Aharony e al. (2010), he Chinese governmen has been aking acions o minimize he negaive impac of relaed pary sales beween lised companies and heir paren companies. For example, since 2006, he Chinese governmen has been encouraging unlised paren companies o ake he enire eniy public, in order o creae a sandard-alone lised company raher han a paren-subsidiary group which involves pervasive relaed pary sales. Our resuls also sugges ha financial analyss should no be unduly willing o rus earnings numbers ha are conaminaed by unreliable relaed pary sales. However, he limiaion is ha financial analyss canno observe he qualiy of 20
relaed pary sales direcly from companies financial disclosures. Thus hey should acquire more privae informaion abou relaed pary sales. 21
APPENDIX: Pearson and Spearman Correlaions Panel A: Correlaion coefficiens among he variables paricipaing in he earning informaiveness analysis a RET EARN ΔEARN RPS LOSS SIZE MBE DEBT RET 0.393*** 0.173*** 0.041 (0.200) -0.299*** -0.178*** -0.112*** -0.043 (0.177) EARN 0.492*** 0.612*** 0.096*** (0.003) -0.684*** 0.180*** -0.060 (0.058) 0.076** (0.017) ΔEARN 0.305*** 0.522*** -0.009 (0.771) 0.476*** -0.011 (0.728) -0.065** (0.043) -0.029 (0.358) RPS 0.004 (0.900) 0.031 (0.326) 0.013 (0.684) -0.059 (0.064) 0.190*** -0.059 (0.063) 0.095*** (0.003) LOSS -0.316*** -0.486*** 0.407*** -0.050 (0.117) -0.131*** 0.063** (0.047) -0.026 (0.407) SIZE -0.148*** 0.195*** -0.020 (0.536) 0.165*** -0.139*** -0.349*** 0.219*** MBE -0.066** (0.039) -0.190*** 0.013 (0.680) -0.051 (0.111) 0.03 (0.319) -0.375*** -0.066** (0.038) DEBT -0.043 (0.182) 0.065** (0.041) 0.014 (0.669) -0.006 (0.854) -0.016 (0.609) 0.206*** -0.090*** (0.005) a Based on daa for 140 sample firms in he period from 1998 o 2004 (980 firm-year observaions). Panel A repors he correlaion marix for he variables in he regression models (1a) and (1b). The upper (lower) diagonal repors he Pearson (Spearman) correlaion coefficiens. p-values are provided in parenheses. ** and *** denoe significance a a level of 5% and 1%, respecively. Variable definiions: RET = firm j s 12-monh cumulaive raw sock reurn over he period beginning 4 monhs following he end of fiscal year -1 and ending 4 monhs afer fiscal year ; EARN = firm j s ne income in year, scaled by he marke value of equiy as of year-end -1; ΔEARN = change in ne income beween year and -1, scaled by he marke value of equiy as of year-end -1; RPS = firm j s sales of goods and services o is relaed paries for year, scaled by oal ne sales for year ; LOSS = a dummy variable, equal o one when firm j s ne income for year is negaive, and zero oherwise; SIZE = he naural log of firm j s oal asses as of year-end ; MBE = he raio of firm j s marke value of equiy o book value of equiy as of year-end ; DEBT = firm j s leverage, measured as he raio of firm j s long-erm deb o oal book value of asses as of year-end. 22
Panel B: Correlaion coefficiens among he variables paricipaing in he analyss earnings forecas analysis a ACCU BIAS RPS SIZE LOSS ΔEARN HORZ FOL DISP ACCU -0.491*** -0.131** (0.041) 0.152** (0.016) -0.573*** 0.359*** -0.096 (0.131) 0.288*** 0.050 (0.429) BIAS -0.967*** 0.126** (0.050) -0.029 (0.646) 0.583*** -0.622*** 0.059 (0.352) -0.138** (0.029) -0.080 (0.210) RPS -0.289*** (0.002) 0.330*** 0.134** (0.034) -0.045 (0.483) 0.073 (0.252) -0.003 (0.964) 0.095 (0.134) 0.022 (0.735) SIZE 0.189*** (0.003) -0.175*** (0.006) 0.124 (0.051) -0.291*** 0.106 (0.094) -0.135** (0.033) 0.508*** 0.394*** LOSS -0.674*** 0.668* -0.023-0.297*** -0.547*** -0.073-0.372*** -0.242*** (0.718) (0.249) ΔEARN 0.688*** -0.720*** -0.021 (0.731) 0.086 (0.176) -0.620*** 0.011 (0.858) 0.205*** (0.001) 0.143** (0.023) HORZ -0.072 (0.257) 0.078 (0.218) 0.061 (0.335) -0.139** (0.028) -0.076 (0.229) 0.044 (0.490) -0.112 (0.078) -0.032 (0.617) FOL 0.181 (0.004) -0.160** (0.012) 0.097 (0.124) 0.486*** -0.254*** 0.151** (0.017) -0.106 (0.095) 0.707*** DISP -0.209*** (0.001) 0.173*** (0.006) 0.056 (0.382) 0.111 (0.081) 0.014 (0.829) 0.017 (0.791) -0.009 (0.885) 0.174*** (0.006) a Based on daa for 140 sample firms in he period from 2002 o 2004 (361 firm-year observaions). Panel B repors he correlaion marix for he variables in he regression model (4). The upper (lower) diagonal repors he Pearson (Spearman) correlaion coefficiens. p-values are provided in parenheses. ** and *** denoe significance a a level of 5% and 1%, respecively. Variable definiions: ACCU = firm j s earnings forecas accuracy in year as defined by Eq (2); BIAS = firm j s earnings forecas bias in year as defined by Eq (3); RPS = firm j s sales of goods and services o is relaed paries, scaled by oal ne sales in year ; SIZE = he naural log of firm j s oal asses as of year-end ; LOSS = a dummy variable, equal o one when firm j s ne income for year is negaive, and zero oherwise; ΔEARN = change in ne income beween year and -1, scaled by he marke value of equiy as of year-end -1; HORZ = firm j s forecas horizon in year, expressed as he number of monhs beween he forecas monh and he end of fiscal year ; FOL = number of financial analyss following firm j in year ; DISP = financial analyss forecas dispersion, measured as he sandard deviaion of firm j s analyss forecass in year deflaed by he sock price per share a he forecas dae. 23
REFERENCES Aharony, J., J. Wang, and H. Yuan. 2010. Tunneling as an incenive for earnings managemen during he IPO process in China. Journal of Accouning and Public Policy 29: 1-26. Brown, L. D. 1993. Earnings forecasing research: Is implicaions for capial marke research. Inernaional Journal of Forecasing 9 (November): 295-320. Brown, L. D. 2001. A emporal analysis of earnings surprises: Profis versus losses. Journal of Accouning Research 39 (Sepember): 221-242. Campbell, J. Y. 1996. Undersanding risk and reurn. Journal of Poliical Economy 104: 298-345. Chaney, P. and D. Jeer. 1992. The effec of size on he magniude of long-window earnings response coefficiens. Conemporary Accouning Research 8, 540-560. Chen, C. J. P., S. Chen, and X. Su. 2001. Is accouning informaion value-relevan in he emerging Chinese sock marke? Journal of Inernaional Accouning, Audiing & Taxaion 10: 1-22. Collins, D., and S.P. Kohari. 1989. An analysis of ineremporal and cross-secional deerminans of earnings response coefficiens. Journal of Accouning and Economics 11, 143-181. Das, S., C. Levine, and K. Sivaramakrishnan. 1998. Earnings predicabiliy and bias in analyss earnings forecass. The Accouning Review 73 (April): 277-294. Dechow, P. M., and C. M. Schrand. 2004. Earnings Qualiy. The Research Foundaion of CFA Insiue, USA. Dhaliwal, D., Lee, K., and N. Fargher. 1991. The associaion beween unexpeced earnings and abnormal securiy reurns in he presence of financial leverage. Conemporary Accouning Research 8, 20-41. Duru, A. and D. M. Reeb. 2002. Inernaional diversificaion and analyss forecas accuracy and bias. The Accouning Review 77: 415-433. Eason, P., and T. Harris. 1991. Earnings as an explanaory variable for reurns. Journal of Accouning Research 29, 19-36. Financial Accouning Sandards Board (FASB). Saemen of Financial Accouning Sandard (SFAS) No. 57. 1982. Relaed pary ransacions. 24
Francis, J., K. Schipper, and L. Vincen. 2005. Earnings and dividend informaiveness when cash flow righs are separaed from voing righs. Journal of Accouning and Economics 39, 329-360. Freeman, R. 1987. The associaion beween accouning earnings and securiy reurns for large and small firms. Journal of Accouning and Economics 55, 615-646. Gordon, E., and E. Henry. 2003. Relaed pary ransacions and earnings managemen. Working paper. Rugers Universiy. Gordon, E., E. Henry., and D. Palia. 2004. Relaed pary ransacions: Associaions wih corporae governance and firm value. Working paper. Rugers Universiy. Hayn, C., 1995. The informaion conen of losses. Journal of Accouning and Economics 11, 123-153. Hwang, L., C. Jan, and S. Basu. 1996. Loss firms and analyss earnings forecas errors. Journal of Financial Saemen Analysis 1 (Winer): 18-30. Jian, M., and T. Wong. 2010. Propping hrough relaed pary ransacions. Review of Accouning Sudies 15: 70-105. Lang, M., and Ro. Lundholm. 1996. Corporae disclosure policy and analys behavior. The Accouning Review 71 (Ocober): 467-492. Lys, T., and L. Soo. 1995. Analyss forecas precision as a response o compeiion. Journal of Accouning, Audiing, and Finance 10 (Fall): 751-763. Neer, J., W. Wasserman, and M. Kuner. 1983. Applied Regression Models. Homewood, IL: Richard D. Irwin. Richardson, S., S. Teoh, and P. Wysocki. 2004. The walkdown o beaable analys forecass: The roles of equiy issuance and insider rading incenives. Conemporary Accouning Research 21 (Winer): 885-924. Teoh, S.H., and T.J. Wong. 2002. Why new issues and high-accrual firms underperform: The role of analyss creduliy. The Review of Financial Sudies 15: 869-900. 25
TABLE 1 Sample Firm Composiion by Indusry Indusry Two-digi SIC Code Number of Sample Firms Percenage of Toal Populaion a Median Toal Asses of Sample Firms as of 12.31.1998 (million RMB) Median Toal Asses of Toal Populaion as of 12.31.1998 (million RMB) Food and obacco 1, 2, 9, 20, 21, 54 Basic indusries including peroleum 10, 12, 13, 14, 24, 26, 28, 29, 33 Consrucion 15, 16, 17, 32, 52 16 28% (57) 740.58 841.48 22 34% (64) 682.16 706.00 28 35% (79) 1,076.57 837.22 Texiles and rade 22, 23, 31, 5 33% (15) 1,307.51 796.91 51, 53, 56, 59 Consumer durables 25, 30, 36, 23 29% (78) 1,031.70 1047.58 37, 39, 50, 55, 57 Capial goods 34, 35, 38 7 35% (20) 928.10 934.44 Transporaion 40, 41, 42, 44, 45, 47 9 53% (17) 910.99 785.52 Services 72, 73 75, 25 76% (33) 787.52 949.87 76, 80, 82, 87, 89 Conglomerae No 5 25% (20) 877.67 938.73 specific SIC code Enire sample 140 37% (383) 901.33 881.63 a The number of lised firms in he enire populaion per indusry is in parenheses. The able shows he sample composiion of 140 firms randomly seleced from he enire populaion of 383 firms lised on he Shanghai Sock Exchange in 1997, classified by nine major indusries (wo-digi SIC code). The indusry classificaion is based on Campbell (1996). As he number of firms in he peroleum indusries (SIC code 13, 29) is small, we combine hem wih he basic indusries. The sample excludes he uiliy indusries (SIC code 46, 48, 49) and he financial services indusries (SIC 60-69). 26
TABLE 2 Summary Saisics for he Variables in Regression Models (1a) and (1b) a Mean Values for Sd. Variable Mean Median he 243 Nonsample Firms b Dev. 12-monh sock reurn (RET) 0.1178 0.0140 0.5035 0.0935 Earnings as % of marke value (EARN) 1.68% 1.93% 3.59% 1.95% ΔEarnings as % of marke value (ΔEARN) 0.00% -0.05% 3.61% 0.00% RP sales as % of oal ne sales (RPS) 11.56% 0.03% 20.05% NA c % RP sales > 0 51.47% -- -- NA Toal RP sales (million RMB) 141.70 0.28 627.30 NA Toal ne sales (million RMB) 1,213.93 599.65 1,648.67 1,169.78 % Earnings < 0 (LOSS) 8.63% -- -- 9.58% Toal asses (million RMB) 2,002.77 1,285.52 2,210.20 2,313.83 Marke-o-book raio (MBE) 3.90 3.38 2.39 4.33 Long-erm deb as % of oal asses (DEBT) 7.75% 2.74% 11.18% 7.15% Equiy marke value (million RMB) 3,110.28 2,369.61 2,860.71 3310.23 a Based on daa for 140 sample firms in he period from 1998 o 2004 (980 firm-year observaions). b The enire populaion of 343 firms less our 140 sample firms. c No Available. Variable definiions: 12-monh sock reurn (RET) is firm j s sock reurn for year, beginning four monhs following he end of year -1 and ending four monhs afer he end of year ; Earnings as % of marke value (EARN) is firm j s ne income (before exraordinary iems) in year scaled by is marke value of equiy as of year-end -1; ΔEarnings as % of marke value (ΔEARN) is firm j s ne income in year minus ne income in year -1, scaled by is marke value of equiy as of year-end -1; RP sales as % of oal ne sales (RPS) is firm j s sales of goods and services o is relaed paries in year, scaled by is oal ne sales in year ; % RP sales >0 is he percenage of firms wih RP sales greaer han zero in year ; Toal RP sales is firm j s oal sales of goods and services o is relaed paries in year ; Toal ne sales is firm j s oal sales revenue ne of sales allowance, sales discoun, and sales reurns in year ; % Earnings < 0 (LOSS) is he percenage of firms wih repored negaive ne income in year ; Toal asses is firm j s oal book value of asses as of year-end ; Marke-o-book raio (MBE) is he raio of firm j s marke value of equiy o book value of equiy as of year-end ; Long-erm deb as % oal asses (DEBT) is he raio of firm j s long-erm deb o oal book value of asses as of year-end ; Equiy marke value is firm j s closing price as of year-end imes oal number of ousanding common shares as of year-end. 27
TABLE 3 The Impac of Relaed Pary Sales on Earnings Informaiveness Regression Resuls of Models (1a) and (1b) a Variable (coefficien) Two-Way Fixed Effecs Pooled Regressions Regressions Model (1a) Model (1b) Model (1a) Model (1b) a Inercep ( ˆ ) -0.122*** -0.114-0.354*** -0.345*** 0 (0.003) (0.003) EARN ( ˆ ) 5.747*** 7.975*** 3.637*** 4.749*** 1 (0.001) (0.004) (0.002) EARN*RPS ( ˆ ) -2.501*** -2.772*** -1.640*** -2.072*** 2 (0.001) (0.002) (0.008) (0.007) ΔEARN ( ˆ ) 3.781** 2.400 3 (0.048) (0.090) ΔEARN*RPS ( ˆ ) -1.161-1.324 4 (0.312) (0.122) EARN*LOSS ( ˆ 1 ) -8.798*** -8.392*** -11.535*** -9.004*** EARN*SIZE ( ˆ 2 ) 0.177 0.109 0.500*** 0.466*** (0.053) (0.303) EARN*MBE ( ˆ 3 ) 1.063*** 0.875*** -0.457*** -0.911*** (0.003) (0.038) (0.001) EARN*DEBT ( ˆ 4 ) -13.234** -17.337*** -4.997-3.369 (0.012) (0.005) (0.313) (0.574) ΔEARN*LOSS ( ˆ 1 ) 0.084 (0.970) 3.587** (0.032) ΔEARN*SIZE ( ˆ 2 ) -0.065 (0.526) -0.074 (0.330) ΔEARN*MBE ( ˆ 3 ) -0.358 (0.303) -0.623** (0.018) ΔEARN*DEBT ( ˆ 4 ) -12.122 (0.219) 0.217 (0.976) Two-way fixed effecs (firm and ime dummies) No included No included Included Included No. of observaions 980 980 980 980 Adj. R-square 0.221 0.221 0.697 0.706 EARN*RPS + ΔEARN*RPS ( ˆ + 2 ˆ 4 ) -3.933** (0.030) -3.396** (0.019) Based on daa for 980 firm-year observaions (140 sample firms for he seven-year period from 1998 o 2004). The able repors he resuls for regression models (1a) and (1b): RET 4 k 0 EARN 2EARN * RPS k EARN * CON 1 k 1, (1a) 28
RET 0 EARN 2EARN RPS EARN 4 EARN * * 1 3 RPS 4 k 1 4 k k k EARN * CON k EARN * CON k 1, (1b) Variable definiions: RET is firm j s 12-monh sock reurn for year, beginning four monhs following he end of year -1 and ending four monhs afer he end of year ; EARN is firm j s ne income (before exraordinary iems) in year scaled by is marke value of equiy as of year-end -1; ΔEARN is firm j s ne income in year minus ne income in year -1, scaled by is marke value of equiy as of year-end -1; RPS is firm j s sales of goods and services o is relaed paries in year, scaled by is oal ne sales in year ; LOSS akes value of one if EARN is posiive and zero oherwise; SIZE is he naural log of firm j s oal asses in year ; MBE is he raio of firm j s marke value of equiy o book value of equiy as of year-end ; DEBT is he raio of firm j s long-erm deb o oal book value of asses as of year-end ; p-values are repored in parenheses. ** and *** denoe wo-ailed significance a he 5% and 1% level, respecively. 29